git.oblomov.eu Git - linux-2.6/blob - fs/btrfs/extent_io.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_io.h"
  16 #include "extent_map.h"
  17
  18 /* temporary define until extent_map moves out of btrfs */
  19 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
  20                                        unsigned long extra_flags,
  21                                        void (*ctor)(void *, struct kmem_cache *,
  22                                                     unsigned long));
  23
  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 static spinlock_t leak_lock = SPIN_LOCK_UNLOCKED;
  30
  31 #define BUFFER_LRU_MAX 64
  32
  33 struct tree_entry {
  34         u64 start;
  35         u64 end;
  36         struct rb_node rb_node;
  37 };
  38
  39 struct extent_page_data {
  40         struct bio *bio;
  41         struct extent_io_tree *tree;
  42         get_extent_t *get_extent;
  43 };
  44
  45 int __init extent_io_init(void)
  46 {
  47         extent_state_cache = btrfs_cache_create("extent_state",
  48                                             sizeof(struct extent_state), 0,
  49                                             NULL);
  50         if (!extent_state_cache)
  51                 return -ENOMEM;
  52
  53         extent_buffer_cache = btrfs_cache_create("extent_buffers",
  54                                             sizeof(struct extent_buffer), 0,
  55                                             NULL);
  56         if (!extent_buffer_cache)
  57                 goto free_state_cache;
  58         return 0;
  59
  60 free_state_cache:
  61         kmem_cache_destroy(extent_state_cache);
  62         return -ENOMEM;
  63 }
  64
  65 void extent_io_exit(void)
  66 {
  67         struct extent_state *state;
  68         struct extent_buffer *eb;
  69
  70         while (!list_empty(&states)) {
  71                 state = list_entry(states.next, struct extent_state, leak_list);
  72                 printk("state leak: start %Lu end %Lu state %lu in tree %p refs %d\n", state->start, state->end, state->state, state->tree, atomic_read(&state->refs));
  73                 list_del(&state->leak_list);
  74                 kmem_cache_free(extent_state_cache, state);
  75
  76         }
  77
  78         while (!list_empty(&buffers)) {
  79                 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
  80                 printk("buffer leak start %Lu len %lu refs %d\n", eb->start, eb->len, atomic_read(&eb->refs));
  81                 list_del(&eb->leak_list);
  82                 kmem_cache_free(extent_buffer_cache, eb);
  83         }
  84         if (extent_state_cache)
  85                 kmem_cache_destroy(extent_state_cache);
  86         if (extent_buffer_cache)
  87                 kmem_cache_destroy(extent_buffer_cache);
  88 }
  89
  90 void extent_io_tree_init(struct extent_io_tree *tree,
  91                           struct address_space *mapping, gfp_t mask)
  92 {
  93         tree->state.rb_node = NULL;
  94         tree->ops = NULL;
  95         tree->dirty_bytes = 0;
  96         spin_lock_init(&tree->lock);
  97         spin_lock_init(&tree->lru_lock);
  98         tree->mapping = mapping;
  99         INIT_LIST_HEAD(&tree->buffer_lru);
 100         tree->lru_size = 0;
 101         tree->last = NULL;
 102 }
 103 EXPORT_SYMBOL(extent_io_tree_init);
 104
 105 void extent_io_tree_empty_lru(struct extent_io_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_io_tree_empty_lru);
 116
 117 struct extent_state *alloc_extent_state(gfp_t mask)
 118 {
 119         struct extent_state *state;
 120         unsigned long flags;
 121
 122         state = kmem_cache_alloc(extent_state_cache, mask);
 123         if (!state)
 124                 return state;
 125         state->state = 0;
 126         state->private = 0;
 127         state->tree = NULL;
 128         spin_lock_irqsave(&leak_lock, flags);
 129         list_add(&state->leak_list, &states);
 130         spin_unlock_irqrestore(&leak_lock, flags);
 131
 132         atomic_set(&state->refs, 1);
 133         init_waitqueue_head(&state->wq);
 134         return state;
 135 }
 136 EXPORT_SYMBOL(alloc_extent_state);
 137
 138 void free_extent_state(struct extent_state *state)
 139 {
 140         if (!state)
 141                 return;
 142         if (atomic_dec_and_test(&state->refs)) {
 143                 unsigned long flags;
 144                 WARN_ON(state->tree);
 145                 spin_lock_irqsave(&leak_lock, flags);
 146                 list_del(&state->leak_list);
 147                 spin_unlock_irqrestore(&leak_lock, flags);
 148                 kmem_cache_free(extent_state_cache, state);
 149         }
 150 }
 151 EXPORT_SYMBOL(free_extent_state);
 152
 153 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
 154                                    struct rb_node *node)
 155 {
 156         struct rb_node ** p = &root->rb_node;
 157         struct rb_node * parent = NULL;
 158         struct tree_entry *entry;
 159
 160         while(*p) {
 161                 parent = *p;
 162                 entry = rb_entry(parent, struct tree_entry, rb_node);
 163
 164                 if (offset < entry->start)
 165                         p = &(*p)->rb_left;
 166                 else if (offset > entry->end)
 167                         p = &(*p)->rb_right;
 168                 else
 169                         return parent;
 170         }
 171
 172         entry = rb_entry(node, struct tree_entry, rb_node);
 173         rb_link_node(node, parent, p);
 174         rb_insert_color(node, root);
 175         return NULL;
 176 }
 177
 178 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
 179                                      struct rb_node **prev_ret,
 180                                      struct rb_node **next_ret)
 181 {
 182         struct rb_root *root = &tree->state;
 183         struct rb_node * n = root->rb_node;
 184         struct rb_node *prev = NULL;
 185         struct rb_node *orig_prev = NULL;
 186         struct tree_entry *entry;
 187         struct tree_entry *prev_entry = NULL;
 188
 189         if (tree->last) {
 190                 struct extent_state *state;
 191                 state = tree->last;
 192                 if (state->start <= offset && offset <= state->end)
 193                         return &tree->last->rb_node;
 194         }
 195         while(n) {
 196                 entry = rb_entry(n, struct tree_entry, rb_node);
 197                 prev = n;
 198                 prev_entry = entry;
 199
 200                 if (offset < entry->start)
 201                         n = n->rb_left;
 202                 else if (offset > entry->end)
 203                         n = n->rb_right;
 204                 else {
 205                         tree->last = rb_entry(n, struct extent_state, rb_node);
 206                         return n;
 207                 }
 208         }
 209
 210         if (prev_ret) {
 211                 orig_prev = prev;
 212                 while(prev && offset > prev_entry->end) {
 213                         prev = rb_next(prev);
 214                         prev_entry = rb_entry(prev, struct tree_entry, rb_node);
 215                 }
 216                 *prev_ret = prev;
 217                 prev = orig_prev;
 218         }
 219
 220         if (next_ret) {
 221                 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
 222                 while(prev && offset < prev_entry->start) {
 223                         prev = rb_prev(prev);
 224                         prev_entry = rb_entry(prev, struct tree_entry, rb_node);
 225                 }
 226                 *next_ret = prev;
 227         }
 228         return NULL;
 229 }
 230
 231 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
 232                                           u64 offset)
 233 {
 234         struct rb_node *prev = NULL;
 235         struct rb_node *ret;
 236
 237         ret = __etree_search(tree, offset, &prev, NULL);
 238         if (!ret) {
 239                 if (prev) {
 240                         tree->last = rb_entry(prev, struct extent_state,
 241                                               rb_node);
 242                 }
 243                 return prev;
 244         }
 245         return ret;
 246 }
 247
 248 /*
 249  * utility function to look for merge candidates inside a given range.
 250  * Any extents with matching state are merged together into a single
 251  * extent in the tree.  Extents with EXTENT_IO in their state field
 252  * are not merged because the end_io handlers need to be able to do
 253  * operations on them without sleeping (or doing allocations/splits).
 254  *
 255  * This should be called with the tree lock held.
 256  */
 257 static int merge_state(struct extent_io_tree *tree,
 258                        struct extent_state *state)
 259 {
 260         struct extent_state *other;
 261         struct rb_node *other_node;
 262
 263         if (state->state & EXTENT_IOBITS)
 264                 return 0;
 265
 266         other_node = rb_prev(&state->rb_node);
 267         if (other_node) {
 268                 other = rb_entry(other_node, struct extent_state, rb_node);
 269                 if (other->end == state->start - 1 &&
 270                     other->state == state->state) {
 271                         state->start = other->start;
 272                         other->tree = NULL;
 273                         if (tree->last == other)
 274                                 tree->last = state;
 275                         rb_erase(&other->rb_node, &tree->state);
 276                         free_extent_state(other);
 277                 }
 278         }
 279         other_node = rb_next(&state->rb_node);
 280         if (other_node) {
 281                 other = rb_entry(other_node, struct extent_state, rb_node);
 282                 if (other->start == state->end + 1 &&
 283                     other->state == state->state) {
 284                         other->start = state->start;
 285                         state->tree = NULL;
 286                         if (tree->last == state)
 287                                 tree->last = other;
 288                         rb_erase(&state->rb_node, &tree->state);
 289                         free_extent_state(state);
 290                 }
 291         }
 292         return 0;
 293 }
 294
 295 static void set_state_cb(struct extent_io_tree *tree,
 296                          struct extent_state *state,
 297                          unsigned long bits)
 298 {
 299         if (tree->ops && tree->ops->set_bit_hook) {
 300                 tree->ops->set_bit_hook(tree->mapping->host, state->start,
 301                                         state->end, state->state, bits);
 302         }
 303 }
 304
 305 static void clear_state_cb(struct extent_io_tree *tree,
 306                            struct extent_state *state,
 307                            unsigned long bits)
 308 {
 309         if (tree->ops && tree->ops->set_bit_hook) {
 310                 tree->ops->clear_bit_hook(tree->mapping->host, state->start,
 311                                           state->end, state->state, bits);
 312         }
 313 }
 314
 315 /*
 316  * insert an extent_state struct into the tree.  'bits' are set on the
 317  * struct before it is inserted.
 318  *
 319  * This may return -EEXIST if the extent is already there, in which case the
 320  * state struct is freed.
 321  *
 322  * The tree lock is not taken internally.  This is a utility function and
 323  * probably isn't what you want to call (see set/clear_extent_bit).
 324  */
 325 static int insert_state(struct extent_io_tree *tree,
 326                         struct extent_state *state, u64 start, u64 end,
 327                         int bits)
 328 {
 329         struct rb_node *node;
 330
 331         if (end < start) {
 332                 printk("end < start %Lu %Lu\n", end, start);
 333                 WARN_ON(1);
 334         }
 335         if (bits & EXTENT_DIRTY)
 336                 tree->dirty_bytes += end - start + 1;
 337         set_state_cb(tree, state, bits);
 338         state->state |= bits;
 339         state->start = start;
 340         state->end = end;
 341         node = tree_insert(&tree->state, end, &state->rb_node);
 342         if (node) {
 343                 struct extent_state *found;
 344                 found = rb_entry(node, struct extent_state, rb_node);
 345                 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, start, end);
 346                 free_extent_state(state);
 347                 return -EEXIST;
 348         }
 349         state->tree = tree;
 350         tree->last = state;
 351         merge_state(tree, state);
 352         return 0;
 353 }
 354
 355 /*
 356  * split a given extent state struct in two, inserting the preallocated
 357  * struct 'prealloc' as the newly created second half.  'split' indicates an
 358  * offset inside 'orig' where it should be split.
 359  *
 360  * Before calling,
 361  * the tree has 'orig' at [orig->start, orig->end].  After calling, there
 362  * are two extent state structs in the tree:
 363  * prealloc: [orig->start, split - 1]
 364  * orig: [ split, orig->end ]
 365  *
 366  * The tree locks are not taken by this function. They need to be held
 367  * by the caller.
 368  */
 369 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
 370                        struct extent_state *prealloc, u64 split)
 371 {
 372         struct rb_node *node;
 373         prealloc->start = orig->start;
 374         prealloc->end = split - 1;
 375         prealloc->state = orig->state;
 376         orig->start = split;
 377
 378         node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
 379         if (node) {
 380                 struct extent_state *found;
 381                 found = rb_entry(node, struct extent_state, rb_node);
 382                 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, prealloc->start, prealloc->end);
 383                 free_extent_state(prealloc);
 384                 return -EEXIST;
 385         }
 386         prealloc->tree = tree;
 387         return 0;
 388 }
 389
 390 /*
 391  * utility function to clear some bits in an extent state struct.
 392  * it will optionally wake up any one waiting on this state (wake == 1), or
 393  * forcibly remove the state from the tree (delete == 1).
 394  *
 395  * If no bits are set on the state struct after clearing things, the
 396  * struct is freed and removed from the tree
 397  */
 398 static int clear_state_bit(struct extent_io_tree *tree,
 399                             struct extent_state *state, int bits, int wake,
 400                             int delete)
 401 {
 402         int ret = state->state & bits;
 403
 404         if ((bits & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
 405                 u64 range = state->end - state->start + 1;
 406                 WARN_ON(range > tree->dirty_bytes);
 407                 tree->dirty_bytes -= range;
 408         }
 409         clear_state_cb(tree, state, bits);
 410         state->state &= ~bits;
 411         if (wake)
 412                 wake_up(&state->wq);
 413         if (delete || state->state == 0) {
 414                 if (state->tree) {
 415                         clear_state_cb(tree, state, state->state);
 416                         if (tree->last == state) {
 417                                 tree->last = extent_state_next(state);
 418                         }
 419                         rb_erase(&state->rb_node, &tree->state);
 420                         state->tree = NULL;
 421                         free_extent_state(state);
 422                 } else {
 423                         WARN_ON(1);
 424                 }
 425         } else {
 426                 merge_state(tree, state);
 427         }
 428         return ret;
 429 }
 430
 431 /*
 432  * clear some bits on a range in the tree.  This may require splitting
 433  * or inserting elements in the tree, so the gfp mask is used to
 434  * indicate which allocations or sleeping are allowed.
 435  *
 436  * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
 437  * the given range from the tree regardless of state (ie for truncate).
 438  *
 439  * the range [start, end] is inclusive.
 440  *
 441  * This takes the tree lock, and returns < 0 on error, > 0 if any of the
 442  * bits were already set, or zero if none of the bits were already set.
