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

   1 #include <linux/bitops.h>
   2 #include <linux/slab.h>
   3 #include <linux/bio.h>
   4 #include <linux/mm.h>
   5 #include <linux/gfp.h>
   6 #include <linux/pagemap.h>
   7 #include <linux/page-flags.h>
   8 #include <linux/module.h>
   9 #include <linux/spinlock.h>
  10 #include <linux/blkdev.h>
  11 #include <linux/swap.h>
  12 #include <linux/version.h>
  13 #include <linux/writeback.h>
  14 #include <linux/pagevec.h>
  15 #include "extent_io.h"
  16 #include "extent_map.h"
  17
  18 /* temporary define until extent_map moves out of btrfs */
  19 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
  20                                        unsigned long extra_flags,
  21                                        void (*ctor)(void *, struct kmem_cache *,
  22                                                     unsigned long));
  23
  24 static struct kmem_cache *extent_state_cache;
  25 static struct kmem_cache *extent_buffer_cache;
  26
  27 static LIST_HEAD(buffers);
  28 static LIST_HEAD(states);
  29 static spinlock_t leak_lock = SPIN_LOCK_UNLOCKED;
  30
  31 #define BUFFER_LRU_MAX 64
  32
  33 struct tree_entry {
  34         u64 start;
  35         u64 end;
  36         struct rb_node rb_node;
  37 };
  38
  39 struct extent_page_data {
  40         struct bio *bio;
  41         struct extent_io_tree *tree;
  42         get_extent_t *get_extent;
  43 };
  44
  45 int __init extent_io_init(void)
  46 {
  47         extent_state_cache = btrfs_cache_create("extent_state",
  48                                             sizeof(struct extent_state), 0,
  49                                             NULL);
  50         if (!extent_state_cache)
  51                 return -ENOMEM;
  52
  53         extent_buffer_cache = btrfs_cache_create("extent_buffers",
  54                                             sizeof(struct extent_buffer), 0,
  55                                             NULL);
  56         if (!extent_buffer_cache)
  57                 goto free_state_cache;
  58         return 0;
  59
  60 free_state_cache:
  61         kmem_cache_destroy(extent_state_cache);
  62         return -ENOMEM;
  63 }
  64
  65 void extent_io_exit(void)
  66 {
  67         struct extent_state *state;
  68         struct extent_buffer *eb;
  69
  70         while (!list_empty(&states)) {
  71                 state = list_entry(states.next, struct extent_state, leak_list);
  72                 printk("state leak: start %Lu end %Lu state %lu in tree %p refs %d\n", state->start, state->end, state->state, state->tree, atomic_read(&state->refs));
  73                 list_del(&state->leak_list);
  74                 kmem_cache_free(extent_state_cache, state);
  75
  76         }
  77
  78         while (!list_empty(&buffers)) {
  79                 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
  80                 printk("buffer leak start %Lu len %lu refs %d\n", eb->start, eb->len, atomic_read(&eb->refs));
  81                 list_del(&eb->leak_list);
  82                 kmem_cache_free(extent_buffer_cache, eb);
  83         }
  84         if (extent_state_cache)
  85                 kmem_cache_destroy(extent_state_cache);
  86         if (extent_buffer_cache)
  87                 kmem_cache_destroy(extent_buffer_cache);
  88 }
  89
  90 void extent_io_tree_init(struct extent_io_tree *tree,
  91                           struct address_space *mapping, gfp_t mask)
  92 {
  93         tree->state.rb_node = NULL;
  94         tree->ops = NULL;
  95         tree->dirty_bytes = 0;
  96         spin_lock_init(&tree->lock);
  97         spin_lock_init(&tree->lru_lock);
  98         tree->mapping = mapping;
  99         INIT_LIST_HEAD(&tree->buffer_lru);
 100         tree->lru_size = 0;
 101         tree->last = NULL;
 102 }
 103 EXPORT_SYMBOL(extent_io_tree_init);
 104
 105 void extent_io_tree_empty_lru(struct extent_io_tree *tree)
 106 {
 107         struct extent_buffer *eb;
 108         while(!list_empty(&tree->buffer_lru)) {
 109                 eb = list_entry(tree->buffer_lru.next, struct extent_buffer,
 110                                 lru);
 111                 list_del_init(&eb->lru);
 112                 free_extent_buffer(eb);
 113         }
 114 }
 115 EXPORT_SYMBOL(extent_io_tree_empty_lru);
 116
 117 struct extent_state *alloc_extent_state(gfp_t mask)
 118 {
 119         struct extent_state *state;
 120         unsigned long flags;
 121
 122         state = kmem_cache_alloc(extent_state_cache, mask);
 123         if (!state)
 124                 return state;
 125         state->state = 0;
