git.oblomov.eu Git - linux-2.6/blob - fs/btrfs/extent_io.c

   1 #include <linux/bitops.h>
   2 #include <linux/slab.h>
   3 #include <linux/bio.h>
   4 #include <linux/mm.h>
   5 #include <linux/gfp.h>
   6 #include <linux/pagemap.h>
   7 #include <linux/page-flags.h>
   8 #include <linux/module.h>
   9 #include <linux/spinlock.h>
  10 #include <linux/blkdev.h>
  11 #include <linux/swap.h>
  12 #include <linux/version.h>
  13 #include <linux/writeback.h>
  14 #include <linux/pagevec.h>
  15 #include "extent_io.h"
  16 #include "extent_map.h"
  17
  18 /* temporary define until extent_map moves out of btrfs */
  19 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
  20                                        unsigned long extra_flags,
  21                                        void (*ctor)(void *, struct kmem_cache *,
  22                                                     unsigned long));
  23
  24 static struct kmem_cache *extent_state_cache;
  25 static struct kmem_cache *extent_buffer_cache;
  26
  27 static LIST_HEAD(buffers);
  28 static LIST_HEAD(states);
  29 static spinlock_t leak_lock = SPIN_LOCK_UNLOCKED;
  30
  31 #define BUFFER_LRU_MAX 64
  32
  33 struct tree_entry {
  34         u64 start;
  35         u64 end;
  36         struct rb_node rb_node;
  37 };
  38
  39 struct extent_page_data {
  40         struct bio *bio;
  41         struct extent_io_tree *tree;
  42         get_extent_t *get_extent;
  43 };
  44
  45 int __init extent_io_init(void)
  46 {
  47         extent_state_cache = btrfs_cache_create("extent_state",
  48                                             sizeof(struct extent_state), 0,
  49                                             NULL);
  50         if (!extent_state_cache)
  51                 return -ENOMEM;
  52
  53         extent_buffer_cache = btrfs_cache_create("extent_buffers",
  54                                             sizeof(struct extent_buffer), 0,
  55                                             NULL);
  56         if (!extent_buffer_cache)
  57                 goto free_state_cache;
  58         return 0;
  59
  60 free_state_cache:
  61         kmem_cache_destroy(extent_state_cache);
  62         return -ENOMEM;
  63 }
  64
  65 void extent_io_exit(void)
  66 {
  67         struct extent_state *state;
  68         struct extent_buffer *eb;
  69
  70         while (!list_empty(&states)) {
  71                 state = list_entry(states.next, struct extent_state, leak_list);
  72                 printk("state leak: start %Lu end %Lu state %lu in tree %p refs %d\n", state->start, state->end, state->state, state->tree, atomic_read(&state->refs));
  73                 list_del(&state->leak_list);
  74                 kmem_cache_free(extent_state_cache, state);
  75
  76         }
  77
  78         while (!list_empty(&buffers)) {
  79                 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
  80                 printk("buffer leak start %Lu len %lu refs %d\n", eb->start, eb->len, atomic_read(&eb->refs));
  81                 list_del(&eb->leak_list);
  82                 kmem_cache_free(extent_buffer_cache, eb);
  83         }
  84         if (extent_state_cache)
  85                 kmem_cache_destroy(extent_state_cache);
  86         if (extent_buffer_cache)
  87                 kmem_cache_destroy(extent_buffer_cache);
  88 }
  89
  90 void extent_io_tree_init(struct extent_io_tree *tree,
  91                           struct address_space *mapping, gfp_t mask)
  92 {
  93         tree->state.rb_node = NULL;
  94         tree->ops = NULL;
  95         tree->dirty_bytes = 0;
  96         spin_lock_init(&tree->lock);
  97         spin_lock_init(&tree->lru_lock);
  98         tree->mapping = mapping;
  99         INIT_LIST_HEAD(&tree->buffer_lru);
 100         tree->lru_size = 0;
 101         tree->last = NULL;
 102 }
 103 EXPORT_SYMBOL(extent_io_tree_init);
 104
 105 void extent_io_tree_empty_lru(struct extent_io_tree *tree)
 106 {
 107         struct extent_buffer *eb;
 108         while(!list_empty(&tree->buffer_lru)) {
 109                 eb = list_entry(tree->buffer_lru.next, struct extent_buffer,
 110                                 lru);
 111                 list_del_init(&eb->lru);
 112                 free_extent_buffer(eb);
 113         }
 114 }
 115 EXPORT_SYMBOL(extent_io_tree_empty_lru);
 116
 117 struct extent_state *alloc_extent_state(gfp_t mask)
 118 {
 119         struct extent_state *state;
 120         unsigned long flags;
 121
 122         state = kmem_cache_alloc(extent_state_cache, mask);
 123         if (!state)
 124                 return state;
 125         state->state = 0;
 126         state->private = 0;
 127         state->tree = NULL;
 128         spin_lock_irqsave(&leak_lock, flags);
 129         list_add(&state->leak_list, &states);
 130         spin_unlock_irqrestore(&leak_lock, flags);
 131
 132         atomic_set(&state->refs, 1);
 133         init_waitqueue_head(&state->wq);
 134         return state;
 135 }
 136 EXPORT_SYMBOL(alloc_extent_state);
 137
 138 void free_extent_state(struct extent_state *state)
 139 {
 140         if (!state)
 141                 return;
 142         if (atomic_dec_and_test(&state->refs)) {
 143                 unsigned long flags;
 144                 WARN_ON(state->tree);
 145                 spin_lock_irqsave(&leak_lock, flags);
 146                 list_del(&state->leak_list);
 147                 spin_unlock_irqrestore(&leak_lock, flags);
 148                 kmem_cache_free(extent_state_cache, state);
 149         }
 150 }
 151 EXPORT_SYMBOL(free_extent_state);
 152
 153 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
 154                                    struct rb_node *node)
 155 {
 156         struct rb_node ** p = &root->rb_node;
 157         struct rb_node * parent = NULL;
 158         struct tree_entry *entry;
 159
 160         while(*p) {
 161                 parent = *p;
 162                 entry = rb_entry(parent, struct tree_entry, rb_node);
 163
 164                 if (offset < entry->start)
 165                         p = &(*p)->rb_left;
 166                 else if (offset > entry->end)
 167                         p = &(*p)->rb_right;
 168                 else
 169                         return parent;
 170         }
 171
 172         entry = rb_entry(node, struct tree_entry, rb_node);
 173         rb_link_node(node, parent, p);
 174         rb_insert_color(node, root);
 175         return NULL;
 176 }
 177
 178 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
 179                                      struct rb_node **prev_ret,
 180                                      struct rb_node **next_ret)
 181 {
 182         struct rb_root *root = &tree->state;
 183         struct rb_node * n = root->rb_node;
 184         struct rb_node *prev = NULL;
 185         struct rb_node *orig_prev = NULL;
 186         struct tree_entry *entry;
 187         struct tree_entry *prev_entry = NULL;
 188
 189         if (tree->last) {
 190                 struct extent_state *state;
 191                 state = tree->last;
 192                 if (state->start <= offset && offset <= state->end)
 193                         return &tree->last->rb_node;
 194         }
 195         while(n) {
 196                 entry = rb_entry(n, struct tree_entry, rb_node);
 197                 prev = n;
 198                 prev_entry = entry;
 199
 200                 if (offset < entry->start)
 201                         n = n->rb_left;
 202                 else if (offset > entry->end)
 203                         n = n->rb_right;
 204                 else {
 205                         tree->last = rb_entry(n, struct extent_state, rb_node);
 206                         return n;
 207                 }
 208         }
 209
 210         if (prev_ret) {
 211                 orig_prev = prev;
 212                 while(prev && offset > prev_entry->end) {
 213                         prev = rb_next(prev);
 214                         prev_entry = rb_entry(prev, struct tree_entry, rb_node);
 215                 }
 216                 *prev_ret = prev;
 217                 prev = orig_prev;
 218         }
 219
 220         if (next_ret) {
 221                 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
 222                 while(prev && offset < prev_entry->start) {
 223                         prev = rb_prev(prev);
 224                         prev_entry = rb_entry(prev, struct tree_entry, rb_node);
 225                 }
 226                 *next_ret = prev;
 227         }
 228         return NULL;
 229 }
 230
 231 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
 232                                           u64 offset)
 233 {
 234         struct rb_node *prev = NULL;
 235         struct rb_node *ret;
 236
 237         ret = __etree_search(tree, offset, &prev, NULL);
 238         if (!ret) {
 239                 if (prev) {
 240                         tree->last = rb_entry(prev, struct extent_state,
 241                                               rb_node);
 242                 }
 243                 return prev;
 244         }
 245         return ret;
 246 }
 247
 248 /*
 249  * utility function to look for merge candidates inside a given range.
 250  * Any extents with matching state are merged together into a single
 251  * extent in the tree.  Extents with EXTENT_IO in their state field
 252  * are not merged because the end_io handlers need to be able to do
 253  * operations on them without sleeping (or doing allocations/splits).
 254  *
 255  * This should be called with the tree lock held.
 256  */
 257 static int merge_state(struct extent_io_tree *tree,
 258                        struct extent_state *state)
 259 {
 260         struct extent_state *other;
 261         struct rb_node *other_node;
 262
 263         if (state->state & EXTENT_IOBITS)
 264                 return 0;
 265
 266         other_node = rb_prev(&state->rb_node);
 267         if (other_node) {
 268                 other = rb_entry(other_node, struct extent_state, rb_node);
 269                 if (other->end == state->start - 1 &&
 270                     other->state == state->state) {
 271                         state->start = other->start;
 272                         other->tree = NULL;
 273                         if (tree->last == other)
 274                                 tree->last = state;
 275                         rb_erase(&other->rb_node, &tree->state);
 276                         free_extent_state(other);
 277                 }
 278         }
 279         other_node = rb_next(&state->rb_node);
 280         if (other_node) {
 281                 other = rb_entry(other_node, struct extent_state, rb_node);
 282                 if (other->start == state->end + 1 &&
 283                     other->state == state->state) {
 284                         other->start = state->start;
 285                         state->tree = NULL;
 286                         if (tree->last == state)
 287                                 tree->last = other;
 288                         rb_erase(&state->rb_node, &tree->state);
 289                         free_extent_state(state);
 290                 }
 291         }
 292         return 0;
 293 }
 294
 295 static void set_state_cb(struct extent_io_tree *tree,
 296                          struct extent_state *state,
 297                          unsigned long bits)
 298 {
 299         if (tree->ops && tree->ops->set_bit_hook) {
 300                 tree->ops->set_bit_hook(tree->mapping->host, state->start,
 301                                         state->end, state->state, bits);
 302         }
 303 }
 304
 305 static void clear_state_cb(struct extent_io_tree *tree,
 306                            struct extent_state *state,
 307                            unsigned long bits)
 308 {
 309         if (tree->ops && tree->ops->set_bit_hook) {
 310                 tree->ops->clear_bit_hook(tree->mapping->host, state->start,
 311                                           state->end, state->state, bits);
 312         }
 313 }
 314
 315 /*
 316  * insert an extent_state struct into the tree.  'bits' are set on the
 317  * struct before it is inserted.
 318  *
 319  * This may return -EEXIST if the extent is already there, in which case the
 320  * state struct is freed.
 321  *
 322  * The tree lock is not taken internally.  This is a utility function and
 323  * probably isn't what you want to call (see set/clear_extent_bit).
 324  */
 325 static int insert_state(struct extent_io_tree *tree,
 326                         struct extent_state *state, u64 start, u64 end,
 327                         int bits)
 328 {
 329         struct rb_node *node;
 330
 331         if (end < start) {
 332                 printk("end < start %Lu %Lu\n", end, start);
 333                 WARN_ON(1);
 334         }
 335         if (bits & EXTENT_DIRTY)
 336                 tree->dirty_bytes += end - start + 1;
 337         set_state_cb(tree, state, bits);
 338         state->state |= bits;
 339         state->start = start;
 340         state->end = end;
 341         node = tree_insert(&tree->state, end, &state->rb_node);
 342         if (node) {
 343                 struct extent_state *found;
 344                 found = rb_entry(node, struct extent_state, rb_node);
 345                 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, start, end);
 346                 free_extent_state(state);
 347                 return -EEXIST;
 348         }
 349         state->tree = tree;
 350         tree->last = state;
 351         merge_state(tree, state);
 352         return 0;
 353 }
 354
 355 /*
 356  * split a given extent state struct in two, inserting the preallocated
 357  * struct 'prealloc' as the newly created second half.  'split' indicates an
 358  * offset inside 'orig' where it should be split.
 359  *
 360  * Before calling,
 361  * the tree has 'orig' at [orig->start, orig->end].  After calling, there
 362  * are two extent state structs in the tree:
 363  * prealloc: [orig->start, split - 1]
 364  * orig: [ split, orig->end ]
 365  *
 366  * The tree locks are not taken by this function. They need to be held
 367  * by the caller.
 368  */
 369 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
 370                        struct extent_state *prealloc, u64 split)
 371 {
 372         struct rb_node *node;
 373         prealloc->start = orig->start;
 374         prealloc->end = split - 1;
 375         prealloc->state = orig->state;
 376         orig->start = split;
 377
 378         node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
 379         if (node) {
 380                 struct extent_state *found;
 381                 found = rb_entry(node, struct extent_state, rb_node);
 382                 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, prealloc->start, prealloc->end);
 383                 free_extent_state(prealloc);
 384                 return -EEXIST;
 385         }
 386         prealloc->tree = tree;
 387         return 0;
 388 }
 389
 390 /*
 391  * utility function to clear some bits in an extent state struct.
 392  * it will optionally wake up any one waiting on this state (wake == 1), or
 393  * forcibly remove the state from the tree (delete == 1).
 394  *
 395  * If no bits are set on the state struct after clearing things, the
 396  * struct is freed and removed from the tree
 397  */
 398 static int clear_state_bit(struct extent_io_tree *tree,
 399                             struct extent_state *state, int bits, int wake,
 400                             int delete)
 401 {
 402         int ret = state->state & bits;
 403
 404         if ((bits & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
 405                 u64 range = state->end - state->start + 1;
 406                 WARN_ON(range > tree->dirty_bytes);
 407                 tree->dirty_bytes -= range;
 408         }
 409         clear_state_cb(tree, state, bits);
 410         state->state &= ~bits;
 411         if (wake)
 412                 wake_up(&state->wq);
 413         if (delete || state->state == 0) {
 414                 if (state->tree) {
 415                         clear_state_cb(tree, state, state->state);
 416                         if (tree->last == state) {
 417                                 tree->last = extent_state_next(state);
 418                         }
 419                         rb_erase(&state->rb_node, &tree->state);
 420                         state->tree = NULL;
 421                         free_extent_state(state);
 422                 } else {
 423                         WARN_ON(1);
 424                 }
 425         } else {
 426                 merge_state(tree, state);
 427         }
 428         return ret;
 429 }
 430
 431 /*
 432  * clear some bits on a range in the tree.  This may require splitting
 433  * or inserting elements in the tree, so the gfp mask is used to
 434  * indicate which allocations or sleeping are allowed.
