git.oblomov.eu Git - linux-2.6/blob - fs/btrfs/extent_map.c

   1 #include <linux/bitops.h>
   2 #include <linux/slab.h>
   3 #include <linux/bio.h>
   4 #include <linux/mm.h>
   5 #include <linux/gfp.h>
   6 #include <linux/pagemap.h>
   7 #include <linux/page-flags.h>
   8 #include <linux/module.h>
   9 #include <linux/spinlock.h>
  10 #include <linux/blkdev.h>
  11 #include <linux/swap.h>
  12 #include <linux/version.h>
  13 #include <linux/writeback.h>
  14 #include <linux/pagevec.h>
  15 #include "extent_map.h"
  16
  17 /* temporary define until extent_map moves out of btrfs */
  18 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
  19                                        unsigned long extra_flags,
  20                                        void (*ctor)(void *, struct kmem_cache *,
  21                                                     unsigned long));
  22
  23 static struct kmem_cache *extent_map_cache;
  24 static struct kmem_cache *extent_state_cache;
  25 static struct kmem_cache *extent_buffer_cache;
  26
  27 static LIST_HEAD(buffers);
  28 static LIST_HEAD(states);
  29
  30 static spinlock_t state_lock = SPIN_LOCK_UNLOCKED;
  31 #define BUFFER_LRU_MAX 64
  32
  33 struct tree_entry {
  34         u64 start;
  35         u64 end;
  36         int in_tree;
  37         struct rb_node rb_node;
  38 };
  39
  40 struct extent_page_data {
  41         struct bio *bio;
  42         struct extent_map_tree *tree;
  43         get_extent_t *get_extent;
  44 };
  45
  46 void __init extent_map_init(void)
  47 {
  48         extent_map_cache = btrfs_cache_create("extent_map",
  49                                             sizeof(struct extent_map), 0,
  50                                             NULL);
  51         extent_state_cache = btrfs_cache_create("extent_state",
  52                                             sizeof(struct extent_state), 0,
  53                                             NULL);
  54         extent_buffer_cache = btrfs_cache_create("extent_buffers",
  55                                             sizeof(struct extent_buffer), 0,
  56                                             NULL);
  57 }
  58
  59 void __exit extent_map_exit(void)
  60 {
  61         struct extent_state *state;
  62
  63         while (!list_empty(&states)) {
  64                 state = list_entry(states.next, struct extent_state, list);
  65                 printk("state leak: start %Lu end %Lu state %lu in tree %d refs %d\n", state->start, state->end, state->state, state->in_tree, atomic_read(&state->refs));
  66                 list_del(&state->list);
  67                 kmem_cache_free(extent_state_cache, state);
  68
  69         }
  70
  71         if (extent_map_cache)
  72                 kmem_cache_destroy(extent_map_cache);
  73         if (extent_state_cache)
  74                 kmem_cache_destroy(extent_state_cache);
  75         if (extent_buffer_cache)
  76                 kmem_cache_destroy(extent_buffer_cache);
  77 }
  78
  79 void extent_map_tree_init(struct extent_map_tree *tree,
  80                           struct address_space *mapping, gfp_t mask)
  81 {
  82         tree->map.rb_node = NULL;
  83         tree->state.rb_node = NULL;
  84         tree->ops = NULL;
  85         rwlock_init(&tree->lock);
  86         spin_lock_init(&tree->lru_lock);
  87         tree->mapping = mapping;
  88         INIT_LIST_HEAD(&tree->buffer_lru);
  89         tree->lru_size = 0;
  90 }
  91 EXPORT_SYMBOL(extent_map_tree_init);
  92
  93 void extent_map_tree_empty_lru(struct extent_map_tree *tree)
  94 {
  95         struct extent_buffer *eb;
  96         while(!list_empty(&tree->buffer_lru)) {
  97                 eb = list_entry(tree->buffer_lru.next, struct extent_buffer,
  98                                 lru);
  99                 list_del_init(&eb->lru);
 100                 free_extent_buffer(eb);
 101         }
 102 }
 103 EXPORT_SYMBOL(extent_map_tree_empty_lru);
 104
 105 struct extent_map *alloc_extent_map(gfp_t mask)
 106 {
 107         struct extent_map *em;
 108         em = kmem_cache_alloc(extent_map_cache, mask);
 109         if (!em || IS_ERR(em))
 110                 return em;
 111         em->in_tree = 0;
 112         atomic_set(&em->refs, 1);
 113         return em;
 114 }
 115 EXPORT_SYMBOL(alloc_extent_map);
 116
 117 void free_extent_map(struct extent_map *em)
 118 {
 119         if (!em)
 120                 return;
 121         if (atomic_dec_and_test(&em->refs)) {
 122                 WARN_ON(em->in_tree);
 123                 kmem_cache_free(extent_map_cache, em);
 124         }
 125 }
 126 EXPORT_SYMBOL(free_extent_map);
 127
 128
 129 struct extent_state *alloc_extent_state(gfp_t mask)
 130 {
 131         struct extent_state *state;
 132         unsigned long flags;
 133
 134         state = kmem_cache_alloc(extent_state_cache, mask);
 135         if (!state || IS_ERR(state))
 136                 return state;
 137         state->state = 0;
 138         state->in_tree = 0;
 139         state->private = 0;
 140
 141         spin_lock_irqsave(&state_lock, flags);
 142         list_add(&state->list, &states);
 143         spin_unlock_irqrestore(&state_lock, flags);
 144
 145         atomic_set(&state->refs, 1);
 146         init_waitqueue_head(&state->wq);
 147         return state;
 148 }
 149 EXPORT_SYMBOL(alloc_extent_state);
 150
 151 void free_extent_state(struct extent_state *state)
 152 {
 153         unsigned long flags;
 154         if (!state)
 155                 return;
 156         if (atomic_dec_and_test(&state->refs)) {
 157                 WARN_ON(state->in_tree);
 158                 spin_lock_irqsave(&state_lock, flags);
 159                 list_del(&state->list);
 160                 spin_unlock_irqrestore(&state_lock, flags);
 161                 kmem_cache_free(extent_state_cache, state);
 162         }
 163 }
 164 EXPORT_SYMBOL(free_extent_state);
 165
 166 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
 167                                    struct rb_node *node)
 168 {
 169         struct rb_node ** p = &root->rb_node;
 170         struct rb_node * parent = NULL;
 171         struct tree_entry *entry;
 172
 173         while(*p) {
 174                 parent = *p;
 175                 entry = rb_entry(parent, struct tree_entry, rb_node);
 176
 177                 if (offset < entry->start)
 178                         p = &(*p)->rb_left;
 179                 else if (offset > entry->end)
 180                         p = &(*p)->rb_right;
 181                 else
 182                         return parent;
 183         }
 184
 185         entry = rb_entry(node, struct tree_entry, rb_node);
 186         entry->in_tree = 1;
 187         rb_link_node(node, parent, p);
 188         rb_insert_color(node, root);
 189         return NULL;
 190 }
 191
 192 static struct rb_node *__tree_search(struct rb_root *root, u64 offset,
 193                                    struct rb_node **prev_ret)
 194 {
 195         struct rb_node * n = root->rb_node;
 196         struct rb_node *prev = NULL;
 197         struct tree_entry *entry;
 198         struct tree_entry *prev_entry = NULL;
 199
 200         while(n) {
 201                 entry = rb_entry(n, struct tree_entry, rb_node);
 202                 prev = n;
 203                 prev_entry = entry;
 204
 205                 if (offset < entry->start)
 206                         n = n->rb_left;
 207                 else if (offset > entry->end)
 208                         n = n->rb_right;
 209                 else
 210                         return n;
 211         }
 212         if (!prev_ret)
 213                 return NULL;
 214         while(prev && offset > prev_entry->end) {
 215                 prev = rb_next(prev);
 216                 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
 217         }
 218         *prev_ret = prev;
 219         return NULL;
 220 }
 221
 222 static inline struct rb_node *tree_search(struct rb_root *root, u64 offset)
 223 {
 224         struct rb_node *prev;
 225         struct rb_node *ret;
 226         ret = __tree_search(root, offset, &prev);
 227         if (!ret)
 228                 return prev;
 229         return ret;
 230 }
 231
 232 static int tree_delete(struct rb_root *root, u64 offset)
 233 {
 234         struct rb_node *node;
 235         struct tree_entry *entry;
 236
 237         node = __tree_search(root, offset, NULL);
 238         if (!node)
 239                 return -ENOENT;
 240         entry = rb_entry(node, struct tree_entry, rb_node);
 241         entry->in_tree = 0;
 242         rb_erase(node, root);
 243         return 0;
 244 }
 245
 246 /*
 247  * add_extent_mapping tries a simple backward merge with existing
 248  * mappings.  The extent_map struct passed in will be inserted into
 249  * the tree directly (no copies made, just a reference taken).
 250  */
 251 int add_extent_mapping(struct extent_map_tree *tree,
 252                        struct extent_map *em)
 253 {
 254         int ret = 0;
 255         struct extent_map *prev = NULL;
 256         struct rb_node *rb;
 257
 258         write_lock_irq(&tree->lock);
 259         rb = tree_insert(&tree->map, em->end, &em->rb_node);
 260         if (rb) {
 261                 prev = rb_entry(rb, struct extent_map, rb_node);
 262                 printk("found extent map %Lu %Lu on insert of %Lu %Lu\n", prev->start, prev->end, em->start, em->end);
 263                 ret = -EEXIST;
 264                 goto out;
 265         }
 266         atomic_inc(&em->refs);
 267         if (em->start != 0) {
 268                 rb = rb_prev(&em->rb_node);
 269                 if (rb)
 270                         prev = rb_entry(rb, struct extent_map, rb_node);
 271                 if (prev && prev->end + 1 == em->start &&
 272                     ((em->block_start == EXTENT_MAP_HOLE &&
 273                       prev->block_start == EXTENT_MAP_HOLE) ||
 274                      (em->block_start == EXTENT_MAP_INLINE &&
 275                       prev->block_start == EXTENT_MAP_INLINE) ||
 276                      (em->block_start == EXTENT_MAP_DELALLOC &&
 277                       prev->block_start == EXTENT_MAP_DELALLOC) ||
 278                      (em->block_start < EXTENT_MAP_DELALLOC - 1 &&
 279                       em->block_start == prev->block_end + 1))) {
 280                         em->start = prev->start;
 281                         em->block_start = prev->block_start;
 282                         rb_erase(&prev->rb_node, &tree->map);
 283                         prev->in_tree = 0;
 284                         free_extent_map(prev);
 285                 }
 286          }
 287 out:
 288         write_unlock_irq(&tree->lock);
 289         return ret;
 290 }
 291 EXPORT_SYMBOL(add_extent_mapping);
 292
 293 /*
 294  * lookup_extent_mapping returns the first extent_map struct in the
 295  * tree that intersects the [start, end] (inclusive) range.  There may
 296  * be additional objects in the tree that intersect, so check the object
 297  * returned carefully to make sure you don't need additional lookups.
 298  */
 299 struct extent_map *lookup_extent_mapping(struct extent_map_tree *tree,
 300                                          u64 start, u64 end)
 301 {
 302         struct extent_map *em;
 303         struct rb_node *rb_node;
 304
 305         read_lock_irq(&tree->lock);
 306         rb_node = tree_search(&tree->map, start);
 307         if (!rb_node) {
 308                 em = NULL;
 309                 goto out;
 310         }
 311         if (IS_ERR(rb_node)) {
 312                 em = ERR_PTR(PTR_ERR(rb_node));
 313                 goto out;
 314         }
 315         em = rb_entry(rb_node, struct extent_map, rb_node);
 316         if (em->end < start || em->start > end) {
 317                 em = NULL;
 318                 goto out;
 319         }
 320         atomic_inc(&em->refs);
 321 out:
 322         read_unlock_irq(&tree->lock);
 323         return em;
 324 }
 325 EXPORT_SYMBOL(lookup_extent_mapping);
 326
 327 /*
 328  * removes an extent_map struct from the tree.  No reference counts are
 329  * dropped, and no checks are done to  see if the range is in use
 330  */
 331 int remove_extent_mapping(struct extent_map_tree *tree, struct extent_map *em)
 332 {
 333         int ret;
 334
 335         write_lock_irq(&tree->lock);
 336         ret = tree_delete(&tree->map, em->end);
 337         write_unlock_irq(&tree->lock);
 338         return ret;
 339 }
 340 EXPORT_SYMBOL(remove_extent_mapping);
 341
 342 /*
 343  * utility function to look for merge candidates inside a given range.