 443  */
 444 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
 445                      int bits, int wake, int delete, gfp_t mask)
 446 {
 447         struct extent_state *state;
 448         struct extent_state *prealloc = NULL;
 449         struct rb_node *node;
 450         unsigned long flags;
 451         int err;
 452         int set = 0;
 453
 454 again:
 455         if (!prealloc && (mask & __GFP_WAIT)) {
 456                 prealloc = alloc_extent_state(mask);
 457                 if (!prealloc)
 458                         return -ENOMEM;
 459         }
 460
 461         spin_lock_irqsave(&tree->lock, flags);
 462         /*
 463          * this search will find the extents that end after
 464          * our range starts
 465          */
 466         node = tree_search(tree, start);
 467         if (!node)
 468                 goto out;
 469         state = rb_entry(node, struct extent_state, rb_node);
 470         if (state->start > end)
 471                 goto out;
 472         WARN_ON(state->end < start);
 473
 474         /*
 475          *     | ---- desired range ---- |
 476          *  | state | or
 477          *  | ------------- state -------------- |
 478          *
 479          * We need to split the extent we found, and may flip
 480          * bits on second half.
 481          *
 482          * If the extent we found extends past our range, we
 483          * just split and search again.  It'll get split again
 484          * the next time though.
 485          *
 486          * If the extent we found is inside our range, we clear
 487          * the desired bit on it.
 488          */
 489
 490         if (state->start < start) {
 491                 if (!prealloc)
 492                         prealloc = alloc_extent_state(GFP_ATOMIC);
 493                 err = split_state(tree, state, prealloc, start);
 494                 BUG_ON(err == -EEXIST);
 495                 prealloc = NULL;
 496                 if (err)
 497                         goto out;
 498                 if (state->end <= end) {
 499                         start = state->end + 1;
 500                         set |= clear_state_bit(tree, state, bits,
 501                                         wake, delete);
 502                 } else {
 503                         start = state->start;
 504                 }
 505                 goto search_again;
 506         }
 507         /*
 508          * | ---- desired range ---- |
 509          *                        | state |
 510          * We need to split the extent, and clear the bit
 511          * on the first half
 512          */
 513         if (state->start <= end && state->end > end) {
 514                 if (!prealloc)
 515                         prealloc = alloc_extent_state(GFP_ATOMIC);
 516                 err = split_state(tree, state, prealloc, end + 1);
 517                 BUG_ON(err == -EEXIST);
 518
 519                 if (wake)
 520                         wake_up(&state->wq);
 521                 set |= clear_state_bit(tree, prealloc, bits,
 522                                        wake, delete);
 523                 prealloc = NULL;
 524                 goto out;
 525         }
 526
 527         start = state->end + 1;
 528         set |= clear_state_bit(tree, state, bits, wake, delete);
 529         goto search_again;
 530
 531 out:
 532         spin_unlock_irqrestore(&tree->lock, flags);
 533         if (prealloc)
 534                 free_extent_state(prealloc);
 535
 536         return set;
 537
 538 search_again:
 539         if (start > end)
 540                 goto out;
 541         spin_unlock_irqrestore(&tree->lock, flags);
 542         if (mask & __GFP_WAIT)
 543                 cond_resched();
 544         goto again;
 545 }
 546 EXPORT_SYMBOL(clear_extent_bit);
 547
 548 static int wait_on_state(struct extent_io_tree *tree,
 549                          struct extent_state *state)
 550 {
 551         DEFINE_WAIT(wait);
 552         prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
 553         spin_unlock_irq(&tree->lock);
 554         schedule();
 555         spin_lock_irq(&tree->lock);
 556         finish_wait(&state->wq, &wait);
 557         return 0;
 558 }
 559
 560 /*
 561  * waits for one or more bits to clear on a range in the state tree.
 562  * The range [start, end] is inclusive.
 563  * The tree lock is taken by this function
 564  */
 565 int wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits)
 566 {
 567         struct extent_state *state;
 568         struct rb_node *node;
 569
 570         spin_lock_irq(&tree->lock);
 571 again:
 572         while (1) {
 573                 /*
 574                  * this search will find all the extents that end after
 575                  * our range starts
 576                  */
 577                 node = tree_search(tree, start);
 578                 if (!node)
 579                         break;
 580
 581                 state = rb_entry(node, struct extent_state, rb_node);
 582
 583                 if (state->start > end)
 584                         goto out;
 585
 586                 if (state->state & bits) {
 587                         start = state->start;
 588                         atomic_inc(&state->refs);
 589                         wait_on_state(tree, state);
 590                         free_extent_state(state);
 591                         goto again;
 592                 }
 593                 start = state->end + 1;
 594
 595                 if (start > end)
 596                         break;
 597
 598                 if (need_resched()) {
 599                         spin_unlock_irq(&tree->lock);
 600                         cond_resched();
 601                         spin_lock_irq(&tree->lock);
 602                 }
 603         }
 604 out:
 605         spin_unlock_irq(&tree->lock);
 606         return 0;
 607 }
 608 EXPORT_SYMBOL(wait_extent_bit);
 609
 610 static void set_state_bits(struct extent_io_tree *tree,
 611                            struct extent_state *state,
 612                            int bits)
 613 {
 614         if ((bits & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
 615                 u64 range = state->end - state->start + 1;
 616                 tree->dirty_bytes += range;
 617         }
 618         set_state_cb(tree, state, bits);
 619         state->state |= bits;
 620 }
 621
 622 /*
 623  * set some bits on a range in the tree.  This may require allocations
 624  * or sleeping, so the gfp mask is used to indicate what is allowed.
 625  *
 626  * If 'exclusive' == 1, this will fail with -EEXIST if some part of the
 627  * range already has the desired bits set.  The start of the existing
 628  * range is returned in failed_start in this case.
 629  *
 630  * [start, end] is inclusive
 631  * This takes the tree lock.
 632  */
 633 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits,
 634                    int exclusive, u64 *failed_start, gfp_t mask)
 635 {
 636         struct extent_state *state;
 637         struct extent_state *prealloc = NULL;
 638         struct rb_node *node;
 639         unsigned long flags;
 640         int err = 0;
 641         int set;
 642         u64 last_start;
 643         u64 last_end;
 644 again:
 645         if (!prealloc && (mask & __GFP_WAIT)) {
 646                 prealloc = alloc_extent_state(mask);
 647                 if (!prealloc)
 648                         return -ENOMEM;
 649         }
 650
 651         spin_lock_irqsave(&tree->lock, flags);
 652         /*
 653          * this search will find all the extents that end after
 654          * our range starts.
 655          */
 656         node = tree_search(tree, start);
 657         if (!node) {
 658                 err = insert_state(tree, prealloc, start, end, bits);
 659                 prealloc = NULL;
 660                 BUG_ON(err == -EEXIST);
 661                 goto out;
 662         }
 663
 664         state = rb_entry(node, struct extent_state, rb_node);
 665         last_start = state->start;
 666         last_end = state->end;
 667
 668         /*
 669          * | ---- desired range ---- |
 670          * | state |
 671          *
 672          * Just lock what we found and keep going
 673          */
 674         if (state->start == start && state->end <= end) {
 675                 set = state->state & bits;
 676                 if (set && exclusive) {
 677                         *failed_start = state->start;
 678                         err = -EEXIST;
 679                         goto out;
 680                 }
 681                 set_state_bits(tree, state, bits);
 682                 start = state->end + 1;
 683                 merge_state(tree, state);
 684                 goto search_again;
 685         }
 686
 687         /*
 688          *     | ---- desired range ---- |
 689          * | state |
 690          *   or
 691          * | ------------- state -------------- |
 692          *
 693          * We need to split the extent we found, and may flip bits on
 694          * second half.
 695          *
 696          * If the extent we found extends past our
 697          * range, we just split and search again.  It'll get split
 698          * again the next time though.
 699          *
 700          * If the extent we found is inside our range, we set the
 701          * desired bit on it.
 702          */
 703         if (state->start < start) {
 704                 set = state->state & bits;
 705                 if (exclusive && set) {
 706                         *failed_start = start;
 707                         err = -EEXIST;
 708                         goto out;
 709                 }
 710                 err = split_state(tree, state, prealloc, start);
 711                 BUG_ON(err == -EEXIST);
 712                 prealloc = NULL;
 713                 if (err)
 714                         goto out;
 715                 if (state->end <= end) {
 716                         set_state_bits(tree, state, bits);
 717                         start = state->end + 1;
 718                         merge_state(tree, state);
 719                 } else {
 720                         start = state->start;
 721                 }
 722                 goto search_again;
 723         }
 724         /*
 725          * | ---- desired range ---- |
 726          *     | state | or               | state |
 727          *
 728          * There's a hole, we need to insert something in it and
 729          * ignore the extent we found.
 730          */
 731         if (state->start > start) {
 732                 u64 this_end;
 733                 if (end < last_start)
 734                         this_end = end;
 735                 else
 736                         this_end = last_start -1;
 737                 err = insert_state(tree, prealloc, start, this_end,
 738                                    bits);
 739                 prealloc = NULL;
 740                 BUG_ON(err == -EEXIST);
 741                 if (err)
 742                         goto out;
 743                 start = this_end + 1;
 744                 goto search_again;
 745         }
 746         /*
 747          * | ---- desired range ---- |
 748          *                        | state |
 749          * We need to split the extent, and set the bit
 750          * on the first half
 751          */
 752         if (state->start <= end && state->end > end) {
 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, end + 1);
 760                 BUG_ON(err == -EEXIST);
 761
 762                 set_state_bits(tree, prealloc, bits);
 763                 merge_state(tree, prealloc);
 764                 prealloc = NULL;
 765                 goto out;
 766         }
 767
 768         goto search_again;
 769
 770 out:
 771         spin_unlock_irqrestore(&tree->lock, flags);
 772         if (prealloc)
 773                 free_extent_state(prealloc);
 774
 775         return err;
 776
 777 search_again:
 778         if (start > end)
 779                 goto out;
 780         spin_unlock_irqrestore(&tree->lock, flags);
 781         if (mask & __GFP_WAIT)
 782                 cond_resched();
 783         goto again;
 784 }
 785 EXPORT_SYMBOL(set_extent_bit);
 786
 787 /* wrappers around set/clear extent bit */
 788 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
 789                      gfp_t mask)
 790 {
 791         return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL,
 792                               mask);
 793 }
 794 EXPORT_SYMBOL(set_extent_dirty);
 795
 796 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
 797                     int bits, gfp_t mask)
 798 {
 799         return set_extent_bit(tree, start, end, bits, 0, NULL,
 800                               mask);
 801 }
 802 EXPORT_SYMBOL(set_extent_bits);
 803
 804 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
 805                       int bits, gfp_t mask)
 806 {
 807         return clear_extent_bit(tree, start, end, bits, 0, 0, mask);
 808 }
 809 EXPORT_SYMBOL(clear_extent_bits);
 810
 811 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
 812                      gfp_t mask)
 813 {
 814         return set_extent_bit(tree, start, end,
 815                               EXTENT_DELALLOC | EXTENT_DIRTY, 0, NULL,
 816                               mask);
 817 }
 818 EXPORT_SYMBOL(set_extent_delalloc);
 819
 820 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
 821                        gfp_t mask)
 822 {
 823         return clear_extent_bit(tree, start, end,
 824                                 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, mask);
 825 }
 826 EXPORT_SYMBOL(clear_extent_dirty);
 827
 828 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
 829                      gfp_t mask)
 830 {
 831         return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL,
 832                               mask);
 833 }
 834 EXPORT_SYMBOL(set_extent_new);
 835
 836 int clear_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
 837                        gfp_t mask)
 838 {
 839         return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0, mask);
 840 }
 841 EXPORT_SYMBOL(clear_extent_new);
 842
 843 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
 844                         gfp_t mask)
 845 {
 846         return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, NULL,
 847                               mask);
 848 }
 849 EXPORT_SYMBOL(set_extent_uptodate);
 850
 851 int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
 852                           gfp_t mask)
 853 {
 854         return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, mask);
 855 }
 856 EXPORT_SYMBOL(clear_extent_uptodate);
 857
 858 int set_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end,
 859                          gfp_t mask)
 860 {
 861         return set_extent_bit(tree, start, end, EXTENT_WRITEBACK,
 862                               0, NULL, mask);
 863 }
 864 EXPORT_SYMBOL(set_extent_writeback);
 865
 866 int clear_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end,
 867                            gfp_t mask)
 868 {
 869         return clear_extent_bit(tree, start, end, EXTENT_WRITEBACK, 1, 0, mask);
 870 }
 871 EXPORT_SYMBOL(clear_extent_writeback);
 872
 873 int wait_on_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end)
 874 {
 875         return wait_extent_bit(tree, start, end, EXTENT_WRITEBACK);
 876 }
 877 EXPORT_SYMBOL(wait_on_extent_writeback);
 878
 879 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
 880 {
 881         int err;
 882         u64 failed_start;
 883         while (1) {
 884                 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1,
 885                                      &failed_start, mask);
 886                 if (err == -EEXIST && (mask & __GFP_WAIT)) {
 887                         wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
 888                         start = failed_start;
 889                 } else {
 890                         break;
 891                 }
 892                 WARN_ON(start > end);
 893         }
 894         return err;
 895 }
 896 EXPORT_SYMBOL(lock_extent);
 897
 898 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end,
 899                   gfp_t mask)
 900 {
 901         return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, mask);
 902 }
 903 EXPORT_SYMBOL(unlock_extent);
 904
 905 /*
 906  * helper function to set pages and extents in the tree dirty
 907  */
 908 int set_range_dirty(struct extent_io_tree *tree, u64 start, u64 end)
 909 {
 910         unsigned long index = start >> PAGE_CACHE_SHIFT;
 911         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
 912         struct page *page;
 913
 914         while (index <= end_index) {
 915                 page = find_get_page(tree->mapping, index);
 916                 BUG_ON(!page);
 917                 __set_page_dirty_nobuffers(page);
 918                 page_cache_release(page);
 919                 index++;
 920         }
 921         set_extent_dirty(tree, start, end, GFP_NOFS);
 922         return 0;
 923 }
 924 EXPORT_SYMBOL(set_range_dirty);
 925
 926 /*
 927  * helper function to set both pages and extents in the tree writeback
 928  */
 929 int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
 930 {
 931         unsigned long index = start >> PAGE_CACHE_SHIFT;
 932         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
 933         struct page *page;
 934
 935         while (index <= end_index) {
 936                 page = find_get_page(tree->mapping, index);
 937                 BUG_ON(!page);
 938                 set_page_writeback(page);
 939                 page_cache_release(page);
 940                 index++;
 941         }
 942         set_extent_writeback(tree, start, end, GFP_NOFS);
 943         return 0;
 944 }
 945 EXPORT_SYMBOL(set_range_writeback);
 946
 947 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
 948                           u64 *start_ret, u64 *end_ret, int bits)
 949 {
 950         struct rb_node *node;
 951         struct extent_state *state;
 952         int ret = 1;
 953
 954         spin_lock_irq(&tree->lock);
 955         /*
 956          * this search will find all the extents that end after
 957          * our range starts.