 435  *
 436  * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
 437  * the given range from the tree regardless of state (ie for truncate).
 438  *
 439  * the range [start, end] is inclusive.
 440  *
 441  * This takes the tree lock, and returns < 0 on error, > 0 if any of the
 442  * bits were already set, or zero if none of the bits were already set.
 443  */
 444 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
 445                      int bits, int wake, int delete, gfp_t mask)
 446 {
 447         struct extent_state *state;
 448         struct extent_state *prealloc = NULL;
 449         struct rb_node *node;
 450         unsigned long flags;
 451         int err;
 452         int set = 0;
 453
 454 again:
 455         if (!prealloc && (mask & __GFP_WAIT)) {
 456                 prealloc = alloc_extent_state(mask);
 457                 if (!prealloc)
 458                         return -ENOMEM;
 459         }
 460
 461         spin_lock_irqsave(&tree->lock, flags);
 462         /*
 463          * this search will find the extents that end after
 464          * our range starts
 465          */
 466         node = tree_search(tree, start);
 467         if (!node)
 468                 goto out;
 469         state = rb_entry(node, struct extent_state, rb_node);
 470         if (state->start > end)
 471                 goto out;
 472         WARN_ON(state->end < start);
 473
 474         /*
 475          *     | ---- desired range ---- |
 476          *  | state | or
 477          *  | ------------- state -------------- |
 478          *
 479          * We need to split the extent we found, and may flip
 480          * bits on second half.
 481          *
 482          * If the extent we found extends past our range, we
 483          * just split and search again.  It'll get split again
 484          * the next time though.
 485          *
 486          * If the extent we found is inside our range, we clear
 487          * the desired bit on it.
 488          */
 489
 490         if (state->start < start) {
 491                 if (!prealloc)
 492                         prealloc = alloc_extent_state(GFP_ATOMIC);
 493                 err = split_state(tree, state, prealloc, start);
 494                 BUG_ON(err == -EEXIST);
 495                 prealloc = NULL;
 496                 if (err)
 497                         goto out;
 498                 if (state->end <= end) {
 499                         start = state->end + 1;
 500                         set |= clear_state_bit(tree, state, bits,
 501                                         wake, delete);
 502                 } else {
 503                         start = state->start;
 504                 }
 505                 goto search_again;
 506         }
 507         /*
 508          * | ---- desired range ---- |
 509          *                        | state |
 510          * We need to split the extent, and clear the bit
 511          * on the first half
 512          */
 513         if (state->start <= end && state->end > end) {
 514                 if (!prealloc)
 515                         prealloc = alloc_extent_state(GFP_ATOMIC);
 516                 err = split_state(tree, state, prealloc, end + 1);
 517                 BUG_ON(err == -EEXIST);
 518
 519                 if (wake)
 520                         wake_up(&state->wq);
 521                 set |= clear_state_bit(tree, prealloc, bits,
 522                                        wake, delete);
 523                 prealloc = NULL;
 524                 goto out;
 525         }
 526
 527         start = state->end + 1;
 528         set |= clear_state_bit(tree, state, bits, wake, delete);
 529         goto search_again;
 530
 531 out:
 532         spin_unlock_irqrestore(&tree->lock, flags);
 533         if (prealloc)
 534                 free_extent_state(prealloc);
 535
 536         return set;
 537
 538 search_again:
 539         if (start > end)
 540                 goto out;
 541         spin_unlock_irqrestore(&tree->lock, flags);
 542         if (mask & __GFP_WAIT)
 543                 cond_resched();
 544         goto again;
 545 }
 546 EXPORT_SYMBOL(clear_extent_bit);
 547
 548 static int wait_on_state(struct extent_io_tree *tree,
 549                          struct extent_state *state)
 550 {
 551         DEFINE_WAIT(wait);
 552         prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
 553         spin_unlock_irq(&tree->lock);
 554         schedule();
 555         spin_lock_irq(&tree->lock);
 556         finish_wait(&state->wq, &wait);
 557         return 0;
 558 }
 559
 560 /*
 561  * waits for one or more bits to clear on a range in the state tree.
 562  * The range [start, end] is inclusive.
 563  * The tree lock is taken by this function
 564  */
 565 int wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits)
 566 {
 567         struct extent_state *state;
 568         struct rb_node *node;
 569
 570         spin_lock_irq(&tree->lock);
 571 again:
 572         while (1) {
 573                 /*
 574                  * this search will find all the extents that end after
 575                  * our range starts
 576                  */
 577                 node = tree_search(tree, start);
 578                 if (!node)
 579                         break;
 580
 581                 state = rb_entry(node, struct extent_state, rb_node);
 582
 583                 if (state->start > end)
 584                         goto out;
 585
 586                 if (state->state & bits) {
 587                         start = state->start;
 588                         atomic_inc(&state->refs);
 589                         wait_on_state(tree, state);
 590                         free_extent_state(state);
 591                         goto again;
 592                 }
 593                 start = state->end + 1;
 594
 595                 if (start > end)
 596                         break;
 597
 598                 if (need_resched()) {
 599                         spin_unlock_irq(&tree->lock);
 600                         cond_resched();
 601                         spin_lock_irq(&tree->lock);
 602                 }
 603         }
 604 out:
 605         spin_unlock_irq(&tree->lock);
 606         return 0;
 607 }
 608 EXPORT_SYMBOL(wait_extent_bit);
 609
 610 static void set_state_bits(struct extent_io_tree *tree,
 611                            struct extent_state *state,
 612                            int bits)
 613 {
 614         if ((bits & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
 615                 u64 range = state->end - state->start + 1;
 616                 tree->dirty_bytes += range;
 617         }
 618         set_state_cb(tree, state, bits);
 619         state->state |= bits;
 620 }
 621
 622 /*
 623  * set some bits on a range in the tree.  This may require allocations
 624  * or sleeping, so the gfp mask is used to indicate what is allowed.
 625  *
 626  * If 'exclusive' == 1, this will fail with -EEXIST if some part of the
 627  * range already has the desired bits set.  The start of the existing
 628  * range is returned in failed_start in this case.
 629  *
 630  * [start, end] is inclusive
 631  * This takes the tree lock.
 632  */
 633 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits,
 634                    int exclusive, u64 *failed_start, gfp_t mask)
 635 {
 636         struct extent_state *state;
 637         struct extent_state *prealloc = NULL;
 638         struct rb_node *node;
 639         unsigned long flags;
 640         int err = 0;
 641         int set;
 642         u64 last_start;
 643         u64 last_end;
 644 again:
 645         if (!prealloc && (mask & __GFP_WAIT)) {
 646                 prealloc = alloc_extent_state(mask);
 647                 if (!prealloc)
 648                         return -ENOMEM;
 649         }
 650
 651         spin_lock_irqsave(&tree->lock, flags);
 652         /*
 653          * this search will find all the extents that end after
 654          * our range starts.
 655          */
 656         node = tree_search(tree, start);
 657         if (!node) {
 658                 err = insert_state(tree, prealloc, start, end, bits);
 659                 prealloc = NULL;
 660                 BUG_ON(err == -EEXIST);
 661                 goto out;
 662         }
 663
 664         state = rb_entry(node, struct extent_state, rb_node);
 665         last_start = state->start;
 666         last_end = state->end;
 667
 668         /*
 669          * | ---- desired range ---- |
 670          * | state |
 671          *
 672          * Just lock what we found and keep going
 673          */
 674         if (state->start == start && state->end <= end) {
 675                 set = state->state & bits;
 676                 if (set && exclusive) {
 677                         *failed_start = state->start;
 678                         err = -EEXIST;
 679                         goto out;
 680                 }
 681                 set_state_bits(tree, state, bits);
 682                 start = state->end + 1;
 683                 merge_state(tree, state);
 684                 goto search_again;
 685         }
 686
 687         /*
 688          *     | ---- desired range ---- |
 689          * | state |
 690          *   or
 691          * | ------------- state -------------- |
 692          *
 693          * We need to split the extent we found, and may flip bits on
 694          * second half.
 695          *
 696          * If the extent we found extends past our
 697          * range, we just split and search again.  It'll get split
 698          * again the next time though.
 699          *
 700          * If the extent we found is inside our range, we set the
 701          * desired bit on it.
 702          */
 703         if (state->start < start) {
 704                 set = state->state & bits;
 705                 if (exclusive && set) {
 706                         *failed_start = start;
 707                         err = -EEXIST;
 708                         goto out;
 709                 }
 710                 err = split_state(tree, state, prealloc, start);
 711                 BUG_ON(err == -EEXIST);
 712                 prealloc = NULL;
 713                 if (err)
 714                         goto out;
 715                 if (state->end <= end) {
 716                         set_state_bits(tree, state, bits);
 717                         start = state->end + 1;
 718                         merge_state(tree, state);
 719                 } else {
 720                         start = state->start;
 721                 }
 722                 goto search_again;
 723         }
 724         /*
 725          * | ---- desired range ---- |
 726          *     | state | or               | state |
 727          *
 728          * There's a hole, we need to insert something in it and
 729          * ignore the extent we found.
 730          */
 731         if (state->start > start) {
 732                 u64 this_end;
 733                 if (end < last_start)
 734                         this_end = end;
 735                 else
 736                         this_end = last_start -1;
 737                 err = insert_state(tree, prealloc, start, this_end,
 738                                    bits);
 739                 prealloc = NULL;
 740                 BUG_ON(err == -EEXIST);
 741                 if (err)
 742                         goto out;
 743                 start = this_end + 1;
 744                 goto search_again;
 745         }
 746         /*
 747          * | ---- desired range ---- |
 748          *                        | state |
 749          * We need to split the extent, and set the bit
 750          * on the first half
 751          */
 752         if (state->start <= end && state->end > end) {
 753                 set = state->state & bits;
 754                 if (exclusive && set) {
 755                         *failed_start = start;
 756                         err = -EEXIST;
 757                         goto out;
 758                 }
 759                 err = split_state(tree, state, prealloc, end + 1);
 760                 BUG_ON(err == -EEXIST);
 761
 762                 set_state_bits(tree, prealloc, bits);
 763                 merge_state(tree, prealloc);
 764                 prealloc = NULL;
 765                 goto out;
 766         }
 767
 768         goto search_again;
 769
 770 out:
 771         spin_unlock_irqrestore(&tree->lock, flags);
 772         if (prealloc)
 773                 free_extent_state(prealloc);
 774
 775         return err;
 776
 777 search_again:
 778         if (start > end)
 779                 goto out;
 780         spin_unlock_irqrestore(&tree->lock, flags);
 781         if (mask & __GFP_WAIT)
 782                 cond_resched();
 783         goto again;
 784 }
 785 EXPORT_SYMBOL(set_extent_bit);
 786
 787 /* wrappers around set/clear extent bit */
 788 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
 789                      gfp_t mask)
 790 {
 791         return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL,
 792                               mask);
 793 }
 794 EXPORT_SYMBOL(set_extent_dirty);
 795
 796 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
 797                     int bits, gfp_t mask)
 798 {
 799         return set_extent_bit(tree, start, end, bits, 0, NULL,
 800                               mask);
 801 }
 802 EXPORT_SYMBOL(set_extent_bits);
 803
 804 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
 805                       int bits, gfp_t mask)
 806 {
 807         return clear_extent_bit(tree, start, end, bits, 0, 0, mask);
 808 }
 809 EXPORT_SYMBOL(clear_extent_bits);
 810
 811 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
 812                      gfp_t mask)
 813 {
 814         return set_extent_bit(tree, start, end,
 815                               EXTENT_DELALLOC | EXTENT_DIRTY, 0, NULL,
 816                               mask);
 817 }
 818 EXPORT_SYMBOL(set_extent_delalloc);
 819
 820 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
 821                        gfp_t mask)
 822 {
 823         return clear_extent_bit(tree, start, end,
 824                                 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, mask);
 825 }
 826 EXPORT_SYMBOL(clear_extent_dirty);
 827
 828 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
 829                      gfp_t mask)
 830 {
 831         return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL,
 832                               mask);
 833 }
 834 EXPORT_SYMBOL(set_extent_new);
 835
 836 int clear_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
 837                        gfp_t mask)
 838 {
 839         return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0, mask);
 840 }
 841 EXPORT_SYMBOL(clear_extent_new);
 842
 843 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
 844                         gfp_t mask)
 845 {
 846         return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, NULL,
 847                               mask);
 848 }
 849 EXPORT_SYMBOL(set_extent_uptodate);
 850
 851 int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
 852                           gfp_t mask)
 853 {
 854         return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, mask);
 855 }
 856 EXPORT_SYMBOL(clear_extent_uptodate);
 857
 858 int set_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end,
 859                          gfp_t mask)
 860 {
 861         return set_extent_bit(tree, start, end, EXTENT_WRITEBACK,
 862                               0, NULL, mask);
 863 }
 864 EXPORT_SYMBOL(set_extent_writeback);
 865
 866 int clear_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end,
 867                            gfp_t mask)
 868 {
 869         return clear_extent_bit(tree, start, end, EXTENT_WRITEBACK, 1, 0, mask);
 870 }
 871 EXPORT_SYMBOL(clear_extent_writeback);
 872
 873 int wait_on_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end)
 874 {
 875         return wait_extent_bit(tree, start, end, EXTENT_WRITEBACK);
 876 }
 877 EXPORT_SYMBOL(wait_on_extent_writeback);
 878
 879 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
 880 {
 881         int err;
 882         u64 failed_start;
 883         while (1) {
 884                 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1,
 885                                      &failed_start, mask);
 886                 if (err == -EEXIST && (mask & __GFP_WAIT)) {
 887                         wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
 888                         start = failed_start;
 889                 } else {
 890                         break;
 891                 }
 892                 WARN_ON(start > end);
 893         }
 894         return err;
 895 }
 896 EXPORT_SYMBOL(lock_extent);
 897
 898 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end,
 899                   gfp_t mask)
 900 {
 901         return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, mask);
 902 }
 903 EXPORT_SYMBOL(unlock_extent);
 904
 905 /*
 906  * helper function to set pages and extents in the tree dirty
 907  */
 908 int set_range_dirty(struct extent_io_tree *tree, u64 start, u64 end)
 909 {
 910         unsigned long index = start >> PAGE_CACHE_SHIFT;
 911         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
 912         struct page *page;
 913
 914         while (index <= end_index) {
 915                 page = find_get_page(tree->mapping, index);
 916                 BUG_ON(!page);
 917                 __set_page_dirty_nobuffers(page);
 918                 page_cache_release(page);
 919                 index++;
 920         }
 921         set_extent_dirty(tree, start, end, GFP_NOFS);
 922         return 0;
 923 }
 924 EXPORT_SYMBOL(set_range_dirty);
 925
 926 /*
 927  * helper function to set both pages and extents in the tree writeback
 928  */
 929 int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
 930 {
 931         unsigned long index = start >> PAGE_CACHE_SHIFT;
 932         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
 933         struct page *page;
 934
 935         while (index <= end_index) {
 936                 page = find_get_page(tree->mapping, index);
 937                 BUG_ON(!page);
 938                 set_page_writeback(page);
 939                 page_cache_release(page);
 940                 index++;
 941         }
 942         set_extent_writeback(tree, start, end, GFP_NOFS);
 943         return 0;
 944 }
 945 EXPORT_SYMBOL(set_range_writeback);
 946
 947 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
 948                           u64 *start_ret, u64 *end_ret, int bits)
 949 {
 950         struct rb_node *node;
 951         struct extent_state *state;
 952         int ret = 1;
 953
 954         spin_lock_irq(&tree->lock);
 955         /*
 956          * this search will find all the extents that end after
 957          * our range starts.
