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                 if (!found)
1029                         *end = (u64)-1;
1030                 goto out;
1031         }
1032
1033         while(1) {
1034                 state = rb_entry(node, struct extent_state, rb_node);
1035                 if (found && state->start != cur_start) {
1036                         goto out;
1037                 }
1038                 if (!(state->state & EXTENT_DELALLOC)) {
1039                         if (!found)
1040                                 *end = state->end;
1041                         goto out;
1042                 }
1043                 if (!found) {
1044                         struct extent_state *prev_state;
1045                         struct rb_node *prev_node = node;
1046                         while(1) {
1047                                 prev_node = rb_prev(prev_node);
1048                                 if (!prev_node)
1049                                         break;
1050                                 prev_state = rb_entry(prev_node,
1051                                                       struct extent_state,
1052                                                       rb_node);
1053                                 if (!(prev_state->state & EXTENT_DELALLOC))
1054                                         break;
1055                                 state = prev_state;
1056                                 node = prev_node;
1057                         }
1058                 }
1059                 if (state->state & EXTENT_LOCKED) {
1060                         DEFINE_WAIT(wait);
1061                         atomic_inc(&state->refs);
1062                         prepare_to_wait(&state->wq, &wait,
1063                                         TASK_UNINTERRUPTIBLE);
1064                         spin_unlock_irq(&tree->lock);
1065                         schedule();
1066                         spin_lock_irq(&tree->lock);
1067                         finish_wait(&state->wq, &wait);
1068                         free_extent_state(state);
1069                         goto search_again;
1070                 }
1071                 set_state_cb(tree, state, EXTENT_LOCKED);
1072                 state->state |= EXTENT_LOCKED;
1073                 if (!found)
1074                         *start = state->start;
1075                 found++;
1076                 *end = state->end;
1077                 cur_start = state->end + 1;
1078                 node = rb_next(node);
1079                 if (!node)
1080                         break;
1081                 total_bytes += state->end - state->start + 1;
1082                 if (total_bytes >= max_bytes)
1083                         break;
1084         }
1085 out:
1086         spin_unlock_irq(&tree->lock);
1087         return found;
1088 }
1089
1090 u64 count_range_bits(struct extent_io_tree *tree,
1091                      u64 *start, u64 search_end, u64 max_bytes,
1092                      unsigned long bits)
1093 {
1094         struct rb_node *node;
1095         struct extent_state *state;
1096         u64 cur_start = *start;
1097         u64 total_bytes = 0;
1098         int found = 0;
1099
1100         if (search_end <= cur_start) {
1101                 printk("search_end %Lu start %Lu\n", search_end, cur_start);
1102                 WARN_ON(1);
1103                 return 0;
1104         }
1105
1106         spin_lock_irq(&tree->lock);
1107         if (cur_start == 0 && bits == EXTENT_DIRTY) {
1108                 total_bytes = tree->dirty_bytes;
1109                 goto out;
1110         }
1111         /*
1112          * this search will find all the extents that end after
1113          * our range starts.
1114          */
1115         node = tree_search(tree, cur_start);
1116         if (!node) {
1117                 goto out;
1118         }
1119
1120         while(1) {
1121                 state = rb_entry(node, struct extent_state, rb_node);
1122                 if (state->start > search_end)
1123                         break;
1124                 if (state->end >= cur_start && (state->state & bits)) {
1125                         total_bytes += min(search_end, state->end) + 1 -
1126                                        max(cur_start, state->start);
1127                         if (total_bytes >= max_bytes)
1128                                 break;
1129                         if (!found) {
1130                                 *start = state->start;
1131                                 found = 1;
1132                         }
1133                 }
1134                 node = rb_next(node);
1135                 if (!node)
1136                         break;
1137         }
1138 out:
1139         spin_unlock_irq(&tree->lock);
1140         return total_bytes;
1141 }
1142 /*
1143  * helper function to lock both pages and extents in the tree.
1144  * pages must be locked first.
1145  */
1146 int lock_range(struct extent_io_tree *tree, u64 start, u64 end)
1147 {
1148         unsigned long index = start >> PAGE_CACHE_SHIFT;
1149         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1150         struct page *page;
1151         int err;
1152
1153         while (index <= end_index) {
1154                 page = grab_cache_page(tree->mapping, index);
1155                 if (!page) {
1156                         err = -ENOMEM;
1157                         goto failed;
1158                 }
1159                 if (IS_ERR(page)) {
1160                         err = PTR_ERR(page);
1161                         goto failed;
1162                 }
1163                 index++;
1164         }
1165         lock_extent(tree, start, end, GFP_NOFS);
1166         return 0;
1167
1168 failed:
1169         /*
1170          * we failed above in getting the page at 'index', so we undo here
1171          * up to but not including the page at 'index'
1172          */
1173         end_index = index;
1174         index = start >> PAGE_CACHE_SHIFT;
1175         while (index < end_index) {
1176                 page = find_get_page(tree->mapping, index);
1177                 unlock_page(page);
1178                 page_cache_release(page);
1179                 index++;
1180         }
1181         return err;
1182 }
1183 EXPORT_SYMBOL(lock_range);
1184
1185 /*
1186  * helper function to unlock both pages and extents in the tree.
1187  */
1188 int unlock_range(struct extent_io_tree *tree, u64 start, u64 end)
1189 {
1190         unsigned long index = start >> PAGE_CACHE_SHIFT;
1191         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1192         struct page *page;
1193
1194         while (index <= end_index) {
1195                 page = find_get_page(tree->mapping, index);
1196                 unlock_page(page);
1197                 page_cache_release(page);
1198                 index++;
1199         }
1200         unlock_extent(tree, start, end, GFP_NOFS);
1201         return 0;
1202 }
1203 EXPORT_SYMBOL(unlock_range);
1204
1205 int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1206 {
1207         struct rb_node *node;
1208         struct extent_state *state;
1209         int ret = 0;
1210
1211         spin_lock_irq(&tree->lock);
1212         /*
1213          * this search will find all the extents that end after
1214          * our range starts.
1215          */
1216         node = tree_search(tree, start);
1217         if (!node) {
1218                 ret = -ENOENT;
1219                 goto out;
1220         }
1221         state = rb_entry(node, struct extent_state, rb_node);
1222         if (state->start != start) {
1223                 ret = -ENOENT;
1224                 goto out;
1225         }
1226         state->private = private;
1227 out:
1228         spin_unlock_irq(&tree->lock);
1229         return ret;
1230 }
1231
1232 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1233 {
1234         struct rb_node *node;
1235         struct extent_state *state;
1236         int ret = 0;
1237
1238         spin_lock_irq(&tree->lock);
1239         /*
1240          * this search will find all the extents that end after
1241          * our range starts.
1242          */
1243         node = tree_search(tree, start);
1244         if (!node) {
1245                 ret = -ENOENT;
1246                 goto out;
1247         }
1248         state = rb_entry(node, struct extent_state, rb_node);
1249         if (state->start != start) {
1250                 ret = -ENOENT;
1251                 goto out;
1252         }
1253         *private = state->private;
1254 out:
1255         spin_unlock_irq(&tree->lock);
1256         return ret;
1257 }
1258
1259 /*
1260  * searches a range in the state tree for a given mask.
1261  * If 'filled' == 1, this returns 1 only if every extent in the tree
1262  * has the bits set.  Otherwise, 1 is returned if any bit in the
1263  * range is found set.
1264  */
1265 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1266                    int bits, int filled)
1267 {
1268         struct extent_state *state = NULL;
1269         struct rb_node *node;
1270         int bitset = 0;
1271         unsigned long flags;
1272
1273         spin_lock_irqsave(&tree->lock, flags);
1274         node = tree_search(tree, start);
1275         while (node && start <= end) {
1276                 state = rb_entry(node, struct extent_state, rb_node);
1277
1278                 if (filled && state->start > start) {
1279                         bitset = 0;
1280                         break;
1281                 }
1282
1283                 if (state->start > end)
1284                         break;
1285
1286                 if (state->state & bits) {
1287                         bitset = 1;
1288                         if (!filled)
1289                                 break;
1290                 } else if (filled) {
1291                         bitset = 0;
1292                         break;
1293                 }
1294                 start = state->end + 1;
1295                 if (start > end)
1296                         break;
1297                 node = rb_next(node);
1298                 if (!node) {
1299                         if (filled)
1300                                 bitset = 0;
1301                         break;
1302                 }
1303         }
1304         spin_unlock_irqrestore(&tree->lock, flags);
1305         return bitset;
1306 }
1307 EXPORT_SYMBOL(test_range_bit);
1308
1309 /*
1310  * helper function to set a given page up to date if all the
1311  * extents in the tree for that page are up to date
1312  */
1313 static int check_page_uptodate(struct extent_io_tree *tree,
1314                                struct page *page)
1315 {
1316         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1317         u64 end = start + PAGE_CACHE_SIZE - 1;
1318         if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1))
1319                 SetPageUptodate(page);
1320         return 0;
1321 }
1322
1323 /*
1324  * helper function to unlock a page if all the extents in the tree
1325  * for that page are unlocked
1326  */
1327 static int check_page_locked(struct extent_io_tree *tree,
1328                              struct page *page)
1329 {
1330         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1331         u64 end = start + PAGE_CACHE_SIZE - 1;
1332         if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0))
1333                 unlock_page(page);
1334         return 0;
1335 }
1336
1337 /*
1338  * helper function to end page writeback if all the extents
1339  * in the tree for that page are done with writeback
1340  */
1341 static int check_page_writeback(struct extent_io_tree *tree,
1342                              struct page *page)
1343 {
1344         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1345         u64 end = start + PAGE_CACHE_SIZE - 1;
1346         if (!test_range_bit(tree, start, end, EXTENT_WRITEBACK, 0))
1347                 end_page_writeback(page);
1348         return 0;
1349 }
1350
1351 /* lots and lots of room for performance fixes in the end_bio funcs */
1352
1353 /*
1354  * after a writepage IO is done, we need to:
1355  * clear the uptodate bits on error
1356  * clear the writeback bits in the extent tree for this IO
1357  * end_page_writeback if the page has no more pending IO
1358  *
1359  * Scheduling is not allowed, so the extent state tree is expected
1360  * to have one and only one object corresponding to this IO.