 126         state->private = 0;
 127         state->tree = NULL;
 128         spin_lock_irqsave(&leak_lock, flags);
 129         list_add(&state->leak_list, &states);
 130         spin_unlock_irqrestore(&leak_lock, flags);
 131
 132         atomic_set(&state->refs, 1);
 133         init_waitqueue_head(&state->wq);
 134         return state;
 135 }
 136 EXPORT_SYMBOL(alloc_extent_state);
 137
 138 void free_extent_state(struct extent_state *state)
 139 {
 140         if (!state)
 141                 return;
 142         if (atomic_dec_and_test(&state->refs)) {
 143                 unsigned long flags;
 144                 WARN_ON(state->tree);
 145                 spin_lock_irqsave(&leak_lock, flags);
 146                 list_del(&state->leak_list);
 147                 spin_unlock_irqrestore(&leak_lock, flags);
 148                 kmem_cache_free(extent_state_cache, state);
 149         }
 150 }
 151 EXPORT_SYMBOL(free_extent_state);
 152
 153 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
 154                                    struct rb_node *node)
 155 {
 156         struct rb_node ** p = &root->rb_node;
 157         struct rb_node * parent = NULL;
 158         struct tree_entry *entry;
 159
 160         while(*p) {
 161                 parent = *p;
 162                 entry = rb_entry(parent, struct tree_entry, rb_node);
 163
 164                 if (offset < entry->start)
 165                         p = &(*p)->rb_left;
 166                 else if (offset > entry->end)
 167                         p = &(*p)->rb_right;
 168                 else
 169                         return parent;
 170         }
 171
 172         entry = rb_entry(node, struct tree_entry, rb_node);
 173         rb_link_node(node, parent, p);
 174         rb_insert_color(node, root);
 175         return NULL;
 176 }
 177
 178 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
 179                                      struct rb_node **prev_ret,
 180                                      struct rb_node **next_ret)
 181 {
 182         struct rb_root *root = &tree->state;
 183         struct rb_node * n = root->rb_node;
 184         struct rb_node *prev = NULL;
 185         struct rb_node *orig_prev = NULL;
 186         struct tree_entry *entry;
 187         struct tree_entry *prev_entry = NULL;
 188
 189         if (tree->last) {
 190                 struct extent_state *state;
 191                 state = tree->last;
 192                 if (state->start <= offset && offset <= state->end)
 193                         return &tree->last->rb_node;
 194         }
 195         while(n) {
 196                 entry = rb_entry(n, struct tree_entry, rb_node);
 197                 prev = n;
 198                 prev_entry = entry;
 199
 200                 if (offset < entry->start)
 201                         n = n->rb_left;
 202                 else if (offset > entry->end)
 203                         n = n->rb_right;
 204                 else {
 205                         tree->last = rb_entry(n, struct extent_state, rb_node);
 206                         return n;
 207                 }
 208         }
 209
 210         if (prev_ret) {
 211                 orig_prev = prev;
 212                 while(prev && offset > prev_entry->end) {
 213                         prev = rb_next(prev);
 214                         prev_entry = rb_entry(prev, struct tree_entry, rb_node);
 215                 }
 216                 *prev_ret = prev;
 217                 prev = orig_prev;
 218         }
 219
 220         if (next_ret) {
 221                 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
 222                 while(prev && offset < prev_entry->start) {
 223                         prev = rb_prev(prev);
 224                         prev_entry = rb_entry(prev, struct tree_entry, rb_node);
 225                 }
 226                 *next_ret = prev;
 227         }
 228         return NULL;
 229 }
 230
 231 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
 232                                           u64 offset)
 233 {
 234         struct rb_node *prev = NULL;
 235         struct rb_node *ret;
 236
 237         ret = __etree_search(tree, offset, &prev, NULL);
 238         if (!ret) {
 239                 if (prev) {
 240                         tree->last = rb_entry(prev, struct extent_state,
 241                                               rb_node);
 242                 }
 243                 return prev;
 244         }
 245         return ret;
 246 }
 247
 248 /*
 249  * utility function to look for merge candidates inside a given range.
