SAFE public projects git trees. - safe/jmp/linux-2.6/blob - fs/btrfs/extent_io.c

   1 #include <linux/bitops.h>
   2 #include <linux/slab.h>
   3 #include <linux/bio.h>
   4 #include <linux/mm.h>
   5 #include <linux/gfp.h>
   6 #include <linux/pagemap.h>
   7 #include <linux/page-flags.h>
   8 #include <linux/module.h>
   9 #include <linux/spinlock.h>
  10 #include <linux/blkdev.h>
  11 #include <linux/swap.h>
  12 #include <linux/version.h>
  13 #include <linux/writeback.h>
  14 #include <linux/pagevec.h>
  15 #include "extent_io.h"
  16 #include "extent_map.h"
  17
  18 /* temporary define until extent_map moves out of btrfs */
  19 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
  20                                        unsigned long extra_flags,
  21                                        void (*ctor)(void *, struct kmem_cache *,
  22                                                     unsigned long));
  23
  24 static struct kmem_cache *extent_state_cache;
  25 static struct kmem_cache *extent_buffer_cache;
  26
  27 static LIST_HEAD(buffers);
  28 static LIST_HEAD(states);
  29 static spinlock_t leak_lock = SPIN_LOCK_UNLOCKED;
  30
  31 #define BUFFER_LRU_MAX 64
  32
  33 struct tree_entry {
  34         u64 start;
  35         u64 end;
  36         struct rb_node rb_node;
  37 };
  38
  39 struct extent_page_data {
  40         struct bio *bio;
  41         struct extent_io_tree *tree;
  42         get_extent_t *get_extent;
  43 };
  44
  45 int __init extent_io_init(void)
  46 {
  47         extent_state_cache = btrfs_cache_create("extent_state",
  48                                             sizeof(struct extent_state), 0,
  49                                             NULL);
  50         if (!extent_state_cache)
  51                 return -ENOMEM;
  52
  53         extent_buffer_cache = btrfs_cache_create("extent_buffers",
  54                                             sizeof(struct extent_buffer), 0,
  55                                             NULL);
  56         if (!extent_buffer_cache)
  57                 goto free_state_cache;
  58         return 0;
  59
  60 free_state_cache:
  61         kmem_cache_destroy(extent_state_cache);
  62         return -ENOMEM;
  63 }
  64
  65 void extent_io_exit(void)
  66 {
  67         struct extent_state *state;
  68         struct extent_buffer *eb;
  69
  70         while (!list_empty(&states)) {
  71                 state = list_entry(states.next, struct extent_state, leak_list);
  72                 printk("state leak: start %Lu end %Lu state %lu in tree %p refs %d\n", state->start, state->end, state->state, state->tree, atomic_read(&state->refs));
  73                 list_del(&state->leak_list);
  74                 kmem_cache_free(extent_state_cache, state);
  75
  76         }
  77
  78         while (!list_empty(&buffers)) {
  79                 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
  80                 printk("buffer leak start %Lu len %lu refs %d\n", eb->start, eb->len, atomic_read(&eb->refs));
  81                 list_del(&eb->leak_list);
  82                 kmem_cache_free(extent_buffer_cache, eb);
  83         }
  84         if (extent_state_cache)
  85                 kmem_cache_destroy(extent_state_cache);
  86         if (extent_buffer_cache)
  87                 kmem_cache_destroy(extent_buffer_cache);
  88 }
  89
  90 void extent_io_tree_init(struct extent_io_tree *tree,
  91                           struct address_space *mapping, gfp_t mask)
  92 {
  93         tree->state.rb_node = NULL;
  94         tree->ops = NULL;
  95         tree->dirty_bytes = 0;
  96         spin_lock_init(&tree->lock);
  97         spin_lock_init(&tree->lru_lock);
  98         tree->mapping = mapping;
  99         INIT_LIST_HEAD(&tree->buffer_lru);
 100         tree->lru_size = 0;
 101         tree->last = NULL;
 102 }
 103 EXPORT_SYMBOL(extent_io_tree_init);
 104
 105 void extent_io_tree_empty_lru(struct extent_io_tree *tree)
 106 {
 107         struct extent_buffer *eb;
 108         while(!list_empty(&tree->buffer_lru)) {
 109                 eb = list_entry(tree->buffer_lru.next, struct extent_buffer,
 110                                 lru);
 111                 list_del_init(&eb->lru);
 112                 free_extent_buffer(eb);
 113         }
 114 }
 115 EXPORT_SYMBOL(extent_io_tree_empty_lru);
 116
 117 struct extent_state *alloc_extent_state(gfp_t mask)
 118 {
 119         struct extent_state *state;
 120         unsigned long flags;
 121
 122         state = kmem_cache_alloc(extent_state_cache, mask);
 123         if (!state)
 124                 return state;
 125         state->state = 0;
 126         state->private = 0;
 127         state->tree = NULL;
 128         spin_lock_irqsave(&leak_lock, flags);
 129         list_add(&state->leak_list, &states);
 130         spin_unlock_irqrestore(&leak_lock, flags);
 131
 132         atomic_set(&state->refs, 1);
 133         init_waitqueue_head(&state->wq);
 134         return state;
 135 }
 136 EXPORT_SYMBOL(alloc_extent_state);
 137
 138 void free_extent_state(struct extent_state *state)
 139 {
 140         if (!state)
 141                 return;
 142         if (atomic_dec_and_test(&state->refs)) {
 143                 unsigned long flags;
 144                 WARN_ON(state->tree);
 145                 spin_lock_irqsave(&leak_lock, flags);
 146                 list_del(&state->leak_list);
 147                 spin_unlock_irqrestore(&leak_lock, flags);
 148                 kmem_cache_free(extent_state_cache, state);
 149         }
 150 }
 151 EXPORT_SYMBOL(free_extent_state);
 152
 153 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
 154                                    struct rb_node *node)
 155 {
 156         struct rb_node ** p = &root->rb_node;
 157         struct rb_node * parent = NULL;
 158         struct tree_entry *entry;
 159
 160         while(*p) {
 161                 parent = *p;
 162                 entry = rb_entry(parent, struct tree_entry, rb_node);
 163
 164                 if (offset < entry->start)
 165                         p = &(*p)->rb_left;
 166                 else if (offset > entry->end)
 167                         p = &(*p)->rb_right;
 168                 else
 169                         return parent;
 170         }
 171
 172         entry = rb_entry(node, struct tree_entry, rb_node);
 173         rb_link_node(node, parent, p);
 174         rb_insert_color(node, root);
 175         return NULL;
 176 }
 177
 178 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
 179                                      struct rb_node **prev_ret,
 180                                      struct rb_node **next_ret)
 181 {
 182         struct rb_root *root = &tree->state;
 183         struct rb_node * n = root->rb_node;
 184         struct rb_node *prev = NULL;
 185         struct rb_node *orig_prev = NULL;
 186         struct tree_entry *entry;
 187         struct tree_entry *prev_entry = NULL;
 188
 189         if (tree->last) {
 190                 struct extent_state *state;
 191                 state = tree->last;
 192                 if (state->start <= offset && offset <= state->end)
 193                         return &tree->last->rb_node;
 194         }
 195         while(n) {
 196                 entry = rb_entry(n, struct tree_entry, rb_node);
 197                 prev = n;
 198                 prev_entry = entry;
 199
 200                 if (offset < entry->start)
 201                         n = n->rb_left;
 202                 else if (offset > entry->end)
 203                         n = n->rb_right;
 204                 else {
 205                         tree->last = rb_entry(n, struct extent_state, rb_node);
 206                         return n;
 207                 }
 208         }
 209
 210         if (prev_ret) {
 211                 orig_prev = prev;
 212                 while(prev && offset > prev_entry->end) {
 213                         prev = rb_next(prev);
 214                         prev_entry = rb_entry(prev, struct tree_entry, rb_node);
 215                 }
 216                 *prev_ret = prev;
 217                 prev = orig_prev;
 218         }
 219
 220         if (next_ret) {
 221                 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
 222                 while(prev && offset < prev_entry->start) {
 223                         prev = rb_prev(prev);
 224                         prev_entry = rb_entry(prev, struct tree_entry, rb_node);
 225                 }
 226                 *next_ret = prev;
 227         }
 228         return NULL;
 229 }
 230
 231 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
 232                                           u64 offset)
 233 {
 234         struct rb_node *prev = NULL;
 235         struct rb_node *ret;
 236
 237         ret = __etree_search(tree, offset, &prev, NULL);
 238         if (!ret) {
 239                 if (prev) {
 240                         tree->last = rb_entry(prev, struct extent_state,
 241                                               rb_node);
 242                 }
 243                 return prev;
 244         }
 245         return ret;
 246 }
 247
 248 /*
 249  * utility function to look for merge candidates inside a given range.
 250  * Any extents with matching state are merged together into a single
 251  * extent in the tree.  Extents with EXTENT_IO in their state field
 252  * are not merged because the end_io handlers need to be able to do
 253  * operations on them without sleeping (or doing allocations/splits).
 254  *
 255  * This should be called with the tree lock held.
 256  */
 257 static int merge_state(struct extent_io_tree *tree,
 258                        struct extent_state *state)
 259 {
 260         struct extent_state *other;
 261         struct rb_node *other_node;
 262
 263         if (state->state & EXTENT_IOBITS)
 264                 return 0;
 265
 266         other_node = rb_prev(&state->rb_node);
 267         if (other_node) {
 268                 other = rb_entry(other_node, struct extent_state, rb_node);
 269                 if (other->end == state->start - 1 &&
 270                     other->state == state->state) {
 271                         state->start = other->start;
 272                         other->tree = NULL;
 273                         if (tree->last == other)
 274                                 tree->last = state;
 275                         rb_erase(&other->rb_node, &tree->state);
 276                         free_extent_state(other);
 277                 }
 278         }
 279         other_node = rb_next(&state->rb_node);
 280         if (other_node) {
 281                 other = rb_entry(other_node, struct extent_state, rb_node);
 282                 if (other->start == state->end + 1 &&
 283                     other->state == state->state) {
 284                         other->start = state->start;
 285                         state->tree = NULL;
 286                         if (tree->last == state)
 287                                 tree->last = other;
 288                         rb_erase(&state->rb_node, &tree->state);
 289                         free_extent_state(state);
 290                 }
 291         }
 292         return 0;
 293 }
 294
 295 static void set_state_cb(struct extent_io_tree *tree,
 296                          struct extent_state *state,
 297                          unsigned long bits)
 298 {
 299         if (tree->ops && tree->ops->set_bit_hook) {
 300                 tree->ops->set_bit_hook(tree->mapping->host, state->start,
 301                                         state->end, state->state, bits);
 302         }
 303 }
 304
 305 static void clear_state_cb(struct extent_io_tree *tree,
 306                            struct extent_state *state,
 307                            unsigned long bits)
 308 {
 309         if (tree->ops && tree->ops->set_bit_hook) {
 310                 tree->ops->clear_bit_hook(tree->mapping->host, state->start,
 311                                           state->end, state->state, bits);
 312         }
 313 }
 314
 315 /*
 316  * insert an extent_state struct into the tree.  'bits' are set on the
 317  * struct before it is inserted.
 318  *
 319  * This may return -EEXIST if the extent is already there, in which case the
 320  * state struct is freed.
 321  *
 322  * The tree lock is not taken internally.  This is a utility function and
 323  * probably isn't what you want to call (see set/clear_extent_bit).
 324  */
 325 static int insert_state(struct extent_io_tree *tree,
 326                         struct extent_state *state, u64 start, u64 end,
 327                         int bits)
 328 {
 329         struct rb_node *node;
 330
 331         if (end < start) {
 332                 printk("end < start %Lu %Lu\n", end, start);
 333                 WARN_ON(1);
 334         }
 335         if (bits & EXTENT_DIRTY)
 336                 tree->dirty_bytes += end - start + 1;
 337         set_state_cb(tree, state, bits);
 338         state->state |= bits;
 339         state->start = start;
 340         state->end = end;
 341         node = tree_insert(&tree->state, end, &state->rb_node);
 342         if (node) {
 343                 struct extent_state *found;
 344                 found = rb_entry(node, struct extent_state, rb_node);
 345                 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, start, end);
 346                 free_extent_state(state);
 347                 return -EEXIST;
 348         }
 349         state->tree = tree;
 350         tree->last = state;
 351         merge_state(tree, state);
 352         return 0;
 353 }
 354
 355 /*
 356  * split a given extent state struct in two, inserting the preallocated
 357  * struct 'prealloc' as the newly created second half.  'split' indicates an
 358  * offset inside 'orig' where it should be split.
 359  *
 360  * Before calling,
 361  * the tree has 'orig' at [orig->start, orig->end].  After calling, there
 362  * are two extent state structs in the tree:
 363  * prealloc: [orig->start, split - 1]
 364  * orig: [ split, orig->end ]
 365  *
 366  * The tree locks are not taken by this function. They need to be held
 367  * by the caller.
 368  */
 369 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
 370                        struct extent_state *prealloc, u64 split)
 371 {
 372         struct rb_node *node;
 373         prealloc->start = orig->start;
 374         prealloc->end = split - 1;
 375         prealloc->state = orig->state;
 376         orig->start = split;
 377
 378         node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
 379         if (node) {
 380                 struct extent_state *found;
 381                 found = rb_entry(node, struct extent_state, rb_node);
 382                 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, prealloc->start, prealloc->end);
 383                 free_extent_state(prealloc);
 384                 return -EEXIST;
 385         }
 386         prealloc->tree = tree;
 387         return 0;
 388 }
 389
 390 /*
 391  * utility function to clear some bits in an extent state struct.
 392  * it will optionally wake up any one waiting on this state (wake == 1), or
 393  * forcibly remove the state from the tree (delete == 1).
