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 int __init extent_map_init(void)
  46 {
  47         extent_map_cache = btrfs_cache_create("extent_map",
  48                                             sizeof(struct extent_map), 0,
  49                                             NULL);
  50         if (!extent_map_cache)
  51                 return -ENOMEM;
  52         extent_state_cache = btrfs_cache_create("extent_state",
  53                                             sizeof(struct extent_state), 0,
  54                                             NULL);
  55         if (!extent_state_cache)
  56                 goto free_map_cache;
  57         extent_buffer_cache = btrfs_cache_create("extent_buffers",
  58                                             sizeof(struct extent_buffer), 0,
  59                                             NULL);
  60         if (!extent_buffer_cache)
  61                 goto free_state_cache;
  62         return 0;
  63
  64 free_state_cache:
  65         kmem_cache_destroy(extent_state_cache);
  66 free_map_cache:
  67         kmem_cache_destroy(extent_map_cache);
  68         return -ENOMEM;
  69 }
  70
  71 void __exit extent_map_exit(void)
  72 {
  73         struct extent_state *state;
  74
  75         while (!list_empty(&states)) {
  76                 state = list_entry(states.next, struct extent_state, list);
  77                 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));
  78                 list_del(&state->list);
  79                 kmem_cache_free(extent_state_cache, state);
  80
  81         }
  82
  83         if (extent_map_cache)
  84                 kmem_cache_destroy(extent_map_cache);
  85         if (extent_state_cache)
  86                 kmem_cache_destroy(extent_state_cache);
  87         if (extent_buffer_cache)
  88                 kmem_cache_destroy(extent_buffer_cache);
  89 }
  90
  91 void extent_map_tree_init(struct extent_map_tree *tree,
  92                           struct address_space *mapping, gfp_t mask)
  93 {
  94         tree->map.rb_node = NULL;
  95         tree->state.rb_node = NULL;
  96         tree->ops = NULL;
  97         rwlock_init(&tree->lock);
  98         spin_lock_init(&tree->lru_lock);
  99         tree->mapping = mapping;
 100         INIT_LIST_HEAD(&tree->buffer_lru);
 101         tree->lru_size = 0;
 102 }
 103 EXPORT_SYMBOL(extent_map_tree_init);
 104
 105 void extent_map_tree_empty_lru(struct extent_map_tree *tree)
 106 {
 107         struct extent_buffer *eb;
 108         while(!list_empty(&tree->buffer_lru)) {
 109                 eb = list_entry(tree->buffer_lru.next, struct extent_buffer,
 110                                 lru);
 111                 list_del_init(&eb->lru);
 112                 free_extent_buffer(eb);
 113         }
 114 }
 115 EXPORT_SYMBOL(extent_map_tree_empty_lru);
 116
 117 struct extent_map *alloc_extent_map(gfp_t mask)
 118 {
 119         struct extent_map *em;
 120         em = kmem_cache_alloc(extent_map_cache, mask);
 121         if (!em || IS_ERR(em))
 122                 return em;
 123         em->in_tree = 0;
 124         atomic_set(&em->refs, 1);
 125         return em;
 126 }
 127 EXPORT_SYMBOL(alloc_extent_map);
 128
 129 void free_extent_map(struct extent_map *em)
 130 {
 131         if (!em)
 132                 return;
 133         if (atomic_dec_and_test(&em->refs)) {
 134                 WARN_ON(em->in_tree);
 135                 kmem_cache_free(extent_map_cache, em);
 136         }
 137 }
 138 EXPORT_SYMBOL(free_extent_map);
 139
 140
 141 struct extent_state *alloc_extent_state(gfp_t mask)
 142 {
 143         struct extent_state *state;
 144         unsigned long flags;
 145
 146         state = kmem_cache_alloc(extent_state_cache, mask);
 147         if (!state || IS_ERR(state))
 148                 return state;
 149         state->state = 0;
 150         state->in_tree = 0;
 151         state->private = 0;
 152
 153         spin_lock_irqsave(&state_lock, flags);
 154         list_add(&state->list, &states);
 155         spin_unlock_irqrestore(&state_lock, flags);
 156
 157         atomic_set(&state->refs, 1);
 158         init_waitqueue_head(&state->wq);
 159         return state;
 160 }
 161 EXPORT_SYMBOL(alloc_extent_state);
 162
 163 void free_extent_state(struct extent_state *state)
 164 {
 165         unsigned long flags;
 166         if (!state)
 167                 return;
 168         if (atomic_dec_and_test(&state->refs)) {
 169                 WARN_ON(state->in_tree);
 170                 spin_lock_irqsave(&state_lock, flags);
 171                 list_del(&state->list);
 172                 spin_unlock_irqrestore(&state_lock, flags);
 173                 kmem_cache_free(extent_state_cache, state);
 174         }
 175 }
 176 EXPORT_SYMBOL(free_extent_state);
 177
 178 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
 179                                    struct rb_node *node)
 180 {
 181         struct rb_node ** p = &root->rb_node;
 182         struct rb_node * parent = NULL;
 183         struct tree_entry *entry;
 184
 185         while(*p) {
 186                 parent = *p;
 187                 entry = rb_entry(parent, struct tree_entry, rb_node);
 188
 189                 if (offset < entry->start)
 190                         p = &(*p)->rb_left;
 191                 else if (offset > entry->end)
 192                         p = &(*p)->rb_right;
 193                 else
 194                         return parent;
 195         }
 196
 197         entry = rb_entry(node, struct tree_entry, rb_node);
 198         entry->in_tree = 1;
 199         rb_link_node(node, parent, p);
 200         rb_insert_color(node, root);
 201         return NULL;
 202 }
 203
 204 static struct rb_node *__tree_search(struct rb_root *root, u64 offset,
 205                                    struct rb_node **prev_ret)
 206 {
 207         struct rb_node * n = root->rb_node;
 208         struct rb_node *prev = NULL;
 209         struct tree_entry *entry;
 210         struct tree_entry *prev_entry = NULL;
 211
 212         while(n) {
 213                 entry = rb_entry(n, struct tree_entry, rb_node);
 214                 prev = n;
 215                 prev_entry = entry;
 216
 217                 if (offset < entry->start)
 218                         n = n->rb_left;
 219                 else if (offset > entry->end)
 220                         n = n->rb_right;
 221                 else
 222                         return n;
 223         }
 224         if (!prev_ret)
 225                 return NULL;
 226         while(prev && offset > prev_entry->end) {
 227                 prev = rb_next(prev);
 228                 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
 229         }
 230         *prev_ret = prev;
 231         return NULL;
 232 }
 233
 234 static inline struct rb_node *tree_search(struct rb_root *root, u64 offset)
 235 {
 236         struct rb_node *prev;
 237         struct rb_node *ret;
 238         ret = __tree_search(root, offset, &prev);
 239         if (!ret)
 240                 return prev;
 241         return ret;
 242 }
 243
 244 static int tree_delete(struct rb_root *root, u64 offset)
 245 {
 246         struct rb_node *node;
 247         struct tree_entry *entry;
 248
 249         node = __tree_search(root, offset, NULL);
 250         if (!node)
 251                 return -ENOENT;
 252         entry = rb_entry(node, struct tree_entry, rb_node);
 253         entry->in_tree = 0;
 254         rb_erase(node, root);
 255         return 0;
 256 }
 257
 258 /*
 259  * add_extent_mapping tries a simple backward merge with existing
 260  * mappings.  The extent_map struct passed in will be inserted into
 261  * the tree directly (no copies made, just a reference taken).
 262  */
 263 int add_extent_mapping(struct extent_map_tree *tree,
 264                        struct extent_map *em)
 265 {
 266         int ret = 0;
 267         struct extent_map *prev = NULL;
 268         struct rb_node *rb;
 269
 270         write_lock_irq(&tree->lock);
 271         rb = tree_insert(&tree->map, em->end, &em->rb_node);
 272         if (rb) {
 273                 prev = rb_entry(rb, struct extent_map, rb_node);
 274                 printk("found extent map %Lu %Lu on insert of %Lu %Lu\n", prev->start, prev->end, em->start, em->end);
 275                 ret = -EEXIST;
 276                 goto out;
 277         }
 278         atomic_inc(&em->refs);
 279         if (em->start != 0) {
 280                 rb = rb_prev(&em->rb_node);
 281                 if (rb)
 282                         prev = rb_entry(rb, struct extent_map, rb_node);
 283                 if (prev && prev->end + 1 == em->start &&
 284                     ((em->block_start == EXTENT_MAP_HOLE &&
 285                       prev->block_start == EXTENT_MAP_HOLE) ||
 286                      (em->block_start == EXTENT_MAP_INLINE &&
 287                       prev->block_start == EXTENT_MAP_INLINE) ||
 288                      (em->block_start == EXTENT_MAP_DELALLOC &&
 289                       prev->block_start == EXTENT_MAP_DELALLOC) ||
 290                      (em->block_start < EXTENT_MAP_DELALLOC - 1 &&
 291                       em->block_start == prev->block_end + 1))) {
 292                         em->start = prev->start;
 293                         em->block_start = prev->block_start;
 294                         rb_erase(&prev->rb_node, &tree->map);
 295                         prev->in_tree = 0;
 296                         free_extent_map(prev);
 297                 }
 298          }
 299 out:
 300         write_unlock_irq(&tree->lock);
 301         return ret;
 302 }
 303 EXPORT_SYMBOL(add_extent_mapping);
 304
 305 /*
 306  * lookup_extent_mapping returns the first extent_map struct in the
 307  * tree that intersects the [start, end] (inclusive) range.  There may
 308  * be additional objects in the tree that intersect, so check the object
 309  * returned carefully to make sure you don't need additional lookups.
 310  */
 311 struct extent_map *lookup_extent_mapping(struct extent_map_tree *tree,
 312                                          u64 start, u64 end)
 313 {
 314         struct extent_map *em;
 315         struct rb_node *rb_node;
 316
 317         read_lock_irq(&tree->lock);
 318         rb_node = tree_search(&tree->map, start);
 319         if (!rb_node) {
 320                 em = NULL;
 321                 goto out;
 322         }
 323         if (IS_ERR(rb_node)) {
 324                 em = ERR_PTR(PTR_ERR(rb_node));
 325                 goto out;
 326         }
 327         em = rb_entry(rb_node, struct extent_map, rb_node);
 328         if (em->end < start || em->start > end) {
 329                 em = NULL;
 330                 goto out;
 331         }
 332         atomic_inc(&em->refs);
 333 out:
 334         read_unlock_irq(&tree->lock);
 335         return em;
 336 }
 337 EXPORT_SYMBOL(lookup_extent_mapping);
 338
 339 /*
 340  * removes an extent_map struct from the tree.  No reference counts are
 341  * dropped, and no checks are done to  see if the range is in use
 342  */
 343 int remove_extent_mapping(struct extent_map_tree *tree, struct extent_map *em)
 344 {
 345         int ret;
 346
 347         write_lock_irq(&tree->lock);
 348         ret = tree_delete(&tree->map, em->end);
 349         write_unlock_irq(&tree->lock);
 350         return ret;
 351 }
 352 EXPORT_SYMBOL(remove_extent_mapping);
 353
 354 /*
 355  * utility function to look for merge candidates inside a given range.
 356  * Any extents with matching state are merged together into a single
 357  * extent in the tree.  Extents with EXTENT_IO in their state field
 358  * are not merged because the end_io handlers need to be able to do
 359  * operations on them without sleeping (or doing allocations/splits).
 360  *
 361  * This should be called with the tree lock held.
 362  */
 363 static int merge_state(struct extent_map_tree *tree,
 364                        struct extent_state *state)
 365 {
 366         struct extent_state *other;
 367         struct rb_node *other_node;
 368
 369         if (state->state & EXTENT_IOBITS)
 370                 return 0;
 371
 372         other_node = rb_prev(&state->rb_node);
 373         if (other_node) {
 374                 other = rb_entry(other_node, struct extent_state, rb_node);
 375                 if (other->end == state->start - 1 &&
 376                     other->state == state->state) {
 377                         state->start = other->start;
 378                         other->in_tree = 0;
 379                         rb_erase(&other->rb_node, &tree->state);
 380                         free_extent_state(other);
 381                 }
 382         }
 383         other_node = rb_next(&state->rb_node);
 384         if (other_node) {
 385                 other = rb_entry(other_node, struct extent_state, rb_node);
 386                 if (other->start == state->end + 1 &&
 387                     other->state == state->state) {
 388                         other->start = state->start;
 389                         state->in_tree = 0;
 390                         rb_erase(&state->rb_node, &tree->state);
 391                         free_extent_state(state);
 392                 }
 393         }
 394         return 0;
 395 }
 396
 397 /*
 398  * insert an extent_state struct into the tree.  'bits' are set on the
 399  * struct before it is inserted.