 344  * Any extents with matching state are merged together into a single
 345  * extent in the tree.  Extents with EXTENT_IO in their state field
 346  * are not merged because the end_io handlers need to be able to do
 347  * operations on them without sleeping (or doing allocations/splits).
 348  *
 349  * This should be called with the tree lock held.
 350  */
 351 static int merge_state(struct extent_map_tree *tree,
 352                        struct extent_state *state)
 353 {
 354         struct extent_state *other;
 355         struct rb_node *other_node;
 356
 357         if (state->state & EXTENT_IOBITS)
 358                 return 0;
 359
 360         other_node = rb_prev(&state->rb_node);
 361         if (other_node) {
 362                 other = rb_entry(other_node, struct extent_state, rb_node);
 363                 if (other->end == state->start - 1 &&
 364                     other->state == state->state) {
 365                         state->start = other->start;
 366                         other->in_tree = 0;
 367                         rb_erase(&other->rb_node, &tree->state);
 368                         free_extent_state(other);
 369                 }
 370         }
 371         other_node = rb_next(&state->rb_node);
 372         if (other_node) {
 373                 other = rb_entry(other_node, struct extent_state, rb_node);
 374                 if (other->start == state->end + 1 &&
 375                     other->state == state->state) {
 376                         other->start = state->start;
 377                         state->in_tree = 0;
 378                         rb_erase(&state->rb_node, &tree->state);
 379                         free_extent_state(state);
 380                 }
 381         }
 382         return 0;
 383 }
 384
 385 /*
 386  * insert an extent_state struct into the tree.  'bits' are set on the
 387  * struct before it is inserted.
 388  *
 389  * This may return -EEXIST if the extent is already there, in which case the
 390  * state struct is freed.
 391  *
 392  * The tree lock is not taken internally.  This is a utility function and
 393  * probably isn't what you want to call (see set/clear_extent_bit).
 394  */
 395 static int insert_state(struct extent_map_tree *tree,
 396                         struct extent_state *state, u64 start, u64 end,
 397                         int bits)
 398 {
 399         struct rb_node *node;
 400
 401         if (end < start) {
 402                 printk("end < start %Lu %Lu\n", end, start);
 403                 WARN_ON(1);
 404         }
 405         state->state |= bits;
 406         state->start = start;
 407         state->end = end;
 408         node = tree_insert(&tree->state, end, &state->rb_node);
 409         if (node) {
 410                 struct extent_state *found;
 411                 found = rb_entry(node, struct extent_state, rb_node);
 412                 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, start, end);
 413                 free_extent_state(state);
 414                 return -EEXIST;
 415         }
 416         merge_state(tree, state);
 417         return 0;
 418 }
 419
 420 /*
 421  * split a given extent state struct in two, inserting the preallocated
 422  * struct 'prealloc' as the newly created second half.  'split' indicates an
 423  * offset inside 'orig' where it should be split.
 424  *
 425  * Before calling,
 426  * the tree has 'orig' at [orig->start, orig->end].  After calling, there
 427  * are two extent state structs in the tree:
 428  * prealloc: [orig->start, split - 1]
 429  * orig: [ split, orig->end ]
 430  *
 431  * The tree locks are not taken by this function. They need to be held
 432  * by the caller.
 433  */
 434 static int split_state(struct extent_map_tree *tree, struct extent_state *orig,
 435                        struct extent_state *prealloc, u64 split)
 436 {
 437         struct rb_node *node;
 438         prealloc->start = orig->start;
 439         prealloc->end = split - 1;
 440         prealloc->state = orig->state;
 441         orig->start = split;
 442
 443         node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
 444         if (node) {
 445                 struct extent_state *found;
 446                 found = rb_entry(node, struct extent_state, rb_node);
 447                 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, prealloc->start, prealloc->end);
 448                 free_extent_state(prealloc);
 449                 return -EEXIST;
 450         }
 451         return 0;
 452 }
 453
 454 /*
 455  * utility function to clear some bits in an extent state struct.
 456  * it will optionally wake up any one waiting on this state (wake == 1), or
 457  * forcibly remove the state from the tree (delete == 1).
 458  *
 459  * If no bits are set on the state struct after clearing things, the
 460  * struct is freed and removed from the tree
 461  */
 462 static int clear_state_bit(struct extent_map_tree *tree,
 463                             struct extent_state *state, int bits, int wake,
 464                             int delete)
 465 {
 466         int ret = state->state & bits;
 467         state->state &= ~bits;
 468         if (wake)
 469                 wake_up(&state->wq);
 470         if (delete || state->state == 0) {
 471                 if (state->in_tree) {
 472                         rb_erase(&state->rb_node, &tree->state);
 473                         state->in_tree = 0;
 474                         free_extent_state(state);
 475                 } else {
 476                         WARN_ON(1);
 477                 }
 478         } else {
 479                 merge_state(tree, state);
 480         }
 481         return ret;
 482 }
 483
 484 /*
 485  * clear some bits on a range in the tree.  This may require splitting
 486  * or inserting elements in the tree, so the gfp mask is used to
 487  * indicate which allocations or sleeping are allowed.
 488  *
 489  * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
 490  * the given range from the tree regardless of state (ie for truncate).
 491  *
 492  * the range [start, end] is inclusive.
 493  *
 494  * This takes the tree lock, and returns < 0 on error, > 0 if any of the
 495  * bits were already set, or zero if none of the bits were already set.
 496  */
 497 int clear_extent_bit(struct extent_map_tree *tree, u64 start, u64 end,
 498                      int bits, int wake, int delete, gfp_t mask)
 499 {
 500         struct extent_state *state;
 501         struct extent_state *prealloc = NULL;
 502         struct rb_node *node;
 503         unsigned long flags;
 504         int err;
 505         int set = 0;
 506
 507 again:
 508         if (!prealloc && (mask & __GFP_WAIT)) {
 509                 prealloc = alloc_extent_state(mask);
 510                 if (!prealloc)
 511                         return -ENOMEM;
 512         }
 513
 514         write_lock_irqsave(&tree->lock, flags);
 515         /*
 516          * this search will find the extents that end after
 517          * our range starts
 518          */
 519         node = tree_search(&tree->state, start);
 520         if (!node)
 521                 goto out;
 522         state = rb_entry(node, struct extent_state, rb_node);
 523         if (state->start > end)
 524                 goto out;
 525         WARN_ON(state->end < start);
 526
 527         /*
 528          *     | ---- desired range ---- |
 529          *  | state | or
 530          *  | ------------- state -------------- |
 531          *
 532          * We need to split the extent we found, and may flip
 533          * bits on second half.
 534          *
 535          * If the extent we found extends past our range, we
 536          * just split and search again.  It'll get split again
 537          * the next time though.
 538          *
 539          * If the extent we found is inside our range, we clear
 540          * the desired bit on it.
 541          */
 542
 543         if (state->start < start) {
 544                 err = split_state(tree, state, prealloc, start);
 545                 BUG_ON(err == -EEXIST);
 546                 prealloc = NULL;
 547                 if (err)
 548                         goto out;
 549                 if (state->end <= end) {
 550                         start = state->end + 1;
 551                         set |= clear_state_bit(tree, state, bits,
 552                                         wake, delete);
 553                 } else {
 554                         start = state->start;
 555                 }
 556                 goto search_again;
 557         }
 558         /*
 559          * | ---- desired range ---- |
 560          *                        | state |
 561          * We need to split the extent, and clear the bit
 562          * on the first half
 563          */
 564         if (state->start <= end && state->end > end) {
 565                 err = split_state(tree, state, prealloc, end + 1);
 566                 BUG_ON(err == -EEXIST);
 567
 568                 if (wake)
 569                         wake_up(&state->wq);
 570                 set |= clear_state_bit(tree, prealloc, bits,
 571                                        wake, delete);
 572                 prealloc = NULL;
 573                 goto out;
 574         }
 575
 576         start = state->end + 1;
 577         set |= clear_state_bit(tree, state, bits, wake, delete);
 578         goto search_again;
 579
 580 out:
 581         write_unlock_irqrestore(&tree->lock, flags);
 582         if (prealloc)
 583                 free_extent_state(prealloc);
 584
 585         return set;
 586
 587 search_again:
 588         if (start > end)
 589                 goto out;
 590         write_unlock_irqrestore(&tree->lock, flags);
 591         if (mask & __GFP_WAIT)
 592                 cond_resched();
 593         goto again;
 594 }
 595 EXPORT_SYMBOL(clear_extent_bit);
 596
 597 static int wait_on_state(struct extent_map_tree *tree,
 598                          struct extent_state *state)
 599 {
 600         DEFINE_WAIT(wait);
 601         prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
 602         read_unlock_irq(&tree->lock);
 603         schedule();
 604         read_lock_irq(&tree->lock);
 605         finish_wait(&state->wq, &wait);
 606         return 0;
 607 }
 608
 609 /*
 610  * waits for one or more bits to clear on a range in the state tree.
 611  * The range [start, end] is inclusive.
 612  * The tree lock is taken by this function
 613  */
 614 int wait_extent_bit(struct extent_map_tree *tree, u64 start, u64 end, int bits)
 615 {
 616         struct extent_state *state;
 617         struct rb_node *node;
 618
 619         read_lock_irq(&tree->lock);
 620 again:
 621         while (1) {
 622                 /*
 623                  * this search will find all the extents that end after
 624                  * our range starts
 625                  */
 626                 node = tree_search(&tree->state, start);
 627                 if (!node)
 628                         break;
 629
 630                 state = rb_entry(node, struct extent_state, rb_node);
 631
 632                 if (state->start > end)
 633                         goto out;
 634
 635                 if (state->state & bits) {
 636                         start = state->start;
 637                         atomic_inc(&state->refs);
 638                         wait_on_state(tree, state);
 639                         free_extent_state(state);
 640                         goto again;
 641                 }
 642                 start = state->end + 1;
 643
 644                 if (start > end)
 645                         break;
 646
 647                 if (need_resched()) {
 648                         read_unlock_irq(&tree->lock);
 649                         cond_resched();
 650                         read_lock_irq(&tree->lock);
 651                 }
 652         }
 653 out:
 654         read_unlock_irq(&tree->lock);
 655         return 0;
 656 }
 657 EXPORT_SYMBOL(wait_extent_bit);
 658
 659 /*
 660  * set some bits on a range in the tree.  This may require allocations
 661  * or sleeping, so the gfp mask is used to indicate what is allowed.
 662  *
 663  * If 'exclusive' == 1, this will fail with -EEXIST if some part of the
 664  * range already has the desired bits set.  The start of the existing
 665  * range is returned in failed_start in this case.
 666  *
 667  * [start, end] is inclusive
 668  * This takes the tree lock.
 669  */
 670 int set_extent_bit(struct extent_map_tree *tree, u64 start, u64 end, int bits,
 671                    int exclusive, u64 *failed_start, gfp_t mask)
 672 {
 673         struct extent_state *state;
 674         struct extent_state *prealloc = NULL;
 675         struct rb_node *node;
 676         unsigned long flags;
 677         int err = 0;
 678         int set;
 679         u64 last_start;
 680         u64 last_end;
 681 again:
 682         if (!prealloc && (mask & __GFP_WAIT)) {
 683                 prealloc = alloc_extent_state(mask);
 684                 if (!prealloc)
 685                         return -ENOMEM;
 686         }
 687
 688         write_lock_irqsave(&tree->lock, flags);
 689         /*
 690          * this search will find all the extents that end after
 691          * our range starts.
 692          */
 693         node = tree_search(&tree->state, start);
 694         if (!node) {
 695                 err = insert_state(tree, prealloc, start, end, bits);
 696                 prealloc = NULL;
 697                 BUG_ON(err == -EEXIST);
 698                 goto out;
 699         }
 700
 701         state = rb_entry(node, struct extent_state, rb_node);
 702         last_start = state->start;
 703         last_end = state->end;
 704
 705         /*
 706          * | ---- desired range ---- |
 707          * | state |
 708          *
 709          * Just lock what we found and keep going
 710          */
 711         if (state->start == start && state->end <= end) {
 712                 set = state->state & bits;
 713                 if (set && exclusive) {
 714                         *failed_start = state->start;
 715                         err = -EEXIST;
 716                         goto out;
 717                 }
 718                 state->state |= bits;
 719                 start = state->end + 1;
 720                 merge_state(tree, state);
 721                 goto search_again;
 722         }
 723
 724         /*
 725          *     | ---- desired range ---- |
 726          * | state |
 727          *   or
 728          * | ------------- state -------------- |
 729          *
 730          * We need to split the extent we found, and may flip bits on
 731          * second half.