 394  *
 395  * If no bits are set on the state struct after clearing things, the
 396  * struct is freed and removed from the tree
 397  */
 398 static int clear_state_bit(struct extent_io_tree *tree,
 399                             struct extent_state *state, int bits, int wake,
 400                             int delete)
 401 {
 402         int ret = state->state & bits;
 403
 404         if ((bits & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
 405                 u64 range = state->end - state->start + 1;
 406                 WARN_ON(range > tree->dirty_bytes);
 407                 tree->dirty_bytes -= range;
 408         }
 409         clear_state_cb(tree, state, bits);
 410         state->state &= ~bits;
 411         if (wake)
 412                 wake_up(&state->wq);
 413         if (delete || state->state == 0) {
 414                 if (state->tree) {
 415                         clear_state_cb(tree, state, state->state);
 416                         if (tree->last == state) {
 417                                 tree->last = extent_state_next(state);
 418                         }
 419                         rb_erase(&state->rb_node, &tree->state);
 420                         state->tree = NULL;
 421                         free_extent_state(state);
 422                 } else {
 423                         WARN_ON(1);
 424                 }
 425         } else {
 426                 merge_state(tree, state);
 427         }
 428         return ret;
 429 }
 430
 431 /*
 432  * clear some bits on a range in the tree.  This may require splitting
 433  * or inserting elements in the tree, so the gfp mask is used to
 434  * indicate which allocations or sleeping are allowed.
 435  *
 436  * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
 437  * the given range from the tree regardless of state (ie for truncate).
 438  *
 439  * the range [start, end] is inclusive.
 440  *
 441  * This takes the tree lock, and returns < 0 on error, > 0 if any of the
 442  * bits were already set, or zero if none of the bits were already set.
 443  */
 444 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
 445                      int bits, int wake, int delete, gfp_t mask)
 446 {
 447         struct extent_state *state;
 448         struct extent_state *prealloc = NULL;
 449         struct rb_node *node;
 450         unsigned long flags;
 451         int err;
 452         int set = 0;
 453
 454 again:
 455         if (!prealloc && (mask & __GFP_WAIT)) {
 456                 prealloc = alloc_extent_state(mask);
 457                 if (!prealloc)
 458                         return -ENOMEM;
 459         }
 460
 461         spin_lock_irqsave(&tree->lock, flags);
 462         /*
 463          * this search will find the extents that end after
 464          * our range starts
 465          */
 466         node = tree_search(tree, start);
 467         if (!node)
 468                 goto out;
 469         state = rb_entry(node, struct extent_state, rb_node);
 470         if (state->start > end)
 471                 goto out;
 472         WARN_ON(state->end < start);
 473
 474         /*
 475          *     | ---- desired range ---- |
 476          *  | state | or
 477          *  | ------------- state -------------- |
 478          *
 479          * We need to split the extent we found, and may flip
 480          * bits on second half.
 481          *
 482          * If the extent we found extends past our range, we
 483          * just split and search again.  It'll get split again
 484          * the next time though.
 485          *
 486          * If the extent we found is inside our range, we clear
 487          * the desired bit on it.
 488          */
 489
 490         if (state->start < start) {
 491                 if (!prealloc)
 492                         prealloc = alloc_extent_state(GFP_ATOMIC);
 493                 err = split_state(tree, state, prealloc, start);
 494                 BUG_ON(err == -EEXIST);
 495                 prealloc = NULL;
 496                 if (err)
 497                         goto out;
 498                 if (state->end <= end) {
 499                         start = state->end + 1;
 500                         set |= clear_state_bit(tree, state, bits,
 501                                         wake, delete);
 502                 } else {
 503                         start = state->start;
 504                 }
 505                 goto search_again;
 506         }
 507         /*
 508          * | ---- desired range ---- |
 509          *                        | state |
 510          * We need to split the extent, and clear the bit
 511          * on the first half
 512          */
 513         if (state->start <= end && state->end > end) {
 514                 if (!prealloc)
 515                         prealloc = alloc_extent_state(GFP_ATOMIC);
 516                 err = split_state(tree, state, prealloc, end + 1);
 517                 BUG_ON(err == -EEXIST);
 518
 519                 if (wake)
 520                         wake_up(&state->wq);
 521                 set |= clear_state_bit(tree, prealloc, bits,
 522                                        wake, delete);
 523                 prealloc = NULL;
 524                 goto out;
 525         }
 526
 527         start = state->end + 1;
 528         set |= clear_state_bit(tree, state, bits, wake, delete);
 529         goto search_again;
 530
 531 out:
 532         spin_unlock_irqrestore(&tree->lock, flags);
 533         if (prealloc)
 534                 free_extent_state(prealloc);
 535
 536         return set;
 537
 538 search_again:
 539         if (start > end)
 540                 goto out;
 541         spin_unlock_irqrestore(&tree->lock, flags);
 542         if (mask & __GFP_WAIT)
 543                 cond_resched();
 544         goto again;
 545 }
 546 EXPORT_SYMBOL(clear_extent_bit);
 547
 548 static int wait_on_state(struct extent_io_tree *tree,
 549                          struct extent_state *state)
 550 {
 551         DEFINE_WAIT(wait);
 552         prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
 553         spin_unlock_irq(&tree->lock);
 554         schedule();
 555         spin_lock_irq(&tree->lock);
 556         finish_wait(&state->wq, &wait);
 557         return 0;
 558 }
 559
 560 /*
 561  * waits for one or more bits to clear on a range in the state tree.
 562  * The range [start, end] is inclusive.
 563  * The tree lock is taken by this function
 564  */
 565 int wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits)
 566 {
 567         struct extent_state *state;
 568         struct rb_node *node;
 569
 570         spin_lock_irq(&tree->lock);
 571 again:
 572         while (1) {
 573                 /*
 574                  * this search will find all the extents that end after
 575                  * our range starts
 576                  */
 577                 node = tree_search(tree, start);
 578                 if (!node)
 579                         break;
 580
 581                 state = rb_entry(node, struct extent_state, rb_node);
 582
 583                 if (state->start > end)
 584                         goto out;
 585
 586                 if (state->state & bits) {
 587                         start = state->start;
 588                         atomic_inc(&state->refs);
 589                         wait_on_state(tree, state);
 590                         free_extent_state(state);
 591                         goto again;
 592                 }
 593                 start = state->end + 1;
 594
 595                 if (start > end)
 596                         break;
 597
 598                 if (need_resched()) {
 599                         spin_unlock_irq(&tree->lock);
 600                         cond_resched();
 601                         spin_lock_irq(&tree->lock);
 602                 }
 603         }
 604 out:
 605         spin_unlock_irq(&tree->lock);
 606         return 0;
 607 }
 608 EXPORT_SYMBOL(wait_extent_bit);
 609
 610 static void set_state_bits(struct extent_io_tree *tree,
 611                            struct extent_state *state,
 612                            int bits)
 613 {
 614         if ((bits & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
 615                 u64 range = state->end - state->start + 1;
 616                 tree->dirty_bytes += range;
 617         }
 618         set_state_cb(tree, state, bits);
 619         state->state |= bits;
 620 }
 621
 622 /*
 623  * set some bits on a range in the tree.  This may require allocations
 624  * or sleeping, so the gfp mask is used to indicate what is allowed.
 625  *
 626  * If 'exclusive' == 1, this will fail with -EEXIST if some part of the
 627  * range already has the desired bits set.  The start of the existing
 628  * range is returned in failed_start in this case.
 629  *
 630  * [start, end] is inclusive
 631  * This takes the tree lock.
 632  */
 633 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits,
 634                    int exclusive, u64 *failed_start, gfp_t mask)
 635 {
 636         struct extent_state *state;
 637         struct extent_state *prealloc = NULL;
 638         struct rb_node *node;
 639         unsigned long flags;
 640         int err = 0;
 641         int set;
 642         u64 last_start;
 643         u64 last_end;
 644 again:
 645         if (!prealloc && (mask & __GFP_WAIT)) {
 646                 prealloc = alloc_extent_state(mask);
 647                 if (!prealloc)
 648                         return -ENOMEM;
 649         }
 650
 651         spin_lock_irqsave(&tree->lock, flags);
 652         /*
 653          * this search will find all the extents that end after
 654          * our range starts.
 655          */
 656         node = tree_search(tree, start);
 657         if (!node) {
 658                 err = insert_state(tree, prealloc, start, end, bits);
 659                 prealloc = NULL;
 660                 BUG_ON(err == -EEXIST);
 661                 goto out;
 662         }
 663
 664         state = rb_entry(node, struct extent_state, rb_node);
 665         last_start = state->start;
 666         last_end = state->end;
 667
 668         /*
 669          * | ---- desired range ---- |
 670          * | state |
 671          *
 672          * Just lock what we found and keep going
 673          */
 674         if (state->start == start && state->end <= end) {
 675                 set = state->state & bits;
 676                 if (set && exclusive) {
 677                         *failed_start = state->start;
 678                         err = -EEXIST;
 679                         goto out;
 680                 }
 681                 set_state_bits(tree, state, bits);
 682                 start = state->end + 1;
 683                 merge_state(tree, state);
 684                 goto search_again;
 685         }
 686
 687         /*
 688          *     | ---- desired range ---- |
 689          * | state |
 690          *   or
 691          * | ------------- state -------------- |
 692          *
 693          * We need to split the extent we found, and may flip bits on
 694          * second half.
 695          *
 696          * If the extent we found extends past our
 697          * range, we just split and search again.  It'll get split
 698          * again the next time though.
 699          *
 700          * If the extent we found is inside our range, we set the
 701          * desired bit on it.
 702          */
 703         if (state->start < start) {
 704                 set = state->state & bits;
 705                 if (exclusive && set) {
 706                         *failed_start = start;
 707                         err = -EEXIST;
 708                         goto out;
 709                 }
 710                 err = split_state(tree, state, prealloc, start);
 711                 BUG_ON(err == -EEXIST);
 712                 prealloc = NULL;
 713                 if (err)
 714                         goto out;
 715                 if (state->end <= end) {
 716                         set_state_bits(tree, state, bits);
 717                         start = state->end + 1;
 718                         merge_state(tree, state);
 719                 } else {
 720                         start = state->start;
 721                 }
 722                 goto search_again;
 723         }
 724         /*
 725          * | ---- desired range ---- |
 726          *     | state | or               | state |
 727          *
 728          * There's a hole, we need to insert something in it and
 729          * ignore the extent we found.
 730          */
 731         if (state->start > start) {
 732                 u64 this_end;
 733                 if (end < last_start)
 734                         this_end = end;
 735                 else
 736                         this_end = last_start -1;
 737                 err = insert_state(tree, prealloc, start, this_end,
 738                                    bits);
 739                 prealloc = NULL;
 740                 BUG_ON(err == -EEXIST);
 741                 if (err)
 742                         goto out;
 743                 start = this_end + 1;
 744                 goto search_again;
 745         }
 746         /*
 747          * | ---- desired range ---- |
 748          *                        | state |
 749          * We need to split the extent, and set the bit
 750          * on the first half
 751          */
 752         if (state->start <= end && state->end > end) {
 753                 set = state->state & bits;
 754                 if (exclusive && set) {
 755                         *failed_start = start;
 756                         err = -EEXIST;
 757                         goto out;
 758                 }
 759                 err = split_state(tree, state, prealloc, end + 1);
 760                 BUG_ON(err == -EEXIST);
 761
 762                 set_state_bits(tree, prealloc, bits);
 763                 merge_state(tree, prealloc);
 764                 prealloc = NULL;
 765                 goto out;
 766         }
 767
 768         goto search_again;
 769
 770 out:
 771         spin_unlock_irqrestore(&tree->lock, flags);
 772         if (prealloc)
 773                 free_extent_state(prealloc);
 774
 775         return err;
 776
 777 search_again:
 778         if (start > end)
 779                 goto out;
 780         spin_unlock_irqrestore(&tree->lock, flags);
 781         if (mask & __GFP_WAIT)
 782                 cond_resched();
 783         goto again;
 784 }
 785 EXPORT_SYMBOL(set_extent_bit);
 786
 787 /* wrappers around set/clear extent bit */
 788 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
 789                      gfp_t mask)
 790 {
 791         return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL,
 792                               mask);
 793 }
 794 EXPORT_SYMBOL(set_extent_dirty);
 795
 796 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
 797                     int bits, gfp_t mask)
 798 {
 799         return set_extent_bit(tree, start, end, bits, 0, NULL,
 800                               mask);
 801 }
 802 EXPORT_SYMBOL(set_extent_bits);
 803
 804 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
 805                       int bits, gfp_t mask)
 806 {
 807         return clear_extent_bit(tree, start, end, bits, 0, 0, mask);
 808 }
 809 EXPORT_SYMBOL(clear_extent_bits);
 810
 811 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
 812                      gfp_t mask)
 813 {
 814         return set_extent_bit(tree, start, end,
 815                               EXTENT_DELALLOC | EXTENT_DIRTY, 0, NULL,
 816                               mask);
 817 }
 818 EXPORT_SYMBOL(set_extent_delalloc);
 819
 820 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
 821                        gfp_t mask)
 822 {
 823         return clear_extent_bit(tree, start, end,
 824                                 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, mask);
 825 }
 826 EXPORT_SYMBOL(clear_extent_dirty);
 827
 828 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
 829                      gfp_t mask)
 830 {
 831         return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL,
 832                               mask);
 833 }
 834 EXPORT_SYMBOL(set_extent_new);
 835
 836 int clear_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
 837                        gfp_t mask)
 838 {
 839         return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0, mask);
 840 }
 841 EXPORT_SYMBOL(clear_extent_new);
 842
 843 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
 844                         gfp_t mask)
 845 {
 846         return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, NULL,
 847                               mask);
 848 }
 849 EXPORT_SYMBOL(set_extent_uptodate);
 850
 851 int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
 852                           gfp_t mask)
 853 {
 854         return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, mask);
 855 }
 856 EXPORT_SYMBOL(clear_extent_uptodate);
 857
 858 int set_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end,
 859                          gfp_t mask)
 860 {
 861         return set_extent_bit(tree, start, end, EXTENT_WRITEBACK,
 862                               0, NULL, mask);
 863 }
 864 EXPORT_SYMBOL(set_extent_writeback);
 865
 866 int clear_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end,
 867                            gfp_t mask)
 868 {
 869         return clear_extent_bit(tree, start, end, EXTENT_WRITEBACK, 1, 0, mask);
 870 }
 871 EXPORT_SYMBOL(clear_extent_writeback);
 872
 873 int wait_on_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end)
 874 {
 875         return wait_extent_bit(tree, start, end, EXTENT_WRITEBACK);
 876 }
 877 EXPORT_SYMBOL(wait_on_extent_writeback);
 878
 879 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
 880 {
 881         int err;
 882         u64 failed_start;
 883         while (1) {
 884                 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1,
 885                                      &failed_start, mask);
 886                 if (err == -EEXIST && (mask & __GFP_WAIT)) {
 887                         wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
 888                         start = failed_start;
 889                 } else {
 890                         break;
 891                 }
 892                 WARN_ON(start > end);
 893         }
 894         return err;
 895 }
 896 EXPORT_SYMBOL(lock_extent);
 897
 898 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end,
 899                   gfp_t mask)
 900 {
 901         return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, mask);
 902 }
 903 EXPORT_SYMBOL(unlock_extent);
 904
 905 /*
 906  * helper function to set pages and extents in the tree dirty
 907  */
 908 int set_range_dirty(struct extent_io_tree *tree, u64 start, u64 end)
 909 {
 910         unsigned long index = start >> PAGE_CACHE_SHIFT;
 911         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
 912         struct page *page;
 913
 914         while (index <= end_index) {
 915                 page = find_get_page(tree->mapping, index);
 916                 BUG_ON(!page);
 917                 __set_page_dirty_nobuffers(page);
 918                 page_cache_release(page);
 919                 index++;
 920         }
 921         set_extent_dirty(tree, start, end, GFP_NOFS);
 922         return 0;
 923 }
 924 EXPORT_SYMBOL(set_range_dirty);
 925
 926 /*
 927  * helper function to set both pages and extents in the tree writeback
 928  */
 929 int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
 930 {
 931         unsigned long index = start >> PAGE_CACHE_SHIFT;
 932         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
 933         struct page *page;
 934
 935         while (index <= end_index) {
 936                 page = find_get_page(tree->mapping, index);
 937                 BUG_ON(!page);
 938                 set_page_writeback(page);
 939                 page_cache_release(page);
 940                 index++;
 941         }
 942         set_extent_writeback(tree, start, end, GFP_NOFS);
 943         return 0;
 944 }
 945 EXPORT_SYMBOL(set_range_writeback);
 946
 947 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
 948                           u64 *start_ret, u64 *end_ret, int bits)
 949 {
 950         struct rb_node *node;
 951         struct extent_state *state;
 952         int ret = 1;
 953
 954         spin_lock_irq(&tree->lock);
 955         /*
 956          * this search will find all the extents that end after
 957          * our range starts.