 958          */
 959         node = tree_search(tree, start);
 960         if (!node) {
 961                 goto out;
 962         }
 963
 964         while(1) {
 965                 state = rb_entry(node, struct extent_state, rb_node);
 966                 if (state->end >= start && (state->state & bits)) {
 967                         *start_ret = state->start;
 968                         *end_ret = state->end;
 969                         ret = 0;
 970                         break;
 971                 }
 972                 node = rb_next(node);
 973                 if (!node)
 974                         break;
 975         }
 976 out:
 977         spin_unlock_irq(&tree->lock);
 978         return ret;
 979 }
 980 EXPORT_SYMBOL(find_first_extent_bit);
 981
 982 struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree,
 983                                                  u64 start, int bits)
 984 {
 985         struct rb_node *node;
 986         struct extent_state *state;
 987
 988         /*
 989          * this search will find all the extents that end after
 990          * our range starts.
 991          */
 992         node = tree_search(tree, start);
 993         if (!node) {
 994                 goto out;
 995         }
 996
 997         while(1) {
 998                 state = rb_entry(node, struct extent_state, rb_node);
 999                 if (state->end >= start && (state->state & bits)) {
1000                         return state;
1001                 }
1002                 node = rb_next(node);
1003                 if (!node)
1004                         break;
1005         }
1006 out:
1007         return NULL;
1008 }
1009 EXPORT_SYMBOL(find_first_extent_bit_state);
1010
1011 u64 find_lock_delalloc_range(struct extent_io_tree *tree,
1012                              u64 *start, u64 *end, u64 max_bytes)
1013 {
1014         struct rb_node *node;
1015         struct extent_state *state;
1016         u64 cur_start = *start;
1017         u64 found = 0;
1018         u64 total_bytes = 0;
1019
1020         spin_lock_irq(&tree->lock);
1021         /*
1022          * this search will find all the extents that end after
1023          * our range starts.
1024          */
1025 search_again:
1026         node = tree_search(tree, cur_start);
1027         if (!node) {
1028                 *end = (u64)-1;
1029                 goto out;
1030         }
1031
1032         while(1) {
1033                 state = rb_entry(node, struct extent_state, rb_node);
1034                 if (found && state->start != cur_start) {
1035                         goto out;
1036                 }
1037                 if (!(state->state & EXTENT_DELALLOC)) {
1038                         if (!found)
1039                                 *end = state->end;
1040                         goto out;
1041                 }
1042                 if (!found) {
1043                         struct extent_state *prev_state;
1044                         struct rb_node *prev_node = node;
1045                         while(1) {
1046                                 prev_node = rb_prev(prev_node);
1047                                 if (!prev_node)
1048                                         break;
1049                                 prev_state = rb_entry(prev_node,
1050                                                       struct extent_state,
1051                                                       rb_node);
1052                                 if (!(prev_state->state & EXTENT_DELALLOC))
1053                                         break;
1054                                 state = prev_state;
1055                                 node = prev_node;
1056                         }
1057                 }
1058                 if (state->state & EXTENT_LOCKED) {
1059                         DEFINE_WAIT(wait);
1060                         atomic_inc(&state->refs);
1061                         prepare_to_wait(&state->wq, &wait,
1062                                         TASK_UNINTERRUPTIBLE);
1063                         spin_unlock_irq(&tree->lock);
1064                         schedule();
1065                         spin_lock_irq(&tree->lock);
1066                         finish_wait(&state->wq, &wait);
1067                         free_extent_state(state);
1068                         goto search_again;
1069                 }
1070                 set_state_cb(tree, state, EXTENT_LOCKED);
1071                 state->state |= EXTENT_LOCKED;
1072                 if (!found)
1073                         *start = state->start;
1074                 found++;
1075                 *end = state->end;
1076                 cur_start = state->end + 1;
1077                 node = rb_next(node);
1078                 if (!node)
1079                         break;
1080                 total_bytes += state->end - state->start + 1;
1081                 if (total_bytes >= max_bytes)
1082                         break;
1083         }
1084 out:
1085         spin_unlock_irq(&tree->lock);
1086         return found;
1087 }
1088
1089 u64 count_range_bits(struct extent_io_tree *tree,
1090                      u64 *start, u64 search_end, u64 max_bytes,
1091                      unsigned long bits)
1092 {
1093         struct rb_node *node;
1094         struct extent_state *state;
1095         u64 cur_start = *start;
1096         u64 total_bytes = 0;
1097         int found = 0;
1098
1099         if (search_end <= cur_start) {
1100                 printk("search_end %Lu start %Lu\n", search_end, cur_start);
1101                 WARN_ON(1);
1102                 return 0;
1103         }
1104
1105         spin_lock_irq(&tree->lock);
1106         if (cur_start == 0 && bits == EXTENT_DIRTY) {
1107                 total_bytes = tree->dirty_bytes;
1108                 goto out;
1109         }
1110         /*
1111          * this search will find all the extents that end after
1112          * our range starts.
1113          */
1114         node = tree_search(tree, cur_start);
1115         if (!node) {
1116                 goto out;
1117         }
1118
1119         while(1) {
1120                 state = rb_entry(node, struct extent_state, rb_node);
1121                 if (state->start > search_end)
1122                         break;
1123                 if (state->end >= cur_start && (state->state & bits)) {
1124                         total_bytes += min(search_end, state->end) + 1 -
1125                                        max(cur_start, state->start);
1126                         if (total_bytes >= max_bytes)
1127                                 break;
1128                         if (!found) {
1129                                 *start = state->start;
1130                                 found = 1;
1131                         }
1132                 }
1133                 node = rb_next(node);
1134                 if (!node)
1135                         break;
1136         }
1137 out:
1138         spin_unlock_irq(&tree->lock);
1139         return total_bytes;
1140 }
1141 /*
1142  * helper function to lock both pages and extents in the tree.
1143  * pages must be locked first.
1144  */
1145 int lock_range(struct extent_io_tree *tree, u64 start, u64 end)
1146 {
1147         unsigned long index = start >> PAGE_CACHE_SHIFT;
1148         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1149         struct page *page;
1150         int err;
1151
1152         while (index <= end_index) {
1153                 page = grab_cache_page(tree->mapping, index);
1154                 if (!page) {
1155                         err = -ENOMEM;
1156                         goto failed;
1157                 }
1158                 if (IS_ERR(page)) {
1159                         err = PTR_ERR(page);
1160                         goto failed;
1161                 }
1162                 index++;
1163         }
1164         lock_extent(tree, start, end, GFP_NOFS);
1165         return 0;
1166
1167 failed:
1168         /*
1169          * we failed above in getting the page at 'index', so we undo here
1170          * up to but not including the page at 'index'
1171          */
1172         end_index = index;
1173         index = start >> PAGE_CACHE_SHIFT;
1174         while (index < end_index) {
1175                 page = find_get_page(tree->mapping, index);
1176                 unlock_page(page);
1177                 page_cache_release(page);
1178                 index++;
1179         }
1180         return err;
1181 }
1182 EXPORT_SYMBOL(lock_range);
1183
1184 /*
1185  * helper function to unlock both pages and extents in the tree.
1186  */
1187 int unlock_range(struct extent_io_tree *tree, u64 start, u64 end)
1188 {
1189         unsigned long index = start >> PAGE_CACHE_SHIFT;
1190         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1191         struct page *page;
1192
1193         while (index <= end_index) {
1194                 page = find_get_page(tree->mapping, index);
1195                 unlock_page(page);
1196                 page_cache_release(page);
1197                 index++;
1198         }
1199         unlock_extent(tree, start, end, GFP_NOFS);
1200         return 0;
1201 }
1202 EXPORT_SYMBOL(unlock_range);
1203
1204 int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1205 {
1206         struct rb_node *node;
1207         struct extent_state *state;
1208         int ret = 0;
1209
1210         spin_lock_irq(&tree->lock);
1211         /*
1212          * this search will find all the extents that end after
1213          * our range starts.
1214          */
1215         node = tree_search(tree, start);
1216         if (!node) {
1217                 ret = -ENOENT;
1218                 goto out;
1219         }
1220         state = rb_entry(node, struct extent_state, rb_node);
1221         if (state->start != start) {
1222                 ret = -ENOENT;
1223                 goto out;
1224         }
1225         state->private = private;
1226 out:
1227         spin_unlock_irq(&tree->lock);
1228         return ret;
1229 }
1230
1231 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1232 {
1233         struct rb_node *node;
1234         struct extent_state *state;
1235         int ret = 0;
1236
1237         spin_lock_irq(&tree->lock);
1238         /*
1239          * this search will find all the extents that end after
1240          * our range starts.
1241          */
1242         node = tree_search(tree, start);
1243         if (!node) {
1244                 ret = -ENOENT;
1245                 goto out;
1246         }
1247         state = rb_entry(node, struct extent_state, rb_node);
1248         if (state->start != start) {
1249                 ret = -ENOENT;
1250                 goto out;
1251         }
1252         *private = state->private;
1253 out:
1254         spin_unlock_irq(&tree->lock);
1255         return ret;
1256 }
1257
1258 /*
1259  * searches a range in the state tree for a given mask.
1260  * If 'filled' == 1, this returns 1 only if every extent in the tree
1261  * has the bits set.  Otherwise, 1 is returned if any bit in the
1262  * range is found set.
1263  */
1264 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1265                    int bits, int filled)
1266 {
1267         struct extent_state *state = NULL;
1268         struct rb_node *node;
1269         int bitset = 0;
1270         unsigned long flags;
1271
1272         spin_lock_irqsave(&tree->lock, flags);
1273         node = tree_search(tree, start);
1274         while (node && start <= end) {
1275                 state = rb_entry(node, struct extent_state, rb_node);
1276
1277                 if (filled && state->start > start) {
1278                         bitset = 0;
1279                         break;
1280                 }
1281
1282                 if (state->start > end)
1283                         break;
1284
1285                 if (state->state & bits) {
1286                         bitset = 1;
1287                         if (!filled)
1288                                 break;
1289                 } else if (filled) {
1290                         bitset = 0;
1291                         break;
1292                 }
1293                 start = state->end + 1;
1294                 if (start > end)
1295                         break;
1296                 node = rb_next(node);
1297                 if (!node) {
1298                         if (filled)
1299                                 bitset = 0;
1300                         break;
1301                 }
1302         }
1303         spin_unlock_irqrestore(&tree->lock, flags);
1304         return bitset;
1305 }
1306 EXPORT_SYMBOL(test_range_bit);
1307
1308 /*
1309  * helper function to set a given page up to date if all the
1310  * extents in the tree for that page are up to date
1311  */
1312 static int check_page_uptodate(struct extent_io_tree *tree,
1313                                struct page *page)
1314 {
1315         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1316         u64 end = start + PAGE_CACHE_SIZE - 1;
1317         if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1))
1318                 SetPageUptodate(page);
1319         return 0;
1320 }
1321
1322 /*
1323  * helper function to unlock a page if all the extents in the tree
1324  * for that page are unlocked
1325  */
1326 static int check_page_locked(struct extent_io_tree *tree,
1327                              struct page *page)
1328 {
1329         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1330         u64 end = start + PAGE_CACHE_SIZE - 1;
1331         if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0))
1332                 unlock_page(page);
1333         return 0;
1334 }
1335
1336 /*
1337  * helper function to end page writeback if all the extents
1338  * in the tree for that page are done with writeback
1339  */
1340 static int check_page_writeback(struct extent_io_tree *tree,
1341                              struct page *page)
1342 {
1343         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1344         u64 end = start + PAGE_CACHE_SIZE - 1;
1345         if (!test_range_bit(tree, start, end, EXTENT_WRITEBACK, 0))
1346                 end_page_writeback(page);
1347         return 0;
1348 }
1349
1350 /* lots and lots of room for performance fixes in the end_bio funcs */
1351
1352 /*
1353  * after a writepage IO is done, we need to:
1354  * clear the uptodate bits on error
1355  * clear the writeback bits in the extent tree for this IO
1356  * end_page_writeback if the page has no more pending IO
1357  *
1358  * Scheduling is not allowed, so the extent state tree is expected
1359  * to have one and only one object corresponding to this IO.
1360  */
1361 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1362 static void end_bio_extent_writepage(struct bio *bio, int err)
1363 #else
1364 static int end_bio_extent_writepage(struct bio *bio,
1365                                    unsigned int bytes_done, int err)
1366 #endif
1367 {
1368         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1369         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1370         struct extent_state *state = bio->bi_private;
1371         struct extent_io_tree *tree = state->tree;
1372         struct rb_node *node;
1373         u64 start;
1374         u64 end;
1375         u64 cur;
1376         int whole_page;
1377         unsigned long flags;
1378
1379 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1380         if (bio->bi_size)
1381                 return 1;
1382 #endif
1383         do {
1384                 struct page *page = bvec->bv_page;
1385                 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1386                          bvec->bv_offset;
1387                 end = start + bvec->bv_len - 1;
1388
1389                 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1390                         whole_page = 1;
1391                 else
1392                         whole_page = 0;
1393
1394                 if (--bvec >= bio->bi_io_vec)
1395                         prefetchw(&bvec->bv_page->flags);
1396
1397                 if (!uptodate) {
1398                         clear_extent_uptodate(tree, start, end, GFP_ATOMIC);
1399                         ClearPageUptodate(page);
1400                         SetPageError(page);
1401                 }
1402
1403                 if (tree->ops && tree->ops->writepage_end_io_hook) {
1404                         tree->ops->writepage_end_io_hook(page, start, end,
1405                                                          state);
1406                 }
1407
1408                 /*
1409                  * bios can get merged in funny ways, and so we need to
1410                  * be careful with the state variable.  We know the
1411                  * state won't be merged with others because it has
1412                  * WRITEBACK set, but we can't be sure each biovec is
1413                  * sequential in the file.  So, if our cached state
1414                  * doesn't match the expected end, search the tree
1415                  * for the correct one.