1361  */
1362 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1363 static void end_bio_extent_writepage(struct bio *bio, int err)
1364 #else
1365 static int end_bio_extent_writepage(struct bio *bio,
1366                                    unsigned int bytes_done, int err)
1367 #endif
1368 {
1369         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1370         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1371         struct extent_state *state = bio->bi_private;
1372         struct extent_io_tree *tree = state->tree;
1373         struct rb_node *node;
1374         u64 start;
1375         u64 end;
1376         u64 cur;
1377         int whole_page;
1378         unsigned long flags;
1379
1380 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1381         if (bio->bi_size)
1382                 return 1;
1383 #endif
1384         do {
1385                 struct page *page = bvec->bv_page;
1386                 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1387                          bvec->bv_offset;
1388                 end = start + bvec->bv_len - 1;
1389
1390                 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1391                         whole_page = 1;
1392                 else
1393                         whole_page = 0;
1394
1395                 if (--bvec >= bio->bi_io_vec)
1396                         prefetchw(&bvec->bv_page->flags);
1397
1398                 if (!uptodate) {
1399                         clear_extent_uptodate(tree, start, end, GFP_ATOMIC);
1400                         ClearPageUptodate(page);
1401                         SetPageError(page);
1402                 }
1403
1404                 if (tree->ops && tree->ops->writepage_end_io_hook) {
1405                         tree->ops->writepage_end_io_hook(page, start, end,
1406                                                          state);
1407                 }
1408
1409                 /*
1410                  * bios can get merged in funny ways, and so we need to
1411                  * be careful with the state variable.  We know the
1412                  * state won't be merged with others because it has
1413                  * WRITEBACK set, but we can't be sure each biovec is
1414                  * sequential in the file.  So, if our cached state
1415                  * doesn't match the expected end, search the tree
1416                  * for the correct one.
1417                  */
1418
1419                 spin_lock_irqsave(&tree->lock, flags);
1420                 if (!state || state->end != end) {
1421                         state = NULL;
1422                         node = __etree_search(tree, start, NULL, NULL);
1423                         if (node) {
1424                                 state = rb_entry(node, struct extent_state,
1425                                                  rb_node);
1426                                 if (state->end != end ||
1427                                     !(state->state & EXTENT_WRITEBACK))
1428                                         state = NULL;
1429                         }
1430                         if (!state) {
1431                                 spin_unlock_irqrestore(&tree->lock, flags);
1432                                 clear_extent_writeback(tree, start,
1433                                                        end, GFP_ATOMIC);
1434                                 goto next_io;
1435                         }
1436                 }
1437                 cur = end;
1438                 while(1) {
1439                         struct extent_state *clear = state;
1440                         cur = state->start;
1441                         node = rb_prev(&state->rb_node);
1442                         if (node) {
1443                                 state = rb_entry(node,
1444                                                  struct extent_state,
1445                                                  rb_node);
1446                         } else {
1447                                 state = NULL;
1448                         }
1449
1450                         clear_state_bit(tree, clear, EXTENT_WRITEBACK,
1451                                         1, 0);
1452                         if (cur == start)
1453                                 break;
1454                         if (cur < start) {
1455                                 WARN_ON(1);
1456                                 break;
1457                         }
1458                         if (!node)
1459                                 break;
1460                 }
1461                 /* before releasing the lock, make sure the next state
1462                  * variable has the expected bits set and corresponds
1463                  * to the correct offsets in the file
1464                  */
1465                 if (state && (state->end + 1 != start ||
1466                     !(state->state & EXTENT_WRITEBACK))) {
1467                         state = NULL;
1468                 }
1469                 spin_unlock_irqrestore(&tree->lock, flags);
1470 next_io:
1471
1472                 if (whole_page)
1473                         end_page_writeback(page);
1474                 else
1475                         check_page_writeback(tree, page);
1476         } while (bvec >= bio->bi_io_vec);
1477         bio_put(bio);
1478 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1479         return 0;
1480 #endif
1481 }
1482
1483 /*
1484  * after a readpage IO is done, we need to:
1485  * clear the uptodate bits on error
1486  * set the uptodate bits if things worked
1487  * set the page up to date if all extents in the tree are uptodate
1488  * clear the lock bit in the extent tree
1489  * unlock the page if there are no other extents locked for it
1490  *
1491  * Scheduling is not allowed, so the extent state tree is expected
1492  * to have one and only one object corresponding to this IO.
1493  */
1494 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1495 static void end_bio_extent_readpage(struct bio *bio, int err)
1496 #else
1497 static int end_bio_extent_readpage(struct bio *bio,
1498                                    unsigned int bytes_done, int err)
1499 #endif
1500 {
1501         int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1502         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1503         struct extent_state *state = bio->bi_private;
1504         struct extent_io_tree *tree = state->tree;
1505         struct rb_node *node;
1506         u64 start;
1507         u64 end;
1508         u64 cur;
1509         unsigned long flags;
1510         int whole_page;
1511         int ret;
1512
1513 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1514         if (bio->bi_size)
1515                 return 1;
1516 #endif
1517
1518         do {
1519                 struct page *page = bvec->bv_page;
1520                 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1521                         bvec->bv_offset;
1522                 end = start + bvec->bv_len - 1;
1523
1524                 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1525                         whole_page = 1;
1526                 else
1527                         whole_page = 0;
1528
1529                 if (--bvec >= bio->bi_io_vec)
1530                         prefetchw(&bvec->bv_page->flags);
1531
1532                 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
1533                         ret = tree->ops->readpage_end_io_hook(page, start, end,
1534                                                               state);
1535                         if (ret)
1536                                 uptodate = 0;
1537                 }
1538                 if (!uptodate && tree->ops &&
1539                     tree->ops->readpage_io_failed_hook) {
1540                         ret = tree->ops->readpage_io_failed_hook(bio, page,
1541                                                          start, end, state);
1542                         if (ret == 0) {
1543                                 state = NULL;
1544                                 uptodate =
1545                                         test_bit(BIO_UPTODATE, &bio->bi_flags);
1546                                 continue;
1547                         }
1548                 }
1549
1550                 spin_lock_irqsave(&tree->lock, flags);
1551                 if (!state || state->end != end) {
1552                         state = NULL;
1553                         node = __etree_search(tree, start, NULL, NULL);
1554                         if (node) {
1555                                 state = rb_entry(node, struct extent_state,
1556                                                  rb_node);
1557                                 if (state->end != end ||
1558                                     !(state->state & EXTENT_LOCKED))
1559                                         state = NULL;
1560                         }
1561                         if (!state) {
1562                                 spin_unlock_irqrestore(&tree->lock, flags);
1563                                 if (uptodate)
1564                                         set_extent_uptodate(tree, start, end,
1565                                                             GFP_ATOMIC);
1566                                 unlock_extent(tree, start, end, GFP_ATOMIC);
1567                                 goto next_io;
1568                         }
1569                 }
1570
1571                 cur = end;
1572                 while(1) {
1573                         struct extent_state *clear = state;
1574                         cur = state->start;
1575                         node = rb_prev(&state->rb_node);
1576                         if (node) {
1577                                 state = rb_entry(node,
1578                                          struct extent_state,
1579                                          rb_node);
1580                         } else {
1581                                 state = NULL;
1582                         }
1583                         if (uptodate) {
1584                                 set_state_cb(tree, clear, EXTENT_UPTODATE);
1585                                 clear->state |= EXTENT_UPTODATE;
1586                         }
1587                         clear_state_bit(tree, clear, EXTENT_LOCKED,
1588                                         1, 0);
1589                         if (cur == start)
1590                                 break;
1591                         if (cur < start) {
1592                                 WARN_ON(1);
1593                                 break;
1594                         }
1595                         if (!node)
1596                                 break;
1597                 }
1598                 /* before releasing the lock, make sure the next state
1599                  * variable has the expected bits set and corresponds
1600                  * to the correct offsets in the file
1601                  */
1602                 if (state && (state->end + 1 != start ||
1603                     !(state->state & EXTENT_LOCKED))) {
1604                         state = NULL;
1605                 }
1606                 spin_unlock_irqrestore(&tree->lock, flags);
1607 next_io:
1608                 if (whole_page) {
1609                         if (uptodate) {
1610                                 SetPageUptodate(page);
1611                         } else {
1612                                 ClearPageUptodate(page);
1613                                 SetPageError(page);
1614                         }
1615                         unlock_page(page);
1616                 } else {
1617                         if (uptodate) {
1618                                 check_page_uptodate(tree, page);
1619                         } else {
1620                                 ClearPageUptodate(page);
1621                                 SetPageError(page);
1622                         }
1623                         check_page_locked(tree, page);
1624                 }
1625         } while (bvec >= bio->bi_io_vec);
1626
1627         bio_put(bio);
1628 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1629         return 0;
1630 #endif
1631 }
1632
1633 /*
1634  * IO done from prepare_write is pretty simple, we just unlock
1635  * the structs in the extent tree when done, and set the uptodate bits
1636  * as appropriate.