 958          */
 959         node = tree_search(tree, start);
 960         if (!node) {
 961                 goto out;
 962         }
 963
 964         while(1) {
 965                 state = rb_entry(node, struct extent_state, rb_node);
 966                 if (state->end >= start && (state->state & bits)) {
 967                         *start_ret = state->start;
 968                         *end_ret = state->end;
 969                         ret = 0;
 970                         break;
 971                 }
 972                 node = rb_next(node);
 973                 if (!node)
 974                         break;
 975         }
 976 out:
 977         spin_unlock_irq(&tree->lock);
 978         return ret;
 979 }
 980 EXPORT_SYMBOL(find_first_extent_bit);
 981
 982 struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree,
 983                                                  u64 start, int bits)
 984 {
 985         struct rb_node *node;
 986         struct extent_state *state;
 987
 988         /*
 989          * this search will find all the extents that end after
 990          * our range starts.
 991          */
 992         node = tree_search(tree, start);
 993         if (!node) {
 994                 goto out;
 995         }
 996
 997         while(1) {
 998                 state = rb_entry(node, struct extent_state, rb_node);
 999                 if (state->end >= start && (state->state & bits)) {
1000                         return state;
1001                 }
1002                 node = rb_next(node);
1003                 if (!node)
1004                         break;
1005         }
1006 out:
1007         return NULL;
1008 }
1009 EXPORT_SYMBOL(find_first_extent_bit_state);
1010
1011 u64 find_lock_delalloc_range(struct extent_io_tree *tree,
1012                              u64 *start, u64 *end, u64 max_bytes)
1013 {
1014         struct rb_node *node;
1015         struct extent_state *state;
1016         u64 cur_start = *start;
1017         u64 found = 0;
1018         u64 total_bytes = 0;
1019
1020         spin_lock_irq(&tree->lock);
1021         /*
1022          * this search will find all the extents that end after
1023          * our range starts.
1024          */
1025 search_again:
1026         node = tree_search(tree, cur_start);
1027         if (!node) {
1028                 *end = (u64)-1;
1029                 goto out;
1030         }
1031
1032         while(1) {
1033                 state = rb_entry(node, struct extent_state, rb_node);
1034                 if (found && state->start != cur_start) {
1035                         goto out;
1036                 }
1037                 if (!(state->state & EXTENT_DELALLOC)) {
1038                         if (!found)
1039                                 *end = state->end;
1040                         goto out;
1041                 }
1042                 if (!found) {
1043                         struct extent_state *prev_state;
1044                         struct rb_node *prev_node = node;
1045                         while(1) {
1046                                 prev_node = rb_prev(prev_node);
1047                                 if (!prev_node)
1048                                         break;
1049                                 prev_state = rb_entry(prev_node,
1050                                                       struct extent_state,
1051                                                       rb_node);
1052                                 if (!(prev_state->state & EXTENT_DELALLOC))
1053                                         break;
1054                                 state = prev_state;
1055                                 node = prev_node;
1056                         }
1057                 }
1058                 if (state->state & EXTENT_LOCKED) {
1059                         DEFINE_WAIT(wait);
1060                         atomic_inc(&state->refs);
1061                         prepare_to_wait(&state->wq, &wait,
1062                                         TASK_UNINTERRUPTIBLE);
1063                         spin_unlock_irq(&tree->lock);
1064                         schedule();
1065                         spin_lock_irq(&tree->lock);
1066                         finish_wait(&state->wq, &wait);
1067                         free_extent_state(state);
1068                         goto search_again;
1069                 }
1070                 set_state_cb(tree, state, EXTENT_LOCKED);
1071                 state->state |= EXTENT_LOCKED;
1072                 if (!found)
1073                         *start = state->start;
1074                 found++;
1075                 *end = state->end;
1076                 cur_start = state->end + 1;
1077                 node = rb_next(node);
1078                 if (!node)
1079                         break;
1080                 total_bytes += state->end - state->start + 1;
1081                 if (total_bytes >= max_bytes)
1082                         break;
1083         }
1084 out:
1085         spin_unlock_irq(&tree->lock);
1086         return found;
1087 }
1088
1089 u64 count_range_bits(struct extent_io_tree *tree,
1090                      u64 *start, u64 search_end, u64 max_bytes,
1091                      unsigned long bits)
1092 {
1093         struct rb_node *node;
1094         struct extent_state *state;
1095         u64 cur_start = *start;
1096         u64 total_bytes = 0;
1097         int found = 0;
1098
1099         if (search_end <= cur_start) {
1100                 printk("search_end %Lu start %Lu\n", search_end, cur_start);
1101                 WARN_ON(1);
1102                 return 0;
1103         }
1104
1105         spin_lock_irq(&tree->lock);
1106         if (cur_start == 0 && bits == EXTENT_DIRTY) {
1107                 total_bytes = tree->dirty_bytes;
1108                 goto out;
1109         }
1110         /*
1111          * this search will find all the extents that end after
1112          * our range starts.
1113          */
1114         node = tree_search(tree, cur_start);
1115         if (!node) {
1116                 goto out;
1117         }
1118
1119         while(1) {
1120                 state = rb_entry(node, struct extent_state, rb_node);
1121                 if (state->start > search_end)
1122                         break;
1123                 if (state->end >= cur_start && (state->state & bits)) {
1124                         total_bytes += min(search_end, state->end) + 1 -
1125                                        max(cur_start, state->start);
1126                         if (total_bytes >= max_bytes)
1127                                 break;
1128                         if (!found) {
1129                                 *start = state->start;
1130                                 found = 1;
1131                         }
1132                 }
1133                 node = rb_next(node);
1134                 if (!node)
1135                         break;
1136         }
1137 out:
1138         spin_unlock_irq(&tree->lock);
1139         return total_bytes;
1140 }
1141 /*
1142  * helper function to lock both pages and extents in the tree.
1143  * pages must be locked first.
1144  */
1145 int lock_range(struct extent_io_tree *tree, u64 start, u64 end)
1146 {
1147         unsigned long index = start >> PAGE_CACHE_SHIFT;
1148         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1149         struct page *page;
1150         int err;
1151
1152         while (index <= end_index) {
1153                 page = grab_cache_page(tree->mapping, index);
1154                 if (!page) {
1155                         err = -ENOMEM;
1156                         goto failed;
1157                 }
1158                 if (IS_ERR(page)) {
1159                         err = PTR_ERR(page);
1160                         goto failed;
1161                 }
1162                 index++;
1163         }
1164         lock_extent(tree, start, end, GFP_NOFS);
1165         return 0;
1166
1167 failed:
1168         /*
1169          * we failed above in getting the page at 'index', so we undo here
1170          * up to but not including the page at 'index'
1171          */
1172         end_index = index;
1173         index = start >> PAGE_CACHE_SHIFT;
1174         while (index < end_index) {
1175                 page = find_get_page(tree->mapping, index);
1176                 unlock_page(page);
1177                 page_cache_release(page);
1178                 index++;
1179         }
1180         return err;
1181 }
1182 EXPORT_SYMBOL(lock_range);
1183
1184 /*
1185  * helper function to unlock both pages and extents in the tree.
1186  */
1187 int unlock_range(struct extent_io_tree *tree, u64 start, u64 end)
1188 {
1189         unsigned long index = start >> PAGE_CACHE_SHIFT;
1190         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1191         struct page *page;
1192
1193         while (index <= end_index) {
1194                 page = find_get_page(tree->mapping, index);
1195                 unlock_page(page);
1196                 page_cache_release(page);
1197                 index++;
1198         }
1199         unlock_extent(tree, start, end, GFP_NOFS);
1200         return 0;
1201 }
1202 EXPORT_SYMBOL(unlock_range);
1203
1204 int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1205 {
1206         struct rb_node *node;
1207         struct extent_state *state;
1208         int ret = 0;
1209
1210         spin_lock_irq(&tree->lock);
1211         /*
1212          * this search will find all the extents that end after
1213          * our range starts.
1214          */
1215         node = tree_search(tree, start);
1216         if (!node) {
1217                 ret = -ENOENT;
1218                 goto out;
1219         }
1220         state = rb_entry(node, struct extent_state, rb_node);
1221         if (state->start != start) {
1222                 ret = -ENOENT;
1223                 goto out;
1224         }
1225         state->private = private;
1226 out:
1227         spin_unlock_irq(&tree->lock);
1228         return ret;
1229 }
1230
1231 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1232 {
1233         struct rb_node *node;
1234         struct extent_state *state;
1235         int ret = 0;
1236
1237         spin_lock_irq(&tree->lock);
1238         /*
1239          * this search will find all the extents that end after
1240          * our range starts.
1241          */
1242         node = tree_search(tree, start);
1243         if (!node) {
1244                 ret = -ENOENT;
1245                 goto out;
1246         }
1247         state = rb_entry(node, struct extent_state, rb_node);
1248         if (state->start != start) {
1249                 ret = -ENOENT;
1250                 goto out;
1251         }
1252         *private = state->private;
1253 out:
1254         spin_unlock_irq(&tree->lock);
1255         return ret;
1256 }
1257
1258 /*
1259  * searches a range in the state tree for a given mask.
1260  * If 'filled' == 1, this returns 1 only if every extent in the tree
1261  * has the bits set.  Otherwise, 1 is returned if any bit in the
1262  * range is found set.
1263  */
1264 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1265                    int bits, int filled)
1266 {
1267         struct extent_state *state = NULL;
1268         struct rb_node *node;
1269         int bitset = 0;
1270         unsigned long flags;
1271
1272         spin_lock_irqsave(&tree->lock, flags);
1273         node = tree_search(tree, start);
1274         while (node && start <= end) {
1275                 state = rb_entry(node, struct extent_state, rb_node);
1276
1277                 if (filled && state->start > start) {
1278                         bitset = 0;
1279                         break;
1280                 }
1281
1282                 if (state->start > end)
1283                         break;
1284
1285                 if (state->state & bits) {
1286                         bitset = 1;
1287                         if (!filled)
1288                                 break;
1289                 } else if (filled) {
1290                         bitset = 0;
1291                         break;
1292                 }
1293                 start = state->end + 1;
1294                 if (start > end)
1295                         break;
1296                 node = rb_next(node);
1297                 if (!node) {
1298                         if (filled)
1299                                 bitset = 0;
1300                         break;
1301                 }
1302         }
1303         spin_unlock_irqrestore(&tree->lock, flags);
1304         return bitset;
1305 }
1306 EXPORT_SYMBOL(test_range_bit);
1307
1308 /*
1309  * helper function to set a given page up to date if all the
1310  * extents in the tree for that page are up to date
1311  */
1312 static int check_page_uptodate(struct extent_io_tree *tree,
1313                                struct page *page)
1314 {
1315         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1316         u64 end = start + PAGE_CACHE_SIZE - 1;
1317         if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1))
1318                 SetPageUptodate(page);
1319         return 0;
1320 }
1321
1322 /*
1323  * helper function to unlock a page if all the extents in the tree
1324  * for that page are unlocked
1325  */
1326 static int check_page_locked(struct extent_io_tree *tree,
1327                              struct page *page)
1328 {
1329         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1330         u64 end = start + PAGE_CACHE_SIZE - 1;
1331         if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0))
1332                 unlock_page(page);
1333         return 0;
1334 }
1335
1336 /*
1337  * helper function to end page writeback if all the extents
1338  * in the tree for that page are done with writeback
1339  */
1340 static int check_page_writeback(struct extent_io_tree *tree,
1341                              struct page *page)
1342 {
1343         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1344         u64 end = start + PAGE_CACHE_SIZE - 1;
1345         if (!test_range_bit(tree, start, end, EXTENT_WRITEBACK, 0))
1346                 end_page_writeback(page);
1347         return 0;
1348 }
1349
1350 /* lots and lots of room for performance fixes in the end_bio funcs */
1351
1352 /*
1353  * after a writepage IO is done, we need to:
1354  * clear the uptodate bits on error
1355  * clear the writeback bits in the extent tree for this IO
1356  * end_page_writeback if the page has no more pending IO
1357  *
1358  * Scheduling is not allowed, so the extent state tree is expected
1359  * to have one and only one object corresponding to this IO.