1416                  */
1417
1418                 spin_lock_irqsave(&tree->lock, flags);
1419                 if (!state || state->end != end) {
1420                         state = NULL;
1421                         node = __etree_search(tree, start, NULL, NULL);
1422                         if (node) {
1423                                 state = rb_entry(node, struct extent_state,
1424                                                  rb_node);
1425                                 if (state->end != end ||
1426                                     !(state->state & EXTENT_WRITEBACK))
1427                                         state = NULL;
1428                         }
1429                         if (!state) {
1430                                 spin_unlock_irqrestore(&tree->lock, flags);
1431                                 clear_extent_writeback(tree, start,
1432                                                        end, GFP_ATOMIC);
1433                                 goto next_io;
1434                         }
1435                 }
1436                 cur = end;
1437                 while(1) {
1438                         struct extent_state *clear = state;
1439                         cur = state->start;
1440                         node = rb_prev(&state->rb_node);
1441                         if (node) {
1442                                 state = rb_entry(node,
1443                                                  struct extent_state,
1444                                                  rb_node);
1445                         } else {
1446                                 state = NULL;
1447                         }
1448
1449                         clear_state_bit(tree, clear, EXTENT_WRITEBACK,
1450                                         1, 0);
1451                         if (cur == start)
1452                                 break;
1453                         if (cur < start) {
1454                                 WARN_ON(1);
1455                                 break;
1456                         }
1457                         if (!node)
1458                                 break;
1459                 }
1460                 /* before releasing the lock, make sure the next state
1461                  * variable has the expected bits set and corresponds
1462                  * to the correct offsets in the file
1463                  */
1464                 if (state && (state->end + 1 != start ||
1465                     !(state->state & EXTENT_WRITEBACK))) {
1466                         state = NULL;
1467                 }
1468                 spin_unlock_irqrestore(&tree->lock, flags);
1469 next_io:
1470
1471                 if (whole_page)
1472                         end_page_writeback(page);
1473                 else
1474                         check_page_writeback(tree, page);
1475         } while (bvec >= bio->bi_io_vec);
1476         bio_put(bio);
1477 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1478         return 0;
1479 #endif
1480 }
1481
1482 /*
1483  * after a readpage IO is done, we need to:
1484  * clear the uptodate bits on error
1485  * set the uptodate bits if things worked
1486  * set the page up to date if all extents in the tree are uptodate
1487  * clear the lock bit in the extent tree
1488  * unlock the page if there are no other extents locked for it
1489  *
1490  * Scheduling is not allowed, so the extent state tree is expected
1491  * to have one and only one object corresponding to this IO.
1492  */
1493 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1494 static void end_bio_extent_readpage(struct bio *bio, int err)
1495 #else
1496 static int end_bio_extent_readpage(struct bio *bio,
1497                                    unsigned int bytes_done, int err)
1498 #endif
1499 {
1500         int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1501         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1502         struct extent_state *state = bio->bi_private;
1503         struct extent_io_tree *tree = state->tree;
1504         struct rb_node *node;
1505         u64 start;
1506         u64 end;
1507         u64 cur;
1508         unsigned long flags;
1509         int whole_page;
1510         int ret;
1511
1512 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1513         if (bio->bi_size)
1514                 return 1;
1515 #endif
1516
1517         do {
1518                 struct page *page = bvec->bv_page;
1519                 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1520                         bvec->bv_offset;
1521                 end = start + bvec->bv_len - 1;
1522
1523                 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1524                         whole_page = 1;
1525                 else
1526                         whole_page = 0;
1527
1528                 if (--bvec >= bio->bi_io_vec)
1529                         prefetchw(&bvec->bv_page->flags);
1530
1531                 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
1532                         ret = tree->ops->readpage_end_io_hook(page, start, end,
1533                                                               state);
1534                         if (ret)
1535                                 uptodate = 0;
1536                 }
1537
1538                 spin_lock_irqsave(&tree->lock, flags);
1539                 if (!state || state->end != end) {
1540                         state = NULL;
1541                         node = __etree_search(tree, start, NULL, NULL);
1542                         if (node) {
1543                                 state = rb_entry(node, struct extent_state,
1544                                                  rb_node);
1545                                 if (state->end != end ||
1546                                     !(state->state & EXTENT_LOCKED))
1547                                         state = NULL;
1548                         }
1549                         if (!state && uptodate) {
1550                                 spin_unlock_irqrestore(&tree->lock, flags);
1551                                 set_extent_uptodate(tree, start, end,
1552                                                     GFP_ATOMIC);
1553                                 unlock_extent(tree, start, end, GFP_ATOMIC);
1554                                 goto next_io;
1555                         }
1556                 }
1557
1558                 cur = end;
1559                 while(1) {
1560                         struct extent_state *clear = state;
1561                         cur = state->start;
1562                         node = rb_prev(&state->rb_node);
1563                         if (node) {
1564                                 state = rb_entry(node,
1565                                          struct extent_state,
1566                                          rb_node);
1567                         } else {
1568                                 state = NULL;
1569                         }
1570                         if (uptodate) {
1571                                 set_state_cb(tree, clear, EXTENT_UPTODATE);
1572                                 clear->state |= EXTENT_UPTODATE;
1573                         }
1574                         clear_state_bit(tree, clear, EXTENT_LOCKED,
1575                                         1, 0);
1576                         if (cur == start)
1577                                 break;
1578                         if (cur < start) {
1579                                 WARN_ON(1);
1580                                 break;
1581                         }
1582                         if (!node)
1583                                 break;
1584                 }
1585                 /* before releasing the lock, make sure the next state
1586                  * variable has the expected bits set and corresponds
1587                  * to the correct offsets in the file
1588                  */
1589                 if (state && (state->end + 1 != start ||
1590                     !(state->state & EXTENT_LOCKED))) {
1591                         state = NULL;
1592                 }
1593                 spin_unlock_irqrestore(&tree->lock, flags);
1594 next_io:
1595                 if (whole_page) {
1596                         if (uptodate) {
1597                                 SetPageUptodate(page);
1598                         } else {
1599                                 ClearPageUptodate(page);
1600                                 SetPageError(page);
1601                         }
1602                         unlock_page(page);
1603                 } else {
1604                         if (uptodate) {
1605                                 check_page_uptodate(tree, page);
1606                         } else {
1607                                 ClearPageUptodate(page);
1608                                 SetPageError(page);
1609                         }
1610                         check_page_locked(tree, page);
1611                 }
1612         } while (bvec >= bio->bi_io_vec);
1613
1614         bio_put(bio);
1615 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1616         return 0;
1617 #endif
1618 }
1619
1620 /*
1621  * IO done from prepare_write is pretty simple, we just unlock
1622  * the structs in the extent tree when done, and set the uptodate bits
1623  * as appropriate.
1624  */
1625 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1626 static void end_bio_extent_preparewrite(struct bio *bio, int err)
1627 #else
1628 static int end_bio_extent_preparewrite(struct bio *bio,
1629                                        unsigned int bytes_done, int err)
1630 #endif
1631 {
1632         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1633         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1634         struct extent_state *state = bio->bi_private;
1635         struct extent_io_tree *tree = state->tree;
1636         u64 start;
1637         u64 end;
1638
1639 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1640         if (bio->bi_size)
1641                 return 1;
1642 #endif
1643
1644         do {
1645                 struct page *page = bvec->bv_page;
1646                 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1647                         bvec->bv_offset;
1648                 end = start + bvec->bv_len - 1;
1649
1650                 if (--bvec >= bio->bi_io_vec)
1651                         prefetchw(&bvec->bv_page->flags);
1652
1653                 if (uptodate) {
1654                         set_extent_uptodate(tree, start, end, GFP_ATOMIC);
1655                 } else {
1656                         ClearPageUptodate(page);
1657                         SetPageError(page);
1658                 }
1659
1660                 unlock_extent(tree, start, end, GFP_ATOMIC);
1661
1662         } while (bvec >= bio->bi_io_vec);
1663
1664         bio_put(bio);
1665 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1666         return 0;
1667 #endif
1668 }
1669
1670 static struct bio *
1671 extent_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
1672                  gfp_t gfp_flags)
1673 {
1674         struct bio *bio;
1675
1676         bio = bio_alloc(gfp_flags, nr_vecs);
1677
1678         if (bio == NULL && (current->flags & PF_MEMALLOC)) {
1679                 while (!bio && (nr_vecs /= 2))
1680                         bio = bio_alloc(gfp_flags, nr_vecs);
1681         }
1682
1683         if (bio) {
1684                 bio->bi_bdev = bdev;
1685                 bio->bi_sector = first_sector;
1686         }
1687         return bio;
1688 }
1689
1690 static int submit_one_bio(int rw, struct bio *bio, int mirror_num)
1691 {
1692         u64 maxsector;
1693         int ret = 0;
1694         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1695         struct page *page = bvec->bv_page;
1696         struct extent_io_tree *tree = bio->bi_private;
1697         struct rb_node *node;
1698         struct extent_state *state;
1699         u64 start;
1700         u64 end;
1701
1702         start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
1703         end = start + bvec->bv_len - 1;
1704
1705         spin_lock_irq(&tree->lock);
1706         node = __etree_search(tree, start, NULL, NULL);
1707         BUG_ON(!node);
1708         state = rb_entry(node, struct extent_state, rb_node);
1709         while(state->end < end) {
1710                 node = rb_next(node);
1711                 state = rb_entry(node, struct extent_state, rb_node);
1712         }
1713         BUG_ON(state->end != end);
1714         spin_unlock_irq(&tree->lock);
1715
1716         bio->bi_private = state;
1717
1718         bio_get(bio);
1719
1720         maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
1721         if (maxsector < bio->bi_sector) {
1722                 printk("sector too large max %Lu got %llu\n", maxsector,
1723                         (unsigned long long)bio->bi_sector);
1724                 WARN_ON(1);
1725         }
1726         if (tree->ops && tree->ops->submit_bio_hook)
1727                 tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
1728                                            mirror_num);
1729         else
1730                 submit_bio(rw, bio);
1731         if (bio_flagged(bio, BIO_EOPNOTSUPP))
1732                 ret = -EOPNOTSUPP;
1733         bio_put(bio);
1734         return ret;
1735 }
1736
1737 static int submit_extent_page(int rw, struct extent_io_tree *tree,
1738                               struct page *page, sector_t sector,
1739                               size_t size, unsigned long offset,
1740                               struct block_device *bdev,
1741                               struct bio **bio_ret,
1742                               unsigned long max_pages,
1743                               bio_end_io_t end_io_func,
1744                               int mirror_num)
1745 {
1746         int ret = 0;
1747         struct bio *bio;
1748         int nr;
1749
1750         if (bio_ret && *bio_ret) {
1751                 bio = *bio_ret;
1752                 if (bio->bi_sector + (bio->bi_size >> 9) != sector ||
1753                     (tree->ops && tree->ops->merge_bio_hook &&
1754                      tree->ops->merge_bio_hook(page, offset, size, bio)) ||
1755                     bio_add_page(bio, page, size, offset) < size) {
1756                         ret = submit_one_bio(rw, bio, mirror_num);
1757                         bio = NULL;
1758                 } else {
1759                         return 0;
1760                 }
1761         }
1762         nr = bio_get_nr_vecs(bdev);
1763         bio = extent_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
1764         if (!bio) {
1765                 printk("failed to allocate bio nr %d\n", nr);
1766         }
1767
1768
1769         bio_add_page(bio, page, size, offset);
1770         bio->bi_end_io = end_io_func;
1771         bio->bi_private = tree;
1772
1773         if (bio_ret) {
1774                 *bio_ret = bio;
1775         } else {
1776                 ret = submit_one_bio(rw, bio, mirror_num);
1777         }
1778
1779         return ret;
1780 }
1781