 250  * Any extents with matching state are merged together into a single
 251  * extent in the tree.  Extents with EXTENT_IO in their state field
 252  * are not merged because the end_io handlers need to be able to do
 253  * operations on them without sleeping (or doing allocations/splits).
 254  *
 255  * This should be called with the tree lock held.
 256  */
 257 static int merge_state(struct extent_io_tree *tree,
 258                        struct extent_state *state)
 259 {
 260         struct extent_state *other;
 261         struct rb_node *other_node;
 262
 263         if (state->state & EXTENT_IOBITS)
 264                 return 0;
 265
 266         other_node = rb_prev(&state->rb_node);
 267         if (other_node) {
 268                 other = rb_entry(other_node, struct extent_state, rb_node);
 269                 if (other->end == state->start - 1 &&
 270                     other->state == state->state) {
 271                         state->start = other->start;
 272                         other->tree = NULL;
 273                         if (tree->last == other)
 274                                 tree->last = state;
 275                         rb_erase(&other->rb_node, &tree->state);
 276                         free_extent_state(other);
 277                 }
 278         }
 279         other_node = rb_next(&state->rb_node);
 280         if (other_node) {
 281                 other = rb_entry(other_node, struct extent_state, rb_node);
 282                 if (other->start == state->end + 1 &&
 283                     other->state == state->state) {
 284                         other->start = state->start;
 285                         state->tree = NULL;
 286                         if (tree->last == state)
 287                                 tree->last = other;
 288                         rb_erase(&state->rb_node, &tree->state);
 289                         free_extent_state(state);
 290                 }
 291         }
 292         return 0;
 293 }
 294
 295 static void set_state_cb(struct extent_io_tree *tree,
 296                          struct extent_state *state,
 297                          unsigned long bits)
 298 {
 299         if (tree->ops && tree->ops->set_bit_hook) {
 300                 tree->ops->set_bit_hook(tree->mapping->host, state->start,
 301                                         state->end, state->state, bits);
 302         }
 303 }
 304
 305 static void clear_state_cb(struct extent_io_tree *tree,
 306                            struct extent_state *state,
 307                            unsigned long bits)
 308 {
 309         if (tree->ops && tree->ops->set_bit_hook) {
 310                 tree->ops->clear_bit_hook(tree->mapping->host, state->start,
 311                                           state->end, state->state, bits);
 312         }
 313 }
 314
 315 /*
 316  * insert an extent_state struct into the tree.  'bits' are set on the
 317  * struct before it is inserted.
 318  *
 319  * This may return -EEXIST if the extent is already there, in which case the
 320  * state struct is freed.
 321  *
 322  * The tree lock is not taken internally.  This is a utility function and
 323  * probably isn't what you want to call (see set/clear_extent_bit).
 324  */
 325 static int insert_state(struct extent_io_tree *tree,
 326                         struct extent_state *state, u64 start, u64 end,
 327                         int bits)
 328 {
 329         struct rb_node *node;
 330
 331         if (end < start) {
 332                 printk("end < start %Lu %Lu\n", end, start);
 333                 WARN_ON(1);
 334         }
 335         if (bits & EXTENT_DIRTY)
 336                 tree->dirty_bytes += end - start + 1;
 337         set_state_cb(tree, state, bits);
 338         state->state |= bits;
 339         state->start = start;
 340         state->end = end;
 341         node = tree_insert(&tree->state, end, &state->rb_node);
 342         if (node) {
 343                 struct extent_state *found;
 344                 found = rb_entry(node, struct extent_state, rb_node);
 345                 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, start, end);
 346                 free_extent_state(state);
 347                 return -EEXIST;
 348         }
 349         state->tree = tree;
 350         tree->last = state;
 351         merge_state(tree, state);
 352         return 0;
 353 }
 354
 355 /*
 356  * split a given extent state struct in two, inserting the preallocated
 357  * struct 'prealloc' as the newly created second half.  'split' indicates an
 358  * offset inside 'orig' where it should be split.
 359  *
 360  * Before calling,
 361  * the tree has 'orig' at [orig->start, orig->end].  After calling, there
 362  * are two extent state structs in the tree:
 363  * prealloc: [orig->start, split - 1]
 364  * orig: [ split, orig->end ]
 365  *
 366  * The tree locks are not taken by this function. They need to be held
 367  * by the caller.