1360  */
1361 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1362 static void end_bio_extent_writepage(struct bio *bio, int err)
1363 #else
1364 static int end_bio_extent_writepage(struct bio *bio,
1365                                    unsigned int bytes_done, int err)
1366 #endif
1367 {
1368         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1369         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1370         struct extent_state *state = bio->bi_private;
1371         struct extent_io_tree *tree = state->tree;
1372         struct rb_node *node;
1373         u64 start;
1374         u64 end;
1375         u64 cur;
1376         int whole_page;
1377         unsigned long flags;
1378
1379 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1380         if (bio->bi_size)
1381                 return 1;
1382 #endif
1383         do {
1384                 struct page *page = bvec->bv_page;
1385                 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1386                          bvec->bv_offset;
1387                 end = start + bvec->bv_len - 1;
1388
1389                 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1390                         whole_page = 1;
1391                 else
1392                         whole_page = 0;
1393
1394                 if (--bvec >= bio->bi_io_vec)
1395                         prefetchw(&bvec->bv_page->flags);
1396
1397                 if (!uptodate) {
1398                         clear_extent_uptodate(tree, start, end, GFP_ATOMIC);
1399                         ClearPageUptodate(page);
1400                         SetPageError(page);
1401                 }
1402
1403                 if (tree->ops && tree->ops->writepage_end_io_hook) {
1404                         tree->ops->writepage_end_io_hook(page, start, end,
1405                                                          state);
1406                 }
1407
1408                 /*
1409                  * bios can get merged in funny ways, and so we need to
1410                  * be careful with the state variable.  We know the
1411                  * state won't be merged with others because it has
1412                  * WRITEBACK set, but we can't be sure each biovec is
1413                  * sequential in the file.  So, if our cached state
1414                  * doesn't match the expected end, search the tree
1415                  * for the correct one.
1416                  */
1417
1418                 spin_lock_irqsave(&tree->lock, flags);
1419                 if (!state || state->end != end) {
1420                         state = NULL;
1421                         node = __etree_search(tree, start, NULL, NULL);
1422                         if (node) {
1423                                 state = rb_entry(node, struct extent_state,
1424                                                  rb_node);
1425                                 if (state->end != end ||
1426                                     !(state->state & EXTENT_WRITEBACK))
1427                                         state = NULL;
1428                         }
1429                         if (!state) {
1430                                 spin_unlock_irqrestore(&tree->lock, flags);
1431                                 clear_extent_writeback(tree, start,
1432                                                        end, GFP_ATOMIC);
1433                                 goto next_io;
1434                         }
1435                 }
1436                 cur = end;
1437                 while(1) {
1438                         struct extent_state *clear = state;
1439                         cur = state->start;
1440                         node = rb_prev(&state->rb_node);
1441                         if (node) {
1442                                 state = rb_entry(node,
1443                                                  struct extent_state,
1444                                                  rb_node);
1445                         } else {
1446                                 state = NULL;
1447                         }
1448
1449                         clear_state_bit(tree, clear, EXTENT_WRITEBACK,
1450                                         1, 0);
1451                         if (cur == start)
1452                                 break;
1453                         if (cur < start) {
1454                                 WARN_ON(1);
1455                                 break;
1456                         }
1457                         if (!node)
1458                                 break;
1459                 }
1460                 /* before releasing the lock, make sure the next state
1461                  * variable has the expected bits set and corresponds
1462                  * to the correct offsets in the file
1463                  */
1464                 if (state && (state->end + 1 != start ||
1465                     !(state->state & EXTENT_WRITEBACK))) {
1466                         state = NULL;
1467                 }
1468                 spin_unlock_irqrestore(&tree->lock, flags);
1469 next_io:
1470
1471                 if (whole_page)
1472                         end_page_writeback(page);
1473                 else
1474                         check_page_writeback(tree, page);
1475         } while (bvec >= bio->bi_io_vec);
1476         bio_put(bio);
1477 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1478         return 0;
1479 #endif
1480 }
1481
1482 /*
1483  * after a readpage IO is done, we need to:
1484  * clear the uptodate bits on error
1485  * set the uptodate bits if things worked
1486  * set the page up to date if all extents in the tree are uptodate
1487  * clear the lock bit in the extent tree
1488  * unlock the page if there are no other extents locked for it
1489  *
1490  * Scheduling is not allowed, so the extent state tree is expected
1491  * to have one and only one object corresponding to this IO.
1492  */
1493 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1494 static void end_bio_extent_readpage(struct bio *bio, int err)
1495 #else
1496 static int end_bio_extent_readpage(struct bio *bio,
1497                                    unsigned int bytes_done, int err)
1498 #endif
1499 {
1500         int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1501         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1502         struct extent_state *state = bio->bi_private;
1503         struct extent_io_tree *tree = state->tree;
1504         struct rb_node *node;
1505         u64 start;
1506         u64 end;
1507         u64 cur;
1508         unsigned long flags;
1509         int whole_page;
1510         int ret;
1511
1512 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1513         if (bio->bi_size)
1514                 return 1;
1515 #endif
1516
1517         do {
1518                 struct page *page = bvec->bv_page;
1519                 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1520                         bvec->bv_offset;
1521                 end = start + bvec->bv_len - 1;
1522
1523                 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1524                         whole_page = 1;
1525                 else
1526                         whole_page = 0;
1527
1528                 if (--bvec >= bio->bi_io_vec)
1529                         prefetchw(&bvec->bv_page->flags);
1530
1531                 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
1532                         ret = tree->ops->readpage_end_io_hook(page, start, end,
1533                                                               state);
1534                         if (ret)
1535                                 uptodate = 0;
1536                 }
1537
1538                 spin_lock_irqsave(&tree->lock, flags);
1539                 if (!state || state->end != end) {
1540                         state = NULL;
1541                         node = __etree_search(tree, start, NULL, NULL);
1542                         if (node) {
1543                                 state = rb_entry(node, struct extent_state,
1544                                                  rb_node);
1545                                 if (state->end != end ||
1546                                     !(state->state & EXTENT_LOCKED))
1547                                         state = NULL;
1548                         }
1549                         if (!state) {
1550                                 spin_unlock_irqrestore(&tree->lock, flags);
1551                                 set_extent_uptodate(tree, start, end,
1552                                                     GFP_ATOMIC);
1553                                 unlock_extent(tree, start, end, GFP_ATOMIC);
1554                                 goto next_io;
1555                         }
1556                 }
1557
1558                 cur = end;
1559                 while(1) {
1560                         struct extent_state *clear = state;
1561                         cur = state->start;
1562                         node = rb_prev(&state->rb_node);
1563                         if (node) {
1564                                 state = rb_entry(node,
1565                                          struct extent_state,
1566                                          rb_node);
1567                         } else {
1568                                 state = NULL;
1569                         }
1570                         set_state_cb(tree, clear, EXTENT_UPTODATE);
1571                         clear->state |= EXTENT_UPTODATE;
1572                         clear_state_bit(tree, clear, EXTENT_LOCKED,
1573                                         1, 0);
1574                         if (cur == start)
1575                                 break;
1576                         if (cur < start) {
1577                                 WARN_ON(1);
1578                                 break;
1579                         }
1580                         if (!node)
1581                                 break;
1582                 }
1583                 /* before releasing the lock, make sure the next state
1584                  * variable has the expected bits set and corresponds
1585                  * to the correct offsets in the file
1586                  */
1587                 if (state && (state->end + 1 != start ||
1588                     !(state->state & EXTENT_LOCKED))) {
1589                         state = NULL;
1590                 }
1591                 spin_unlock_irqrestore(&tree->lock, flags);
1592 next_io:
1593                 if (whole_page) {
1594                         if (uptodate) {
1595                                 SetPageUptodate(page);
1596                         } else {
1597                                 ClearPageUptodate(page);
1598                                 SetPageError(page);
1599                         }
1600                         unlock_page(page);
1601                 } else {
1602                         if (uptodate) {
1603                                 check_page_uptodate(tree, page);
1604                         } else {
1605                                 ClearPageUptodate(page);
1606                                 SetPageError(page);
1607                         }
1608                         check_page_locked(tree, page);
1609                 }
1610         } while (bvec >= bio->bi_io_vec);
1611
1612         bio_put(bio);
1613 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1614         return 0;
1615 #endif
1616 }
1617
1618 /*
1619  * IO done from prepare_write is pretty simple, we just unlock
1620  * the structs in the extent tree when done, and set the uptodate bits
1621  * as appropriate.
1622  */
1623 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1624 static void end_bio_extent_preparewrite(struct bio *bio, int err)
1625 #else
1626 static int end_bio_extent_preparewrite(struct bio *bio,
1627                                        unsigned int bytes_done, int err)
1628 #endif
1629 {
1630         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1631         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1632         struct extent_state *state = bio->bi_private;
1633         struct extent_io_tree *tree = state->tree;
1634         u64 start;
1635         u64 end;
1636
1637 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1638         if (bio->bi_size)
1639                 return 1;
1640 #endif
1641
1642         do {
1643                 struct page *page = bvec->bv_page;
1644                 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1645                         bvec->bv_offset;
1646                 end = start + bvec->bv_len - 1;
1647
1648                 if (--bvec >= bio->bi_io_vec)
1649                         prefetchw(&bvec->bv_page->flags);
1650
1651                 if (uptodate) {
1652                         set_extent_uptodate(tree, start, end, GFP_ATOMIC);
1653                 } else {
1654                         ClearPageUptodate(page);
1655                         SetPageError(page);
1656                 }
1657
1658                 unlock_extent(tree, start, end, GFP_ATOMIC);
1659
1660         } while (bvec >= bio->bi_io_vec);
1661
1662         bio_put(bio);
1663 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1664         return 0;
1665 #endif
1666 }
1667
1668 static struct bio *
1669 extent_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
1670                  gfp_t gfp_flags)
1671 {
1672         struct bio *bio;
1673
1674         bio = bio_alloc(gfp_flags, nr_vecs);
1675
1676         if (bio == NULL && (current->flags & PF_MEMALLOC)) {
1677                 while (!bio && (nr_vecs /= 2))
1678                         bio = bio_alloc(gfp_flags, nr_vecs);
1679         }
1680
1681         if (bio) {
1682                 bio->bi_bdev = bdev;
1683                 bio->bi_sector = first_sector;
1684         }
1685         return bio;
1686 }
1687
1688 static int submit_one_bio(int rw, struct bio *bio)
1689 {
1690         u64 maxsector;
1691         int ret = 0;
1692         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1693         struct page *page = bvec->bv_page;
1694         struct extent_io_tree *tree = bio->bi_private;
1695         struct rb_node *node;
1696         struct extent_state *state;
1697         u64 start;
1698         u64 end;
1699
1700         start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
1701         end = start + bvec->bv_len - 1;
1702
1703         spin_lock_irq(&tree->lock);
1704         node = __etree_search(tree, start, NULL, NULL);
1705         BUG_ON(!node);
1706         state = rb_entry(node, struct extent_state, rb_node);
1707         while(state->end < end) {
1708                 node = rb_next(node);
1709                 state = rb_entry(node, struct extent_state, rb_node);
1710         }
1711         BUG_ON(state->end != end);
1712         spin_unlock_irq(&tree->lock);
1713
1714         bio->bi_private = state;
1715
1716         bio_get(bio);
1717
1718         maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
1719         if (maxsector < bio->bi_sector) {
1720                 printk("sector too large max %Lu got %llu\n", maxsector,
1721                         (unsigned long long)bio->bi_sector);
1722                 WARN_ON(1);
1723         }
1724         if (tree->ops && tree->ops->submit_bio_hook)
1725                 tree->ops->submit_bio_hook(page->mapping->host, rw, bio);
1726         else
1727                 submit_bio(rw, bio);
1728         if (bio_flagged(bio, BIO_EOPNOTSUPP))
1729                 ret = -EOPNOTSUPP;
1730         bio_put(bio);
1731         return ret;
1732 }
1733
1734 static int submit_extent_page(int rw, struct extent_io_tree *tree,
1735                               struct page *page, sector_t sector,
1736                               size_t size, unsigned long offset,
1737                               struct block_device *bdev,
1738                               struct bio **bio_ret,
1739                               unsigned long max_pages,
1740                               bio_end_io_t end_io_func)
1741 {
1742         int ret = 0;
1743         struct bio *bio;
1744         int nr;
1745
1746         if (bio_ret && *bio_ret) {
1747                 bio = *bio_ret;
