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->buffer.rb_node = NULL;
  95         tree->ops = NULL;
  96         tree->dirty_bytes = 0;
  97         spin_lock_init(&tree->lock);
  98         spin_lock_init(&tree->buffer_lock);
  99         tree->mapping = mapping;
 100         tree->last = NULL;
 101 }
 102 EXPORT_SYMBOL(extent_io_tree_init);
 103
 104 struct extent_state *alloc_extent_state(gfp_t mask)
 105 {
 106         struct extent_state *state;
 107         unsigned long flags;
 108
 109         state = kmem_cache_alloc(extent_state_cache, mask);
 110         if (!state)
 111                 return state;
 112         state->state = 0;
 113         state->private = 0;
 114         state->tree = NULL;
 115         spin_lock_irqsave(&leak_lock, flags);
 116         list_add(&state->leak_list, &states);
 117         spin_unlock_irqrestore(&leak_lock, flags);
 118
 119         atomic_set(&state->refs, 1);
 120         init_waitqueue_head(&state->wq);
 121         return state;
 122 }
 123 EXPORT_SYMBOL(alloc_extent_state);
 124
 125 void free_extent_state(struct extent_state *state)
 126 {
 127         if (!state)
 128                 return;
 129         if (atomic_dec_and_test(&state->refs)) {
 130                 unsigned long flags;
 131                 WARN_ON(state->tree);
 132                 spin_lock_irqsave(&leak_lock, flags);
 133                 list_del(&state->leak_list);
 134                 spin_unlock_irqrestore(&leak_lock, flags);
 135                 kmem_cache_free(extent_state_cache, state);
 136         }
 137 }
 138 EXPORT_SYMBOL(free_extent_state);
 139
 140 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
 141                                    struct rb_node *node)
 142 {
 143         struct rb_node ** p = &root->rb_node;
 144         struct rb_node * parent = NULL;
 145         struct tree_entry *entry;
 146
 147         while(*p) {
 148                 parent = *p;
 149                 entry = rb_entry(parent, struct tree_entry, rb_node);
 150
 151                 if (offset < entry->start)
 152                         p = &(*p)->rb_left;
 153                 else if (offset > entry->end)
 154                         p = &(*p)->rb_right;
 155                 else
 156                         return parent;
 157         }
 158
 159         entry = rb_entry(node, struct tree_entry, rb_node);
 160         rb_link_node(node, parent, p);
 161         rb_insert_color(node, root);
 162         return NULL;
 163 }
 164
 165 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
 166                                      struct rb_node **prev_ret,
 167                                      struct rb_node **next_ret)
 168 {
 169         struct rb_root *root = &tree->state;
 170         struct rb_node * n = root->rb_node;
 171         struct rb_node *prev = NULL;
 172         struct rb_node *orig_prev = NULL;
 173         struct tree_entry *entry;
 174         struct tree_entry *prev_entry = NULL;
 175
 176         if (tree->last) {
 177                 struct extent_state *state;
 178                 state = tree->last;
 179                 if (state->start <= offset && offset <= state->end)
 180                         return &tree->last->rb_node;
 181         }
 182         while(n) {
 183                 entry = rb_entry(n, struct tree_entry, rb_node);
 184                 prev = n;
 185                 prev_entry = entry;
 186
 187                 if (offset < entry->start)
 188                         n = n->rb_left;
 189                 else if (offset > entry->end)
 190                         n = n->rb_right;
 191                 else {
 192                         tree->last = rb_entry(n, struct extent_state, rb_node);
 193                         return n;
 194                 }
 195         }
 196
 197         if (prev_ret) {
 198                 orig_prev = prev;
 199                 while(prev && offset > prev_entry->end) {
 200                         prev = rb_next(prev);
 201                         prev_entry = rb_entry(prev, struct tree_entry, rb_node);
 202                 }
 203                 *prev_ret = prev;
 204                 prev = orig_prev;
 205         }
 206
 207         if (next_ret) {
 208                 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
 209                 while(prev && offset < prev_entry->start) {
 210                         prev = rb_prev(prev);
 211                         prev_entry = rb_entry(prev, struct tree_entry, rb_node);
 212                 }
 213                 *next_ret = prev;
 214         }
 215         return NULL;
 216 }
 217
 218 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
 219                                           u64 offset)
 220 {
 221         struct rb_node *prev = NULL;
 222         struct rb_node *ret;
 223
 224         ret = __etree_search(tree, offset, &prev, NULL);
 225         if (!ret) {
 226                 if (prev) {
 227                         tree->last = rb_entry(prev, struct extent_state,
 228                                               rb_node);
 229                 }
 230                 return prev;
 231         }
 232         return ret;
 233 }
 234
 235 static struct extent_buffer *buffer_tree_insert(struct extent_io_tree *tree,
 236                                           u64 offset, struct rb_node *node)
 237 {
 238         struct rb_root *root = &tree->buffer;
 239         struct rb_node ** p = &root->rb_node;
 240         struct rb_node * parent = NULL;
 241         struct extent_buffer *eb;
 242
 243         while(*p) {
 244                 parent = *p;
 245                 eb = rb_entry(parent, struct extent_buffer, rb_node);
 246
 247                 if (offset < eb->start)
 248                         p = &(*p)->rb_left;
 249                 else if (offset > eb->start)
 250                         p = &(*p)->rb_right;
 251                 else
 252                         return eb;
 253         }
 254
 255         rb_link_node(node, parent, p);
 256         rb_insert_color(node, root);
 257         return NULL;
 258 }
 259
 260 static struct extent_buffer *buffer_search(struct extent_io_tree *tree,
 261                                            u64 offset)
 262 {
 263         struct rb_root *root = &tree->buffer;
 264         struct rb_node * n = root->rb_node;
 265         struct extent_buffer *eb;
 266
 267         while(n) {
 268                 eb = rb_entry(n, struct extent_buffer, rb_node);
 269                 if (offset < eb->start)
 270                         n = n->rb_left;
 271                 else if (offset > eb->start)
 272                         n = n->rb_right;
 273                 else
 274                         return eb;
 275         }
 276         return NULL;
 277 }
 278
 279 /*
 280  * utility function to look for merge candidates inside a given range.
 281  * Any extents with matching state are merged together into a single
 282  * extent in the tree.  Extents with EXTENT_IO in their state field
 283  * are not merged because the end_io handlers need to be able to do
 284  * operations on them without sleeping (or doing allocations/splits).
 285  *
 286  * This should be called with the tree lock held.
 287  */
 288 static int merge_state(struct extent_io_tree *tree,
 289                        struct extent_state *state)
 290 {
 291         struct extent_state *other;
 292         struct rb_node *other_node;
 293
 294         if (state->state & EXTENT_IOBITS)
 295                 return 0;
 296
 297         other_node = rb_prev(&state->rb_node);
 298         if (other_node) {
 299                 other = rb_entry(other_node, struct extent_state, rb_node);
 300                 if (other->end == state->start - 1 &&
 301                     other->state == state->state) {
 302                         state->start = other->start;
 303                         other->tree = NULL;
 304                         if (tree->last == other)
 305                                 tree->last = state;
 306                         rb_erase(&other->rb_node, &tree->state);
 307                         free_extent_state(other);
 308                 }
 309         }
 310         other_node = rb_next(&state->rb_node);
 311         if (other_node) {
 312                 other = rb_entry(other_node, struct extent_state, rb_node);
 313                 if (other->start == state->end + 1 &&
 314                     other->state == state->state) {
 315                         other->start = state->start;
 316                         state->tree = NULL;
 317                         if (tree->last == state)
 318                                 tree->last = other;
 319                         rb_erase(&state->rb_node, &tree->state);
 320                         free_extent_state(state);
 321                 }
 322         }
 323         return 0;
 324 }
 325
 326 static void set_state_cb(struct extent_io_tree *tree,
 327                          struct extent_state *state,
 328                          unsigned long bits)
 329 {
 330         if (tree->ops && tree->ops->set_bit_hook) {
 331                 tree->ops->set_bit_hook(tree->mapping->host, state->start,
 332                                         state->end, state->state, bits);
 333         }
 334 }
 335
 336 static void clear_state_cb(struct extent_io_tree *tree,
 337                            struct extent_state *state,
 338                            unsigned long bits)
 339 {
 340         if (tree->ops && tree->ops->set_bit_hook) {
 341                 tree->ops->clear_bit_hook(tree->mapping->host, state->start,
 342                                           state->end, state->state, bits);
 343         }
 344 }
 345
 346 /*
 347  * insert an extent_state struct into the tree.  'bits' are set on the
 348  * struct before it is inserted.
 349  *
 350  * This may return -EEXIST if the extent is already there, in which case the
 351  * state struct is freed.
 352  *
 353  * The tree lock is not taken internally.  This is a utility function and
 354  * probably isn't what you want to call (see set/clear_extent_bit).
 355  */
 356 static int insert_state(struct extent_io_tree *tree,
 357                         struct extent_state *state, u64 start, u64 end,
 358                         int bits)
 359 {
 360         struct rb_node *node;
 361
 362         if (end < start) {
 363                 printk("end < start %Lu %Lu\n", end, start);
 364                 WARN_ON(1);
 365         }
 366         if (bits & EXTENT_DIRTY)
 367                 tree->dirty_bytes += end - start + 1;
 368         set_state_cb(tree, state, bits);
 369         state->state |= bits;
 370         state->start = start;
 371         state->end = end;
 372         node = tree_insert(&tree->state, end, &state->rb_node);
 373         if (node) {
 374                 struct extent_state *found;
 375                 found = rb_entry(node, struct extent_state, rb_node);
 376                 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, start, end);
 377                 free_extent_state(state);
 378                 return -EEXIST;
 379         }
 380         state->tree = tree;
 381         tree->last = state;
 382         merge_state(tree, state);
 383         return 0;
 384 }
 385
 386 /*
 387  * split a given extent state struct in two, inserting the preallocated
 388  * struct 'prealloc' as the newly created second half.  'split' indicates an
 389  * offset inside 'orig' where it should be split.
 390  *
 391  * Before calling,
 392  * the tree has 'orig' at [orig->start, orig->end].  After calling, there
 393  * are two extent state structs in the tree:
 394  * prealloc: [orig->start, split - 1]
 395  * orig: [ split, orig->end ]
 396  *
 397  * The tree locks are not taken by this function. They need to be held
 398  * by the caller.
 399  */
 400 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
 401                        struct extent_state *prealloc, u64 split)
 402 {
 403         struct rb_node *node;
 404         prealloc->start = orig->start;
 405         prealloc->end = split - 1;
 406         prealloc->state = orig->state;
 407         orig->start = split;
 408
 409         node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
 410         if (node) {
 411                 struct extent_state *found;
 412                 found = rb_entry(node, struct extent_state, rb_node);
 413                 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, prealloc->start, prealloc->end);
 414                 free_extent_state(prealloc);
 415                 return -EEXIST;
 416         }
 417         prealloc->tree = tree;
 418         return 0;
 419 }
 420
 421 /*
 422  * utility function to clear some bits in an extent state struct.
 423  * it will optionally wake up any one waiting on this state (wake == 1), or
 424  * forcibly remove the state from the tree (delete == 1).
 425  *
 426  * If no bits are set on the state struct after clearing things, the
 427  * struct is freed and removed from the tree
 428  */
 429 static int clear_state_bit(struct extent_io_tree *tree,
 430                             struct extent_state *state, int bits, int wake,
 431                             int delete)
 432 {
 433         int ret = state->state & bits;
 434
 435         if ((bits & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
 436                 u64 range = state->end - state->start + 1;
 437                 WARN_ON(range > tree->dirty_bytes);
 438                 tree->dirty_bytes -= range;
 439         }
 440         clear_state_cb(tree, state, bits);
 441         state->state &= ~bits;
 442         if (wake)
 443                 wake_up(&state->wq);
 444         if (delete || state->state == 0) {
 445                 if (state->tree) {
 446                         clear_state_cb(tree, state, state->state);
 447                         if (tree->last == state) {
 448                                 tree->last = extent_state_next(state);
 449                         }
 450                         rb_erase(&state->rb_node, &tree->state);
 451                         state->tree = NULL;
 452                         free_extent_state(state);
 453                 } else {
 454                         WARN_ON(1);
 455                 }
 456         } else {
 457                 merge_state(tree, state);
 458         }
 459         return ret;
 460 }
 461
 462 /*
 463  * clear some bits on a range in the tree.  This may require splitting
 464  * or inserting elements in the tree, so the gfp mask is used to
 465  * indicate which allocations or sleeping are allowed.
 466  *
 467  * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
 468  * the given range from the tree regardless of state (ie for truncate).
 469  *
 470  * the range [start, end] is inclusive.
 471  *
 472  * This takes the tree lock, and returns < 0 on error, > 0 if any of the
 473  * bits were already set, or zero if none of the bits were already set.
 474  */
 475 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
 476                      int bits, int wake, int delete, gfp_t mask)
 477 {
 478         struct extent_state *state;
 479         struct extent_state *prealloc = NULL;
 480         struct rb_node *node;
 481         unsigned long flags;
 482         int err;
 483         int set = 0;
 484
 485 again:
 486         if (!prealloc && (mask & __GFP_WAIT)) {
 487                 prealloc = alloc_extent_state(mask);
 488                 if (!prealloc)
 489                         return -ENOMEM;
 490         }
 491
 492         spin_lock_irqsave(&tree->lock, flags);
 493         /*
 494          * this search will find the extents that end after
 495          * our range starts
 496          */
 497         node = tree_search(tree, start);
 498         if (!node)
 499                 goto out;
 500         state = rb_entry(node, struct extent_state, rb_node);
 501         if (state->start > end)
 502                 goto out;
 503         WARN_ON(state->end < start);
 504
 505         /*
 506          *     | ---- desired range ---- |
 507          *  | state | or
 508          *  | ------------- state -------------- |
 509          *
 510          * We need to split the extent we found, and may flip
 511          * bits on second half.
 512          *
 513          * If the extent we found extends past our range, we
 514          * just split and search again.  It'll get split again
 515          * the next time though.
 516          *
 517          * If the extent we found is inside our range, we clear
 518          * the desired bit on it.