 368  */
 369 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
 370                        struct extent_state *prealloc, u64 split)
 371 {
 372         struct rb_node *node;
 373         prealloc->start = orig->start;
 374         prealloc->end = split - 1;
 375         prealloc->state = orig->state;
 376         orig->start = split;
 377
 378         node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
 379         if (node) {
 380                 struct extent_state *found;
 381                 found = rb_entry(node, struct extent_state, rb_node);
 382                 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, prealloc->start, prealloc->end);
 383                 free_extent_state(prealloc);
 384                 return -EEXIST;
 385         }
 386         prealloc->tree = tree;
 387         return 0;
 388 }
 389
 390 /*
 391  * utility function to clear some bits in an extent state struct.
 392  * it will optionally wake up any one waiting on this state (wake == 1), or
 393  * forcibly remove the state from the tree (delete == 1).
 394  *
 395  * If no bits are set on the state struct after clearing things, the
 396  * struct is freed and removed from the tree
 397  */
 398 static int clear_state_bit(struct extent_io_tree *tree,
 399                             struct extent_state *state, int bits, int wake,
 400                             int delete)
 401 {
 402         int ret = state->state & bits;
 403
 404         if ((bits & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
 405                 u64 range = state->end - state->start + 1;
 406                 WARN_ON(range > tree->dirty_bytes);
 407                 tree->dirty_bytes -= range;
 408         }
 409         clear_state_cb(tree, state, bits);
 410         state->state &= ~bits;
 411         if (wake)
 412                 wake_up(&state->wq);
 413         if (delete || state->state == 0) {
 414                 if (state->tree) {
 415                         clear_state_cb(tree, state, state->state);
 416                         if (tree->last == state) {
 417                                 tree->last = extent_state_next(state);
 418                         }
 419                         rb_erase(&state->rb_node, &tree->state);
 420                         state->tree = NULL;
 421                         free_extent_state(state);
 422                 } else {
 423                         WARN_ON(1);
 424                 }
 425         } else {
 426                 merge_state(tree, state);
 427         }
 428         return ret;
 429 }
 430
 431 /*
 432  * clear some bits on a range in the tree.  This may require splitting
 433  * or inserting elements in the tree, so the gfp mask is used to
 434  * indicate which allocations or sleeping are allowed.
 435  *
 436  * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
 437  * the given range from the tree regardless of state (ie for truncate).
 438  *
 439  * the range [start, end] is inclusive.
 440  *
 441  * This takes the tree lock, and returns < 0 on error, > 0 if any of the
 442  * bits were already set, or zero if none of the bits were already set.
 443  */
 444 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
 445                      int bits, int wake, int delete, gfp_t mask)
 446 {
 447         struct extent_state *state;
 448         struct extent_state *prealloc = NULL;
 449         struct rb_node *node;
 450         unsigned long flags;
 451         int err;
 452         int set = 0;
 453
 454 again:
 455         if (!prealloc && (mask & __GFP_WAIT)) {
 456                 prealloc = alloc_extent_state(mask);
 457                 if (!prealloc)
 458                         return -ENOMEM;
 459         }
 460
 461         spin_lock_irqsave(&tree->lock, flags);
 462         /*
 463          * this search will find the extents that end after
 464          * our range starts
 465          */
 466         node = tree_search(tree, start);
 467         if (!node)
 468                 goto out;
 469         state = rb_entry(node, struct extent_state, rb_node);
 470         if (state->start > end)
 471                 goto out;
 472         WARN_ON(state->end < start);
 473
 474         /*
 475          *     | ---- desired range ---- |
 476          *  | state | or
 477          *  | ------------- state -------------- |
 478          *
 479          * We need to split the extent we found, and may flip
 480          * bits on second half.
 481          *
 482          * If the extent we found extends past our range, we
 483          * just split and search again.  It'll get split again
 484          * the next time though.
 485          *
 486          * If the extent we found is inside our range, we clear
 487          * the desired bit on it.