1782 void set_page_extent_mapped(struct page *page)
1783 {
1784         if (!PagePrivate(page)) {
1785                 SetPagePrivate(page);
1786                 WARN_ON(!page->mapping->a_ops->invalidatepage);
1787                 set_page_private(page, EXTENT_PAGE_PRIVATE);
1788                 page_cache_get(page);
1789         }
1790 }
1791
1792 void set_page_extent_head(struct page *page, unsigned long len)
1793 {
1794         set_page_private(page, EXTENT_PAGE_PRIVATE_FIRST_PAGE | len << 2);
1795 }
1796
1797 /*
1798  * basic readpage implementation.  Locked extent state structs are inserted
1799  * into the tree that are removed when the IO is done (by the end_io
1800  * handlers)
1801  */
1802 static int __extent_read_full_page(struct extent_io_tree *tree,
1803                                    struct page *page,
1804                                    get_extent_t *get_extent,
1805                                    struct bio **bio, int mirror_num)
1806 {
1807         struct inode *inode = page->mapping->host;
1808         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1809         u64 page_end = start + PAGE_CACHE_SIZE - 1;
1810         u64 end;
1811         u64 cur = start;
1812         u64 extent_offset;
1813         u64 last_byte = i_size_read(inode);
1814         u64 block_start;
1815         u64 cur_end;
1816         sector_t sector;
1817         struct extent_map *em;
1818         struct block_device *bdev;
1819         int ret;
1820         int nr = 0;
1821         size_t page_offset = 0;
1822         size_t iosize;
1823         size_t blocksize = inode->i_sb->s_blocksize;
1824
1825         set_page_extent_mapped(page);
1826
1827         end = page_end;
1828         lock_extent(tree, start, end, GFP_NOFS);
1829
1830         while (cur <= end) {
1831                 if (cur >= last_byte) {
1832                         char *userpage;
1833                         iosize = PAGE_CACHE_SIZE - page_offset;
1834                         userpage = kmap_atomic(page, KM_USER0);
1835                         memset(userpage + page_offset, 0, iosize);
1836                         flush_dcache_page(page);
1837                         kunmap_atomic(userpage, KM_USER0);
1838                         set_extent_uptodate(tree, cur, cur + iosize - 1,
1839                                             GFP_NOFS);
1840                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1841                         break;
1842                 }
1843                 em = get_extent(inode, page, page_offset, cur,
1844                                 end - cur + 1, 0);
1845                 if (IS_ERR(em) || !em) {
1846                         SetPageError(page);
1847                         unlock_extent(tree, cur, end, GFP_NOFS);
1848                         break;
1849                 }
1850
1851                 extent_offset = cur - em->start;
1852                 BUG_ON(extent_map_end(em) <= cur);
1853                 BUG_ON(end < cur);
1854
1855                 iosize = min(extent_map_end(em) - cur, end - cur + 1);
1856                 cur_end = min(extent_map_end(em) - 1, end);
1857                 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
1858                 sector = (em->block_start + extent_offset) >> 9;
1859                 bdev = em->bdev;
1860                 block_start = em->block_start;
1861                 free_extent_map(em);
1862                 em = NULL;
1863
1864                 /* we've found a hole, just zero and go on */
1865                 if (block_start == EXTENT_MAP_HOLE) {
1866                         char *userpage;
1867                         userpage = kmap_atomic(page, KM_USER0);
1868                         memset(userpage + page_offset, 0, iosize);
1869                         flush_dcache_page(page);
1870                         kunmap_atomic(userpage, KM_USER0);
1871
1872                         set_extent_uptodate(tree, cur, cur + iosize - 1,
1873                                             GFP_NOFS);
1874                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1875                         cur = cur + iosize;
1876                         page_offset += iosize;
1877                         continue;
1878                 }
1879                 /* the get_extent function already copied into the page */
1880                 if (test_range_bit(tree, cur, cur_end, EXTENT_UPTODATE, 1)) {
1881                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1882                         cur = cur + iosize;
1883                         page_offset += iosize;
1884                         continue;
1885                 }
1886                 /* we have an inline extent but it didn't get marked up
1887                  * to date.  Error out
1888                  */
1889                 if (block_start == EXTENT_MAP_INLINE) {
1890                         SetPageError(page);
1891                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1892                         cur = cur + iosize;
1893                         page_offset += iosize;
1894                         continue;
1895                 }
1896
1897                 ret = 0;
1898                 if (tree->ops && tree->ops->readpage_io_hook) {
1899                         ret = tree->ops->readpage_io_hook(page, cur,
1900                                                           cur + iosize - 1);
1901                 }
1902                 if (!ret) {
1903                         unsigned long nr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
1904                         nr -= page->index;
1905                         ret = submit_extent_page(READ, tree, page,
1906                                          sector, iosize, page_offset,
1907                                          bdev, bio, nr,
1908                                          end_bio_extent_readpage, mirror_num);
1909                 }
1910                 if (ret)
1911                         SetPageError(page);
1912                 cur = cur + iosize;
1913                 page_offset += iosize;
1914                 nr++;
1915         }
1916         if (!nr) {
1917                 if (!PageError(page))
1918                         SetPageUptodate(page);
1919                 unlock_page(page);
1920         }
1921         return 0;
1922 }
1923
1924 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
1925                             get_extent_t *get_extent)
1926 {
1927         struct bio *bio = NULL;
1928         int ret;
1929
1930         ret = __extent_read_full_page(tree, page, get_extent, &bio, 0);
1931         if (bio)
1932                 submit_one_bio(READ, bio, 0);
1933         return ret;
1934 }
1935 EXPORT_SYMBOL(extent_read_full_page);
1936
1937 /*
1938  * the writepage semantics are similar to regular writepage.  extent
1939  * records are inserted to lock ranges in the tree, and as dirty areas
1940  * are found, they are marked writeback.  Then the lock bits are removed
1941  * and the end_io handler clears the writeback ranges
1942  */
1943 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
1944                               void *data)
1945 {
1946         struct inode *inode = page->mapping->host;
1947         struct extent_page_data *epd = data;
1948         struct extent_io_tree *tree = epd->tree;
1949         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1950         u64 delalloc_start;
1951         u64 page_end = start + PAGE_CACHE_SIZE - 1;
1952         u64 end;
1953         u64 cur = start;
1954         u64 extent_offset;
1955         u64 last_byte = i_size_read(inode);
1956         u64 block_start;
1957         u64 iosize;
1958         sector_t sector;
1959         struct extent_map *em;
1960         struct block_device *bdev;
1961         int ret;
1962         int nr = 0;
1963         size_t page_offset = 0;
1964         size_t blocksize;
1965         loff_t i_size = i_size_read(inode);
1966         unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
1967         u64 nr_delalloc;
1968         u64 delalloc_end;
1969
1970         WARN_ON(!PageLocked(page));
1971         if (page->index > end_index) {
1972                 clear_extent_dirty(tree, start, page_end, GFP_NOFS);
1973                 unlock_page(page);
1974                 return 0;
1975         }
1976
1977         if (page->index == end_index) {
1978                 char *userpage;
1979
1980                 size_t offset = i_size & (PAGE_CACHE_SIZE - 1);
1981
1982                 userpage = kmap_atomic(page, KM_USER0);
1983                 memset(userpage + offset, 0, PAGE_CACHE_SIZE - offset);
1984                 flush_dcache_page(page);
1985                 kunmap_atomic(userpage, KM_USER0);
1986         }
1987
1988         set_page_extent_mapped(page);
1989
1990         delalloc_start = start;
1991         delalloc_end = 0;
1992         while(delalloc_end < page_end) {
1993                 nr_delalloc = find_lock_delalloc_range(tree, &delalloc_start,
1994                                                        &delalloc_end,
1995                                                        128 * 1024 * 1024);
1996                 if (nr_delalloc == 0) {
1997                         delalloc_start = delalloc_end + 1;
1998                         continue;
1999                 }
2000                 tree->ops->fill_delalloc(inode, delalloc_start,
2001                                          delalloc_end);
2002                 clear_extent_bit(tree, delalloc_start,
2003                                  delalloc_end,
2004                                  EXTENT_LOCKED | EXTENT_DELALLOC,
2005                                  1, 0, GFP_NOFS);
2006                 delalloc_start = delalloc_end + 1;
2007         }
2008         lock_extent(tree, start, page_end, GFP_NOFS);
2009
2010         end = page_end;
2011         if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0)) {
2012                 printk("found delalloc bits after lock_extent\n");
2013         }
2014
2015         if (last_byte <= start) {
2016                 clear_extent_dirty(tree, start, page_end, GFP_NOFS);
2017                 goto done;
2018         }
2019
2020         set_extent_uptodate(tree, start, page_end, GFP_NOFS);
2021         blocksize = inode->i_sb->s_blocksize;
2022
2023         while (cur <= end) {
2024                 if (cur >= last_byte) {
2025                         clear_extent_dirty(tree, cur, page_end, GFP_NOFS);
2026                         break;
2027                 }
2028                 em = epd->get_extent(inode, page, page_offset, cur,
2029                                      end - cur + 1, 1);
2030                 if (IS_ERR(em) || !em) {
2031                         SetPageError(page);
2032                         break;
2033                 }
2034
2035                 extent_offset = cur - em->start;
2036                 BUG_ON(extent_map_end(em) <= cur);
2037                 BUG_ON(end < cur);
2038                 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2039                 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2040                 sector = (em->block_start + extent_offset) >> 9;
2041                 bdev = em->bdev;
2042                 block_start = em->block_start;
2043                 free_extent_map(em);
2044                 em = NULL;
2045
2046                 if (block_start == EXTENT_MAP_HOLE ||
2047                     block_start == EXTENT_MAP_INLINE) {
2048                         clear_extent_dirty(tree, cur,
2049                                            cur + iosize - 1, GFP_NOFS);
2050                         cur = cur + iosize;
2051                         page_offset += iosize;
2052                         continue;
2053                 }
2054
2055                 /* leave this out until we have a page_mkwrite call */
2056                 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
2057                                    EXTENT_DIRTY, 0)) {
2058                         cur = cur + iosize;
2059                         page_offset += iosize;
2060                         continue;
2061                 }
2062                 clear_extent_dirty(tree, cur, cur + iosize - 1, GFP_NOFS);
2063                 if (tree->ops && tree->ops->writepage_io_hook) {
2064                         ret = tree->ops->writepage_io_hook(page, cur,
2065                                                 cur + iosize - 1);
2066                 } else {
2067                         ret = 0;
2068                 }
2069                 if (ret)
2070                         SetPageError(page);
2071                 else {
2072                         unsigned long max_nr = end_index + 1;
2073                         set_range_writeback(tree, cur, cur + iosize - 1);
2074                         if (!PageWriteback(page)) {
2075                                 printk("warning page %lu not writeback, "
2076                                        "cur %llu end %llu\n", page->index,
2077                                        (unsigned long long)cur,
2078                                        (unsigned long long)end);
2079                         }
2080
2081                         ret = submit_extent_page(WRITE, tree, page, sector,
2082                                                  iosize, page_offset, bdev,
2083                                                  &epd->bio, max_nr,
2084                                                  end_bio_extent_writepage, 0);
2085                         if (ret)
2086                                 SetPageError(page);
2087                 }
2088                 cur = cur + iosize;
2089                 page_offset += iosize;
2090                 nr++;
2091         }
2092 done:
2093         if (nr == 0) {
2094                 /* make sure the mapping tag for page dirty gets cleared */
2095                 set_page_writeback(page);
2096                 end_page_writeback(page);
2097         }
2098         unlock_extent(tree, start, page_end, GFP_NOFS);
2099         unlock_page(page);
2100         return 0;
2101 }
2102
2103 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
2104
2105 /* Taken directly from 2.6.23 for 2.6.18 back port */
2106 typedef int (*writepage_t)(struct page *page, struct writeback_control *wbc,
2107                                 void *data);
2108
2109 /**
2110  * write_cache_pages - walk the list of dirty pages of the given address space
2111  * and write all of them.
2112  * @mapping: address space structure to write
2113  * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2114  * @writepage: function called for each page
2115  * @data: data passed to writepage function
2116  *
2117  * If a page is already under I/O, write_cache_pages() skips it, even
2118  * if it's dirty.  This is desirable behaviour for memory-cleaning writeback,
2119  * but it is INCORRECT for data-integrity system calls such as fsync().  fsync()
2120  * and msync() need to guarantee that all the data which was dirty at the time
2121  * the call was made get new I/O started against them.  If wbc->sync_mode is
2122  * WB_SYNC_ALL then we were called for data integrity and we must wait for
2123  * existing IO to complete.