 958          */
 959         node = tree_search(tree, start);
 960         if (!node) {
 961                 goto out;
 962         }
 963
 964         while(1) {
 965                 state = rb_entry(node, struct extent_state, rb_node);
 966                 if (state->end >= start && (state->state & bits)) {
 967                         *start_ret = state->start;
 968                         *end_ret = state->end;
 969                         ret = 0;
 970                         break;
 971                 }
 972                 node = rb_next(node);
 973                 if (!node)
 974                         break;
 975         }
 976 out:
 977         spin_unlock_irq(&tree->lock);
 978         return ret;
 979 }
 980 EXPORT_SYMBOL(find_first_extent_bit);
 981
 982 struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree,
 983                                                  u64 start, int bits)
 984 {
 985         struct rb_node *node;
 986         struct extent_state *state;
 987
 988         /*
 989          * this search will find all the extents that end after
 990          * our range starts.
 991          */
 992         node = tree_search(tree, start);
 993         if (!node) {
 994                 goto out;
 995         }
 996
 997         while(1) {
 998                 state = rb_entry(node, struct extent_state, rb_node);
 999                 if (state->end >= start && (state->state & bits)) {
1000                         return state;
1001                 }
1002                 node = rb_next(node);
1003                 if (!node)
1004                         break;
1005         }
1006 out:
1007         return NULL;
1008 }
1009 EXPORT_SYMBOL(find_first_extent_bit_state);
1010
1011 u64 find_lock_delalloc_range(struct extent_io_tree *tree,
1012                              u64 *start, u64 *end, u64 max_bytes)
1013 {
1014         struct rb_node *node;
1015         struct extent_state *state;
1016         u64 cur_start = *start;
1017         u64 found = 0;
1018         u64 total_bytes = 0;
1019
1020         spin_lock_irq(&tree->lock);
1021         /*
1022          * this search will find all the extents that end after
1023          * our range starts.
1024          */
1025 search_again:
1026         node = tree_search(tree, cur_start);
1027         if (!node) {
1028                 *end = (u64)-1;
1029                 goto out;
1030         }
1031
1032         while(1) {
1033                 state = rb_entry(node, struct extent_state, rb_node);
1034                 if (found && state->start != cur_start) {
1035                         goto out;
1036                 }
1037                 if (!(state->state & EXTENT_DELALLOC)) {
1038                         if (!found)
1039                                 *end = state->end;
1040                         goto out;
1041                 }
1042                 if (!found) {
1043                         struct extent_state *prev_state;
1044                         struct rb_node *prev_node = node;
1045                         while(1) {
1046                                 prev_node = rb_prev(prev_node);
1047                                 if (!prev_node)
1048                                         break;
1049                                 prev_state = rb_entry(prev_node,
1050                                                       struct extent_state,
1051                                                       rb_node);
1052                                 if (!(prev_state->state & EXTENT_DELALLOC))
1053                                         break;
1054                                 state = prev_state;
1055                                 node = prev_node;
1056                         }
1057                 }
1058                 if (state->state & EXTENT_LOCKED) {
1059                         DEFINE_WAIT(wait);
1060                         atomic_inc(&state->refs);
1061                         prepare_to_wait(&state->wq, &wait,
1062                                         TASK_UNINTERRUPTIBLE);
1063                         spin_unlock_irq(&tree->lock);
1064                         schedule();
1065                         spin_lock_irq(&tree->lock);
1066                         finish_wait(&state->wq, &wait);
1067                         free_extent_state(state);
1068                         goto search_again;
1069                 }
1070                 set_state_cb(tree, state, EXTENT_LOCKED);
1071                 state->state |= EXTENT_LOCKED;
1072                 if (!found)
1073                         *start = state->start;
1074                 found++;
1075                 *end = state->end;
1076                 cur_start = state->end + 1;
1077                 node = rb_next(node);
1078                 if (!node)
1079                         break;
1080                 total_bytes += state->end - state->start + 1;
1081                 if (total_bytes >= max_bytes)
1082                         break;
1083         }
1084 out:
1085         spin_unlock_irq(&tree->lock);
1086         return found;
1087 }
1088
1089 u64 count_range_bits(struct extent_io_tree *tree,
1090                      u64 *start, u64 search_end, u64 max_bytes,
1091                      unsigned long bits)
1092 {
1093         struct rb_node *node;
1094         struct extent_state *state;
1095         u64 cur_start = *start;
1096         u64 total_bytes = 0;
1097         int found = 0;
1098
1099         if (search_end <= cur_start) {
1100                 printk("search_end %Lu start %Lu\n", search_end, cur_start);
1101                 WARN_ON(1);
1102                 return 0;
1103         }
1104
1105         spin_lock_irq(&tree->lock);
1106         if (cur_start == 0 && bits == EXTENT_DIRTY) {
1107                 total_bytes = tree->dirty_bytes;
1108                 goto out;
1109         }
1110         /*
1111          * this search will find all the extents that end after
1112          * our range starts.
1113          */
1114         node = tree_search(tree, cur_start);
1115         if (!node) {
1116                 goto out;
1117         }
1118
1119         while(1) {
1120                 state = rb_entry(node, struct extent_state, rb_node);
1121                 if (state->start > search_end)
1122                         break;
1123                 if (state->end >= cur_start && (state->state & bits)) {
1124                         total_bytes += min(search_end, state->end) + 1 -
1125                                        max(cur_start, state->start);
1126                         if (total_bytes >= max_bytes)
1127                                 break;
1128                         if (!found) {
1129                                 *start = state->start;
1130                                 found = 1;
1131                         }
1132                 }
1133                 node = rb_next(node);
1134                 if (!node)
1135                         break;
1136         }
1137 out:
1138         spin_unlock_irq(&tree->lock);
1139         return total_bytes;
1140 }
1141 /*
1142  * helper function to lock both pages and extents in the tree.
1143  * pages must be locked first.
1144  */
1145 int lock_range(struct extent_io_tree *tree, u64 start, u64 end)
1146 {
1147         unsigned long index = start >> PAGE_CACHE_SHIFT;
1148         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1149         struct page *page;
1150         int err;
1151
1152         while (index <= end_index) {
1153                 page = grab_cache_page(tree->mapping, index);
1154                 if (!page) {
1155                         err = -ENOMEM;
1156                         goto failed;
1157                 }
1158                 if (IS_ERR(page)) {
1159                         err = PTR_ERR(page);
1160                         goto failed;
1161                 }
1162                 index++;
1163         }
1164         lock_extent(tree, start, end, GFP_NOFS);
1165         return 0;
1166
1167 failed:
1168         /*
1169          * we failed above in getting the page at 'index', so we undo here
1170          * up to but not including the page at 'index'
1171          */
1172         end_index = index;
1173         index = start >> PAGE_CACHE_SHIFT;
1174         while (index < end_index) {
1175                 page = find_get_page(tree->mapping, index);
1176                 unlock_page(page);
1177                 page_cache_release(page);
1178                 index++;
1179         }
1180         return err;
1181 }
1182 EXPORT_SYMBOL(lock_range);
1183
1184 /*
1185  * helper function to unlock both pages and extents in the tree.
1186  */
1187 int unlock_range(struct extent_io_tree *tree, u64 start, u64 end)
1188 {
1189         unsigned long index = start >> PAGE_CACHE_SHIFT;
1190         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1191         struct page *page;
1192
1193         while (index <= end_index) {
1194                 page = find_get_page(tree->mapping, index);
1195                 unlock_page(page);
1196                 page_cache_release(page);
1197                 index++;
1198         }
1199         unlock_extent(tree, start, end, GFP_NOFS);
1200         return 0;
1201 }
1202 EXPORT_SYMBOL(unlock_range);
1203
1204 int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1205 {
1206         struct rb_node *node;
1207         struct extent_state *state;
1208         int ret = 0;
1209
1210         spin_lock_irq(&tree->lock);
1211         /*
1212          * this search will find all the extents that end after
1213          * our range starts.
1214          */
1215         node = tree_search(tree, start);
1216         if (!node) {
1217                 ret = -ENOENT;
1218                 goto out;
1219         }
1220         state = rb_entry(node, struct extent_state, rb_node);
1221         if (state->start != start) {
1222                 ret = -ENOENT;
1223                 goto out;
1224         }
1225         state->private = private;
1226 out:
1227         spin_unlock_irq(&tree->lock);
1228         return ret;
1229 }
1230
1231 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1232 {
1233         struct rb_node *node;
1234         struct extent_state *state;
1235         int ret = 0;
1236
1237         spin_lock_irq(&tree->lock);
1238         /*
1239          * this search will find all the extents that end after
1240          * our range starts.
1241          */
1242         node = tree_search(tree, start);
1243         if (!node) {
1244                 ret = -ENOENT;
1245                 goto out;
1246         }
1247         state = rb_entry(node, struct extent_state, rb_node);
1248         if (state->start != start) {
1249                 ret = -ENOENT;
1250                 goto out;
1251         }
1252         *private = state->private;
1253 out:
1254         spin_unlock_irq(&tree->lock);
1255         return ret;
1256 }
1257
1258 /*
1259  * searches a range in the state tree for a given mask.
1260  * If 'filled' == 1, this returns 1 only if every extent in the tree
1261  * has the bits set.  Otherwise, 1 is returned if any bit in the
1262  * range is found set.
1263  */
1264 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1265                    int bits, int filled)
1266 {
1267         struct extent_state *state = NULL;
1268         struct rb_node *node;
1269         int bitset = 0;
1270         unsigned long flags;
1271
1272         spin_lock_irqsave(&tree->lock, flags);
1273         node = tree_search(tree, start);
1274         while (node && start <= end) {
1275                 state = rb_entry(node, struct extent_state, rb_node);
1276
1277                 if (filled && state->start > start) {
1278                         bitset = 0;
1279                         break;
1280                 }
1281
1282                 if (state->start > end)
1283                         break;
1284
1285                 if (state->state & bits) {
1286                         bitset = 1;
1287                         if (!filled)
1288                                 break;
1289                 } else if (filled) {
1290                         bitset = 0;
1291                         break;
1292                 }
1293                 start = state->end + 1;
1294                 if (start > end)
1295                         break;
1296                 node = rb_next(node);
1297                 if (!node) {
1298                         if (filled)
1299                                 bitset = 0;
1300                         break;
1301                 }
1302         }
1303         spin_unlock_irqrestore(&tree->lock, flags);
1304         return bitset;
1305 }
1306 EXPORT_SYMBOL(test_range_bit);
1307
1308 /*
1309  * helper function to set a given page up to date if all the
1310  * extents in the tree for that page are up to date
1311  */
1312 static int check_page_uptodate(struct extent_io_tree *tree,
1313                                struct page *page)
1314 {
1315         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1316         u64 end = start + PAGE_CACHE_SIZE - 1;
1317         if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1))
1318                 SetPageUptodate(page);
1319         return 0;
1320 }
1321
1322 /*
1323  * helper function to unlock a page if all the extents in the tree
1324  * for that page are unlocked
1325  */
1326 static int check_page_locked(struct extent_io_tree *tree,
1327                              struct page *page)
1328 {
1329         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1330         u64 end = start + PAGE_CACHE_SIZE - 1;
1331         if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0))
1332                 unlock_page(page);
1333         return 0;
1334 }
1335
1336 /*
1337  * helper function to end page writeback if all the extents
1338  * in the tree for that page are done with writeback
1339  */
1340 static int check_page_writeback(struct extent_io_tree *tree,
1341                              struct page *page)
1342 {
1343         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1344         u64 end = start + PAGE_CACHE_SIZE - 1;
1345         if (!test_range_bit(tree, start, end, EXTENT_WRITEBACK, 0))
1346                 end_page_writeback(page);
1347         return 0;
1348 }
1349
1350 /* lots and lots of room for performance fixes in the end_bio funcs */
1351
1352 /*
1353  * after a writepage IO is done, we need to:
1354  * clear the uptodate bits on error
1355  * clear the writeback bits in the extent tree for this IO
1356  * end_page_writeback if the page has no more pending IO
1357  *
1358  * Scheduling is not allowed, so the extent state tree is expected
1359  * to have one and only one object corresponding to this IO.