 732          *
 733          * If the extent we found extends past our
 734          * range, we just split and search again.  It'll get split
 735          * again the next time though.
 736          *
 737          * If the extent we found is inside our range, we set the
 738          * desired bit on it.
 739          */
 740         if (state->start < start) {
 741                 set = state->state & bits;
 742                 if (exclusive && set) {
 743                         *failed_start = start;
 744                         err = -EEXIST;
 745                         goto out;
 746                 }
 747                 err = split_state(tree, state, prealloc, start);
 748                 BUG_ON(err == -EEXIST);
 749                 prealloc = NULL;
 750                 if (err)
 751                         goto out;
 752                 if (state->end <= end) {
 753                         state->state |= bits;
 754                         start = state->end + 1;
 755                         merge_state(tree, state);
 756                 } else {
 757                         start = state->start;
 758                 }
 759                 goto search_again;
 760         }
 761         /*
 762          * | ---- desired range ---- |
 763          *     | state | or               | state |
 764          *
 765          * There's a hole, we need to insert something in it and
 766          * ignore the extent we found.
 767          */
 768         if (state->start > start) {
 769                 u64 this_end;
 770                 if (end < last_start)
 771                         this_end = end;
 772                 else
 773                         this_end = last_start -1;
 774                 err = insert_state(tree, prealloc, start, this_end,
 775                                    bits);
 776                 prealloc = NULL;
 777                 BUG_ON(err == -EEXIST);
 778                 if (err)
 779                         goto out;
 780                 start = this_end + 1;
 781                 goto search_again;
 782         }
 783         /*
 784          * | ---- desired range ---- |
 785          *                        | state |
 786          * We need to split the extent, and set the bit
 787          * on the first half
 788          */
 789         if (state->start <= end && state->end > end) {
 790                 set = state->state & bits;
 791                 if (exclusive && set) {
 792                         *failed_start = start;
 793                         err = -EEXIST;
 794                         goto out;
 795                 }
 796                 err = split_state(tree, state, prealloc, end + 1);
 797                 BUG_ON(err == -EEXIST);
 798
 799                 prealloc->state |= bits;
 800                 merge_state(tree, prealloc);
 801                 prealloc = NULL;
 802                 goto out;
 803         }
 804
 805         goto search_again;
 806
 807 out:
 808         write_unlock_irqrestore(&tree->lock, flags);
 809         if (prealloc)
 810                 free_extent_state(prealloc);
 811
 812         return err;
 813
 814 search_again:
 815         if (start > end)
 816                 goto out;
 817         write_unlock_irqrestore(&tree->lock, flags);
 818         if (mask & __GFP_WAIT)
 819                 cond_resched();
 820         goto again;
 821 }
 822 EXPORT_SYMBOL(set_extent_bit);
 823
 824 /* wrappers around set/clear extent bit */
 825 int set_extent_dirty(struct extent_map_tree *tree, u64 start, u64 end,
 826                      gfp_t mask)
 827 {
 828         return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL,
 829                               mask);
 830 }
 831 EXPORT_SYMBOL(set_extent_dirty);
 832
 833 int set_extent_bits(struct extent_map_tree *tree, u64 start, u64 end,
 834                     int bits, gfp_t mask)
 835 {
 836         return set_extent_bit(tree, start, end, bits, 0, NULL,
 837                               mask);
 838 }
 839 EXPORT_SYMBOL(set_extent_bits);
 840
 841 int clear_extent_bits(struct extent_map_tree *tree, u64 start, u64 end,
 842                       int bits, gfp_t mask)
 843 {
 844         return clear_extent_bit(tree, start, end, bits, 0, 0, mask);
 845 }
 846 EXPORT_SYMBOL(clear_extent_bits);
 847
 848 int set_extent_delalloc(struct extent_map_tree *tree, u64 start, u64 end,
 849                      gfp_t mask)
 850 {
 851         return set_extent_bit(tree, start, end,
 852                               EXTENT_DELALLOC | EXTENT_DIRTY, 0, NULL,
 853                               mask);
 854 }
 855 EXPORT_SYMBOL(set_extent_delalloc);
 856
 857 int clear_extent_dirty(struct extent_map_tree *tree, u64 start, u64 end,
 858                        gfp_t mask)
 859 {
 860         return clear_extent_bit(tree, start, end,
 861                                 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, mask);
 862 }
 863 EXPORT_SYMBOL(clear_extent_dirty);
 864
 865 int set_extent_new(struct extent_map_tree *tree, u64 start, u64 end,
 866                      gfp_t mask)
 867 {
 868         return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL,
 869                               mask);
 870 }
 871 EXPORT_SYMBOL(set_extent_new);
 872
 873 int clear_extent_new(struct extent_map_tree *tree, u64 start, u64 end,
 874                        gfp_t mask)
 875 {
 876         return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0, mask);
 877 }
 878 EXPORT_SYMBOL(clear_extent_new);
 879
 880 int set_extent_uptodate(struct extent_map_tree *tree, u64 start, u64 end,
 881                         gfp_t mask)
 882 {
 883         return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, NULL,
 884                               mask);
 885 }
 886 EXPORT_SYMBOL(set_extent_uptodate);
 887
 888 int clear_extent_uptodate(struct extent_map_tree *tree, u64 start, u64 end,
 889                           gfp_t mask)
 890 {
 891         return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, mask);
 892 }
 893 EXPORT_SYMBOL(clear_extent_uptodate);
 894
 895 int set_extent_writeback(struct extent_map_tree *tree, u64 start, u64 end,
 896                          gfp_t mask)
 897 {
 898         return set_extent_bit(tree, start, end, EXTENT_WRITEBACK,
 899                               0, NULL, mask);
 900 }
 901 EXPORT_SYMBOL(set_extent_writeback);
 902
 903 int clear_extent_writeback(struct extent_map_tree *tree, u64 start, u64 end,
 904                            gfp_t mask)
 905 {
 906         return clear_extent_bit(tree, start, end, EXTENT_WRITEBACK, 1, 0, mask);
 907 }
 908 EXPORT_SYMBOL(clear_extent_writeback);
 909
 910 int wait_on_extent_writeback(struct extent_map_tree *tree, u64 start, u64 end)
 911 {
 912         return wait_extent_bit(tree, start, end, EXTENT_WRITEBACK);
 913 }
 914 EXPORT_SYMBOL(wait_on_extent_writeback);
 915
 916 /*
 917  * locks a range in ascending order, waiting for any locked regions
 918  * it hits on the way.  [start,end] are inclusive, and this will sleep.
 919  */
 920 int lock_extent(struct extent_map_tree *tree, u64 start, u64 end, gfp_t mask)
 921 {
 922         int err;
 923         u64 failed_start;
 924         while (1) {
 925                 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1,
 926                                      &failed_start, mask);
 927                 if (err == -EEXIST && (mask & __GFP_WAIT)) {
 928                         wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
 929                         start = failed_start;
 930                 } else {
 931                         break;
 932                 }
 933                 WARN_ON(start > end);
 934         }
 935         return err;
 936 }
 937 EXPORT_SYMBOL(lock_extent);
 938
 939 int unlock_extent(struct extent_map_tree *tree, u64 start, u64 end,
 940                   gfp_t mask)
 941 {
 942         return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, mask);
 943 }
 944 EXPORT_SYMBOL(unlock_extent);
 945
 946 /*
 947  * helper function to set pages and extents in the tree dirty
 948  */
 949 int set_range_dirty(struct extent_map_tree *tree, u64 start, u64 end)
 950 {
 951         unsigned long index = start >> PAGE_CACHE_SHIFT;
 952         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
 953         struct page *page;
 954
 955         while (index <= end_index) {
 956                 page = find_get_page(tree->mapping, index);
 957                 BUG_ON(!page);
 958                 __set_page_dirty_nobuffers(page);
 959                 page_cache_release(page);
 960                 index++;
 961         }
 962         set_extent_dirty(tree, start, end, GFP_NOFS);
 963         return 0;
 964 }
 965 EXPORT_SYMBOL(set_range_dirty);
 966
 967 /*
 968  * helper function to set both pages and extents in the tree writeback
 969  */
 970 int set_range_writeback(struct extent_map_tree *tree, u64 start, u64 end)
 971 {
 972         unsigned long index = start >> PAGE_CACHE_SHIFT;
 973         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
 974         struct page *page;
 975
 976         while (index <= end_index) {
 977                 page = find_get_page(tree->mapping, index);
 978                 BUG_ON(!page);
 979                 set_page_writeback(page);
 980                 page_cache_release(page);
 981                 index++;
 982         }
 983         set_extent_writeback(tree, start, end, GFP_NOFS);
 984         return 0;
 985 }
 986 EXPORT_SYMBOL(set_range_writeback);
 987
 988 int find_first_extent_bit(struct extent_map_tree *tree, u64 start,
 989                           u64 *start_ret, u64 *end_ret, int bits)
 990 {
 991         struct rb_node *node;
 992         struct extent_state *state;
 993         int ret = 1;
 994
 995         read_lock_irq(&tree->lock);
 996         /*
 997          * this search will find all the extents that end after
 998          * our range starts.
 999          */
1000         node = tree_search(&tree->state, start);
1001         if (!node || IS_ERR(node)) {
1002                 goto out;
1003         }
1004
1005         while(1) {
1006                 state = rb_entry(node, struct extent_state, rb_node);
1007                 if (state->end >= start && (state->state & bits)) {
1008                         *start_ret = state->start;
1009                         *end_ret = state->end;
1010                         ret = 0;
1011                         break;
1012                 }
1013                 node = rb_next(node);
1014                 if (!node)
1015                         break;
1016         }
1017 out:
1018         read_unlock_irq(&tree->lock);
1019         return ret;
1020 }
1021 EXPORT_SYMBOL(find_first_extent_bit);
1022
1023 u64 find_lock_delalloc_range(struct extent_map_tree *tree,
1024                              u64 start, u64 lock_start, u64 *end, u64 max_bytes)
1025 {
1026         struct rb_node *node;
1027         struct extent_state *state;
1028         u64 cur_start = start;
1029         u64 found = 0;
1030         u64 total_bytes = 0;
1031
1032         write_lock_irq(&tree->lock);
1033         /*
1034          * this search will find all the extents that end after
1035          * our range starts.
1036          */
1037 search_again:
1038         node = tree_search(&tree->state, cur_start);
1039         if (!node || IS_ERR(node)) {
1040                 goto out;
1041         }
1042
1043         while(1) {
1044                 state = rb_entry(node, struct extent_state, rb_node);
1045                 if (state->start != cur_start) {
1046                         goto out;
1047                 }
1048                 if (!(state->state & EXTENT_DELALLOC)) {
1049                         goto out;
1050                 }
1051                 if (state->start >= lock_start) {
1052                         if (state->state & EXTENT_LOCKED) {
1053                                 DEFINE_WAIT(wait);
1054                                 atomic_inc(&state->refs);
1055                                 prepare_to_wait(&state->wq, &wait,
1056                                                 TASK_UNINTERRUPTIBLE);
1057                                 write_unlock_irq(&tree->lock);
1058                                 schedule();
1059                                 write_lock_irq(&tree->lock);
1060                                 finish_wait(&state->wq, &wait);
1061                                 free_extent_state(state);
1062                                 goto search_again;
1063                         }
1064                         state->state |= EXTENT_LOCKED;
1065                 }
1066                 found++;
1067                 *end = state->end;
1068                 cur_start = state->end + 1;
1069                 node = rb_next(node);
1070                 if (!node)
1071                         break;
1072                 total_bytes += state->end - state->start + 1;
1073                 if (total_bytes >= max_bytes)
1074                         break;
1075         }
1076 out:
1077         write_unlock_irq(&tree->lock);
1078         return found;
1079 }
1080
1081 /*
1082  * helper function to lock both pages and extents in the tree.
1083  * pages must be locked first.