2124  */
2125 static int write_cache_pages(struct address_space *mapping,
2126                       struct writeback_control *wbc, writepage_t writepage,
2127                       void *data)
2128 {
2129         struct backing_dev_info *bdi = mapping->backing_dev_info;
2130         int ret = 0;
2131         int done = 0;
2132         struct pagevec pvec;
2133         int nr_pages;
2134         pgoff_t index;
2135         pgoff_t end;            /* Inclusive */
2136         int scanned = 0;
2137         int range_whole = 0;
2138
2139         if (wbc->nonblocking && bdi_write_congested(bdi)) {
2140                 wbc->encountered_congestion = 1;
2141                 return 0;
2142         }
2143
2144         pagevec_init(&pvec, 0);
2145         if (wbc->range_cyclic) {
2146                 index = mapping->writeback_index; /* Start from prev offset */
2147                 end = -1;
2148         } else {
2149                 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2150                 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2151                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2152                         range_whole = 1;
2153                 scanned = 1;
2154         }
2155 retry:
2156         while (!done && (index <= end) &&
2157                (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
2158                                               PAGECACHE_TAG_DIRTY,
2159                                               min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
2160                 unsigned i;
2161
2162                 scanned = 1;
2163                 for (i = 0; i < nr_pages; i++) {
2164                         struct page *page = pvec.pages[i];
2165
2166                         /*
2167                          * At this point we hold neither mapping->tree_lock nor
2168                          * lock on the page itself: the page may be truncated or
2169                          * invalidated (changing page->mapping to NULL), or even
2170                          * swizzled back from swapper_space to tmpfs file
2171                          * mapping
2172                          */
2173                         lock_page(page);
2174
2175                         if (unlikely(page->mapping != mapping)) {
2176                                 unlock_page(page);
2177                                 continue;
2178                         }
2179
2180                         if (!wbc->range_cyclic && page->index > end) {
2181                                 done = 1;
2182                                 unlock_page(page);
2183                                 continue;
2184                         }
2185
2186                         if (wbc->sync_mode != WB_SYNC_NONE)
2187                                 wait_on_page_writeback(page);
2188
2189                         if (PageWriteback(page) ||
2190                             !clear_page_dirty_for_io(page)) {
2191                                 unlock_page(page);
2192                                 continue;
2193                         }
2194
2195                         ret = (*writepage)(page, wbc, data);
2196
2197                         if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
2198                                 unlock_page(page);
2199                                 ret = 0;
2200                         }
2201                         if (ret || (--(wbc->nr_to_write) <= 0))
2202                                 done = 1;
2203                         if (wbc->nonblocking && bdi_write_congested(bdi)) {
2204                                 wbc->encountered_congestion = 1;
2205                                 done = 1;
2206                         }
2207                 }
2208                 pagevec_release(&pvec);
2209                 cond_resched();
2210         }
2211         if (!scanned && !done) {
2212                 /*
2213                  * We hit the last page and there is more work to be done: wrap
2214                  * back to the start of the file
2215                  */
2216                 scanned = 1;
2217                 index = 0;
2218                 goto retry;
2219         }
2220         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2221                 mapping->writeback_index = index;
2222         return ret;
2223 }
2224 #endif
2225
2226 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
2227                           get_extent_t *get_extent,
2228                           struct writeback_control *wbc)
2229 {
2230         int ret;
2231         struct address_space *mapping = page->mapping;
2232         struct extent_page_data epd = {
2233                 .bio = NULL,
2234                 .tree = tree,
2235                 .get_extent = get_extent,
2236         };
2237         struct writeback_control wbc_writepages = {
2238                 .bdi            = wbc->bdi,
2239                 .sync_mode      = WB_SYNC_NONE,
2240                 .older_than_this = NULL,
2241                 .nr_to_write    = 64,
2242                 .range_start    = page_offset(page) + PAGE_CACHE_SIZE,
2243                 .range_end      = (loff_t)-1,
2244         };
2245
2246
2247         ret = __extent_writepage(page, wbc, &epd);
2248
2249         write_cache_pages(mapping, &wbc_writepages, __extent_writepage, &epd);
2250         if (epd.bio) {
2251                 submit_one_bio(WRITE, epd.bio, 0);
2252         }
2253         return ret;
2254 }
2255 EXPORT_SYMBOL(extent_write_full_page);
2256
2257
2258 int extent_writepages(struct extent_io_tree *tree,
2259                       struct address_space *mapping,
2260                       get_extent_t *get_extent,
2261                       struct writeback_control *wbc)
2262 {
2263         int ret = 0;
2264         struct extent_page_data epd = {
2265                 .bio = NULL,
2266                 .tree = tree,
2267                 .get_extent = get_extent,
2268         };
2269
2270         ret = write_cache_pages(mapping, wbc, __extent_writepage, &epd);
2271         if (epd.bio) {
2272                 submit_one_bio(WRITE, epd.bio, 0);
2273         }
2274         return ret;
2275 }
2276 EXPORT_SYMBOL(extent_writepages);
2277
2278 int extent_readpages(struct extent_io_tree *tree,
2279                      struct address_space *mapping,
2280                      struct list_head *pages, unsigned nr_pages,
2281                      get_extent_t get_extent)
2282 {
2283         struct bio *bio = NULL;
2284         unsigned page_idx;
2285         struct pagevec pvec;
2286
2287         pagevec_init(&pvec, 0);
2288         for (page_idx = 0; page_idx < nr_pages; page_idx++) {
2289                 struct page *page = list_entry(pages->prev, struct page, lru);
2290
2291                 prefetchw(&page->flags);
2292                 list_del(&page->lru);
2293                 /*
2294                  * what we want to do here is call add_to_page_cache_lru,
2295                  * but that isn't exported, so we reproduce it here
2296                  */
2297                 if (!add_to_page_cache(page, mapping,
2298                                         page->index, GFP_KERNEL)) {
2299
2300                         /* open coding of lru_cache_add, also not exported */
2301                         page_cache_get(page);
2302                         if (!pagevec_add(&pvec, page))
2303                                 __pagevec_lru_add(&pvec);
2304                         __extent_read_full_page(tree, page, get_extent,
2305                                                 &bio, 0);
2306                 }
2307                 page_cache_release(page);
2308         }
2309         if (pagevec_count(&pvec))
2310                 __pagevec_lru_add(&pvec);
2311         BUG_ON(!list_empty(pages));
2312         if (bio)
2313                 submit_one_bio(READ, bio, 0);
2314         return 0;
2315 }
2316 EXPORT_SYMBOL(extent_readpages);
2317
2318 /*
2319  * basic invalidatepage code, this waits on any locked or writeback
2320  * ranges corresponding to the page, and then deletes any extent state
2321  * records from the tree
2322  */
2323 int extent_invalidatepage(struct extent_io_tree *tree,
2324                           struct page *page, unsigned long offset)
2325 {
2326         u64 start = ((u64)page->index << PAGE_CACHE_SHIFT);
2327         u64 end = start + PAGE_CACHE_SIZE - 1;
2328         size_t blocksize = page->mapping->host->i_sb->s_blocksize;
2329
2330         start += (offset + blocksize -1) & ~(blocksize - 1);
2331         if (start > end)
2332                 return 0;
2333
2334         lock_extent(tree, start, end, GFP_NOFS);
2335         wait_on_extent_writeback(tree, start, end);
2336         clear_extent_bit(tree, start, end,
2337                          EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC,
2338                          1, 1, GFP_NOFS);
2339         return 0;
2340 }
2341 EXPORT_SYMBOL(extent_invalidatepage);
2342
2343 /*
2344  * simple commit_write call, set_range_dirty is used to mark both
2345  * the pages and the extent records as dirty
2346  */
2347 int extent_commit_write(struct extent_io_tree *tree,
2348                         struct inode *inode, struct page *page,
2349                         unsigned from, unsigned to)
2350 {
2351         loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
2352
2353         set_page_extent_mapped(page);
2354         set_page_dirty(page);
2355
2356         if (pos > inode->i_size) {
2357                 i_size_write(inode, pos);
2358                 mark_inode_dirty(inode);
2359         }
2360         return 0;
2361 }
2362 EXPORT_SYMBOL(extent_commit_write);
2363
2364 int extent_prepare_write(struct extent_io_tree *tree,
2365                          struct inode *inode, struct page *page,
2366                          unsigned from, unsigned to, get_extent_t *get_extent)
2367 {
2368         u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2369         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
2370         u64 block_start;
2371         u64 orig_block_start;
2372         u64 block_end;
2373         u64 cur_end;
2374         struct extent_map *em;
2375         unsigned blocksize = 1 << inode->i_blkbits;
2376         size_t page_offset = 0;
2377         size_t block_off_start;
2378         size_t block_off_end;
2379         int err = 0;
2380         int iocount = 0;
2381         int ret = 0;
2382         int isnew;
2383
2384         set_page_extent_mapped(page);
2385
2386         block_start = (page_start + from) & ~((u64)blocksize - 1);
2387         block_end = (page_start + to - 1) | (blocksize - 1);
2388         orig_block_start = block_start;
2389
2390         lock_extent(tree, page_start, page_end, GFP_NOFS);
2391         while(block_start <= block_end) {
2392                 em = get_extent(inode, page, page_offset, block_start,
2393                                 block_end - block_start + 1, 1);
2394                 if (IS_ERR(em) || !em) {
2395                         goto err;
2396                 }
2397                 cur_end = min(block_end, extent_map_end(em) - 1);
2398                 block_off_start = block_start & (PAGE_CACHE_SIZE - 1);
2399                 block_off_end = block_off_start + blocksize;
2400                 isnew = clear_extent_new(tree, block_start, cur_end, GFP_NOFS);
2401
2402                 if (!PageUptodate(page) && isnew &&
2403                     (block_off_end > to || block_off_start < from)) {
2404                         void *kaddr;
2405
2406                         kaddr = kmap_atomic(page, KM_USER0);
2407                         if (block_off_end > to)
2408                                 memset(kaddr + to, 0, block_off_end - to);
2409                         if (block_off_start < from)
2410                                 memset(kaddr + block_off_start, 0,
2411                                        from - block_off_start);
2412                         flush_dcache_page(page);
2413                         kunmap_atomic(kaddr, KM_USER0);
2414                 }
2415                 if ((em->block_start != EXTENT_MAP_HOLE &&
2416                      em->block_start != EXTENT_MAP_INLINE) &&
2417                     !isnew && !PageUptodate(page) &&
2418                     (block_off_end > to || block_off_start < from) &&
2419                     !test_range_bit(tree, block_start, cur_end,
2420                                     EXTENT_UPTODATE, 1)) {
2421                         u64 sector;
2422                         u64 extent_offset = block_start - em->start;
2423                         size_t iosize;
2424                         sector = (em->block_start + extent_offset) >> 9;
2425                         iosize = (cur_end - block_start + blocksize) &
2426                                 ~((u64)blocksize - 1);
2427                         /*
2428                          * we've already got the extent locked, but we
2429                          * need to split the state such that our end_bio
2430                          * handler can clear the lock.
2431                          */
2432                         set_extent_bit(tree, block_start,
2433                                        block_start + iosize - 1,
2434                                        EXTENT_LOCKED, 0, NULL, GFP_NOFS);
2435                         ret = submit_extent_page(READ, tree, page,
2436                                          sector, iosize, page_offset, em->bdev,
2437                                          NULL, 1,
2438                                          end_bio_extent_preparewrite, 0);
2439                         iocount++;
2440                         block_start = block_start + iosize;
2441                 } else {
2442                         set_extent_uptodate(tree, block_start, cur_end,
2443                                             GFP_NOFS);
2444                         unlock_extent(tree, block_start, cur_end, GFP_NOFS);
2445                         block_start = cur_end + 1;
2446                 }
2447                 page_offset = block_start & (PAGE_CACHE_SIZE - 1);
2448                 free_extent_map(em);
2449         }
2450         if (iocount) {
2451                 wait_extent_bit(tree, orig_block_start,
2452                                 block_end, EXTENT_LOCKED);
2453         }
2454         check_page_uptodate(tree, page);
2455 err:
2456         /* FIXME, zero out newly allocated blocks on error */
2457         return err;
2458 }
2459 EXPORT_SYMBOL(extent_prepare_write);
2460
2461 /*
2462  * a helper for releasepage.  As long as there are no locked extents
2463  * in the range corresponding to the page, both state records and extent
2464  * map records are removed
2465  */
2466 int try_release_extent_mapping(struct extent_map_tree *map,
2467                                struct extent_io_tree *tree, struct page *page,
2468                                gfp_t mask)
2469 {
2470         struct extent_map *em;
2471         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2472         u64 end = start + PAGE_CACHE_SIZE - 1;
2473         u64 orig_start = start;
2474         int ret = 1;
2475         if ((mask & __GFP_WAIT) &&
2476             page->mapping->host->i_size > 16 * 1024 * 1024) {
2477                 u64 len;
2478                 while (start <= end) {
2479                         len = end - start + 1;
2480                         spin_lock(&map->lock);
2481                         em = lookup_extent_mapping(map, start, len);
2482                         if (!em || IS_ERR(em)) {
2483                                 spin_unlock(&map->lock);
2484                                 break;
2485                         }
2486                         if (em->start != start) {
2487                                 spin_unlock(&map->lock);
2488                                 free_extent_map(em);
2489                                 break;
2490                         }
2491                         if (!test_range_bit(tree, em->start,