 519          */
 520
 521         if (state->start < start) {
 522                 if (!prealloc)
 523                         prealloc = alloc_extent_state(GFP_ATOMIC);
 524                 err = split_state(tree, state, prealloc, start);
 525                 BUG_ON(err == -EEXIST);
 526                 prealloc = NULL;
 527                 if (err)
 528                         goto out;
 529                 if (state->end <= end) {
 530                         start = state->end + 1;
 531                         set |= clear_state_bit(tree, state, bits,
 532                                         wake, delete);
 533                 } else {
 534                         start = state->start;
 535                 }
 536                 goto search_again;
 537         }
 538         /*
 539          * | ---- desired range ---- |
 540          *                        | state |
 541          * We need to split the extent, and clear the bit
 542          * on the first half
 543          */
 544         if (state->start <= end && state->end > end) {
 545                 if (!prealloc)
 546                         prealloc = alloc_extent_state(GFP_ATOMIC);
 547                 err = split_state(tree, state, prealloc, end + 1);
 548                 BUG_ON(err == -EEXIST);
 549
 550                 if (wake)
 551                         wake_up(&state->wq);
 552                 set |= clear_state_bit(tree, prealloc, bits,
 553                                        wake, delete);
 554                 prealloc = NULL;
 555                 goto out;
 556         }
 557
 558         start = state->end + 1;
 559         set |= clear_state_bit(tree, state, bits, wake, delete);
 560         goto search_again;
 561
 562 out:
 563         spin_unlock_irqrestore(&tree->lock, flags);
 564         if (prealloc)
 565                 free_extent_state(prealloc);
 566
 567         return set;
 568
 569 search_again:
 570         if (start > end)
 571                 goto out;
 572         spin_unlock_irqrestore(&tree->lock, flags);
 573         if (mask & __GFP_WAIT)
 574                 cond_resched();
 575         goto again;
 576 }
 577 EXPORT_SYMBOL(clear_extent_bit);
 578
 579 static int wait_on_state(struct extent_io_tree *tree,
 580                          struct extent_state *state)
 581 {
 582         DEFINE_WAIT(wait);
 583         prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
 584         spin_unlock_irq(&tree->lock);
 585         schedule();
 586         spin_lock_irq(&tree->lock);
 587         finish_wait(&state->wq, &wait);
 588         return 0;
 589 }
 590
 591 /*
 592  * waits for one or more bits to clear on a range in the state tree.
 593  * The range [start, end] is inclusive.
 594  * The tree lock is taken by this function
 595  */
 596 int wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits)
 597 {
 598         struct extent_state *state;
 599         struct rb_node *node;
 600
 601         spin_lock_irq(&tree->lock);
 602 again:
 603         while (1) {
 604                 /*
 605                  * this search will find all the extents that end after
 606                  * our range starts
 607                  */
 608                 node = tree_search(tree, start);
 609                 if (!node)
 610                         break;
 611
 612                 state = rb_entry(node, struct extent_state, rb_node);
 613
 614                 if (state->start > end)
 615                         goto out;
 616
 617                 if (state->state & bits) {
 618                         start = state->start;
 619                         atomic_inc(&state->refs);
 620                         wait_on_state(tree, state);
 621                         free_extent_state(state);
 622                         goto again;
 623                 }
 624                 start = state->end + 1;
 625
 626                 if (start > end)
 627                         break;
 628
 629                 if (need_resched()) {
 630                         spin_unlock_irq(&tree->lock);
 631                         cond_resched();
 632                         spin_lock_irq(&tree->lock);
 633                 }
 634         }
 635 out:
 636         spin_unlock_irq(&tree->lock);
 637         return 0;
 638 }
 639 EXPORT_SYMBOL(wait_extent_bit);
 640
 641 static void set_state_bits(struct extent_io_tree *tree,
 642                            struct extent_state *state,
 643                            int bits)
 644 {
 645         if ((bits & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
 646                 u64 range = state->end - state->start + 1;
 647                 tree->dirty_bytes += range;
 648         }
 649         set_state_cb(tree, state, bits);
 650         state->state |= bits;
 651 }
 652
 653 /*
 654  * set some bits on a range in the tree.  This may require allocations
 655  * or sleeping, so the gfp mask is used to indicate what is allowed.
 656  *
 657  * If 'exclusive' == 1, this will fail with -EEXIST if some part of the
 658  * range already has the desired bits set.  The start of the existing
 659  * range is returned in failed_start in this case.
 660  *
 661  * [start, end] is inclusive
 662  * This takes the tree lock.
 663  */
 664 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits,
 665                    int exclusive, u64 *failed_start, gfp_t mask)
 666 {
 667         struct extent_state *state;
 668         struct extent_state *prealloc = NULL;
 669         struct rb_node *node;
 670         unsigned long flags;
 671         int err = 0;
 672         int set;
 673         u64 last_start;
 674         u64 last_end;
 675 again:
 676         if (!prealloc && (mask & __GFP_WAIT)) {
 677                 prealloc = alloc_extent_state(mask);
 678                 if (!prealloc)
 679                         return -ENOMEM;
 680         }
 681
 682         spin_lock_irqsave(&tree->lock, flags);
 683         /*
 684          * this search will find all the extents that end after
 685          * our range starts.
 686          */
 687         node = tree_search(tree, start);
 688         if (!node) {
 689                 err = insert_state(tree, prealloc, start, end, bits);
 690                 prealloc = NULL;
 691                 BUG_ON(err == -EEXIST);
 692                 goto out;
 693         }
 694
 695         state = rb_entry(node, struct extent_state, rb_node);
 696         last_start = state->start;
 697         last_end = state->end;
 698
 699         /*
 700          * | ---- desired range ---- |
 701          * | state |
 702          *
 703          * Just lock what we found and keep going
 704          */
 705         if (state->start == start && state->end <= end) {
 706                 set = state->state & bits;
 707                 if (set && exclusive) {
 708                         *failed_start = state->start;
 709                         err = -EEXIST;
 710                         goto out;
 711                 }
 712                 set_state_bits(tree, state, bits);
 713                 start = state->end + 1;
 714                 merge_state(tree, state);
 715                 goto search_again;
 716         }
 717
 718         /*
 719          *     | ---- desired range ---- |
 720          * | state |
 721          *   or
 722          * | ------------- state -------------- |
 723          *
 724          * We need to split the extent we found, and may flip bits on
 725          * second half.
 726          *
 727          * If the extent we found extends past our
 728          * range, we just split and search again.  It'll get split
 729          * again the next time though.
 730          *
 731          * If the extent we found is inside our range, we set the
 732          * desired bit on it.
 733          */
 734         if (state->start < start) {
 735                 set = state->state & bits;
 736                 if (exclusive && set) {
 737                         *failed_start = start;
 738                         err = -EEXIST;
 739                         goto out;
 740                 }
 741                 err = split_state(tree, state, prealloc, start);
 742                 BUG_ON(err == -EEXIST);
 743                 prealloc = NULL;
 744                 if (err)
 745                         goto out;
 746                 if (state->end <= end) {
 747                         set_state_bits(tree, state, bits);
 748                         start = state->end + 1;
 749                         merge_state(tree, state);
 750                 } else {
 751                         start = state->start;
 752                 }
 753                 goto search_again;
 754         }
 755         /*
 756          * | ---- desired range ---- |
 757          *     | state | or               | state |
 758          *
 759          * There's a hole, we need to insert something in it and
 760          * ignore the extent we found.
 761          */
 762         if (state->start > start) {
 763                 u64 this_end;
 764                 if (end < last_start)
 765                         this_end = end;
 766                 else
 767                         this_end = last_start -1;
 768                 err = insert_state(tree, prealloc, start, this_end,
 769                                    bits);
 770                 prealloc = NULL;
 771                 BUG_ON(err == -EEXIST);
 772                 if (err)
 773                         goto out;
 774                 start = this_end + 1;
 775                 goto search_again;
 776         }
 777         /*
 778          * | ---- desired range ---- |
 779          *                        | state |
 780          * We need to split the extent, and set the bit
 781          * on the first half
 782          */
 783         if (state->start <= end && state->end > end) {
 784                 set = state->state & bits;
 785                 if (exclusive && set) {
 786                         *failed_start = start;
 787                         err = -EEXIST;
 788                         goto out;
 789                 }
 790                 err = split_state(tree, state, prealloc, end + 1);
 791                 BUG_ON(err == -EEXIST);
 792
 793                 set_state_bits(tree, prealloc, bits);
 794                 merge_state(tree, prealloc);
 795                 prealloc = NULL;
 796                 goto out;
 797         }
 798
 799         goto search_again;
 800
 801 out:
 802         spin_unlock_irqrestore(&tree->lock, flags);
 803         if (prealloc)
 804                 free_extent_state(prealloc);
 805
 806         return err;
 807
 808 search_again:
 809         if (start > end)
 810                 goto out;
 811         spin_unlock_irqrestore(&tree->lock, flags);
 812         if (mask & __GFP_WAIT)
 813                 cond_resched();
 814         goto again;
 815 }
 816 EXPORT_SYMBOL(set_extent_bit);
 817
 818 /* wrappers around set/clear extent bit */
 819 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
 820                      gfp_t mask)
 821 {
 822         return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL,
 823                               mask);
 824 }
 825 EXPORT_SYMBOL(set_extent_dirty);
 826
 827 int set_extent_ordered(struct extent_io_tree *tree, u64 start, u64 end,
 828                        gfp_t mask)
 829 {
 830         return set_extent_bit(tree, start, end, EXTENT_ORDERED, 0, NULL, mask);
 831 }
 832 EXPORT_SYMBOL(set_extent_ordered);
 833
 834 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
 835                     int bits, gfp_t mask)
 836 {
 837         return set_extent_bit(tree, start, end, bits, 0, NULL,
 838                               mask);
 839 }
 840 EXPORT_SYMBOL(set_extent_bits);
 841
 842 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
 843                       int bits, gfp_t mask)
 844 {
 845         return clear_extent_bit(tree, start, end, bits, 0, 0, mask);
 846 }
 847 EXPORT_SYMBOL(clear_extent_bits);
 848
 849 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
 850                      gfp_t mask)
 851 {
 852         return set_extent_bit(tree, start, end,
 853                               EXTENT_DELALLOC | EXTENT_DIRTY,
 854                               0, NULL, mask);
 855 }
 856 EXPORT_SYMBOL(set_extent_delalloc);
 857
 858 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
 859                        gfp_t mask)
 860 {
 861         return clear_extent_bit(tree, start, end,
 862                                 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, mask);
 863 }
 864 EXPORT_SYMBOL(clear_extent_dirty);
 865
 866 int clear_extent_ordered(struct extent_io_tree *tree, u64 start, u64 end,
 867                          gfp_t mask)
 868 {
 869         return clear_extent_bit(tree, start, end, EXTENT_ORDERED, 1, 0, mask);
 870 }
 871 EXPORT_SYMBOL(clear_extent_ordered);
 872
 873 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
 874                      gfp_t mask)
 875 {
 876         return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL,
 877                               mask);
 878 }
 879 EXPORT_SYMBOL(set_extent_new);
 880
 881 int clear_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
 882                        gfp_t mask)
 883 {
 884         return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0, mask);
 885 }
 886 EXPORT_SYMBOL(clear_extent_new);
 887
 888 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
 889                         gfp_t mask)
 890 {
 891         return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, NULL,
 892                               mask);
 893 }
 894 EXPORT_SYMBOL(set_extent_uptodate);
 895
 896 int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
 897                           gfp_t mask)
 898 {
 899         return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, mask);
 900 }
 901 EXPORT_SYMBOL(clear_extent_uptodate);
 902
 903 int set_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end,
 904                          gfp_t mask)
 905 {
 906         return set_extent_bit(tree, start, end, EXTENT_WRITEBACK,
 907                               0, NULL, mask);
 908 }
 909 EXPORT_SYMBOL(set_extent_writeback);
 910
 911 int clear_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end,
 912                            gfp_t mask)
 913 {
 914         return clear_extent_bit(tree, start, end, EXTENT_WRITEBACK, 1, 0, mask);
 915 }
 916 EXPORT_SYMBOL(clear_extent_writeback);
 917
 918 int wait_on_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end)
 919 {
 920         return wait_extent_bit(tree, start, end, EXTENT_WRITEBACK);
 921 }
 922 EXPORT_SYMBOL(wait_on_extent_writeback);
 923
 924 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
 925 {
 926         int err;
 927         u64 failed_start;
 928         while (1) {
 929                 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1,
 930                                      &failed_start, mask);
 931                 if (err == -EEXIST && (mask & __GFP_WAIT)) {
 932                         wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
 933                         start = failed_start;
 934                 } else {
 935                         break;
 936                 }
 937                 WARN_ON(start > end);
 938         }
 939         return err;
 940 }
 941 EXPORT_SYMBOL(lock_extent);
 942
 943 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end,
 944                   gfp_t mask)
 945 {
 946         return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, mask);
 947 }
 948 EXPORT_SYMBOL(unlock_extent);
 949
 950 /*
 951  * helper function to set pages and extents in the tree dirty
 952  */
 953 int set_range_dirty(struct extent_io_tree *tree, u64 start, u64 end)
 954 {
 955         unsigned long index = start >> PAGE_CACHE_SHIFT;
 956         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
 957         struct page *page;
 958
 959         while (index <= end_index) {
 960                 page = find_get_page(tree->mapping, index);
 961                 BUG_ON(!page);
 962                 __set_page_dirty_nobuffers(page);
 963                 page_cache_release(page);
 964                 index++;
 965         }
 966         set_extent_dirty(tree, start, end, GFP_NOFS);
 967         return 0;
 968 }
 969 EXPORT_SYMBOL(set_range_dirty);
 970
 971 /*
 972  * helper function to set both pages and extents in the tree writeback
 973  */
 974 int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
 975 {
 976         unsigned long index = start >> PAGE_CACHE_SHIFT;
 977         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
 978         struct page *page;
 979
 980         while (index <= end_index) {
 981                 page = find_get_page(tree->mapping, index);
 982                 BUG_ON(!page);
 983                 set_page_writeback(page);
 984                 page_cache_release(page);
 985                 index++;
 986         }
 987         set_extent_writeback(tree, start, end, GFP_NOFS);
 988         return 0;
 989 }
 990 EXPORT_SYMBOL(set_range_writeback);
 991
 992 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
 993                           u64 *start_ret, u64 *end_ret, int bits)
 994 {
 995         struct rb_node *node;
 996         struct extent_state *state;
 997         int ret = 1;
 998
 999         spin_lock_irq(&tree->lock);
1000         /*
1001          * this search will find all the extents that end after
1002          * our range starts.