1748                 if (bio->bi_sector + (bio->bi_size >> 9) != sector ||
1749                     (tree->ops && tree->ops->merge_bio_hook &&
1750                      tree->ops->merge_bio_hook(page, offset, size, bio)) ||
1751                     bio_add_page(bio, page, size, offset) < size) {
1752                         ret = submit_one_bio(rw, bio);
1753                         bio = NULL;
1754                 } else {
1755                         return 0;
1756                 }
1757         }
1758         nr = bio_get_nr_vecs(bdev);
1759         bio = extent_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
1760         if (!bio) {
1761                 printk("failed to allocate bio nr %d\n", nr);
1762         }
1763
1764
1765         bio_add_page(bio, page, size, offset);
1766         bio->bi_end_io = end_io_func;
1767         bio->bi_private = tree;
1768
1769         if (bio_ret) {
1770                 *bio_ret = bio;
1771         } else {
1772                 ret = submit_one_bio(rw, bio);
1773         }
1774
1775         return ret;
1776 }
1777
1778 void set_page_extent_mapped(struct page *page)
1779 {
1780         if (!PagePrivate(page)) {
1781                 SetPagePrivate(page);
1782                 WARN_ON(!page->mapping->a_ops->invalidatepage);
1783                 set_page_private(page, EXTENT_PAGE_PRIVATE);
1784                 page_cache_get(page);
1785         }
1786 }
1787
1788 void set_page_extent_head(struct page *page, unsigned long len)
1789 {
1790         set_page_private(page, EXTENT_PAGE_PRIVATE_FIRST_PAGE | len << 2);
1791 }
1792
1793 /*
1794  * basic readpage implementation.  Locked extent state structs are inserted
1795  * into the tree that are removed when the IO is done (by the end_io
1796  * handlers)
1797  */
1798 static int __extent_read_full_page(struct extent_io_tree *tree,
1799                                    struct page *page,
1800                                    get_extent_t *get_extent,
1801                                    struct bio **bio)
1802 {
1803         struct inode *inode = page->mapping->host;
1804         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1805         u64 page_end = start + PAGE_CACHE_SIZE - 1;
1806         u64 end;
1807         u64 cur = start;
1808         u64 extent_offset;
1809         u64 last_byte = i_size_read(inode);
1810         u64 block_start;
1811         u64 cur_end;
1812         sector_t sector;
1813         struct extent_map *em;
1814         struct block_device *bdev;
1815         int ret;
1816         int nr = 0;
1817         size_t page_offset = 0;
1818         size_t iosize;
1819         size_t blocksize = inode->i_sb->s_blocksize;
1820
1821         set_page_extent_mapped(page);
1822
1823         end = page_end;
1824         lock_extent(tree, start, end, GFP_NOFS);
1825
1826         while (cur <= end) {
1827                 if (cur >= last_byte) {
1828                         char *userpage;
1829                         iosize = PAGE_CACHE_SIZE - page_offset;
1830                         userpage = kmap_atomic(page, KM_USER0);
1831                         memset(userpage + page_offset, 0, iosize);
1832                         flush_dcache_page(page);
1833                         kunmap_atomic(userpage, KM_USER0);
1834                         set_extent_uptodate(tree, cur, cur + iosize - 1,
1835                                             GFP_NOFS);
1836                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1837                         break;
1838                 }
1839                 em = get_extent(inode, page, page_offset, cur,
1840                                 end - cur + 1, 0);
1841                 if (IS_ERR(em) || !em) {
1842                         SetPageError(page);
1843                         unlock_extent(tree, cur, end, GFP_NOFS);
1844                         break;
1845                 }
1846
1847                 extent_offset = cur - em->start;
1848                 BUG_ON(extent_map_end(em) <= cur);
1849                 BUG_ON(end < cur);
1850
1851                 iosize = min(extent_map_end(em) - cur, end - cur + 1);
1852                 cur_end = min(extent_map_end(em) - 1, end);
1853                 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
1854                 sector = (em->block_start + extent_offset) >> 9;
1855                 bdev = em->bdev;
1856                 block_start = em->block_start;
1857                 free_extent_map(em);
1858                 em = NULL;
1859
1860                 /* we've found a hole, just zero and go on */
1861                 if (block_start == EXTENT_MAP_HOLE) {
1862                         char *userpage;
1863                         userpage = kmap_atomic(page, KM_USER0);
1864                         memset(userpage + page_offset, 0, iosize);
1865                         flush_dcache_page(page);
1866                         kunmap_atomic(userpage, KM_USER0);
1867
1868                         set_extent_uptodate(tree, cur, cur + iosize - 1,
1869                                             GFP_NOFS);
1870                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1871                         cur = cur + iosize;
1872                         page_offset += iosize;
1873                         continue;
1874                 }
1875                 /* the get_extent function already copied into the page */
1876                 if (test_range_bit(tree, cur, cur_end, EXTENT_UPTODATE, 1)) {
1877                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1878                         cur = cur + iosize;
1879                         page_offset += iosize;
1880                         continue;
1881                 }
1882                 /* we have an inline extent but it didn't get marked up
1883                  * to date.  Error out
1884                  */
1885                 if (block_start == EXTENT_MAP_INLINE) {
1886                         SetPageError(page);
1887                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1888                         cur = cur + iosize;
1889                         page_offset += iosize;
1890                         continue;
1891                 }
1892
1893                 ret = 0;
1894                 if (tree->ops && tree->ops->readpage_io_hook) {
1895                         ret = tree->ops->readpage_io_hook(page, cur,
1896                                                           cur + iosize - 1);
1897                 }
1898                 if (!ret) {
1899                         unsigned long nr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
1900                         nr -= page->index;
1901                         ret = submit_extent_page(READ, tree, page,
1902                                          sector, iosize, page_offset,
1903                                          bdev, bio, nr,
1904                                          end_bio_extent_readpage);
1905                 }
1906                 if (ret)
1907                         SetPageError(page);
1908                 cur = cur + iosize;
1909                 page_offset += iosize;
1910                 nr++;
1911         }
1912         if (!nr) {
1913                 if (!PageError(page))
1914                         SetPageUptodate(page);
1915                 unlock_page(page);
1916         }
1917         return 0;
1918 }
1919
1920 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
1921                             get_extent_t *get_extent)
1922 {
1923         struct bio *bio = NULL;
1924         int ret;
1925
1926         ret = __extent_read_full_page(tree, page, get_extent, &bio);
1927         if (bio)
1928                 submit_one_bio(READ, bio);
1929         return ret;
1930 }
1931 EXPORT_SYMBOL(extent_read_full_page);
1932
1933 /*
1934  * the writepage semantics are similar to regular writepage.  extent
1935  * records are inserted to lock ranges in the tree, and as dirty areas
1936  * are found, they are marked writeback.  Then the lock bits are removed
1937  * and the end_io handler clears the writeback ranges
1938  */
1939 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
1940                               void *data)
1941 {
1942         struct inode *inode = page->mapping->host;
1943         struct extent_page_data *epd = data;
1944         struct extent_io_tree *tree = epd->tree;
1945         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1946         u64 delalloc_start;
1947         u64 page_end = start + PAGE_CACHE_SIZE - 1;
1948         u64 end;
1949         u64 cur = start;
1950         u64 extent_offset;
1951         u64 last_byte = i_size_read(inode);
1952         u64 block_start;
1953         u64 iosize;
1954         sector_t sector;
1955         struct extent_map *em;
1956         struct block_device *bdev;
1957         int ret;
1958         int nr = 0;
1959         size_t page_offset = 0;
1960         size_t blocksize;
1961         loff_t i_size = i_size_read(inode);
1962         unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
1963         u64 nr_delalloc;
1964         u64 delalloc_end;
1965
1966         WARN_ON(!PageLocked(page));
1967         if (page->index > end_index) {
1968                 clear_extent_dirty(tree, start, page_end, GFP_NOFS);
1969                 unlock_page(page);
1970                 return 0;
1971         }
1972
1973         if (page->index == end_index) {
1974                 char *userpage;
1975
1976                 size_t offset = i_size & (PAGE_CACHE_SIZE - 1);
1977
1978                 userpage = kmap_atomic(page, KM_USER0);
1979                 memset(userpage + offset, 0, PAGE_CACHE_SIZE - offset);
1980                 flush_dcache_page(page);
1981                 kunmap_atomic(userpage, KM_USER0);
1982         }
1983
1984         set_page_extent_mapped(page);
1985
1986         delalloc_start = start;
1987         delalloc_end = 0;
1988         while(delalloc_end < page_end) {
1989                 nr_delalloc = find_lock_delalloc_range(tree, &delalloc_start,
1990                                                        &delalloc_end,
1991                                                        128 * 1024 * 1024);
1992                 if (nr_delalloc == 0) {
1993                         delalloc_start = delalloc_end + 1;
1994                         continue;
1995                 }
1996                 tree->ops->fill_delalloc(inode, delalloc_start,
1997                                          delalloc_end);
1998                 clear_extent_bit(tree, delalloc_start,
1999                                  delalloc_end,
2000                                  EXTENT_LOCKED | EXTENT_DELALLOC,
2001                                  1, 0, GFP_NOFS);
2002                 delalloc_start = delalloc_end + 1;
2003         }
2004         lock_extent(tree, start, page_end, GFP_NOFS);
2005
2006         end = page_end;
2007         if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0)) {
2008                 printk("found delalloc bits after lock_extent\n");
2009         }
2010
2011         if (last_byte <= start) {
2012                 clear_extent_dirty(tree, start, page_end, GFP_NOFS);
2013                 goto done;
2014         }
2015
2016         set_extent_uptodate(tree, start, page_end, GFP_NOFS);
2017         blocksize = inode->i_sb->s_blocksize;
2018
2019         while (cur <= end) {
2020                 if (cur >= last_byte) {
2021                         clear_extent_dirty(tree, cur, page_end, GFP_NOFS);
2022                         break;
2023                 }
2024                 em = epd->get_extent(inode, page, page_offset, cur,
2025                                      end - cur + 1, 1);
2026                 if (IS_ERR(em) || !em) {
2027                         SetPageError(page);
2028                         break;
2029                 }
2030
2031                 extent_offset = cur - em->start;
2032                 BUG_ON(extent_map_end(em) <= cur);
2033                 BUG_ON(end < cur);
2034                 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2035                 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2036                 sector = (em->block_start + extent_offset) >> 9;
2037                 bdev = em->bdev;
2038                 block_start = em->block_start;
2039                 free_extent_map(em);
2040                 em = NULL;
2041
2042                 if (block_start == EXTENT_MAP_HOLE ||
2043                     block_start == EXTENT_MAP_INLINE) {
2044                         clear_extent_dirty(tree, cur,
2045                                            cur + iosize - 1, GFP_NOFS);
2046                         cur = cur + iosize;
2047                         page_offset += iosize;
2048                         continue;
2049                 }
2050
2051                 /* leave this out until we have a page_mkwrite call */
2052                 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
2053                                    EXTENT_DIRTY, 0)) {
2054                         cur = cur + iosize;
2055                         page_offset += iosize;
2056                         continue;
2057                 }
2058                 clear_extent_dirty(tree, cur, cur + iosize - 1, GFP_NOFS);
2059                 if (tree->ops && tree->ops->writepage_io_hook) {
2060                         ret = tree->ops->writepage_io_hook(page, cur,
2061                                                 cur + iosize - 1);
2062                 } else {
2063                         ret = 0;
2064                 }
2065                 if (ret)
2066                         SetPageError(page);
2067                 else {
2068                         unsigned long max_nr = end_index + 1;
2069                         set_range_writeback(tree, cur, cur + iosize - 1);
2070                         if (!PageWriteback(page)) {
2071                                 printk("warning page %lu not writeback, "
2072                                        "cur %llu end %llu\n", page->index,
2073                                        (unsigned long long)cur,
2074                                        (unsigned long long)end);
2075                         }
2076
2077                         ret = submit_extent_page(WRITE, tree, page, sector,
2078                                                  iosize, page_offset, bdev,
2079                                                  &epd->bio, max_nr,
2080                                                  end_bio_extent_writepage);
2081                         if (ret)
2082                                 SetPageError(page);
2083                 }
2084                 cur = cur + iosize;
2085                 page_offset += iosize;
2086                 nr++;
2087         }
2088 done:
2089         if (nr == 0) {
2090                 /* make sure the mapping tag for page dirty gets cleared */
2091                 set_page_writeback(page);
2092                 end_page_writeback(page);
2093         }
2094         unlock_extent(tree, start, page_end, GFP_NOFS);
2095         unlock_page(page);
2096         return 0;
2097 }
2098
2099 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
2100
2101 /* Taken directly from 2.6.23 for 2.6.18 back port */
2102 typedef int (*writepage_t)(struct page *page, struct writeback_control *wbc,
2103                                 void *data);
2104
2105 /**
2106  * write_cache_pages - walk the list of dirty pages of the given address space
2107  * and write all of them.
2108  * @mapping: address space structure to write
2109  * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2110  * @writepage: function called for each page
2111  * @data: data passed to writepage function
2112  *
2113  * If a page is already under I/O, write_cache_pages() skips it, even
2114  * if it's dirty.  This is desirable behaviour for memory-cleaning writeback,
2115  * but it is INCORRECT for data-integrity system calls such as fsync().  fsync()
2116  * and msync() need to guarantee that all the data which was dirty at the time
2117  * the call was made get new I/O started against them.  If wbc->sync_mode is
2118  * WB_SYNC_ALL then we were called for data integrity and we must wait for
2119  * existing IO to complete.