 400  *
 401  * This may return -EEXIST if the extent is already there, in which case the
 402  * state struct is freed.
 403  *
 404  * The tree lock is not taken internally.  This is a utility function and
 405  * probably isn't what you want to call (see set/clear_extent_bit).
 406  */
 407 static int insert_state(struct extent_map_tree *tree,
 408                         struct extent_state *state, u64 start, u64 end,
 409                         int bits)
 410 {
 411         struct rb_node *node;
 412
 413         if (end < start) {
 414                 printk("end < start %Lu %Lu\n", end, start);
 415                 WARN_ON(1);
 416         }
 417         state->state |= bits;
 418         state->start = start;
 419         state->end = end;
 420         node = tree_insert(&tree->state, end, &state->rb_node);
 421         if (node) {
 422                 struct extent_state *found;
 423                 found = rb_entry(node, struct extent_state, rb_node);
 424                 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, start, end);
 425                 free_extent_state(state);
 426                 return -EEXIST;
 427         }
 428         merge_state(tree, state);
 429         return 0;
 430 }
 431
 432 /*
 433  * split a given extent state struct in two, inserting the preallocated
 434  * struct 'prealloc' as the newly created second half.  'split' indicates an
 435  * offset inside 'orig' where it should be split.
 436  *
 437  * Before calling,
 438  * the tree has 'orig' at [orig->start, orig->end].  After calling, there
 439  * are two extent state structs in the tree:
 440  * prealloc: [orig->start, split - 1]
 441  * orig: [ split, orig->end ]
 442  *
 443  * The tree locks are not taken by this function. They need to be held
 444  * by the caller.
 445  */
 446 static int split_state(struct extent_map_tree *tree, struct extent_state *orig,
 447                        struct extent_state *prealloc, u64 split)
 448 {
 449         struct rb_node *node;
 450         prealloc->start = orig->start;
 451         prealloc->end = split - 1;
 452         prealloc->state = orig->state;
 453         orig->start = split;
 454
 455         node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
 456         if (node) {
 457                 struct extent_state *found;
 458                 found = rb_entry(node, struct extent_state, rb_node);
 459                 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, prealloc->start, prealloc->end);
 460                 free_extent_state(prealloc);
 461                 return -EEXIST;
 462         }
 463         return 0;
 464 }
 465
 466 /*
 467  * utility function to clear some bits in an extent state struct.
 468  * it will optionally wake up any one waiting on this state (wake == 1), or
 469  * forcibly remove the state from the tree (delete == 1).
 470  *
 471  * If no bits are set on the state struct after clearing things, the
 472  * struct is freed and removed from the tree
 473  */
 474 static int clear_state_bit(struct extent_map_tree *tree,
 475                             struct extent_state *state, int bits, int wake,
 476                             int delete)
 477 {
 478         int ret = state->state & bits;
 479         state->state &= ~bits;
 480         if (wake)
 481                 wake_up(&state->wq);
 482         if (delete || state->state == 0) {
 483                 if (state->in_tree) {
 484                         rb_erase(&state->rb_node, &tree->state);
 485                         state->in_tree = 0;
 486                         free_extent_state(state);
 487                 } else {
 488                         WARN_ON(1);
 489                 }
 490         } else {
 491                 merge_state(tree, state);
 492         }
 493         return ret;
 494 }
 495
 496 /*
 497  * clear some bits on a range in the tree.  This may require splitting
 498  * or inserting elements in the tree, so the gfp mask is used to
 499  * indicate which allocations or sleeping are allowed.
 500  *
 501  * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
 502  * the given range from the tree regardless of state (ie for truncate).
 503  *
 504  * the range [start, end] is inclusive.
 505  *
 506  * This takes the tree lock, and returns < 0 on error, > 0 if any of the
 507  * bits were already set, or zero if none of the bits were already set.
 508  */
 509 int clear_extent_bit(struct extent_map_tree *tree, u64 start, u64 end,
 510                      int bits, int wake, int delete, gfp_t mask)
 511 {
 512         struct extent_state *state;
 513         struct extent_state *prealloc = NULL;
 514         struct rb_node *node;
 515         unsigned long flags;
 516         int err;
 517         int set = 0;
 518
 519 again:
 520         if (!prealloc && (mask & __GFP_WAIT)) {
 521                 prealloc = alloc_extent_state(mask);
 522                 if (!prealloc)
 523                         return -ENOMEM;
 524         }
 525
 526         write_lock_irqsave(&tree->lock, flags);
 527         /*
 528          * this search will find the extents that end after
 529          * our range starts
 530          */
 531         node = tree_search(&tree->state, start);
 532         if (!node)
 533                 goto out;
 534         state = rb_entry(node, struct extent_state, rb_node);
 535         if (state->start > end)
 536                 goto out;
 537         WARN_ON(state->end < start);
 538
 539         /*
 540          *     | ---- desired range ---- |
 541          *  | state | or
 542          *  | ------------- state -------------- |
 543          *
 544          * We need to split the extent we found, and may flip
 545          * bits on second half.
 546          *
 547          * If the extent we found extends past our range, we
 548          * just split and search again.  It'll get split again
 549          * the next time though.
 550          *
 551          * If the extent we found is inside our range, we clear
 552          * the desired bit on it.
 553          */
 554
 555         if (state->start < start) {
 556                 err = split_state(tree, state, prealloc, start);
 557                 BUG_ON(err == -EEXIST);
 558                 prealloc = NULL;
 559                 if (err)
 560                         goto out;
 561                 if (state->end <= end) {
 562                         start = state->end + 1;
 563                         set |= clear_state_bit(tree, state, bits,
 564                                         wake, delete);
 565                 } else {
 566                         start = state->start;
 567                 }
 568                 goto search_again;
 569         }
 570         /*
 571          * | ---- desired range ---- |
 572          *                        | state |
 573          * We need to split the extent, and clear the bit
 574          * on the first half
 575          */
 576         if (state->start <= end && state->end > end) {
 577                 err = split_state(tree, state, prealloc, end + 1);
 578                 BUG_ON(err == -EEXIST);
 579
 580                 if (wake)
 581                         wake_up(&state->wq);
 582                 set |= clear_state_bit(tree, prealloc, bits,
 583                                        wake, delete);
 584                 prealloc = NULL;
 585                 goto out;
 586         }
 587
 588         start = state->end + 1;
 589         set |= clear_state_bit(tree, state, bits, wake, delete);
 590         goto search_again;
 591
 592 out:
 593         write_unlock_irqrestore(&tree->lock, flags);
 594         if (prealloc)
 595                 free_extent_state(prealloc);
 596
 597         return set;
 598
 599 search_again:
 600         if (start > end)
 601                 goto out;
 602         write_unlock_irqrestore(&tree->lock, flags);
 603         if (mask & __GFP_WAIT)
 604                 cond_resched();
 605         goto again;
 606 }
 607 EXPORT_SYMBOL(clear_extent_bit);
 608
 609 static int wait_on_state(struct extent_map_tree *tree,
 610                          struct extent_state *state)
 611 {
 612         DEFINE_WAIT(wait);
 613         prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
 614         read_unlock_irq(&tree->lock);
 615         schedule();
 616         read_lock_irq(&tree->lock);
 617         finish_wait(&state->wq, &wait);
 618         return 0;
 619 }
 620
 621 /*
 622  * waits for one or more bits to clear on a range in the state tree.
 623  * The range [start, end] is inclusive.
 624  * The tree lock is taken by this function
 625  */
 626 int wait_extent_bit(struct extent_map_tree *tree, u64 start, u64 end, int bits)
 627 {
 628         struct extent_state *state;
 629         struct rb_node *node;
 630
 631         read_lock_irq(&tree->lock);
 632 again:
 633         while (1) {
 634                 /*
 635                  * this search will find all the extents that end after
 636                  * our range starts
 637                  */
 638                 node = tree_search(&tree->state, start);
 639                 if (!node)
 640                         break;
 641
 642                 state = rb_entry(node, struct extent_state, rb_node);
 643
 644                 if (state->start > end)
 645                         goto out;
 646
 647                 if (state->state & bits) {
 648                         start = state->start;
 649                         atomic_inc(&state->refs);
 650                         wait_on_state(tree, state);
 651                         free_extent_state(state);
 652                         goto again;
 653                 }
 654                 start = state->end + 1;
 655
 656                 if (start > end)
 657                         break;
 658
 659                 if (need_resched()) {
 660                         read_unlock_irq(&tree->lock);
 661                         cond_resched();
 662                         read_lock_irq(&tree->lock);
 663                 }
 664         }
 665 out:
 666         read_unlock_irq(&tree->lock);
 667         return 0;
 668 }
 669 EXPORT_SYMBOL(wait_extent_bit);
 670
 671 /*
 672  * set some bits on a range in the tree.  This may require allocations
 673  * or sleeping, so the gfp mask is used to indicate what is allowed.
 674  *
 675  * If 'exclusive' == 1, this will fail with -EEXIST if some part of the
 676  * range already has the desired bits set.  The start of the existing
 677  * range is returned in failed_start in this case.
 678  *
 679  * [start, end] is inclusive
 680  * This takes the tree lock.
 681  */
 682 int set_extent_bit(struct extent_map_tree *tree, u64 start, u64 end, int bits,
 683                    int exclusive, u64 *failed_start, gfp_t mask)
 684 {
 685         struct extent_state *state;
 686         struct extent_state *prealloc = NULL;
 687         struct rb_node *node;
 688         unsigned long flags;
 689         int err = 0;
 690         int set;
 691         u64 last_start;
 692         u64 last_end;
 693 again:
 694         if (!prealloc && (mask & __GFP_WAIT)) {
 695                 prealloc = alloc_extent_state(mask);
 696                 if (!prealloc)
 697                         return -ENOMEM;
 698         }
 699
 700         write_lock_irqsave(&tree->lock, flags);
 701         /*
 702          * this search will find all the extents that end after
 703          * our range starts.
 704          */
 705         node = tree_search(&tree->state, start);
 706         if (!node) {
 707                 err = insert_state(tree, prealloc, start, end, bits);
 708                 prealloc = NULL;
 709                 BUG_ON(err == -EEXIST);
 710                 goto out;
 711         }
 712
 713         state = rb_entry(node, struct extent_state, rb_node);
 714         last_start = state->start;
 715         last_end = state->end;
 716
 717         /*
 718          * | ---- desired range ---- |
 719          * | state |
 720          *
 721          * Just lock what we found and keep going
 722          */
 723         if (state->start == start && state->end <= end) {
 724                 set = state->state & bits;
 725                 if (set && exclusive) {
 726                         *failed_start = state->start;
 727                         err = -EEXIST;
 728                         goto out;
 729                 }
 730                 state->state |= bits;
 731                 start = state->end + 1;
 732                 merge_state(tree, state);
 733                 goto search_again;
 734         }
 735
 736         /*
 737          *     | ---- desired range ---- |
 738          * | state |
 739          *   or
 740          * | ------------- state -------------- |
 741          *
 742          * We need to split the extent we found, and may flip bits on
 743          * second half.
 744          *
 745          * If the extent we found extends past our
 746          * range, we just split and search again.  It'll get split
 747          * again the next time though.
 748          *
 749          * If the extent we found is inside our range, we set the
 750          * desired bit on it.
 751          */
 752         if (state->start < start) {
 753                 set = state->state & bits;
 754                 if (exclusive && set) {
 755                         *failed_start = start;
 756                         err = -EEXIST;
 757                         goto out;
 758                 }
 759                 err = split_state(tree, state, prealloc, start);
 760                 BUG_ON(err == -EEXIST);
 761                 prealloc = NULL;
 762                 if (err)
 763                         goto out;
 764                 if (state->end <= end) {
 765                         state->state |= bits;
 766                         start = state->end + 1;
 767                         merge_state(tree, state);
 768                 } else {
 769                         start = state->start;
 770                 }
 771                 goto search_again;
 772         }
 773         /*
 774          * | ---- desired range ---- |
 775          *     | state | or               | state |
 776          *
 777          * There's a hole, we need to insert something in it and
 778          * ignore the extent we found.