 488          */
 489
 490         if (state->start < start) {
 491                 if (!prealloc)
 492                         prealloc = alloc_extent_state(GFP_ATOMIC);
 493                 err = split_state(tree, state, prealloc, start);
 494                 BUG_ON(err == -EEXIST);
 495                 prealloc = NULL;
 496                 if (err)
 497                         goto out;
 498                 if (state->end <= end) {
 499                         start = state->end + 1;
 500                         set |= clear_state_bit(tree, state, bits,
 501                                         wake, delete);
 502                 } else {
 503                         start = state->start;
 504                 }
 505                 goto search_again;
 506         }
 507         /*
 508          * | ---- desired range ---- |
 509          *                        | state |
 510          * We need to split the extent, and clear the bit
 511          * on the first half
 512          */
 513         if (state->start <= end && state->end > end) {
 514                 if (!prealloc)
 515                         prealloc = alloc_extent_state(GFP_ATOMIC);
 516                 err = split_state(tree, state, prealloc, end + 1);
 517                 BUG_ON(err == -EEXIST);
 518
 519                 if (wake)
 520                         wake_up(&state->wq);
 521                 set |= clear_state_bit(tree, prealloc, bits,
 522                                        wake, delete);
 523                 prealloc = NULL;
 524                 goto out;
 525         }
 526
 527         start = state->end + 1;
 528         set |= clear_state_bit(tree, state, bits, wake, delete);
 529         goto search_again;
 530
 531 out:
 532         spin_unlock_irqrestore(&tree->lock, flags);
 533         if (prealloc)
 534                 free_extent_state(prealloc);
 535
 536         return set;
 537
 538 search_again:
 539         if (start > end)
 540                 goto out;
 541         spin_unlock_irqrestore(&tree->lock, flags);
 542         if (mask & __GFP_WAIT)
 543                 cond_resched();
 544         goto again;
 545 }
 546 EXPORT_SYMBOL(clear_extent_bit);
 547
 548 static int wait_on_state(struct extent_io_tree *tree,
 549                          struct extent_state *state)
 550 {
 551         DEFINE_WAIT(wait);
 552         prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
 553         spin_unlock_irq(&tree->lock);
 554         schedule();
 555         spin_lock_irq(&tree->lock);
 556         finish_wait(&state->wq, &wait);
 557         return 0;
 558 }
 559
 560 /*
 561  * waits for one or more bits to clear on a range in the state tree.
 562  * The range [start, end] is inclusive.
 563  * The tree lock is taken by this function
 564  */
 565 int wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits)
 566 {
 567         struct extent_state *state;
 568         struct rb_node *node;
 569
 570         spin_lock_irq(&tree->lock);
 571 again:
 572         while (1) {
 573                 /*
 574                  * this search will find all the extents that end after
 575                  * our range starts
 576                  */
 577                 node = tree_search(tree, start);
 578                 if (!node)
 579                         break;
 580
 581                 state = rb_entry(node, struct extent_state, rb_node);
 582
 583                 if (state->start > end)
 584                         goto out;
 585
 586                 if (state->state & bits) {
 587                         start = state->start;
 588                         atomic_inc(&state->refs);
 589                         wait_on_state(tree, state);
 590                         free_extent_state(state);
 591                         goto again;
 592                 }
 593                 start = state->end + 1;
 594
 595                 if (start > end)
 596                         break;
 597
 598                 if (need_resched()) {
 599                         spin_unlock_irq(&tree->lock);
 600                         cond_resched();
 601                         spin_lock_irq(&tree->lock);
 602                 }
 603         }
 604 out:
 605         spin_unlock_irq(&tree->lock);
 606         return 0;
 607 }
 608 EXPORT_SYMBOL(wait_extent_bit);
 609
 610 static void set_state_bits(struct extent_io_tree *tree,
 611                            struct extent_state *state,
 612                            int bits)
 613 {
 614         if ((bits & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
 615                 u64 range = state->end - state->start + 1;
 616                 tree->dirty_bytes += range;
 617         }
 618         set_state_cb(tree, state, bits);
 619         state->state |= bits;
 620 }
 621
 622 /*
 623  * set some bits on a range in the tree.  This may require allocations
 624  * or sleeping, so the gfp mask is used to indicate what is allowed.
 625  *
 626  * If 'exclusive' == 1, this will fail with -EEXIST if some part of the
 627  * range already has the desired bits set.  The start of the existing
 628  * range is returned in failed_start in this case.
 629  *
 630  * [start, end] is inclusive
 631  * This takes the tree lock.
 632  */
 633 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits,
 634                    int exclusive, u64 *failed_start, gfp_t mask)
 635 {
 636         struct extent_state *state;
 637         struct extent_state *prealloc = NULL;
 638         struct rb_node *node;
 639         unsigned long flags;
 640         int err = 0;
 641         int set;
 642         u64 last_start;
 643         u64 last_end;
 644 again:
 645         if (!prealloc && (mask & __GFP_WAIT)) {
 646                 prealloc = alloc_extent_state(mask);
 647                 if (!prealloc)
 648                         return -ENOMEM;
 649         }
 650
 651         spin_lock_irqsave(&tree->lock, flags);
 652         /*
 653          * this search will find all the extents that end after
 654          * our range starts.