1637  */
1638 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1639 static void end_bio_extent_preparewrite(struct bio *bio, int err)
1640 #else
1641 static int end_bio_extent_preparewrite(struct bio *bio,
1642                                        unsigned int bytes_done, int err)
1643 #endif
1644 {
1645         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1646         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1647         struct extent_state *state = bio->bi_private;
1648         struct extent_io_tree *tree = state->tree;
1649         u64 start;
1650         u64 end;
1651
1652 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1653         if (bio->bi_size)
1654                 return 1;
1655 #endif
1656
1657         do {
1658                 struct page *page = bvec->bv_page;
1659                 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1660                         bvec->bv_offset;
1661                 end = start + bvec->bv_len - 1;
1662
1663                 if (--bvec >= bio->bi_io_vec)
1664                         prefetchw(&bvec->bv_page->flags);
1665
1666                 if (uptodate) {
1667                         set_extent_uptodate(tree, start, end, GFP_ATOMIC);
1668                 } else {
1669                         ClearPageUptodate(page);
1670                         SetPageError(page);
1671                 }
1672
1673                 unlock_extent(tree, start, end, GFP_ATOMIC);
1674
1675         } while (bvec >= bio->bi_io_vec);
1676
1677         bio_put(bio);
1678 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1679         return 0;
1680 #endif
1681 }
1682
1683 static struct bio *
1684 extent_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
1685                  gfp_t gfp_flags)
1686 {
1687         struct bio *bio;
1688
1689         bio = bio_alloc(gfp_flags, nr_vecs);
1690
1691         if (bio == NULL && (current->flags & PF_MEMALLOC)) {
1692                 while (!bio && (nr_vecs /= 2))
1693                         bio = bio_alloc(gfp_flags, nr_vecs);
1694         }
1695
1696         if (bio) {
1697                 bio->bi_bdev = bdev;
1698                 bio->bi_sector = first_sector;
1699         }
1700         return bio;
1701 }
1702
1703 static int submit_one_bio(int rw, struct bio *bio, int mirror_num)
1704 {
1705         int ret = 0;
1706         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1707         struct page *page = bvec->bv_page;
1708         struct extent_io_tree *tree = bio->bi_private;
1709         struct rb_node *node;
1710         struct extent_state *state;
1711         u64 start;
1712         u64 end;
1713
1714         start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
1715         end = start + bvec->bv_len - 1;
1716
1717         spin_lock_irq(&tree->lock);
1718         node = __etree_search(tree, start, NULL, NULL);
1719         BUG_ON(!node);
1720         state = rb_entry(node, struct extent_state, rb_node);
1721         while(state->end < end) {
1722                 node = rb_next(node);
1723                 state = rb_entry(node, struct extent_state, rb_node);
1724         }
1725         BUG_ON(state->end != end);
1726         spin_unlock_irq(&tree->lock);
1727
1728         bio->bi_private = state;
1729
1730         bio_get(bio);
1731
1732         if (tree->ops && tree->ops->submit_bio_hook)
1733                 tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
1734                                            mirror_num);
1735         else
1736                 submit_bio(rw, bio);
1737         if (bio_flagged(bio, BIO_EOPNOTSUPP))
1738                 ret = -EOPNOTSUPP;
1739         bio_put(bio);
1740         return ret;
1741 }
1742
1743 static int submit_extent_page(int rw, struct extent_io_tree *tree,
1744                               struct page *page, sector_t sector,
1745                               size_t size, unsigned long offset,
1746                               struct block_device *bdev,
1747                               struct bio **bio_ret,
1748                               unsigned long max_pages,
1749                               bio_end_io_t end_io_func,
1750                               int mirror_num)
1751 {
1752         int ret = 0;
1753         struct bio *bio;
1754         int nr;
1755
1756         if (bio_ret && *bio_ret) {
1757                 bio = *bio_ret;
1758                 if (bio->bi_sector + (bio->bi_size >> 9) != sector ||
1759                     (tree->ops && tree->ops->merge_bio_hook &&
1760                      tree->ops->merge_bio_hook(page, offset, size, bio)) ||
1761                     bio_add_page(bio, page, size, offset) < size) {
1762                         ret = submit_one_bio(rw, bio, mirror_num);
1763                         bio = NULL;
1764                 } else {
1765                         return 0;
1766                 }
1767         }
1768         nr = bio_get_nr_vecs(bdev);
1769         bio = extent_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
1770         if (!bio) {
1771                 printk("failed to allocate bio nr %d\n", nr);
1772         }
1773
1774
1775         bio_add_page(bio, page, size, offset);
1776         bio->bi_end_io = end_io_func;
1777         bio->bi_private = tree;
1778
1779         if (bio_ret) {
1780                 *bio_ret = bio;
1781         } else {
1782                 ret = submit_one_bio(rw, bio, mirror_num);
1783         }
1784
1785         return ret;
1786 }
1787
1788 void set_page_extent_mapped(struct page *page)
1789 {
1790         if (!PagePrivate(page)) {
1791                 SetPagePrivate(page);
1792                 WARN_ON(!page->mapping->a_ops->invalidatepage);
1793                 set_page_private(page, EXTENT_PAGE_PRIVATE);
1794                 page_cache_get(page);
1795         }
1796 }
1797
1798 void set_page_extent_head(struct page *page, unsigned long len)
1799 {
1800         set_page_private(page, EXTENT_PAGE_PRIVATE_FIRST_PAGE | len << 2);
1801 }
1802
1803 /*
1804  * basic readpage implementation.  Locked extent state structs are inserted
1805  * into the tree that are removed when the IO is done (by the end_io
1806  * handlers)
1807  */
1808 static int __extent_read_full_page(struct extent_io_tree *tree,
1809                                    struct page *page,
1810                                    get_extent_t *get_extent,
1811                                    struct bio **bio, int mirror_num)
1812 {
1813         struct inode *inode = page->mapping->host;
1814         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1815         u64 page_end = start + PAGE_CACHE_SIZE - 1;
1816         u64 end;
1817         u64 cur = start;
1818         u64 extent_offset;
1819         u64 last_byte = i_size_read(inode);
1820         u64 block_start;
1821         u64 cur_end;
1822         sector_t sector;
1823         struct extent_map *em;
1824         struct block_device *bdev;
1825         int ret;
1826         int nr = 0;
1827         size_t page_offset = 0;
1828         size_t iosize;
1829         size_t blocksize = inode->i_sb->s_blocksize;
1830
1831         set_page_extent_mapped(page);
1832
1833         end = page_end;
1834         lock_extent(tree, start, end, GFP_NOFS);
1835
1836         while (cur <= end) {
1837                 if (cur >= last_byte) {
1838                         char *userpage;
1839                         iosize = PAGE_CACHE_SIZE - page_offset;
1840                         userpage = kmap_atomic(page, KM_USER0);
1841                         memset(userpage + page_offset, 0, iosize);
1842                         flush_dcache_page(page);
1843                         kunmap_atomic(userpage, KM_USER0);
1844                         set_extent_uptodate(tree, cur, cur + iosize - 1,
1845                                             GFP_NOFS);
1846                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1847                         break;
1848                 }
1849                 em = get_extent(inode, page, page_offset, cur,
1850                                 end - cur + 1, 0);
1851                 if (IS_ERR(em) || !em) {
1852                         SetPageError(page);
1853                         unlock_extent(tree, cur, end, GFP_NOFS);
1854                         break;
1855                 }
1856
1857                 extent_offset = cur - em->start;
1858                 BUG_ON(extent_map_end(em) <= cur);
1859                 BUG_ON(end < cur);
1860
1861                 iosize = min(extent_map_end(em) - cur, end - cur + 1);
1862                 cur_end = min(extent_map_end(em) - 1, end);
1863                 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
1864                 sector = (em->block_start + extent_offset) >> 9;
1865                 bdev = em->bdev;
1866                 block_start = em->block_start;
1867                 free_extent_map(em);
1868                 em = NULL;
1869
1870                 /* we've found a hole, just zero and go on */
1871                 if (block_start == EXTENT_MAP_HOLE) {
1872                         char *userpage;
1873                         userpage = kmap_atomic(page, KM_USER0);
1874                         memset(userpage + page_offset, 0, iosize);
1875                         flush_dcache_page(page);
1876                         kunmap_atomic(userpage, KM_USER0);
1877
1878                         set_extent_uptodate(tree, cur, cur + iosize - 1,
1879                                             GFP_NOFS);
1880                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1881                         cur = cur + iosize;
1882                         page_offset += iosize;
1883                         continue;
1884                 }
1885                 /* the get_extent function already copied into the page */
1886                 if (test_range_bit(tree, cur, cur_end, EXTENT_UPTODATE, 1)) {
1887                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1888                         cur = cur + iosize;
1889                         page_offset += iosize;
1890                         continue;
1891                 }
1892                 /* we have an inline extent but it didn't get marked up
1893                  * to date.  Error out
1894                  */
1895                 if (block_start == EXTENT_MAP_INLINE) {
1896                         SetPageError(page);
1897                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1898                         cur = cur + iosize;
1899                         page_offset += iosize;
1900                         continue;
1901                 }
1902
1903                 ret = 0;
1904                 if (tree->ops && tree->ops->readpage_io_hook) {
1905                         ret = tree->ops->readpage_io_hook(page, cur,
1906                                                           cur + iosize - 1);
1907                 }
1908                 if (!ret) {
1909                         unsigned long nr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
1910                         nr -= page->index;
1911                         ret = submit_extent_page(READ, tree, page,
1912                                          sector, iosize, page_offset,
1913                                          bdev, bio, nr,
1914                                          end_bio_extent_readpage, mirror_num);
1915                 }
1916                 if (ret)
1917                         SetPageError(page);
1918                 cur = cur + iosize;
1919                 page_offset += iosize;
1920                 nr++;
1921         }
1922         if (!nr) {
1923                 if (!PageError(page))
1924                         SetPageUptodate(page);
1925                 unlock_page(page);
1926         }
1927         return 0;
1928 }
1929
1930 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
1931                             get_extent_t *get_extent)
1932 {
1933         struct bio *bio = NULL;
1934         int ret;
1935
1936         ret = __extent_read_full_page(tree, page, get_extent, &bio, 0);
1937         if (bio)
1938                 submit_one_bio(READ, bio, 0);
1939         return ret;
1940 }
1941 EXPORT_SYMBOL(extent_read_full_page);
1942
1943 /*
1944  * the writepage semantics are similar to regular writepage.  extent
1945  * records are inserted to lock ranges in the tree, and as dirty areas
1946  * are found, they are marked writeback.  Then the lock bits are removed
1947  * and the end_io handler clears the writeback ranges
1948  */
1949 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
1950                               void *data)
1951 {
1952         struct inode *inode = page->mapping->host;
1953         struct extent_page_data *epd = data;
1954         struct extent_io_tree *tree = epd->tree;
1955         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1956         u64 delalloc_start;
1957         u64 page_end = start + PAGE_CACHE_SIZE - 1;
1958         u64 end;
1959         u64 cur = start;
1960         u64 extent_offset;
1961         u64 last_byte = i_size_read(inode);
1962         u64 block_start;
1963         u64 iosize;
1964         sector_t sector;