2492                                             extent_map_end(em) - 1,
2493                                             EXTENT_LOCKED, 0)) {
2494                                 remove_extent_mapping(map, em);
2495                                 /* once for the rb tree */
2496                                 free_extent_map(em);
2497                         }
2498                         start = extent_map_end(em);
2499                         spin_unlock(&map->lock);
2500
2501                         /* once for us */
2502                         free_extent_map(em);
2503                 }
2504         }
2505         if (test_range_bit(tree, orig_start, end, EXTENT_IOBITS, 0))
2506                 ret = 0;
2507         else {
2508                 if ((mask & GFP_NOFS) == GFP_NOFS)
2509                         mask = GFP_NOFS;
2510                 clear_extent_bit(tree, orig_start, end, EXTENT_UPTODATE,
2511                                  1, 1, mask);
2512         }
2513         return ret;
2514 }
2515 EXPORT_SYMBOL(try_release_extent_mapping);
2516
2517 sector_t extent_bmap(struct address_space *mapping, sector_t iblock,
2518                 get_extent_t *get_extent)
2519 {
2520         struct inode *inode = mapping->host;
2521         u64 start = iblock << inode->i_blkbits;
2522         sector_t sector = 0;
2523         struct extent_map *em;
2524
2525         em = get_extent(inode, NULL, 0, start, (1 << inode->i_blkbits), 0);
2526         if (!em || IS_ERR(em))
2527                 return 0;
2528
2529         if (em->block_start == EXTENT_MAP_INLINE ||
2530             em->block_start == EXTENT_MAP_HOLE)
2531                 goto out;
2532
2533         sector = (em->block_start + start - em->start) >> inode->i_blkbits;
2534 out:
2535         free_extent_map(em);
2536         return sector;
2537 }
2538
2539 static int add_lru(struct extent_io_tree *tree, struct extent_buffer *eb)
2540 {
2541         if (list_empty(&eb->lru)) {
2542                 extent_buffer_get(eb);
2543                 list_add(&eb->lru, &tree->buffer_lru);
2544                 tree->lru_size++;
2545                 if (tree->lru_size >= BUFFER_LRU_MAX) {
2546                         struct extent_buffer *rm;
2547                         rm = list_entry(tree->buffer_lru.prev,
2548                                         struct extent_buffer, lru);
2549                         tree->lru_size--;
2550                         list_del_init(&rm->lru);
2551                         free_extent_buffer(rm);
2552                 }
2553         } else
2554                 list_move(&eb->lru, &tree->buffer_lru);
2555         return 0;
2556 }
2557 static struct extent_buffer *find_lru(struct extent_io_tree *tree,
2558                                       u64 start, unsigned long len)
2559 {
2560         struct list_head *lru = &tree->buffer_lru;
2561         struct list_head *cur = lru->next;
2562         struct extent_buffer *eb;
2563
2564         if (list_empty(lru))
2565                 return NULL;
2566
2567         do {
2568                 eb = list_entry(cur, struct extent_buffer, lru);
2569                 if (eb->start == start && eb->len == len) {
2570                         extent_buffer_get(eb);
2571                         return eb;
2572                 }
2573                 cur = cur->next;
2574         } while (cur != lru);
2575         return NULL;
2576 }
2577
2578 static inline unsigned long num_extent_pages(u64 start, u64 len)
2579 {
2580         return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
2581                 (start >> PAGE_CACHE_SHIFT);
2582 }
2583
2584 static inline struct page *extent_buffer_page(struct extent_buffer *eb,
2585                                               unsigned long i)
2586 {
2587         struct page *p;
2588         struct address_space *mapping;
2589
2590         if (i == 0)
2591                 return eb->first_page;
2592         i += eb->start >> PAGE_CACHE_SHIFT;
2593         mapping = eb->first_page->mapping;
2594         read_lock_irq(&mapping->tree_lock);
2595         p = radix_tree_lookup(&mapping->page_tree, i);
2596         read_unlock_irq(&mapping->tree_lock);
2597         return p;
2598 }
2599
2600 int release_extent_buffer_tail_pages(struct extent_buffer *eb)
2601 {
2602         unsigned long num_pages = num_extent_pages(eb->start, eb->len);
2603         struct page *page;
2604         unsigned long i;
2605
2606         if (num_pages == 1)
2607                 return 0;
2608         for (i = 1; i < num_pages; i++) {
2609                 page = extent_buffer_page(eb, i);
2610                 page_cache_release(page);
2611         }
2612         return 0;
2613 }
2614
2615
2616 int invalidate_extent_lru(struct extent_io_tree *tree, u64 start,
2617                           unsigned long len)
2618 {
2619         struct list_head *lru = &tree->buffer_lru;
2620         struct list_head *cur = lru->next;
2621         struct extent_buffer *eb;
2622         int found = 0;
2623
2624         spin_lock(&tree->lru_lock);
2625         if (list_empty(lru))
2626                 goto out;
2627
2628         do {
2629                 eb = list_entry(cur, struct extent_buffer, lru);
2630                 if (eb->start <= start && eb->start + eb->len > start) {
2631                         eb->flags &= ~EXTENT_UPTODATE;
2632                 }
2633                 cur = cur->next;
2634         } while (cur != lru);
2635 out:
2636         spin_unlock(&tree->lru_lock);
2637         return found;
2638 }
2639
2640 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
2641                                                    u64 start,
2642                                                    unsigned long len,
2643                                                    gfp_t mask)
2644 {
2645         struct extent_buffer *eb = NULL;
2646         unsigned long flags;
2647
2648         spin_lock(&tree->lru_lock);
2649         eb = find_lru(tree, start, len);
2650         spin_unlock(&tree->lru_lock);
2651         if (eb) {
2652                 return eb;
2653         }
2654
2655         eb = kmem_cache_zalloc(extent_buffer_cache, mask);
2656         INIT_LIST_HEAD(&eb->lru);
2657         eb->start = start;
2658         eb->len = len;
2659         spin_lock_irqsave(&leak_lock, flags);
2660         list_add(&eb->leak_list, &buffers);
2661         spin_unlock_irqrestore(&leak_lock, flags);
2662         atomic_set(&eb->refs, 1);
2663
2664         return eb;
2665 }
2666
2667 static void __free_extent_buffer(struct extent_buffer *eb)
2668 {
2669         unsigned long flags;
2670         spin_lock_irqsave(&leak_lock, flags);
2671         list_del(&eb->leak_list);
2672         spin_unlock_irqrestore(&leak_lock, flags);
2673         kmem_cache_free(extent_buffer_cache, eb);
2674 }
2675
2676 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
2677                                           u64 start, unsigned long len,
2678                                           struct page *page0,
2679                                           gfp_t mask)
2680 {
2681         unsigned long num_pages = num_extent_pages(start, len);
2682         unsigned long i;
2683         unsigned long index = start >> PAGE_CACHE_SHIFT;
2684         struct extent_buffer *eb;
2685         struct page *p;
2686         struct address_space *mapping = tree->mapping;
2687         int uptodate = 1;
2688
2689         eb = __alloc_extent_buffer(tree, start, len, mask);
2690         if (!eb)
2691                 return NULL;
2692
2693         if (eb->flags & EXTENT_BUFFER_FILLED)
2694                 goto lru_add;
2695
2696         if (page0) {
2697                 eb->first_page = page0;
2698                 i = 1;
2699                 index++;
2700                 page_cache_get(page0);
2701                 mark_page_accessed(page0);
2702                 set_page_extent_mapped(page0);
2703                 set_page_extent_head(page0, len);
2704                 uptodate = PageUptodate(page0);
2705         } else {
2706                 i = 0;
2707         }
2708         for (; i < num_pages; i++, index++) {
2709                 p = find_or_create_page(mapping, index, mask | __GFP_HIGHMEM);
2710                 if (!p) {
2711                         WARN_ON(1);
2712                         goto fail;
2713                 }
2714                 set_page_extent_mapped(p);
2715                 mark_page_accessed(p);
2716                 if (i == 0) {
2717                         eb->first_page = p;
2718                         set_page_extent_head(p, len);
2719                 } else {
2720                         set_page_private(p, EXTENT_PAGE_PRIVATE);
2721                 }
2722                 if (!PageUptodate(p))
2723                         uptodate = 0;
2724                 unlock_page(p);
2725         }
2726         if (uptodate)
2727                 eb->flags |= EXTENT_UPTODATE;
2728         eb->flags |= EXTENT_BUFFER_FILLED;
2729
2730 lru_add:
2731         spin_lock(&tree->lru_lock);
2732         add_lru(tree, eb);
2733         spin_unlock(&tree->lru_lock);
2734         return eb;
2735
2736 fail:
2737         spin_lock(&tree->lru_lock);
2738         list_del_init(&eb->lru);
2739         spin_unlock(&tree->lru_lock);
2740         if (!atomic_dec_and_test(&eb->refs))
2741                 return NULL;
2742         for (index = 1; index < i; index++) {
2743                 page_cache_release(extent_buffer_page(eb, index));
2744         }
2745         if (i > 0)
2746                 page_cache_release(extent_buffer_page(eb, 0));
2747         __free_extent_buffer(eb);
2748         return NULL;
2749 }
2750 EXPORT_SYMBOL(alloc_extent_buffer);
2751
2752 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
2753                                          u64 start, unsigned long len,
2754                                           gfp_t mask)
2755 {
2756         unsigned long num_pages = num_extent_pages(start, len);
2757         unsigned long i;
2758         unsigned long index = start >> PAGE_CACHE_SHIFT;
2759         struct extent_buffer *eb;
2760         struct page *p;
2761         struct address_space *mapping = tree->mapping;
2762         int uptodate = 1;
2763
2764         eb = __alloc_extent_buffer(tree, start, len, mask);
2765         if (!eb)
2766                 return NULL;
2767
2768         if (eb->flags & EXTENT_BUFFER_FILLED)
2769                 goto lru_add;
2770
2771         for (i = 0; i < num_pages; i++, index++) {
2772                 p = find_lock_page(mapping, index);
2773                 if (!p) {
2774                         goto fail;
2775                 }
2776                 set_page_extent_mapped(p);
2777                 mark_page_accessed(p);
2778
2779                 if (i == 0) {
2780                         eb->first_page = p;
2781                         set_page_extent_head(p, len);
2782                 } else {
2783                         set_page_private(p, EXTENT_PAGE_PRIVATE);
2784                 }
2785
2786                 if (!PageUptodate(p))
2787                         uptodate = 0;
2788                 unlock_page(p);
2789         }
2790         if (uptodate)
2791                 eb->flags |= EXTENT_UPTODATE;
2792         eb->flags |= EXTENT_BUFFER_FILLED;
2793
2794 lru_add:
2795         spin_lock(&tree->lru_lock);
2796         add_lru(tree, eb);
2797         spin_unlock(&tree->lru_lock);
2798         return eb;
2799 fail:
2800         spin_lock(&tree->lru_lock);
2801         list_del_init(&eb->lru);
2802         spin_unlock(&tree->lru_lock);
2803         if (!atomic_dec_and_test(&eb->refs))
2804                 return NULL;
2805         for (index = 1; index < i; index++) {
2806                 page_cache_release(extent_buffer_page(eb, index));
2807         }
2808         if (i > 0)
2809                 page_cache_release(extent_buffer_page(eb, 0));
2810         __free_extent_buffer(eb);
2811         return NULL;
2812 }
2813 EXPORT_SYMBOL(find_extent_buffer);
2814
2815 void free_extent_buffer(struct extent_buffer *eb)
2816 {
2817         unsigned long i;
2818         unsigned long num_pages;
2819
2820         if (!eb)
2821                 return;
2822
2823         if (!atomic_dec_and_test(&eb->refs))
2824                 return;
2825
2826         WARN_ON(!list_empty(&eb->lru));
2827         num_pages = num_extent_pages(eb->start, eb->len);
2828
2829         for (i = 1; i < num_pages; i++) {
2830                 page_cache_release(extent_buffer_page(eb, i));
2831         }
2832         page_cache_release(extent_buffer_page(eb, 0));
2833         __free_extent_buffer(eb);
2834 }
2835 EXPORT_SYMBOL(free_extent_buffer);
2836
2837 int clear_extent_buffer_dirty(struct extent_io_tree *tree,
2838                               struct extent_buffer *eb)
2839 {
2840         int set;
2841         unsigned long i;
2842         unsigned long num_pages;
2843         struct page *page;
2844
2845         u64 start = eb->start;
2846         u64 end = start + eb->len - 1;
2847
2848         set = clear_extent_dirty(tree, start, end, GFP_NOFS);
2849         num_pages = num_extent_pages(eb->start, eb->len);
2850
2851         for (i = 0; i < num_pages; i++) {
2852                 page = extent_buffer_page(eb, i);
2853                 lock_page(page);
2854                 if (i == 0)
2855                         set_page_extent_head(page, eb->len);
2856                 else
2857                         set_page_private(page, EXTENT_PAGE_PRIVATE);
2858
2859                 /*
2860                  * if we're on the last page or the first page and the
2861                  * block isn't aligned on a page boundary, do extra checks
2862                  * to make sure we don't clean page that is partially dirty
2863                  */
2864                 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
2865                     ((i == num_pages - 1) &&
2866                      ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
2867                         start = (u64)page->index << PAGE_CACHE_SHIFT;
2868                         end  = start + PAGE_CACHE_SIZE - 1;
2869                         if (test_range_bit(tree, start, end,
2870                                            EXTENT_DIRTY, 0)) {
2871                                 unlock_page(page);
2872                                 continue;
2873                         }
2874                 }
2875                 clear_page_dirty_for_io(page);
2876                 read_lock_irq(&page->mapping->tree_lock);
2877                 if (!PageDirty(page)) {
2878                         radix_tree_tag_clear(&page->mapping->page_tree,
2879                                                 page_index(page),
2880                                                 PAGECACHE_TAG_DIRTY);
2881                 }
2882                 read_unlock_irq(&page->mapping->tree_lock);
2883                 unlock_page(page);
2884         }
2885         return 0;
2886 }
2887 EXPORT_SYMBOL(clear_extent_buffer_dirty);
2888
2889 int wait_on_extent_buffer_writeback(struct extent_io_tree *tree,
2890                                     struct extent_buffer *eb)
2891 {
2892         return wait_on_extent_writeback(tree, eb->start,
2893                                         eb->start + eb->len - 1);
2894 }
2895 EXPORT_SYMBOL(wait_on_extent_buffer_writeback);
2896
2897 int set_extent_buffer_dirty(struct extent_io_tree *tree,
2898                              struct extent_buffer *eb)
2899 {
2900         unsigned long i;
2901         unsigned long num_pages;
2902
2903         num_pages = num_extent_pages(eb->start, eb->len);
2904         for (i = 0; i < num_pages; i++) {
2905                 struct page *page = extent_buffer_page(eb, i);
2906                 /* writepage may need to do something special for the
2907                  * first page, we have to make sure page->private is
2908                  * properly set.  releasepage may drop page->private
2909                  * on us if the page isn't already dirty.