1360  */
1361 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1362 static void end_bio_extent_writepage(struct bio *bio, int err)
1363 #else
1364 static int end_bio_extent_writepage(struct bio *bio,
1365                                    unsigned int bytes_done, int err)
1366 #endif
1367 {
1368         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1369         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1370         struct extent_state *state = bio->bi_private;
1371         struct extent_io_tree *tree = state->tree;
1372         struct rb_node *node;
1373         u64 start;
1374         u64 end;
1375         u64 cur;
1376         int whole_page;
1377         unsigned long flags;
1378
1379 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1380         if (bio->bi_size)
1381                 return 1;
1382 #endif
1383         do {
1384                 struct page *page = bvec->bv_page;
1385                 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1386                          bvec->bv_offset;
1387                 end = start + bvec->bv_len - 1;
1388
1389                 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1390                         whole_page = 1;
1391                 else
1392                         whole_page = 0;
1393
1394                 if (--bvec >= bio->bi_io_vec)
1395                         prefetchw(&bvec->bv_page->flags);
1396
1397                 if (!uptodate) {
1398                         clear_extent_uptodate(tree, start, end, GFP_ATOMIC);
1399                         ClearPageUptodate(page);
1400                         SetPageError(page);
1401                 }
1402
1403                 if (tree->ops && tree->ops->writepage_end_io_hook) {
1404                         tree->ops->writepage_end_io_hook(page, start, end,
1405                                                          state);
1406                 }
1407
1408                 /*
1409                  * bios can get merged in funny ways, and so we need to
1410                  * be careful with the state variable.  We know the
1411                  * state won't be merged with others because it has
1412                  * WRITEBACK set, but we can't be sure each biovec is
1413                  * sequential in the file.  So, if our cached state
1414                  * doesn't match the expected end, search the tree
1415                  * for the correct one.
1416                  */
1417
1418                 spin_lock_irqsave(&tree->lock, flags);
1419                 if (!state || state->end != end) {
1420                         state = NULL;
1421                         node = __etree_search(tree, start, NULL, NULL);
1422                         if (node) {
1423                                 state = rb_entry(node, struct extent_state,
1424                                                  rb_node);
1425                                 if (state->end != end ||
1426                                     !(state->state & EXTENT_WRITEBACK))
1427                                         state = NULL;
1428                         }
1429                         if (!state) {
1430                                 spin_unlock_irqrestore(&tree->lock, flags);
1431                                 clear_extent_writeback(tree, start,
1432                                                        end, GFP_ATOMIC);
1433                                 goto next_io;
1434                         }
1435                 }
1436                 cur = end;
1437                 while(1) {
1438                         struct extent_state *clear = state;
1439                         cur = state->start;
1440                         node = rb_prev(&state->rb_node);
1441                         if (node) {
1442                                 state = rb_entry(node,
1443                                                  struct extent_state,
1444                                                  rb_node);
1445                         } else {
1446                                 state = NULL;
1447                         }
1448
1449                         clear_state_bit(tree, clear, EXTENT_WRITEBACK,
1450                                         1, 0);
1451                         if (cur == start)
1452                                 break;
1453                         if (cur < start) {
1454                                 WARN_ON(1);
1455                                 break;
1456                         }
1457                         if (!node)
1458                                 break;
1459                 }
1460                 /* before releasing the lock, make sure the next state
1461                  * variable has the expected bits set and corresponds
1462                  * to the correct offsets in the file
1463                  */
1464                 if (state && (state->end + 1 != start ||
1465                     !(state->state & EXTENT_WRITEBACK))) {
1466                         state = NULL;
1467                 }
1468                 spin_unlock_irqrestore(&tree->lock, flags);
1469 next_io:
1470
1471                 if (whole_page)
1472                         end_page_writeback(page);
1473                 else
1474                         check_page_writeback(tree, page);
1475         } while (bvec >= bio->bi_io_vec);
1476         bio_put(bio);
1477 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1478         return 0;
1479 #endif
1480 }
1481
1482 /*
1483  * after a readpage IO is done, we need to:
1484  * clear the uptodate bits on error
1485  * set the uptodate bits if things worked
1486  * set the page up to date if all extents in the tree are uptodate
1487  * clear the lock bit in the extent tree
1488  * unlock the page if there are no other extents locked for it
1489  *
1490  * Scheduling is not allowed, so the extent state tree is expected
1491  * to have one and only one object corresponding to this IO.
1492  */
1493 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1494 static void end_bio_extent_readpage(struct bio *bio, int err)
1495 #else
1496 static int end_bio_extent_readpage(struct bio *bio,
1497                                    unsigned int bytes_done, int err)
1498 #endif
1499 {
1500         int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1501         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1502         struct extent_state *state = bio->bi_private;
1503         struct extent_io_tree *tree = state->tree;
1504         struct rb_node *node;
1505         u64 start;
1506         u64 end;
1507         u64 cur;
1508         unsigned long flags;
1509         int whole_page;
1510         int ret;
1511
1512 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1513         if (bio->bi_size)
1514                 return 1;
1515 #endif
1516
1517         do {
1518                 struct page *page = bvec->bv_page;
1519                 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1520                         bvec->bv_offset;
1521                 end = start + bvec->bv_len - 1;
1522
1523                 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1524                         whole_page = 1;
1525                 else
1526                         whole_page = 0;
1527
1528                 if (--bvec >= bio->bi_io_vec)
1529                         prefetchw(&bvec->bv_page->flags);
1530
1531                 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
1532                         ret = tree->ops->readpage_end_io_hook(page, start, end,
1533                                                               state);
1534                         if (ret)
1535                                 uptodate = 0;
1536                 }
1537                 if (!uptodate && tree->ops &&
1538                     tree->ops->readpage_io_failed_hook) {
1539                         ret = tree->ops->readpage_io_failed_hook(bio, page,
1540                                                          start, end, state);
1541                         if (ret == 0) {
1542                                 state = NULL;
1543                                 continue;
1544                         }
1545                 }
1546
1547                 spin_lock_irqsave(&tree->lock, flags);
1548                 if (!state || state->end != end) {
1549                         state = NULL;
1550                         node = __etree_search(tree, start, NULL, NULL);
1551                         if (node) {
1552                                 state = rb_entry(node, struct extent_state,
1553                                                  rb_node);
1554                                 if (state->end != end ||
1555                                     !(state->state & EXTENT_LOCKED))
1556                                         state = NULL;
1557                         }
1558                         if (!state && uptodate) {
1559                                 spin_unlock_irqrestore(&tree->lock, flags);
1560                                 set_extent_uptodate(tree, start, end,
1561                                                     GFP_ATOMIC);
1562                                 unlock_extent(tree, start, end, GFP_ATOMIC);
1563                                 goto next_io;
1564                         }
1565                 }
1566
1567                 cur = end;
1568                 while(1) {
1569                         struct extent_state *clear = state;
1570                         cur = state->start;
1571                         node = rb_prev(&state->rb_node);
1572                         if (node) {
1573                                 state = rb_entry(node,
1574                                          struct extent_state,
1575                                          rb_node);
1576                         } else {
1577                                 state = NULL;
1578                         }
1579                         if (uptodate) {
1580                                 set_state_cb(tree, clear, EXTENT_UPTODATE);
1581                                 clear->state |= EXTENT_UPTODATE;
1582                         }
1583                         clear_state_bit(tree, clear, EXTENT_LOCKED,
1584                                         1, 0);
1585                         if (cur == start)
1586                                 break;
1587                         if (cur < start) {
1588                                 WARN_ON(1);
1589                                 break;
1590                         }
1591                         if (!node)
1592                                 break;
1593                 }
1594                 /* before releasing the lock, make sure the next state
1595                  * variable has the expected bits set and corresponds
1596                  * to the correct offsets in the file
1597                  */
1598                 if (state && (state->end + 1 != start ||
1599                     !(state->state & EXTENT_LOCKED))) {
1600                         state = NULL;
1601                 }
1602                 spin_unlock_irqrestore(&tree->lock, flags);
1603 next_io:
1604                 if (whole_page) {
1605                         if (uptodate) {
1606                                 SetPageUptodate(page);
1607                         } else {
1608                                 ClearPageUptodate(page);
1609                                 SetPageError(page);
1610                         }
1611                         unlock_page(page);
1612                 } else {
1613                         if (uptodate) {
1614                                 check_page_uptodate(tree, page);
1615                         } else {
1616                                 ClearPageUptodate(page);
1617                                 SetPageError(page);
1618                         }
1619                         check_page_locked(tree, page);
1620                 }
1621         } while (bvec >= bio->bi_io_vec);
1622
1623         bio_put(bio);
1624 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1625         return 0;
1626 #endif
1627 }
1628
1629 /*
1630  * IO done from prepare_write is pretty simple, we just unlock
1631  * the structs in the extent tree when done, and set the uptodate bits
1632  * as appropriate.
1633  */
1634 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1635 static void end_bio_extent_preparewrite(struct bio *bio, int err)
1636 #else
1637 static int end_bio_extent_preparewrite(struct bio *bio,
1638                                        unsigned int bytes_done, int err)
1639 #endif
1640 {
1641         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1642         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1643         struct extent_state *state = bio->bi_private;
1644         struct extent_io_tree *tree = state->tree;
1645         u64 start;
1646         u64 end;
1647
1648 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1649         if (bio->bi_size)
1650                 return 1;
1651 #endif
1652
1653         do {
1654                 struct page *page = bvec->bv_page;
1655                 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1656                         bvec->bv_offset;
1657                 end = start + bvec->bv_len - 1;
1658
1659                 if (--bvec >= bio->bi_io_vec)
1660                         prefetchw(&bvec->bv_page->flags);
1661
1662                 if (uptodate) {
1663                         set_extent_uptodate(tree, start, end, GFP_ATOMIC);
1664                 } else {
1665                         ClearPageUptodate(page);
1666                         SetPageError(page);
1667                 }
1668
1669                 unlock_extent(tree, start, end, GFP_ATOMIC);
1670
1671         } while (bvec >= bio->bi_io_vec);
1672
1673         bio_put(bio);
1674 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1675         return 0;
1676 #endif
1677 }
1678
1679 static struct bio *
1680 extent_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
1681                  gfp_t gfp_flags)
1682 {
1683         struct bio *bio;
1684
1685         bio = bio_alloc(gfp_flags, nr_vecs);
1686
1687         if (bio == NULL && (current->flags & PF_MEMALLOC)) {
1688                 while (!bio && (nr_vecs /= 2))
1689                         bio = bio_alloc(gfp_flags, nr_vecs);
1690         }
1691
1692         if (bio) {
1693                 bio->bi_bdev = bdev;
1694                 bio->bi_sector = first_sector;
1695         }
1696         return bio;
1697 }
1698
1699 static int submit_one_bio(int rw, struct bio *bio, int mirror_num)
1700 {
1701         u64 maxsector;
1702         int ret = 0;
1703         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1704         struct page *page = bvec->bv_page;
1705         struct extent_io_tree *tree = bio->bi_private;
1706         struct rb_node *node;
1707         struct extent_state *state;
1708         u64 start;
1709         u64 end;
1710
1711         start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
1712         end = start + bvec->bv_len - 1;
1713
1714         spin_lock_irq(&tree->lock);
1715         node = __etree_search(tree, start, NULL, NULL);
1716         BUG_ON(!node);
1717         state = rb_entry(node, struct extent_state, rb_node);
1718         while(state->end < end) {
1719                 node = rb_next(node);
1720                 state = rb_entry(node, struct extent_state, rb_node);
1721         }
1722         BUG_ON(state->end != end);
1723         spin_unlock_irq(&tree->lock);
1724
1725         bio->bi_private = state;
1726
1727         bio_get(bio);
1728
1729         maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
1730         if (maxsector < bio->bi_sector) {
1731                 printk("sector too large max %Lu got %llu\n", maxsector,
1732                         (unsigned long long)bio->bi_sector);
1733                 WARN_ON(1);
1734         }
1735         if (tree->ops && tree->ops->submit_bio_hook)
1736                 tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
1737                                            mirror_num);
1738         else
1739                 submit_bio(rw, bio);
1740         if (bio_flagged(bio, BIO_EOPNOTSUPP))
1741                 ret = -EOPNOTSUPP;
1742         bio_put(bio);
1743         return ret;
1744 }
1745
1746 static int submit_extent_page(int rw, struct extent_io_tree *tree,
1747                               struct page *page, sector_t sector,
1748                               size_t size, unsigned long offset,
1749                               struct block_device *bdev,
1750                               struct bio **bio_ret,