 655          */
 656         node = tree_search(tree, start);
 657         if (!node) {
 658                 err = insert_state(tree, prealloc, start, end, bits);
 659                 prealloc = NULL;
 660                 BUG_ON(err == -EEXIST);
 661                 goto out;
 662         }
 663
 664         state = rb_entry(node, struct extent_state, rb_node);
 665         last_start = state->start;
 666         last_end = state->end;
 667
 668         /*
 669          * | ---- desired range ---- |
 670          * | state |
 671          *
 672          * Just lock what we found and keep going
 673          */
 674         if (state->start == start && state->end <= end) {
 675                 set = state->state & bits;
 676                 if (set && exclusive) {
 677                         *failed_start = state->start;
 678                         err = -EEXIST;
 679                         goto out;
 680                 }
 681                 set_state_bits(tree, state, bits);
 682                 start = state->end + 1;
 683                 merge_state(tree, state);
 684                 goto search_again;
 685         }
 686
 687         /*
 688          *     | ---- desired range ---- |
 689          * | state |
 690          *   or
 691          * | ------------- state -------------- |
 692          *
 693          * We need to split the extent we found, and may flip bits on
 694          * second half.
 695          *
 696          * If the extent we found extends past our
 697          * range, we just split and search again.  It'll get split
 698          * again the next time though.
 699          *
 700          * If the extent we found is inside our range, we set the
 701          * desired bit on it.
 702          */
 703         if (state->start < start) {
 704                 set = state->state & bits;
 705                 if (exclusive && set) {
 706                         *failed_start = start;
 707                         err = -EEXIST;
 708                         goto out;
 709                 }
 710                 err = split_state(tree, state, prealloc, start);
 711                 BUG_ON(err == -EEXIST);
 712                 prealloc = NULL;
 713                 if (err)
 714                         goto out;
 715                 if (state->end <= end) {
 716                         set_state_bits(tree, state, bits);
 717                         start = state->end + 1;
 718                         merge_state(tree, state);
 719                 } else {
 720                         start = state->start;
 721                 }
 722                 goto search_again;
 723         }
 724         /*
 725          * | ---- desired range ---- |
 726          *     | state | or               | state |
 727          *
 728          * There's a hole, we need to insert something in it and
 729          * ignore the extent we found.
 730          */
 731         if (state->start > start) {
 732                 u64 this_end;
 733                 if (end < last_start)
 734                         this_end = end;
 735                 else
 736                         this_end = last_start -1;
 737                 err = insert_state(tree, prealloc, start, this_end,
 738                                    bits);
 739                 prealloc = NULL;
 740                 BUG_ON(err == -EEXIST);
 741                 if (err)
 742                         goto out;
 743                 start = this_end + 1;
 744                 goto search_again;
 745         }
 746         /*
 747          * | ---- desired range ---- |
 748          *                        | state |
 749          * We need to split the extent, and set the bit
 750          * on the first half
 751          */
 752         if (state->start <= end && state->end > end) {
 753                 set = state->state & bits;
 754                 if (exclusive && set) {
 755                         *failed_start = start;
 756                         err = -EEXIST;
 757                         goto out;
 758                 }
 759                 err = split_state(tree, state, prealloc, end + 1);
 760                 BUG_ON(err == -EEXIST);
 761
 762                 set_state_bits(tree, prealloc, bits);
 763                 merge_state(tree, prealloc);
 764                 prealloc = NULL;
 765                 goto out;
 766         }
 767
 768         goto search_again;
 769
 770 out:
 771         spin_unlock_irqrestore(&tree->lock, flags);
 772         if (prealloc)
 773                 free_extent_state(prealloc);
 774
 775         return err;
 776
 777 search_again:
 778         if (start > end)
 779                 goto out;
 780         spin_unlock_irqrestore(&tree->lock, flags);
 781         if (mask & __GFP_WAIT)
 782                 cond_resched();
 783         goto again;
 784 }
 785 EXPORT_SYMBOL(set_extent_bit);
 786
 787 /* wrappers around set/clear extent bit */
 788 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
 789                      gfp_t mask)
 790 {
 791         return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL,
 792                               mask);
 793 }
 794 EXPORT_SYMBOL(set_extent_dirty);
 795
 796 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
 797                     int bits, gfp_t mask)
 798 {
 799         return set_extent_bit(tree, start, end, bits, 0, NULL,
 800                               mask);
 801 }
 802 EXPORT_SYMBOL(set_extent_bits);
 803
 804 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