1003          */
1004         node = tree_search(tree, start);
1005         if (!node) {
1006                 goto out;
1007         }
1008
1009         while(1) {
1010                 state = rb_entry(node, struct extent_state, rb_node);
1011                 if (state->end >= start && (state->state & bits)) {
1012                         *start_ret = state->start;
1013                         *end_ret = state->end;
1014                         ret = 0;
1015                         break;
1016                 }
1017                 node = rb_next(node);
1018                 if (!node)
1019                         break;
1020         }
1021 out:
1022         spin_unlock_irq(&tree->lock);
1023         return ret;
1024 }
1025 EXPORT_SYMBOL(find_first_extent_bit);
1026
1027 struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree,
1028                                                  u64 start, int bits)
1029 {
1030         struct rb_node *node;
1031         struct extent_state *state;
1032
1033         /*
1034          * this search will find all the extents that end after
1035          * our range starts.
1036          */
1037         node = tree_search(tree, start);
1038         if (!node) {
1039                 goto out;
1040         }
1041
1042         while(1) {
1043                 state = rb_entry(node, struct extent_state, rb_node);
1044                 if (state->end >= start && (state->state & bits)) {
1045                         return state;
1046                 }
1047                 node = rb_next(node);
1048                 if (!node)
1049                         break;
1050         }
1051 out:
1052         return NULL;
1053 }
1054 EXPORT_SYMBOL(find_first_extent_bit_state);
1055
1056 u64 find_lock_delalloc_range(struct extent_io_tree *tree,
1057                              u64 *start, u64 *end, u64 max_bytes)
1058 {
1059         struct rb_node *node;
1060         struct extent_state *state;
1061         u64 cur_start = *start;
1062         u64 found = 0;
1063         u64 total_bytes = 0;
1064
1065         spin_lock_irq(&tree->lock);
1066         /*
1067          * this search will find all the extents that end after
1068          * our range starts.
1069          */
1070 search_again:
1071         node = tree_search(tree, cur_start);
1072         if (!node) {
1073                 if (!found)
1074                         *end = (u64)-1;
1075                 goto out;
1076         }
1077
1078         while(1) {
1079                 state = rb_entry(node, struct extent_state, rb_node);
1080                 if (found && state->start != cur_start) {
1081                         goto out;
1082                 }
1083                 if (!(state->state & EXTENT_DELALLOC)) {
1084                         if (!found)
1085                                 *end = state->end;
1086                         goto out;
1087                 }
1088                 if (!found) {
1089                         struct extent_state *prev_state;
1090                         struct rb_node *prev_node = node;
1091                         while(1) {
1092                                 prev_node = rb_prev(prev_node);
1093                                 if (!prev_node)
1094                                         break;
1095                                 prev_state = rb_entry(prev_node,
1096                                                       struct extent_state,
1097                                                       rb_node);
1098                                 if (!(prev_state->state & EXTENT_DELALLOC))
1099                                         break;
1100                                 state = prev_state;
1101                                 node = prev_node;
1102                         }
1103                 }
1104                 if (state->state & EXTENT_LOCKED) {
1105                         DEFINE_WAIT(wait);
1106                         atomic_inc(&state->refs);
1107                         prepare_to_wait(&state->wq, &wait,
1108                                         TASK_UNINTERRUPTIBLE);
1109                         spin_unlock_irq(&tree->lock);
1110                         schedule();
1111                         spin_lock_irq(&tree->lock);
1112                         finish_wait(&state->wq, &wait);
1113                         free_extent_state(state);
1114                         goto search_again;
1115                 }
1116                 set_state_cb(tree, state, EXTENT_LOCKED);
1117                 state->state |= EXTENT_LOCKED;
1118                 if (!found)
1119                         *start = state->start;
1120                 found++;
1121                 *end = state->end;
1122                 cur_start = state->end + 1;
1123                 node = rb_next(node);
1124                 if (!node)
1125                         break;
1126                 total_bytes += state->end - state->start + 1;
1127                 if (total_bytes >= max_bytes)
1128                         break;
1129         }
1130 out:
1131         spin_unlock_irq(&tree->lock);
1132         return found;
1133 }
1134
1135 u64 count_range_bits(struct extent_io_tree *tree,
1136                      u64 *start, u64 search_end, u64 max_bytes,
1137                      unsigned long bits)
1138 {
1139         struct rb_node *node;
1140         struct extent_state *state;
1141         u64 cur_start = *start;
1142         u64 total_bytes = 0;
1143         int found = 0;
1144
1145         if (search_end <= cur_start) {
1146                 printk("search_end %Lu start %Lu\n", search_end, cur_start);
1147                 WARN_ON(1);
1148                 return 0;
1149         }
1150
1151         spin_lock_irq(&tree->lock);
1152         if (cur_start == 0 && bits == EXTENT_DIRTY) {
1153                 total_bytes = tree->dirty_bytes;
1154                 goto out;
1155         }
1156         /*
1157          * this search will find all the extents that end after
1158          * our range starts.
1159          */
1160         node = tree_search(tree, cur_start);
1161         if (!node) {
1162                 goto out;
1163         }
1164
1165         while(1) {
1166                 state = rb_entry(node, struct extent_state, rb_node);
1167                 if (state->start > search_end)
1168                         break;
1169                 if (state->end >= cur_start && (state->state & bits)) {
1170                         total_bytes += min(search_end, state->end) + 1 -
1171                                        max(cur_start, state->start);
1172                         if (total_bytes >= max_bytes)
1173                                 break;
1174                         if (!found) {
1175                                 *start = state->start;
1176                                 found = 1;
1177                         }
1178                 }
1179                 node = rb_next(node);
1180                 if (!node)
1181                         break;
1182         }
1183 out:
1184         spin_unlock_irq(&tree->lock);
1185         return total_bytes;
1186 }
1187 /*
1188  * helper function to lock both pages and extents in the tree.
1189  * pages must be locked first.
1190  */
1191 int lock_range(struct extent_io_tree *tree, u64 start, u64 end)
1192 {
1193         unsigned long index = start >> PAGE_CACHE_SHIFT;
1194         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1195         struct page *page;
1196         int err;
1197
1198         while (index <= end_index) {
1199                 page = grab_cache_page(tree->mapping, index);
1200                 if (!page) {
1201                         err = -ENOMEM;
1202                         goto failed;
1203                 }
1204                 if (IS_ERR(page)) {
1205                         err = PTR_ERR(page);
1206                         goto failed;
1207                 }
1208                 index++;
1209         }
1210         lock_extent(tree, start, end, GFP_NOFS);
1211         return 0;
1212
1213 failed:
1214         /*
1215          * we failed above in getting the page at 'index', so we undo here
1216          * up to but not including the page at 'index'
1217          */
1218         end_index = index;
1219         index = start >> PAGE_CACHE_SHIFT;
1220         while (index < end_index) {
1221                 page = find_get_page(tree->mapping, index);
1222                 unlock_page(page);
1223                 page_cache_release(page);
1224                 index++;
1225         }
1226         return err;
1227 }
1228 EXPORT_SYMBOL(lock_range);
1229
1230 /*
1231  * helper function to unlock both pages and extents in the tree.
1232  */
1233 int unlock_range(struct extent_io_tree *tree, u64 start, u64 end)
1234 {
1235         unsigned long index = start >> PAGE_CACHE_SHIFT;
1236         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1237         struct page *page;
1238
1239         while (index <= end_index) {
1240                 page = find_get_page(tree->mapping, index);
1241                 unlock_page(page);
1242                 page_cache_release(page);
1243                 index++;
1244         }
1245         unlock_extent(tree, start, end, GFP_NOFS);
1246         return 0;
1247 }
1248 EXPORT_SYMBOL(unlock_range);
1249
1250 int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1251 {
1252         struct rb_node *node;
1253         struct extent_state *state;
1254         int ret = 0;
1255
1256         spin_lock_irq(&tree->lock);
1257         /*
1258          * this search will find all the extents that end after
1259          * our range starts.
1260          */
1261         node = tree_search(tree, start);
1262         if (!node) {
1263                 ret = -ENOENT;
1264                 goto out;
1265         }
1266         state = rb_entry(node, struct extent_state, rb_node);
1267         if (state->start != start) {
1268                 ret = -ENOENT;
1269                 goto out;
1270         }
1271         state->private = private;
1272 out:
1273         spin_unlock_irq(&tree->lock);
1274         return ret;
1275 }
1276
1277 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1278 {
1279         struct rb_node *node;
1280         struct extent_state *state;
1281         int ret = 0;
1282
1283         spin_lock_irq(&tree->lock);
1284         /*
1285          * this search will find all the extents that end after
1286          * our range starts.
1287          */
1288         node = tree_search(tree, start);
1289         if (!node) {
1290                 ret = -ENOENT;
1291                 goto out;
1292         }
1293         state = rb_entry(node, struct extent_state, rb_node);
1294         if (state->start != start) {
1295                 ret = -ENOENT;
1296                 goto out;
1297         }
1298         *private = state->private;
1299 out:
1300         spin_unlock_irq(&tree->lock);
1301         return ret;
1302 }
1303
1304 /*
1305  * searches a range in the state tree for a given mask.
1306  * If 'filled' == 1, this returns 1 only if every extent in the tree
1307  * has the bits set.  Otherwise, 1 is returned if any bit in the
1308  * range is found set.
1309  */
1310 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1311                    int bits, int filled)
1312 {
1313         struct extent_state *state = NULL;
1314         struct rb_node *node;
1315         int bitset = 0;
1316         unsigned long flags;
1317
1318         spin_lock_irqsave(&tree->lock, flags);
1319         node = tree_search(tree, start);
1320         while (node && start <= end) {
1321                 state = rb_entry(node, struct extent_state, rb_node);
1322
1323                 if (filled && state->start > start) {
1324                         bitset = 0;
1325                         break;
1326                 }
1327
1328                 if (state->start > end)
1329                         break;
1330
1331                 if (state->state & bits) {
1332                         bitset = 1;
1333                         if (!filled)
1334                                 break;
1335                 } else if (filled) {
1336                         bitset = 0;
1337                         break;
1338                 }
1339                 start = state->end + 1;
1340                 if (start > end)
1341                         break;
1342                 node = rb_next(node);
1343                 if (!node) {
1344                         if (filled)
1345                                 bitset = 0;
1346                         break;
1347                 }
1348         }
1349         spin_unlock_irqrestore(&tree->lock, flags);
1350         return bitset;
1351 }
1352 EXPORT_SYMBOL(test_range_bit);
1353
1354 /*
1355  * helper function to set a given page up to date if all the
1356  * extents in the tree for that page are up to date
1357  */
1358 static int check_page_uptodate(struct extent_io_tree *tree,
1359                                struct page *page)
1360 {
1361         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1362         u64 end = start + PAGE_CACHE_SIZE - 1;
1363         if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1))
1364                 SetPageUptodate(page);
1365         return 0;
1366 }
1367
1368 /*
1369  * helper function to unlock a page if all the extents in the tree
1370  * for that page are unlocked
1371  */
1372 static int check_page_locked(struct extent_io_tree *tree,
1373                              struct page *page)
1374 {
1375         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1376         u64 end = start + PAGE_CACHE_SIZE - 1;
1377         if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0))
1378                 unlock_page(page);
1379         return 0;
1380 }
1381
1382 /*
1383  * helper function to end page writeback if all the extents
1384  * in the tree for that page are done with writeback
1385  */
1386 static int check_page_writeback(struct extent_io_tree *tree,
1387                              struct page *page)
1388 {
1389         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1390         u64 end = start + PAGE_CACHE_SIZE - 1;
1391         if (!test_range_bit(tree, start, end, EXTENT_WRITEBACK, 0))
1392                 end_page_writeback(page);
1393         return 0;
1394 }
1395
1396 /* lots and lots of room for performance fixes in the end_bio funcs */
1397
1398 /*
1399  * after a writepage IO is done, we need to:
1400  * clear the uptodate bits on error
1401  * clear the writeback bits in the extent tree for this IO
1402  * end_page_writeback if the page has no more pending IO
1403  *
1404  * Scheduling is not allowed, so the extent state tree is expected
1405  * to have one and only one object corresponding to this IO.
1406  */
1407 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1408 static void end_bio_extent_writepage(struct bio *bio, int err)
1409 #else
1410 static int end_bio_extent_writepage(struct bio *bio,
1411                                    unsigned int bytes_done, int err)
1412 #endif
1413 {
1414         int uptodate = err == 0;
1415         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1416         struct extent_state *state = bio->bi_private;
1417         struct extent_io_tree *tree = state->tree;
1418         struct rb_node *node;
1419         u64 start;
1420         u64 end;
1421         u64 cur;
1422         int whole_page;
1423         int ret;
1424         unsigned long flags;
1425
1426 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1427         if (bio->bi_size)
1428                 return 1;
1429 #endif
1430         do {
1431                 struct page *page = bvec->bv_page;
1432                 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1433                          bvec->bv_offset;
1434                 end = start + bvec->bv_len - 1;
1435
1436                 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1437                         whole_page = 1;
1438                 else
1439                         whole_page = 0;
1440
1441                 if (--bvec >= bio->bi_io_vec)
1442                         prefetchw(&bvec->bv_page->flags);
1443                 if (tree->ops && tree->ops->writepage_end_io_hook) {
1444                         ret = tree->ops->writepage_end_io_hook(page, start,
1445                                                        end, state, uptodate);
1446                         if (ret)
1447                                 uptodate = 0;
1448                 }
1449
1450                 if (!uptodate && tree->ops &&
1451                     tree->ops->writepage_io_failed_hook) {
1452                         ret = tree->ops->writepage_io_failed_hook(bio, page,
1453                                                          start, end, state);
1454                         if (ret == 0) {
1455                                 state = NULL;
1456                                 uptodate = (err == 0);
1457                                 continue;
1458                         }
1459                 }
1460
1461                 if (!uptodate) {
1462                         clear_extent_uptodate(tree, start, end, GFP_ATOMIC);
1463                         ClearPageUptodate(page);
1464                         SetPageError(page);
1465                 }
1466
1467                 /*
1468                  * bios can get merged in funny ways, and so we need to
1469                  * be careful with the state variable.  We know the
1470                  * state won't be merged with others because it has
1471                  * WRITEBACK set, but we can't be sure each biovec is
1472                  * sequential in the file.  So, if our cached state
1473                  * doesn't match the expected end, search the tree
1474                  * for the correct one.