1751                               unsigned long max_pages,
1752                               bio_end_io_t end_io_func,
1753                               int mirror_num)
1754 {
1755         int ret = 0;
1756         struct bio *bio;
1757         int nr;
1758
1759         if (bio_ret && *bio_ret) {
1760                 bio = *bio_ret;
1761                 if (bio->bi_sector + (bio->bi_size >> 9) != sector ||
1762                     (tree->ops && tree->ops->merge_bio_hook &&
1763                      tree->ops->merge_bio_hook(page, offset, size, bio)) ||
1764                     bio_add_page(bio, page, size, offset) < size) {
1765                         ret = submit_one_bio(rw, bio, mirror_num);
1766                         bio = NULL;
1767                 } else {
1768                         return 0;
1769                 }
1770         }
1771         nr = bio_get_nr_vecs(bdev);
1772         bio = extent_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
1773         if (!bio) {
1774                 printk("failed to allocate bio nr %d\n", nr);
1775         }
1776
1777
1778         bio_add_page(bio, page, size, offset);
1779         bio->bi_end_io = end_io_func;
1780         bio->bi_private = tree;
1781
1782         if (bio_ret) {
1783                 *bio_ret = bio;
1784         } else {
1785                 ret = submit_one_bio(rw, bio, mirror_num);
1786         }
1787
1788         return ret;
1789 }
1790
1791 void set_page_extent_mapped(struct page *page)
1792 {
1793         if (!PagePrivate(page)) {
1794                 SetPagePrivate(page);
1795                 WARN_ON(!page->mapping->a_ops->invalidatepage);
1796                 set_page_private(page, EXTENT_PAGE_PRIVATE);
1797                 page_cache_get(page);
1798         }
1799 }
1800
1801 void set_page_extent_head(struct page *page, unsigned long len)
1802 {
1803         set_page_private(page, EXTENT_PAGE_PRIVATE_FIRST_PAGE | len << 2);
1804 }
1805
1806 /*
1807  * basic readpage implementation.  Locked extent state structs are inserted
1808  * into the tree that are removed when the IO is done (by the end_io
1809  * handlers)
1810  */
1811 static int __extent_read_full_page(struct extent_io_tree *tree,
1812                                    struct page *page,
1813                                    get_extent_t *get_extent,
1814                                    struct bio **bio, int mirror_num)
1815 {
1816         struct inode *inode = page->mapping->host;
1817         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1818         u64 page_end = start + PAGE_CACHE_SIZE - 1;
1819         u64 end;
1820         u64 cur = start;
1821         u64 extent_offset;
1822         u64 last_byte = i_size_read(inode);
1823         u64 block_start;
1824         u64 cur_end;
1825         sector_t sector;
1826         struct extent_map *em;
1827         struct block_device *bdev;
1828         int ret;
1829         int nr = 0;
1830         size_t page_offset = 0;
1831         size_t iosize;
1832         size_t blocksize = inode->i_sb->s_blocksize;
1833
1834         set_page_extent_mapped(page);
1835
1836         end = page_end;
1837         lock_extent(tree, start, end, GFP_NOFS);
1838
1839         while (cur <= end) {
1840                 if (cur >= last_byte) {
1841                         char *userpage;
1842                         iosize = PAGE_CACHE_SIZE - page_offset;
1843                         userpage = kmap_atomic(page, KM_USER0);
1844                         memset(userpage + page_offset, 0, iosize);
1845                         flush_dcache_page(page);
1846                         kunmap_atomic(userpage, KM_USER0);
1847                         set_extent_uptodate(tree, cur, cur + iosize - 1,
1848                                             GFP_NOFS);
1849                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1850                         break;
1851                 }
1852                 em = get_extent(inode, page, page_offset, cur,
1853                                 end - cur + 1, 0);
1854                 if (IS_ERR(em) || !em) {
1855                         SetPageError(page);
1856                         unlock_extent(tree, cur, end, GFP_NOFS);
1857                         break;
1858                 }
1859
1860                 extent_offset = cur - em->start;
1861                 BUG_ON(extent_map_end(em) <= cur);
1862                 BUG_ON(end < cur);
1863
1864                 iosize = min(extent_map_end(em) - cur, end - cur + 1);
1865                 cur_end = min(extent_map_end(em) - 1, end);
1866                 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
1867                 sector = (em->block_start + extent_offset) >> 9;
1868                 bdev = em->bdev;
1869                 block_start = em->block_start;
1870                 free_extent_map(em);
1871                 em = NULL;
1872
1873                 /* we've found a hole, just zero and go on */
1874                 if (block_start == EXTENT_MAP_HOLE) {
1875                         char *userpage;
1876                         userpage = kmap_atomic(page, KM_USER0);
1877                         memset(userpage + page_offset, 0, iosize);
1878                         flush_dcache_page(page);
1879                         kunmap_atomic(userpage, KM_USER0);
1880
1881                         set_extent_uptodate(tree, cur, cur + iosize - 1,
1882                                             GFP_NOFS);
1883                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1884                         cur = cur + iosize;
1885                         page_offset += iosize;
1886                         continue;
1887                 }
1888                 /* the get_extent function already copied into the page */
1889                 if (test_range_bit(tree, cur, cur_end, EXTENT_UPTODATE, 1)) {
1890                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1891                         cur = cur + iosize;
1892                         page_offset += iosize;
1893                         continue;
1894                 }
1895                 /* we have an inline extent but it didn't get marked up
1896                  * to date.  Error out
1897                  */
1898                 if (block_start == EXTENT_MAP_INLINE) {
1899                         SetPageError(page);
1900                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1901                         cur = cur + iosize;
1902                         page_offset += iosize;
1903                         continue;
1904                 }
1905
1906                 ret = 0;
1907                 if (tree->ops && tree->ops->readpage_io_hook) {
1908                         ret = tree->ops->readpage_io_hook(page, cur,
1909                                                           cur + iosize - 1);
1910                 }
1911                 if (!ret) {
1912                         unsigned long nr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
1913                         nr -= page->index;
1914                         ret = submit_extent_page(READ, tree, page,
1915                                          sector, iosize, page_offset,
1916                                          bdev, bio, nr,
1917                                          end_bio_extent_readpage, mirror_num);
1918                 }
1919                 if (ret)
1920                         SetPageError(page);
1921                 cur = cur + iosize;
1922                 page_offset += iosize;
1923                 nr++;
1924         }
1925         if (!nr) {
1926                 if (!PageError(page))
1927                         SetPageUptodate(page);
1928                 unlock_page(page);
1929         }
1930         return 0;
1931 }
1932
1933 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
1934                             get_extent_t *get_extent)
1935 {
1936         struct bio *bio = NULL;
1937         int ret;
1938
1939         ret = __extent_read_full_page(tree, page, get_extent, &bio, 0);
1940         if (bio)
1941                 submit_one_bio(READ, bio, 0);
1942         return ret;
1943 }
1944 EXPORT_SYMBOL(extent_read_full_page);
1945
1946 /*
1947  * the writepage semantics are similar to regular writepage.  extent
1948  * records are inserted to lock ranges in the tree, and as dirty areas
1949  * are found, they are marked writeback.  Then the lock bits are removed
1950  * and the end_io handler clears the writeback ranges
1951  */
1952 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
1953                               void *data)
1954 {
1955         struct inode *inode = page->mapping->host;
1956         struct extent_page_data *epd = data;
1957         struct extent_io_tree *tree = epd->tree;
1958         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1959         u64 delalloc_start;
1960         u64 page_end = start + PAGE_CACHE_SIZE - 1;
1961         u64 end;
1962         u64 cur = start;
1963         u64 extent_offset;
1964         u64 last_byte = i_size_read(inode);
1965         u64 block_start;
1966         u64 iosize;
1967         sector_t sector;
1968         struct extent_map *em;
1969         struct block_device *bdev;
1970         int ret;
1971         int nr = 0;
1972         size_t page_offset = 0;
1973         size_t blocksize;
1974         loff_t i_size = i_size_read(inode);
1975         unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
1976         u64 nr_delalloc;
1977         u64 delalloc_end;
1978
1979         WARN_ON(!PageLocked(page));
1980         if (page->index > end_index) {
1981                 clear_extent_dirty(tree, start, page_end, GFP_NOFS);
1982                 unlock_page(page);
1983                 return 0;
1984         }
1985
1986         if (page->index == end_index) {
1987                 char *userpage;
1988
1989                 size_t offset = i_size & (PAGE_CACHE_SIZE - 1);
1990
1991                 userpage = kmap_atomic(page, KM_USER0);
1992                 memset(userpage + offset, 0, PAGE_CACHE_SIZE - offset);
1993                 flush_dcache_page(page);
1994                 kunmap_atomic(userpage, KM_USER0);
1995         }
1996
1997         set_page_extent_mapped(page);
1998
1999         delalloc_start = start;
2000         delalloc_end = 0;
2001         while(delalloc_end < page_end) {
2002                 nr_delalloc = find_lock_delalloc_range(tree, &delalloc_start,
2003                                                        &delalloc_end,
2004                                                        128 * 1024 * 1024);
2005                 if (nr_delalloc == 0) {
2006                         delalloc_start = delalloc_end + 1;
2007                         continue;
2008                 }
2009                 tree->ops->fill_delalloc(inode, delalloc_start,
2010                                          delalloc_end);
2011                 clear_extent_bit(tree, delalloc_start,
2012                                  delalloc_end,
2013                                  EXTENT_LOCKED | EXTENT_DELALLOC,
2014                                  1, 0, GFP_NOFS);
2015                 delalloc_start = delalloc_end + 1;
2016         }
2017         lock_extent(tree, start, page_end, GFP_NOFS);
2018
2019         end = page_end;
2020         if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0)) {
2021                 printk("found delalloc bits after lock_extent\n");
2022         }
2023
2024         if (last_byte <= start) {
2025                 clear_extent_dirty(tree, start, page_end, GFP_NOFS);
2026                 goto done;
2027         }
2028
2029         set_extent_uptodate(tree, start, page_end, GFP_NOFS);
2030         blocksize = inode->i_sb->s_blocksize;
2031
2032         while (cur <= end) {
2033                 if (cur >= last_byte) {
2034                         clear_extent_dirty(tree, cur, page_end, GFP_NOFS);
2035                         break;
2036                 }
2037                 em = epd->get_extent(inode, page, page_offset, cur,
2038                                      end - cur + 1, 1);
2039                 if (IS_ERR(em) || !em) {
2040                         SetPageError(page);
2041                         break;
2042                 }
2043
2044                 extent_offset = cur - em->start;
2045                 BUG_ON(extent_map_end(em) <= cur);
2046                 BUG_ON(end < cur);
2047                 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2048                 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2049                 sector = (em->block_start + extent_offset) >> 9;
2050                 bdev = em->bdev;
2051                 block_start = em->block_start;
2052                 free_extent_map(em);
2053                 em = NULL;
2054
2055                 if (block_start == EXTENT_MAP_HOLE ||
2056                     block_start == EXTENT_MAP_INLINE) {
2057                         clear_extent_dirty(tree, cur,
2058                                            cur + iosize - 1, GFP_NOFS);
2059                         cur = cur + iosize;
2060                         page_offset += iosize;
2061                         continue;
2062                 }
2063
2064                 /* leave this out until we have a page_mkwrite call */
2065                 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
2066                                    EXTENT_DIRTY, 0)) {
2067                         cur = cur + iosize;
2068                         page_offset += iosize;
2069                         continue;
2070                 }
2071                 clear_extent_dirty(tree, cur, cur + iosize - 1, GFP_NOFS);
2072                 if (tree->ops && tree->ops->writepage_io_hook) {
2073                         ret = tree->ops->writepage_io_hook(page, cur,
2074                                                 cur + iosize - 1);
2075                 } else {
2076                         ret = 0;
2077                 }
2078                 if (ret)
2079                         SetPageError(page);
2080                 else {
2081                         unsigned long max_nr = end_index + 1;
2082                         set_range_writeback(tree, cur, cur + iosize - 1);
2083                         if (!PageWriteback(page)) {
2084                                 printk("warning page %lu not writeback, "
2085                                        "cur %llu end %llu\n", page->index,
2086                                        (unsigned long long)cur,
2087                                        (unsigned long long)end);
2088                         }
2089
2090                         ret = submit_extent_page(WRITE, tree, page, sector,
2091                                                  iosize, page_offset, bdev,
2092                                                  &epd->bio, max_nr,
2093                                                  end_bio_extent_writepage, 0);
2094                         if (ret)
2095                                 SetPageError(page);
2096                 }
2097                 cur = cur + iosize;
2098                 page_offset += iosize;
2099                 nr++;
2100         }
2101 done:
2102         if (nr == 0) {
2103                 /* make sure the mapping tag for page dirty gets cleared */
2104                 set_page_writeback(page);
2105                 end_page_writeback(page);
2106         }
2107         unlock_extent(tree, start, page_end, GFP_NOFS);
2108         unlock_page(page);
2109         return 0;
2110 }
2111
2112 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,20)
2113 /* Taken directly from 2.6.23 for 2.6.18 back port */
2114 typedef int (*writepage_t)(struct page *page, struct writeback_control *wbc,
2115                                 void *data);
2116
2117 /**
2118  * write_cache_pages - walk the list of dirty pages of the given address space
2119  * and write all of them.
2120  * @mapping: address space structure to write
2121  * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2122  * @writepage: function called for each page
2123  * @data: data passed to writepage function
2124  *
2125  * If a page is already under I/O, write_cache_pages() skips it, even
2126  * if it's dirty.  This is desirable behaviour for memory-cleaning writeback,
2127  * but it is INCORRECT for data-integrity system calls such as fsync().  fsync()
2128  * and msync() need to guarantee that all the data which was dirty at the time
2129  * the call was made get new I/O started against them.  If wbc->sync_mode is
2130  * WB_SYNC_ALL then we were called for data integrity and we must wait for
2131  * existing IO to complete.