2120  */
2121 static int write_cache_pages(struct address_space *mapping,
2122                       struct writeback_control *wbc, writepage_t writepage,
2123                       void *data)
2124 {
2125         struct backing_dev_info *bdi = mapping->backing_dev_info;
2126         int ret = 0;
2127         int done = 0;
2128         struct pagevec pvec;
2129         int nr_pages;
2130         pgoff_t index;
2131         pgoff_t end;            /* Inclusive */
2132         int scanned = 0;
2133         int range_whole = 0;
2134
2135         if (wbc->nonblocking && bdi_write_congested(bdi)) {
2136                 wbc->encountered_congestion = 1;
2137                 return 0;
2138         }
2139
2140         pagevec_init(&pvec, 0);
2141         if (wbc->range_cyclic) {
2142                 index = mapping->writeback_index; /* Start from prev offset */
2143                 end = -1;
2144         } else {
2145                 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2146                 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2147                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2148                         range_whole = 1;
2149                 scanned = 1;
2150         }
2151 retry:
2152         while (!done && (index <= end) &&
2153                (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
2154                                               PAGECACHE_TAG_DIRTY,
2155                                               min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
2156                 unsigned i;
2157
2158                 scanned = 1;
2159                 for (i = 0; i < nr_pages; i++) {
2160                         struct page *page = pvec.pages[i];
2161
2162                         /*
2163                          * At this point we hold neither mapping->tree_lock nor
2164                          * lock on the page itself: the page may be truncated or
2165                          * invalidated (changing page->mapping to NULL), or even
2166                          * swizzled back from swapper_space to tmpfs file
2167                          * mapping
2168                          */
2169                         lock_page(page);
2170
2171                         if (unlikely(page->mapping != mapping)) {
2172                                 unlock_page(page);
2173                                 continue;
2174                         }
2175
2176                         if (!wbc->range_cyclic && page->index > end) {
2177                                 done = 1;
2178                                 unlock_page(page);
2179                                 continue;
2180                         }
2181
2182                         if (wbc->sync_mode != WB_SYNC_NONE)
2183                                 wait_on_page_writeback(page);
2184
2185                         if (PageWriteback(page) ||
2186                             !clear_page_dirty_for_io(page)) {
2187                                 unlock_page(page);
2188                                 continue;
2189                         }
2190
2191                         ret = (*writepage)(page, wbc, data);
2192
2193                         if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
2194                                 unlock_page(page);
2195                                 ret = 0;
2196                         }
2197                         if (ret || (--(wbc->nr_to_write) <= 0))
2198                                 done = 1;
2199                         if (wbc->nonblocking && bdi_write_congested(bdi)) {
2200                                 wbc->encountered_congestion = 1;
2201                                 done = 1;
2202                         }
2203                 }
2204                 pagevec_release(&pvec);
2205                 cond_resched();
2206         }
2207         if (!scanned && !done) {
2208                 /*
2209                  * We hit the last page and there is more work to be done: wrap
2210                  * back to the start of the file
2211                  */
2212                 scanned = 1;
2213                 index = 0;
2214                 goto retry;
2215         }
2216         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2217                 mapping->writeback_index = index;
2218         return ret;
2219 }
2220 #endif
2221
2222 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
2223                           get_extent_t *get_extent,
2224                           struct writeback_control *wbc)
2225 {
2226         int ret;
2227         struct address_space *mapping = page->mapping;
2228         struct extent_page_data epd = {
2229                 .bio = NULL,
2230                 .tree = tree,
2231                 .get_extent = get_extent,
2232         };
2233         struct writeback_control wbc_writepages = {
2234                 .bdi            = wbc->bdi,
2235                 .sync_mode      = WB_SYNC_NONE,
2236                 .older_than_this = NULL,
2237                 .nr_to_write    = 64,
2238                 .range_start    = page_offset(page) + PAGE_CACHE_SIZE,
2239                 .range_end      = (loff_t)-1,
2240         };
2241
2242
2243         ret = __extent_writepage(page, wbc, &epd);
2244
2245         write_cache_pages(mapping, &wbc_writepages, __extent_writepage, &epd);
2246         if (epd.bio) {
2247                 submit_one_bio(WRITE, epd.bio);
2248         }
2249         return ret;
2250 }
2251 EXPORT_SYMBOL(extent_write_full_page);
2252
2253
2254 int extent_writepages(struct extent_io_tree *tree,
2255                       struct address_space *mapping,
2256                       get_extent_t *get_extent,
2257                       struct writeback_control *wbc)
2258 {
2259         int ret = 0;
2260         struct extent_page_data epd = {
2261                 .bio = NULL,
2262                 .tree = tree,
2263                 .get_extent = get_extent,
2264         };
2265
2266         ret = write_cache_pages(mapping, wbc, __extent_writepage, &epd);
2267         if (epd.bio) {
2268                 submit_one_bio(WRITE, epd.bio);
2269         }
2270         return ret;
2271 }
2272 EXPORT_SYMBOL(extent_writepages);
2273
2274 int extent_readpages(struct extent_io_tree *tree,
2275                      struct address_space *mapping,
2276                      struct list_head *pages, unsigned nr_pages,
2277                      get_extent_t get_extent)
2278 {
2279         struct bio *bio = NULL;
2280         unsigned page_idx;
2281         struct pagevec pvec;
2282
2283         pagevec_init(&pvec, 0);
2284         for (page_idx = 0; page_idx < nr_pages; page_idx++) {
2285                 struct page *page = list_entry(pages->prev, struct page, lru);
2286
2287                 prefetchw(&page->flags);
2288                 list_del(&page->lru);
2289                 /*
2290                  * what we want to do here is call add_to_page_cache_lru,
2291                  * but that isn't exported, so we reproduce it here
2292                  */
2293                 if (!add_to_page_cache(page, mapping,
2294                                         page->index, GFP_KERNEL)) {
2295
2296                         /* open coding of lru_cache_add, also not exported */
2297                         page_cache_get(page);
2298                         if (!pagevec_add(&pvec, page))
2299                                 __pagevec_lru_add(&pvec);
2300                         __extent_read_full_page(tree, page, get_extent, &bio);
2301                 }
2302                 page_cache_release(page);
2303         }
2304         if (pagevec_count(&pvec))
2305                 __pagevec_lru_add(&pvec);
2306         BUG_ON(!list_empty(pages));
2307         if (bio)
2308                 submit_one_bio(READ, bio);
2309         return 0;
2310 }
2311 EXPORT_SYMBOL(extent_readpages);
2312
2313 /*
2314  * basic invalidatepage code, this waits on any locked or writeback
2315  * ranges corresponding to the page, and then deletes any extent state
2316  * records from the tree
2317  */
2318 int extent_invalidatepage(struct extent_io_tree *tree,
2319                           struct page *page, unsigned long offset)
2320 {
2321         u64 start = ((u64)page->index << PAGE_CACHE_SHIFT);
2322         u64 end = start + PAGE_CACHE_SIZE - 1;
2323         size_t blocksize = page->mapping->host->i_sb->s_blocksize;
2324
2325         start += (offset + blocksize -1) & ~(blocksize - 1);
2326         if (start > end)
2327                 return 0;
2328
2329         lock_extent(tree, start, end, GFP_NOFS);
2330         wait_on_extent_writeback(tree, start, end);
2331         clear_extent_bit(tree, start, end,
2332                          EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC,
2333                          1, 1, GFP_NOFS);
2334         return 0;
2335 }
2336 EXPORT_SYMBOL(extent_invalidatepage);
2337
2338 /*
2339  * simple commit_write call, set_range_dirty is used to mark both
2340  * the pages and the extent records as dirty
2341  */
2342 int extent_commit_write(struct extent_io_tree *tree,
2343                         struct inode *inode, struct page *page,
2344                         unsigned from, unsigned to)
2345 {
2346         loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
2347
2348         set_page_extent_mapped(page);
2349         set_page_dirty(page);
2350
2351         if (pos > inode->i_size) {
2352                 i_size_write(inode, pos);
2353                 mark_inode_dirty(inode);
2354         }
2355         return 0;
2356 }
2357 EXPORT_SYMBOL(extent_commit_write);
2358
2359 int extent_prepare_write(struct extent_io_tree *tree,
2360                          struct inode *inode, struct page *page,
2361                          unsigned from, unsigned to, get_extent_t *get_extent)
2362 {
2363         u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2364         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
2365         u64 block_start;
2366         u64 orig_block_start;
2367         u64 block_end;
2368         u64 cur_end;
2369         struct extent_map *em;
2370         unsigned blocksize = 1 << inode->i_blkbits;
2371         size_t page_offset = 0;
2372         size_t block_off_start;
2373         size_t block_off_end;
2374         int err = 0;
2375         int iocount = 0;
2376         int ret = 0;
2377         int isnew;
2378
2379         set_page_extent_mapped(page);
2380
2381         block_start = (page_start + from) & ~((u64)blocksize - 1);
2382         block_end = (page_start + to - 1) | (blocksize - 1);
2383         orig_block_start = block_start;
2384
2385         lock_extent(tree, page_start, page_end, GFP_NOFS);
2386         while(block_start <= block_end) {
2387                 em = get_extent(inode, page, page_offset, block_start,
2388                                 block_end - block_start + 1, 1);
2389                 if (IS_ERR(em) || !em) {
2390                         goto err;
2391                 }
2392                 cur_end = min(block_end, extent_map_end(em) - 1);
2393                 block_off_start = block_start & (PAGE_CACHE_SIZE - 1);
2394                 block_off_end = block_off_start + blocksize;
2395                 isnew = clear_extent_new(tree, block_start, cur_end, GFP_NOFS);
2396
2397                 if (!PageUptodate(page) && isnew &&
2398                     (block_off_end > to || block_off_start < from)) {
2399                         void *kaddr;
2400
2401                         kaddr = kmap_atomic(page, KM_USER0);
2402                         if (block_off_end > to)
2403                                 memset(kaddr + to, 0, block_off_end - to);
2404                         if (block_off_start < from)
2405                                 memset(kaddr + block_off_start, 0,
2406                                        from - block_off_start);
2407                         flush_dcache_page(page);
2408                         kunmap_atomic(kaddr, KM_USER0);
2409                 }
2410                 if ((em->block_start != EXTENT_MAP_HOLE &&
2411                      em->block_start != EXTENT_MAP_INLINE) &&
2412                     !isnew && !PageUptodate(page) &&
2413                     (block_off_end > to || block_off_start < from) &&
2414                     !test_range_bit(tree, block_start, cur_end,
2415                                     EXTENT_UPTODATE, 1)) {
2416                         u64 sector;
2417                         u64 extent_offset = block_start - em->start;
2418                         size_t iosize;
2419                         sector = (em->block_start + extent_offset) >> 9;
2420                         iosize = (cur_end - block_start + blocksize) &
2421                                 ~((u64)blocksize - 1);
2422                         /*
2423                          * we've already got the extent locked, but we
2424                          * need to split the state such that our end_bio
2425                          * handler can clear the lock.
2426                          */
2427                         set_extent_bit(tree, block_start,
2428                                        block_start + iosize - 1,
2429                                        EXTENT_LOCKED, 0, NULL, GFP_NOFS);
2430                         ret = submit_extent_page(READ, tree, page,
2431                                          sector, iosize, page_offset, em->bdev,
2432                                          NULL, 1,
2433                                          end_bio_extent_preparewrite);
2434                         iocount++;
2435                         block_start = block_start + iosize;
2436                 } else {
2437                         set_extent_uptodate(tree, block_start, cur_end,
2438                                             GFP_NOFS);
2439                         unlock_extent(tree, block_start, cur_end, GFP_NOFS);
2440                         block_start = cur_end + 1;
2441                 }
2442                 page_offset = block_start & (PAGE_CACHE_SIZE - 1);
2443                 free_extent_map(em);
2444         }
2445         if (iocount) {
2446                 wait_extent_bit(tree, orig_block_start,
2447                                 block_end, EXTENT_LOCKED);
2448         }
2449         check_page_uptodate(tree, page);
2450 err:
2451         /* FIXME, zero out newly allocated blocks on error */
2452         return err;
2453 }
2454 EXPORT_SYMBOL(extent_prepare_write);
2455
2456 /*
2457  * a helper for releasepage.  As long as there are no locked extents
2458  * in the range corresponding to the page, both state records and extent
2459  * map records are removed
2460  */
2461 int try_release_extent_mapping(struct extent_map_tree *map,
2462                                struct extent_io_tree *tree, struct page *page,
2463                                gfp_t mask)
2464 {
2465         struct extent_map *em;
2466         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2467         u64 end = start + PAGE_CACHE_SIZE - 1;
2468         u64 orig_start = start;
2469         int ret = 1;
2470         if ((mask & __GFP_WAIT) &&
2471             page->mapping->host->i_size > 16 * 1024 * 1024) {
2472                 u64 len;
2473                 while (start <= end) {
2474                         len = end - start + 1;
2475                         spin_lock(&map->lock);
2476                         em = lookup_extent_mapping(map, start, len);
2477                         if (!em || IS_ERR(em)) {
2478                                 spin_unlock(&map->lock);
2479                                 break;
2480                         }
2481                         if (em->start != start) {
2482                                 spin_unlock(&map->lock);