2910                  */
2911                 if (i == 0) {
2912                         lock_page(page);
2913                         set_page_extent_head(page, eb->len);
2914                 } else if (PagePrivate(page) &&
2915                            page->private != EXTENT_PAGE_PRIVATE) {
2916                         lock_page(page);
2917                         set_page_extent_mapped(page);
2918                         unlock_page(page);
2919                 }
2920                 __set_page_dirty_nobuffers(extent_buffer_page(eb, i));
2921                 if (i == 0)
2922                         unlock_page(page);
2923         }
2924         return set_extent_dirty(tree, eb->start,
2925                                 eb->start + eb->len - 1, GFP_NOFS);
2926 }
2927 EXPORT_SYMBOL(set_extent_buffer_dirty);
2928
2929 int set_extent_buffer_uptodate(struct extent_io_tree *tree,
2930                                 struct extent_buffer *eb)
2931 {
2932         unsigned long i;
2933         struct page *page;
2934         unsigned long num_pages;
2935
2936         num_pages = num_extent_pages(eb->start, eb->len);
2937
2938         set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
2939                             GFP_NOFS);
2940         for (i = 0; i < num_pages; i++) {
2941                 page = extent_buffer_page(eb, i);
2942                 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
2943                     ((i == num_pages - 1) &&
2944                      ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
2945                         check_page_uptodate(tree, page);
2946                         continue;
2947                 }
2948                 SetPageUptodate(page);
2949         }
2950         return 0;
2951 }
2952 EXPORT_SYMBOL(set_extent_buffer_uptodate);
2953
2954 int extent_range_uptodate(struct extent_io_tree *tree,
2955                           u64 start, u64 end)
2956 {
2957         struct page *page;
2958         int ret;
2959         int pg_uptodate = 1;
2960         int uptodate;
2961         unsigned long index;
2962
2963         ret = test_range_bit(tree, start, end, EXTENT_UPTODATE, 1);
2964         if (ret)
2965                 return 1;
2966         while(start <= end) {
2967                 index = start >> PAGE_CACHE_SHIFT;
2968                 page = find_get_page(tree->mapping, index);
2969                 uptodate = PageUptodate(page);
2970                 page_cache_release(page);
2971                 if (!uptodate) {
2972                         pg_uptodate = 0;
2973                         break;
2974                 }
2975                 start += PAGE_CACHE_SIZE;
2976         }
2977         return pg_uptodate;
2978 }
2979
2980 int extent_buffer_uptodate(struct extent_io_tree *tree,
2981                            struct extent_buffer *eb)
2982 {
2983         int ret = 0;
2984         int ret2;
2985         unsigned long num_pages;
2986         unsigned long i;
2987         struct page *page;
2988         int pg_uptodate = 1;
2989
2990         if (eb->flags & EXTENT_UPTODATE)
2991                 ret = 1;
2992
2993         ret2  = test_range_bit(tree, eb->start, eb->start + eb->len - 1,
2994                            EXTENT_UPTODATE, 1);
2995
2996         num_pages = num_extent_pages(eb->start, eb->len);
2997         for (i = 0; i < num_pages; i++) {
2998                 page = extent_buffer_page(eb, i);
2999                 if (!PageUptodate(page)) {
3000                         pg_uptodate = 0;
3001                         break;
3002                 }
3003         }
3004         if ((ret || ret2) && !pg_uptodate) {
3005 printk("uptodate error2 eb %Lu ret %d ret2 %d pg_uptodate %d\n", eb->start, ret, ret2, pg_uptodate);
3006                 WARN_ON(1);
3007         }
3008         return (ret || ret2);
3009 }
3010 EXPORT_SYMBOL(extent_buffer_uptodate);
3011
3012 int read_extent_buffer_pages(struct extent_io_tree *tree,
3013                              struct extent_buffer *eb,
3014                              u64 start, int wait,
3015                              get_extent_t *get_extent, int mirror_num)
3016 {
3017         unsigned long i;
3018         unsigned long start_i;
3019         struct page *page;
3020         int err;
3021         int ret = 0;
3022         int locked_pages = 0;
3023         int all_uptodate = 1;
3024         int inc_all_pages = 0;
3025         unsigned long num_pages;
3026         struct bio *bio = NULL;
3027
3028         if (eb->flags & EXTENT_UPTODATE)
3029                 return 0;
3030
3031         if (test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3032                            EXTENT_UPTODATE, 1)) {
3033                 return 0;
3034         }
3035
3036         if (start) {
3037                 WARN_ON(start < eb->start);
3038                 start_i = (start >> PAGE_CACHE_SHIFT) -
3039                         (eb->start >> PAGE_CACHE_SHIFT);
3040         } else {
3041                 start_i = 0;
3042         }
3043
3044         num_pages = num_extent_pages(eb->start, eb->len);
3045         for (i = start_i; i < num_pages; i++) {
3046                 page = extent_buffer_page(eb, i);
3047                 if (!wait) {
3048                         if (TestSetPageLocked(page))
3049                                 goto unlock_exit;
3050                 } else {
3051                         lock_page(page);
3052                 }
3053                 locked_pages++;
3054                 if (!PageUptodate(page)) {
3055                         all_uptodate = 0;
3056                 }
3057         }
3058         if (all_uptodate) {
3059                 if (start_i == 0)
3060                         eb->flags |= EXTENT_UPTODATE;
3061                 goto unlock_exit;
3062         }
3063
3064         for (i = start_i; i < num_pages; i++) {
3065                 page = extent_buffer_page(eb, i);
3066                 if (inc_all_pages)
3067                         page_cache_get(page);
3068                 if (!PageUptodate(page)) {
3069                         if (start_i == 0)
3070                                 inc_all_pages = 1;
3071                         ClearPageError(page);
3072                         err = __extent_read_full_page(tree, page,
3073                                                       get_extent, &bio,
3074                                                       mirror_num);
3075                         if (err) {
3076                                 ret = err;
3077                         }
3078                 } else {
3079                         unlock_page(page);
3080                 }
3081         }
3082
3083         if (bio)
3084                 submit_one_bio(READ, bio, mirror_num);
3085
3086         if (ret || !wait) {
3087                 return ret;
3088         }
3089         for (i = start_i; i < num_pages; i++) {
3090                 page = extent_buffer_page(eb, i);
3091                 wait_on_page_locked(page);
3092                 if (!PageUptodate(page)) {
3093                         ret = -EIO;
3094                 }
3095         }
3096         if (!ret)
3097                 eb->flags |= EXTENT_UPTODATE;
3098         return ret;
3099
3100 unlock_exit:
3101         i = start_i;
3102         while(locked_pages > 0) {
3103                 page = extent_buffer_page(eb, i);
3104                 i++;
3105                 unlock_page(page);
3106                 locked_pages--;
3107         }
3108         return ret;
3109 }
3110 EXPORT_SYMBOL(read_extent_buffer_pages);
3111
3112 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
3113                         unsigned long start,
3114                         unsigned long len)
3115 {
3116         size_t cur;
3117         size_t offset;
3118         struct page *page;
3119         char *kaddr;
3120         char *dst = (char *)dstv;
3121         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3122         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3123
3124         WARN_ON(start > eb->len);
3125         WARN_ON(start + len > eb->start + eb->len);
3126
3127         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3128
3129         while(len > 0) {
3130                 page = extent_buffer_page(eb, i);
3131
3132                 cur = min(len, (PAGE_CACHE_SIZE - offset));
3133                 kaddr = kmap_atomic(page, KM_USER1);
3134                 memcpy(dst, kaddr + offset, cur);
3135                 kunmap_atomic(kaddr, KM_USER1);
3136
3137                 dst += cur;
3138                 len -= cur;
3139                 offset = 0;
3140                 i++;
3141         }
3142 }
3143 EXPORT_SYMBOL(read_extent_buffer);
3144
3145 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
3146                                unsigned long min_len, char **token, char **map,
3147                                unsigned long *map_start,
3148                                unsigned long *map_len, int km)
3149 {
3150         size_t offset = start & (PAGE_CACHE_SIZE - 1);
3151         char *kaddr;
3152         struct page *p;
3153         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3154         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3155         unsigned long end_i = (start_offset + start + min_len - 1) >>
3156                 PAGE_CACHE_SHIFT;
3157
3158         if (i != end_i)
3159                 return -EINVAL;
3160
3161         if (i == 0) {
3162                 offset = start_offset;
3163                 *map_start = 0;
3164         } else {
3165                 offset = 0;
3166                 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
3167         }
3168         if (start + min_len > eb->len) {
3169 printk("bad mapping eb start %Lu len %lu, wanted %lu %lu\n", eb->start, eb->len, start, min_len);
3170                 WARN_ON(1);
3171         }
3172
3173         p = extent_buffer_page(eb, i);
3174         kaddr = kmap_atomic(p, km);
3175         *token = kaddr;
3176         *map = kaddr + offset;
3177         *map_len = PAGE_CACHE_SIZE - offset;
3178         return 0;
3179 }
3180 EXPORT_SYMBOL(map_private_extent_buffer);
3181
3182 int map_extent_buffer(struct extent_buffer *eb, unsigned long start,
3183                       unsigned long min_len,
3184                       char **token, char **map,
3185                       unsigned long *map_start,
3186                       unsigned long *map_len, int km)
3187 {
3188         int err;
3189         int save = 0;
3190         if (eb->map_token) {
3191                 unmap_extent_buffer(eb, eb->map_token, km);
3192                 eb->map_token = NULL;
3193                 save = 1;
3194         }
3195         err = map_private_extent_buffer(eb, start, min_len, token, map,
3196                                        map_start, map_len, km);
3197         if (!err && save) {
3198                 eb->map_token = *token;
3199                 eb->kaddr = *map;
3200                 eb->map_start = *map_start;
3201                 eb->map_len = *map_len;
3202         }
3203         return err;
3204 }
3205 EXPORT_SYMBOL(map_extent_buffer);
3206
3207 void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km)
3208 {
3209         kunmap_atomic(token, km);
3210 }
3211 EXPORT_SYMBOL(unmap_extent_buffer);
3212
3213 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
3214                           unsigned long start,
3215                           unsigned long len)
3216 {
3217         size_t cur;
3218         size_t offset;
3219         struct page *page;
3220         char *kaddr;
3221         char *ptr = (char *)ptrv;
3222         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3223         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3224         int ret = 0;
3225
3226         WARN_ON(start > eb->len);
3227         WARN_ON(start + len > eb->start + eb->len);
3228
3229         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3230
3231         while(len > 0) {
3232                 page = extent_buffer_page(eb, i);
3233
3234                 cur = min(len, (PAGE_CACHE_SIZE - offset));
3235
3236                 kaddr = kmap_atomic(page, KM_USER0);
3237                 ret = memcmp(ptr, kaddr + offset, cur);
3238                 kunmap_atomic(kaddr, KM_USER0);
3239                 if (ret)
3240                         break;
3241
3242                 ptr += cur;
3243                 len -= cur;
3244                 offset = 0;
3245                 i++;
3246         }
3247         return ret;
3248 }
3249 EXPORT_SYMBOL(memcmp_extent_buffer);
3250
3251 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
3252                          unsigned long start, unsigned long len)
3253 {
3254         size_t cur;
3255         size_t offset;
3256         struct page *page;
3257         char *kaddr;
3258         char *src = (char *)srcv;
3259         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3260         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3261
3262         WARN_ON(start > eb->len);
3263         WARN_ON(start + len > eb->start + eb->len);
3264
3265         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3266
3267         while(len > 0) {
3268                 page = extent_buffer_page(eb, i);
3269                 WARN_ON(!PageUptodate(page));
3270
3271                 cur = min(len, PAGE_CACHE_SIZE - offset);
3272                 kaddr = kmap_atomic(page, KM_USER1);
3273                 memcpy(kaddr + offset, src, cur);
3274                 kunmap_atomic(kaddr, KM_USER1);
3275
3276                 src += cur;
3277                 len -= cur;
3278                 offset = 0;
3279                 i++;
3280         }
3281 }
3282 EXPORT_SYMBOL(write_extent_buffer);
3283
3284 void memset_extent_buffer(struct extent_buffer *eb, char c,
3285                           unsigned long start, unsigned long len)
3286 {
3287         size_t cur;
3288         size_t offset;
3289         struct page *page;
3290         char *kaddr;
3291         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3292         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3293
3294         WARN_ON(start > eb->len);
3295         WARN_ON(start + len > eb->start + eb->len);
3296
3297         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3298
3299         while(len > 0) {
3300                 page = extent_buffer_page(eb, i);
3301                 WARN_ON(!PageUptodate(page));
3302
3303                 cur = min(len, PAGE_CACHE_SIZE - offset);
3304                 kaddr = kmap_atomic(page, KM_USER0);
3305                 memset(kaddr + offset, c, cur);
3306                 kunmap_atomic(kaddr, KM_USER0);
3307
3308                 len -= cur;
3309                 offset = 0;
3310                 i++;
3311         }
3312 }
3313 EXPORT_SYMBOL(memset_extent_buffer);
3314
3315 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
3316                         unsigned long dst_offset, unsigned long src_offset,
3317                         unsigned long len)
3318 {
3319         u64 dst_len = dst->len;
3320         size_t cur;
3321         size_t offset;
3322         struct page *page;
3323         char *kaddr;
3324         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3325         unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3326
3327         WARN_ON(src->len != dst_len);
3328
3329         offset = (start_offset + dst_offset) &
3330                 ((unsigned long)PAGE_CACHE_SIZE - 1);
3331
3332         while(len > 0) {
3333                 page = extent_buffer_page(dst, i);
3334                 WARN_ON(!PageUptodate(page));
3335
3336                 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
3337
3338                 kaddr = kmap_atomic(page, KM_USER0);
3339                 read_extent_buffer(src, kaddr + offset, src_offset, cur);
3340                 kunmap_atomic(kaddr, KM_USER0);
3341
3342                 src_offset += cur;
3343                 len -= cur;
3344                 offset = 0;
3345                 i++;
3346         }
3347 }
3348 EXPORT_SYMBOL(copy_extent_buffer);
3349
3350 static void move_pages(struct page *dst_page, struct page *src_page,
3351                        unsigned long dst_off, unsigned long src_off,
3352                        unsigned long len)
3353 {
3354         char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3355         if (dst_page == src_page) {
3356                 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
3357         } else {
3358                 char *src_kaddr = kmap_atomic(src_page, KM_USER1);
3359                 char *p = dst_kaddr + dst_off + len;
3360                 char *s = src_kaddr + src_off + len;
3361
3362                 while (len--)
3363                         *--p = *--s;
3364
3365                 kunmap_atomic(src_kaddr, KM_USER1);
3366         }
3367         kunmap_atomic(dst_kaddr, KM_USER0);
3368 }
3369
3370 static void copy_pages(struct page *dst_page, struct page *src_page,
3371                        unsigned long dst_off, unsigned long src_off,
3372                        unsigned long len)
3373 {
3374         char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3375         char *src_kaddr;
3376
3377         if (dst_page != src_page)
3378                 src_kaddr = kmap_atomic(src_page, KM_USER1);
3379         else
3380                 src_kaddr = dst_kaddr;
3381
3382         memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
3383         kunmap_atomic(dst_kaddr, KM_USER0);
3384         if (dst_page != src_page)
3385                 kunmap_atomic(src_kaddr, KM_USER1);
3386 }
3387
3388 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3389                            unsigned long src_offset, unsigned long len)
3390 {
3391         size_t cur;
3392         size_t dst_off_in_page;
3393         size_t src_off_in_page;
3394         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3395         unsigned long dst_i;
3396         unsigned long src_i;
3397
3398         if (src_offset + len > dst->len) {
3399                 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
3400                        src_offset, len, dst->len);
3401                 BUG_ON(1);
3402         }
3403         if (dst_offset + len > dst->len) {
3404                 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
3405                        dst_offset, len, dst->len);
3406                 BUG_ON(1);
3407         }
3408
3409         while(len > 0) {
3410                 dst_off_in_page = (start_offset + dst_offset) &
3411                         ((unsigned long)PAGE_CACHE_SIZE - 1);
3412                 src_off_in_page = (start_offset + src_offset) &
3413                         ((unsigned long)PAGE_CACHE_SIZE - 1);
3414
3415                 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3416                 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
3417
3418                 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
3419                                                src_off_in_page));
3420                 cur = min_t(unsigned long, cur,
3421                         (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
3422
3423                 copy_pages(extent_buffer_page(dst, dst_i),
3424                            extent_buffer_page(dst, src_i),
3425                            dst_off_in_page, src_off_in_page, cur);
3426
3427                 src_offset += cur;
3428                 dst_offset += cur;
3429                 len -= cur;
3430         }
3431 }
3432 EXPORT_SYMBOL(memcpy_extent_buffer);
3433
3434 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3435                            unsigned long src_offset, unsigned long len)
3436 {
3437         size_t cur;
3438         size_t dst_off_in_page;
3439         size_t src_off_in_page;
3440         unsigned long dst_end = dst_offset + len - 1;
3441         unsigned long src_end = src_offset + len - 1;
3442         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3443         unsigned long dst_i;
3444         unsigned long src_i;
3445
3446         if (src_offset + len > dst->len) {
3447                 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
3448                        src_offset, len, dst->len);
3449                 BUG_ON(1);
3450         }
3451         if (dst_offset + len > dst->len) {
3452                 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
3453                        dst_offset, len, dst->len);
3454                 BUG_ON(1);
3455         }
3456         if (dst_offset < src_offset) {
3457                 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
3458                 return;
3459         }
3460         while(len > 0) {
3461                 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
3462                 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
3463
3464                 dst_off_in_page = (start_offset + dst_end) &
3465                         ((unsigned long)PAGE_CACHE_SIZE - 1);
3466                 src_off_in_page = (start_offset + src_end) &
3467                         ((unsigned long)PAGE_CACHE_SIZE - 1);
3468
3469                 cur = min_t(unsigned long, len, src_off_in_page + 1);
3470                 cur = min(cur, dst_off_in_page + 1);
3471                 move_pages(extent_buffer_page(dst, dst_i),
3472                            extent_buffer_page(dst, src_i),
3473                            dst_off_in_page - cur + 1,
3474                            src_off_in_page - cur + 1, cur);
3475
3476                 dst_end -= cur;
3477                 src_end -= cur;
3478                 len -= cur;
3479         }
3480 }
3481 EXPORT_SYMBOL(memmove_extent_buffer);