2132  */
2133 static int write_cache_pages(struct address_space *mapping,
2134                       struct writeback_control *wbc, writepage_t writepage,
2135                       void *data)
2136 {
2137         struct backing_dev_info *bdi = mapping->backing_dev_info;
2138         int ret = 0;
2139         int done = 0;
2140         struct pagevec pvec;
2141         int nr_pages;
2142         pgoff_t index;
2143         pgoff_t end;            /* Inclusive */
2144         int scanned = 0;
2145         int range_whole = 0;
2146
2147         if (wbc->nonblocking && bdi_write_congested(bdi)) {
2148                 wbc->encountered_congestion = 1;
2149                 return 0;
2150         }
2151
2152         pagevec_init(&pvec, 0);
2153         if (wbc->range_cyclic) {
2154                 index = mapping->writeback_index; /* Start from prev offset */
2155                 end = -1;
2156         } else {
2157                 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2158                 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2159                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2160                         range_whole = 1;
2161                 scanned = 1;
2162         }
2163 retry:
2164         while (!done && (index <= end) &&
2165                (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
2166                                               PAGECACHE_TAG_DIRTY,
2167                                               min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
2168                 unsigned i;
2169
2170                 scanned = 1;
2171                 for (i = 0; i < nr_pages; i++) {
2172                         struct page *page = pvec.pages[i];
2173
2174                         /*
2175                          * At this point we hold neither mapping->tree_lock nor
2176                          * lock on the page itself: the page may be truncated or
2177                          * invalidated (changing page->mapping to NULL), or even
2178                          * swizzled back from swapper_space to tmpfs file
2179                          * mapping
2180                          */
2181                         lock_page(page);
2182
2183                         if (unlikely(page->mapping != mapping)) {
2184                                 unlock_page(page);
2185                                 continue;
2186                         }
2187
2188                         if (!wbc->range_cyclic && page->index > end) {
2189                                 done = 1;
2190                                 unlock_page(page);
2191                                 continue;
2192                         }
2193
2194                         if (wbc->sync_mode != WB_SYNC_NONE)
2195                                 wait_on_page_writeback(page);
2196
2197                         if (PageWriteback(page) ||
2198                             !clear_page_dirty_for_io(page)) {
2199                                 unlock_page(page);
2200                                 continue;
2201                         }
2202
2203                         ret = (*writepage)(page, wbc, data);
2204
2205                         if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
2206                                 unlock_page(page);
2207                                 ret = 0;
2208                         }
2209                         if (ret || (--(wbc->nr_to_write) <= 0))
2210                                 done = 1;
2211                         if (wbc->nonblocking && bdi_write_congested(bdi)) {
2212                                 wbc->encountered_congestion = 1;
2213                                 done = 1;
2214                         }
2215                 }
2216                 pagevec_release(&pvec);
2217                 cond_resched();
2218         }
2219         if (!scanned && !done) {
2220                 /*
2221                  * We hit the last page and there is more work to be done: wrap
2222                  * back to the start of the file
2223                  */
2224                 scanned = 1;
2225                 index = 0;
2226                 goto retry;
2227         }
2228         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2229                 mapping->writeback_index = index;
2230         return ret;
2231 }
2232 #endif
2233
2234 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
2235                           get_extent_t *get_extent,
2236                           struct writeback_control *wbc)
2237 {
2238         int ret;
2239         struct address_space *mapping = page->mapping;
2240         struct extent_page_data epd = {
2241                 .bio = NULL,
2242                 .tree = tree,
2243                 .get_extent = get_extent,
2244         };
2245         struct writeback_control wbc_writepages = {
2246                 .bdi            = wbc->bdi,
2247                 .sync_mode      = WB_SYNC_NONE,
2248                 .older_than_this = NULL,
2249                 .nr_to_write    = 64,
2250                 .range_start    = page_offset(page) + PAGE_CACHE_SIZE,
2251                 .range_end      = (loff_t)-1,
2252         };
2253
2254
2255         ret = __extent_writepage(page, wbc, &epd);
2256
2257         write_cache_pages(mapping, &wbc_writepages, __extent_writepage, &epd);
2258         if (epd.bio) {
2259                 submit_one_bio(WRITE, epd.bio, 0);
2260         }
2261         return ret;
2262 }
2263 EXPORT_SYMBOL(extent_write_full_page);
2264
2265
2266 int extent_writepages(struct extent_io_tree *tree,
2267                       struct address_space *mapping,
2268                       get_extent_t *get_extent,
2269                       struct writeback_control *wbc)
2270 {
2271         int ret = 0;
2272         struct extent_page_data epd = {
2273                 .bio = NULL,
2274                 .tree = tree,
2275                 .get_extent = get_extent,
2276         };
2277
2278         ret = write_cache_pages(mapping, wbc, __extent_writepage, &epd);
2279         if (epd.bio) {
2280                 submit_one_bio(WRITE, epd.bio, 0);
2281         }
2282         return ret;
2283 }
2284 EXPORT_SYMBOL(extent_writepages);
2285
2286 int extent_readpages(struct extent_io_tree *tree,
2287                      struct address_space *mapping,
2288                      struct list_head *pages, unsigned nr_pages,
2289                      get_extent_t get_extent)
2290 {
2291         struct bio *bio = NULL;
2292         unsigned page_idx;
2293         struct pagevec pvec;
2294
2295         pagevec_init(&pvec, 0);
2296         for (page_idx = 0; page_idx < nr_pages; page_idx++) {
2297                 struct page *page = list_entry(pages->prev, struct page, lru);
2298
2299                 prefetchw(&page->flags);
2300                 list_del(&page->lru);
2301                 /*
2302                  * what we want to do here is call add_to_page_cache_lru,
2303                  * but that isn't exported, so we reproduce it here
2304                  */
2305                 if (!add_to_page_cache(page, mapping,
2306                                         page->index, GFP_KERNEL)) {
2307
2308                         /* open coding of lru_cache_add, also not exported */
2309                         page_cache_get(page);
2310                         if (!pagevec_add(&pvec, page))
2311                                 __pagevec_lru_add(&pvec);
2312                         __extent_read_full_page(tree, page, get_extent,
2313                                                 &bio, 0);
2314                 }
2315                 page_cache_release(page);
2316         }
2317         if (pagevec_count(&pvec))
2318                 __pagevec_lru_add(&pvec);
2319         BUG_ON(!list_empty(pages));
2320         if (bio)
2321                 submit_one_bio(READ, bio, 0);
2322         return 0;
2323 }
2324 EXPORT_SYMBOL(extent_readpages);
2325
2326 /*
2327  * basic invalidatepage code, this waits on any locked or writeback
2328  * ranges corresponding to the page, and then deletes any extent state
2329  * records from the tree
2330  */
2331 int extent_invalidatepage(struct extent_io_tree *tree,
2332                           struct page *page, unsigned long offset)
2333 {
2334         u64 start = ((u64)page->index << PAGE_CACHE_SHIFT);
2335         u64 end = start + PAGE_CACHE_SIZE - 1;
2336         size_t blocksize = page->mapping->host->i_sb->s_blocksize;
2337
2338         start += (offset + blocksize -1) & ~(blocksize - 1);
2339         if (start > end)
2340                 return 0;
2341
2342         lock_extent(tree, start, end, GFP_NOFS);
2343         wait_on_extent_writeback(tree, start, end);
2344         clear_extent_bit(tree, start, end,
2345                          EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC,
2346                          1, 1, GFP_NOFS);
2347         return 0;
2348 }
2349 EXPORT_SYMBOL(extent_invalidatepage);
2350
2351 /*
2352  * simple commit_write call, set_range_dirty is used to mark both
2353  * the pages and the extent records as dirty
2354  */
2355 int extent_commit_write(struct extent_io_tree *tree,
2356                         struct inode *inode, struct page *page,
2357                         unsigned from, unsigned to)
2358 {
2359         loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
2360
2361         set_page_extent_mapped(page);
2362         set_page_dirty(page);
2363
2364         if (pos > inode->i_size) {
2365                 i_size_write(inode, pos);
2366                 mark_inode_dirty(inode);
2367         }
2368         return 0;
2369 }
2370 EXPORT_SYMBOL(extent_commit_write);
2371
2372 int extent_prepare_write(struct extent_io_tree *tree,
2373                          struct inode *inode, struct page *page,
2374                          unsigned from, unsigned to, get_extent_t *get_extent)
2375 {
2376         u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2377         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
2378         u64 block_start;
2379         u64 orig_block_start;
2380         u64 block_end;
2381         u64 cur_end;
2382         struct extent_map *em;
2383         unsigned blocksize = 1 << inode->i_blkbits;
2384         size_t page_offset = 0;
2385         size_t block_off_start;
2386         size_t block_off_end;
2387         int err = 0;
2388         int iocount = 0;
2389         int ret = 0;
2390         int isnew;
2391
2392         set_page_extent_mapped(page);
2393
2394         block_start = (page_start + from) & ~((u64)blocksize - 1);
2395         block_end = (page_start + to - 1) | (blocksize - 1);
2396         orig_block_start = block_start;
2397
2398         lock_extent(tree, page_start, page_end, GFP_NOFS);
2399         while(block_start <= block_end) {
2400                 em = get_extent(inode, page, page_offset, block_start,
2401                                 block_end - block_start + 1, 1);
2402                 if (IS_ERR(em) || !em) {
2403                         goto err;
2404                 }
2405                 cur_end = min(block_end, extent_map_end(em) - 1);
2406                 block_off_start = block_start & (PAGE_CACHE_SIZE - 1);
2407                 block_off_end = block_off_start + blocksize;
2408                 isnew = clear_extent_new(tree, block_start, cur_end, GFP_NOFS);
2409
2410                 if (!PageUptodate(page) && isnew &&
2411                     (block_off_end > to || block_off_start < from)) {
2412                         void *kaddr;
2413
2414                         kaddr = kmap_atomic(page, KM_USER0);
2415                         if (block_off_end > to)
2416                                 memset(kaddr + to, 0, block_off_end - to);
2417                         if (block_off_start < from)
2418                                 memset(kaddr + block_off_start, 0,
2419                                        from - block_off_start);
2420                         flush_dcache_page(page);
2421                         kunmap_atomic(kaddr, KM_USER0);
2422                 }
2423                 if ((em->block_start != EXTENT_MAP_HOLE &&
2424                      em->block_start != EXTENT_MAP_INLINE) &&
2425                     !isnew && !PageUptodate(page) &&
2426                     (block_off_end > to || block_off_start < from) &&
2427                     !test_range_bit(tree, block_start, cur_end,
2428                                     EXTENT_UPTODATE, 1)) {
2429                         u64 sector;
2430                         u64 extent_offset = block_start - em->start;
2431                         size_t iosize;
2432                         sector = (em->block_start + extent_offset) >> 9;
2433                         iosize = (cur_end - block_start + blocksize) &
2434                                 ~((u64)blocksize - 1);
2435                         /*
2436                          * we've already got the extent locked, but we
2437                          * need to split the state such that our end_bio
2438                          * handler can clear the lock.
2439                          */
2440                         set_extent_bit(tree, block_start,
2441                                        block_start + iosize - 1,
2442                                        EXTENT_LOCKED, 0, NULL, GFP_NOFS);
2443                         ret = submit_extent_page(READ, tree, page,
2444                                          sector, iosize, page_offset, em->bdev,
2445                                          NULL, 1,
2446                                          end_bio_extent_preparewrite, 0);
2447                         iocount++;
2448                         block_start = block_start + iosize;
2449                 } else {
2450                         set_extent_uptodate(tree, block_start, cur_end,
2451                                             GFP_NOFS);
2452                         unlock_extent(tree, block_start, cur_end, GFP_NOFS);
2453                         block_start = cur_end + 1;
2454                 }
2455                 page_offset = block_start & (PAGE_CACHE_SIZE - 1);
2456                 free_extent_map(em);
2457         }
2458         if (iocount) {
2459                 wait_extent_bit(tree, orig_block_start,
2460                                 block_end, EXTENT_LOCKED);
2461         }
2462         check_page_uptodate(tree, page);
2463 err:
2464         /* FIXME, zero out newly allocated blocks on error */
2465         return err;
2466 }
2467 EXPORT_SYMBOL(extent_prepare_write);
2468
2469 /*
2470  * a helper for releasepage.  As long as there are no locked extents
2471  * in the range corresponding to the page, both state records and extent
2472  * map records are removed
2473  */
2474 int try_release_extent_mapping(struct extent_map_tree *map,
2475                                struct extent_io_tree *tree, struct page *page,
2476                                gfp_t mask)
2477 {
2478         struct extent_map *em;
2479         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2480         u64 end = start + PAGE_CACHE_SIZE - 1;
2481         u64 orig_start = start;
2482         int ret = 1;
2483         if ((mask & __GFP_WAIT) &&
2484             page->mapping->host->i_size > 16 * 1024 * 1024) {
2485                 u64 len;
2486                 while (start <= end) {
2487                         len = end - start + 1;
2488                         spin_lock(&map->lock);
2489                         em = lookup_extent_mapping(map, start, len);
2490                         if (!em || IS_ERR(em)) {
2491                                 spin_unlock(&map->lock);
2492                                 break;
2493                         }
2494                         if (em->start != start) {
2495                                 spin_unlock(&map->lock);
2496                                 free_extent_map(em);
2497                                 break;
2498                         }
2499                         if (!test_range_bit(tree, em->start,