1084  */
1085 int lock_range(struct extent_map_tree *tree, u64 start, u64 end)
1086 {
1087         unsigned long index = start >> PAGE_CACHE_SHIFT;
1088         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1089         struct page *page;
1090         int err;
1091
1092         while (index <= end_index) {
1093                 page = grab_cache_page(tree->mapping, index);
1094                 if (!page) {
1095                         err = -ENOMEM;
1096                         goto failed;
1097                 }
1098                 if (IS_ERR(page)) {
1099                         err = PTR_ERR(page);
1100                         goto failed;
1101                 }
1102                 index++;
1103         }
1104         lock_extent(tree, start, end, GFP_NOFS);
1105         return 0;
1106
1107 failed:
1108         /*
1109          * we failed above in getting the page at 'index', so we undo here
1110          * up to but not including the page at 'index'
1111          */
1112         end_index = index;
1113         index = start >> PAGE_CACHE_SHIFT;
1114         while (index < end_index) {
1115                 page = find_get_page(tree->mapping, index);
1116                 unlock_page(page);
1117                 page_cache_release(page);
1118                 index++;
1119         }
1120         return err;
1121 }
1122 EXPORT_SYMBOL(lock_range);
1123
1124 /*
1125  * helper function to unlock both pages and extents in the tree.
1126  */
1127 int unlock_range(struct extent_map_tree *tree, u64 start, u64 end)
1128 {
1129         unsigned long index = start >> PAGE_CACHE_SHIFT;
1130         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1131         struct page *page;
1132
1133         while (index <= end_index) {
1134                 page = find_get_page(tree->mapping, index);
1135                 unlock_page(page);
1136                 page_cache_release(page);
1137                 index++;
1138         }
1139         unlock_extent(tree, start, end, GFP_NOFS);
1140         return 0;
1141 }
1142 EXPORT_SYMBOL(unlock_range);
1143
1144 int set_state_private(struct extent_map_tree *tree, u64 start, u64 private)
1145 {
1146         struct rb_node *node;
1147         struct extent_state *state;
1148         int ret = 0;
1149
1150         write_lock_irq(&tree->lock);
1151         /*
1152          * this search will find all the extents that end after
1153          * our range starts.
1154          */
1155         node = tree_search(&tree->state, start);
1156         if (!node || IS_ERR(node)) {
1157                 ret = -ENOENT;
1158                 goto out;
1159         }
1160         state = rb_entry(node, struct extent_state, rb_node);
1161         if (state->start != start) {
1162                 ret = -ENOENT;
1163                 goto out;
1164         }
1165         state->private = private;
1166 out:
1167         write_unlock_irq(&tree->lock);
1168         return ret;
1169 }
1170
1171 int get_state_private(struct extent_map_tree *tree, u64 start, u64 *private)
1172 {
1173         struct rb_node *node;
1174         struct extent_state *state;
1175         int ret = 0;
1176
1177         read_lock_irq(&tree->lock);
1178         /*
1179          * this search will find all the extents that end after
1180          * our range starts.
1181          */
1182         node = tree_search(&tree->state, start);
1183         if (!node || IS_ERR(node)) {
1184                 ret = -ENOENT;
1185                 goto out;
1186         }
1187         state = rb_entry(node, struct extent_state, rb_node);
1188         if (state->start != start) {
1189                 ret = -ENOENT;
1190                 goto out;
1191         }
1192         *private = state->private;
1193 out:
1194         read_unlock_irq(&tree->lock);
1195         return ret;
1196 }
1197
1198 /*
1199  * searches a range in the state tree for a given mask.
1200  * If 'filled' == 1, this returns 1 only if ever extent in the tree
1201  * has the bits set.  Otherwise, 1 is returned if any bit in the
1202  * range is found set.
1203  */
1204 int test_range_bit(struct extent_map_tree *tree, u64 start, u64 end,
1205                    int bits, int filled)
1206 {
1207         struct extent_state *state = NULL;
1208         struct rb_node *node;
1209         int bitset = 0;
1210
1211         read_lock_irq(&tree->lock);
1212         node = tree_search(&tree->state, start);
1213         while (node && start <= end) {
1214                 state = rb_entry(node, struct extent_state, rb_node);
1215
1216                 if (filled && state->start > start) {
1217                         bitset = 0;
1218                         break;
1219                 }
1220
1221                 if (state->start > end)
1222                         break;
1223
1224                 if (state->state & bits) {
1225                         bitset = 1;
1226                         if (!filled)
1227                                 break;
1228                 } else if (filled) {
1229                         bitset = 0;
1230                         break;
1231                 }
1232                 start = state->end + 1;
1233                 if (start > end)
1234                         break;
1235                 node = rb_next(node);
1236         }
1237         read_unlock_irq(&tree->lock);
1238         return bitset;
1239 }
1240 EXPORT_SYMBOL(test_range_bit);
1241
1242 /*
1243  * helper function to set a given page up to date if all the
1244  * extents in the tree for that page are up to date
1245  */
1246 static int check_page_uptodate(struct extent_map_tree *tree,
1247                                struct page *page)
1248 {
1249         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1250         u64 end = start + PAGE_CACHE_SIZE - 1;
1251         if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1))
1252                 SetPageUptodate(page);
1253         return 0;
1254 }
1255
1256 /*
1257  * helper function to unlock a page if all the extents in the tree
1258  * for that page are unlocked
1259  */
1260 static int check_page_locked(struct extent_map_tree *tree,
1261                              struct page *page)
1262 {
1263         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1264         u64 end = start + PAGE_CACHE_SIZE - 1;
1265         if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0))
1266                 unlock_page(page);
1267         return 0;
1268 }
1269
1270 /*
1271  * helper function to end page writeback if all the extents
1272  * in the tree for that page are done with writeback
1273  */
1274 static int check_page_writeback(struct extent_map_tree *tree,
1275                              struct page *page)
1276 {
1277         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1278         u64 end = start + PAGE_CACHE_SIZE - 1;
1279         if (!test_range_bit(tree, start, end, EXTENT_WRITEBACK, 0))
1280                 end_page_writeback(page);
1281         return 0;
1282 }
1283
1284 /* lots and lots of room for performance fixes in the end_bio funcs */
1285
1286 /*
1287  * after a writepage IO is done, we need to:
1288  * clear the uptodate bits on error
1289  * clear the writeback bits in the extent tree for this IO
1290  * end_page_writeback if the page has no more pending IO
1291  *
1292  * Scheduling is not allowed, so the extent state tree is expected
1293  * to have one and only one object corresponding to this IO.
1294  */
1295 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1296 static void end_bio_extent_writepage(struct bio *bio, int err)
1297 #else
1298 static int end_bio_extent_writepage(struct bio *bio,
1299                                    unsigned int bytes_done, int err)
1300 #endif
1301 {
1302         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1303         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1304         struct extent_map_tree *tree = bio->bi_private;
1305         u64 start;
1306         u64 end;
1307         int whole_page;
1308
1309 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1310         if (bio->bi_size)
1311                 return 1;
1312 #endif
1313
1314         do {
1315                 struct page *page = bvec->bv_page;
1316                 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1317                          bvec->bv_offset;
1318                 end = start + bvec->bv_len - 1;
1319
1320                 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1321                         whole_page = 1;
1322                 else
1323                         whole_page = 0;
1324
1325                 if (--bvec >= bio->bi_io_vec)
1326                         prefetchw(&bvec->bv_page->flags);
1327
1328                 if (!uptodate) {
1329                         clear_extent_uptodate(tree, start, end, GFP_ATOMIC);
1330                         ClearPageUptodate(page);
1331                         SetPageError(page);
1332                 }
1333                 clear_extent_writeback(tree, start, end, GFP_ATOMIC);
1334
1335                 if (whole_page)
1336                         end_page_writeback(page);
1337                 else
1338                         check_page_writeback(tree, page);
1339                 if (tree->ops && tree->ops->writepage_end_io_hook)
1340                         tree->ops->writepage_end_io_hook(page, start, end);
1341         } while (bvec >= bio->bi_io_vec);
1342
1343         bio_put(bio);
1344 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1345         return 0;
1346 #endif
1347 }
1348
1349 /*
1350  * after a readpage IO is done, we need to:
1351  * clear the uptodate bits on error
1352  * set the uptodate bits if things worked
1353  * set the page up to date if all extents in the tree are uptodate
1354  * clear the lock bit in the extent tree
1355  * unlock the page if there are no other extents locked for it
1356  *
1357  * Scheduling is not allowed, so the extent state tree is expected
1358  * to have one and only one object corresponding to this IO.
1359  */
1360 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1361 static void end_bio_extent_readpage(struct bio *bio, int err)
1362 #else
1363 static int end_bio_extent_readpage(struct bio *bio,
1364                                    unsigned int bytes_done, int err)
1365 #endif
1366 {
1367         int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1368         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1369         struct extent_map_tree *tree = bio->bi_private;
1370         u64 start;
1371         u64 end;
1372         int whole_page;
1373         int ret;
1374
1375 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1376         if (bio->bi_size)
1377                 return 1;
1378 #endif
1379
1380         do {
1381                 struct page *page = bvec->bv_page;
1382                 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1383                         bvec->bv_offset;
1384                 end = start + bvec->bv_len - 1;
1385
1386                 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1387                         whole_page = 1;
1388                 else
1389                         whole_page = 0;
1390
1391                 if (--bvec >= bio->bi_io_vec)
1392                         prefetchw(&bvec->bv_page->flags);
1393
1394                 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
1395                         ret = tree->ops->readpage_end_io_hook(page, start, end);
1396                         if (ret)
1397                                 uptodate = 0;
1398                 }
1399                 if (uptodate) {
1400                         set_extent_uptodate(tree, start, end, GFP_ATOMIC);
1401                         if (whole_page)
1402                                 SetPageUptodate(page);
1403                         else
1404                                 check_page_uptodate(tree, page);
1405                 } else {
1406                         ClearPageUptodate(page);
1407                         SetPageError(page);
1408                 }
1409
1410                 unlock_extent(tree, start, end, GFP_ATOMIC);
1411
1412                 if (whole_page)
1413                         unlock_page(page);
1414                 else
1415                         check_page_locked(tree, page);
1416         } while (bvec >= bio->bi_io_vec);
1417
1418         bio_put(bio);
1419 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1420         return 0;
1421 #endif
1422 }
1423
1424 /*
1425  * IO done from prepare_write is pretty simple, we just unlock
1426  * the structs in the extent tree when done, and set the uptodate bits
1427  * as appropriate.