1475                  */
1476
1477                 spin_lock_irqsave(&tree->lock, flags);
1478                 if (!state || state->end != end) {
1479                         state = NULL;
1480                         node = __etree_search(tree, start, NULL, NULL);
1481                         if (node) {
1482                                 state = rb_entry(node, struct extent_state,
1483                                                  rb_node);
1484                                 if (state->end != end ||
1485                                     !(state->state & EXTENT_WRITEBACK))
1486                                         state = NULL;
1487                         }
1488                         if (!state) {
1489                                 spin_unlock_irqrestore(&tree->lock, flags);
1490                                 clear_extent_writeback(tree, start,
1491                                                        end, GFP_ATOMIC);
1492                                 goto next_io;
1493                         }
1494                 }
1495                 cur = end;
1496                 while(1) {
1497                         struct extent_state *clear = state;
1498                         cur = state->start;
1499                         node = rb_prev(&state->rb_node);
1500                         if (node) {
1501                                 state = rb_entry(node,
1502                                                  struct extent_state,
1503                                                  rb_node);
1504                         } else {
1505                                 state = NULL;
1506                         }
1507
1508                         clear_state_bit(tree, clear, EXTENT_WRITEBACK,
1509                                         1, 0);
1510                         if (cur == start)
1511                                 break;
1512                         if (cur < start) {
1513                                 WARN_ON(1);
1514                                 break;
1515                         }
1516                         if (!node)
1517                                 break;
1518                 }
1519                 /* before releasing the lock, make sure the next state
1520                  * variable has the expected bits set and corresponds
1521                  * to the correct offsets in the file
1522                  */
1523                 if (state && (state->end + 1 != start ||
1524                     !(state->state & EXTENT_WRITEBACK))) {
1525                         state = NULL;
1526                 }
1527                 spin_unlock_irqrestore(&tree->lock, flags);
1528 next_io:
1529
1530                 if (whole_page)
1531                         end_page_writeback(page);
1532                 else
1533                         check_page_writeback(tree, page);
1534         } while (bvec >= bio->bi_io_vec);
1535         bio_put(bio);
1536 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1537         return 0;
1538 #endif
1539 }
1540
1541 /*
1542  * after a readpage IO is done, we need to:
1543  * clear the uptodate bits on error
1544  * set the uptodate bits if things worked
1545  * set the page up to date if all extents in the tree are uptodate
1546  * clear the lock bit in the extent tree
1547  * unlock the page if there are no other extents locked for it
1548  *
1549  * Scheduling is not allowed, so the extent state tree is expected
1550  * to have one and only one object corresponding to this IO.
1551  */
1552 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1553 static void end_bio_extent_readpage(struct bio *bio, int err)
1554 #else
1555 static int end_bio_extent_readpage(struct bio *bio,
1556                                    unsigned int bytes_done, int err)
1557 #endif
1558 {
1559         int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1560         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1561         struct extent_state *state = bio->bi_private;
1562         struct extent_io_tree *tree = state->tree;
1563         struct rb_node *node;
1564         u64 start;
1565         u64 end;
1566         u64 cur;
1567         unsigned long flags;
1568         int whole_page;
1569         int ret;
1570
1571 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1572         if (bio->bi_size)
1573                 return 1;
1574 #endif
1575
1576         do {
1577                 struct page *page = bvec->bv_page;
1578                 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1579                         bvec->bv_offset;
1580                 end = start + bvec->bv_len - 1;
1581
1582                 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1583                         whole_page = 1;
1584                 else
1585                         whole_page = 0;
1586
1587                 if (--bvec >= bio->bi_io_vec)
1588                         prefetchw(&bvec->bv_page->flags);
1589
1590                 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
1591                         ret = tree->ops->readpage_end_io_hook(page, start, end,
1592                                                               state);
1593                         if (ret)
1594                                 uptodate = 0;
1595                 }
1596                 if (!uptodate && tree->ops &&
1597                     tree->ops->readpage_io_failed_hook) {
1598                         ret = tree->ops->readpage_io_failed_hook(bio, page,
1599                                                          start, end, state);
1600                         if (ret == 0) {
1601                                 state = NULL;
1602                                 uptodate =
1603                                         test_bit(BIO_UPTODATE, &bio->bi_flags);
1604                                 continue;
1605                         }
1606                 }
1607
1608                 spin_lock_irqsave(&tree->lock, flags);
1609                 if (!state || state->end != end) {
1610                         state = NULL;
1611                         node = __etree_search(tree, start, NULL, NULL);
1612                         if (node) {
1613                                 state = rb_entry(node, struct extent_state,
1614                                                  rb_node);
1615                                 if (state->end != end ||
1616                                     !(state->state & EXTENT_LOCKED))
1617                                         state = NULL;
1618                         }
1619                         if (!state) {
1620                                 spin_unlock_irqrestore(&tree->lock, flags);
1621                                 if (uptodate)
1622                                         set_extent_uptodate(tree, start, end,
1623                                                             GFP_ATOMIC);
1624                                 unlock_extent(tree, start, end, GFP_ATOMIC);
1625                                 goto next_io;
1626                         }
1627                 }
1628
1629                 cur = end;
1630                 while(1) {
1631                         struct extent_state *clear = state;
1632                         cur = state->start;
1633                         node = rb_prev(&state->rb_node);
1634                         if (node) {
1635                                 state = rb_entry(node,
1636                                          struct extent_state,
1637                                          rb_node);
1638                         } else {
1639                                 state = NULL;
1640                         }
1641                         if (uptodate) {
1642                                 set_state_cb(tree, clear, EXTENT_UPTODATE);
1643                                 clear->state |= EXTENT_UPTODATE;
1644                         }
1645                         clear_state_bit(tree, clear, EXTENT_LOCKED,
1646                                         1, 0);
1647                         if (cur == start)
1648                                 break;
1649                         if (cur < start) {
1650                                 WARN_ON(1);
1651                                 break;
1652                         }
1653                         if (!node)
1654                                 break;
1655                 }
1656                 /* before releasing the lock, make sure the next state
1657                  * variable has the expected bits set and corresponds
1658                  * to the correct offsets in the file
1659                  */
1660                 if (state && (state->end + 1 != start ||
1661                     !(state->state & EXTENT_LOCKED))) {
1662                         state = NULL;
1663                 }
1664                 spin_unlock_irqrestore(&tree->lock, flags);
1665 next_io:
1666                 if (whole_page) {
1667                         if (uptodate) {
1668                                 SetPageUptodate(page);
1669                         } else {
1670                                 ClearPageUptodate(page);
1671                                 SetPageError(page);
1672                         }
1673                         unlock_page(page);
1674                 } else {
1675                         if (uptodate) {
1676                                 check_page_uptodate(tree, page);
1677                         } else {
1678                                 ClearPageUptodate(page);
1679                                 SetPageError(page);
1680                         }
1681                         check_page_locked(tree, page);
1682                 }
1683         } while (bvec >= bio->bi_io_vec);
1684
1685         bio_put(bio);
1686 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1687         return 0;
1688 #endif
1689 }
1690
1691 /*
1692  * IO done from prepare_write is pretty simple, we just unlock
1693  * the structs in the extent tree when done, and set the uptodate bits
1694  * as appropriate.
1695  */
1696 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1697 static void end_bio_extent_preparewrite(struct bio *bio, int err)
1698 #else
1699 static int end_bio_extent_preparewrite(struct bio *bio,
1700                                        unsigned int bytes_done, int err)
1701 #endif
1702 {
1703         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1704         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1705         struct extent_state *state = bio->bi_private;
1706         struct extent_io_tree *tree = state->tree;
1707         u64 start;
1708         u64 end;
1709
1710 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1711         if (bio->bi_size)
1712                 return 1;
1713 #endif
1714
1715         do {
1716                 struct page *page = bvec->bv_page;
1717                 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1718                         bvec->bv_offset;
1719                 end = start + bvec->bv_len - 1;
1720
1721                 if (--bvec >= bio->bi_io_vec)
1722                         prefetchw(&bvec->bv_page->flags);
1723
1724                 if (uptodate) {
1725                         set_extent_uptodate(tree, start, end, GFP_ATOMIC);
1726                 } else {
1727                         ClearPageUptodate(page);
1728                         SetPageError(page);
1729                 }
1730
1731                 unlock_extent(tree, start, end, GFP_ATOMIC);
1732
1733         } while (bvec >= bio->bi_io_vec);
1734
1735         bio_put(bio);
1736 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1737         return 0;
1738 #endif
1739 }
1740
1741 static struct bio *
1742 extent_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
1743                  gfp_t gfp_flags)
1744 {
1745         struct bio *bio;
1746
1747         bio = bio_alloc(gfp_flags, nr_vecs);
1748
1749         if (bio == NULL && (current->flags & PF_MEMALLOC)) {
1750                 while (!bio && (nr_vecs /= 2))
1751                         bio = bio_alloc(gfp_flags, nr_vecs);
1752         }
1753
1754         if (bio) {
1755                 bio->bi_size = 0;
1756                 bio->bi_bdev = bdev;
1757                 bio->bi_sector = first_sector;
1758         }
1759         return bio;
1760 }
1761
1762 static int submit_one_bio(int rw, struct bio *bio, int mirror_num)
1763 {
1764         int ret = 0;
1765         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1766         struct page *page = bvec->bv_page;
1767         struct extent_io_tree *tree = bio->bi_private;
1768         struct rb_node *node;
1769         struct extent_state *state;
1770         u64 start;
1771         u64 end;
1772
1773         start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
1774         end = start + bvec->bv_len - 1;
1775
1776         spin_lock_irq(&tree->lock);
1777         node = __etree_search(tree, start, NULL, NULL);
1778         BUG_ON(!node);
1779         state = rb_entry(node, struct extent_state, rb_node);
1780         while(state->end < end) {
1781                 node = rb_next(node);
1782                 state = rb_entry(node, struct extent_state, rb_node);
1783         }
1784         BUG_ON(state->end != end);
1785         spin_unlock_irq(&tree->lock);
1786
1787         bio->bi_private = state;
1788
1789         bio_get(bio);
1790
1791         if (tree->ops && tree->ops->submit_bio_hook)
1792                 tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
1793                                            mirror_num);
1794         else
1795                 submit_bio(rw, bio);
1796         if (bio_flagged(bio, BIO_EOPNOTSUPP))
1797                 ret = -EOPNOTSUPP;
1798         bio_put(bio);
1799         return ret;
1800 }
1801
1802 static int submit_extent_page(int rw, struct extent_io_tree *tree,
1803                               struct page *page, sector_t sector,
1804                               size_t size, unsigned long offset,
1805                               struct block_device *bdev,
1806                               struct bio **bio_ret,
1807                               unsigned long max_pages,
1808                               bio_end_io_t end_io_func,
1809                               int mirror_num)
1810 {
1811         int ret = 0;
1812         struct bio *bio;
1813         int nr;
1814
1815         if (bio_ret && *bio_ret) {
1816                 bio = *bio_ret;
1817                 if (bio->bi_sector + (bio->bi_size >> 9) != sector ||
1818                     (tree->ops && tree->ops->merge_bio_hook &&
1819                      tree->ops->merge_bio_hook(page, offset, size, bio)) ||
1820                     bio_add_page(bio, page, size, offset) < size) {
1821                         ret = submit_one_bio(rw, bio, mirror_num);
1822                         bio = NULL;
1823                 } else {
1824                         return 0;
1825                 }
1826         }
1827         nr = bio_get_nr_vecs(bdev);
1828         bio = extent_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
1829         if (!bio) {
1830                 printk("failed to allocate bio nr %d\n", nr);
1831         }
1832
1833
1834         bio_add_page(bio, page, size, offset);
1835         bio->bi_end_io = end_io_func;
1836         bio->bi_private = tree;
1837
1838         if (bio_ret) {
1839                 *bio_ret = bio;
1840         } else {
1841                 ret = submit_one_bio(rw, bio, mirror_num);
1842         }
1843
1844         return ret;
1845 }
1846
1847 void set_page_extent_mapped(struct page *page)
1848 {
1849         if (!PagePrivate(page)) {
1850                 SetPagePrivate(page);
1851                 page_cache_get(page);
1852                 set_page_private(page, EXTENT_PAGE_PRIVATE);
1853         }
1854 }
1855
1856 void set_page_extent_head(struct page *page, unsigned long len)
1857 {
1858         set_page_private(page, EXTENT_PAGE_PRIVATE_FIRST_PAGE | len << 2);
1859 }
1860
1861 /*
1862  * basic readpage implementation.  Locked extent state structs are inserted
1863  * into the tree that are removed when the IO is done (by the end_io
1864  * handlers)
1865  */
1866 static int __extent_read_full_page(struct extent_io_tree *tree,
1867                                    struct page *page,
1868                                    get_extent_t *get_extent,
1869                                    struct bio **bio, int mirror_num)