2500                                             extent_map_end(em) - 1,
2501                                             EXTENT_LOCKED, 0)) {
2502                                 remove_extent_mapping(map, em);
2503                                 /* once for the rb tree */
2504                                 free_extent_map(em);
2505                         }
2506                         start = extent_map_end(em);
2507                         spin_unlock(&map->lock);
2508
2509                         /* once for us */
2510                         free_extent_map(em);
2511                 }
2512         }
2513         if (test_range_bit(tree, orig_start, end, EXTENT_IOBITS, 0))
2514                 ret = 0;
2515         else {
2516                 if ((mask & GFP_NOFS) == GFP_NOFS)
2517                         mask = GFP_NOFS;
2518                 clear_extent_bit(tree, orig_start, end, EXTENT_UPTODATE,
2519                                  1, 1, mask);
2520         }
2521         return ret;
2522 }
2523 EXPORT_SYMBOL(try_release_extent_mapping);
2524
2525 sector_t extent_bmap(struct address_space *mapping, sector_t iblock,
2526                 get_extent_t *get_extent)
2527 {
2528         struct inode *inode = mapping->host;
2529         u64 start = iblock << inode->i_blkbits;
2530         sector_t sector = 0;
2531         struct extent_map *em;
2532
2533         em = get_extent(inode, NULL, 0, start, (1 << inode->i_blkbits), 0);
2534         if (!em || IS_ERR(em))
2535                 return 0;
2536
2537         if (em->block_start == EXTENT_MAP_INLINE ||
2538             em->block_start == EXTENT_MAP_HOLE)
2539                 goto out;
2540
2541         sector = (em->block_start + start - em->start) >> inode->i_blkbits;
2542 out:
2543         free_extent_map(em);
2544         return sector;
2545 }
2546
2547 static int add_lru(struct extent_io_tree *tree, struct extent_buffer *eb)
2548 {
2549         if (list_empty(&eb->lru)) {
2550                 extent_buffer_get(eb);
2551                 list_add(&eb->lru, &tree->buffer_lru);
2552                 tree->lru_size++;
2553                 if (tree->lru_size >= BUFFER_LRU_MAX) {
2554                         struct extent_buffer *rm;
2555                         rm = list_entry(tree->buffer_lru.prev,
2556                                         struct extent_buffer, lru);
2557                         tree->lru_size--;
2558                         list_del_init(&rm->lru);
2559                         free_extent_buffer(rm);
2560                 }
2561         } else
2562                 list_move(&eb->lru, &tree->buffer_lru);
2563         return 0;
2564 }
2565 static struct extent_buffer *find_lru(struct extent_io_tree *tree,
2566                                       u64 start, unsigned long len)
2567 {
2568         struct list_head *lru = &tree->buffer_lru;
2569         struct list_head *cur = lru->next;
2570         struct extent_buffer *eb;
2571
2572         if (list_empty(lru))
2573                 return NULL;
2574
2575         do {
2576                 eb = list_entry(cur, struct extent_buffer, lru);
2577                 if (eb->start == start && eb->len == len) {
2578                         extent_buffer_get(eb);
2579                         return eb;
2580                 }
2581                 cur = cur->next;
2582         } while (cur != lru);
2583         return NULL;
2584 }
2585
2586 static inline unsigned long num_extent_pages(u64 start, u64 len)
2587 {
2588         return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
2589                 (start >> PAGE_CACHE_SHIFT);
2590 }
2591
2592 static inline struct page *extent_buffer_page(struct extent_buffer *eb,
2593                                               unsigned long i)
2594 {
2595         struct page *p;
2596         struct address_space *mapping;
2597
2598         if (i == 0)
2599                 return eb->first_page;
2600         i += eb->start >> PAGE_CACHE_SHIFT;
2601         mapping = eb->first_page->mapping;
2602         read_lock_irq(&mapping->tree_lock);
2603         p = radix_tree_lookup(&mapping->page_tree, i);
2604         read_unlock_irq(&mapping->tree_lock);
2605         return p;
2606 }
2607
2608 int release_extent_buffer_tail_pages(struct extent_buffer *eb)
2609 {
2610         unsigned long num_pages = num_extent_pages(eb->start, eb->len);
2611         struct page *page;
2612         unsigned long i;
2613
2614         if (num_pages == 1)
2615                 return 0;
2616         for (i = 1; i < num_pages; i++) {
2617                 page = extent_buffer_page(eb, i);
2618                 page_cache_release(page);
2619         }
2620         return 0;
2621 }
2622
2623
2624 int invalidate_extent_lru(struct extent_io_tree *tree, u64 start,
2625                           unsigned long len)
2626 {
2627         struct list_head *lru = &tree->buffer_lru;
2628         struct list_head *cur = lru->next;
2629         struct extent_buffer *eb;
2630         int found = 0;
2631
2632         spin_lock(&tree->lru_lock);
2633         if (list_empty(lru))
2634                 goto out;
2635
2636         do {
2637                 eb = list_entry(cur, struct extent_buffer, lru);
2638                 if (eb->start <= start && eb->start + eb->len > start) {
2639                         eb->flags &= ~EXTENT_UPTODATE;
2640                 }
2641                 cur = cur->next;
2642         } while (cur != lru);
2643 out:
2644         spin_unlock(&tree->lru_lock);
2645         return found;
2646 }
2647
2648 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
2649                                                    u64 start,
2650                                                    unsigned long len,
2651                                                    gfp_t mask)
2652 {
2653         struct extent_buffer *eb = NULL;
2654         unsigned long flags;
2655
2656         spin_lock(&tree->lru_lock);
2657         eb = find_lru(tree, start, len);
2658         spin_unlock(&tree->lru_lock);
2659         if (eb) {
2660                 return eb;
2661         }
2662
2663         eb = kmem_cache_zalloc(extent_buffer_cache, mask);
2664         INIT_LIST_HEAD(&eb->lru);
2665         eb->start = start;
2666         eb->len = len;
2667         spin_lock_irqsave(&leak_lock, flags);
2668         list_add(&eb->leak_list, &buffers);
2669         spin_unlock_irqrestore(&leak_lock, flags);
2670         atomic_set(&eb->refs, 1);
2671
2672         return eb;
2673 }
2674
2675 static void __free_extent_buffer(struct extent_buffer *eb)
2676 {
2677         unsigned long flags;
2678         spin_lock_irqsave(&leak_lock, flags);
2679         list_del(&eb->leak_list);
2680         spin_unlock_irqrestore(&leak_lock, flags);
2681         kmem_cache_free(extent_buffer_cache, eb);
2682 }
2683
2684 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
2685                                           u64 start, unsigned long len,
2686                                           struct page *page0,
2687                                           gfp_t mask)
2688 {
2689         unsigned long num_pages = num_extent_pages(start, len);
2690         unsigned long i;
2691         unsigned long index = start >> PAGE_CACHE_SHIFT;
2692         struct extent_buffer *eb;
2693         struct page *p;
2694         struct address_space *mapping = tree->mapping;
2695         int uptodate = 1;
2696
2697         eb = __alloc_extent_buffer(tree, start, len, mask);
2698         if (!eb)
2699                 return NULL;
2700
2701         if (eb->flags & EXTENT_BUFFER_FILLED)
2702                 goto lru_add;
2703
2704         if (page0) {
2705                 eb->first_page = page0;
2706                 i = 1;
2707                 index++;
2708                 page_cache_get(page0);
2709                 mark_page_accessed(page0);
2710                 set_page_extent_mapped(page0);
2711                 set_page_extent_head(page0, len);
2712                 uptodate = PageUptodate(page0);
2713         } else {
2714                 i = 0;
2715         }
2716         for (; i < num_pages; i++, index++) {
2717                 p = find_or_create_page(mapping, index, mask | __GFP_HIGHMEM);
2718                 if (!p) {
2719                         WARN_ON(1);
2720                         goto fail;
2721                 }
2722                 set_page_extent_mapped(p);
2723                 mark_page_accessed(p);
2724                 if (i == 0) {
2725                         eb->first_page = p;
2726                         set_page_extent_head(p, len);
2727                 } else {
2728                         set_page_private(p, EXTENT_PAGE_PRIVATE);
2729                 }
2730                 if (!PageUptodate(p))
2731                         uptodate = 0;
2732                 unlock_page(p);
2733         }
2734         if (uptodate)
2735                 eb->flags |= EXTENT_UPTODATE;
2736         eb->flags |= EXTENT_BUFFER_FILLED;
2737
2738 lru_add:
2739         spin_lock(&tree->lru_lock);
2740         add_lru(tree, eb);
2741         spin_unlock(&tree->lru_lock);
2742         return eb;
2743
2744 fail:
2745         spin_lock(&tree->lru_lock);
2746         list_del_init(&eb->lru);
2747         spin_unlock(&tree->lru_lock);
2748         if (!atomic_dec_and_test(&eb->refs))
2749                 return NULL;
2750         for (index = 1; index < i; index++) {
2751                 page_cache_release(extent_buffer_page(eb, index));
2752         }
2753         if (i > 0)
2754                 page_cache_release(extent_buffer_page(eb, 0));
2755         __free_extent_buffer(eb);
2756         return NULL;
2757 }
2758 EXPORT_SYMBOL(alloc_extent_buffer);
2759
2760 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
2761                                          u64 start, unsigned long len,
2762                                           gfp_t mask)
2763 {
2764         unsigned long num_pages = num_extent_pages(start, len);
2765         unsigned long i;
2766         unsigned long index = start >> PAGE_CACHE_SHIFT;
2767         struct extent_buffer *eb;
2768         struct page *p;
2769         struct address_space *mapping = tree->mapping;
2770         int uptodate = 1;
2771
2772         eb = __alloc_extent_buffer(tree, start, len, mask);
2773         if (!eb)
2774                 return NULL;
2775
2776         if (eb->flags & EXTENT_BUFFER_FILLED)
2777                 goto lru_add;
2778
2779         for (i = 0; i < num_pages; i++, index++) {
2780                 p = find_lock_page(mapping, index);
2781                 if (!p) {
2782                         goto fail;
2783                 }
2784                 set_page_extent_mapped(p);
2785                 mark_page_accessed(p);
2786
2787                 if (i == 0) {
2788                         eb->first_page = p;
2789                         set_page_extent_head(p, len);
2790                 } else {
2791                         set_page_private(p, EXTENT_PAGE_PRIVATE);
2792                 }
2793
2794                 if (!PageUptodate(p))
2795                         uptodate = 0;
2796                 unlock_page(p);
2797         }
2798         if (uptodate)
2799                 eb->flags |= EXTENT_UPTODATE;
2800         eb->flags |= EXTENT_BUFFER_FILLED;
2801
2802 lru_add:
2803         spin_lock(&tree->lru_lock);
2804         add_lru(tree, eb);
2805         spin_unlock(&tree->lru_lock);
2806         return eb;
2807 fail:
2808         spin_lock(&tree->lru_lock);
2809         list_del_init(&eb->lru);
2810         spin_unlock(&tree->lru_lock);
2811         if (!atomic_dec_and_test(&eb->refs))
2812                 return NULL;
2813         for (index = 1; index < i; index++) {
2814                 page_cache_release(extent_buffer_page(eb, index));
2815         }
2816         if (i > 0)
2817                 page_cache_release(extent_buffer_page(eb, 0));
2818         __free_extent_buffer(eb);
2819         return NULL;
2820 }
2821 EXPORT_SYMBOL(find_extent_buffer);
2822
2823 void free_extent_buffer(struct extent_buffer *eb)
2824 {
2825         unsigned long i;
2826         unsigned long num_pages;
2827
2828         if (!eb)
2829                 return;
2830
2831         if (!atomic_dec_and_test(&eb->refs))
2832                 return;
2833
2834         WARN_ON(!list_empty(&eb->lru));
2835         num_pages = num_extent_pages(eb->start, eb->len);
2836
2837         for (i = 1; i < num_pages; i++) {
2838                 page_cache_release(extent_buffer_page(eb, i));
2839         }
2840         page_cache_release(extent_buffer_page(eb, 0));
2841         __free_extent_buffer(eb);
2842 }
2843 EXPORT_SYMBOL(free_extent_buffer);
2844
2845 int clear_extent_buffer_dirty(struct extent_io_tree *tree,
2846                               struct extent_buffer *eb)
2847 {
2848         int set;
2849         unsigned long i;
2850         unsigned long num_pages;
2851         struct page *page;
2852
2853         u64 start = eb->start;
2854         u64 end = start + eb->len - 1;
2855
2856         set = clear_extent_dirty(tree, start, end, GFP_NOFS);
2857         num_pages = num_extent_pages(eb->start, eb->len);
2858
2859         for (i = 0; i < num_pages; i++) {
2860                 page = extent_buffer_page(eb, i);
2861                 lock_page(page);
2862                 if (i == 0)
2863                         set_page_extent_head(page, eb->len);
2864                 else
2865                         set_page_private(page, EXTENT_PAGE_PRIVATE);
2866
2867                 /*
2868                  * if we're on the last page or the first page and the
2869                  * block isn't aligned on a page boundary, do extra checks
2870                  * to make sure we don't clean page that is partially dirty
2871                  */
2872                 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
2873                     ((i == num_pages - 1) &&
2874                      ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
2875                         start = (u64)page->index << PAGE_CACHE_SHIFT;
2876                         end  = start + PAGE_CACHE_SIZE - 1;
2877                         if (test_range_bit(tree, start, end,
2878                                            EXTENT_DIRTY, 0)) {
2879                                 unlock_page(page);
2880                                 continue;
2881                         }
2882                 }
2883                 clear_page_dirty_for_io(page);
2884                 read_lock_irq(&page->mapping->tree_lock);
2885                 if (!PageDirty(page)) {
2886                         radix_tree_tag_clear(&page->mapping->page_tree,
2887                                                 page_index(page),
2888                                                 PAGECACHE_TAG_DIRTY);
2889                 }
2890                 read_unlock_irq(&page->mapping->tree_lock);
2891                 unlock_page(page);
2892         }
2893         return 0;
2894 }
2895 EXPORT_SYMBOL(clear_extent_buffer_dirty);
2896
2897 int wait_on_extent_buffer_writeback(struct extent_io_tree *tree,
2898                                     struct extent_buffer *eb)
2899 {
2900         return wait_on_extent_writeback(tree, eb->start,
2901                                         eb->start + eb->len - 1);
2902 }
2903 EXPORT_SYMBOL(wait_on_extent_buffer_writeback);
2904
2905 int set_extent_buffer_dirty(struct extent_io_tree *tree,
2906                              struct extent_buffer *eb)
2907 {
2908         unsigned long i;
2909         unsigned long num_pages;
2910
2911         num_pages = num_extent_pages(eb->start, eb->len);
2912         for (i = 0; i < num_pages; i++) {
2913                 struct page *page = extent_buffer_page(eb, i);
2914                 /* writepage may need to do something special for the
2915                  * first page, we have to make sure page->private is
2916                  * properly set.  releasepage may drop page->private
2917                  * on us if the page isn't already dirty.