1428  */
1429 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1430 static void end_bio_extent_preparewrite(struct bio *bio, int err)
1431 #else
1432 static int end_bio_extent_preparewrite(struct bio *bio,
1433                                        unsigned int bytes_done, int err)
1434 #endif
1435 {
1436         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1437         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1438         struct extent_map_tree *tree = bio->bi_private;
1439         u64 start;
1440         u64 end;
1441
1442 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1443         if (bio->bi_size)
1444                 return 1;
1445 #endif
1446
1447         do {
1448                 struct page *page = bvec->bv_page;
1449                 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1450                         bvec->bv_offset;
1451                 end = start + bvec->bv_len - 1;
1452
1453                 if (--bvec >= bio->bi_io_vec)
1454                         prefetchw(&bvec->bv_page->flags);
1455
1456                 if (uptodate) {
1457                         set_extent_uptodate(tree, start, end, GFP_ATOMIC);
1458                 } else {
1459                         ClearPageUptodate(page);
1460                         SetPageError(page);
1461                 }
1462
1463                 unlock_extent(tree, start, end, GFP_ATOMIC);
1464
1465         } while (bvec >= bio->bi_io_vec);
1466
1467         bio_put(bio);
1468 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1469         return 0;
1470 #endif
1471 }
1472
1473 static struct bio *
1474 extent_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
1475                  gfp_t gfp_flags)
1476 {
1477         struct bio *bio;
1478
1479         bio = bio_alloc(gfp_flags, nr_vecs);
1480
1481         if (bio == NULL && (current->flags & PF_MEMALLOC)) {
1482                 while (!bio && (nr_vecs /= 2))
1483                         bio = bio_alloc(gfp_flags, nr_vecs);
1484         }
1485
1486         if (bio) {
1487                 bio->bi_bdev = bdev;
1488                 bio->bi_sector = first_sector;
1489         }
1490         return bio;
1491 }
1492
1493 static int submit_one_bio(int rw, struct bio *bio)
1494 {
1495         int ret = 0;
1496         bio_get(bio);
1497         submit_bio(rw, bio);
1498         if (bio_flagged(bio, BIO_EOPNOTSUPP))
1499                 ret = -EOPNOTSUPP;
1500         bio_put(bio);
1501         return ret;
1502 }
1503
1504 static int submit_extent_page(int rw, struct extent_map_tree *tree,
1505                               struct page *page, sector_t sector,
1506                               size_t size, unsigned long offset,
1507                               struct block_device *bdev,
1508                               struct bio **bio_ret,
1509                               unsigned long max_pages,
1510                               bio_end_io_t end_io_func)
1511 {
1512         int ret = 0;
1513         struct bio *bio;
1514         int nr;
1515
1516         if (bio_ret && *bio_ret) {
1517                 bio = *bio_ret;
1518                 if (bio->bi_sector + (bio->bi_size >> 9) != sector ||
1519                     bio_add_page(bio, page, size, offset) < size) {
1520                         ret = submit_one_bio(rw, bio);
1521                         bio = NULL;
1522                 } else {
1523                         return 0;
1524                 }
1525         }
1526         nr = min_t(int, max_pages, bio_get_nr_vecs(bdev));
1527         bio = extent_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
1528         if (!bio) {
1529                 printk("failed to allocate bio nr %d\n", nr);
1530         }
1531         bio_add_page(bio, page, size, offset);
1532         bio->bi_end_io = end_io_func;
1533         bio->bi_private = tree;
1534         if (bio_ret) {
1535                 *bio_ret = bio;
1536         } else {
1537                 ret = submit_one_bio(rw, bio);
1538         }
1539
1540         return ret;
1541 }
1542
1543 void set_page_extent_mapped(struct page *page)
1544 {
1545         if (!PagePrivate(page)) {
1546                 SetPagePrivate(page);
1547                 WARN_ON(!page->mapping->a_ops->invalidatepage);
1548                 set_page_private(page, EXTENT_PAGE_PRIVATE);
1549                 page_cache_get(page);
1550         }
1551 }
1552
1553 /*
1554  * basic readpage implementation.  Locked extent state structs are inserted
1555  * into the tree that are removed when the IO is done (by the end_io
1556  * handlers)
1557  */
1558 static int __extent_read_full_page(struct extent_map_tree *tree,
1559                                    struct page *page,
1560                                    get_extent_t *get_extent,
1561                                    struct bio **bio)
1562 {
1563         struct inode *inode = page->mapping->host;
1564         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1565         u64 page_end = start + PAGE_CACHE_SIZE - 1;
1566         u64 end;
1567         u64 cur = start;
1568         u64 extent_offset;
1569         u64 last_byte = i_size_read(inode);
1570         u64 block_start;
1571         u64 cur_end;
1572         sector_t sector;
1573         struct extent_map *em;
1574         struct block_device *bdev;
1575         int ret;
1576         int nr = 0;
1577         size_t page_offset = 0;
1578         size_t iosize;
1579         size_t blocksize = inode->i_sb->s_blocksize;
1580
1581         set_page_extent_mapped(page);
1582
1583         end = page_end;
1584         lock_extent(tree, start, end, GFP_NOFS);
1585
1586         while (cur <= end) {
1587                 if (cur >= last_byte) {
1588                         iosize = PAGE_CACHE_SIZE - page_offset;
1589                         zero_user_page(page, page_offset, iosize, KM_USER0);
1590                         set_extent_uptodate(tree, cur, cur + iosize - 1,
1591                                             GFP_NOFS);
1592                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1593                         break;
1594                 }
1595                 em = get_extent(inode, page, page_offset, cur, end, 0);
1596                 if (IS_ERR(em) || !em) {
1597                         SetPageError(page);
1598                         unlock_extent(tree, cur, end, GFP_NOFS);
1599                         break;
1600                 }
1601
1602                 extent_offset = cur - em->start;
1603                 BUG_ON(em->end < cur);
1604                 BUG_ON(end < cur);
1605
1606                 iosize = min(em->end - cur, end - cur) + 1;
1607                 cur_end = min(em->end, end);
1608                 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
1609                 sector = (em->block_start + extent_offset) >> 9;
1610                 bdev = em->bdev;
1611                 block_start = em->block_start;
1612                 free_extent_map(em);
1613                 em = NULL;
1614
1615                 /* we've found a hole, just zero and go on */
1616                 if (block_start == EXTENT_MAP_HOLE) {
1617                         zero_user_page(page, page_offset, iosize, KM_USER0);
1618                         set_extent_uptodate(tree, cur, cur + iosize - 1,
1619                                             GFP_NOFS);
1620                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1621                         cur = cur + iosize;
1622                         page_offset += iosize;
1623                         continue;
1624                 }
1625                 /* the get_extent function already copied into the page */
1626                 if (test_range_bit(tree, cur, cur_end, EXTENT_UPTODATE, 1)) {
1627                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1628                         cur = cur + iosize;
1629                         page_offset += iosize;
1630                         continue;
1631                 }
1632
1633                 ret = 0;
1634                 if (tree->ops && tree->ops->readpage_io_hook) {
1635                         ret = tree->ops->readpage_io_hook(page, cur,
1636                                                           cur + iosize - 1);
1637                 }
1638                 if (!ret) {
1639                         unsigned long nr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
1640                         nr -= page->index;
1641                         ret = submit_extent_page(READ, tree, page,
1642                                          sector, iosize, page_offset,
1643                                          bdev, bio, nr,
1644                                          end_bio_extent_readpage);
1645                 }
1646                 if (ret)
1647                         SetPageError(page);
1648                 cur = cur + iosize;
1649                 page_offset += iosize;
1650                 nr++;
1651         }
1652         if (!nr) {
1653                 if (!PageError(page))
1654                         SetPageUptodate(page);
1655                 unlock_page(page);
1656         }
1657         return 0;
1658 }
1659
1660 int extent_read_full_page(struct extent_map_tree *tree, struct page *page,
1661                             get_extent_t *get_extent)
1662 {
1663         struct bio *bio = NULL;
1664         int ret;
1665
1666         ret = __extent_read_full_page(tree, page, get_extent, &bio);
1667         if (bio)
1668                 submit_one_bio(READ, bio);
1669         return ret;
1670 }
1671 EXPORT_SYMBOL(extent_read_full_page);
1672
1673 /*
1674  * the writepage semantics are similar to regular writepage.  extent
1675  * records are inserted to lock ranges in the tree, and as dirty areas
1676  * are found, they are marked writeback.  Then the lock bits are removed
1677  * and the end_io handler clears the writeback ranges
1678  */
1679 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
1680                               void *data)
1681 {
1682         struct inode *inode = page->mapping->host;
1683         struct extent_page_data *epd = data;
1684         struct extent_map_tree *tree = epd->tree;
1685         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1686         u64 page_end = start + PAGE_CACHE_SIZE - 1;
1687         u64 end;
1688         u64 cur = start;
1689         u64 extent_offset;
1690         u64 last_byte = i_size_read(inode);
1691         u64 block_start;
1692         u64 iosize;
1693         sector_t sector;
1694         struct extent_map *em;
1695         struct block_device *bdev;
1696         int ret;
1697         int nr = 0;
1698         size_t page_offset = 0;
1699         size_t blocksize;
1700         loff_t i_size = i_size_read(inode);
1701         unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
1702         u64 nr_delalloc;
1703         u64 delalloc_end;
1704
1705         WARN_ON(!PageLocked(page));
1706         if (page->index > end_index) {
1707                 clear_extent_dirty(tree, start, page_end, GFP_NOFS);
1708                 unlock_page(page);
1709                 return 0;
1710         }
1711
1712         if (page->index == end_index) {
1713                 size_t offset = i_size & (PAGE_CACHE_SIZE - 1);
1714                 zero_user_page(page, offset,
1715                                PAGE_CACHE_SIZE - offset, KM_USER0);
1716         }
1717
1718         set_page_extent_mapped(page);
1719
1720         lock_extent(tree, start, page_end, GFP_NOFS);
1721         nr_delalloc = find_lock_delalloc_range(tree, start, page_end + 1,
1722                                                &delalloc_end,
1723                                                128 * 1024 * 1024);
1724         if (nr_delalloc) {
1725                 tree->ops->fill_delalloc(inode, start, delalloc_end);
1726                 if (delalloc_end >= page_end + 1) {
1727                         clear_extent_bit(tree, page_end + 1, delalloc_end,
1728                                          EXTENT_LOCKED | EXTENT_DELALLOC,
1729                                          1, 0, GFP_NOFS);
1730                 }
1731                 clear_extent_bit(tree, start, page_end, EXTENT_DELALLOC,
1732                                  0, 0, GFP_NOFS);
1733                 if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0)) {
1734                         printk("found delalloc bits after clear extent_bit\n");
1735                 }
1736         } else if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0)) {
1737                 printk("found delalloc bits after find_delalloc_range returns 0\n");
1738         }
1739
1740         end = page_end;
1741         if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0)) {
1742                 printk("found delalloc bits after lock_extent\n");
1743         }
1744
1745         if (last_byte <= start) {
1746                 clear_extent_dirty(tree, start, page_end, GFP_NOFS);
1747                 goto done;
1748         }
1749
1750         set_extent_uptodate(tree, start, page_end, GFP_NOFS);
1751         blocksize = inode->i_sb->s_blocksize;
1752
1753         while (cur <= end) {
1754                 if (cur >= last_byte) {
1755                         clear_extent_dirty(tree, cur, page_end, GFP_NOFS);
1756                         break;
1757                 }
1758                 em = epd->get_extent(inode, page, page_offset, cur, end, 1);
1759                 if (IS_ERR(em) || !em) {
1760                         SetPageError(page);
1761                         break;
1762                 }
1763
1764                 extent_offset = cur - em->start;
1765                 BUG_ON(em->end < cur);
1766                 BUG_ON(end < cur);
1767                 iosize = min(em->end - cur, end - cur) + 1;
1768                 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
1769                 sector = (em->block_start + extent_offset) >> 9;
1770                 bdev = em->bdev;
1771                 block_start = em->block_start;
1772                 free_extent_map(em);
1773                 em = NULL;
1774
1775                 if (block_start == EXTENT_MAP_HOLE ||
1776                     block_start == EXTENT_MAP_INLINE) {
1777                         clear_extent_dirty(tree, cur,
1778                                            cur + iosize - 1, GFP_NOFS);
1779                         cur = cur + iosize;
1780                         page_offset += iosize;
1781                         continue;
1782                 }
1783
1784                 /* leave this out until we have a page_mkwrite call */
1785                 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
1786                                    EXTENT_DIRTY, 0)) {
1787                         cur = cur + iosize;