1965         struct extent_map *em;
1966         struct block_device *bdev;
1967         int ret;
1968         int nr = 0;
1969         size_t page_offset = 0;
1970         size_t blocksize;
1971         loff_t i_size = i_size_read(inode);
1972         unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
1973         u64 nr_delalloc;
1974         u64 delalloc_end;
1975
1976         WARN_ON(!PageLocked(page));
1977         if (page->index > end_index) {
1978                 clear_extent_dirty(tree, start, page_end, GFP_NOFS);
1979                 unlock_page(page);
1980                 return 0;
1981         }
1982
1983         if (page->index == end_index) {
1984                 char *userpage;
1985
1986                 size_t offset = i_size & (PAGE_CACHE_SIZE - 1);
1987
1988                 userpage = kmap_atomic(page, KM_USER0);
1989                 memset(userpage + offset, 0, PAGE_CACHE_SIZE - offset);
1990                 flush_dcache_page(page);
1991                 kunmap_atomic(userpage, KM_USER0);
1992         }
1993
1994         set_page_extent_mapped(page);
1995
1996         delalloc_start = start;
1997         delalloc_end = 0;
1998         while(delalloc_end < page_end) {
1999                 nr_delalloc = find_lock_delalloc_range(tree, &delalloc_start,
2000                                                        &delalloc_end,
2001                                                        128 * 1024 * 1024);
2002                 if (nr_delalloc == 0) {
2003                         delalloc_start = delalloc_end + 1;
2004                         continue;
2005                 }
2006                 tree->ops->fill_delalloc(inode, delalloc_start,
2007                                          delalloc_end);
2008                 clear_extent_bit(tree, delalloc_start,
2009                                  delalloc_end,
2010                                  EXTENT_LOCKED | EXTENT_DELALLOC,
2011                                  1, 0, GFP_NOFS);
2012                 delalloc_start = delalloc_end + 1;
2013         }
2014         lock_extent(tree, start, page_end, GFP_NOFS);
2015
2016         end = page_end;
2017         if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0)) {
2018                 printk("found delalloc bits after lock_extent\n");
2019         }
2020
2021         if (last_byte <= start) {
2022                 clear_extent_dirty(tree, start, page_end, GFP_NOFS);
2023                 goto done;
2024         }
2025
2026         set_extent_uptodate(tree, start, page_end, GFP_NOFS);
2027         blocksize = inode->i_sb->s_blocksize;
2028
2029         while (cur <= end) {
2030                 if (cur >= last_byte) {
2031                         clear_extent_dirty(tree, cur, page_end, GFP_NOFS);
2032                         break;
2033                 }
2034                 em = epd->get_extent(inode, page, page_offset, cur,
2035                                      end - cur + 1, 1);
2036                 if (IS_ERR(em) || !em) {
2037                         SetPageError(page);
2038                         break;
2039                 }
2040
2041                 extent_offset = cur - em->start;
2042                 BUG_ON(extent_map_end(em) <= cur);
2043                 BUG_ON(end < cur);
2044                 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2045                 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2046                 sector = (em->block_start + extent_offset) >> 9;
2047                 bdev = em->bdev;
2048                 block_start = em->block_start;
2049                 free_extent_map(em);
2050                 em = NULL;
2051
2052                 if (block_start == EXTENT_MAP_HOLE ||
2053                     block_start == EXTENT_MAP_INLINE) {
2054                         clear_extent_dirty(tree, cur,
2055                                            cur + iosize - 1, GFP_NOFS);
2056                         cur = cur + iosize;
2057                         page_offset += iosize;
2058                         continue;
2059                 }
2060
2061                 /* leave this out until we have a page_mkwrite call */
2062                 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
2063                                    EXTENT_DIRTY, 0)) {
2064                         cur = cur + iosize;
2065                         page_offset += iosize;
2066                         continue;
2067                 }
2068                 clear_extent_dirty(tree, cur, cur + iosize - 1, GFP_NOFS);
2069                 if (tree->ops && tree->ops->writepage_io_hook) {
2070                         ret = tree->ops->writepage_io_hook(page, cur,
2071                                                 cur + iosize - 1);
2072                 } else {
2073                         ret = 0;
2074                 }
2075                 if (ret)
2076                         SetPageError(page);
2077                 else {
2078                         unsigned long max_nr = end_index + 1;
2079                         set_range_writeback(tree, cur, cur + iosize - 1);
2080                         if (!PageWriteback(page)) {
2081                                 printk("warning page %lu not writeback, "
2082                                        "cur %llu end %llu\n", page->index,
2083                                        (unsigned long long)cur,
2084                                        (unsigned long long)end);
2085                         }
2086
2087                         ret = submit_extent_page(WRITE, tree, page, sector,
2088                                                  iosize, page_offset, bdev,
2089                                                  &epd->bio, max_nr,
2090                                                  end_bio_extent_writepage, 0);
2091                         if (ret)
2092                                 SetPageError(page);
2093                 }
2094                 cur = cur + iosize;
2095                 page_offset += iosize;
2096                 nr++;
2097         }
2098 done:
2099         if (nr == 0) {
2100                 /* make sure the mapping tag for page dirty gets cleared */
2101                 set_page_writeback(page);
2102                 end_page_writeback(page);
2103         }
2104         unlock_extent(tree, start, page_end, GFP_NOFS);
2105         unlock_page(page);
2106         return 0;
2107 }
2108
2109 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,20)
2110 /* Taken directly from 2.6.23 for 2.6.18 back port */
2111 typedef int (*writepage_t)(struct page *page, struct writeback_control *wbc,
2112                                 void *data);
2113
2114 /**
2115  * write_cache_pages - walk the list of dirty pages of the given address space
2116  * and write all of them.
2117  * @mapping: address space structure to write
2118  * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2119  * @writepage: function called for each page
2120  * @data: data passed to writepage function
2121  *
2122  * If a page is already under I/O, write_cache_pages() skips it, even
2123  * if it's dirty.  This is desirable behaviour for memory-cleaning writeback,
2124  * but it is INCORRECT for data-integrity system calls such as fsync().  fsync()
2125  * and msync() need to guarantee that all the data which was dirty at the time
2126  * the call was made get new I/O started against them.  If wbc->sync_mode is
2127  * WB_SYNC_ALL then we were called for data integrity and we must wait for
2128  * existing IO to complete.
2129  */
2130 static int write_cache_pages(struct address_space *mapping,
2131                       struct writeback_control *wbc, writepage_t writepage,
2132                       void *data)
2133 {
2134         struct backing_dev_info *bdi = mapping->backing_dev_info;
2135         int ret = 0;
2136         int done = 0;
2137         struct pagevec pvec;
2138         int nr_pages;
2139         pgoff_t index;
2140         pgoff_t end;            /* Inclusive */
2141         int scanned = 0;
2142         int range_whole = 0;
2143
2144         if (wbc->nonblocking && bdi_write_congested(bdi)) {
2145                 wbc->encountered_congestion = 1;
2146                 return 0;
2147         }
2148
2149         pagevec_init(&pvec, 0);
2150         if (wbc->range_cyclic) {
2151                 index = mapping->writeback_index; /* Start from prev offset */
2152                 end = -1;
2153         } else {
2154                 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2155                 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2156                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2157                         range_whole = 1;
2158                 scanned = 1;
2159         }
2160 retry:
2161         while (!done && (index <= end) &&
2162                (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
2163                                               PAGECACHE_TAG_DIRTY,
2164                                               min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
2165                 unsigned i;
2166
2167                 scanned = 1;
2168                 for (i = 0; i < nr_pages; i++) {
2169                         struct page *page = pvec.pages[i];
2170
2171                         /*
2172                          * At this point we hold neither mapping->tree_lock nor
2173                          * lock on the page itself: the page may be truncated or
2174                          * invalidated (changing page->mapping to NULL), or even
2175                          * swizzled back from swapper_space to tmpfs file
2176                          * mapping
2177                          */
2178                         lock_page(page);
2179
2180                         if (unlikely(page->mapping != mapping)) {
2181                                 unlock_page(page);
2182                                 continue;
2183                         }
2184
2185                         if (!wbc->range_cyclic && page->index > end) {
2186                                 done = 1;
2187                                 unlock_page(page);
2188                                 continue;
2189                         }
2190
2191                         if (wbc->sync_mode != WB_SYNC_NONE)
2192                                 wait_on_page_writeback(page);
2193
2194                         if (PageWriteback(page) ||
2195                             !clear_page_dirty_for_io(page)) {
2196                                 unlock_page(page);
2197                                 continue;
2198                         }
2199
2200                         ret = (*writepage)(page, wbc, data);
2201
2202                         if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
2203                                 unlock_page(page);
2204                                 ret = 0;
2205                         }
2206                         if (ret || (--(wbc->nr_to_write) <= 0))
2207                                 done = 1;
2208                         if (wbc->nonblocking && bdi_write_congested(bdi)) {
2209                                 wbc->encountered_congestion = 1;
2210                                 done = 1;
2211                         }
2212                 }
2213                 pagevec_release(&pvec);
2214                 cond_resched();
2215         }
2216         if (!scanned && !done) {
2217                 /*
2218                  * We hit the last page and there is more work to be done: wrap
2219                  * back to the start of the file
2220                  */
2221                 scanned = 1;
2222                 index = 0;
2223                 goto retry;
2224         }
2225         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2226                 mapping->writeback_index = index;
2227         return ret;
2228 }
2229 #endif
2230
2231 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
2232                           get_extent_t *get_extent,
2233                           struct writeback_control *wbc)
2234 {
2235         int ret;
2236         struct address_space *mapping = page->mapping;
2237         struct extent_page_data epd = {
2238                 .bio = NULL,
2239                 .tree = tree,
2240                 .get_extent = get_extent,
2241         };
2242         struct writeback_control wbc_writepages = {
2243                 .bdi            = wbc->bdi,
2244                 .sync_mode      = WB_SYNC_NONE,
2245                 .older_than_this = NULL,
2246                 .nr_to_write    = 64,
2247                 .range_start    = page_offset(page) + PAGE_CACHE_SIZE,
2248                 .range_end      = (loff_t)-1,
2249         };
2250
2251
2252         ret = __extent_writepage(page, wbc, &epd);
2253
2254         write_cache_pages(mapping, &wbc_writepages, __extent_writepage, &epd);
2255         if (epd.bio) {
2256                 submit_one_bio(WRITE, epd.bio, 0);
2257         }
2258         return ret;
2259 }
2260 EXPORT_SYMBOL(extent_write_full_page);
2261
2262
2263 int extent_writepages(struct extent_io_tree *tree,
2264                       struct address_space *mapping,
2265                       get_extent_t *get_extent,