2918                  */
2919                 if (i == 0) {
2920                         lock_page(page);
2921                         set_page_extent_head(page, eb->len);
2922                 } else if (PagePrivate(page) &&
2923                            page->private != EXTENT_PAGE_PRIVATE) {
2924                         lock_page(page);
2925                         set_page_extent_mapped(page);
2926                         unlock_page(page);
2927                 }
2928                 __set_page_dirty_nobuffers(extent_buffer_page(eb, i));
2929                 if (i == 0)
2930                         unlock_page(page);
2931         }
2932         return set_extent_dirty(tree, eb->start,
2933                                 eb->start + eb->len - 1, GFP_NOFS);
2934 }
2935 EXPORT_SYMBOL(set_extent_buffer_dirty);
2936
2937 int set_extent_buffer_uptodate(struct extent_io_tree *tree,
2938                                 struct extent_buffer *eb)
2939 {
2940         unsigned long i;
2941         struct page *page;
2942         unsigned long num_pages;
2943
2944         num_pages = num_extent_pages(eb->start, eb->len);
2945
2946         set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
2947                             GFP_NOFS);
2948         for (i = 0; i < num_pages; i++) {
2949                 page = extent_buffer_page(eb, i);
2950                 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
2951                     ((i == num_pages - 1) &&
2952                      ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
2953                         check_page_uptodate(tree, page);
2954                         continue;
2955                 }
2956                 SetPageUptodate(page);
2957         }
2958         return 0;
2959 }
2960 EXPORT_SYMBOL(set_extent_buffer_uptodate);
2961
2962 int extent_range_uptodate(struct extent_io_tree *tree,
2963                           u64 start, u64 end)
2964 {
2965         struct page *page;
2966         int ret;
2967         int pg_uptodate = 1;
2968         int uptodate;
2969         unsigned long index;
2970
2971         ret = test_range_bit(tree, start, end, EXTENT_UPTODATE, 1);
2972         if (ret)
2973                 return 1;
2974         while(start <= end) {
2975                 index = start >> PAGE_CACHE_SHIFT;
2976                 page = find_get_page(tree->mapping, index);
2977                 uptodate = PageUptodate(page);
2978                 page_cache_release(page);
2979                 if (!uptodate) {
2980                         pg_uptodate = 0;
2981                         break;
2982                 }
2983                 start += PAGE_CACHE_SIZE;
2984         }
2985         return pg_uptodate;
2986 }
2987
2988 int extent_buffer_uptodate(struct extent_io_tree *tree,
2989                            struct extent_buffer *eb)
2990 {
2991         int ret = 0;
2992         int ret2;
2993         unsigned long num_pages;
2994         unsigned long i;
2995         struct page *page;
2996         int pg_uptodate = 1;
2997
2998         if (eb->flags & EXTENT_UPTODATE)
2999                 ret = 1;
3000
3001         ret2  = test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3002                            EXTENT_UPTODATE, 1);
3003
3004         num_pages = num_extent_pages(eb->start, eb->len);
3005         for (i = 0; i < num_pages; i++) {
3006                 page = extent_buffer_page(eb, i);
3007                 if (!PageUptodate(page)) {
3008                         pg_uptodate = 0;
3009                         break;
3010                 }
3011         }
3012         if ((ret || ret2) && !pg_uptodate) {
3013 printk("uptodate error2 eb %Lu ret %d ret2 %d pg_uptodate %d\n", eb->start, ret, ret2, pg_uptodate);
3014                 WARN_ON(1);
3015         }
3016         return (ret || ret2);
3017 }
3018 EXPORT_SYMBOL(extent_buffer_uptodate);
3019
3020 int read_extent_buffer_pages(struct extent_io_tree *tree,
3021                              struct extent_buffer *eb,
3022                              u64 start, int wait,
3023                              get_extent_t *get_extent, int mirror_num)
3024 {
3025         unsigned long i;
3026         unsigned long start_i;
3027         struct page *page;
3028         int err;
3029         int ret = 0;
3030         int locked_pages = 0;
3031         int all_uptodate = 1;
3032         int inc_all_pages = 0;
3033         unsigned long num_pages;
3034         struct bio *bio = NULL;
3035
3036         if (eb->flags & EXTENT_UPTODATE)
3037                 return 0;
3038
3039         if (test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3040                            EXTENT_UPTODATE, 1)) {
3041                 return 0;
3042         }
3043
3044         if (start) {
3045                 WARN_ON(start < eb->start);
3046                 start_i = (start >> PAGE_CACHE_SHIFT) -
3047                         (eb->start >> PAGE_CACHE_SHIFT);
3048         } else {
3049                 start_i = 0;
3050         }
3051
3052         num_pages = num_extent_pages(eb->start, eb->len);
3053         for (i = start_i; i < num_pages; i++) {
3054                 page = extent_buffer_page(eb, i);
3055                 if (!wait) {
3056                         if (TestSetPageLocked(page))
3057                                 goto unlock_exit;
3058                 } else {
3059                         lock_page(page);
3060                 }
3061                 locked_pages++;
3062                 if (!PageUptodate(page)) {
3063                         all_uptodate = 0;
3064                 }
3065         }
3066         if (all_uptodate) {
3067                 if (start_i == 0)
3068                         eb->flags |= EXTENT_UPTODATE;
3069                 goto unlock_exit;
3070         }
3071
3072         for (i = start_i; i < num_pages; i++) {
3073                 page = extent_buffer_page(eb, i);
3074                 if (inc_all_pages)
3075                         page_cache_get(page);
3076                 if (!PageUptodate(page)) {
3077                         if (start_i == 0)
3078                                 inc_all_pages = 1;
3079                         ClearPageError(page);
3080                         err = __extent_read_full_page(tree, page,
3081                                                       get_extent, &bio,
3082                                                       mirror_num);
3083                         if (err) {
3084                                 ret = err;
3085                         }
3086                 } else {
3087                         unlock_page(page);
3088                 }
3089         }
3090
3091         if (bio)
3092                 submit_one_bio(READ, bio, mirror_num);
3093
3094         if (ret || !wait) {
3095                 return ret;
3096         }
3097         for (i = start_i; i < num_pages; i++) {
3098                 page = extent_buffer_page(eb, i);
3099                 wait_on_page_locked(page);
3100                 if (!PageUptodate(page)) {
3101                         ret = -EIO;
3102                 }
3103         }
3104         if (!ret)
3105                 eb->flags |= EXTENT_UPTODATE;
3106         return ret;
3107
3108 unlock_exit:
3109         i = start_i;
3110         while(locked_pages > 0) {
3111                 page = extent_buffer_page(eb, i);
3112                 i++;
3113                 unlock_page(page);
3114                 locked_pages--;
3115         }
3116         return ret;
3117 }
3118 EXPORT_SYMBOL(read_extent_buffer_pages);
3119
3120 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
3121                         unsigned long start,
3122                         unsigned long len)
3123 {
3124         size_t cur;
3125         size_t offset;
3126         struct page *page;
3127         char *kaddr;
3128         char *dst = (char *)dstv;
3129         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3130         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3131
3132         WARN_ON(start > eb->len);
3133         WARN_ON(start + len > eb->start + eb->len);
3134
3135         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3136
3137         while(len > 0) {
3138                 page = extent_buffer_page(eb, i);
3139
3140                 cur = min(len, (PAGE_CACHE_SIZE - offset));
3141                 kaddr = kmap_atomic(page, KM_USER1);
3142                 memcpy(dst, kaddr + offset, cur);
3143                 kunmap_atomic(kaddr, KM_USER1);
3144
3145                 dst += cur;
3146                 len -= cur;
3147                 offset = 0;
3148                 i++;
3149         }
3150 }
3151 EXPORT_SYMBOL(read_extent_buffer);
3152
3153 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
3154                                unsigned long min_len, char **token, char **map,
3155                                unsigned long *map_start,
3156                                unsigned long *map_len, int km)
3157 {
3158         size_t offset = start & (PAGE_CACHE_SIZE - 1);
3159         char *kaddr;
3160         struct page *p;
3161         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3162         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3163         unsigned long end_i = (start_offset + start + min_len - 1) >>
3164                 PAGE_CACHE_SHIFT;
3165
3166         if (i != end_i)
3167                 return -EINVAL;
3168
3169         if (i == 0) {
3170                 offset = start_offset;
3171                 *map_start = 0;
3172         } else {
3173                 offset = 0;
3174                 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
3175         }
3176         if (start + min_len > eb->len) {
3177 printk("bad mapping eb start %Lu len %lu, wanted %lu %lu\n", eb->start, eb->len, start, min_len);
3178                 WARN_ON(1);
3179         }
3180
3181         p = extent_buffer_page(eb, i);
3182         kaddr = kmap_atomic(p, km);
3183         *token = kaddr;
3184         *map = kaddr + offset;
3185         *map_len = PAGE_CACHE_SIZE - offset;
3186         return 0;
3187 }
3188 EXPORT_SYMBOL(map_private_extent_buffer);
3189
3190 int map_extent_buffer(struct extent_buffer *eb, unsigned long start,
3191                       unsigned long min_len,
3192                       char **token, char **map,
3193                       unsigned long *map_start,
3194                       unsigned long *map_len, int km)
3195 {
3196         int err;
3197         int save = 0;
3198         if (eb->map_token) {
3199                 unmap_extent_buffer(eb, eb->map_token, km);
3200                 eb->map_token = NULL;
3201                 save = 1;
3202         }
3203         err = map_private_extent_buffer(eb, start, min_len, token, map,
3204                                        map_start, map_len, km);
3205         if (!err && save) {
3206                 eb->map_token = *token;
3207                 eb->kaddr = *map;
3208                 eb->map_start = *map_start;
3209                 eb->map_len = *map_len;
3210         }
3211         return err;
3212 }
3213 EXPORT_SYMBOL(map_extent_buffer);
3214
3215 void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km)
3216 {
3217         kunmap_atomic(token, km);
3218 }
3219 EXPORT_SYMBOL(unmap_extent_buffer);
3220
3221 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
3222                           unsigned long start,
3223                           unsigned long len)
3224 {
3225         size_t cur;
3226         size_t offset;
3227         struct page *page;
3228         char *kaddr;
3229         char *ptr = (char *)ptrv;
3230         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3231         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3232         int ret = 0;
3233
3234         WARN_ON(start > eb->len);
3235         WARN_ON(start + len > eb->start + eb->len);
3236
3237         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3238
3239         while(len > 0) {
3240                 page = extent_buffer_page(eb, i);
3241
3242                 cur = min(len, (PAGE_CACHE_SIZE - offset));
3243
3244                 kaddr = kmap_atomic(page, KM_USER0);
3245                 ret = memcmp(ptr, kaddr + offset, cur);
3246                 kunmap_atomic(kaddr, KM_USER0);
3247                 if (ret)
3248                         break;
3249
3250                 ptr += cur;
3251                 len -= cur;
3252                 offset = 0;
3253                 i++;
3254         }
3255         return ret;
3256 }
3257 EXPORT_SYMBOL(memcmp_extent_buffer);
3258
3259 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
3260                          unsigned long start, unsigned long len)
3261 {
3262         size_t cur;
3263         size_t offset;
3264         struct page *page;
3265         char *kaddr;
3266         char *src = (char *)srcv;
3267         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3268         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3269
3270         WARN_ON(start > eb->len);
3271         WARN_ON(start + len > eb->start + eb->len);
3272
3273         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3274
3275         while(len > 0) {
3276                 page = extent_buffer_page(eb, i);
3277                 WARN_ON(!PageUptodate(page));
3278
3279                 cur = min(len, PAGE_CACHE_SIZE - offset);
3280                 kaddr = kmap_atomic(page, KM_USER1);
3281                 memcpy(kaddr + offset, src, cur);
3282                 kunmap_atomic(kaddr, KM_USER1);
3283
3284                 src += cur;
3285                 len -= cur;
3286                 offset = 0;
3287                 i++;
3288         }
3289 }
3290 EXPORT_SYMBOL(write_extent_buffer);
3291
3292 void memset_extent_buffer(struct extent_buffer *eb, char c,
3293                           unsigned long start, unsigned long len)
3294 {
3295         size_t cur;
3296         size_t offset;
3297         struct page *page;
3298         char *kaddr;
3299         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3300         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3301
3302         WARN_ON(start > eb->len);
3303         WARN_ON(start + len > eb->start + eb->len);
3304
3305         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3306
3307         while(len > 0) {
3308                 page = extent_buffer_page(eb, i);
3309                 WARN_ON(!PageUptodate(page));
3310
3311                 cur = min(len, PAGE_CACHE_SIZE - offset);
3312                 kaddr = kmap_atomic(page, KM_USER0);
3313                 memset(kaddr + offset, c, cur);
3314                 kunmap_atomic(kaddr, KM_USER0);
3315
3316                 len -= cur;
3317                 offset = 0;
3318                 i++;
3319         }
3320 }
3321 EXPORT_SYMBOL(memset_extent_buffer);
3322
3323 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
3324                         unsigned long dst_offset, unsigned long src_offset,
3325                         unsigned long len)
3326 {
3327         u64 dst_len = dst->len;
3328         size_t cur;
3329         size_t offset;
3330         struct page *page;
3331         char *kaddr;
3332         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3333         unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3334
3335         WARN_ON(src->len != dst_len);
3336
3337         offset = (start_offset + dst_offset) &
3338                 ((unsigned long)PAGE_CACHE_SIZE - 1);
3339
3340         while(len > 0) {
3341                 page = extent_buffer_page(dst, i);
3342                 WARN_ON(!PageUptodate(page));
3343
3344                 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
3345
3346                 kaddr = kmap_atomic(page, KM_USER0);
3347                 read_extent_buffer(src, kaddr + offset, src_offset, cur);
3348                 kunmap_atomic(kaddr, KM_USER0);
3349
3350                 src_offset += cur;
3351                 len -= cur;
3352                 offset = 0;
3353                 i++;
3354         }
3355 }
3356 EXPORT_SYMBOL(copy_extent_buffer);
3357
3358 static void move_pages(struct page *dst_page, struct page *src_page,
3359                        unsigned long dst_off, unsigned long src_off,
3360                        unsigned long len)
3361 {
3362         char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3363         if (dst_page == src_page) {
3364                 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
3365         } else {
3366                 char *src_kaddr = kmap_atomic(src_page, KM_USER1);
3367                 char *p = dst_kaddr + dst_off + len;
3368                 char *s = src_kaddr + src_off + len;
3369
3370                 while (len--)
3371                         *--p = *--s;
3372
3373                 kunmap_atomic(src_kaddr, KM_USER1);
3374         }
3375         kunmap_atomic(dst_kaddr, KM_USER0);
3376 }
3377
3378 static void copy_pages(struct page *dst_page, struct page *src_page,
3379                        unsigned long dst_off, unsigned long src_off,
3380                        unsigned long len)
3381 {
3382         char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3383         char *src_kaddr;
3384
3385         if (dst_page != src_page)
3386                 src_kaddr = kmap_atomic(src_page, KM_USER1);
3387         else
3388                 src_kaddr = dst_kaddr;
3389
3390         memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
3391         kunmap_atomic(dst_kaddr, KM_USER0);
3392         if (dst_page != src_page)
3393                 kunmap_atomic(src_kaddr, KM_USER1);
3394 }
3395
3396 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3397                            unsigned long src_offset, unsigned long len)
3398 {
3399         size_t cur;
3400         size_t dst_off_in_page;
3401         size_t src_off_in_page;
3402         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3403         unsigned long dst_i;
3404         unsigned long src_i;
3405
3406         if (src_offset + len > dst->len) {
3407                 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
3408                        src_offset, len, dst->len);
3409                 BUG_ON(1);
3410         }
3411         if (dst_offset + len > dst->len) {
3412                 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
3413                        dst_offset, len, dst->len);
3414                 BUG_ON(1);
3415         }
3416
3417         while(len > 0) {
3418                 dst_off_in_page = (start_offset + dst_offset) &
3419                         ((unsigned long)PAGE_CACHE_SIZE - 1);
3420                 src_off_in_page = (start_offset + src_offset) &
3421                         ((unsigned long)PAGE_CACHE_SIZE - 1);
3422
3423                 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3424                 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
3425
3426                 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
3427                                                src_off_in_page));
3428                 cur = min_t(unsigned long, cur,
3429                         (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
3430
3431                 copy_pages(extent_buffer_page(dst, dst_i),
3432                            extent_buffer_page(dst, src_i),
3433                            dst_off_in_page, src_off_in_page, cur);
3434
3435                 src_offset += cur;
3436                 dst_offset += cur;
3437                 len -= cur;
3438         }
3439 }
3440 EXPORT_SYMBOL(memcpy_extent_buffer);
3441
3442 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3443                            unsigned long src_offset, unsigned long len)
3444 {
3445         size_t cur;
3446         size_t dst_off_in_page;
3447         size_t src_off_in_page;
3448         unsigned long dst_end = dst_offset + len - 1;
3449         unsigned long src_end = src_offset + len - 1;
3450         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3451         unsigned long dst_i;
3452         unsigned long src_i;
3453
3454         if (src_offset + len > dst->len) {
3455                 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
3456                        src_offset, len, dst->len);
3457                 BUG_ON(1);
3458         }
3459         if (dst_offset + len > dst->len) {
3460                 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
3461                        dst_offset, len, dst->len);
3462                 BUG_ON(1);
3463         }
3464         if (dst_offset < src_offset) {
3465                 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
3466                 return;
3467         }
3468         while(len > 0) {
3469                 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
3470                 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
3471
3472                 dst_off_in_page = (start_offset + dst_end) &
3473                         ((unsigned long)PAGE_CACHE_SIZE - 1);
3474                 src_off_in_page = (start_offset + src_end) &
3475                         ((unsigned long)PAGE_CACHE_SIZE - 1);
3476
3477                 cur = min_t(unsigned long, len, src_off_in_page + 1);
3478                 cur = min(cur, dst_off_in_page + 1);
3479                 move_pages(extent_buffer_page(dst, dst_i),
3480                            extent_buffer_page(dst, src_i),
3481                            dst_off_in_page - cur + 1,
3482                            src_off_in_page - cur + 1, cur);
3483
3484                 dst_end -= cur;
3485                 src_end -= cur;
3486                 len -= cur;
3487         }
3488 }
3489 EXPORT_SYMBOL(memmove_extent_buffer);