1788                         page_offset += iosize;
1789                         continue;
1790                 }
1791                 clear_extent_dirty(tree, cur, cur + iosize - 1, GFP_NOFS);
1792                 if (tree->ops && tree->ops->writepage_io_hook) {
1793                         ret = tree->ops->writepage_io_hook(page, cur,
1794                                                 cur + iosize - 1);
1795                 } else {
1796                         ret = 0;
1797                 }
1798                 if (ret)
1799                         SetPageError(page);
1800                 else {
1801                         unsigned long nr = end_index + 1;
1802                         set_range_writeback(tree, cur, cur + iosize - 1);
1803
1804                         ret = submit_extent_page(WRITE, tree, page, sector,
1805                                                  iosize, page_offset, bdev,
1806                                                  &epd->bio, nr,
1807                                                  end_bio_extent_writepage);
1808                         if (ret)
1809                                 SetPageError(page);
1810                 }
1811                 cur = cur + iosize;
1812                 page_offset += iosize;
1813                 nr++;
1814         }
1815 done:
1816         unlock_extent(tree, start, page_end, GFP_NOFS);
1817         unlock_page(page);
1818         return 0;
1819 }
1820
1821 int extent_write_full_page(struct extent_map_tree *tree, struct page *page,
1822                           get_extent_t *get_extent,
1823                           struct writeback_control *wbc)
1824 {
1825         int ret;
1826         struct extent_page_data epd = {
1827                 .bio = NULL,
1828                 .tree = tree,
1829                 .get_extent = get_extent,
1830         };
1831
1832         ret = __extent_writepage(page, wbc, &epd);
1833         if (epd.bio)
1834                 submit_one_bio(WRITE, epd.bio);
1835         return ret;
1836 }
1837 EXPORT_SYMBOL(extent_write_full_page);
1838
1839 int extent_writepages(struct extent_map_tree *tree,
1840                       struct address_space *mapping,
1841                       get_extent_t *get_extent,
1842                       struct writeback_control *wbc)
1843 {
1844         int ret;
1845         struct extent_page_data epd = {
1846                 .bio = NULL,
1847                 .tree = tree,
1848                 .get_extent = get_extent,
1849         };
1850
1851         ret = write_cache_pages(mapping, wbc, __extent_writepage, &epd);
1852         if (epd.bio)
1853                 submit_one_bio(WRITE, epd.bio);
1854         return ret;
1855 }
1856 EXPORT_SYMBOL(extent_writepages);
1857
1858 int extent_readpages(struct extent_map_tree *tree,
1859                      struct address_space *mapping,
1860                      struct list_head *pages, unsigned nr_pages,
1861                      get_extent_t get_extent)
1862 {
1863         struct bio *bio = NULL;
1864         unsigned page_idx;
1865         struct pagevec pvec;
1866
1867         pagevec_init(&pvec, 0);
1868         for (page_idx = 0; page_idx < nr_pages; page_idx++) {
1869                 struct page *page = list_entry(pages->prev, struct page, lru);
1870
1871                 prefetchw(&page->flags);
1872                 list_del(&page->lru);
1873                 /*
1874                  * what we want to do here is call add_to_page_cache_lru,
1875                  * but that isn't exported, so we reproduce it here
1876                  */
1877                 if (!add_to_page_cache(page, mapping,
1878                                         page->index, GFP_KERNEL)) {
1879
1880                         /* open coding of lru_cache_add, also not exported */
1881                         page_cache_get(page);
1882                         if (!pagevec_add(&pvec, page))
1883                                 __pagevec_lru_add(&pvec);
1884                         __extent_read_full_page(tree, page, get_extent, &bio);
1885                 }
1886                 page_cache_release(page);
1887         }
1888         if (pagevec_count(&pvec))
1889                 __pagevec_lru_add(&pvec);
1890         BUG_ON(!list_empty(pages));
1891         if (bio)
1892                 submit_one_bio(READ, bio);
1893         return 0;
1894 }
1895 EXPORT_SYMBOL(extent_readpages);
1896
1897 /*
1898  * basic invalidatepage code, this waits on any locked or writeback
1899  * ranges corresponding to the page, and then deletes any extent state
1900  * records from the tree
1901  */
1902 int extent_invalidatepage(struct extent_map_tree *tree,
1903                           struct page *page, unsigned long offset)
1904 {
1905         u64 start = ((u64)page->index << PAGE_CACHE_SHIFT);
1906         u64 end = start + PAGE_CACHE_SIZE - 1;
1907         size_t blocksize = page->mapping->host->i_sb->s_blocksize;
1908
1909         start += (offset + blocksize -1) & ~(blocksize - 1);
1910         if (start > end)
1911                 return 0;
1912
1913         lock_extent(tree, start, end, GFP_NOFS);
1914         wait_on_extent_writeback(tree, start, end);
1915         clear_extent_bit(tree, start, end,
1916                          EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC,
1917                          1, 1, GFP_NOFS);
1918         return 0;
1919 }
1920 EXPORT_SYMBOL(extent_invalidatepage);
1921
1922 /*
1923  * simple commit_write call, set_range_dirty is used to mark both
1924  * the pages and the extent records as dirty
1925  */
1926 int extent_commit_write(struct extent_map_tree *tree,
1927                         struct inode *inode, struct page *page,
1928                         unsigned from, unsigned to)
1929 {
1930         loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1931
1932         set_page_extent_mapped(page);
1933         set_page_dirty(page);
1934
1935         if (pos > inode->i_size) {
1936                 i_size_write(inode, pos);
1937                 mark_inode_dirty(inode);
1938         }
1939         return 0;
1940 }
1941 EXPORT_SYMBOL(extent_commit_write);
1942
1943 int extent_prepare_write(struct extent_map_tree *tree,
1944                          struct inode *inode, struct page *page,
1945                          unsigned from, unsigned to, get_extent_t *get_extent)
1946 {
1947         u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
1948         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
1949         u64 block_start;
1950         u64 orig_block_start;
1951         u64 block_end;
1952         u64 cur_end;
1953         struct extent_map *em;
1954         unsigned blocksize = 1 << inode->i_blkbits;
1955         size_t page_offset = 0;
1956         size_t block_off_start;
1957         size_t block_off_end;
1958         int err = 0;
1959         int iocount = 0;
1960         int ret = 0;
1961         int isnew;
1962
1963         set_page_extent_mapped(page);
1964
1965         block_start = (page_start + from) & ~((u64)blocksize - 1);
1966         block_end = (page_start + to - 1) | (blocksize - 1);
1967         orig_block_start = block_start;
1968
1969         lock_extent(tree, page_start, page_end, GFP_NOFS);
1970         while(block_start <= block_end) {
1971                 em = get_extent(inode, page, page_offset, block_start,
1972                                 block_end, 1);
1973                 if (IS_ERR(em) || !em) {
1974                         goto err;
1975                 }
1976                 cur_end = min(block_end, em->end);
1977                 block_off_start = block_start & (PAGE_CACHE_SIZE - 1);
1978                 block_off_end = block_off_start + blocksize;
1979                 isnew = clear_extent_new(tree, block_start, cur_end, GFP_NOFS);
1980
1981                 if (!PageUptodate(page) && isnew &&
1982                     (block_off_end > to || block_off_start < from)) {
1983                         void *kaddr;
1984
1985                         kaddr = kmap_atomic(page, KM_USER0);
1986                         if (block_off_end > to)
1987                                 memset(kaddr + to, 0, block_off_end - to);
1988                         if (block_off_start < from)
1989                                 memset(kaddr + block_off_start, 0,
1990                                        from - block_off_start);
1991                         flush_dcache_page(page);
1992                         kunmap_atomic(kaddr, KM_USER0);
1993                 }
1994                 if (!isnew && !PageUptodate(page) &&
1995                     (block_off_end > to || block_off_start < from) &&
1996                     !test_range_bit(tree, block_start, cur_end,
1997                                     EXTENT_UPTODATE, 1)) {
1998                         u64 sector;
1999                         u64 extent_offset = block_start - em->start;
2000                         size_t iosize;
2001                         sector = (em->block_start + extent_offset) >> 9;
2002                         iosize = (cur_end - block_start + blocksize - 1) &
2003                                 ~((u64)blocksize - 1);
2004                         /*
2005                          * we've already got the extent locked, but we
2006                          * need to split the state such that our end_bio
2007                          * handler can clear the lock.
2008                          */
2009                         set_extent_bit(tree, block_start,
2010                                        block_start + iosize - 1,
2011                                        EXTENT_LOCKED, 0, NULL, GFP_NOFS);
2012                         ret = submit_extent_page(READ, tree, page,
2013                                          sector, iosize, page_offset, em->bdev,
2014                                          NULL, 1,
2015                                          end_bio_extent_preparewrite);
2016                         iocount++;
2017                         block_start = block_start + iosize;
2018                 } else {
2019                         set_extent_uptodate(tree, block_start, cur_end,
2020                                             GFP_NOFS);
2021                         unlock_extent(tree, block_start, cur_end, GFP_NOFS);
2022                         block_start = cur_end + 1;
2023                 }
2024                 page_offset = block_start & (PAGE_CACHE_SIZE - 1);
2025                 free_extent_map(em);
2026         }
2027         if (iocount) {
2028                 wait_extent_bit(tree, orig_block_start,
2029                                 block_end, EXTENT_LOCKED);
2030         }
2031         check_page_uptodate(tree, page);
2032 err:
2033         /* FIXME, zero out newly allocated blocks on error */
2034         return err;
2035 }
2036 EXPORT_SYMBOL(extent_prepare_write);
2037
2038 /*
2039  * a helper for releasepage.  As long as there are no locked extents
2040  * in the range corresponding to the page, both state records and extent
2041  * map records are removed
2042  */
2043 int try_release_extent_mapping(struct extent_map_tree *tree, struct page *page)
2044 {
2045         struct extent_map *em;
2046         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2047         u64 end = start + PAGE_CACHE_SIZE - 1;
2048         u64 orig_start = start;
2049         int ret = 1;
2050
2051         while (start <= end) {
2052                 em = lookup_extent_mapping(tree, start, end);
2053                 if (!em || IS_ERR(em))
2054                         break;
2055                 if (!test_range_bit(tree, em->start, em->end,
2056                                     EXTENT_LOCKED, 0)) {
2057                         remove_extent_mapping(tree, em);
2058                         /* once for the rb tree */
2059                         free_extent_map(em);
2060                 }
2061                 start = em->end + 1;
2062                 /* once for us */
2063                 free_extent_map(em);
2064         }
2065         if (test_range_bit(tree, orig_start, end, EXTENT_LOCKED, 0))
2066                 ret = 0;
2067         else
2068                 clear_extent_bit(tree, orig_start, end, EXTENT_UPTODATE,
2069                                  1, 1, GFP_NOFS);
2070         return ret;
2071 }
2072 EXPORT_SYMBOL(try_release_extent_mapping);
2073
2074 sector_t extent_bmap(struct address_space *mapping, sector_t iblock,
2075                 get_extent_t *get_extent)
2076 {
2077         struct inode *inode = mapping->host;
2078         u64 start = iblock << inode->i_blkbits;
2079         u64 end = start + (1 << inode->i_blkbits) - 1;
2080         sector_t sector = 0;
2081         struct extent_map *em;
2082
2083         em = get_extent(inode, NULL, 0, start, end, 0);
2084         if (!em || IS_ERR(em))
2085                 return 0;
2086
2087         if (em->block_start == EXTENT_MAP_INLINE ||
2088             em->block_start == EXTENT_MAP_HOLE)
2089                 goto out;
2090
2091         sector = (em->block_start + start - em->start) >> inode->i_blkbits;
2092 out:
2093         free_extent_map(em);
2094         return sector;
2095 }
2096
2097 static int add_lru(struct extent_map_tree *tree, struct extent_buffer *eb)
2098 {
2099         if (list_empty(&eb->lru)) {
2100                 extent_buffer_get(eb);
2101                 list_add(&eb->lru, &tree->buffer_lru);
2102                 tree->lru_size++;
2103                 if (tree->lru_size >= BUFFER_LRU_MAX) {
2104                         struct extent_buffer *rm;
2105                         rm = list_entry(tree->buffer_lru.prev,
2106                                         struct extent_buffer, lru);
2107                         tree->lru_size--;
2108                         list_del_init(&rm->lru);
2109                         free_extent_buffer(rm);
2110                 }
2111         } else
2112                 list_move(&eb->lru, &tree->buffer_lru);
2113         return 0;
2114 }
2115 static struct extent_buffer *find_lru(struct extent_map_tree *tree,
2116                                       u64 start, unsigned long len)
2117 {
2118         struct list_head *lru = &tree->buffer_lru;
2119         struct list_head *cur = lru->next;
2120         struct extent_buffer *eb;
2121
2122         if (list_empty(lru))
2123                 return NULL;
2124
2125         do {
2126                 eb = list_entry(cur, struct extent_buffer, lru);
2127                 if (eb->start == start && eb->len == len) {
2128                         extent_buffer_get(eb);
2129                         return eb;
2130                 }
2131                 cur = cur->next;
2132         } while (cur != lru);
2133         return NULL;
2134 }
2135
2136 static inline unsigned long num_extent_pages(u64 start, u64 len)
2137 {