2266                       struct writeback_control *wbc)
2267 {
2268         int ret = 0;
2269         struct extent_page_data epd = {
2270                 .bio = NULL,
2271                 .tree = tree,
2272                 .get_extent = get_extent,
2273         };
2274
2275         ret = write_cache_pages(mapping, wbc, __extent_writepage, &epd);
2276         if (epd.bio) {
2277                 submit_one_bio(WRITE, epd.bio, 0);
2278         }
2279         return ret;
2280 }
2281 EXPORT_SYMBOL(extent_writepages);
2282
2283 int extent_readpages(struct extent_io_tree *tree,
2284                      struct address_space *mapping,
2285                      struct list_head *pages, unsigned nr_pages,
2286                      get_extent_t get_extent)
2287 {
2288         struct bio *bio = NULL;
2289         unsigned page_idx;
2290         struct pagevec pvec;
2291
2292         pagevec_init(&pvec, 0);
2293         for (page_idx = 0; page_idx < nr_pages; page_idx++) {
2294                 struct page *page = list_entry(pages->prev, struct page, lru);
2295
2296                 prefetchw(&page->flags);
2297                 list_del(&page->lru);
2298                 /*
2299                  * what we want to do here is call add_to_page_cache_lru,
2300                  * but that isn't exported, so we reproduce it here
2301                  */
2302                 if (!add_to_page_cache(page, mapping,
2303                                         page->index, GFP_KERNEL)) {
2304
2305                         /* open coding of lru_cache_add, also not exported */
2306                         page_cache_get(page);
2307                         if (!pagevec_add(&pvec, page))
2308                                 __pagevec_lru_add(&pvec);
2309                         __extent_read_full_page(tree, page, get_extent,
2310                                                 &bio, 0);
2311                 }
2312                 page_cache_release(page);
2313         }
2314         if (pagevec_count(&pvec))
2315                 __pagevec_lru_add(&pvec);
2316         BUG_ON(!list_empty(pages));
2317         if (bio)
2318                 submit_one_bio(READ, bio, 0);
2319         return 0;
2320 }
2321 EXPORT_SYMBOL(extent_readpages);
2322
2323 /*
2324  * basic invalidatepage code, this waits on any locked or writeback
2325  * ranges corresponding to the page, and then deletes any extent state
2326  * records from the tree
2327  */
2328 int extent_invalidatepage(struct extent_io_tree *tree,
2329                           struct page *page, unsigned long offset)
2330 {
2331         u64 start = ((u64)page->index << PAGE_CACHE_SHIFT);
2332         u64 end = start + PAGE_CACHE_SIZE - 1;
2333         size_t blocksize = page->mapping->host->i_sb->s_blocksize;
2334
2335         start += (offset + blocksize -1) & ~(blocksize - 1);
2336         if (start > end)
2337                 return 0;
2338
2339         lock_extent(tree, start, end, GFP_NOFS);
2340         wait_on_extent_writeback(tree, start, end);
2341         clear_extent_bit(tree, start, end,
2342                          EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC,
2343                          1, 1, GFP_NOFS);
2344         return 0;
2345 }
2346 EXPORT_SYMBOL(extent_invalidatepage);
2347
2348 /*
2349  * simple commit_write call, set_range_dirty is used to mark both
2350  * the pages and the extent records as dirty
2351  */
2352 int extent_commit_write(struct extent_io_tree *tree,
2353                         struct inode *inode, struct page *page,
2354                         unsigned from, unsigned to)
2355 {
2356         loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
2357
2358         set_page_extent_mapped(page);
2359         set_page_dirty(page);
2360
2361         if (pos > inode->i_size) {
2362                 i_size_write(inode, pos);
2363                 mark_inode_dirty(inode);
2364         }
2365         return 0;
2366 }
2367 EXPORT_SYMBOL(extent_commit_write);
2368
2369 int extent_prepare_write(struct extent_io_tree *tree,
2370                          struct inode *inode, struct page *page,
2371                          unsigned from, unsigned to, get_extent_t *get_extent)
2372 {
2373         u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2374         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
2375         u64 block_start;
2376         u64 orig_block_start;
2377         u64 block_end;
2378         u64 cur_end;
2379         struct extent_map *em;
2380         unsigned blocksize = 1 << inode->i_blkbits;
2381         size_t page_offset = 0;
2382         size_t block_off_start;
2383         size_t block_off_end;
2384         int err = 0;
2385         int iocount = 0;
2386         int ret = 0;
2387         int isnew;
2388
2389         set_page_extent_mapped(page);
2390
2391         block_start = (page_start + from) & ~((u64)blocksize - 1);
2392         block_end = (page_start + to - 1) | (blocksize - 1);
2393         orig_block_start = block_start;
2394
2395         lock_extent(tree, page_start, page_end, GFP_NOFS);
2396         while(block_start <= block_end) {
2397                 em = get_extent(inode, page, page_offset, block_start,
2398                                 block_end - block_start + 1, 1);
2399                 if (IS_ERR(em) || !em) {
2400                         goto err;
2401                 }
2402                 cur_end = min(block_end, extent_map_end(em) - 1);
2403                 block_off_start = block_start & (PAGE_CACHE_SIZE - 1);
2404                 block_off_end = block_off_start + blocksize;
2405                 isnew = clear_extent_new(tree, block_start, cur_end, GFP_NOFS);
2406
2407                 if (!PageUptodate(page) && isnew &&
2408                     (block_off_end > to || block_off_start < from)) {
2409                         void *kaddr;
2410
2411                         kaddr = kmap_atomic(page, KM_USER0);
2412                         if (block_off_end > to)
2413                                 memset(kaddr + to, 0, block_off_end - to);
2414                         if (block_off_start < from)
2415                                 memset(kaddr + block_off_start, 0,
2416                                        from - block_off_start);
2417                         flush_dcache_page(page);
2418                         kunmap_atomic(kaddr, KM_USER0);
2419                 }
2420                 if ((em->block_start != EXTENT_MAP_HOLE &&
2421                      em->block_start != EXTENT_MAP_INLINE) &&
2422                     !isnew && !PageUptodate(page) &&
2423                     (block_off_end > to || block_off_start < from) &&
2424                     !test_range_bit(tree, block_start, cur_end,
2425                                     EXTENT_UPTODATE, 1)) {
2426                         u64 sector;
2427                         u64 extent_offset = block_start - em->start;
2428                         size_t iosize;
2429                         sector = (em->block_start + extent_offset) >> 9;
2430                         iosize = (cur_end - block_start + blocksize) &
2431                                 ~((u64)blocksize - 1);
2432                         /*
2433                          * we've already got the extent locked, but we
2434                          * need to split the state such that our end_bio
2435                          * handler can clear the lock.
2436                          */
2437                         set_extent_bit(tree, block_start,
2438                                        block_start + iosize - 1,
2439                                        EXTENT_LOCKED, 0, NULL, GFP_NOFS);
2440                         ret = submit_extent_page(READ, tree, page,
2441                                          sector, iosize, page_offset, em->bdev,
2442                                          NULL, 1,
2443                                          end_bio_extent_preparewrite, 0);
2444                         iocount++;
2445                         block_start = block_start + iosize;
2446                 } else {
2447                         set_extent_uptodate(tree, block_start, cur_end,
2448                                             GFP_NOFS);
2449                         unlock_extent(tree, block_start, cur_end, GFP_NOFS);
2450                         block_start = cur_end + 1;
2451                 }
2452                 page_offset = block_start & (PAGE_CACHE_SIZE - 1);
2453                 free_extent_map(em);
2454         }
2455         if (iocount) {
2456                 wait_extent_bit(tree, orig_block_start,
2457                                 block_end, EXTENT_LOCKED);
2458         }
2459         check_page_uptodate(tree, page);
2460 err:
2461         /* FIXME, zero out newly allocated blocks on error */
2462         return err;
2463 }
2464 EXPORT_SYMBOL(extent_prepare_write);
2465
2466 /*
2467  * a helper for releasepage.  As long as there are no locked extents
2468  * in the range corresponding to the page, both state records and extent
2469  * map records are removed
2470  */
2471 int try_release_extent_mapping(struct extent_map_tree *map,
2472                                struct extent_io_tree *tree, struct page *page,
2473                                gfp_t mask)
2474 {
2475         struct extent_map *em;
2476         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2477         u64 end = start + PAGE_CACHE_SIZE - 1;
2478         u64 orig_start = start;
2479         int ret = 1;
2480         if ((mask & __GFP_WAIT) &&
2481             page->mapping->host->i_size > 16 * 1024 * 1024) {
2482                 u64 len;
2483                 while (start <= end) {
2484                         len = end - start + 1;
2485                         spin_lock(&map->lock);
2486                         em = lookup_extent_mapping(map, start, len);
2487                         if (!em || IS_ERR(em)) {
2488                                 spin_unlock(&map->lock);
2489                                 break;
2490                         }
2491                         if (em->start != start) {
2492                                 spin_unlock(&map->lock);
2493                                 free_extent_map(em);
2494                                 break;
2495                         }
2496                         if (!test_range_bit(tree, em->start,
2497                                             extent_map_end(em) - 1,
2498                                             EXTENT_LOCKED, 0)) {
2499                                 remove_extent_mapping(map, em);
2500                                 /* once for the rb tree */
2501                                 free_extent_map(em);
2502                         }
2503                         start = extent_map_end(em);
2504                         spin_unlock(&map->lock);
2505
2506                         /* once for us */
2507                         free_extent_map(em);
2508                 }
2509         }
2510         if (test_range_bit(tree, orig_start, end, EXTENT_IOBITS, 0))
2511                 ret = 0;
2512         else {
2513                 if ((mask & GFP_NOFS) == GFP_NOFS)
2514                         mask = GFP_NOFS;
2515                 clear_extent_bit(tree, orig_start, end, EXTENT_UPTODATE,
2516                                  1, 1, mask);
2517         }
2518         return ret;
2519 }
2520 EXPORT_SYMBOL(try_release_extent_mapping);
2521
2522 sector_t extent_bmap(struct address_space *mapping, sector_t iblock,
2523                 get_extent_t *get_extent)
2524 {
2525         struct inode *inode = mapping->host;
2526         u64 start = iblock << inode->i_blkbits;
2527         sector_t sector = 0;
2528         struct extent_map *em;
2529
2530         em = get_extent(inode, NULL, 0, start, (1 << inode->i_blkbits), 0);
2531         if (!em || IS_ERR(em))
2532                 return 0;
2533
2534         if (em->block_start == EXTENT_MAP_INLINE ||
2535             em->block_start == EXTENT_MAP_HOLE)
2536                 goto out;
2537
2538         sector = (em->block_start + start - em->start) >> inode->i_blkbits;
2539 out:
2540         free_extent_map(em);
2541         return sector;
2542 }
2543
2544 static int add_lru(struct extent_io_tree *tree, struct extent_buffer *eb)
2545 {
2546         if (list_empty(&eb->lru)) {
2547                 extent_buffer_get(eb);
2548                 list_add(&eb->lru, &tree->buffer_lru);
2549                 tree->lru_size++;
2550                 if (tree->lru_size >= BUFFER_LRU_MAX) {
2551                         struct extent_buffer *rm;
2552                         rm = list_entry(tree->buffer_lru.prev,
2553                                         struct extent_buffer, lru);
2554                         tree->lru_size--;
2555                         list_del_init(&rm->lru);
2556                         free_extent_buffer(rm);
2557                 }
2558         } else
2559                 list_move(&eb->lru, &tree->buffer_lru);
2560         return 0;
2561 }
2562 static struct extent_buffer *find_lru(struct extent_io_tree *tree,
2563                                       u64 start, unsigned long len)
2564 {