 779          */
 780         if (state->start > start) {
 781                 u64 this_end;
 782                 if (end < last_start)
 783                         this_end = end;
 784                 else
 785                         this_end = last_start -1;
 786                 err = insert_state(tree, prealloc, start, this_end,
 787                                    bits);
 788                 prealloc = NULL;
 789                 BUG_ON(err == -EEXIST);
 790                 if (err)
 791                         goto out;
 792                 start = this_end + 1;
 793                 goto search_again;
 794         }
 795         /*
 796          * | ---- desired range ---- |
 797          *                        | state |
 798          * We need to split the extent, and set the bit
 799          * on the first half
 800          */
 801         if (state->start <= end && state->end > end) {
 802                 set = state->state & bits;
 803                 if (exclusive && set) {
 804                         *failed_start = start;
 805                         err = -EEXIST;
 806                         goto out;
 807                 }
 808                 err = split_state(tree, state, prealloc, end + 1);
 809                 BUG_ON(err == -EEXIST);
 810
 811                 prealloc->state |= bits;
 812                 merge_state(tree, prealloc);
 813                 prealloc = NULL;
 814                 goto out;
 815         }
 816
 817         goto search_again;
 818
 819 out:
 820         write_unlock_irqrestore(&tree->lock, flags);
 821         if (prealloc)
 822                 free_extent_state(prealloc);
 823
 824         return err;
 825
 826 search_again:
 827         if (start > end)
 828                 goto out;
 829         write_unlock_irqrestore(&tree->lock, flags);
 830         if (mask & __GFP_WAIT)
 831                 cond_resched();
 832         goto again;
 833 }
 834 EXPORT_SYMBOL(set_extent_bit);
 835
 836 /* wrappers around set/clear extent bit */
 837 int set_extent_dirty(struct extent_map_tree *tree, u64 start, u64 end,
 838                      gfp_t mask)
 839 {
 840         return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL,
 841                               mask);
 842 }
 843 EXPORT_SYMBOL(set_extent_dirty);
 844
 845 int set_extent_bits(struct extent_map_tree *tree, u64 start, u64 end,
 846                     int bits, gfp_t mask)
 847 {
 848         return set_extent_bit(tree, start, end, bits, 0, NULL,
 849                               mask);
 850 }
 851 EXPORT_SYMBOL(set_extent_bits);
 852
 853 int clear_extent_bits(struct extent_map_tree *tree, u64 start, u64 end,
 854                       int bits, gfp_t mask)
 855 {
 856         return clear_extent_bit(tree, start, end, bits, 0, 0, mask);
 857 }
 858 EXPORT_SYMBOL(clear_extent_bits);
 859
 860 int set_extent_delalloc(struct extent_map_tree *tree, u64 start, u64 end,
 861                      gfp_t mask)
 862 {
 863         return set_extent_bit(tree, start, end,
 864                               EXTENT_DELALLOC | EXTENT_DIRTY, 0, NULL,
 865                               mask);
 866 }
 867 EXPORT_SYMBOL(set_extent_delalloc);
 868
 869 int clear_extent_dirty(struct extent_map_tree *tree, u64 start, u64 end,
 870                        gfp_t mask)
 871 {
 872         return clear_extent_bit(tree, start, end,
 873                                 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, mask);
 874 }
 875 EXPORT_SYMBOL(clear_extent_dirty);
 876
 877 int set_extent_new(struct extent_map_tree *tree, u64 start, u64 end,
 878                      gfp_t mask)
 879 {
 880         return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL,
 881                               mask);
 882 }
 883 EXPORT_SYMBOL(set_extent_new);
 884
 885 int clear_extent_new(struct extent_map_tree *tree, u64 start, u64 end,
 886                        gfp_t mask)
 887 {
 888         return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0, mask);
 889 }
 890 EXPORT_SYMBOL(clear_extent_new);
 891
 892 int set_extent_uptodate(struct extent_map_tree *tree, u64 start, u64 end,
 893                         gfp_t mask)
 894 {
 895         return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, NULL,
 896                               mask);
 897 }
 898 EXPORT_SYMBOL(set_extent_uptodate);
 899
 900 int clear_extent_uptodate(struct extent_map_tree *tree, u64 start, u64 end,
 901                           gfp_t mask)
 902 {
 903         return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, mask);
 904 }
 905 EXPORT_SYMBOL(clear_extent_uptodate);
 906
 907 int set_extent_writeback(struct extent_map_tree *tree, u64 start, u64 end,
 908                          gfp_t mask)
 909 {
 910         return set_extent_bit(tree, start, end, EXTENT_WRITEBACK,
 911                               0, NULL, mask);
 912 }
 913 EXPORT_SYMBOL(set_extent_writeback);
 914
 915 int clear_extent_writeback(struct extent_map_tree *tree, u64 start, u64 end,
 916                            gfp_t mask)
 917 {
 918         return clear_extent_bit(tree, start, end, EXTENT_WRITEBACK, 1, 0, mask);
 919 }
 920 EXPORT_SYMBOL(clear_extent_writeback);
 921
 922 int wait_on_extent_writeback(struct extent_map_tree *tree, u64 start, u64 end)
 923 {
 924         return wait_extent_bit(tree, start, end, EXTENT_WRITEBACK);
 925 }
 926 EXPORT_SYMBOL(wait_on_extent_writeback);
 927
 928 /*
 929  * locks a range in ascending order, waiting for any locked regions
 930  * it hits on the way.  [start,end] are inclusive, and this will sleep.
 931  */
 932 int lock_extent(struct extent_map_tree *tree, u64 start, u64 end, gfp_t mask)
 933 {
 934         int err;
 935         u64 failed_start;
 936         while (1) {
 937                 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1,
 938                                      &failed_start, mask);
 939                 if (err == -EEXIST && (mask & __GFP_WAIT)) {
 940                         wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
 941                         start = failed_start;
 942                 } else {
 943                         break;
 944                 }
 945                 WARN_ON(start > end);
 946         }
 947         return err;
 948 }
 949 EXPORT_SYMBOL(lock_extent);
 950
 951 int unlock_extent(struct extent_map_tree *tree, u64 start, u64 end,
 952                   gfp_t mask)
 953 {
 954         return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, mask);
 955 }
 956 EXPORT_SYMBOL(unlock_extent);
 957
 958 /*
 959  * helper function to set pages and extents in the tree dirty
 960  */
 961 int set_range_dirty(struct extent_map_tree *tree, u64 start, u64 end)
 962 {
 963         unsigned long index = start >> PAGE_CACHE_SHIFT;
 964         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
 965         struct page *page;
 966
 967         while (index <= end_index) {
 968                 page = find_get_page(tree->mapping, index);
 969                 BUG_ON(!page);
 970                 __set_page_dirty_nobuffers(page);
 971                 page_cache_release(page);
 972                 index++;
 973         }
 974         set_extent_dirty(tree, start, end, GFP_NOFS);
 975         return 0;
 976 }
 977 EXPORT_SYMBOL(set_range_dirty);
 978
 979 /*
 980  * helper function to set both pages and extents in the tree writeback
 981  */
 982 int set_range_writeback(struct extent_map_tree *tree, u64 start, u64 end)
 983 {
 984         unsigned long index = start >> PAGE_CACHE_SHIFT;
 985         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
 986         struct page *page;
 987
 988         while (index <= end_index) {
 989                 page = find_get_page(tree->mapping, index);
 990                 BUG_ON(!page);
 991                 set_page_writeback(page);
 992                 page_cache_release(page);
 993                 index++;
 994         }
 995         set_extent_writeback(tree, start, end, GFP_NOFS);
 996         return 0;
 997 }
 998 EXPORT_SYMBOL(set_range_writeback);
 999
1000 int find_first_extent_bit(struct extent_map_tree *tree, u64 start,
1001                           u64 *start_ret, u64 *end_ret, int bits)
1002 {
1003         struct rb_node *node;
1004         struct extent_state *state;
1005         int ret = 1;
1006
1007         read_lock_irq(&tree->lock);
1008         /*
1009          * this search will find all the extents that end after
1010          * our range starts.
1011          */
1012         node = tree_search(&tree->state, start);
1013         if (!node || IS_ERR(node)) {
1014                 goto out;
1015         }
1016
1017         while(1) {
1018                 state = rb_entry(node, struct extent_state, rb_node);
1019                 if (state->end >= start && (state->state & bits)) {
1020                         *start_ret = state->start;
1021                         *end_ret = state->end;
1022                         ret = 0;
1023                         break;
1024                 }
1025                 node = rb_next(node);
1026                 if (!node)
1027                         break;
1028         }
1029 out:
1030         read_unlock_irq(&tree->lock);
1031         return ret;
1032 }
1033 EXPORT_SYMBOL(find_first_extent_bit);
1034
1035 u64 find_lock_delalloc_range(struct extent_map_tree *tree,
1036                              u64 *start, u64 *end, u64 max_bytes)
1037 {
1038         struct rb_node *node;
1039         struct extent_state *state;
1040         u64 cur_start = *start;
1041         u64 found = 0;
1042         u64 total_bytes = 0;
1043
1044         write_lock_irq(&tree->lock);
1045         /*
1046          * this search will find all the extents that end after
1047          * our range starts.
1048          */
1049 search_again:
1050         node = tree_search(&tree->state, cur_start);
1051         if (!node || IS_ERR(node)) {
1052                 goto out;
1053         }
1054
1055         while(1) {
1056                 state = rb_entry(node, struct extent_state, rb_node);
1057                 if (found && state->start != cur_start) {
1058                         goto out;
1059                 }
1060                 if (!(state->state & EXTENT_DELALLOC)) {
1061                         goto out;
1062                 }
1063                 if (!found) {
1064                         struct extent_state *prev_state;
1065                         struct rb_node *prev_node = node;
1066                         while(1) {
1067                                 prev_node = rb_prev(prev_node);
1068                                 if (!prev_node)
1069                                         break;
1070                                 prev_state = rb_entry(prev_node,
1071                                                       struct extent_state,
1072                                                       rb_node);
1073                                 if (!(prev_state->state & EXTENT_DELALLOC))
1074                                         break;
1075                                 state = prev_state;
1076                                 node = prev_node;
1077                         }
1078                 }
1079                 if (state->state & EXTENT_LOCKED) {
1080                         DEFINE_WAIT(wait);
1081                         atomic_inc(&state->refs);
1082                         prepare_to_wait(&state->wq, &wait,
1083                                         TASK_UNINTERRUPTIBLE);
1084                         write_unlock_irq(&tree->lock);
1085                         schedule();
1086                         write_lock_irq(&tree->lock);
1087                         finish_wait(&state->wq, &wait);
1088                         free_extent_state(state);
1089                         goto search_again;
1090                 }
1091                 state->state |= EXTENT_LOCKED;
1092                 if (!found)
1093                         *start = state->start;
1094                 found++;
1095                 *end = state->end;
1096                 cur_start = state->end + 1;
1097                 node = rb_next(node);
1098                 if (!node)
1099                         break;
1100                 total_bytes += state->end - state->start + 1;
1101                 if (total_bytes >= max_bytes)
1102                         break;
1103         }
1104 out:
1105         write_unlock_irq(&tree->lock);
1106         return found;
1107 }
1108
1109 /*
1110  * helper function to lock both pages and extents in the tree.
1111  * pages must be locked first.
1112  */
1113 int lock_range(struct extent_map_tree *tree, u64 start, u64 end)
1114 {
1115         unsigned long index = start >> PAGE_CACHE_SHIFT;
1116         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1117         struct page *page;
1118         int err;
1119
1120         while (index <= end_index) {
1121                 page = grab_cache_page(tree->mapping, index);
1122                 if (!page) {
1123                         err = -ENOMEM;
1124                         goto failed;
1125                 }
1126                 if (IS_ERR(page)) {
1127                         err = PTR_ERR(page);
1128                         goto failed;
1129                 }
1130                 index++;
1131         }
1132         lock_extent(tree, start, end, GFP_NOFS);
1133         return 0;
1134
1135 failed:
1136         /*
1137          * we failed above in getting the page at 'index', so we undo here
1138          * up to but not including the page at 'index'
1139          */
1140         end_index = index;
1141         index = start >> PAGE_CACHE_SHIFT;
1142         while (index < end_index) {
1143                 page = find_get_page(tree->mapping, index);
1144                 unlock_page(page);
1145                 page_cache_release(page);
1146                 index++;
1147         }
1148         return err;
1149 }
1150 EXPORT_SYMBOL(lock_range);
1151
1152 /*
1153  * helper function to unlock both pages and extents in the tree.