2138         return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
2139                 (start >> PAGE_CACHE_SHIFT);
2140 }
2141
2142 static inline struct page *extent_buffer_page(struct extent_buffer *eb,
2143                                               unsigned long i)
2144 {
2145         struct page *p;
2146         struct address_space *mapping;
2147
2148         if (i == 0)
2149                 return eb->first_page;
2150         i += eb->start >> PAGE_CACHE_SHIFT;
2151         mapping = eb->first_page->mapping;
2152         read_lock_irq(&mapping->tree_lock);
2153         p = radix_tree_lookup(&mapping->page_tree, i);
2154         read_unlock_irq(&mapping->tree_lock);
2155         return p;
2156 }
2157
2158 static struct extent_buffer *__alloc_extent_buffer(struct extent_map_tree *tree,
2159                                                    u64 start,
2160                                                    unsigned long len,
2161                                                    gfp_t mask)
2162 {
2163         struct extent_buffer *eb = NULL;
2164
2165         spin_lock(&tree->lru_lock);
2166         eb = find_lru(tree, start, len);
2167         spin_unlock(&tree->lru_lock);
2168         if (eb) {
2169                 return eb;
2170         }
2171
2172         eb = kmem_cache_zalloc(extent_buffer_cache, mask);
2173         INIT_LIST_HEAD(&eb->lru);
2174         eb->start = start;
2175         eb->len = len;
2176         atomic_set(&eb->refs, 1);
2177
2178         return eb;
2179 }
2180
2181 static void __free_extent_buffer(struct extent_buffer *eb)
2182 {
2183         kmem_cache_free(extent_buffer_cache, eb);
2184 }
2185
2186 struct extent_buffer *alloc_extent_buffer(struct extent_map_tree *tree,
2187                                           u64 start, unsigned long len,
2188                                           struct page *page0,
2189                                           gfp_t mask)
2190 {
2191         unsigned long num_pages = num_extent_pages(start, len);
2192         unsigned long i;
2193         unsigned long index = start >> PAGE_CACHE_SHIFT;
2194         struct extent_buffer *eb;
2195         struct page *p;
2196         struct address_space *mapping = tree->mapping;
2197         int uptodate = 1;
2198
2199         eb = __alloc_extent_buffer(tree, start, len, mask);
2200         if (!eb || IS_ERR(eb))
2201                 return NULL;
2202
2203         if (eb->flags & EXTENT_BUFFER_FILLED)
2204                 goto lru_add;
2205
2206         if (page0) {
2207                 eb->first_page = page0;
2208                 i = 1;
2209                 index++;
2210                 page_cache_get(page0);
2211                 mark_page_accessed(page0);
2212                 set_page_extent_mapped(page0);
2213                 WARN_ON(!PageUptodate(page0));
2214                 set_page_private(page0, EXTENT_PAGE_PRIVATE_FIRST_PAGE |
2215                                  len << 2);
2216         } else {
2217                 i = 0;
2218         }
2219         for (; i < num_pages; i++, index++) {
2220                 p = find_or_create_page(mapping, index, mask | __GFP_HIGHMEM);
2221                 if (!p) {
2222                         WARN_ON(1);
2223                         goto fail;
2224                 }
2225                 set_page_extent_mapped(p);
2226                 mark_page_accessed(p);
2227                 if (i == 0) {
2228                         eb->first_page = p;
2229                         set_page_private(p, EXTENT_PAGE_PRIVATE_FIRST_PAGE |
2230                                          len << 2);
2231                 } else {
2232                         set_page_private(p, EXTENT_PAGE_PRIVATE);
2233                 }
2234                 if (!PageUptodate(p))
2235                         uptodate = 0;
2236                 unlock_page(p);
2237         }
2238         if (uptodate)
2239                 eb->flags |= EXTENT_UPTODATE;
2240         eb->flags |= EXTENT_BUFFER_FILLED;
2241
2242 lru_add:
2243         spin_lock(&tree->lru_lock);
2244         add_lru(tree, eb);
2245         spin_unlock(&tree->lru_lock);
2246         return eb;
2247
2248 fail:
2249         spin_lock(&tree->lru_lock);
2250         list_del_init(&eb->lru);
2251         spin_unlock(&tree->lru_lock);
2252         if (!atomic_dec_and_test(&eb->refs))
2253                 return NULL;
2254         for (index = 1; index < i; index++) {
2255                 page_cache_release(extent_buffer_page(eb, index));
2256         }
2257         if (i > 0)
2258                 page_cache_release(extent_buffer_page(eb, 0));
2259         __free_extent_buffer(eb);
2260         return NULL;
2261 }
2262 EXPORT_SYMBOL(alloc_extent_buffer);
2263
2264 struct extent_buffer *find_extent_buffer(struct extent_map_tree *tree,
2265                                          u64 start, unsigned long len,
2266                                           gfp_t mask)
2267 {
2268         unsigned long num_pages = num_extent_pages(start, len);
2269         unsigned long i;
2270         unsigned long index = start >> PAGE_CACHE_SHIFT;
2271         struct extent_buffer *eb;
2272         struct page *p;
2273         struct address_space *mapping = tree->mapping;
2274         int uptodate = 1;
2275
2276         eb = __alloc_extent_buffer(tree, start, len, mask);
2277         if (!eb || IS_ERR(eb))
2278                 return NULL;
2279
2280         if (eb->flags & EXTENT_BUFFER_FILLED)
2281                 goto lru_add;
2282
2283         for (i = 0; i < num_pages; i++, index++) {
2284                 p = find_lock_page(mapping, index);
2285                 if (!p) {
2286                         goto fail;
2287                 }
2288                 set_page_extent_mapped(p);
2289                 mark_page_accessed(p);
2290
2291                 if (i == 0) {
2292                         eb->first_page = p;
2293                         set_page_private(p, EXTENT_PAGE_PRIVATE_FIRST_PAGE |
2294                                          len << 2);
2295                 } else {
2296                         set_page_private(p, EXTENT_PAGE_PRIVATE);
2297                 }
2298
2299                 if (!PageUptodate(p))
2300                         uptodate = 0;
2301                 unlock_page(p);
2302         }
2303         if (uptodate)
2304                 eb->flags |= EXTENT_UPTODATE;
2305         eb->flags |= EXTENT_BUFFER_FILLED;
2306
2307 lru_add:
2308         spin_lock(&tree->lru_lock);
2309         add_lru(tree, eb);
2310         spin_unlock(&tree->lru_lock);
2311         return eb;
2312 fail:
2313         spin_lock(&tree->lru_lock);
2314         list_del_init(&eb->lru);
2315         spin_unlock(&tree->lru_lock);
2316         if (!atomic_dec_and_test(&eb->refs))
2317                 return NULL;
2318         for (index = 1; index < i; index++) {
2319                 page_cache_release(extent_buffer_page(eb, index));
2320         }
2321         if (i > 0)
2322                 page_cache_release(extent_buffer_page(eb, 0));
2323         __free_extent_buffer(eb);
2324         return NULL;
2325 }
2326 EXPORT_SYMBOL(find_extent_buffer);
2327
2328 void free_extent_buffer(struct extent_buffer *eb)
2329 {
2330         unsigned long i;
2331         unsigned long num_pages;
2332
2333         if (!eb)
2334                 return;
2335
2336         if (!atomic_dec_and_test(&eb->refs))
2337                 return;
2338
2339         WARN_ON(!list_empty(&eb->lru));
2340         num_pages = num_extent_pages(eb->start, eb->len);
2341
2342         for (i = 1; i < num_pages; i++) {
2343                 page_cache_release(extent_buffer_page(eb, i));
2344         }
2345         page_cache_release(extent_buffer_page(eb, 0));
2346         __free_extent_buffer(eb);
2347 }
2348 EXPORT_SYMBOL(free_extent_buffer);
2349
2350 int clear_extent_buffer_dirty(struct extent_map_tree *tree,
2351                               struct extent_buffer *eb)
2352 {
2353         int set;
2354         unsigned long i;
2355         unsigned long num_pages;
2356         struct page *page;
2357
2358         u64 start = eb->start;
2359         u64 end = start + eb->len - 1;
2360
2361         set = clear_extent_dirty(tree, start, end, GFP_NOFS);
2362         num_pages = num_extent_pages(eb->start, eb->len);
2363
2364         for (i = 0; i < num_pages; i++) {
2365                 page = extent_buffer_page(eb, i);
2366                 lock_page(page);
2367                 /*
2368                  * if we're on the last page or the first page and the
2369                  * block isn't aligned on a page boundary, do extra checks
2370                  * to make sure we don't clean page that is partially dirty
2371                  */
2372                 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
2373                     ((i == num_pages - 1) &&
2374                      ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
2375                         start = (u64)page->index << PAGE_CACHE_SHIFT;
2376                         end  = start + PAGE_CACHE_SIZE - 1;
2377                         if (test_range_bit(tree, start, end,
2378                                            EXTENT_DIRTY, 0)) {
2379                                 unlock_page(page);
2380                                 continue;
2381                         }
2382                 }
2383                 clear_page_dirty_for_io(page);
2384                 unlock_page(page);
2385         }
2386         return 0;
2387 }
2388 EXPORT_SYMBOL(clear_extent_buffer_dirty);
2389
2390 int wait_on_extent_buffer_writeback(struct extent_map_tree *tree,
2391                                     struct extent_buffer *eb)
2392 {
2393         return wait_on_extent_writeback(tree, eb->start,
2394                                         eb->start + eb->len - 1);
2395 }
2396 EXPORT_SYMBOL(wait_on_extent_buffer_writeback);
2397
2398 int set_extent_buffer_dirty(struct extent_map_tree *tree,
2399                              struct extent_buffer *eb)
2400 {
2401         unsigned long i;
2402         unsigned long num_pages;
2403
2404         num_pages = num_extent_pages(eb->start, eb->len);
2405         for (i = 0; i < num_pages; i++) {
2406                 struct page *page = extent_buffer_page(eb, i);
2407                 /* writepage may need to do something special for the
2408                  * first page, we have to make sure page->private is
2409                  * properly set.  releasepage may drop page->private
2410                  * on us if the page isn't already dirty.
2411                  */
2412                 if (i == 0) {
2413                         lock_page(page);
2414                         set_page_private(page,
2415                                          EXTENT_PAGE_PRIVATE_FIRST_PAGE |
2416                                          eb->len << 2);
2417                 }
2418                 __set_page_dirty_nobuffers(extent_buffer_page(eb, i));
2419                 if (i == 0)
2420                         unlock_page(page);
2421         }
2422         return set_extent_dirty(tree, eb->start,
2423                                 eb->start + eb->len - 1, GFP_NOFS);
2424 }
2425 EXPORT_SYMBOL(set_extent_buffer_dirty);
2426
2427 int set_extent_buffer_uptodate(struct extent_map_tree *tree,
2428                                 struct extent_buffer *eb)
2429 {
2430         unsigned long i;
2431         struct page *page;
2432         unsigned long num_pages;
2433
2434         num_pages = num_extent_pages(eb->start, eb->len);
2435
2436         set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
2437                             GFP_NOFS);
2438         for (i = 0; i < num_pages; i++) {
2439                 page = extent_buffer_page(eb, i);
2440                 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
2441                     ((i == num_pages - 1) &&
2442                      ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
2443                         check_page_uptodate(tree, page);
2444                         continue;
2445                 }
2446                 SetPageUptodate(page);
2447         }
2448         return 0;
2449 }
2450 EXPORT_SYMBOL(set_extent_buffer_uptodate);
2451
2452 int extent_buffer_uptodate(struct extent_map_tree *tree,
2453                              struct extent_buffer *eb)
2454 {
2455         if (eb->flags & EXTENT_UPTODATE)
2456                 return 1;
2457         return test_range_bit(tree, eb->start, eb->start + eb->len - 1,
2458                            EXTENT_UPTODATE, 1);
2459 }
2460 EXPORT_SYMBOL(extent_buffer_uptodate);
2461
2462 int read_extent_buffer_pages(struct extent_map_tree *tree,
2463                              struct extent_buffer *eb,
2464                              u64 start,
2465                              int wait)
2466 {
2467         unsigned long i;
2468         unsigned long start_i;
2469         struct page *page;
2470         int err;
2471         int ret = 0;
2472         unsigned long num_pages;
2473
2474         if (eb->flags & EXTENT_UPTODATE)
2475                 return 0;
2476
2477         if (0 && test_range_bit(tree, eb->start, eb->start + eb->len - 1,
2478                            EXTENT_UPTODATE, 1)) {
2479                 return 0;
2480         }
2481
2482         if (start) {
2483                 WARN_ON(start < eb->start);
2484                 start_i = (start >> PAGE_CACHE_SHIFT) -
2485                         (eb->start >> PAGE_CACHE_SHIFT);
2486         } else {
2487                 start_i = 0;
2488         }
2489
2490         num_pages = num_extent_pages(eb->start, eb->len);
2491         for (i = start_i; i < num_pages; i++) {
2492                 page = extent_buffer_page(eb, i);
2493                 if (PageUptodate(page)) {
2494                         continue;
2495                 }
2496                 if (!wait) {
2497                         if (TestSetPageLocked(page)) {
2498                                 continue;
2499                         }
2500                 } else {
2501                         lock_page(page);
2502                 }
2503                 if (!PageUptodate(page)) {
2504                         err = page->mapping->a_ops->readpage(NULL, page);
2505                         if (err) {
2506                                 ret = err;
2507                         }
2508                 } else {
2509                         unlock_page(page);
2510                 }
2511         }
2512
2513         if (ret || !wait) {
2514                 return ret;
2515         }
2516
2517         for (i = start_i; i < num_pages; i++) {
2518                 page = extent_buffer_page(eb, i);
2519                 wait_on_page_locked(page);
2520                 if (!PageUptodate(page)) {
2521                         ret = -EIO;
2522                 }