2565         struct list_head *lru = &tree->buffer_lru;
2566         struct list_head *cur = lru->next;
2567         struct extent_buffer *eb;
2568
2569         if (list_empty(lru))
2570                 return NULL;
2571
2572         do {
2573                 eb = list_entry(cur, struct extent_buffer, lru);
2574                 if (eb->start == start && eb->len == len) {
2575                         extent_buffer_get(eb);
2576                         return eb;
2577                 }
2578                 cur = cur->next;
2579         } while (cur != lru);
2580         return NULL;
2581 }
2582
2583 static inline unsigned long num_extent_pages(u64 start, u64 len)
2584 {
2585         return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
2586                 (start >> PAGE_CACHE_SHIFT);
2587 }
2588
2589 static inline struct page *extent_buffer_page(struct extent_buffer *eb,
2590                                               unsigned long i)
2591 {
2592         struct page *p;
2593         struct address_space *mapping;
2594
2595         if (i == 0)
2596                 return eb->first_page;
2597         i += eb->start >> PAGE_CACHE_SHIFT;
2598         mapping = eb->first_page->mapping;
2599         read_lock_irq(&mapping->tree_lock);
2600         p = radix_tree_lookup(&mapping->page_tree, i);
2601         read_unlock_irq(&mapping->tree_lock);
2602         return p;
2603 }
2604
2605 int release_extent_buffer_tail_pages(struct extent_buffer *eb)
2606 {
2607         unsigned long num_pages = num_extent_pages(eb->start, eb->len);
2608         struct page *page;
2609         unsigned long i;
2610
2611         if (num_pages == 1)
2612                 return 0;
2613         for (i = 1; i < num_pages; i++) {
2614                 page = extent_buffer_page(eb, i);
2615                 page_cache_release(page);
2616         }
2617         return 0;
2618 }
2619
2620
2621 int invalidate_extent_lru(struct extent_io_tree *tree, u64 start,
2622                           unsigned long len)
2623 {
2624         struct list_head *lru = &tree->buffer_lru;
2625         struct list_head *cur = lru->next;
2626         struct extent_buffer *eb;
2627         int found = 0;
2628
2629         spin_lock(&tree->lru_lock);
2630         if (list_empty(lru))
2631                 goto out;
2632
2633         do {
2634                 eb = list_entry(cur, struct extent_buffer, lru);
2635                 if (eb->start <= start && eb->start + eb->len > start) {
2636                         eb->flags &= ~EXTENT_UPTODATE;
2637                 }
2638                 cur = cur->next;
2639         } while (cur != lru);
2640 out:
2641         spin_unlock(&tree->lru_lock);
2642         return found;
2643 }
2644
2645 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
2646                                                    u64 start,
2647                                                    unsigned long len,
2648                                                    gfp_t mask)
2649 {
2650         struct extent_buffer *eb = NULL;
2651         unsigned long flags;
2652
2653         spin_lock(&tree->lru_lock);
2654         eb = find_lru(tree, start, len);
2655         spin_unlock(&tree->lru_lock);
2656         if (eb) {
2657                 return eb;
2658         }
2659
2660         eb = kmem_cache_zalloc(extent_buffer_cache, mask);
2661         INIT_LIST_HEAD(&eb->lru);
2662         eb->start = start;
2663         eb->len = len;
2664         spin_lock_irqsave(&leak_lock, flags);
2665         list_add(&eb->leak_list, &buffers);
2666         spin_unlock_irqrestore(&leak_lock, flags);
2667         atomic_set(&eb->refs, 1);
2668
2669         return eb;
2670 }
2671
2672 static void __free_extent_buffer(struct extent_buffer *eb)
2673 {
2674         unsigned long flags;
2675         spin_lock_irqsave(&leak_lock, flags);
2676         list_del(&eb->leak_list);
2677         spin_unlock_irqrestore(&leak_lock, flags);
2678         kmem_cache_free(extent_buffer_cache, eb);
2679 }
2680
2681 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
2682                                           u64 start, unsigned long len,
2683                                           struct page *page0,
2684                                           gfp_t mask)
2685 {
2686         unsigned long num_pages = num_extent_pages(start, len);
2687         unsigned long i;
2688         unsigned long index = start >> PAGE_CACHE_SHIFT;
2689         struct extent_buffer *eb;
2690         struct page *p;
2691         struct address_space *mapping = tree->mapping;
2692         int uptodate = 1;
2693
2694         eb = __alloc_extent_buffer(tree, start, len, mask);
2695         if (!eb)
2696                 return NULL;
2697
2698         if (eb->flags & EXTENT_BUFFER_FILLED)
2699                 goto lru_add;
2700
2701         if (page0) {
2702                 eb->first_page = page0;
2703                 i = 1;
2704                 index++;
2705                 page_cache_get(page0);
2706                 mark_page_accessed(page0);
2707                 set_page_extent_mapped(page0);
2708                 set_page_extent_head(page0, len);
2709                 uptodate = PageUptodate(page0);
2710         } else {
2711                 i = 0;
2712         }
2713         for (; i < num_pages; i++, index++) {
2714                 p = find_or_create_page(mapping, index, mask | __GFP_HIGHMEM);
2715                 if (!p) {
2716                         WARN_ON(1);
2717                         goto fail;
2718                 }
2719                 set_page_extent_mapped(p);
2720                 mark_page_accessed(p);
2721                 if (i == 0) {
2722                         eb->first_page = p;
2723                         set_page_extent_head(p, len);
2724                 } else {
2725                         set_page_private(p, EXTENT_PAGE_PRIVATE);
2726                 }
2727                 if (!PageUptodate(p))
2728                         uptodate = 0;
2729                 unlock_page(p);
2730         }
2731         if (uptodate)
2732                 eb->flags |= EXTENT_UPTODATE;
2733         eb->flags |= EXTENT_BUFFER_FILLED;
2734
2735 lru_add:
2736         spin_lock(&tree->lru_lock);
2737         add_lru(tree, eb);
2738         spin_unlock(&tree->lru_lock);
2739         return eb;
2740
2741 fail:
2742         spin_lock(&tree->lru_lock);
2743         list_del_init(&eb->lru);
2744         spin_unlock(&tree->lru_lock);
2745         if (!atomic_dec_and_test(&eb->refs))
2746                 return NULL;
2747         for (index = 1; index < i; index++) {
2748                 page_cache_release(extent_buffer_page(eb, index));
2749         }
2750         if (i > 0)
2751                 page_cache_release(extent_buffer_page(eb, 0));
2752         __free_extent_buffer(eb);
2753         return NULL;
2754 }
2755 EXPORT_SYMBOL(alloc_extent_buffer);
2756
2757 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
2758                                          u64 start, unsigned long len,
2759                                           gfp_t mask)
2760 {
2761         unsigned long num_pages = num_extent_pages(start, len);
2762         unsigned long i;
2763         unsigned long index = start >> PAGE_CACHE_SHIFT;
2764         struct extent_buffer *eb;
2765         struct page *p;
2766         struct address_space *mapping = tree->mapping;
2767         int uptodate = 1;
2768
2769         eb = __alloc_extent_buffer(tree, start, len, mask);
2770         if (!eb)
2771                 return NULL;
2772
2773         if (eb->flags & EXTENT_BUFFER_FILLED)
2774                 goto lru_add;
2775
2776         for (i = 0; i < num_pages; i++, index++) {
2777                 p = find_lock_page(mapping, index);
2778                 if (!p) {
2779                         goto fail;
2780                 }
2781                 set_page_extent_mapped(p);
2782                 mark_page_accessed(p);
2783
2784                 if (i == 0) {
2785                         eb->first_page = p;
2786                         set_page_extent_head(p, len);
2787                 } else {
2788                         set_page_private(p, EXTENT_PAGE_PRIVATE);
2789                 }
2790
2791                 if (!PageUptodate(p))
2792                         uptodate = 0;
2793                 unlock_page(p);
2794         }
2795         if (uptodate)
2796                 eb->flags |= EXTENT_UPTODATE;
2797         eb->flags |= EXTENT_BUFFER_FILLED;
2798
2799 lru_add:
2800         spin_lock(&tree->lru_lock);
2801         add_lru(tree, eb);
2802         spin_unlock(&tree->lru_lock);
2803         return eb;
2804 fail:
2805         spin_lock(&tree->lru_lock);
2806         list_del_init(&eb->lru);
2807         spin_unlock(&tree->lru_lock);
2808         if (!atomic_dec_and_test(&eb->refs))
2809                 return NULL;
2810         for (index = 1; index < i; index++) {
2811                 page_cache_release(extent_buffer_page(eb, index));
2812         }
2813         if (i > 0)
2814                 page_cache_release(extent_buffer_page(eb, 0));
2815         __free_extent_buffer(eb);
2816         return NULL;
2817 }
2818 EXPORT_SYMBOL(find_extent_buffer);
2819
2820 void free_extent_buffer(struct extent_buffer *eb)
2821 {
2822         unsigned long i;
2823         unsigned long num_pages;
2824
2825         if (!eb)
2826                 return;
2827
2828         if (!atomic_dec_and_test(&eb->refs))
2829                 return;
2830
2831         WARN_ON(!list_empty(&eb->lru));
2832         num_pages = num_extent_pages(eb->start, eb->len);
2833
2834         for (i = 1; i < num_pages; i++) {
2835                 page_cache_release(extent_buffer_page(eb, i));
2836         }
2837         page_cache_release(extent_buffer_page(eb, 0));
2838         __free_extent_buffer(eb);
2839 }
2840 EXPORT_SYMBOL(free_extent_buffer);
2841
2842 int clear_extent_buffer_dirty(struct extent_io_tree *tree,
2843                               struct extent_buffer *eb)
2844 {
2845         int set;
2846         unsigned long i;
2847         unsigned long num_pages;
2848         struct page *page;
2849
2850         u64 start = eb->start;
2851         u64 end = start + eb->len - 1;
2852
2853         set = clear_extent_dirty(tree, start, end, GFP_NOFS);
2854         num_pages = num_extent_pages(eb->start, eb->len);
2855
2856         for (i = 0; i < num_pages; i++) {
2857                 page = extent_buffer_page(eb, i);
2858                 lock_page(page);
2859                 if (i == 0)
2860                         set_page_extent_head(page, eb->len);
2861                 else
2862                         set_page_private(page, EXTENT_PAGE_PRIVATE);
2863
2864                 /*
2865                  * if we're on the last page or the first page and the
2866                  * block isn't aligned on a page boundary, do extra checks
2867                  * to make sure we don't clean page that is partially dirty
2868                  */
2869                 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
2870                     ((i == num_pages - 1) &&
2871                      ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
2872                         start = (u64)page->index << PAGE_CACHE_SHIFT;
2873                         end  = start + PAGE_CACHE_SIZE - 1;
2874                         if (test_range_bit(tree, start, end,
2875                                            EXTENT_DIRTY, 0)) {
2876                                 unlock_page(page);
2877                                 continue;
2878                         }
2879                 }
2880                 clear_page_dirty_for_io(page);
2881                 read_lock_irq(&page->mapping->tree_lock);
2882                 if (!PageDirty(page)) {
2883                         radix_tree_tag_clear(&page->mapping->page_tree,
2884                                                 page_index(page),
2885                                                 PAGECACHE_TAG_DIRTY);
2886                 }
2887                 read_unlock_irq(&page->mapping->tree_lock);
2888                 unlock_page(page);
2889         }
2890         return 0;
2891 }
2892 EXPORT_SYMBOL(clear_extent_buffer_dirty);
2893
2894 int wait_on_extent_buffer_writeback(struct extent_io_tree *tree,
2895                                     struct extent_buffer *eb)
2896 {
2897         return wait_on_extent_writeback(tree, eb->start,
2898                                         eb->start + eb->len - 1);
2899 }
2900 EXPORT_SYMBOL(wait_on_extent_buffer_writeback);
2901
2902 int set_extent_buffer_dirty(struct extent_io_tree *tree,
2903                              struct extent_buffer *eb)
2904 {
2905         unsigned long i;
2906         unsigned long num_pages;
2907
2908         num_pages = num_extent_pages(eb->start, eb->len);
2909         for (i = 0; i < num_pages; i++) {
2910                 struct page *page = extent_buffer_page(eb, i);
2911                 /* writepage may need to do something special for the
2912                  * first page, we have to make sure page->private is
2913                  * properly set.  releasepage may drop page->private
2914                  * on us if the page isn't already dirty.