1870 {
1871         struct inode *inode = page->mapping->host;
1872         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1873         u64 page_end = start + PAGE_CACHE_SIZE - 1;
1874         u64 end;
1875         u64 cur = start;
1876         u64 extent_offset;
1877         u64 last_byte = i_size_read(inode);
1878         u64 block_start;
1879         u64 cur_end;
1880         sector_t sector;
1881         struct extent_map *em;
1882         struct block_device *bdev;
1883         int ret;
1884         int nr = 0;
1885         size_t page_offset = 0;
1886         size_t iosize;
1887         size_t blocksize = inode->i_sb->s_blocksize;
1888
1889         set_page_extent_mapped(page);
1890
1891         end = page_end;
1892         lock_extent(tree, start, end, GFP_NOFS);
1893
1894         while (cur <= end) {
1895                 if (cur >= last_byte) {
1896                         char *userpage;
1897                         iosize = PAGE_CACHE_SIZE - page_offset;
1898                         userpage = kmap_atomic(page, KM_USER0);
1899                         memset(userpage + page_offset, 0, iosize);
1900                         flush_dcache_page(page);
1901                         kunmap_atomic(userpage, KM_USER0);
1902                         set_extent_uptodate(tree, cur, cur + iosize - 1,
1903                                             GFP_NOFS);
1904                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1905                         break;
1906                 }
1907                 em = get_extent(inode, page, page_offset, cur,
1908                                 end - cur + 1, 0);
1909                 if (IS_ERR(em) || !em) {
1910                         SetPageError(page);
1911                         unlock_extent(tree, cur, end, GFP_NOFS);
1912                         break;
1913                 }
1914                 extent_offset = cur - em->start;
1915                 if (extent_map_end(em) <= cur) {
1916 printk("bad mapping em [%Lu %Lu] cur %Lu\n", em->start, extent_map_end(em), cur);
1917                 }
1918                 BUG_ON(extent_map_end(em) <= cur);
1919                 if (end < cur) {
1920 printk("2bad mapping end %Lu cur %Lu\n", end, cur);
1921                 }
1922                 BUG_ON(end < cur);
1923
1924                 iosize = min(extent_map_end(em) - cur, end - cur + 1);
1925                 cur_end = min(extent_map_end(em) - 1, end);
1926                 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
1927                 sector = (em->block_start + extent_offset) >> 9;
1928                 bdev = em->bdev;
1929                 block_start = em->block_start;
1930                 free_extent_map(em);
1931                 em = NULL;
1932
1933                 /* we've found a hole, just zero and go on */
1934                 if (block_start == EXTENT_MAP_HOLE) {
1935                         char *userpage;
1936                         userpage = kmap_atomic(page, KM_USER0);
1937                         memset(userpage + page_offset, 0, iosize);
1938                         flush_dcache_page(page);
1939                         kunmap_atomic(userpage, KM_USER0);
1940
1941                         set_extent_uptodate(tree, cur, cur + iosize - 1,
1942                                             GFP_NOFS);
1943                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1944                         cur = cur + iosize;
1945                         page_offset += iosize;
1946                         continue;
1947                 }
1948                 /* the get_extent function already copied into the page */
1949                 if (test_range_bit(tree, cur, cur_end, EXTENT_UPTODATE, 1)) {
1950                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1951                         cur = cur + iosize;
1952                         page_offset += iosize;
1953                         continue;
1954                 }
1955                 /* we have an inline extent but it didn't get marked up
1956                  * to date.  Error out
1957                  */
1958                 if (block_start == EXTENT_MAP_INLINE) {
1959                         SetPageError(page);
1960                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1961                         cur = cur + iosize;
1962                         page_offset += iosize;
1963                         continue;
1964                 }
1965
1966                 ret = 0;
1967                 if (tree->ops && tree->ops->readpage_io_hook) {
1968                         ret = tree->ops->readpage_io_hook(page, cur,
1969                                                           cur + iosize - 1);
1970                 }
1971                 if (!ret) {
1972                         unsigned long nr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
1973                         nr -= page->index;
1974                         ret = submit_extent_page(READ, tree, page,
1975                                          sector, iosize, page_offset,
1976                                          bdev, bio, nr,
1977                                          end_bio_extent_readpage, mirror_num);
1978                 }
1979                 if (ret)
1980                         SetPageError(page);
1981                 cur = cur + iosize;
1982                 page_offset += iosize;
1983                 nr++;
1984         }
1985         if (!nr) {
1986                 if (!PageError(page))
1987                         SetPageUptodate(page);
1988                 unlock_page(page);
1989         }
1990         return 0;
1991 }
1992
1993 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
1994                             get_extent_t *get_extent)
1995 {
1996         struct bio *bio = NULL;
1997         int ret;
1998
1999         ret = __extent_read_full_page(tree, page, get_extent, &bio, 0);
2000         if (bio)
2001                 submit_one_bio(READ, bio, 0);
2002         return ret;
2003 }
2004 EXPORT_SYMBOL(extent_read_full_page);
2005
2006 /*
2007  * the writepage semantics are similar to regular writepage.  extent
2008  * records are inserted to lock ranges in the tree, and as dirty areas
2009  * are found, they are marked writeback.  Then the lock bits are removed
2010  * and the end_io handler clears the writeback ranges
2011  */
2012 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2013                               void *data)
2014 {
2015         struct inode *inode = page->mapping->host;
2016         struct extent_page_data *epd = data;
2017         struct extent_io_tree *tree = epd->tree;
2018         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2019         u64 delalloc_start;
2020         u64 page_end = start + PAGE_CACHE_SIZE - 1;
2021         u64 end;
2022         u64 cur = start;
2023         u64 extent_offset;
2024         u64 last_byte = i_size_read(inode);
2025         u64 block_start;
2026         u64 iosize;
2027         u64 unlock_start;
2028         sector_t sector;
2029         struct extent_map *em;
2030         struct block_device *bdev;
2031         int ret;
2032         int nr = 0;
2033         size_t pg_offset = 0;
2034         size_t blocksize;
2035         loff_t i_size = i_size_read(inode);
2036         unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
2037         u64 nr_delalloc;
2038         u64 delalloc_end;
2039
2040         WARN_ON(!PageLocked(page));
2041         pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
2042         if (page->index > end_index ||
2043            (page->index == end_index && !pg_offset)) {
2044                 page->mapping->a_ops->invalidatepage(page, 0);
2045                 unlock_page(page);
2046                 return 0;
2047         }
2048
2049         if (page->index == end_index) {
2050                 char *userpage;
2051
2052                 userpage = kmap_atomic(page, KM_USER0);
2053                 memset(userpage + pg_offset, 0,
2054                        PAGE_CACHE_SIZE - pg_offset);
2055                 kunmap_atomic(userpage, KM_USER0);
2056                 flush_dcache_page(page);
2057         }
2058         pg_offset = 0;
2059
2060         set_page_extent_mapped(page);
2061
2062         delalloc_start = start;
2063         delalloc_end = 0;
2064         while(delalloc_end < page_end) {
2065                 nr_delalloc = find_lock_delalloc_range(tree, &delalloc_start,
2066                                                        &delalloc_end,
2067                                                        128 * 1024 * 1024);
2068                 if (nr_delalloc == 0) {
2069                         delalloc_start = delalloc_end + 1;
2070                         continue;
2071                 }
2072                 tree->ops->fill_delalloc(inode, delalloc_start,
2073                                          delalloc_end);
2074                 clear_extent_bit(tree, delalloc_start,
2075                                  delalloc_end,
2076                                  EXTENT_LOCKED | EXTENT_DELALLOC,
2077                                  1, 0, GFP_NOFS);
2078                 delalloc_start = delalloc_end + 1;
2079         }
2080         lock_extent(tree, start, page_end, GFP_NOFS);
2081         unlock_start = start;
2082
2083         if (tree->ops && tree->ops->writepage_start_hook) {
2084                 ret = tree->ops->writepage_start_hook(page, start, page_end);
2085                 if (ret == -EAGAIN) {
2086                         unlock_extent(tree, start, page_end, GFP_NOFS);
2087                         redirty_page_for_writepage(wbc, page);
2088                         unlock_page(page);
2089                         return 0;
2090                 }
2091         }
2092
2093         end = page_end;
2094         if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0)) {
2095                 printk("found delalloc bits after lock_extent\n");
2096         }
2097
2098         if (last_byte <= start) {
2099                 clear_extent_dirty(tree, start, page_end, GFP_NOFS);
2100                 unlock_extent(tree, start, page_end, GFP_NOFS);
2101                 if (tree->ops && tree->ops->writepage_end_io_hook)
2102                         tree->ops->writepage_end_io_hook(page, start,
2103                                                          page_end, NULL, 1);
2104                 unlock_start = page_end + 1;
2105                 goto done;
2106         }
2107
2108         set_extent_uptodate(tree, start, page_end, GFP_NOFS);
2109         blocksize = inode->i_sb->s_blocksize;
2110
2111         while (cur <= end) {
2112                 if (cur >= last_byte) {
2113                         clear_extent_dirty(tree, cur, page_end, GFP_NOFS);
2114                         unlock_extent(tree, unlock_start, page_end, GFP_NOFS);
2115                         if (tree->ops && tree->ops->writepage_end_io_hook)
2116                                 tree->ops->writepage_end_io_hook(page, cur,
2117                                                          page_end, NULL, 1);
2118                         unlock_start = page_end + 1;
2119                         break;
2120                 }
2121                 em = epd->get_extent(inode, page, pg_offset, cur,
2122                                      end - cur + 1, 1);
2123                 if (IS_ERR(em) || !em) {
2124                         SetPageError(page);
2125                         break;
2126                 }
2127
2128                 extent_offset = cur - em->start;
2129                 BUG_ON(extent_map_end(em) <= cur);
2130                 BUG_ON(end < cur);
2131                 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2132                 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2133                 sector = (em->block_start + extent_offset) >> 9;
2134                 bdev = em->bdev;
2135                 block_start = em->block_start;
2136                 free_extent_map(em);
2137                 em = NULL;
2138
2139                 if (block_start == EXTENT_MAP_HOLE ||
2140                     block_start == EXTENT_MAP_INLINE) {
2141                         clear_extent_dirty(tree, cur,
2142                                            cur + iosize - 1, GFP_NOFS);
2143
2144                         unlock_extent(tree, unlock_start, cur + iosize -1,
2145                                       GFP_NOFS);
2146
2147                         if (tree->ops && tree->ops->writepage_end_io_hook)
2148                                 tree->ops->writepage_end_io_hook(page, cur,
2149                                                          cur + iosize - 1,
2150                                                          NULL, 1);
2151                         cur = cur + iosize;
2152                         pg_offset += iosize;
2153                         unlock_start = cur;
2154                         continue;
2155                 }
2156
2157                 /* leave this out until we have a page_mkwrite call */
2158                 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
2159                                    EXTENT_DIRTY, 0)) {
2160                         cur = cur + iosize;
2161                         pg_offset += iosize;
2162                         continue;
2163                 }
2164                 clear_extent_dirty(tree, cur, cur + iosize - 1, GFP_NOFS);
2165                 if (tree->ops && tree->ops->writepage_io_hook) {
2166                         ret = tree->ops->writepage_io_hook(page, cur,
2167                                                 cur + iosize - 1);
2168                 } else {
2169                         ret = 0;
2170                 }
2171                 if (ret) {
2172                         SetPageError(page);
2173                 } else {
2174                         unsigned long max_nr = end_index + 1;
2175
2176                         set_range_writeback(tree, cur, cur + iosize - 1);
2177                         if (!PageWriteback(page)) {
2178                                 printk("warning page %lu not writeback, "
2179                                        "cur %llu end %llu\n", page->index,
2180                                        (unsigned long long)cur,
2181                                        (unsigned long long)end);
2182                         }
2183
2184                         ret = submit_extent_page(WRITE, tree, page, sector,
2185                                                  iosize, pg_offset, bdev,
2186                                                  &epd->bio, max_nr,
2187                                                  end_bio_extent_writepage, 0);
2188                         if (ret)
2189                                 SetPageError(page);
2190                 }
2191                 cur = cur + iosize;
2192                 pg_offset += iosize;
2193                 nr++;
2194         }
2195 done:
2196         if (nr == 0) {
2197                 /* make sure the mapping tag for page dirty gets cleared */
2198                 set_page_writeback(page);
2199                 end_page_writeback(page);
2200         }
2201         if (unlock_start <= page_end)
2202                 unlock_extent(tree, unlock_start, page_end, GFP_NOFS);
2203         unlock_page(page);
2204         return 0;
2205 }
2206
2207 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,22)
2208 /* Taken directly from 2.6.23 for 2.6.18 back port */
2209 typedef int (*writepage_t)(struct page *page, struct writeback_control *wbc,
2210                                 void *data);
2211
2212 /**
2213  * write_cache_pages - walk the list of dirty pages of the given address space
2214  * and write all of them.
2215  * @mapping: address space structure to write
2216  * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2217  * @writepage: function called for each page
2218  * @data: data passed to writepage function
2219  *
2220  * If a page is already under I/O, write_cache_pages() skips it, even
2221  * if it's dirty.  This is desirable behaviour for memory-cleaning writeback,
2222  * but it is INCORRECT for data-integrity system calls such as fsync().  fsync()
2223  * and msync() need to guarantee that all the data which was dirty at the time
2224  * the call was made get new I/O started against them.  If wbc->sync_mode is
2225  * WB_SYNC_ALL then we were called for data integrity and we must wait for
2226  * existing IO to complete.