 805                       int bits, gfp_t mask)
 806 {
 807         return clear_extent_bit(tree, start, end, bits, 0, 0, mask);
 808 }
 809 EXPORT_SYMBOL(clear_extent_bits);
 810
 811 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
 812                      gfp_t mask)
 813 {
 814         return set_extent_bit(tree, start, end,
 815                               EXTENT_DELALLOC | EXTENT_DIRTY, 0, NULL,
 816                               mask);
 817 }
 818 EXPORT_SYMBOL(set_extent_delalloc);
 819
 820 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
 821                        gfp_t mask)
 822 {
 823         return clear_extent_bit(tree, start, end,
 824                                 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, mask);
 825 }
 826 EXPORT_SYMBOL(clear_extent_dirty);
 827
 828 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
 829                      gfp_t mask)
 830 {
 831         return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL,
 832                               mask);
 833 }
 834 EXPORT_SYMBOL(set_extent_new);
 835
 836 int clear_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
 837                        gfp_t mask)
 838 {
 839         return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0, mask);
 840 }
 841 EXPORT_SYMBOL(clear_extent_new);
 842
 843 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
 844                         gfp_t mask)
 845 {
 846         return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, NULL,
 847                               mask);
 848 }
 849 EXPORT_SYMBOL(set_extent_uptodate);
 850
 851 int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
 852                           gfp_t mask)
 853 {
 854         return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, mask);
 855 }
 856 EXPORT_SYMBOL(clear_extent_uptodate);
 857
 858 int set_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end,
 859                          gfp_t mask)
 860 {
 861         return set_extent_bit(tree, start, end, EXTENT_WRITEBACK,
 862                               0, NULL, mask);
 863 }
 864 EXPORT_SYMBOL(set_extent_writeback);
 865
 866 int clear_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end,
 867                            gfp_t mask)
 868 {
 869         return clear_extent_bit(tree, start, end, EXTENT_WRITEBACK, 1, 0, mask);
 870 }
 871 EXPORT_SYMBOL(clear_extent_writeback);
 872
 873 int wait_on_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end)
 874 {
 875         return wait_extent_bit(tree, start, end, EXTENT_WRITEBACK);
 876 }
 877 EXPORT_SYMBOL(wait_on_extent_writeback);
 878
 879 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
 880 {
 881         int err;
 882         u64 failed_start;
 883         while (1) {
 884                 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1,
 885                                      &failed_start, mask);
 886                 if (err == -EEXIST && (mask & __GFP_WAIT)) {
 887                         wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
 888                         start = failed_start;
 889                 } else {
 890                         break;
 891                 }
 892                 WARN_ON(start > end);
 893         }
 894         return err;
 895 }
 896 EXPORT_SYMBOL(lock_extent);
 897
 898 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end,
 899                   gfp_t mask)
 900 {
 901         return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, mask);
 902 }
 903 EXPORT_SYMBOL(unlock_extent);
 904
 905 /*
 906  * helper function to set pages and extents in the tree dirty
 907  */
 908 int set_range_dirty(struct extent_io_tree *tree, u64 start, u64 end)
 909 {
 910         unsigned long index = start >> PAGE_CACHE_SHIFT;
 911         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
 912         struct page *page;
 913
 914         while (index <= end_index) {
 915                 page = find_get_page(tree->mapping, index);
 916                 BUG_ON(!page);
 917                 __set_page_dirty_nobuffers(page);
 918                 page_cache_release(page);
 919                 index++;
 920         }
 921         set_extent_dirty(tree, start, end, GFP_NOFS);
 922         return 0;
 923 }
 924 EXPORT_SYMBOL(set_range_dirty);
 925
 926 /*
 927  * helper function to set both pages and extents in the tree writeback
 928  */
 929 int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
 930 {
 931         unsigned long index = start >> PAGE_CACHE_SHIFT;
 932         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
 933         struct page *page;
 934
 935         while (index <= end_index) {
 936                 page = find_get_page(tree->mapping, index);
 937                 BUG_ON(!page);
 938                 set_page_writeback(page);
 939                 page_cache_release(page);
 940                 index++;
 941         }
 942         set_extent_writeback(tree, start, end, GFP_NOFS);
 943         return 0;
 944 }
 945 EXPORT_SYMBOL(set_range_writeback);
 946
 947 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
 948                           u64 *start_ret, u64 *end_ret, int bits)
 949 {
 950         struct rb_node *node;
 951         struct extent_state *state;
 952         int ret = 1;
 953
 954         spin_lock_irq(&tree->lock);
 955         /*
 956          * this search will find all the extents that end after
 957          * our range starts.