1154  */
1155 int unlock_range(struct extent_map_tree *tree, u64 start, u64 end)
1156 {
1157         unsigned long index = start >> PAGE_CACHE_SHIFT;
1158         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1159         struct page *page;
1160
1161         while (index <= end_index) {
1162                 page = find_get_page(tree->mapping, index);
1163                 unlock_page(page);
1164                 page_cache_release(page);
1165                 index++;
1166         }
1167         unlock_extent(tree, start, end, GFP_NOFS);
1168         return 0;
1169 }
1170 EXPORT_SYMBOL(unlock_range);
1171
1172 int set_state_private(struct extent_map_tree *tree, u64 start, u64 private)
1173 {
1174         struct rb_node *node;
1175         struct extent_state *state;
1176         int ret = 0;
1177
1178         write_lock_irq(&tree->lock);
1179         /*
1180          * this search will find all the extents that end after
1181          * our range starts.
1182          */
1183         node = tree_search(&tree->state, start);
1184         if (!node || IS_ERR(node)) {
1185                 ret = -ENOENT;
1186                 goto out;
1187         }
1188         state = rb_entry(node, struct extent_state, rb_node);
1189         if (state->start != start) {
1190                 ret = -ENOENT;
1191                 goto out;
1192         }
1193         state->private = private;
1194 out:
1195         write_unlock_irq(&tree->lock);
1196         return ret;
1197 }
1198
1199 int get_state_private(struct extent_map_tree *tree, u64 start, u64 *private)
1200 {
1201         struct rb_node *node;
1202         struct extent_state *state;
1203         int ret = 0;
1204
1205         read_lock_irq(&tree->lock);
1206         /*
1207          * this search will find all the extents that end after
1208          * our range starts.
1209          */
1210         node = tree_search(&tree->state, start);
1211         if (!node || IS_ERR(node)) {
1212                 ret = -ENOENT;
1213                 goto out;
1214         }
1215         state = rb_entry(node, struct extent_state, rb_node);
1216         if (state->start != start) {
1217                 ret = -ENOENT;
1218                 goto out;
1219         }
1220         *private = state->private;
1221 out:
1222         read_unlock_irq(&tree->lock);
1223         return ret;
1224 }
1225
1226 /*
1227  * searches a range in the state tree for a given mask.
1228  * If 'filled' == 1, this returns 1 only if ever extent in the tree
1229  * has the bits set.  Otherwise, 1 is returned if any bit in the
1230  * range is found set.
1231  */
1232 int test_range_bit(struct extent_map_tree *tree, u64 start, u64 end,
1233                    int bits, int filled)
1234 {
1235         struct extent_state *state = NULL;
1236         struct rb_node *node;
1237         int bitset = 0;
1238
1239         read_lock_irq(&tree->lock);
1240         node = tree_search(&tree->state, start);
1241         while (node && start <= end) {
1242                 state = rb_entry(node, struct extent_state, rb_node);
1243
1244                 if (filled && state->start > start) {
1245                         bitset = 0;
1246                         break;
1247                 }
1248
1249                 if (state->start > end)
1250                         break;
1251
1252                 if (state->state & bits) {
1253                         bitset = 1;
1254                         if (!filled)
1255                                 break;
1256                 } else if (filled) {
1257                         bitset = 0;
1258                         break;
1259                 }
1260                 start = state->end + 1;
1261                 if (start > end)
1262                         break;
1263                 node = rb_next(node);
1264         }
1265         read_unlock_irq(&tree->lock);
1266         return bitset;
1267 }
1268 EXPORT_SYMBOL(test_range_bit);
1269
1270 /*
1271  * helper function to set a given page up to date if all the
1272  * extents in the tree for that page are up to date
1273  */
1274 static int check_page_uptodate(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_UPTODATE, 1))
1280                 SetPageUptodate(page);
1281         return 0;
1282 }
1283
1284 /*
1285  * helper function to unlock a page if all the extents in the tree
1286  * for that page are unlocked
1287  */
1288 static int check_page_locked(struct extent_map_tree *tree,
1289                              struct page *page)
1290 {
1291         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1292         u64 end = start + PAGE_CACHE_SIZE - 1;
1293         if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0))
1294                 unlock_page(page);
1295         return 0;
1296 }
1297
1298 /*
1299  * helper function to end page writeback if all the extents
1300  * in the tree for that page are done with writeback
1301  */
1302 static int check_page_writeback(struct extent_map_tree *tree,
1303                              struct page *page)
1304 {
1305         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1306         u64 end = start + PAGE_CACHE_SIZE - 1;
1307         if (!test_range_bit(tree, start, end, EXTENT_WRITEBACK, 0))
1308                 end_page_writeback(page);
1309         return 0;
1310 }
1311
1312 /* lots and lots of room for performance fixes in the end_bio funcs */
1313
1314 /*
1315  * after a writepage IO is done, we need to:
1316  * clear the uptodate bits on error
1317  * clear the writeback bits in the extent tree for this IO
1318  * end_page_writeback if the page has no more pending IO
1319  *
1320  * Scheduling is not allowed, so the extent state tree is expected
1321  * to have one and only one object corresponding to this IO.
1322  */
1323 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1324 static void end_bio_extent_writepage(struct bio *bio, int err)
1325 #else
1326 static int end_bio_extent_writepage(struct bio *bio,
1327                                    unsigned int bytes_done, int err)
1328 #endif
1329 {
1330         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1331         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1332         struct extent_map_tree *tree = bio->bi_private;
1333         u64 start;
1334         u64 end;
1335         int whole_page;
1336
1337 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1338         if (bio->bi_size)
1339                 return 1;
1340 #endif
1341
1342         do {
1343                 struct page *page = bvec->bv_page;
1344                 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1345                          bvec->bv_offset;
1346                 end = start + bvec->bv_len - 1;
1347
1348                 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1349                         whole_page = 1;
1350                 else
1351                         whole_page = 0;
1352
1353                 if (--bvec >= bio->bi_io_vec)
1354                         prefetchw(&bvec->bv_page->flags);
1355
1356                 if (!uptodate) {
1357                         clear_extent_uptodate(tree, start, end, GFP_ATOMIC);
1358                         ClearPageUptodate(page);
1359                         SetPageError(page);
1360                 }
1361                 clear_extent_writeback(tree, start, end, GFP_ATOMIC);
1362
1363                 if (whole_page)
1364                         end_page_writeback(page);
1365                 else
1366                         check_page_writeback(tree, page);
1367                 if (tree->ops && tree->ops->writepage_end_io_hook)
1368                         tree->ops->writepage_end_io_hook(page, start, end);
1369         } while (bvec >= bio->bi_io_vec);
1370
1371         bio_put(bio);
1372 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1373         return 0;
1374 #endif
1375 }
1376
1377 /*
1378  * after a readpage IO is done, we need to:
1379  * clear the uptodate bits on error
1380  * set the uptodate bits if things worked
1381  * set the page up to date if all extents in the tree are uptodate
1382  * clear the lock bit in the extent tree
1383  * unlock the page if there are no other extents locked for it
1384  *
1385  * Scheduling is not allowed, so the extent state tree is expected
1386  * to have one and only one object corresponding to this IO.
1387  */
1388 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1389 static void end_bio_extent_readpage(struct bio *bio, int err)
1390 #else
1391 static int end_bio_extent_readpage(struct bio *bio,
1392                                    unsigned int bytes_done, int err)
1393 #endif
1394 {
1395         int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1396         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1397         struct extent_map_tree *tree = bio->bi_private;
1398         u64 start;
1399         u64 end;
1400         int whole_page;
1401         int ret;
1402
1403 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1404         if (bio->bi_size)
1405                 return 1;
1406 #endif
1407
1408         do {
1409                 struct page *page = bvec->bv_page;
1410                 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1411                         bvec->bv_offset;
1412                 end = start + bvec->bv_len - 1;
1413
1414                 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1415                         whole_page = 1;
1416                 else
1417                         whole_page = 0;
1418
1419                 if (--bvec >= bio->bi_io_vec)
1420                         prefetchw(&bvec->bv_page->flags);
1421
1422                 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
1423                         ret = tree->ops->readpage_end_io_hook(page, start, end);
1424                         if (ret)
1425                                 uptodate = 0;
1426                 }
1427                 if (uptodate) {
1428                         set_extent_uptodate(tree, start, end, GFP_ATOMIC);
1429                         if (whole_page)
1430                                 SetPageUptodate(page);
1431                         else
1432                                 check_page_uptodate(tree, page);
1433                 } else {
1434                         ClearPageUptodate(page);
1435                         SetPageError(page);
1436                 }
1437
1438                 unlock_extent(tree, start, end, GFP_ATOMIC);
1439
1440                 if (whole_page)
1441                         unlock_page(page);
1442                 else
1443                         check_page_locked(tree, page);
1444         } while (bvec >= bio->bi_io_vec);
1445
1446         bio_put(bio);
1447 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1448         return 0;
1449 #endif
1450 }
1451
1452 /*
1453  * IO done from prepare_write is pretty simple, we just unlock
1454  * the structs in the extent tree when done, and set the uptodate bits
1455  * as appropriate.