2523         }
2524         if (!ret)
2525                 eb->flags |= EXTENT_UPTODATE;
2526         return ret;
2527 }
2528 EXPORT_SYMBOL(read_extent_buffer_pages);
2529
2530 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
2531                         unsigned long start,
2532                         unsigned long len)
2533 {
2534         size_t cur;
2535         size_t offset;
2536         struct page *page;
2537         char *kaddr;
2538         char *dst = (char *)dstv;
2539         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
2540         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
2541         unsigned long num_pages = num_extent_pages(eb->start, eb->len);
2542
2543         WARN_ON(start > eb->len);
2544         WARN_ON(start + len > eb->start + eb->len);
2545
2546         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
2547
2548         while(len > 0) {
2549                 page = extent_buffer_page(eb, i);
2550                 if (!PageUptodate(page)) {
2551                         printk("page %lu not up to date i %lu, total %lu, len %lu\n", page->index, i, num_pages, eb->len);
2552                         WARN_ON(1);
2553                 }
2554                 WARN_ON(!PageUptodate(page));
2555
2556                 cur = min(len, (PAGE_CACHE_SIZE - offset));
2557                 kaddr = kmap_atomic(page, KM_USER1);
2558                 memcpy(dst, kaddr + offset, cur);
2559                 kunmap_atomic(kaddr, KM_USER1);
2560
2561                 dst += cur;
2562                 len -= cur;
2563                 offset = 0;
2564                 i++;
2565         }
2566 }
2567 EXPORT_SYMBOL(read_extent_buffer);
2568
2569 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
2570                                unsigned long min_len, char **token, char **map,
2571                                unsigned long *map_start,
2572                                unsigned long *map_len, int km)
2573 {
2574         size_t offset = start & (PAGE_CACHE_SIZE - 1);
2575         char *kaddr;
2576         struct page *p;
2577         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
2578         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
2579         unsigned long end_i = (start_offset + start + min_len - 1) >>
2580                 PAGE_CACHE_SHIFT;
2581
2582         if (i != end_i)
2583                 return -EINVAL;
2584
2585         if (i == 0) {
2586                 offset = start_offset;
2587                 *map_start = 0;
2588         } else {
2589                 offset = 0;
2590                 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
2591         }
2592         if (start + min_len > eb->len) {
2593 printk("bad mapping eb start %Lu len %lu, wanted %lu %lu\n", eb->start, eb->len, start, min_len);
2594                 WARN_ON(1);
2595         }
2596
2597         p = extent_buffer_page(eb, i);
2598         WARN_ON(!PageUptodate(p));
2599         kaddr = kmap_atomic(p, km);
2600         *token = kaddr;
2601         *map = kaddr + offset;
2602         *map_len = PAGE_CACHE_SIZE - offset;
2603         return 0;
2604 }
2605 EXPORT_SYMBOL(map_private_extent_buffer);
2606
2607 int map_extent_buffer(struct extent_buffer *eb, unsigned long start,
2608                       unsigned long min_len,
2609                       char **token, char **map,
2610                       unsigned long *map_start,
2611                       unsigned long *map_len, int km)
2612 {
2613         int err;
2614         int save = 0;
2615         if (eb->map_token) {
2616                 unmap_extent_buffer(eb, eb->map_token, km);
2617                 eb->map_token = NULL;
2618                 save = 1;
2619         }
2620         err = map_private_extent_buffer(eb, start, min_len, token, map,
2621                                        map_start, map_len, km);
2622         if (!err && save) {
2623                 eb->map_token = *token;
2624                 eb->kaddr = *map;
2625                 eb->map_start = *map_start;
2626                 eb->map_len = *map_len;
2627         }
2628         return err;
2629 }
2630 EXPORT_SYMBOL(map_extent_buffer);
2631
2632 void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km)
2633 {
2634         kunmap_atomic(token, km);
2635 }
2636 EXPORT_SYMBOL(unmap_extent_buffer);
2637
2638 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
2639                           unsigned long start,
2640                           unsigned long len)
2641 {
2642         size_t cur;
2643         size_t offset;
2644         struct page *page;
2645         char *kaddr;
2646         char *ptr = (char *)ptrv;
2647         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
2648         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
2649         int ret = 0;
2650
2651         WARN_ON(start > eb->len);
2652         WARN_ON(start + len > eb->start + eb->len);
2653
2654         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
2655
2656         while(len > 0) {
2657                 page = extent_buffer_page(eb, i);
2658                 WARN_ON(!PageUptodate(page));
2659
2660                 cur = min(len, (PAGE_CACHE_SIZE - offset));
2661
2662                 kaddr = kmap_atomic(page, KM_USER0);
2663                 ret = memcmp(ptr, kaddr + offset, cur);
2664                 kunmap_atomic(kaddr, KM_USER0);
2665                 if (ret)
2666                         break;
2667
2668                 ptr += cur;
2669                 len -= cur;
2670                 offset = 0;
2671                 i++;
2672         }
2673         return ret;
2674 }
2675 EXPORT_SYMBOL(memcmp_extent_buffer);
2676
2677 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
2678                          unsigned long start, unsigned long len)
2679 {
2680         size_t cur;
2681         size_t offset;
2682         struct page *page;
2683         char *kaddr;
2684         char *src = (char *)srcv;
2685         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
2686         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
2687
2688         WARN_ON(start > eb->len);
2689         WARN_ON(start + len > eb->start + eb->len);
2690
2691         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
2692
2693         while(len > 0) {
2694                 page = extent_buffer_page(eb, i);
2695                 WARN_ON(!PageUptodate(page));
2696
2697                 cur = min(len, PAGE_CACHE_SIZE - offset);
2698                 kaddr = kmap_atomic(page, KM_USER1);
2699                 memcpy(kaddr + offset, src, cur);
2700                 kunmap_atomic(kaddr, KM_USER1);
2701
2702                 src += cur;
2703                 len -= cur;
2704                 offset = 0;
2705                 i++;
2706         }
2707 }
2708 EXPORT_SYMBOL(write_extent_buffer);
2709
2710 void memset_extent_buffer(struct extent_buffer *eb, char c,
2711                           unsigned long start, unsigned long len)
2712 {
2713         size_t cur;
2714         size_t offset;
2715         struct page *page;
2716         char *kaddr;
2717         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
2718         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
2719
2720         WARN_ON(start > eb->len);
2721         WARN_ON(start + len > eb->start + eb->len);
2722
2723         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
2724
2725         while(len > 0) {
2726                 page = extent_buffer_page(eb, i);
2727                 WARN_ON(!PageUptodate(page));
2728
2729                 cur = min(len, PAGE_CACHE_SIZE - offset);
2730                 kaddr = kmap_atomic(page, KM_USER0);
2731                 memset(kaddr + offset, c, cur);
2732                 kunmap_atomic(kaddr, KM_USER0);
2733
2734                 len -= cur;
2735                 offset = 0;
2736                 i++;
2737         }
2738 }
2739 EXPORT_SYMBOL(memset_extent_buffer);
2740
2741 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
2742                         unsigned long dst_offset, unsigned long src_offset,
2743                         unsigned long len)
2744 {
2745         u64 dst_len = dst->len;
2746         size_t cur;
2747         size_t offset;
2748         struct page *page;
2749         char *kaddr;
2750         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
2751         unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
2752
2753         WARN_ON(src->len != dst_len);
2754
2755         offset = (start_offset + dst_offset) &
2756                 ((unsigned long)PAGE_CACHE_SIZE - 1);
2757
2758         while(len > 0) {
2759                 page = extent_buffer_page(dst, i);
2760                 WARN_ON(!PageUptodate(page));
2761
2762                 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
2763
2764                 kaddr = kmap_atomic(page, KM_USER0);
2765                 read_extent_buffer(src, kaddr + offset, src_offset, cur);
2766                 kunmap_atomic(kaddr, KM_USER0);
2767
2768                 src_offset += cur;
2769                 len -= cur;
2770                 offset = 0;
2771                 i++;
2772         }
2773 }
2774 EXPORT_SYMBOL(copy_extent_buffer);
2775
2776 static void move_pages(struct page *dst_page, struct page *src_page,
2777                        unsigned long dst_off, unsigned long src_off,
2778                        unsigned long len)
2779 {
2780         char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
2781         if (dst_page == src_page) {
2782                 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
2783         } else {
2784                 char *src_kaddr = kmap_atomic(src_page, KM_USER1);
2785                 char *p = dst_kaddr + dst_off + len;
2786                 char *s = src_kaddr + src_off + len;
2787
2788                 while (len--)
2789                         *--p = *--s;
2790
2791                 kunmap_atomic(src_kaddr, KM_USER1);
2792         }
2793         kunmap_atomic(dst_kaddr, KM_USER0);
2794 }
2795
2796 static void copy_pages(struct page *dst_page, struct page *src_page,
2797                        unsigned long dst_off, unsigned long src_off,
2798                        unsigned long len)
2799 {
2800         char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
2801         char *src_kaddr;
2802
2803         if (dst_page != src_page)
2804                 src_kaddr = kmap_atomic(src_page, KM_USER1);
2805         else
2806                 src_kaddr = dst_kaddr;
2807
2808         memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
2809         kunmap_atomic(dst_kaddr, KM_USER0);
2810         if (dst_page != src_page)
2811                 kunmap_atomic(src_kaddr, KM_USER1);
2812 }
2813
2814 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
2815                            unsigned long src_offset, unsigned long len)
2816 {
2817         size_t cur;
2818         size_t dst_off_in_page;
2819         size_t src_off_in_page;
2820         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
2821         unsigned long dst_i;
2822         unsigned long src_i;
2823
2824         if (src_offset + len > dst->len) {
2825                 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
2826                        src_offset, len, dst->len);
2827                 BUG_ON(1);
2828         }
2829         if (dst_offset + len > dst->len) {
2830                 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
2831                        dst_offset, len, dst->len);
2832                 BUG_ON(1);
2833         }
2834
2835         while(len > 0) {
2836                 dst_off_in_page = (start_offset + dst_offset) &
2837                         ((unsigned long)PAGE_CACHE_SIZE - 1);
2838                 src_off_in_page = (start_offset + src_offset) &
2839                         ((unsigned long)PAGE_CACHE_SIZE - 1);
2840
2841                 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
2842                 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
2843
2844                 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
2845                                                src_off_in_page));
2846                 cur = min_t(unsigned long, cur,
2847                         (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
2848
2849                 copy_pages(extent_buffer_page(dst, dst_i),
2850                            extent_buffer_page(dst, src_i),
2851                            dst_off_in_page, src_off_in_page, cur);
2852
2853                 src_offset += cur;
2854                 dst_offset += cur;
2855                 len -= cur;
2856         }
2857 }
2858 EXPORT_SYMBOL(memcpy_extent_buffer);
2859
2860 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
2861                            unsigned long src_offset, unsigned long len)
2862 {
2863         size_t cur;
2864         size_t dst_off_in_page;
2865         size_t src_off_in_page;
2866         unsigned long dst_end = dst_offset + len - 1;
2867         unsigned long src_end = src_offset + len - 1;
2868         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
2869         unsigned long dst_i;
2870         unsigned long src_i;
2871
2872         if (src_offset + len > dst->len) {
2873                 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
2874                        src_offset, len, dst->len);
2875                 BUG_ON(1);
2876         }
2877         if (dst_offset + len > dst->len) {
2878                 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
2879                        dst_offset, len, dst->len);
2880                 BUG_ON(1);
2881         }
2882         if (dst_offset < src_offset) {
2883                 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
2884                 return;
2885         }
2886         while(len > 0) {
2887                 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
2888                 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
2889
2890                 dst_off_in_page = (start_offset + dst_end) &
2891                         ((unsigned long)PAGE_CACHE_SIZE - 1);
2892                 src_off_in_page = (start_offset + src_end) &
2893                         ((unsigned long)PAGE_CACHE_SIZE - 1);
2894
2895                 cur = min_t(unsigned long, len, src_off_in_page + 1);
2896                 cur = min(cur, dst_off_in_page + 1);
2897                 move_pages(extent_buffer_page(dst, dst_i),
2898                            extent_buffer_page(dst, src_i),
2899                            dst_off_in_page - cur + 1,
2900                            src_off_in_page - cur + 1, cur);
2901
2902                 dst_end -= cur;
2903                 src_end -= cur;
2904                 len -= cur;
2905         }
2906 }
2907 EXPORT_SYMBOL(memmove_extent_buffer);