2227  */
2228 static int write_cache_pages(struct address_space *mapping,
2229                       struct writeback_control *wbc, writepage_t writepage,
2230                       void *data)
2231 {
2232         struct backing_dev_info *bdi = mapping->backing_dev_info;
2233         int ret = 0;
2234         int done = 0;
2235         struct pagevec pvec;
2236         int nr_pages;
2237         pgoff_t index;
2238         pgoff_t end;            /* Inclusive */
2239         int scanned = 0;
2240         int range_whole = 0;
2241
2242         if (wbc->nonblocking && bdi_write_congested(bdi)) {
2243                 wbc->encountered_congestion = 1;
2244                 return 0;
2245         }
2246
2247         pagevec_init(&pvec, 0);
2248         if (wbc->range_cyclic) {
2249                 index = mapping->writeback_index; /* Start from prev offset */
2250                 end = -1;
2251         } else {
2252                 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2253                 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2254                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2255                         range_whole = 1;
2256                 scanned = 1;
2257         }
2258 retry:
2259         while (!done && (index <= end) &&
2260                (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
2261                                               PAGECACHE_TAG_DIRTY,
2262                                               min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
2263                 unsigned i;
2264
2265                 scanned = 1;
2266                 for (i = 0; i < nr_pages; i++) {
2267                         struct page *page = pvec.pages[i];
2268
2269                         /*
2270                          * At this point we hold neither mapping->tree_lock nor
2271                          * lock on the page itself: the page may be truncated or
2272                          * invalidated (changing page->mapping to NULL), or even
2273                          * swizzled back from swapper_space to tmpfs file
2274                          * mapping
2275                          */
2276                         lock_page(page);
2277
2278                         if (unlikely(page->mapping != mapping)) {
2279                                 unlock_page(page);
2280                                 continue;
2281                         }
2282
2283                         if (!wbc->range_cyclic && page->index > end) {
2284                                 done = 1;
2285                                 unlock_page(page);
2286                                 continue;
2287                         }
2288
2289                         if (wbc->sync_mode != WB_SYNC_NONE)
2290                                 wait_on_page_writeback(page);
2291
2292                         if (PageWriteback(page) ||
2293                             !clear_page_dirty_for_io(page)) {
2294                                 unlock_page(page);
2295                                 continue;
2296                         }
2297
2298                         ret = (*writepage)(page, wbc, data);
2299
2300                         if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
2301                                 unlock_page(page);
2302                                 ret = 0;
2303                         }
2304                         if (ret || (--(wbc->nr_to_write) <= 0))
2305                                 done = 1;
2306                         if (wbc->nonblocking && bdi_write_congested(bdi)) {
2307                                 wbc->encountered_congestion = 1;
2308                                 done = 1;
2309                         }
2310                 }
2311                 pagevec_release(&pvec);
2312                 cond_resched();
2313         }
2314         if (!scanned && !done) {
2315                 /*
2316                  * We hit the last page and there is more work to be done: wrap
2317                  * back to the start of the file
2318                  */
2319                 scanned = 1;
2320                 index = 0;
2321                 goto retry;
2322         }
2323         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2324                 mapping->writeback_index = index;
2325         return ret;
2326 }
2327 #endif
2328
2329 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
2330                           get_extent_t *get_extent,
2331                           struct writeback_control *wbc)
2332 {
2333         int ret;
2334         struct address_space *mapping = page->mapping;
2335         struct extent_page_data epd = {
2336                 .bio = NULL,
2337                 .tree = tree,
2338                 .get_extent = get_extent,
2339         };
2340         struct writeback_control wbc_writepages = {
2341                 .bdi            = wbc->bdi,
2342                 .sync_mode      = WB_SYNC_NONE,
2343                 .older_than_this = NULL,
2344                 .nr_to_write    = 64,
2345                 .range_start    = page_offset(page) + PAGE_CACHE_SIZE,
2346                 .range_end      = (loff_t)-1,
2347         };
2348
2349
2350         ret = __extent_writepage(page, wbc, &epd);
2351
2352         write_cache_pages(mapping, &wbc_writepages, __extent_writepage, &epd);
2353         if (epd.bio) {
2354                 submit_one_bio(WRITE, epd.bio, 0);
2355         }
2356         return ret;
2357 }
2358 EXPORT_SYMBOL(extent_write_full_page);
2359
2360
2361 int extent_writepages(struct extent_io_tree *tree,
2362                       struct address_space *mapping,
2363                       get_extent_t *get_extent,
2364                       struct writeback_control *wbc)
2365 {
2366         int ret = 0;
2367         struct extent_page_data epd = {
2368                 .bio = NULL,
2369                 .tree = tree,
2370                 .get_extent = get_extent,
2371         };
2372
2373         ret = write_cache_pages(mapping, wbc, __extent_writepage, &epd);
2374         if (epd.bio) {
2375                 submit_one_bio(WRITE, epd.bio, 0);
2376         }
2377         return ret;
2378 }
2379 EXPORT_SYMBOL(extent_writepages);
2380
2381 int extent_readpages(struct extent_io_tree *tree,
2382                      struct address_space *mapping,
2383                      struct list_head *pages, unsigned nr_pages,
2384                      get_extent_t get_extent)
2385 {
2386         struct bio *bio = NULL;
2387         unsigned page_idx;
2388         struct pagevec pvec;
2389
2390         pagevec_init(&pvec, 0);
2391         for (page_idx = 0; page_idx < nr_pages; page_idx++) {
2392                 struct page *page = list_entry(pages->prev, struct page, lru);
2393
2394                 prefetchw(&page->flags);
2395                 list_del(&page->lru);
2396                 /*
2397                  * what we want to do here is call add_to_page_cache_lru,
2398                  * but that isn't exported, so we reproduce it here
2399                  */
2400                 if (!add_to_page_cache(page, mapping,
2401                                         page->index, GFP_KERNEL)) {
2402
2403                         /* open coding of lru_cache_add, also not exported */
2404                         page_cache_get(page);
2405                         if (!pagevec_add(&pvec, page))
2406                                 __pagevec_lru_add(&pvec);
2407                         __extent_read_full_page(tree, page, get_extent,
2408                                                 &bio, 0);
2409                 }
2410                 page_cache_release(page);
2411         }
2412         if (pagevec_count(&pvec))
2413                 __pagevec_lru_add(&pvec);
2414         BUG_ON(!list_empty(pages));
2415         if (bio)
2416                 submit_one_bio(READ, bio, 0);
2417         return 0;
2418 }
2419 EXPORT_SYMBOL(extent_readpages);
2420
2421 /*
2422  * basic invalidatepage code, this waits on any locked or writeback
2423  * ranges corresponding to the page, and then deletes any extent state
2424  * records from the tree
2425  */
2426 int extent_invalidatepage(struct extent_io_tree *tree,
2427                           struct page *page, unsigned long offset)
2428 {
2429         u64 start = ((u64)page->index << PAGE_CACHE_SHIFT);
2430         u64 end = start + PAGE_CACHE_SIZE - 1;
2431         size_t blocksize = page->mapping->host->i_sb->s_blocksize;
2432
2433         start += (offset + blocksize -1) & ~(blocksize - 1);
2434         if (start > end)
2435                 return 0;
2436
2437         lock_extent(tree, start, end, GFP_NOFS);
2438         wait_on_extent_writeback(tree, start, end);
2439         clear_extent_bit(tree, start, end,
2440                          EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC,
2441                          1, 1, GFP_NOFS);
2442         return 0;
2443 }
2444 EXPORT_SYMBOL(extent_invalidatepage);
2445
2446 /*
2447  * simple commit_write call, set_range_dirty is used to mark both
2448  * the pages and the extent records as dirty
2449  */
2450 int extent_commit_write(struct extent_io_tree *tree,
2451                         struct inode *inode, struct page *page,
2452                         unsigned from, unsigned to)
2453 {
2454         loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
2455
2456         set_page_extent_mapped(page);
2457         set_page_dirty(page);
2458
2459         if (pos > inode->i_size) {
2460                 i_size_write(inode, pos);
2461                 mark_inode_dirty(inode);
2462         }
2463         return 0;
2464 }
2465 EXPORT_SYMBOL(extent_commit_write);
2466
2467 int extent_prepare_write(struct extent_io_tree *tree,
2468                          struct inode *inode, struct page *page,
2469                          unsigned from, unsigned to, get_extent_t *get_extent)
2470 {
2471         u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2472         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
2473         u64 block_start;
2474         u64 orig_block_start;
2475         u64 block_end;
2476         u64 cur_end;
2477         struct extent_map *em;
2478         unsigned blocksize = 1 << inode->i_blkbits;
2479         size_t page_offset = 0;
2480         size_t block_off_start;
2481         size_t block_off_end;
2482         int err = 0;
2483         int iocount = 0;
2484         int ret = 0;
2485         int isnew;
2486
2487         set_page_extent_mapped(page);
2488
2489         block_start = (page_start + from) & ~((u64)blocksize - 1);
2490         block_end = (page_start + to - 1) | (blocksize - 1);
2491         orig_block_start = block_start;
2492
2493         lock_extent(tree, page_start, page_end, GFP_NOFS);
2494         while(block_start <= block_end) {
2495                 em = get_extent(inode, page, page_offset, block_start,
2496                                 block_end - block_start + 1, 1);
2497                 if (IS_ERR(em) || !em) {
2498                         goto err;
2499                 }
2500                 cur_end = min(block_end, extent_map_end(em) - 1);
2501                 block_off_start = block_start & (PAGE_CACHE_SIZE - 1);
2502                 block_off_end = block_off_start + blocksize;
2503                 isnew = clear_extent_new(tree, block_start, cur_end, GFP_NOFS);
2504
2505                 if (!PageUptodate(page) && isnew &&
2506                     (block_off_end > to || block_off_start < from)) {
2507                         void *kaddr;
2508
2509                         kaddr = kmap_atomic(page, KM_USER0);
2510                         if (block_off_end > to)
2511                                 memset(kaddr + to, 0, block_off_end - to);
2512                         if (block_off_start < from)
2513                                 memset(kaddr + block_off_start, 0,
2514                                        from - block_off_start);
2515                         flush_dcache_page(page);
2516                         kunmap_atomic(kaddr, KM_USER0);
2517                 }
2518                 if ((em->block_start != EXTENT_MAP_HOLE &&
2519                      em->block_start != EXTENT_MAP_INLINE) &&
2520                     !isnew && !PageUptodate(page) &&
2521                     (block_off_end > to || block_off_start < from) &&
2522                     !test_range_bit(tree, block_start, cur_end,
2523                                     EXTENT_UPTODATE, 1)) {
2524                         u64 sector;
2525                         u64 extent_offset = block_start - em->start;
2526                         size_t iosize;
2527                         sector = (em->block_start + extent_offset) >> 9;
2528                         iosize = (cur_end - block_start + blocksize) &
2529                                 ~((u64)blocksize - 1);
2530                         /*
2531                          * we've already got the extent locked, but we
2532                          * need to split the state such that our end_bio
2533                          * handler can clear the lock.
2534                          */
2535                         set_extent_bit(tree, block_start,
2536                                        block_start + iosize - 1,
2537                                        EXTENT_LOCKED, 0, NULL, GFP_NOFS);
2538                         ret = submit_extent_page(READ, tree, page,
2539                                          sector, iosize, page_offset, em->bdev,
2540                                          NULL, 1,
2541                                          end_bio_extent_preparewrite, 0);
2542                         iocount++;
2543                         block_start = block_start + iosize;
2544                 } else {
2545                         set_extent_uptodate(tree, block_start, cur_end,
2546                                             GFP_NOFS);
2547                         unlock_extent(tree, block_start, cur_end, GFP_NOFS);
2548                         block_start = cur_end + 1;
2549                 }
2550                 page_offset = block_start & (PAGE_CACHE_SIZE - 1);
2551                 free_extent_map(em);
2552         }
2553         if (iocount) {
2554                 wait_extent_bit(tree, orig_block_start,
2555                                 block_end, EXTENT_LOCKED);
2556         }
2557         check_page_uptodate(tree, page);
2558 err:
2559         /* FIXME, zero out newly allocated blocks on error */
2560         return err;
2561 }
2562 EXPORT_SYMBOL(extent_prepare_write);
2563
2564 /*
2565  * a helper for releasepage, this tests for areas of the page that
2566  * are locked or under IO and drops the related state bits if it is safe
2567  * to drop the page.