1456  */
1457 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1458 static void end_bio_extent_preparewrite(struct bio *bio, int err)
1459 #else
1460 static int end_bio_extent_preparewrite(struct bio *bio,
1461                                        unsigned int bytes_done, int err)
1462 #endif
1463 {
1464         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1465         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1466         struct extent_map_tree *tree = bio->bi_private;
1467         u64 start;
1468         u64 end;
1469
1470 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1471         if (bio->bi_size)
1472                 return 1;
1473 #endif
1474
1475         do {
1476                 struct page *page = bvec->bv_page;
1477                 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1478                         bvec->bv_offset;
1479                 end = start + bvec->bv_len - 1;
1480
1481                 if (--bvec >= bio->bi_io_vec)
1482                         prefetchw(&bvec->bv_page->flags);
1483
1484                 if (uptodate) {
1485                         set_extent_uptodate(tree, start, end, GFP_ATOMIC);
1486                 } else {
1487                         ClearPageUptodate(page);
1488                         SetPageError(page);
1489                 }
1490
1491                 unlock_extent(tree, start, end, GFP_ATOMIC);
1492
1493         } while (bvec >= bio->bi_io_vec);
1494
1495         bio_put(bio);
1496 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1497         return 0;
1498 #endif
1499 }
1500
1501 static struct bio *
1502 extent_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
1503                  gfp_t gfp_flags)
1504 {
1505         struct bio *bio;
1506
1507         bio = bio_alloc(gfp_flags, nr_vecs);
1508
1509         if (bio == NULL && (current->flags & PF_MEMALLOC)) {
1510                 while (!bio && (nr_vecs /= 2))
1511                         bio = bio_alloc(gfp_flags, nr_vecs);
1512         }
1513
1514         if (bio) {
1515                 bio->bi_bdev = bdev;
1516                 bio->bi_sector = first_sector;
1517         }
1518         return bio;
1519 }
1520
1521 static int submit_one_bio(int rw, struct bio *bio)
1522 {
1523         int ret = 0;
1524         bio_get(bio);
1525         submit_bio(rw, bio);
1526         if (bio_flagged(bio, BIO_EOPNOTSUPP))
1527                 ret = -EOPNOTSUPP;
1528         bio_put(bio);
1529         return ret;
1530 }
1531
1532 static int submit_extent_page(int rw, struct extent_map_tree *tree,
1533                               struct page *page, sector_t sector,
1534                               size_t size, unsigned long offset,
1535                               struct block_device *bdev,
1536                               struct bio **bio_ret,
1537                               unsigned long max_pages,
1538                               bio_end_io_t end_io_func)
1539 {
1540         int ret = 0;
1541         struct bio *bio;
1542         int nr;
1543
1544         if (bio_ret && *bio_ret) {
1545                 bio = *bio_ret;
1546                 if (bio->bi_sector + (bio->bi_size >> 9) != sector ||
1547                     bio_add_page(bio, page, size, offset) < size) {
1548                         ret = submit_one_bio(rw, bio);
1549                         bio = NULL;
1550                 } else {
1551                         return 0;
1552                 }
1553         }
1554         nr = min_t(int, max_pages, bio_get_nr_vecs(bdev));
1555         bio = extent_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
1556         if (!bio) {
1557                 printk("failed to allocate bio nr %d\n", nr);
1558         }
1559         bio_add_page(bio, page, size, offset);
1560         bio->bi_end_io = end_io_func;
1561         bio->bi_private = tree;
1562         if (bio_ret) {
1563                 *bio_ret = bio;
1564         } else {
1565                 ret = submit_one_bio(rw, bio);
1566         }
1567
1568         return ret;
1569 }
1570
1571 void set_page_extent_mapped(struct page *page)
1572 {
1573         if (!PagePrivate(page)) {
1574                 SetPagePrivate(page);
1575                 WARN_ON(!page->mapping->a_ops->invalidatepage);
1576                 set_page_private(page, EXTENT_PAGE_PRIVATE);
1577                 page_cache_get(page);
1578         }
1579 }
1580
1581 /*
1582  * basic readpage implementation.  Locked extent state structs are inserted
1583  * into the tree that are removed when the IO is done (by the end_io
1584  * handlers)
1585  */
1586 static int __extent_read_full_page(struct extent_map_tree *tree,
1587                                    struct page *page,
1588                                    get_extent_t *get_extent,
1589                                    struct bio **bio)
1590 {
1591         struct inode *inode = page->mapping->host;
1592         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1593         u64 page_end = start + PAGE_CACHE_SIZE - 1;
1594         u64 end;
1595         u64 cur = start;
1596         u64 extent_offset;
1597         u64 last_byte = i_size_read(inode);
1598         u64 block_start;
1599         u64 cur_end;
1600         sector_t sector;
1601         struct extent_map *em;
1602         struct block_device *bdev;
1603         int ret;
1604         int nr = 0;
1605         size_t page_offset = 0;
1606         size_t iosize;
1607         size_t blocksize = inode->i_sb->s_blocksize;
1608
1609         set_page_extent_mapped(page);
1610
1611         end = page_end;
1612         lock_extent(tree, start, end, GFP_NOFS);
1613
1614         while (cur <= end) {
1615                 if (cur >= last_byte) {
1616                         iosize = PAGE_CACHE_SIZE - page_offset;
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                         break;
1622                 }
1623                 em = get_extent(inode, page, page_offset, cur, end, 0);
1624                 if (IS_ERR(em) || !em) {
1625                         SetPageError(page);
1626                         unlock_extent(tree, cur, end, GFP_NOFS);
1627                         break;
1628                 }
1629
1630                 extent_offset = cur - em->start;
1631                 BUG_ON(em->end < cur);
1632                 BUG_ON(end < cur);
1633
1634                 iosize = min(em->end - cur, end - cur) + 1;
1635                 cur_end = min(em->end, end);
1636                 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
1637                 sector = (em->block_start + extent_offset) >> 9;
1638                 bdev = em->bdev;
1639                 block_start = em->block_start;
1640                 free_extent_map(em);
1641                 em = NULL;
1642
1643                 /* we've found a hole, just zero and go on */
1644                 if (block_start == EXTENT_MAP_HOLE) {
1645                         zero_user_page(page, page_offset, iosize, KM_USER0);
1646                         set_extent_uptodate(tree, cur, cur + iosize - 1,
1647                                             GFP_NOFS);
1648                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1649                         cur = cur + iosize;
1650                         page_offset += iosize;
1651                         continue;
1652                 }
1653                 /* the get_extent function already copied into the page */
1654                 if (test_range_bit(tree, cur, cur_end, EXTENT_UPTODATE, 1)) {
1655                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1656                         cur = cur + iosize;
1657                         page_offset += iosize;
1658                         continue;
1659                 }
1660
1661                 ret = 0;
1662                 if (tree->ops && tree->ops->readpage_io_hook) {
1663                         ret = tree->ops->readpage_io_hook(page, cur,
1664                                                           cur + iosize - 1);
1665                 }
1666                 if (!ret) {
1667                         unsigned long nr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
1668                         nr -= page->index;
1669                         ret = submit_extent_page(READ, tree, page,
1670                                          sector, iosize, page_offset,
1671                                          bdev, bio, nr,
1672                                          end_bio_extent_readpage);
1673                 }
1674                 if (ret)
1675                         SetPageError(page);
1676                 cur = cur + iosize;
1677                 page_offset += iosize;
1678                 nr++;
1679         }
1680         if (!nr) {
1681                 if (!PageError(page))
1682                         SetPageUptodate(page);
1683                 unlock_page(page);
1684         }
1685         return 0;
1686 }
1687
1688 int extent_read_full_page(struct extent_map_tree *tree, struct page *page,
1689                             get_extent_t *get_extent)
1690 {
1691         struct bio *bio = NULL;
1692         int ret;
1693
1694         ret = __extent_read_full_page(tree, page, get_extent, &bio);
1695         if (bio)
1696                 submit_one_bio(READ, bio);
1697         return ret;
1698 }
1699 EXPORT_SYMBOL(extent_read_full_page);
1700
1701 /*
1702  * the writepage semantics are similar to regular writepage.  extent
1703  * records are inserted to lock ranges in the tree, and as dirty areas
1704  * are found, they are marked writeback.  Then the lock bits are removed
1705  * and the end_io handler clears the writeback ranges
1706  */
1707 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
1708                               void *data)
1709 {
1710         struct inode *inode = page->mapping->host;
1711         struct extent_page_data *epd = data;
1712         struct extent_map_tree *tree = epd->tree;
1713         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1714         u64 delalloc_start;
1715         u64 page_end = start + PAGE_CACHE_SIZE - 1;
1716         u64 end;
1717         u64 cur = start;
1718         u64 extent_offset;
1719         u64 last_byte = i_size_read(inode);
1720         u64 block_start;
1721         u64 iosize;
1722         sector_t sector;
1723         struct extent_map *em;
1724         struct block_device *bdev;
1725         int ret;
1726         int nr = 0;
1727         size_t page_offset = 0;
1728         size_t blocksize;
1729         loff_t i_size = i_size_read(inode);
1730         unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
1731         u64 nr_delalloc;
1732         u64 delalloc_end;
1733
1734         WARN_ON(!PageLocked(page));
1735         if (page->index > end_index) {
1736                 clear_extent_dirty(tree, start, page_end, GFP_NOFS);
1737                 unlock_page(page);
1738                 return 0;
1739         }
1740
1741         if (page->index == end_index) {
1742                 size_t offset = i_size & (PAGE_CACHE_SIZE - 1);
1743                 zero_user_page(page, offset,
1744                                PAGE_CACHE_SIZE - offset, KM_USER0);
1745         }
1746
1747         set_page_extent_mapped(page);
1748
1749         delalloc_start = start;
1750         delalloc_end = 0;
1751         while(delalloc_end < page_end) {
1752                 nr_delalloc = find_lock_delalloc_range(tree, &delalloc_start,
1753                                                        &delalloc_end,
1754                                                        128 * 1024 * 1024);
1755                 if (nr_delalloc <= 0)
1756                         break;
1757                 tree->ops->fill_delalloc(inode, delalloc_start,
1758                                          delalloc_end);
1759                 clear_extent_bit(tree, delalloc_start,
1760                                  delalloc_end,
1761                                  EXTENT_LOCKED | EXTENT_DELALLOC,
1762                                  1, 0, GFP_NOFS);
1763                 delalloc_start = delalloc_end + 1;
1764         }
1765         lock_extent(tree, start, page_end, GFP_NOFS);
1766
1767         end = page_end;
1768         if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0)) {
1769                 printk("found delalloc bits after lock_extent\n");
1770         }
1771
1772         if (last_byte <= start) {
1773                 clear_extent_dirty(tree, start, page_end, GFP_NOFS);
1774                 goto done;
1775         }
1776
1777         set_extent_uptodate(tree, start, page_end, GFP_NOFS);
1778         blocksize = inode->i_sb->s_blocksize;
1779
1780         while (cur <= end) {
1781                 if (cur >= last_byte) {
1782                         clear_extent_dirty(tree, cur, page_end, GFP_NOFS);
1783                         break;
1784                 }
1785                 em = epd->get_extent(inode, page, page_offset, cur, end, 1);
1786                 if (IS_ERR(em) || !em) {
1787                         SetPageError(page);
1788                         break;
1789                 }
1790
1791                 extent_offset = cur - em->start;
1792                 BUG_ON(em->end < cur);
1793                 BUG_ON(end < cur);
1794                 iosize = min(em->end - cur, end - cur) + 1;
1795                 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
1796                 sector = (em->block_start + extent_offset) >> 9;
1797                 bdev = em->bdev;
1798                 block_start = em->block_start;
1799                 free_extent_map(em);
1800                 em = NULL;
1801
1802                 if (block_start == EXTENT_MAP_HOLE ||
1803                     block_start == EXTENT_MAP_INLINE) {