2483                                 free_extent_map(em);
2484                                 break;
2485                         }
2486                         if (!test_range_bit(tree, em->start,
2487                                             extent_map_end(em) - 1,
2488                                             EXTENT_LOCKED, 0)) {
2489                                 remove_extent_mapping(map, em);
2490                                 /* once for the rb tree */
2491                                 free_extent_map(em);
2492                         }
2493                         start = extent_map_end(em);
2494                         spin_unlock(&map->lock);
2495
2496                         /* once for us */
2497                         free_extent_map(em);
2498                 }
2499         }
2500         if (test_range_bit(tree, orig_start, end, EXTENT_IOBITS, 0))
2501                 ret = 0;
2502         else {
2503                 if ((mask & GFP_NOFS) == GFP_NOFS)
2504                         mask = GFP_NOFS;
2505                 clear_extent_bit(tree, orig_start, end, EXTENT_UPTODATE,
2506                                  1, 1, mask);
2507         }
2508         return ret;
2509 }
2510 EXPORT_SYMBOL(try_release_extent_mapping);
2511
2512 sector_t extent_bmap(struct address_space *mapping, sector_t iblock,
2513                 get_extent_t *get_extent)
2514 {
2515         struct inode *inode = mapping->host;
2516         u64 start = iblock << inode->i_blkbits;
2517         sector_t sector = 0;
2518         struct extent_map *em;
2519
2520         em = get_extent(inode, NULL, 0, start, (1 << inode->i_blkbits), 0);
2521         if (!em || IS_ERR(em))
2522                 return 0;
2523
2524         if (em->block_start == EXTENT_MAP_INLINE ||
2525             em->block_start == EXTENT_MAP_HOLE)
2526                 goto out;
2527
2528         sector = (em->block_start + start - em->start) >> inode->i_blkbits;
2529 out:
2530         free_extent_map(em);
2531         return sector;
2532 }
2533
2534 static int add_lru(struct extent_io_tree *tree, struct extent_buffer *eb)
2535 {
2536         if (list_empty(&eb->lru)) {
2537                 extent_buffer_get(eb);
2538                 list_add(&eb->lru, &tree->buffer_lru);
2539                 tree->lru_size++;
2540                 if (tree->lru_size >= BUFFER_LRU_MAX) {
2541                         struct extent_buffer *rm;
2542                         rm = list_entry(tree->buffer_lru.prev,
2543                                         struct extent_buffer, lru);
2544                         tree->lru_size--;
2545                         list_del_init(&rm->lru);
2546                         free_extent_buffer(rm);
2547                 }
2548         } else
2549                 list_move(&eb->lru, &tree->buffer_lru);
2550         return 0;
2551 }
2552 static struct extent_buffer *find_lru(struct extent_io_tree *tree,
2553                                       u64 start, unsigned long len)
2554 {
2555         struct list_head *lru = &tree->buffer_lru;
2556         struct list_head *cur = lru->next;
2557         struct extent_buffer *eb;
2558
2559         if (list_empty(lru))
2560                 return NULL;
2561
2562         do {
2563                 eb = list_entry(cur, struct extent_buffer, lru);
2564                 if (eb->start == start && eb->len == len) {
2565                         extent_buffer_get(eb);
2566                         return eb;
2567                 }
2568                 cur = cur->next;
2569         } while (cur != lru);
2570         return NULL;
2571 }
2572
2573 static inline unsigned long num_extent_pages(u64 start, u64 len)
2574 {
2575         return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
2576                 (start >> PAGE_CACHE_SHIFT);
2577 }
2578
2579 static inline struct page *extent_buffer_page(struct extent_buffer *eb,
2580                                               unsigned long i)
2581 {
2582         struct page *p;
2583         struct address_space *mapping;
2584
2585         if (i == 0)
2586                 return eb->first_page;
2587         i += eb->start >> PAGE_CACHE_SHIFT;
2588         mapping = eb->first_page->mapping;
2589         read_lock_irq(&mapping->tree_lock);
2590         p = radix_tree_lookup(&mapping->page_tree, i);
2591         read_unlock_irq(&mapping->tree_lock);
2592         return p;
2593 }
2594
2595 int invalidate_extent_lru(struct extent_io_tree *tree, u64 start,
2596                           unsigned long len)
2597 {
2598         struct list_head *lru = &tree->buffer_lru;
2599         struct list_head *cur = lru->next;
2600         struct extent_buffer *eb;
2601         int found = 0;
2602
2603         spin_lock(&tree->lru_lock);
2604         if (list_empty(lru))
2605                 goto out;
2606
2607         do {
2608                 eb = list_entry(cur, struct extent_buffer, lru);
2609                 if (eb->start <= start && eb->start + eb->len > start) {
2610                         eb->flags &= ~EXTENT_UPTODATE;
2611                 }
2612                 if (eb->start == start) {
2613                         eb->flags &= ~EXTENT_CSUM;
2614                 }
2615                 cur = cur->next;
2616         } while (cur != lru);
2617 out:
2618         spin_unlock(&tree->lru_lock);
2619         return found;
2620 }
2621
2622 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
2623                                                    u64 start,
2624                                                    unsigned long len,
2625                                                    gfp_t mask)
2626 {
2627         struct extent_buffer *eb = NULL;
2628         unsigned long flags;
2629
2630         spin_lock(&tree->lru_lock);
2631         eb = find_lru(tree, start, len);
2632         spin_unlock(&tree->lru_lock);
2633         if (eb) {
2634                 return eb;
2635         }
2636
2637         eb = kmem_cache_zalloc(extent_buffer_cache, mask);
2638         INIT_LIST_HEAD(&eb->lru);
2639         eb->start = start;
2640         eb->len = len;
2641         spin_lock_irqsave(&leak_lock, flags);
2642         list_add(&eb->leak_list, &buffers);
2643         spin_unlock_irqrestore(&leak_lock, flags);
2644         atomic_set(&eb->refs, 1);
2645
2646         return eb;
2647 }
2648
2649 static void __free_extent_buffer(struct extent_buffer *eb)
2650 {
2651         unsigned long flags;
2652         spin_lock_irqsave(&leak_lock, flags);
2653         list_del(&eb->leak_list);
2654         spin_unlock_irqrestore(&leak_lock, flags);
2655         kmem_cache_free(extent_buffer_cache, eb);
2656 }
2657
2658 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
2659                                           u64 start, unsigned long len,
2660                                           struct page *page0,
2661                                           gfp_t mask)
2662 {
2663         unsigned long num_pages = num_extent_pages(start, len);
2664         unsigned long i;
2665         unsigned long index = start >> PAGE_CACHE_SHIFT;
2666         struct extent_buffer *eb;
2667         struct page *p;
2668         struct address_space *mapping = tree->mapping;
2669         int uptodate = 1;
2670
2671         eb = __alloc_extent_buffer(tree, start, len, mask);
2672         if (!eb)
2673                 return NULL;
2674
2675         if (eb->flags & EXTENT_BUFFER_FILLED)
2676                 goto lru_add;
2677
2678         if (page0) {
2679                 eb->first_page = page0;
2680                 i = 1;
2681                 index++;
2682                 page_cache_get(page0);
2683                 mark_page_accessed(page0);
2684                 set_page_extent_mapped(page0);
2685                 WARN_ON(!PageUptodate(page0));
2686                 set_page_extent_head(page0, len);
2687         } else {
2688                 i = 0;
2689         }
2690         for (; i < num_pages; i++, index++) {
2691                 p = find_or_create_page(mapping, index, mask | __GFP_HIGHMEM);
2692                 if (!p) {
2693                         WARN_ON(1);
2694                         goto fail;
2695                 }
2696                 set_page_extent_mapped(p);
2697                 mark_page_accessed(p);
2698                 if (i == 0) {
2699                         eb->first_page = p;
2700                         set_page_extent_head(p, len);
2701                 } else {
2702                         set_page_private(p, EXTENT_PAGE_PRIVATE);
2703                 }
2704                 if (!PageUptodate(p))
2705                         uptodate = 0;
2706                 unlock_page(p);
2707         }
2708         if (uptodate)
2709                 eb->flags |= EXTENT_UPTODATE;
2710         eb->flags |= EXTENT_BUFFER_FILLED;
2711
2712 lru_add:
2713         spin_lock(&tree->lru_lock);
2714         add_lru(tree, eb);
2715         spin_unlock(&tree->lru_lock);
2716         return eb;
2717
2718 fail:
2719         spin_lock(&tree->lru_lock);
2720         list_del_init(&eb->lru);
2721         spin_unlock(&tree->lru_lock);
2722         if (!atomic_dec_and_test(&eb->refs))
2723                 return NULL;
2724         for (index = 1; index < i; index++) {
2725                 page_cache_release(extent_buffer_page(eb, index));
2726         }
2727         if (i > 0)
2728                 page_cache_release(extent_buffer_page(eb, 0));
2729         __free_extent_buffer(eb);
2730         return NULL;
2731 }
2732 EXPORT_SYMBOL(alloc_extent_buffer);
2733
2734 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
2735                                          u64 start, unsigned long len,
2736                                           gfp_t mask)
2737 {
2738         unsigned long num_pages = num_extent_pages(start, len);
2739         unsigned long i;
2740         unsigned long index = start >> PAGE_CACHE_SHIFT;
2741         struct extent_buffer *eb;
2742         struct page *p;
2743         struct address_space *mapping = tree->mapping;
2744         int uptodate = 1;
2745
2746         eb = __alloc_extent_buffer(tree, start, len, mask);
2747         if (!eb)
2748                 return NULL;
2749
2750         if (eb->flags & EXTENT_BUFFER_FILLED)
2751                 goto lru_add;
2752
2753         for (i = 0; i < num_pages; i++, index++) {
2754                 p = find_lock_page(mapping, index);
2755                 if (!p) {
2756                         goto fail;
2757                 }
2758                 set_page_extent_mapped(p);
2759                 mark_page_accessed(p);
2760
2761                 if (i == 0) {
2762                         eb->first_page = p;
2763                         set_page_extent_head(p, len);
2764                 } else {
2765                         set_page_private(p, EXTENT_PAGE_PRIVATE);
2766                 }
2767
2768                 if (!PageUptodate(p))
2769                         uptodate = 0;
2770                 unlock_page(p);
2771         }
2772         if (uptodate)
2773                 eb->flags |= EXTENT_UPTODATE;
2774         eb->flags |= EXTENT_BUFFER_FILLED;
2775
2776 lru_add:
2777         spin_lock(&tree->lru_lock);
2778         add_lru(tree, eb);
2779         spin_unlock(&tree->lru_lock);
2780         return eb;
2781 fail:
2782         spin_lock(&tree->lru_lock);
2783         list_del_init(&eb->lru);
2784         spin_unlock(&tree->lru_lock);
2785         if (!atomic_dec_and_test(&eb->refs))
2786                 return NULL;
2787         for (index = 1; index < i; index++) {
2788                 page_cache_release(extent_buffer_page(eb, index));
2789         }
2790         if (i > 0)
2791                 page_cache_release(extent_buffer_page(eb, 0));
2792         __free_extent_buffer(eb);
2793         return NULL;
2794 }
2795 EXPORT_SYMBOL(find_extent_buffer);
2796
2797 void free_extent_buffer(struct extent_buffer *eb)
2798 {
2799         unsigned long i;
2800         unsigned long num_pages;
2801
2802         if (!eb)
2803                 return;
2804
2805         if (!atomic_dec_and_test(&eb->refs))
2806                 return;
2807
2808         WARN_ON(!list_empty(&eb->lru));
2809         num_pages = num_extent_pages(eb->start, eb->len);
2810
2811         for (i = 1; i < num_pages; i++) {
2812                 page_cache_release(extent_buffer_page(eb, i));
2813         }
2814         page_cache_release(extent_buffer_page(eb, 0));
2815         __free_extent_buffer(eb);
2816 }
2817 EXPORT_SYMBOL(free_extent_buffer);
2818
2819 int clear_extent_buffer_dirty(struct extent_io_tree *tree,
2820                               struct extent_buffer *eb)
2821 {
2822         int set;
2823         unsigned long i;
2824         unsigned long num_pages;
2825         struct page *page;
2826
2827         u64 start = eb->start;
2828         u64 end = start + eb->len - 1;
2829
2830         set = clear_extent_dirty(tree, start, end, GFP_NOFS);
2831         num_pages = num_extent_pages(eb->start, eb->len);
2832
2833         for (i = 0; i < num_pages; i++) {
2834                 page = extent_buffer_page(eb, i);
2835                 lock_page(page);
2836                 if (i == 0)
2837                         set_page_extent_head(page, eb->len);
2838                 else
2839                         set_page_private(page, EXTENT_PAGE_PRIVATE);
2840
2841                 /*
2842                  * if we're on the last page or the first page and the
2843                  * block isn't aligned on a page boundary, do extra checks
2844                  * to make sure we don't clean page that is partially dirty
2845                  */
2846                 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
2847                     ((i == num_pages - 1) &&
2848                      ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
2849                         start = (u64)page->index << PAGE_CACHE_SHIFT;
2850                         end  = start + PAGE_CACHE_SIZE - 1;
2851                         if (test_range_bit(tree, start, end,
2852                                            EXTENT_DIRTY, 0)) {
2853                                 unlock_page(page);
2854                                 continue;
2855                         }
2856                 }
2857                 clear_page_dirty_for_io(page);
2858                 read_lock_irq(&page->mapping->tree_lock);
2859                 if (!PageDirty(page)) {
2860                         radix_tree_tag_clear(&page->mapping->page_tree,
2861                                                 page_index(page),
2862                                                 PAGECACHE_TAG_DIRTY);
2863                 }
2864                 read_unlock_irq(&page->mapping->tree_lock);
2865                 unlock_page(page);
2866         }
2867         return 0;
2868 }
2869 EXPORT_SYMBOL(clear_extent_buffer_dirty);
2870
2871 int wait_on_extent_buffer_writeback(struct extent_io_tree *tree,
2872                                     struct extent_buffer *eb)
2873 {
2874         return wait_on_extent_writeback(tree, eb->start,
2875                                         eb->start + eb->len - 1);
2876 }
2877 EXPORT_SYMBOL(wait_on_extent_buffer_writeback);
2878
2879 int set_extent_buffer_dirty(struct extent_io_tree *tree,
2880                              struct extent_buffer *eb)
2881 {
2882         unsigned long i;
2883         unsigned long num_pages;
2884
2885         num_pages = num_extent_pages(eb->start, eb->len);
2886         for (i = 0; i < num_pages; i++) {
2887                 struct page *page = extent_buffer_page(eb, i);
2888                 /* writepage may need to do something special for the
2889                  * first page, we have to make sure page->private is
2890                  * properly set.  releasepage may drop page->private
2891                  * on us if the page isn't already dirty.