2915                  */
2916                 if (i == 0) {
2917                         lock_page(page);
2918                         set_page_extent_head(page, eb->len);
2919                 } else if (PagePrivate(page) &&
2920                            page->private != EXTENT_PAGE_PRIVATE) {
2921                         lock_page(page);
2922                         set_page_extent_mapped(page);
2923                         unlock_page(page);
2924                 }
2925                 __set_page_dirty_nobuffers(extent_buffer_page(eb, i));
2926                 if (i == 0)
2927                         unlock_page(page);
2928         }
2929         return set_extent_dirty(tree, eb->start,
2930                                 eb->start + eb->len - 1, GFP_NOFS);
2931 }
2932 EXPORT_SYMBOL(set_extent_buffer_dirty);
2933
2934 int set_extent_buffer_uptodate(struct extent_io_tree *tree,
2935                                 struct extent_buffer *eb)
2936 {
2937         unsigned long i;
2938         struct page *page;
2939         unsigned long num_pages;
2940
2941         num_pages = num_extent_pages(eb->start, eb->len);
2942
2943         set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
2944                             GFP_NOFS);
2945         for (i = 0; i < num_pages; i++) {
2946                 page = extent_buffer_page(eb, i);
2947                 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
2948                     ((i == num_pages - 1) &&
2949                      ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
2950                         check_page_uptodate(tree, page);
2951                         continue;
2952                 }
2953                 SetPageUptodate(page);
2954         }
2955         return 0;
2956 }
2957 EXPORT_SYMBOL(set_extent_buffer_uptodate);
2958
2959 int extent_range_uptodate(struct extent_io_tree *tree,
2960                           u64 start, u64 end)
2961 {
2962         struct page *page;
2963         int ret;
2964         int pg_uptodate = 1;
2965         int uptodate;
2966         unsigned long index;
2967
2968         ret = test_range_bit(tree, start, end, EXTENT_UPTODATE, 1);
2969         if (ret)
2970                 return 1;
2971         while(start <= end) {
2972                 index = start >> PAGE_CACHE_SHIFT;
2973                 page = find_get_page(tree->mapping, index);
2974                 uptodate = PageUptodate(page);
2975                 page_cache_release(page);
2976                 if (!uptodate) {
2977                         pg_uptodate = 0;
2978                         break;
2979                 }
2980                 start += PAGE_CACHE_SIZE;
2981         }
2982         return pg_uptodate;
2983 }
2984
2985 int extent_buffer_uptodate(struct extent_io_tree *tree,
2986                            struct extent_buffer *eb)
2987 {
2988         int ret = 0;
2989         int ret2;
2990         unsigned long num_pages;
2991         unsigned long i;
2992         struct page *page;
2993         int pg_uptodate = 1;
2994
2995         if (eb->flags & EXTENT_UPTODATE)
2996                 ret = 1;
2997
2998         ret2  = test_range_bit(tree, eb->start, eb->start + eb->len - 1,
2999                            EXTENT_UPTODATE, 1);
3000
3001         num_pages = num_extent_pages(eb->start, eb->len);
3002         for (i = 0; i < num_pages; i++) {
3003                 page = extent_buffer_page(eb, i);
3004                 if (!PageUptodate(page)) {
3005                         pg_uptodate = 0;
3006                         break;
3007                 }
3008         }
3009         if ((ret || ret2) && !pg_uptodate) {
3010 printk("uptodate error2 eb %Lu ret %d ret2 %d pg_uptodate %d\n", eb->start, ret, ret2, pg_uptodate);
3011                 WARN_ON(1);
3012         }
3013         return (ret || ret2);
3014 }
3015 EXPORT_SYMBOL(extent_buffer_uptodate);
3016
3017 int read_extent_buffer_pages(struct extent_io_tree *tree,
3018                              struct extent_buffer *eb,
3019                              u64 start, int wait,
3020                              get_extent_t *get_extent, int mirror_num)
3021 {
3022         unsigned long i;
3023         unsigned long start_i;
3024         struct page *page;
3025         int err;
3026         int ret = 0;
3027         int locked_pages = 0;
3028         int all_uptodate = 1;
3029         int inc_all_pages = 0;
3030         unsigned long num_pages;
3031         struct bio *bio = NULL;
3032
3033         if (eb->flags & EXTENT_UPTODATE)
3034                 return 0;
3035
3036         if (test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3037                            EXTENT_UPTODATE, 1)) {
3038                 return 0;
3039         }
3040
3041         if (start) {
3042                 WARN_ON(start < eb->start);
3043                 start_i = (start >> PAGE_CACHE_SHIFT) -
3044                         (eb->start >> PAGE_CACHE_SHIFT);
3045         } else {
3046                 start_i = 0;
3047         }
3048
3049         num_pages = num_extent_pages(eb->start, eb->len);
3050         for (i = start_i; i < num_pages; i++) {
3051                 page = extent_buffer_page(eb, i);
3052                 if (!wait) {
3053                         if (TestSetPageLocked(page))
3054                                 goto unlock_exit;
3055                 } else {
3056                         lock_page(page);
3057                 }
3058                 locked_pages++;
3059                 if (!PageUptodate(page)) {
3060                         all_uptodate = 0;
3061                 }
3062         }
3063         if (all_uptodate) {
3064                 if (start_i == 0)
3065                         eb->flags |= EXTENT_UPTODATE;
3066                 goto unlock_exit;
3067         }
3068
3069         for (i = start_i; i < num_pages; i++) {
3070                 page = extent_buffer_page(eb, i);
3071                 if (inc_all_pages)
3072                         page_cache_get(page);
3073                 if (!PageUptodate(page)) {
3074                         if (start_i == 0)
3075                                 inc_all_pages = 1;
3076                         ClearPageError(page);
3077                         err = __extent_read_full_page(tree, page,
3078                                                       get_extent, &bio,
3079                                                       mirror_num);
3080                         if (err) {
3081                                 ret = err;
3082                         }
3083                 } else {
3084                         unlock_page(page);
3085                 }
3086         }
3087
3088         if (bio)
3089                 submit_one_bio(READ, bio, mirror_num);
3090
3091         if (ret || !wait) {
3092                 return ret;
3093         }
3094         for (i = start_i; i < num_pages; i++) {
3095                 page = extent_buffer_page(eb, i);
3096                 wait_on_page_locked(page);
3097                 if (!PageUptodate(page)) {
3098                         ret = -EIO;
3099                 }
3100         }
3101         if (!ret)
3102                 eb->flags |= EXTENT_UPTODATE;
3103         return ret;
3104
3105 unlock_exit:
3106         i = start_i;
3107         while(locked_pages > 0) {
3108                 page = extent_buffer_page(eb, i);
3109                 i++;
3110                 unlock_page(page);
3111                 locked_pages--;
3112         }
3113         return ret;
3114 }
3115 EXPORT_SYMBOL(read_extent_buffer_pages);
3116
3117 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
3118                         unsigned long start,
3119                         unsigned long len)
3120 {
3121         size_t cur;
3122         size_t offset;
3123         struct page *page;
3124         char *kaddr;
3125         char *dst = (char *)dstv;
3126         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3127         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3128
3129         WARN_ON(start > eb->len);
3130         WARN_ON(start + len > eb->start + eb->len);
3131
3132         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3133
3134         while(len > 0) {
3135                 page = extent_buffer_page(eb, i);
3136
3137                 cur = min(len, (PAGE_CACHE_SIZE - offset));
3138                 kaddr = kmap_atomic(page, KM_USER1);
3139                 memcpy(dst, kaddr + offset, cur);
3140                 kunmap_atomic(kaddr, KM_USER1);
3141
3142                 dst += cur;
3143                 len -= cur;
3144                 offset = 0;
3145                 i++;
3146         }
3147 }
3148 EXPORT_SYMBOL(read_extent_buffer);
3149
3150 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
3151                                unsigned long min_len, char **token, char **map,
3152                                unsigned long *map_start,
3153                                unsigned long *map_len, int km)
3154 {
3155         size_t offset = start & (PAGE_CACHE_SIZE - 1);
3156         char *kaddr;
3157         struct page *p;
3158         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3159         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3160         unsigned long end_i = (start_offset + start + min_len - 1) >>
3161                 PAGE_CACHE_SHIFT;
3162
3163         if (i != end_i)
3164                 return -EINVAL;
3165
3166         if (i == 0) {
3167                 offset = start_offset;
3168                 *map_start = 0;
3169         } else {
3170                 offset = 0;
3171                 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
3172         }
3173         if (start + min_len > eb->len) {
3174 printk("bad mapping eb start %Lu len %lu, wanted %lu %lu\n", eb->start, eb->len, start, min_len);
3175                 WARN_ON(1);
3176         }
3177
3178         p = extent_buffer_page(eb, i);
3179         kaddr = kmap_atomic(p, km);
3180         *token = kaddr;
3181         *map = kaddr + offset;
3182         *map_len = PAGE_CACHE_SIZE - offset;
3183         return 0;
3184 }
3185 EXPORT_SYMBOL(map_private_extent_buffer);
3186
3187 int map_extent_buffer(struct extent_buffer *eb, unsigned long start,
3188                       unsigned long min_len,
3189                       char **token, char **map,
3190                       unsigned long *map_start,
3191                       unsigned long *map_len, int km)
3192 {
3193         int err;
3194         int save = 0;
3195         if (eb->map_token) {
3196                 unmap_extent_buffer(eb, eb->map_token, km);
3197                 eb->map_token = NULL;
3198                 save = 1;
3199         }
3200         err = map_private_extent_buffer(eb, start, min_len, token, map,
3201                                        map_start, map_len, km);
3202         if (!err && save) {