1804                         clear_extent_dirty(tree, cur,
1805                                            cur + iosize - 1, GFP_NOFS);
1806                         cur = cur + iosize;
1807                         page_offset += iosize;
1808                         continue;
1809                 }
1810
1811                 /* leave this out until we have a page_mkwrite call */
1812                 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
1813                                    EXTENT_DIRTY, 0)) {
1814                         cur = cur + iosize;
1815                         page_offset += iosize;
1816                         continue;
1817                 }
1818                 clear_extent_dirty(tree, cur, cur + iosize - 1, GFP_NOFS);
1819                 if (tree->ops && tree->ops->writepage_io_hook) {
1820                         ret = tree->ops->writepage_io_hook(page, cur,
1821                                                 cur + iosize - 1);
1822                 } else {
1823                         ret = 0;
1824                 }
1825                 if (ret)
1826                         SetPageError(page);
1827                 else {
1828                         unsigned long max_nr = end_index + 1;
1829                         set_range_writeback(tree, cur, cur + iosize - 1);
1830                         if (!PageWriteback(page)) {
1831                                 printk("warning page %lu not writeback, "
1832                                        "cur %llu end %llu\n", page->index,
1833                                        (unsigned long long)cur,
1834                                        (unsigned long long)end);
1835                         }
1836
1837                         ret = submit_extent_page(WRITE, tree, page, sector,
1838                                                  iosize, page_offset, bdev,
1839                                                  &epd->bio, max_nr,
1840                                                  end_bio_extent_writepage);
1841                         if (ret)
1842                                 SetPageError(page);
1843                 }
1844                 cur = cur + iosize;
1845                 page_offset += iosize;
1846                 nr++;
1847         }
1848 done:
1849         if (nr == 0) {
1850                 /* make sure the mapping tag for page dirty gets cleared */
1851                 set_page_writeback(page);
1852                 end_page_writeback(page);
1853         }
1854         unlock_extent(tree, start, page_end, GFP_NOFS);
1855         unlock_page(page);
1856         return 0;
1857 }
1858
1859 int extent_write_full_page(struct extent_map_tree *tree, struct page *page,
1860                           get_extent_t *get_extent,
1861                           struct writeback_control *wbc)
1862 {
1863         int ret;
1864         struct address_space *mapping = page->mapping;
1865         struct extent_page_data epd = {
1866                 .bio = NULL,
1867                 .tree = tree,
1868                 .get_extent = get_extent,
1869         };
1870         struct writeback_control wbc_writepages = {
1871                 .bdi            = wbc->bdi,
1872                 .sync_mode      = WB_SYNC_NONE,
1873                 .older_than_this = NULL,
1874                 .nr_to_write    = 64,
1875                 .range_start    = page_offset(page) + PAGE_CACHE_SIZE,
1876                 .range_end      = (loff_t)-1,
1877         };
1878
1879
1880         ret = __extent_writepage(page, wbc, &epd);
1881
1882         write_cache_pages(mapping, &wbc_writepages, __extent_writepage, &epd);
1883         if (epd.bio)
1884                 submit_one_bio(WRITE, epd.bio);
1885         return ret;
1886 }
1887 EXPORT_SYMBOL(extent_write_full_page);
1888
1889 int extent_writepages(struct extent_map_tree *tree,
1890                       struct address_space *mapping,
1891                       get_extent_t *get_extent,
1892                       struct writeback_control *wbc)
1893 {
1894         int ret;
1895         struct extent_page_data epd = {
1896                 .bio = NULL,
1897                 .tree = tree,
1898                 .get_extent = get_extent,
1899         };
1900
1901         ret = write_cache_pages(mapping, wbc, __extent_writepage, &epd);
1902         if (epd.bio)
1903                 submit_one_bio(WRITE, epd.bio);
1904         return ret;
1905 }
1906 EXPORT_SYMBOL(extent_writepages);
1907
1908 int extent_readpages(struct extent_map_tree *tree,
1909                      struct address_space *mapping,
1910                      struct list_head *pages, unsigned nr_pages,
1911                      get_extent_t get_extent)
1912 {
1913         struct bio *bio = NULL;
1914         unsigned page_idx;
1915         struct pagevec pvec;
1916
1917         pagevec_init(&pvec, 0);
1918         for (page_idx = 0; page_idx < nr_pages; page_idx++) {
1919                 struct page *page = list_entry(pages->prev, struct page, lru);
1920
1921                 prefetchw(&page->flags);
1922                 list_del(&page->lru);
1923                 /*
1924                  * what we want to do here is call add_to_page_cache_lru,
1925                  * but that isn't exported, so we reproduce it here
1926                  */
1927                 if (!add_to_page_cache(page, mapping,
1928                                         page->index, GFP_KERNEL)) {
1929
1930                         /* open coding of lru_cache_add, also not exported */
1931                         page_cache_get(page);
1932                         if (!pagevec_add(&pvec, page))
1933                                 __pagevec_lru_add(&pvec);
1934                         __extent_read_full_page(tree, page, get_extent, &bio);
1935                 }
1936                 page_cache_release(page);
1937         }
1938         if (pagevec_count(&pvec))
1939                 __pagevec_lru_add(&pvec);
1940         BUG_ON(!list_empty(pages));
1941         if (bio)
1942                 submit_one_bio(READ, bio);
1943         return 0;
1944 }
1945 EXPORT_SYMBOL(extent_readpages);
1946
1947 /*
1948  * basic invalidatepage code, this waits on any locked or writeback
1949  * ranges corresponding to the page, and then deletes any extent state
1950  * records from the tree
1951  */
1952 int extent_invalidatepage(struct extent_map_tree *tree,
1953                           struct page *page, unsigned long offset)
1954 {
1955         u64 start = ((u64)page->index << PAGE_CACHE_SHIFT);
1956         u64 end = start + PAGE_CACHE_SIZE - 1;
1957         size_t blocksize = page->mapping->host->i_sb->s_blocksize;
1958
1959         start += (offset + blocksize -1) & ~(blocksize - 1);
1960         if (start > end)
1961                 return 0;
1962
1963         lock_extent(tree, start, end, GFP_NOFS);
1964         wait_on_extent_writeback(tree, start, end);
1965         clear_extent_bit(tree, start, end,
1966                          EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC,
1967                          1, 1, GFP_NOFS);
1968         return 0;
1969 }
1970 EXPORT_SYMBOL(extent_invalidatepage);
1971
1972 /*
1973  * simple commit_write call, set_range_dirty is used to mark both
1974  * the pages and the extent records as dirty
1975  */
1976 int extent_commit_write(struct extent_map_tree *tree,
1977                         struct inode *inode, struct page *page,
1978                         unsigned from, unsigned to)
1979 {
1980         loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1981
1982         set_page_extent_mapped(page);
1983         set_page_dirty(page);
1984
1985         if (pos > inode->i_size) {
1986                 i_size_write(inode, pos);
1987                 mark_inode_dirty(inode);
1988         }
1989         return 0;
1990 }
1991 EXPORT_SYMBOL(extent_commit_write);
1992
1993 int extent_prepare_write(struct extent_map_tree *tree,
1994                          struct inode *inode, struct page *page,
1995                          unsigned from, unsigned to, get_extent_t *get_extent)
1996 {
1997         u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
1998         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
1999         u64 block_start;
2000         u64 orig_block_start;
2001         u64 block_end;
2002         u64 cur_end;
2003         struct extent_map *em;
2004         unsigned blocksize = 1 << inode->i_blkbits;
2005         size_t page_offset = 0;
2006         size_t block_off_start;
2007         size_t block_off_end;
2008         int err = 0;
2009         int iocount = 0;
2010         int ret = 0;
2011         int isnew;
2012
2013         set_page_extent_mapped(page);
2014
2015         block_start = (page_start + from) & ~((u64)blocksize - 1);
2016         block_end = (page_start + to - 1) | (blocksize - 1);
2017         orig_block_start = block_start;
2018
2019         lock_extent(tree, page_start, page_end, GFP_NOFS);
2020         while(block_start <= block_end) {
2021                 em = get_extent(inode, page, page_offset, block_start,
2022                                 block_end, 1);
2023                 if (IS_ERR(em) || !em) {
2024                         goto err;
2025                 }
2026                 cur_end = min(block_end, em->end);
2027                 block_off_start = block_start & (PAGE_CACHE_SIZE - 1);
2028                 block_off_end = block_off_start + blocksize;
2029                 isnew = clear_extent_new(tree, block_start, cur_end, GFP_NOFS);
2030
2031                 if (!PageUptodate(page) && isnew &&
2032                     (block_off_end > to || block_off_start < from)) {
2033                         void *kaddr;
2034
2035                         kaddr = kmap_atomic(page, KM_USER0);
2036                         if (block_off_end > to)
2037                                 memset(kaddr + to, 0, block_off_end - to);
2038                         if (block_off_start < from)
2039                                 memset(kaddr + block_off_start, 0,
2040                                        from - block_off_start);
2041                         flush_dcache_page(page);
2042                         kunmap_atomic(kaddr, KM_USER0);
2043                 }
2044                 if (!isnew && !PageUptodate(page) &&
2045                     (block_off_end > to || block_off_start < from) &&
2046                     !test_range_bit(tree, block_start, cur_end,
2047                                     EXTENT_UPTODATE, 1)) {
2048                         u64 sector;
2049                         u64 extent_offset = block_start - em->start;
2050                         size_t iosize;
2051                         sector = (em->block_start + extent_offset) >> 9;
2052                         iosize = (cur_end - block_start + blocksize - 1) &
2053                                 ~((u64)blocksize - 1);
2054                         /*
2055                          * we've already got the extent locked, but we
2056                          * need to split the state such that our end_bio
2057                          * handler can clear the lock.
2058                          */
2059                         set_extent_bit(tree, block_start,
2060                                        block_start + iosize - 1,
2061                                        EXTENT_LOCKED, 0, NULL, GFP_NOFS);
2062                         ret = submit_extent_page(READ, tree, page,
2063                                          sector, iosize, page_offset, em->bdev,
2064                                          NULL, 1,
2065                                          end_bio_extent_preparewrite);
2066                         iocount++;
2067                         block_start = block_start + iosize;
2068                 } else {
2069                         set_extent_uptodate(tree, block_start, cur_end,
2070                                             GFP_NOFS);
2071                         unlock_extent(tree, block_start, cur_end, GFP_NOFS);
2072                         block_start = cur_end + 1;
2073                 }
2074                 page_offset = block_start & (PAGE_CACHE_SIZE - 1);
2075                 free_extent_map(em);
2076         }
2077         if (iocount) {
2078                 wait_extent_bit(tree, orig_block_start,
2079                                 block_end, EXTENT_LOCKED);
2080         }
2081         check_page_uptodate(tree, page);
2082 err:
2083         /* FIXME, zero out newly allocated blocks on error */
2084         return err;
2085 }
2086 EXPORT_SYMBOL(extent_prepare_write);
2087
2088 /*
2089  * a helper for releasepage.  As long as there are no locked extents
2090  * in the range corresponding to the page, both state records and extent
2091  * map records are removed
2092  */
2093 int try_release_extent_mapping(struct extent_map_tree *tree, struct page *page)
2094 {
2095         struct extent_map *em;
2096         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2097         u64 end = start + PAGE_CACHE_SIZE - 1;
2098         u64 orig_start = start;
2099         int ret = 1;
2100
2101         while (start <= end) {
2102                 em = lookup_extent_mapping(tree, start, end);
2103                 if (!em || IS_ERR(em))
2104                         break;
2105                 if (!test_range_bit(tree, em->start, em->end,
2106                                     EXTENT_LOCKED, 0)) {
2107                         remove_extent_mapping(tree, em);
2108                         /* once for the rb tree */
2109                         free_extent_map(em);
2110                 }
2111                 start = em->end + 1;
2112                 /* once for us */
2113                 free_extent_map(em);
2114         }