2568  */
2569 int try_release_extent_state(struct extent_map_tree *map,
2570                              struct extent_io_tree *tree, struct page *page,
2571                              gfp_t mask)
2572 {
2573         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2574         u64 end = start + PAGE_CACHE_SIZE - 1;
2575         int ret = 1;
2576
2577         if (test_range_bit(tree, start, end,
2578                            EXTENT_IOBITS | EXTENT_ORDERED, 0))
2579                 ret = 0;
2580         else {
2581                 if ((mask & GFP_NOFS) == GFP_NOFS)
2582                         mask = GFP_NOFS;
2583                 clear_extent_bit(tree, start, end, EXTENT_UPTODATE,
2584                                  1, 1, mask);
2585         }
2586         return ret;
2587 }
2588 EXPORT_SYMBOL(try_release_extent_state);
2589
2590 /*
2591  * a helper for releasepage.  As long as there are no locked extents
2592  * in the range corresponding to the page, both state records and extent
2593  * map records are removed
2594  */
2595 int try_release_extent_mapping(struct extent_map_tree *map,
2596                                struct extent_io_tree *tree, struct page *page,
2597                                gfp_t mask)
2598 {
2599         struct extent_map *em;
2600         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2601         u64 end = start + PAGE_CACHE_SIZE - 1;
2602
2603         if ((mask & __GFP_WAIT) &&
2604             page->mapping->host->i_size > 16 * 1024 * 1024) {
2605                 u64 len;
2606                 while (start <= end) {
2607                         len = end - start + 1;
2608                         spin_lock(&map->lock);
2609                         em = lookup_extent_mapping(map, start, len);
2610                         if (!em || IS_ERR(em)) {
2611                                 spin_unlock(&map->lock);
2612                                 break;
2613                         }
2614                         if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
2615                             em->start != start) {
2616                                 spin_unlock(&map->lock);
2617                                 free_extent_map(em);
2618                                 break;
2619                         }
2620                         if (!test_range_bit(tree, em->start,
2621                                             extent_map_end(em) - 1,
2622                                             EXTENT_LOCKED, 0)) {
2623                                 remove_extent_mapping(map, em);
2624                                 /* once for the rb tree */
2625                                 free_extent_map(em);
2626                         }
2627                         start = extent_map_end(em);
2628                         spin_unlock(&map->lock);
2629
2630                         /* once for us */
2631                         free_extent_map(em);
2632                 }
2633         }
2634         return try_release_extent_state(map, tree, page, mask);
2635 }
2636 EXPORT_SYMBOL(try_release_extent_mapping);
2637
2638 sector_t extent_bmap(struct address_space *mapping, sector_t iblock,
2639                 get_extent_t *get_extent)
2640 {
2641         struct inode *inode = mapping->host;
2642         u64 start = iblock << inode->i_blkbits;
2643         sector_t sector = 0;
2644         struct extent_map *em;
2645
2646         em = get_extent(inode, NULL, 0, start, (1 << inode->i_blkbits), 0);
2647         if (!em || IS_ERR(em))
2648                 return 0;
2649
2650         if (em->block_start == EXTENT_MAP_INLINE ||
2651             em->block_start == EXTENT_MAP_HOLE)
2652                 goto out;
2653
2654         sector = (em->block_start + start - em->start) >> inode->i_blkbits;
2655 out:
2656         free_extent_map(em);
2657         return sector;
2658 }
2659
2660 static inline struct page *extent_buffer_page(struct extent_buffer *eb,
2661                                               unsigned long i)
2662 {
2663         struct page *p;
2664         struct address_space *mapping;
2665
2666         if (i == 0)
2667                 return eb->first_page;
2668         i += eb->start >> PAGE_CACHE_SHIFT;
2669         mapping = eb->first_page->mapping;
2670         read_lock_irq(&mapping->tree_lock);
2671         p = radix_tree_lookup(&mapping->page_tree, i);
2672         read_unlock_irq(&mapping->tree_lock);
2673         return p;
2674 }
2675
2676 static inline unsigned long num_extent_pages(u64 start, u64 len)
2677 {
2678         return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
2679                 (start >> PAGE_CACHE_SHIFT);
2680 }
2681
2682 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
2683                                                    u64 start,
2684                                                    unsigned long len,
2685                                                    gfp_t mask)
2686 {
2687         struct extent_buffer *eb = NULL;
2688         unsigned long flags;
2689
2690         eb = kmem_cache_zalloc(extent_buffer_cache, mask);
2691         eb->start = start;
2692         eb->len = len;
2693         mutex_init(&eb->mutex);
2694         spin_lock_irqsave(&leak_lock, flags);
2695         list_add(&eb->leak_list, &buffers);
2696         spin_unlock_irqrestore(&leak_lock, flags);
2697         atomic_set(&eb->refs, 1);
2698
2699         return eb;
2700 }
2701
2702 static void __free_extent_buffer(struct extent_buffer *eb)
2703 {
2704         unsigned long flags;
2705         spin_lock_irqsave(&leak_lock, flags);
2706         list_del(&eb->leak_list);
2707         spin_unlock_irqrestore(&leak_lock, flags);
2708         kmem_cache_free(extent_buffer_cache, eb);
2709 }
2710
2711 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
2712                                           u64 start, unsigned long len,
2713                                           struct page *page0,
2714                                           gfp_t mask)
2715 {
2716         unsigned long num_pages = num_extent_pages(start, len);
2717         unsigned long i;
2718         unsigned long index = start >> PAGE_CACHE_SHIFT;
2719         struct extent_buffer *eb;
2720         struct extent_buffer *exists = NULL;
2721         struct page *p;
2722         struct address_space *mapping = tree->mapping;
2723         int uptodate = 1;
2724
2725         spin_lock(&tree->buffer_lock);
2726         eb = buffer_search(tree, start);
2727         if (eb) {
2728                 atomic_inc(&eb->refs);
2729                 spin_unlock(&tree->buffer_lock);
2730                 return eb;
2731         }
2732         spin_unlock(&tree->buffer_lock);
2733
2734         eb = __alloc_extent_buffer(tree, start, len, mask);
2735         if (!eb)
2736                 return NULL;
2737
2738         if (page0) {
2739                 eb->first_page = page0;
2740                 i = 1;
2741                 index++;
2742                 page_cache_get(page0);
2743                 mark_page_accessed(page0);
2744                 set_page_extent_mapped(page0);
2745                 set_page_extent_head(page0, len);
2746                 uptodate = PageUptodate(page0);
2747         } else {
2748                 i = 0;
2749         }
2750         for (; i < num_pages; i++, index++) {
2751                 p = find_or_create_page(mapping, index, mask | __GFP_HIGHMEM);
2752                 if (!p) {
2753                         WARN_ON(1);
2754                         goto free_eb;
2755                 }
2756                 set_page_extent_mapped(p);
2757                 mark_page_accessed(p);
2758                 if (i == 0) {
2759                         eb->first_page = p;
2760                         set_page_extent_head(p, len);
2761                 } else {
2762                         set_page_private(p, EXTENT_PAGE_PRIVATE);
2763                 }
2764                 if (!PageUptodate(p))
2765                         uptodate = 0;
2766                 unlock_page(p);
2767         }
2768         if (uptodate)
2769                 eb->flags |= EXTENT_UPTODATE;
2770         eb->flags |= EXTENT_BUFFER_FILLED;
2771
2772         spin_lock(&tree->buffer_lock);
2773         exists = buffer_tree_insert(tree, start, &eb->rb_node);
2774         if (exists) {
2775                 /* add one reference for the caller */
2776                 atomic_inc(&exists->refs);
2777                 spin_unlock(&tree->buffer_lock);
2778                 goto free_eb;
2779         }
2780         spin_unlock(&tree->buffer_lock);
2781
2782         /* add one reference for the tree */
2783         atomic_inc(&eb->refs);
2784         return eb;
2785
2786 free_eb:
2787         if (!atomic_dec_and_test(&eb->refs))
2788                 return exists;
2789         for (index = 1; index < i; index++)
2790                 page_cache_release(extent_buffer_page(eb, index));
2791         page_cache_release(extent_buffer_page(eb, 0));
2792         __free_extent_buffer(eb);
2793         return exists;
2794 }
2795 EXPORT_SYMBOL(alloc_extent_buffer);
2796
2797 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
2798                                          u64 start, unsigned long len,
2799                                           gfp_t mask)
2800 {
2801         struct extent_buffer *eb;
2802
2803         spin_lock(&tree->buffer_lock);
2804         eb = buffer_search(tree, start);
2805         if (eb)
2806                 atomic_inc(&eb->refs);
2807         spin_unlock(&tree->buffer_lock);
2808
2809         return eb;
2810 }
2811 EXPORT_SYMBOL(find_extent_buffer);
2812
2813 void free_extent_buffer(struct extent_buffer *eb)
2814 {
2815         if (!eb)
2816                 return;
2817
2818         if (!atomic_dec_and_test(&eb->refs))
2819                 return;
2820
2821         WARN_ON(1);
2822 }
2823 EXPORT_SYMBOL(free_extent_buffer);
2824
2825 int clear_extent_buffer_dirty(struct extent_io_tree *tree,
2826                               struct extent_buffer *eb)
2827 {
2828         int set;
2829         unsigned long i;
2830         unsigned long num_pages;
2831         struct page *page;
2832
2833         u64 start = eb->start;
2834         u64 end = start + eb->len - 1;
2835
2836         set = clear_extent_dirty(tree, start, end, GFP_NOFS);
2837         num_pages = num_extent_pages(eb->start, eb->len);
2838
2839         for (i = 0; i < num_pages; i++) {
2840                 page = extent_buffer_page(eb, i);
2841                 lock_page(page);
2842                 if (i == 0)
2843                         set_page_extent_head(page, eb->len);
2844                 else
2845                         set_page_private(page, EXTENT_PAGE_PRIVATE);
2846
2847                 /*
2848                  * if we're on the last page or the first page and the
2849                  * block isn't aligned on a page boundary, do extra checks
2850                  * to make sure we don't clean page that is partially dirty
2851                  */
2852                 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
2853                     ((i == num_pages - 1) &&
2854                      ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
2855                         start = (u64)page->index << PAGE_CACHE_SHIFT;
2856                         end  = start + PAGE_CACHE_SIZE - 1;
2857                         if (test_range_bit(tree, start, end,
2858                                            EXTENT_DIRTY, 0)) {
2859                                 unlock_page(page);
2860                                 continue;
2861                         }
2862                 }
2863                 clear_page_dirty_for_io(page);
2864                 read_lock_irq(&page->mapping->tree_lock);
2865                 if (!PageDirty(page)) {
2866                         radix_tree_tag_clear(&page->mapping->page_tree,
2867                                                 page_index(page),
2868                                                 PAGECACHE_TAG_DIRTY);
2869                 }
2870                 read_unlock_irq(&page->mapping->tree_lock);
2871                 unlock_page(page);
2872         }
2873         return 0;
2874 }
2875 EXPORT_SYMBOL(clear_extent_buffer_dirty);
2876
2877 int wait_on_extent_buffer_writeback(struct extent_io_tree *tree,
2878                                     struct extent_buffer *eb)
2879 {
2880         return wait_on_extent_writeback(tree, eb->start,
2881                                         eb->start + eb->len - 1);
2882 }
2883 EXPORT_SYMBOL(wait_on_extent_buffer_writeback);
2884
2885 int set_extent_buffer_dirty(struct extent_io_tree *tree,
2886                              struct extent_buffer *eb)
2887 {
2888         unsigned long i;
2889         unsigned long num_pages;
2890
2891         num_pages = num_extent_pages(eb->start, eb->len);
2892         for (i = 0; i < num_pages; i++) {
2893                 struct page *page = extent_buffer_page(eb, i);
2894                 /* writepage may need to do something special for the
2895                  * first page, we have to make sure page->private is
2896                  * properly set.  releasepage may drop page->private
2897                  * on us if the page isn't already dirty.
2898                  */
2899                 if (i == 0) {
2900                         lock_page(page);
2901                         set_page_extent_head(page, eb->len);
2902                 } else if (PagePrivate(page) &&
2903                            page->private != EXTENT_PAGE_PRIVATE) {
2904                         lock_page(page);
2905                         set_page_extent_mapped(page);
2906                         unlock_page(page);
2907                 }
2908                 __set_page_dirty_nobuffers(extent_buffer_page(eb, i));
2909                 if (i == 0)
2910                         unlock_page(page);
2911         }
2912         return set_extent_dirty(tree, eb->start,
2913                                 eb->start + eb->len - 1, GFP_NOFS);
2914 }
2915 EXPORT_SYMBOL(set_extent_buffer_dirty);
2916
2917 int clear_extent_buffer_uptodate(struct extent_io_tree *tree,
2918                                 struct extent_buffer *eb)
2919 {
2920         unsigned long i;
2921         struct page *page;
2922         unsigned long num_pages;
2923
2924         num_pages = num_extent_pages(eb->start, eb->len);
2925         eb->flags &= ~EXTENT_UPTODATE;
2926
2927         clear_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
2928                               GFP_NOFS);
2929         for (i = 0; i < num_pages; i++) {
2930                 page = extent_buffer_page(eb, i);
2931                 ClearPageUptodate(page);
2932         }
2933         return 0;
2934 }
2935
2936 int set_extent_buffer_uptodate(struct extent_io_tree *tree,
2937                                 struct extent_buffer *eb)
2938 {
2939         unsigned long i;
2940         struct page *page;
2941         unsigned long num_pages;
2942
2943         num_pages = num_extent_pages(eb->start, eb->len);
2944
2945         set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
2946                             GFP_NOFS);
2947         for (i = 0; i < num_pages; i++) {
2948                 page = extent_buffer_page(eb, i);
2949                 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
2950                     ((i == num_pages - 1) &&
2951                      ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
2952                         check_page_uptodate(tree, page);
2953                         continue;
2954                 }
2955                 SetPageUptodate(page);
2956         }
2957         return 0;
2958 }
2959 EXPORT_SYMBOL(set_extent_buffer_uptodate);
2960
2961 int extent_range_uptodate(struct extent_io_tree *tree,
2962                           u64 start, u64 end)
2963 {
2964         struct page *page;
2965         int ret;
2966         int pg_uptodate = 1;
2967         int uptodate;
2968         unsigned long index;
2969
2970         ret = test_range_bit(tree, start, end, EXTENT_UPTODATE, 1);
2971         if (ret)
2972                 return 1;
2973         while(start <= end) {
2974                 index = start >> PAGE_CACHE_SHIFT;
2975                 page = find_get_page(tree->mapping, index);
2976                 uptodate = PageUptodate(page);
2977                 page_cache_release(page);
2978                 if (!uptodate) {
2979                         pg_uptodate = 0;
2980                         break;
2981                 }
2982                 start += PAGE_CACHE_SIZE;
2983         }
2984         return pg_uptodate;
2985 }
2986
2987 int extent_buffer_uptodate(struct extent_io_tree *tree,
2988                            struct extent_buffer *eb)
2989 {
2990         int ret = 0;
2991         unsigned long num_pages;
2992         unsigned long i;
2993         struct page *page;
2994         int pg_uptodate = 1;
2995
2996         if (eb->flags & EXTENT_UPTODATE)
2997                 return 1;
2998
2999         ret = test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3000                            EXTENT_UPTODATE, 1);
3001         if (ret)
3002                 return ret;
3003
3004         num_pages = num_extent_pages(eb->start, eb->len);
3005         for (i = 0; i < num_pages; i++) {
3006                 page = extent_buffer_page(eb, i);
3007                 if (!PageUptodate(page)) {
3008                         pg_uptodate = 0;
3009                         break;
3010                 }
3011         }
3012         return pg_uptodate;
3013 }
3014 EXPORT_SYMBOL(extent_buffer_uptodate);
3015
3016 int read_extent_buffer_pages(struct extent_io_tree *tree,
3017                              struct extent_buffer *eb,
3018                              u64 start, int wait,
3019                              get_extent_t *get_extent, int mirror_num)
3020 {
3021         unsigned long i;
3022         unsigned long start_i;
3023         struct page *page;
3024         int err;
3025         int ret = 0;
3026         int locked_pages = 0;
3027         int all_uptodate = 1;
3028         int inc_all_pages = 0;
3029         unsigned long num_pages;
3030         struct bio *bio = NULL;
3031
3032         if (eb->flags & EXTENT_UPTODATE)