3203                 eb->map_token = *token;
3204                 eb->kaddr = *map;
3205                 eb->map_start = *map_start;
3206                 eb->map_len = *map_len;
3207         }
3208         return err;
3209 }
3210 EXPORT_SYMBOL(map_extent_buffer);
3211
3212 void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km)
3213 {
3214         kunmap_atomic(token, km);
3215 }
3216 EXPORT_SYMBOL(unmap_extent_buffer);
3217
3218 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
3219                           unsigned long start,
3220                           unsigned long len)
3221 {
3222         size_t cur;
3223         size_t offset;
3224         struct page *page;
3225         char *kaddr;
3226         char *ptr = (char *)ptrv;
3227         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3228         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3229         int ret = 0;
3230
3231         WARN_ON(start > eb->len);
3232         WARN_ON(start + len > eb->start + eb->len);
3233
3234         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3235
3236         while(len > 0) {
3237                 page = extent_buffer_page(eb, i);
3238
3239                 cur = min(len, (PAGE_CACHE_SIZE - offset));
3240
3241                 kaddr = kmap_atomic(page, KM_USER0);
3242                 ret = memcmp(ptr, kaddr + offset, cur);
3243                 kunmap_atomic(kaddr, KM_USER0);
3244                 if (ret)
3245                         break;
3246
3247                 ptr += cur;
3248                 len -= cur;
3249                 offset = 0;
3250                 i++;
3251         }
3252         return ret;
3253 }
3254 EXPORT_SYMBOL(memcmp_extent_buffer);
3255
3256 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
3257                          unsigned long start, unsigned long len)
3258 {
3259         size_t cur;
3260         size_t offset;
3261         struct page *page;
3262         char *kaddr;
3263         char *src = (char *)srcv;
3264         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3265         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3266
3267         WARN_ON(start > eb->len);
3268         WARN_ON(start + len > eb->start + eb->len);
3269
3270         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3271
3272         while(len > 0) {
3273                 page = extent_buffer_page(eb, i);
3274                 WARN_ON(!PageUptodate(page));
3275
3276                 cur = min(len, PAGE_CACHE_SIZE - offset);
3277                 kaddr = kmap_atomic(page, KM_USER1);
3278                 memcpy(kaddr + offset, src, cur);
3279                 kunmap_atomic(kaddr, KM_USER1);
3280
3281                 src += cur;
3282                 len -= cur;
3283                 offset = 0;
3284                 i++;
3285         }
3286 }
3287 EXPORT_SYMBOL(write_extent_buffer);
3288
3289 void memset_extent_buffer(struct extent_buffer *eb, char c,
3290                           unsigned long start, unsigned long len)
3291 {
3292         size_t cur;
3293         size_t offset;
3294         struct page *page;
3295         char *kaddr;
3296         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3297         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3298
3299         WARN_ON(start > eb->len);
3300         WARN_ON(start + len > eb->start + eb->len);
3301
3302         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3303
3304         while(len > 0) {
3305                 page = extent_buffer_page(eb, i);
3306                 WARN_ON(!PageUptodate(page));
3307
3308                 cur = min(len, PAGE_CACHE_SIZE - offset);
3309                 kaddr = kmap_atomic(page, KM_USER0);
3310                 memset(kaddr + offset, c, cur);
3311                 kunmap_atomic(kaddr, KM_USER0);
3312
3313                 len -= cur;
3314                 offset = 0;
3315                 i++;
3316         }
3317 }
3318 EXPORT_SYMBOL(memset_extent_buffer);
3319
3320 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
3321                         unsigned long dst_offset, unsigned long src_offset,
3322                         unsigned long len)
3323 {
3324         u64 dst_len = dst->len;
3325         size_t cur;
3326         size_t offset;
3327         struct page *page;
3328         char *kaddr;
3329         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3330         unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3331
3332         WARN_ON(src->len != dst_len);
3333
3334         offset = (start_offset + dst_offset) &
3335                 ((unsigned long)PAGE_CACHE_SIZE - 1);
3336
3337         while(len > 0) {
3338                 page = extent_buffer_page(dst, i);
3339                 WARN_ON(!PageUptodate(page));
3340
3341                 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
3342
3343                 kaddr = kmap_atomic(page, KM_USER0);
3344                 read_extent_buffer(src, kaddr + offset, src_offset, cur);
3345                 kunmap_atomic(kaddr, KM_USER0);
3346
3347                 src_offset += cur;
3348                 len -= cur;
3349                 offset = 0;
3350                 i++;
3351         }
3352 }
3353 EXPORT_SYMBOL(copy_extent_buffer);
3354
3355 static void move_pages(struct page *dst_page, struct page *src_page,
3356                        unsigned long dst_off, unsigned long src_off,
3357                        unsigned long len)
3358 {
3359         char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3360         if (dst_page == src_page) {
3361                 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
3362         } else {
3363                 char *src_kaddr = kmap_atomic(src_page, KM_USER1);
3364                 char *p = dst_kaddr + dst_off + len;
3365                 char *s = src_kaddr + src_off + len;
3366
3367                 while (len--)
3368                         *--p = *--s;
3369
3370                 kunmap_atomic(src_kaddr, KM_USER1);
3371         }
3372         kunmap_atomic(dst_kaddr, KM_USER0);
3373 }
3374
3375 static void copy_pages(struct page *dst_page, struct page *src_page,
3376                        unsigned long dst_off, unsigned long src_off,
3377                        unsigned long len)
3378 {
3379         char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3380         char *src_kaddr;
3381
3382         if (dst_page != src_page)
3383                 src_kaddr = kmap_atomic(src_page, KM_USER1);
3384         else
3385                 src_kaddr = dst_kaddr;
3386
3387         memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
3388         kunmap_atomic(dst_kaddr, KM_USER0);
3389         if (dst_page != src_page)
3390                 kunmap_atomic(src_kaddr, KM_USER1);
3391 }
3392
3393 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3394                            unsigned long src_offset, unsigned long len)
3395 {
3396         size_t cur;
3397         size_t dst_off_in_page;
3398         size_t src_off_in_page;
3399         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3400         unsigned long dst_i;
3401         unsigned long src_i;
3402
3403         if (src_offset + len > dst->len) {
3404                 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
3405                        src_offset, len, dst->len);
3406                 BUG_ON(1);
3407         }
3408         if (dst_offset + len > dst->len) {
3409                 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
3410                        dst_offset, len, dst->len);
3411                 BUG_ON(1);
3412         }
3413
3414         while(len > 0) {
3415                 dst_off_in_page = (start_offset + dst_offset) &
3416                         ((unsigned long)PAGE_CACHE_SIZE - 1);
3417                 src_off_in_page = (start_offset + src_offset) &
3418                         ((unsigned long)PAGE_CACHE_SIZE - 1);
3419
3420                 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3421                 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
3422
3423                 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
3424                                                src_off_in_page));
3425                 cur = min_t(unsigned long, cur,
3426                         (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
3427
3428                 copy_pages(extent_buffer_page(dst, dst_i),
3429                            extent_buffer_page(dst, src_i),
3430                            dst_off_in_page, src_off_in_page, cur);
3431
3432                 src_offset += cur;
3433                 dst_offset += cur;
3434                 len -= cur;
3435         }
3436 }
3437 EXPORT_SYMBOL(memcpy_extent_buffer);
3438
3439 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3440                            unsigned long src_offset, unsigned long len)
3441 {
3442         size_t cur;
3443         size_t dst_off_in_page;
3444         size_t src_off_in_page;
3445         unsigned long dst_end = dst_offset + len - 1;
3446         unsigned long src_end = src_offset + len - 1;
3447         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3448         unsigned long dst_i;
3449         unsigned long src_i;
3450
3451         if (src_offset + len > dst->len) {
3452                 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
3453                        src_offset, len, dst->len);
3454                 BUG_ON(1);
3455         }
3456         if (dst_offset + len > dst->len) {
3457                 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
3458                        dst_offset, len, dst->len);
3459                 BUG_ON(1);
3460         }
3461         if (dst_offset < src_offset) {
3462                 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
3463                 return;
3464         }
3465         while(len > 0) {
3466                 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
3467                 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
3468
3469                 dst_off_in_page = (start_offset + dst_end) &
3470                         ((unsigned long)PAGE_CACHE_SIZE - 1);
3471                 src_off_in_page = (start_offset + src_end) &
3472                         ((unsigned long)PAGE_CACHE_SIZE - 1);
3473
3474                 cur = min_t(unsigned long, len, src_off_in_page + 1);
3475                 cur = min(cur, dst_off_in_page + 1);
3476                 move_pages(extent_buffer_page(dst, dst_i),
3477                            extent_buffer_page(dst, src_i),
3478                            dst_off_in_page - cur + 1,
3479                            src_off_in_page - cur + 1, cur);
3480
3481                 dst_end -= cur;
3482                 src_end -= cur;
3483                 len -= cur;
3484         }
3485 }
3486 EXPORT_SYMBOL(memmove_extent_buffer);