2115         if (test_range_bit(tree, orig_start, end, EXTENT_LOCKED, 0))
2116                 ret = 0;
2117         else
2118                 clear_extent_bit(tree, orig_start, end, EXTENT_UPTODATE,
2119                                  1, 1, GFP_NOFS);
2120         return ret;
2121 }
2122 EXPORT_SYMBOL(try_release_extent_mapping);
2123
2124 sector_t extent_bmap(struct address_space *mapping, sector_t iblock,
2125                 get_extent_t *get_extent)
2126 {
2127         struct inode *inode = mapping->host;
2128         u64 start = iblock << inode->i_blkbits;
2129         u64 end = start + (1 << inode->i_blkbits) - 1;
2130         sector_t sector = 0;
2131         struct extent_map *em;
2132
2133         em = get_extent(inode, NULL, 0, start, end, 0);
2134         if (!em || IS_ERR(em))
2135                 return 0;
2136
2137         if (em->block_start == EXTENT_MAP_INLINE ||
2138             em->block_start == EXTENT_MAP_HOLE)
2139                 goto out;
2140
2141         sector = (em->block_start + start - em->start) >> inode->i_blkbits;
2142 out:
2143         free_extent_map(em);
2144         return sector;
2145 }
2146
2147 static int add_lru(struct extent_map_tree *tree, struct extent_buffer *eb)
2148 {
2149         if (list_empty(&eb->lru)) {
2150                 extent_buffer_get(eb);
2151                 list_add(&eb->lru, &tree->buffer_lru);
2152                 tree->lru_size++;
2153                 if (tree->lru_size >= BUFFER_LRU_MAX) {
2154                         struct extent_buffer *rm;
2155                         rm = list_entry(tree->buffer_lru.prev,
2156                                         struct extent_buffer, lru);
2157                         tree->lru_size--;
2158                         list_del_init(&rm->lru);
2159                         free_extent_buffer(rm);
2160                 }
2161         } else
2162                 list_move(&eb->lru, &tree->buffer_lru);
2163         return 0;
2164 }
2165 static struct extent_buffer *find_lru(struct extent_map_tree *tree,
2166                                       u64 start, unsigned long len)
2167 {
2168         struct list_head *lru = &tree->buffer_lru;
2169         struct list_head *cur = lru->next;
2170         struct extent_buffer *eb;
2171
2172         if (list_empty(lru))
2173                 return NULL;
2174
2175         do {
2176                 eb = list_entry(cur, struct extent_buffer, lru);
2177                 if (eb->start == start && eb->len == len) {
2178                         extent_buffer_get(eb);
2179                         return eb;
2180                 }
2181                 cur = cur->next;
2182         } while (cur != lru);
2183         return NULL;
2184 }
2185
2186 static inline unsigned long num_extent_pages(u64 start, u64 len)
2187 {
2188         return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
2189                 (start >> PAGE_CACHE_SHIFT);
2190 }
2191
2192 static inline struct page *extent_buffer_page(struct extent_buffer *eb,
2193                                               unsigned long i)
2194 {
2195         struct page *p;
2196         struct address_space *mapping;
2197
2198         if (i == 0)
2199                 return eb->first_page;
2200         i += eb->start >> PAGE_CACHE_SHIFT;
2201         mapping = eb->first_page->mapping;
2202         read_lock_irq(&mapping->tree_lock);
2203         p = radix_tree_lookup(&mapping->page_tree, i);
2204         read_unlock_irq(&mapping->tree_lock);
2205         return p;
2206 }
2207
2208 static struct extent_buffer *__alloc_extent_buffer(struct extent_map_tree *tree,
2209                                                    u64 start,
2210                                                    unsigned long len,
2211                                                    gfp_t mask)
2212 {
2213         struct extent_buffer *eb = NULL;
2214
2215         spin_lock(&tree->lru_lock);
2216         eb = find_lru(tree, start, len);
2217         spin_unlock(&tree->lru_lock);
2218         if (eb) {
2219                 return eb;
2220         }
2221
2222         eb = kmem_cache_zalloc(extent_buffer_cache, mask);
2223         INIT_LIST_HEAD(&eb->lru);
2224         eb->start = start;
2225         eb->len = len;
2226         atomic_set(&eb->refs, 1);
2227
2228         return eb;
2229 }
2230
2231 static void __free_extent_buffer(struct extent_buffer *eb)
2232 {
2233         kmem_cache_free(extent_buffer_cache, eb);
2234 }
2235
2236 struct extent_buffer *alloc_extent_buffer(struct extent_map_tree *tree,
2237                                           u64 start, unsigned long len,
2238                                           struct page *page0,
2239                                           gfp_t mask)
2240 {
2241         unsigned long num_pages = num_extent_pages(start, len);
2242         unsigned long i;
2243         unsigned long index = start >> PAGE_CACHE_SHIFT;
2244         struct extent_buffer *eb;
2245         struct page *p;
2246         struct address_space *mapping = tree->mapping;
2247         int uptodate = 1;
2248
2249         eb = __alloc_extent_buffer(tree, start, len, mask);
2250         if (!eb || IS_ERR(eb))
2251                 return NULL;
2252
2253         if (eb->flags & EXTENT_BUFFER_FILLED)
2254                 goto lru_add;
2255
2256         if (page0) {
2257                 eb->first_page = page0;
2258                 i = 1;
2259                 index++;
2260                 page_cache_get(page0);
2261                 mark_page_accessed(page0);
2262                 set_page_extent_mapped(page0);
2263                 WARN_ON(!PageUptodate(page0));
2264                 set_page_private(page0, EXTENT_PAGE_PRIVATE_FIRST_PAGE |
2265                                  len << 2);
2266         } else {
2267                 i = 0;
2268         }
2269         for (; i < num_pages; i++, index++) {
2270                 p = find_or_create_page(mapping, index, mask | __GFP_HIGHMEM);
2271                 if (!p) {
2272                         WARN_ON(1);
2273                         goto fail;
2274                 }
2275                 set_page_extent_mapped(p);
2276                 mark_page_accessed(p);
2277                 if (i == 0) {
2278                         eb->first_page = p;
2279                         set_page_private(p, EXTENT_PAGE_PRIVATE_FIRST_PAGE |
2280                                          len << 2);
2281                 } else {
2282                         set_page_private(p, EXTENT_PAGE_PRIVATE);
2283                 }
2284                 if (!PageUptodate(p))
2285                         uptodate = 0;
2286                 unlock_page(p);
2287         }
2288         if (uptodate)
2289                 eb->flags |= EXTENT_UPTODATE;
2290         eb->flags |= EXTENT_BUFFER_FILLED;
2291
2292 lru_add:
2293         spin_lock(&tree->lru_lock);
2294         add_lru(tree, eb);
2295         spin_unlock(&tree->lru_lock);
2296         return eb;
2297
2298 fail:
2299         spin_lock(&tree->lru_lock);
2300         list_del_init(&eb->lru);
2301         spin_unlock(&tree->lru_lock);
2302         if (!atomic_dec_and_test(&eb->refs))
2303                 return NULL;
2304         for (index = 1; index < i; index++) {
2305                 page_cache_release(extent_buffer_page(eb, index));
2306         }
2307         if (i > 0)
2308                 page_cache_release(extent_buffer_page(eb, 0));
2309         __free_extent_buffer(eb);
2310         return NULL;
2311 }
2312 EXPORT_SYMBOL(alloc_extent_buffer);
2313
2314 struct extent_buffer *find_extent_buffer(struct extent_map_tree *tree,
2315                                          u64 start, unsigned long len,
2316                                           gfp_t mask)
2317 {
2318         unsigned long num_pages = num_extent_pages(start, len);
2319         unsigned long i;
2320         unsigned long index = start >> PAGE_CACHE_SHIFT;
2321         struct extent_buffer *eb;
2322         struct page *p;
2323         struct address_space *mapping = tree->mapping;
2324         int uptodate = 1;
2325
2326         eb = __alloc_extent_buffer(tree, start, len, mask);
2327         if (!eb || IS_ERR(eb))
2328                 return NULL;
2329
2330         if (eb->flags & EXTENT_BUFFER_FILLED)
2331                 goto lru_add;
2332
2333         for (i = 0; i < num_pages; i++, index++) {
2334                 p = find_lock_page(mapping, index);
2335                 if (!p) {
2336                         goto fail;
2337                 }
2338                 set_page_extent_mapped(p);
2339                 mark_page_accessed(p);
2340
2341                 if (i == 0) {
2342                         eb->first_page = p;
2343                         set_page_private(p, EXTENT_PAGE_PRIVATE_FIRST_PAGE |
2344                                          len << 2);
2345                 } else {
2346                         set_page_private(p, EXTENT_PAGE_PRIVATE);
2347                 }
2348
2349                 if (!PageUptodate(p))
2350                         uptodate = 0;
2351                 unlock_page(p);
2352         }
2353         if (uptodate)
2354                 eb->flags |= EXTENT_UPTODATE;
2355         eb->flags |= EXTENT_BUFFER_FILLED;
2356
2357 lru_add:
2358         spin_lock(&tree->lru_lock);
2359         add_lru(tree, eb);
2360         spin_unlock(&tree->lru_lock);
2361         return eb;
2362 fail:
2363         spin_lock(&tree->lru_lock);
2364         list_del_init(&eb->lru);
2365         spin_unlock(&tree->lru_lock);
2366         if (!atomic_dec_and_test(&eb->refs))
2367                 return NULL;
2368         for (index = 1; index < i; index++) {
2369                 page_cache_release(extent_buffer_page(eb, index));
2370         }
2371         if (i > 0)
2372                 page_cache_release(extent_buffer_page(eb, 0));
2373         __free_extent_buffer(eb);
2374         return NULL;
2375 }
2376 EXPORT_SYMBOL(find_extent_buffer);
2377
2378 void free_extent_buffer(struct extent_buffer *eb)
2379 {
2380         unsigned long i;
2381         unsigned long num_pages;
2382
2383         if (!eb)
2384                 return;
2385
2386         if (!atomic_dec_and_test(&eb->refs))
2387                 return;
2388
2389         WARN_ON(!list_empty(&eb->lru));
2390         num_pages = num_extent_pages(eb->start, eb->len);
2391
2392         for (i = 1; i < num_pages; i++) {
2393                 page_cache_release(extent_buffer_page(eb, i));
2394         }
2395         page_cache_release(extent_buffer_page(eb, 0));
2396         __free_extent_buffer(eb);
2397 }
2398 EXPORT_SYMBOL(free_extent_buffer);
2399
2400 int clear_extent_buffer_dirty(struct extent_map_tree *tree,
2401                               struct extent_buffer *eb)
2402 {
2403         int set;
2404         unsigned long i;
2405         unsigned long num_pages;
2406         struct page *page;
2407
2408         u64 start = eb->start;
2409         u64 end = start + eb->len - 1;
2410
2411         set = clear_extent_dirty(tree, start, end, GFP_NOFS);
2412         num_pages = num_extent_pages(eb->start, eb->len);
2413
2414         for (i = 0; i < num_pages; i++) {
2415                 page = extent_buffer_page(eb, i);
2416                 lock_page(page);
2417                 /*
2418                  * if we're on the last page or the first page and the
2419                  * block isn't aligned on a page boundary, do extra checks
2420                  * to make sure we don't clean page that is partially dirty
2421                  */
2422                 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
2423                     ((i == num_pages - 1) &&
2424                      ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
2425                         start = (u64)page->index << PAGE_CACHE_SHIFT;
2426                         end  = start + PAGE_CACHE_SIZE - 1;
2427                         if (test_range_bit(tree, start, end,
2428                                            EXTENT_DIRTY, 0)) {
2429                                 unlock_page(page);
2430                                 continue;
2431                         }
2432                 }
2433                 clear_page_dirty_for_io(page);
2434                 write_lock_irq(&page->mapping->tree_lock);
2435                 if (!PageDirty(page)) {
2436                         radix_tree_tag_clear(&page->mapping->page_tree,
2437                                                 page_index(page),
2438                                                 PAGECACHE_TAG_DIRTY);
2439                 }
2440                 write_unlock_irq(&page->mapping->tree_lock);
2441                 unlock_page(page);
2442         }
2443         return 0;
2444 }
2445 EXPORT_SYMBOL(clear_extent_buffer_dirty);
2446
2447 int wait_on_extent_buffer_writeback(struct extent_map_tree *tree,
2448                                     struct extent_buffer *eb)
2449 {
2450         return wait_on_extent_writeback(tree, eb->start,
2451                                         eb->start + eb->len - 1);
2452 }
2453 EXPORT_SYMBOL(wait_on_extent_buffer_writeback);
2454
2455 int set_extent_buffer_dirty(struct extent_map_tree *tree,
2456                              struct extent_buffer *eb)
2457 {
2458         unsigned long i;
2459         unsigned long num_pages;
2460
2461         num_pages = num_extent_pages(eb->start, eb->len);
2462         for (i = 0; i < num_pages; i++) {
2463                 struct page *page = extent_buffer_page(eb, i);
2464                 /* writepage may need to do something special for the
2465                  * first page, we have to make sure page->private is
2466                  * properly set.  releasepage may drop page->private
2467                  * on us if the page isn't already dirty.