3033                 return 0;
3034
3035         if (test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3036                            EXTENT_UPTODATE, 1)) {
3037                 return 0;
3038         }
3039
3040         if (start) {
3041                 WARN_ON(start < eb->start);
3042                 start_i = (start >> PAGE_CACHE_SHIFT) -
3043                         (eb->start >> PAGE_CACHE_SHIFT);
3044         } else {
3045                 start_i = 0;
3046         }
3047
3048         num_pages = num_extent_pages(eb->start, eb->len);
3049         for (i = start_i; i < num_pages; i++) {
3050                 page = extent_buffer_page(eb, i);
3051                 if (!wait) {
3052                         if (TestSetPageLocked(page))
3053                                 goto unlock_exit;
3054                 } else {
3055                         lock_page(page);
3056                 }
3057                 locked_pages++;
3058                 if (!PageUptodate(page)) {
3059                         all_uptodate = 0;
3060                 }
3061         }
3062         if (all_uptodate) {
3063                 if (start_i == 0)
3064                         eb->flags |= EXTENT_UPTODATE;
3065                 goto unlock_exit;
3066         }
3067
3068         for (i = start_i; i < num_pages; i++) {
3069                 page = extent_buffer_page(eb, i);
3070                 if (inc_all_pages)
3071                         page_cache_get(page);
3072                 if (!PageUptodate(page)) {
3073                         if (start_i == 0)
3074                                 inc_all_pages = 1;
3075                         ClearPageError(page);
3076                         err = __extent_read_full_page(tree, page,
3077                                                       get_extent, &bio,
3078                                                       mirror_num);
3079                         if (err) {
3080                                 ret = err;
3081                         }
3082                 } else {
3083                         unlock_page(page);
3084                 }
3085         }
3086
3087         if (bio)
3088                 submit_one_bio(READ, bio, mirror_num);
3089
3090         if (ret || !wait) {
3091                 return ret;
3092         }
3093         for (i = start_i; i < num_pages; i++) {
3094                 page = extent_buffer_page(eb, i);
3095                 wait_on_page_locked(page);
3096                 if (!PageUptodate(page)) {
3097                         ret = -EIO;
3098                 }
3099         }
3100         if (!ret)
3101                 eb->flags |= EXTENT_UPTODATE;
3102         return ret;
3103
3104 unlock_exit:
3105         i = start_i;
3106         while(locked_pages > 0) {
3107                 page = extent_buffer_page(eb, i);
3108                 i++;
3109                 unlock_page(page);
3110                 locked_pages--;
3111         }
3112         return ret;
3113 }
3114 EXPORT_SYMBOL(read_extent_buffer_pages);
3115
3116 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
3117                         unsigned long start,
3118                         unsigned long len)
3119 {
3120         size_t cur;
3121         size_t offset;
3122         struct page *page;
3123         char *kaddr;
3124         char *dst = (char *)dstv;
3125         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3126         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3127
3128         WARN_ON(start > eb->len);
3129         WARN_ON(start + len > eb->start + eb->len);
3130
3131         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3132
3133         while(len > 0) {
3134                 page = extent_buffer_page(eb, i);
3135
3136                 cur = min(len, (PAGE_CACHE_SIZE - offset));
3137                 kaddr = kmap_atomic(page, KM_USER1);
3138                 memcpy(dst, kaddr + offset, cur);
3139                 kunmap_atomic(kaddr, KM_USER1);
3140
3141                 dst += cur;
3142                 len -= cur;
3143                 offset = 0;
3144                 i++;
3145         }
3146 }
3147 EXPORT_SYMBOL(read_extent_buffer);
3148
3149 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
3150                                unsigned long min_len, char **token, char **map,
3151                                unsigned long *map_start,
3152                                unsigned long *map_len, int km)
3153 {
3154         size_t offset = start & (PAGE_CACHE_SIZE - 1);
3155         char *kaddr;
3156         struct page *p;
3157         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3158         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3159         unsigned long end_i = (start_offset + start + min_len - 1) >>
3160                 PAGE_CACHE_SHIFT;
3161
3162         if (i != end_i)
3163                 return -EINVAL;
3164
3165         if (i == 0) {
3166                 offset = start_offset;
3167                 *map_start = 0;
3168         } else {
3169                 offset = 0;
3170                 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
3171         }
3172         if (start + min_len > eb->len) {
3173 printk("bad mapping eb start %Lu len %lu, wanted %lu %lu\n", eb->start, eb->len, start, min_len);
3174                 WARN_ON(1);
3175         }
3176
3177         p = extent_buffer_page(eb, i);
3178         kaddr = kmap_atomic(p, km);
3179         *token = kaddr;
3180         *map = kaddr + offset;
3181         *map_len = PAGE_CACHE_SIZE - offset;
3182         return 0;
3183 }
3184 EXPORT_SYMBOL(map_private_extent_buffer);
3185
3186 int map_extent_buffer(struct extent_buffer *eb, unsigned long start,
3187                       unsigned long min_len,
3188                       char **token, char **map,
3189                       unsigned long *map_start,
3190                       unsigned long *map_len, int km)
3191 {
3192         int err;
3193         int save = 0;
3194         if (eb->map_token) {
3195                 unmap_extent_buffer(eb, eb->map_token, km);
3196                 eb->map_token = NULL;
3197                 save = 1;
3198         }
3199         err = map_private_extent_buffer(eb, start, min_len, token, map,
3200                                        map_start, map_len, km);
3201         if (!err && save) {
3202                 eb->map_token = *token;
3203                 eb->kaddr = *map;
3204                 eb->map_start = *map_start;
3205                 eb->map_len = *map_len;
3206         }
3207         return err;
3208 }
3209 EXPORT_SYMBOL(map_extent_buffer);
3210
3211 void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km)
3212 {
3213         kunmap_atomic(token, km);
3214 }
3215 EXPORT_SYMBOL(unmap_extent_buffer);
3216
3217 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
3218                           unsigned long start,
3219                           unsigned long len)
3220 {
3221         size_t cur;
3222         size_t offset;
3223         struct page *page;
3224         char *kaddr;
3225         char *ptr = (char *)ptrv;
3226         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3227         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3228         int ret = 0;
3229
3230         WARN_ON(start > eb->len);
3231         WARN_ON(start + len > eb->start + eb->len);
3232
3233         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3234
3235         while(len > 0) {
3236                 page = extent_buffer_page(eb, i);
3237
3238                 cur = min(len, (PAGE_CACHE_SIZE - offset));
3239
3240                 kaddr = kmap_atomic(page, KM_USER0);
3241                 ret = memcmp(ptr, kaddr + offset, cur);
3242                 kunmap_atomic(kaddr, KM_USER0);
3243                 if (ret)
3244                         break;
3245
3246                 ptr += cur;
3247                 len -= cur;
3248                 offset = 0;
3249                 i++;
3250         }
3251         return ret;
3252 }
3253 EXPORT_SYMBOL(memcmp_extent_buffer);
3254
3255 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
3256                          unsigned long start, unsigned long len)
3257 {
3258         size_t cur;
3259         size_t offset;
3260         struct page *page;
3261         char *kaddr;
3262         char *src = (char *)srcv;
3263         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3264         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3265
3266         WARN_ON(start > eb->len);
3267         WARN_ON(start + len > eb->start + eb->len);
3268
3269         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3270
3271         while(len > 0) {
3272                 page = extent_buffer_page(eb, i);
3273                 WARN_ON(!PageUptodate(page));
3274
3275                 cur = min(len, PAGE_CACHE_SIZE - offset);
3276                 kaddr = kmap_atomic(page, KM_USER1);
3277                 memcpy(kaddr + offset, src, cur);
3278                 kunmap_atomic(kaddr, KM_USER1);
3279
3280                 src += cur;
3281                 len -= cur;
3282                 offset = 0;
3283                 i++;
3284         }
3285 }
3286 EXPORT_SYMBOL(write_extent_buffer);
3287
3288 void memset_extent_buffer(struct extent_buffer *eb, char c,
3289                           unsigned long start, unsigned long len)
3290 {
3291         size_t cur;
3292         size_t offset;
3293         struct page *page;
3294         char *kaddr;
3295         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3296         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3297
3298         WARN_ON(start > eb->len);
3299         WARN_ON(start + len > eb->start + eb->len);
3300
3301         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3302
3303         while(len > 0) {
3304                 page = extent_buffer_page(eb, i);
3305                 WARN_ON(!PageUptodate(page));
3306
3307                 cur = min(len, PAGE_CACHE_SIZE - offset);
3308                 kaddr = kmap_atomic(page, KM_USER0);
3309                 memset(kaddr + offset, c, cur);
3310                 kunmap_atomic(kaddr, KM_USER0);
3311
3312                 len -= cur;
3313                 offset = 0;
3314                 i++;
3315         }
3316 }
3317 EXPORT_SYMBOL(memset_extent_buffer);
3318
3319 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
3320                         unsigned long dst_offset, unsigned long src_offset,
3321                         unsigned long len)
3322 {
3323         u64 dst_len = dst->len;
3324         size_t cur;
3325         size_t offset;
3326         struct page *page;
3327         char *kaddr;
3328         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3329         unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3330
3331         WARN_ON(src->len != dst_len);
3332
3333         offset = (start_offset + dst_offset) &
3334                 ((unsigned long)PAGE_CACHE_SIZE - 1);
3335
3336         while(len > 0) {
3337                 page = extent_buffer_page(dst, i);
3338                 WARN_ON(!PageUptodate(page));
3339
3340                 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
3341
3342                 kaddr = kmap_atomic(page, KM_USER0);
3343                 read_extent_buffer(src, kaddr + offset, src_offset, cur);
3344                 kunmap_atomic(kaddr, KM_USER0);
3345
3346                 src_offset += cur;
3347                 len -= cur;
3348                 offset = 0;
3349                 i++;
3350         }
3351 }
3352 EXPORT_SYMBOL(copy_extent_buffer);
3353
3354 static void move_pages(struct page *dst_page, struct page *src_page,
3355                        unsigned long dst_off, unsigned long src_off,
3356                        unsigned long len)
3357 {
3358         char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3359         if (dst_page == src_page) {
3360                 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
3361         } else {
3362                 char *src_kaddr = kmap_atomic(src_page, KM_USER1);
3363                 char *p = dst_kaddr + dst_off + len;
3364                 char *s = src_kaddr + src_off + len;
3365
3366                 while (len--)
3367                         *--p = *--s;
3368
3369                 kunmap_atomic(src_kaddr, KM_USER1);
3370         }
3371         kunmap_atomic(dst_kaddr, KM_USER0);
3372 }
3373
3374 static void copy_pages(struct page *dst_page, struct page *src_page,
3375                        unsigned long dst_off, unsigned long src_off,
3376                        unsigned long len)
3377 {
3378         char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3379         char *src_kaddr;
3380
3381         if (dst_page != src_page)
3382                 src_kaddr = kmap_atomic(src_page, KM_USER1);
3383         else
3384                 src_kaddr = dst_kaddr;
3385
3386         memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
3387         kunmap_atomic(dst_kaddr, KM_USER0);
3388         if (dst_page != src_page)
3389                 kunmap_atomic(src_kaddr, KM_USER1);
3390 }
3391
3392 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3393                            unsigned long src_offset, unsigned long len)
3394 {
3395         size_t cur;
3396         size_t dst_off_in_page;
3397         size_t src_off_in_page;
3398         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3399         unsigned long dst_i;
3400         unsigned long src_i;
3401
3402         if (src_offset + len > dst->len) {
3403                 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
3404                        src_offset, len, dst->len);
3405                 BUG_ON(1);
3406         }
3407         if (dst_offset + len > dst->len) {
3408                 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
3409                        dst_offset, len, dst->len);
3410                 BUG_ON(1);
3411         }
3412
3413         while(len > 0) {
3414                 dst_off_in_page = (start_offset + dst_offset) &
3415                         ((unsigned long)PAGE_CACHE_SIZE - 1);
3416                 src_off_in_page = (start_offset + src_offset) &
3417                         ((unsigned long)PAGE_CACHE_SIZE - 1);
3418
3419                 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3420                 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
3421
3422                 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
3423                                                src_off_in_page));
3424                 cur = min_t(unsigned long, cur,
3425                         (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
3426
3427                 copy_pages(extent_buffer_page(dst, dst_i),
3428                            extent_buffer_page(dst, src_i),
3429                            dst_off_in_page, src_off_in_page, cur);
3430
3431                 src_offset += cur;
3432                 dst_offset += cur;
3433                 len -= cur;
3434         }
3435 }
3436 EXPORT_SYMBOL(memcpy_extent_buffer);
3437
3438 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3439                            unsigned long src_offset, unsigned long len)
3440 {
3441         size_t cur;
3442         size_t dst_off_in_page;
3443         size_t src_off_in_page;
3444         unsigned long dst_end = dst_offset + len - 1;
3445         unsigned long src_end = src_offset + len - 1;
3446         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3447         unsigned long dst_i;
3448         unsigned long src_i;
3449
3450         if (src_offset + len > dst->len) {
3451                 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
3452                        src_offset, len, dst->len);
3453                 BUG_ON(1);
3454         }
3455         if (dst_offset + len > dst->len) {
3456                 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
3457                        dst_offset, len, dst->len);
3458                 BUG_ON(1);
3459         }
3460         if (dst_offset < src_offset) {
3461                 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
3462                 return;
3463         }
3464         while(len > 0) {
3465                 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
3466                 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
3467
3468                 dst_off_in_page = (start_offset + dst_end) &
3469                         ((unsigned long)PAGE_CACHE_SIZE - 1);
3470                 src_off_in_page = (start_offset + src_end) &
3471                         ((unsigned long)PAGE_CACHE_SIZE - 1);
3472
3473                 cur = min_t(unsigned long, len, src_off_in_page + 1);
3474                 cur = min(cur, dst_off_in_page + 1);
3475                 move_pages(extent_buffer_page(dst, dst_i),
3476                            extent_buffer_page(dst, src_i),
3477                            dst_off_in_page - cur + 1,
3478                            src_off_in_page - cur + 1, cur);
3479
3480                 dst_end -= cur;
3481                 src_end -= cur;
3482                 len -= cur;
3483         }
3484 }
3485 EXPORT_SYMBOL(memmove_extent_buffer);
3486
3487 int try_release_extent_buffer(struct extent_io_tree *tree, struct page *page)
3488 {
3489         u64 start = page_offset(page);
3490         struct extent_buffer *eb;
3491         int ret = 1;
3492         unsigned long i;
3493         unsigned long num_pages;
3494
3495         spin_lock(&tree->buffer_lock);
3496         eb = buffer_search(tree, start);
3497         if (!eb)
3498                 goto out;
3499
3500         if (atomic_read(&eb->refs) > 1) {
3501                 ret = 0;
3502                 goto out;
3503         }
3504         /* at this point we can safely release the extent buffer */
3505         num_pages = num_extent_pages(eb->start, eb->len);
3506         for (i = 0; i < num_pages; i++) {
3507                 struct page *page = extent_buffer_page(eb, i);
3508                 page_cache_release(page);
3509         }
3510         rb_erase(&eb->rb_node, &tree->buffer);
3511         __free_extent_buffer(eb);
3512 out:
3513         spin_unlock(&tree->buffer_lock);
3514         return ret;
3515 }
3516 EXPORT_SYMBOL(try_release_extent_buffer);
3517