2892                  */
2893                 if (i == 0) {
2894                         lock_page(page);
2895                         set_page_extent_head(page, eb->len);
2896                 } else if (PagePrivate(page) &&
2897                            page->private != EXTENT_PAGE_PRIVATE) {
2898                         lock_page(page);
2899                         set_page_extent_mapped(page);
2900                         unlock_page(page);
2901                 }
2902                 __set_page_dirty_nobuffers(extent_buffer_page(eb, i));
2903                 if (i == 0)
2904                         unlock_page(page);
2905         }
2906         return set_extent_dirty(tree, eb->start,
2907                                 eb->start + eb->len - 1, GFP_NOFS);
2908 }
2909 EXPORT_SYMBOL(set_extent_buffer_dirty);
2910
2911 int set_extent_buffer_uptodate(struct extent_io_tree *tree,
2912                                 struct extent_buffer *eb)
2913 {
2914         unsigned long i;
2915         struct page *page;
2916         unsigned long num_pages;
2917
2918         num_pages = num_extent_pages(eb->start, eb->len);
2919
2920         set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
2921                             GFP_NOFS);
2922         for (i = 0; i < num_pages; i++) {
2923                 page = extent_buffer_page(eb, i);
2924                 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
2925                     ((i == num_pages - 1) &&
2926                      ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
2927                         check_page_uptodate(tree, page);
2928                         continue;
2929                 }
2930                 SetPageUptodate(page);
2931         }
2932         return 0;
2933 }
2934 EXPORT_SYMBOL(set_extent_buffer_uptodate);
2935
2936 int extent_buffer_uptodate(struct extent_io_tree *tree,
2937                              struct extent_buffer *eb)
2938 {
2939         int ret = 0;
2940         int ret2;
2941         int num_pages;
2942         int i;
2943         struct page *page;
2944         int pg_uptodate = 1;
2945
2946         if (eb->flags & EXTENT_UPTODATE)
2947                 ret = 1;
2948
2949         ret2  = test_range_bit(tree, eb->start, eb->start + eb->len - 1,
2950                            EXTENT_UPTODATE, 1);
2951
2952         num_pages = num_extent_pages(eb->start, eb->len);
2953         for (i = 0; i < num_pages; i++) {
2954                 page = extent_buffer_page(eb, i);
2955                 if (!PageUptodate(page)) {
2956                         pg_uptodate = 0;
2957                         break;
2958                 }
2959         }
2960         if ((ret || ret2) && !pg_uptodate) {
2961 printk("uptodate error2 eb %Lu ret %d ret2 %d pg_uptodate %d\n", eb->start, ret, ret2, pg_uptodate);
2962                 WARN_ON(1);
2963         }
2964         return (ret || ret2);
2965 }
2966 EXPORT_SYMBOL(extent_buffer_uptodate);
2967
2968 int read_extent_buffer_pages(struct extent_io_tree *tree,
2969                              struct extent_buffer *eb,
2970                              u64 start, int wait,
2971                              get_extent_t *get_extent)
2972 {
2973         unsigned long i;
2974         unsigned long start_i;
2975         struct page *page;
2976         int err;
2977         int ret = 0;
2978         unsigned long num_pages;
2979         struct bio *bio = NULL;
2980
2981         if (eb->flags & EXTENT_UPTODATE)
2982                 return 0;
2983
2984         if (0 && test_range_bit(tree, eb->start, eb->start + eb->len - 1,
2985                            EXTENT_UPTODATE, 1)) {
2986                 return 0;
2987         }
2988
2989         if (start) {
2990                 WARN_ON(start < eb->start);
2991                 start_i = (start >> PAGE_CACHE_SHIFT) -
2992                         (eb->start >> PAGE_CACHE_SHIFT);
2993         } else {
2994                 start_i = 0;
2995         }
2996
2997         num_pages = num_extent_pages(eb->start, eb->len);
2998         for (i = start_i; i < num_pages; i++) {
2999                 page = extent_buffer_page(eb, i);
3000                 if (PageUptodate(page)) {
3001                         continue;
3002                 }
3003                 if (!wait) {
3004                         if (TestSetPageLocked(page)) {
3005                                 continue;
3006                         }
3007                 } else {
3008                         lock_page(page);
3009                 }
3010                 if (!PageUptodate(page)) {
3011                         err = __extent_read_full_page(tree, page,
3012                                                       get_extent, &bio);
3013                         if (err) {
3014                                 ret = err;
3015                         }
3016                 } else {
3017                         unlock_page(page);
3018                 }
3019         }
3020
3021         if (bio)
3022                 submit_one_bio(READ, bio);
3023
3024         if (ret || !wait) {
3025                 return ret;
3026         }
3027         for (i = start_i; i < num_pages; i++) {
3028                 page = extent_buffer_page(eb, i);
3029                 wait_on_page_locked(page);
3030                 if (!PageUptodate(page)) {
3031                         ret = -EIO;
3032                 }
3033         }
3034         if (!ret)
3035                 eb->flags |= EXTENT_UPTODATE;
3036         return ret;
3037 }
3038 EXPORT_SYMBOL(read_extent_buffer_pages);
3039
3040 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
3041                         unsigned long start,
3042                         unsigned long len)
3043 {
3044         size_t cur;
3045         size_t offset;
3046         struct page *page;
3047         char *kaddr;
3048         char *dst = (char *)dstv;
3049         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3050         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3051         unsigned long num_pages = num_extent_pages(eb->start, eb->len);
3052
3053         WARN_ON(start > eb->len);
3054         WARN_ON(start + len > eb->start + eb->len);
3055
3056         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3057
3058         while(len > 0) {
3059                 page = extent_buffer_page(eb, i);
3060                 if (!PageUptodate(page)) {
3061                         printk("page %lu not up to date i %lu, total %lu, len %lu\n", page->index, i, num_pages, eb->len);
3062                         WARN_ON(1);
3063                 }
3064                 WARN_ON(!PageUptodate(page));
3065
3066                 cur = min(len, (PAGE_CACHE_SIZE - offset));
3067                 kaddr = kmap_atomic(page, KM_USER1);
3068                 memcpy(dst, kaddr + offset, cur);
3069                 kunmap_atomic(kaddr, KM_USER1);
3070
3071                 dst += cur;
3072                 len -= cur;
3073                 offset = 0;
3074                 i++;
3075         }
3076 }
3077 EXPORT_SYMBOL(read_extent_buffer);
3078
3079 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
3080                                unsigned long min_len, char **token, char **map,
3081                                unsigned long *map_start,
3082                                unsigned long *map_len, int km)
3083 {
3084         size_t offset = start & (PAGE_CACHE_SIZE - 1);
3085         char *kaddr;
3086         struct page *p;
3087         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3088         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3089         unsigned long end_i = (start_offset + start + min_len - 1) >>
3090                 PAGE_CACHE_SHIFT;
3091
3092         if (i != end_i)
3093                 return -EINVAL;
3094
3095         if (i == 0) {
3096                 offset = start_offset;
3097                 *map_start = 0;
3098         } else {
3099                 offset = 0;
3100                 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
3101         }
3102         if (start + min_len > eb->len) {
3103 printk("bad mapping eb start %Lu len %lu, wanted %lu %lu\n", eb->start, eb->len, start, min_len);
3104                 WARN_ON(1);
3105         }
3106
3107         p = extent_buffer_page(eb, i);
3108         WARN_ON(!PageUptodate(p));
3109         kaddr = kmap_atomic(p, km);
3110         *token = kaddr;
3111         *map = kaddr + offset;
3112         *map_len = PAGE_CACHE_SIZE - offset;
3113         return 0;
3114 }
3115 EXPORT_SYMBOL(map_private_extent_buffer);
3116
3117 int map_extent_buffer(struct extent_buffer *eb, unsigned long start,
3118                       unsigned long min_len,
3119                       char **token, char **map,
3120                       unsigned long *map_start,
3121                       unsigned long *map_len, int km)
3122 {
3123         int err;
3124         int save = 0;
3125         if (eb->map_token) {
3126                 unmap_extent_buffer(eb, eb->map_token, km);
3127                 eb->map_token = NULL;
3128                 save = 1;
3129         }
3130         err = map_private_extent_buffer(eb, start, min_len, token, map,
3131                                        map_start, map_len, km);
3132         if (!err && save) {
3133                 eb->map_token = *token;
3134                 eb->kaddr = *map;
3135                 eb->map_start = *map_start;
3136                 eb->map_len = *map_len;
3137         }
3138         return err;
3139 }
3140 EXPORT_SYMBOL(map_extent_buffer);
3141
3142 void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km)
3143 {
3144         kunmap_atomic(token, km);
3145 }
3146 EXPORT_SYMBOL(unmap_extent_buffer);
3147
3148 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
3149                           unsigned long start,
3150                           unsigned long len)
3151 {
3152         size_t cur;
3153         size_t offset;
3154         struct page *page;
3155         char *kaddr;
3156         char *ptr = (char *)ptrv;
3157         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3158         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3159         int ret = 0;
3160
3161         WARN_ON(start > eb->len);
3162         WARN_ON(start + len > eb->start + eb->len);
3163
3164         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3165
3166         while(len > 0) {
3167                 page = extent_buffer_page(eb, i);
3168                 WARN_ON(!PageUptodate(page));
3169
3170                 cur = min(len, (PAGE_CACHE_SIZE - offset));
3171
3172                 kaddr = kmap_atomic(page, KM_USER0);
3173                 ret = memcmp(ptr, kaddr + offset, cur);
3174                 kunmap_atomic(kaddr, KM_USER0);
3175                 if (ret)
3176                         break;
3177
3178                 ptr += cur;
3179                 len -= cur;
3180                 offset = 0;
3181                 i++;
3182         }
3183         return ret;
3184 }
3185 EXPORT_SYMBOL(memcmp_extent_buffer);
3186
3187 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
3188                          unsigned long start, unsigned long len)
3189 {
3190         size_t cur;
3191         size_t offset;
3192         struct page *page;
3193         char *kaddr;
3194         char *src = (char *)srcv;
3195         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3196         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3197
3198         WARN_ON(start > eb->len);
3199         WARN_ON(start + len > eb->start + eb->len);
3200
3201         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3202
3203         while(len > 0) {
3204                 page = extent_buffer_page(eb, i);
3205                 WARN_ON(!PageUptodate(page));
3206
3207                 cur = min(len, PAGE_CACHE_SIZE - offset);
3208                 kaddr = kmap_atomic(page, KM_USER1);
3209                 memcpy(kaddr + offset, src, cur);
3210                 kunmap_atomic(kaddr, KM_USER1);
3211
3212                 src += cur;
3213                 len -= cur;
3214                 offset = 0;
3215                 i++;
3216         }
3217 }
3218 EXPORT_SYMBOL(write_extent_buffer);
3219
3220 void memset_extent_buffer(struct extent_buffer *eb, char c,
3221                           unsigned long start, unsigned long len)
3222 {
3223         size_t cur;
3224         size_t offset;
3225         struct page *page;
3226         char *kaddr;
3227         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3228         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
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                 WARN_ON(!PageUptodate(page));
3238
3239                 cur = min(len, PAGE_CACHE_SIZE - offset);
3240                 kaddr = kmap_atomic(page, KM_USER0);
3241                 memset(kaddr + offset, c, cur);
3242                 kunmap_atomic(kaddr, KM_USER0);
3243
3244                 len -= cur;
3245                 offset = 0;
3246                 i++;
3247         }
3248 }
3249 EXPORT_SYMBOL(memset_extent_buffer);
3250
3251 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
3252                         unsigned long dst_offset, unsigned long src_offset,
3253                         unsigned long len)
3254 {
3255         u64 dst_len = dst->len;
3256         size_t cur;
3257         size_t offset;
3258         struct page *page;
3259         char *kaddr;
3260         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3261         unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3262
3263         WARN_ON(src->len != dst_len);
3264
3265         offset = (start_offset + dst_offset) &
3266                 ((unsigned long)PAGE_CACHE_SIZE - 1);
3267
3268         while(len > 0) {
3269                 page = extent_buffer_page(dst, i);
3270                 WARN_ON(!PageUptodate(page));
3271
3272                 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
3273
3274                 kaddr = kmap_atomic(page, KM_USER0);
3275                 read_extent_buffer(src, kaddr + offset, src_offset, cur);
3276                 kunmap_atomic(kaddr, KM_USER0);
3277
3278                 src_offset += cur;
3279                 len -= cur;
3280                 offset = 0;
3281                 i++;
3282         }
3283 }
3284 EXPORT_SYMBOL(copy_extent_buffer);
3285
3286 static void move_pages(struct page *dst_page, struct page *src_page,
3287                        unsigned long dst_off, unsigned long src_off,
3288                        unsigned long len)
3289 {
3290         char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3291         if (dst_page == src_page) {
3292                 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
3293         } else {
3294                 char *src_kaddr = kmap_atomic(src_page, KM_USER1);
3295                 char *p = dst_kaddr + dst_off + len;
3296                 char *s = src_kaddr + src_off + len;
3297
3298                 while (len--)
3299                         *--p = *--s;
3300
3301                 kunmap_atomic(src_kaddr, KM_USER1);
3302         }
3303         kunmap_atomic(dst_kaddr, KM_USER0);
3304 }
3305
3306 static void copy_pages(struct page *dst_page, struct page *src_page,
3307                        unsigned long dst_off, unsigned long src_off,
3308                        unsigned long len)
3309 {
3310         char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3311         char *src_kaddr;
3312
3313         if (dst_page != src_page)
3314                 src_kaddr = kmap_atomic(src_page, KM_USER1);
3315         else
3316                 src_kaddr = dst_kaddr;
3317
3318         memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
3319         kunmap_atomic(dst_kaddr, KM_USER0);
3320         if (dst_page != src_page)
3321                 kunmap_atomic(src_kaddr, KM_USER1);
3322 }
3323
3324 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3325                            unsigned long src_offset, unsigned long len)
3326 {
3327         size_t cur;
3328         size_t dst_off_in_page;
3329         size_t src_off_in_page;
3330         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3331         unsigned long dst_i;
3332         unsigned long src_i;
3333
3334         if (src_offset + len > dst->len) {
3335                 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
3336                        src_offset, len, dst->len);
3337                 BUG_ON(1);
3338         }
3339         if (dst_offset + len > dst->len) {
3340                 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
3341                        dst_offset, len, dst->len);
3342                 BUG_ON(1);
3343         }
3344
3345         while(len > 0) {
3346                 dst_off_in_page = (start_offset + dst_offset) &
3347                         ((unsigned long)PAGE_CACHE_SIZE - 1);
3348                 src_off_in_page = (start_offset + src_offset) &
3349                         ((unsigned long)PAGE_CACHE_SIZE - 1);
3350
3351                 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3352                 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
3353
3354                 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
3355                                                src_off_in_page));
3356                 cur = min_t(unsigned long, cur,
3357                         (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
3358
3359                 copy_pages(extent_buffer_page(dst, dst_i),
3360                            extent_buffer_page(dst, src_i),
3361                            dst_off_in_page, src_off_in_page, cur);
3362
3363                 src_offset += cur;
3364                 dst_offset += cur;
3365                 len -= cur;
3366         }
3367 }
3368 EXPORT_SYMBOL(memcpy_extent_buffer);
3369
3370 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3371                            unsigned long src_offset, unsigned long len)
3372 {
3373         size_t cur;
3374         size_t dst_off_in_page;
3375         size_t src_off_in_page;
3376         unsigned long dst_end = dst_offset + len - 1;
3377         unsigned long src_end = src_offset + len - 1;
3378         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3379         unsigned long dst_i;
3380         unsigned long src_i;
3381
3382         if (src_offset + len > dst->len) {
3383                 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
3384                        src_offset, len, dst->len);
3385                 BUG_ON(1);
3386         }
3387         if (dst_offset + len > dst->len) {
3388                 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
3389                        dst_offset, len, dst->len);
3390                 BUG_ON(1);
3391         }
3392         if (dst_offset < src_offset) {
3393                 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
3394                 return;
3395         }
3396         while(len > 0) {
3397                 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
3398                 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
3399
3400                 dst_off_in_page = (start_offset + dst_end) &
3401                         ((unsigned long)PAGE_CACHE_SIZE - 1);
3402                 src_off_in_page = (start_offset + src_end) &
3403                         ((unsigned long)PAGE_CACHE_SIZE - 1);
3404
3405                 cur = min_t(unsigned long, len, src_off_in_page + 1);
3406                 cur = min(cur, dst_off_in_page + 1);
3407                 move_pages(extent_buffer_page(dst, dst_i),
3408                            extent_buffer_page(dst, src_i),
3409                            dst_off_in_page - cur + 1,
3410                            src_off_in_page - cur + 1, cur);
3411
3412                 dst_end -= cur;
3413                 src_end -= cur;
3414                 len -= cur;
3415         }
3416 }
3417 EXPORT_SYMBOL(memmove_extent_buffer);