2468                  */
2469                 if (i == 0) {
2470                         lock_page(page);
2471                         set_page_private(page,
2472                                          EXTENT_PAGE_PRIVATE_FIRST_PAGE |
2473                                          eb->len << 2);
2474                 }
2475                 __set_page_dirty_nobuffers(extent_buffer_page(eb, i));
2476                 if (i == 0)
2477                         unlock_page(page);
2478         }
2479         return set_extent_dirty(tree, eb->start,
2480                                 eb->start + eb->len - 1, GFP_NOFS);
2481 }
2482 EXPORT_SYMBOL(set_extent_buffer_dirty);
2483
2484 int set_extent_buffer_uptodate(struct extent_map_tree *tree,
2485                                 struct extent_buffer *eb)
2486 {
2487         unsigned long i;
2488         struct page *page;
2489         unsigned long num_pages;
2490
2491         num_pages = num_extent_pages(eb->start, eb->len);
2492
2493         set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
2494                             GFP_NOFS);
2495         for (i = 0; i < num_pages; i++) {
2496                 page = extent_buffer_page(eb, i);
2497                 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
2498                     ((i == num_pages - 1) &&
2499                      ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
2500                         check_page_uptodate(tree, page);
2501                         continue;
2502                 }
2503                 SetPageUptodate(page);
2504         }
2505         return 0;
2506 }
2507 EXPORT_SYMBOL(set_extent_buffer_uptodate);
2508
2509 int extent_buffer_uptodate(struct extent_map_tree *tree,
2510                              struct extent_buffer *eb)
2511 {
2512         if (eb->flags & EXTENT_UPTODATE)
2513                 return 1;
2514         return test_range_bit(tree, eb->start, eb->start + eb->len - 1,
2515                            EXTENT_UPTODATE, 1);
2516 }
2517 EXPORT_SYMBOL(extent_buffer_uptodate);
2518
2519 int read_extent_buffer_pages(struct extent_map_tree *tree,
2520                              struct extent_buffer *eb,
2521                              u64 start,
2522                              int wait)
2523 {
2524         unsigned long i;
2525         unsigned long start_i;
2526         struct page *page;
2527         int err;
2528         int ret = 0;
2529         unsigned long num_pages;
2530
2531         if (eb->flags & EXTENT_UPTODATE)
2532                 return 0;
2533
2534         if (0 && test_range_bit(tree, eb->start, eb->start + eb->len - 1,
2535                            EXTENT_UPTODATE, 1)) {
2536                 return 0;
2537         }
2538
2539         if (start) {
2540                 WARN_ON(start < eb->start);
2541                 start_i = (start >> PAGE_CACHE_SHIFT) -
2542                         (eb->start >> PAGE_CACHE_SHIFT);
2543         } else {
2544                 start_i = 0;
2545         }
2546
2547         num_pages = num_extent_pages(eb->start, eb->len);
2548         for (i = start_i; i < num_pages; i++) {
2549                 page = extent_buffer_page(eb, i);
2550                 if (PageUptodate(page)) {
2551                         continue;
2552                 }
2553                 if (!wait) {
2554                         if (TestSetPageLocked(page)) {
2555                                 continue;
2556                         }
2557                 } else {
2558                         lock_page(page);
2559                 }
2560                 if (!PageUptodate(page)) {
2561                         err = page->mapping->a_ops->readpage(NULL, page);
2562                         if (err) {
2563                                 ret = err;
2564                         }
2565                 } else {
2566                         unlock_page(page);
2567                 }
2568         }
2569
2570         if (ret || !wait) {
2571                 return ret;
2572         }
2573
2574         for (i = start_i; i < num_pages; i++) {
2575                 page = extent_buffer_page(eb, i);
2576                 wait_on_page_locked(page);
2577                 if (!PageUptodate(page)) {
2578                         ret = -EIO;
2579                 }
2580         }
2581         if (!ret)
2582                 eb->flags |= EXTENT_UPTODATE;
2583         return ret;
2584 }
2585 EXPORT_SYMBOL(read_extent_buffer_pages);
2586
2587 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
2588                         unsigned long start,
2589                         unsigned long len)
2590 {
2591         size_t cur;
2592         size_t offset;
2593         struct page *page;
2594         char *kaddr;
2595         char *dst = (char *)dstv;
2596         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
2597         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
2598         unsigned long num_pages = num_extent_pages(eb->start, eb->len);
2599
2600         WARN_ON(start > eb->len);
2601         WARN_ON(start + len > eb->start + eb->len);
2602
2603         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
2604
2605         while(len > 0) {
2606                 page = extent_buffer_page(eb, i);
2607                 if (!PageUptodate(page)) {
2608                         printk("page %lu not up to date i %lu, total %lu, len %lu\n", page->index, i, num_pages, eb->len);
2609                         WARN_ON(1);
2610                 }
2611                 WARN_ON(!PageUptodate(page));
2612
2613                 cur = min(len, (PAGE_CACHE_SIZE - offset));
2614                 kaddr = kmap_atomic(page, KM_USER1);
2615                 memcpy(dst, kaddr + offset, cur);
2616                 kunmap_atomic(kaddr, KM_USER1);
2617
2618                 dst += cur;
2619                 len -= cur;
2620                 offset = 0;
2621                 i++;
2622         }
2623 }
2624 EXPORT_SYMBOL(read_extent_buffer);
2625
2626 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
2627                                unsigned long min_len, char **token, char **map,
2628                                unsigned long *map_start,
2629                                unsigned long *map_len, int km)
2630 {
2631         size_t offset = start & (PAGE_CACHE_SIZE - 1);
2632         char *kaddr;
2633         struct page *p;
2634         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
2635         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
2636         unsigned long end_i = (start_offset + start + min_len - 1) >>
2637                 PAGE_CACHE_SHIFT;
2638
2639         if (i != end_i)
2640                 return -EINVAL;
2641
2642         if (i == 0) {
2643                 offset = start_offset;
2644                 *map_start = 0;
2645         } else {
2646                 offset = 0;
2647                 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
2648         }
2649         if (start + min_len > eb->len) {
2650 printk("bad mapping eb start %Lu len %lu, wanted %lu %lu\n", eb->start, eb->len, start, min_len);
2651                 WARN_ON(1);
2652         }
2653
2654         p = extent_buffer_page(eb, i);
2655         WARN_ON(!PageUptodate(p));
2656         kaddr = kmap_atomic(p, km);
2657         *token = kaddr;
2658         *map = kaddr + offset;
2659         *map_len = PAGE_CACHE_SIZE - offset;
2660         return 0;
2661 }
2662 EXPORT_SYMBOL(map_private_extent_buffer);
2663
2664 int map_extent_buffer(struct extent_buffer *eb, unsigned long start,
2665                       unsigned long min_len,
2666                       char **token, char **map,
2667                       unsigned long *map_start,
2668                       unsigned long *map_len, int km)
2669 {
2670         int err;
2671         int save = 0;
2672         if (eb->map_token) {
2673                 unmap_extent_buffer(eb, eb->map_token, km);
2674                 eb->map_token = NULL;
2675                 save = 1;
2676         }
2677         err = map_private_extent_buffer(eb, start, min_len, token, map,
2678                                        map_start, map_len, km);
2679         if (!err && save) {
2680                 eb->map_token = *token;
2681                 eb->kaddr = *map;
2682                 eb->map_start = *map_start;
2683                 eb->map_len = *map_len;
2684         }
2685         return err;
2686 }
2687 EXPORT_SYMBOL(map_extent_buffer);
2688
2689 void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km)
2690 {
2691         kunmap_atomic(token, km);
2692 }
2693 EXPORT_SYMBOL(unmap_extent_buffer);
2694
2695 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
2696                           unsigned long start,
2697                           unsigned long len)
2698 {
2699         size_t cur;
2700         size_t offset;
2701         struct page *page;
2702         char *kaddr;
2703         char *ptr = (char *)ptrv;
2704         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
2705         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
2706         int ret = 0;
2707
2708         WARN_ON(start > eb->len);
2709         WARN_ON(start + len > eb->start + eb->len);
2710
2711         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
2712
2713         while(len > 0) {
2714                 page = extent_buffer_page(eb, i);
2715                 WARN_ON(!PageUptodate(page));
2716
2717                 cur = min(len, (PAGE_CACHE_SIZE - offset));
2718
2719                 kaddr = kmap_atomic(page, KM_USER0);
2720                 ret = memcmp(ptr, kaddr + offset, cur);
2721                 kunmap_atomic(kaddr, KM_USER0);
2722                 if (ret)
2723                         break;
2724
2725                 ptr += cur;
2726                 len -= cur;
2727                 offset = 0;
2728                 i++;
2729         }
2730         return ret;
2731 }
2732 EXPORT_SYMBOL(memcmp_extent_buffer);
2733
2734 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
2735                          unsigned long start, unsigned long len)
2736 {
2737         size_t cur;
2738         size_t offset;
2739         struct page *page;
2740         char *kaddr;
2741         char *src = (char *)srcv;
2742         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
2743         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
2744
2745         WARN_ON(start > eb->len);
2746         WARN_ON(start + len > eb->start + eb->len);
2747
2748         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
2749
2750         while(len > 0) {
2751                 page = extent_buffer_page(eb, i);
2752                 WARN_ON(!PageUptodate(page));
2753
2754                 cur = min(len, PAGE_CACHE_SIZE - offset);
2755                 kaddr = kmap_atomic(page, KM_USER1);
2756                 memcpy(kaddr + offset, src, cur);
2757                 kunmap_atomic(kaddr, KM_USER1);
2758
2759                 src += cur;
2760                 len -= cur;
2761                 offset = 0;
2762                 i++;
2763         }
2764 }
2765 EXPORT_SYMBOL(write_extent_buffer);
2766
2767 void memset_extent_buffer(struct extent_buffer *eb, char c,
2768                           unsigned long start, unsigned long len)
2769 {
2770         size_t cur;
2771         size_t offset;
2772         struct page *page;
2773         char *kaddr;
2774         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
2775         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
2776
2777         WARN_ON(start > eb->len);
2778         WARN_ON(start + len > eb->start + eb->len);
2779
2780         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
2781
2782         while(len > 0) {
2783                 page = extent_buffer_page(eb, i);
2784                 WARN_ON(!PageUptodate(page));
2785
2786                 cur = min(len, PAGE_CACHE_SIZE - offset);
2787                 kaddr = kmap_atomic(page, KM_USER0);
2788                 memset(kaddr + offset, c, cur);
2789                 kunmap_atomic(kaddr, KM_USER0);
2790
2791                 len -= cur;
2792                 offset = 0;
2793                 i++;
2794         }
2795 }
2796 EXPORT_SYMBOL(memset_extent_buffer);
2797
2798 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
2799                         unsigned long dst_offset, unsigned long src_offset,
2800                         unsigned long len)
2801 {
2802         u64 dst_len = dst->len;
2803         size_t cur;
2804         size_t offset;
2805         struct page *page;
2806         char *kaddr;
2807         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
2808         unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
2809
2810         WARN_ON(src->len != dst_len);
2811
2812         offset = (start_offset + dst_offset) &
2813                 ((unsigned long)PAGE_CACHE_SIZE - 1);
2814
2815         while(len > 0) {
2816                 page = extent_buffer_page(dst, i);
2817                 WARN_ON(!PageUptodate(page));
2818
2819                 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
2820
2821                 kaddr = kmap_atomic(page, KM_USER0);
2822                 read_extent_buffer(src, kaddr + offset, src_offset, cur);
2823                 kunmap_atomic(kaddr, KM_USER0);
2824
2825                 src_offset += cur;
2826                 len -= cur;
2827                 offset = 0;
2828                 i++;
2829         }
2830 }
2831 EXPORT_SYMBOL(copy_extent_buffer);
2832
2833 static void move_pages(struct page *dst_page, struct page *src_page,
2834                        unsigned long dst_off, unsigned long src_off,
2835                        unsigned long len)
2836 {
2837         char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
2838         if (dst_page == src_page) {
2839                 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
2840         } else {
2841                 char *src_kaddr = kmap_atomic(src_page, KM_USER1);
2842                 char *p = dst_kaddr + dst_off + len;
2843                 char *s = src_kaddr + src_off + len;
2844
2845                 while (len--)
2846                         *--p = *--s;
2847
2848                 kunmap_atomic(src_kaddr, KM_USER1);
2849         }
2850         kunmap_atomic(dst_kaddr, KM_USER0);
2851 }
2852
2853 static void copy_pages(struct page *dst_page, struct page *src_page,
2854                        unsigned long dst_off, unsigned long src_off,
2855                        unsigned long len)
2856 {
2857         char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
2858         char *src_kaddr;
2859
2860         if (dst_page != src_page)
2861                 src_kaddr = kmap_atomic(src_page, KM_USER1);
2862         else
2863                 src_kaddr = dst_kaddr;
2864
2865         memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
2866         kunmap_atomic(dst_kaddr, KM_USER0);
2867         if (dst_page != src_page)
2868                 kunmap_atomic(src_kaddr, KM_USER1);
2869 }
2870
2871 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
2872                            unsigned long src_offset, unsigned long len)
2873 {
2874         size_t cur;
2875         size_t dst_off_in_page;
2876         size_t src_off_in_page;
2877         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
2878         unsigned long dst_i;
2879         unsigned long src_i;
2880
2881         if (src_offset + len > dst->len) {
2882                 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
2883                        src_offset, len, dst->len);
2884                 BUG_ON(1);
2885         }
2886         if (dst_offset + len > dst->len) {
2887                 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
2888                        dst_offset, len, dst->len);
2889                 BUG_ON(1);
2890         }
2891
2892         while(len > 0) {
2893                 dst_off_in_page = (start_offset + dst_offset) &
2894                         ((unsigned long)PAGE_CACHE_SIZE - 1);
2895                 src_off_in_page = (start_offset + src_offset) &
2896                         ((unsigned long)PAGE_CACHE_SIZE - 1);
2897
2898                 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
2899                 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
2900
2901                 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
2902                                                src_off_in_page));
2903                 cur = min_t(unsigned long, cur,
2904                         (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
2905
2906                 copy_pages(extent_buffer_page(dst, dst_i),
2907                            extent_buffer_page(dst, src_i),
2908                            dst_off_in_page, src_off_in_page, cur);
2909
2910                 src_offset += cur;
2911                 dst_offset += cur;
2912                 len -= cur;
2913         }
2914 }
2915 EXPORT_SYMBOL(memcpy_extent_buffer);
2916
2917 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
2918                            unsigned long src_offset, unsigned long len)
2919 {
2920         size_t cur;
2921         size_t dst_off_in_page;
2922         size_t src_off_in_page;
2923         unsigned long dst_end = dst_offset + len - 1;
2924         unsigned long src_end = src_offset + len - 1;
2925         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
2926         unsigned long dst_i;
2927         unsigned long src_i;
2928
2929         if (src_offset + len > dst->len) {
2930                 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
2931                        src_offset, len, dst->len);
2932                 BUG_ON(1);
2933         }
2934         if (dst_offset + len > dst->len) {
2935                 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
2936                        dst_offset, len, dst->len);
2937                 BUG_ON(1);
2938         }
2939         if (dst_offset < src_offset) {
2940                 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
2941                 return;
2942         }
2943         while(len > 0) {
2944                 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
2945                 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
2946
2947                 dst_off_in_page = (start_offset + dst_end) &
2948                         ((unsigned long)PAGE_CACHE_SIZE - 1);
2949                 src_off_in_page = (start_offset + src_end) &
2950                         ((unsigned long)PAGE_CACHE_SIZE - 1);
2951
2952                 cur = min_t(unsigned long, len, src_off_in_page + 1);
2953                 cur = min(cur, dst_off_in_page + 1);
2954                 move_pages(extent_buffer_page(dst, dst_i),
2955                            extent_buffer_page(dst, src_i),
2956                            dst_off_in_page - cur + 1,
2957                            src_off_in_page - cur + 1, cur);
2958
2959                 dst_end -= cur;
2960                 src_end -= cur;
2961                 len -= cur;
2962         }
2963 }
2964 EXPORT_SYMBOL(memmove_extent_buffer);