 958          */
 959         node = tree_search(tree, start);
 960         if (!node) {
 961                 goto out;
 962         }
 963
 964         while(1) {
 965                 state = rb_entry(node, struct extent_state, rb_node);
 966                 if (state->end >= start && (state->state & bits)) {
 967                         *start_ret = state->start;
 968                         *end_ret = state->end;
 969                         ret = 0;
 970                         break;
 971                 }
 972                 node = rb_next(node);
 973                 if (!node)
 974                         break;
 975         }
 976 out:
 977         spin_unlock_irq(&tree->lock);
 978         return ret;
 979 }
 980 EXPORT_SYMBOL(find_first_extent_bit);
 981
 982 struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree,
 983                                                  u64 start, int bits)
 984 {
 985         struct rb_node *node;
 986         struct extent_state *state;
 987
 988         /*
 989          * this search will find all the extents that end after
 990          * our range starts.
 991          */
 992         node = tree_search(tree, start);
 993         if (!node) {
 994                 goto out;
 995         }
 996
 997         while(1) {
 998                 state = rb_entry(node, struct extent_state, rb_node);
 999                 if (state->end >= start && (state->state & bits)) {
1000                         return state;
1001                 }
1002                 node = rb_next(node);
1003                 if (!node)
1004                         break;
1005         }
1006 out:
1007         return NULL;
1008 }
1009 EXPORT_SYMBOL(find_first_extent_bit_state);
1010
1011 u64 find_lock_delalloc_range(struct extent_io_tree *tree,
1012                              u64 *start, u64 *end, u64 max_bytes)
1013 {
1014         struct rb_node *node;
1015         struct extent_state *state;
1016         u64 cur_start = *start;
1017         u64 found = 0;
1018         u64 total_bytes = 0;
1019
1020         spin_lock_irq(&tree->lock);
1021         /*
1022          * this search will find all the extents that end after
1023          * our range starts.
1024          */
1025 search_again:
1026         node = tree_search(tree, cur_start);
1027         if (!node) {
1028                 *end = (u64)-1;
1029                 goto out;
1030         }
1031
1032         while(1) {
1033                 state = rb_entry(node, struct extent_state, rb_node);
1034                 if (found && state->start != cur_start) {
1035                         goto out;
1036                 }
1037                 if (!(state->state & EXTENT_DELALLOC)) {
1038                         if (!found)
1039                                 *end = state->end;
1040                         goto out;
1041                 }
1042                 if (!found) {
1043                         struct extent_state *prev_state;
1044                         struct rb_node *prev_node = node;
1045                         while(1) {
1046                                 prev_node = rb_prev(prev_node);
1047                                 if (!prev_node)
1048                                         break;
1049                                 prev_state = rb_entry(prev_node,
1050                                                       struct extent_state,
1051                                                       rb_node);
1052                                 if (!(prev_state->state & EXTENT_DELALLOC))
1053                                         break;
1054                                 state = prev_state;
1055                                 node = prev_node;
1056                         }
1057                 }
1058                 if (state->state & EXTENT_LOCKED) {
1059                         DEFINE_WAIT(wait);
1060                         atomic_inc(&state->refs);
1061                         prepare_to_wait(&state->wq, &wait,
1062                                         TASK_UNINTERRUPTIBLE);
1063                         spin_unlock_irq(&tree->lock);
1064                         schedule();
1065                         spin_lock_irq(&tree->lock);
1066                         finish_wait(&state->wq, &wait);
1067                         free_extent_state(state);
1068                         goto search_again;
1069                 }
1070                 set_state_cb(tree, state, EXTENT_LOCKED);
1071                 state->state |= EXTENT_LOCKED;
1072                 if (!found)
1073                         *start = state->start;
1074                 found++;
1075                 *end = state->end;
1076                 cur_start = state->end + 1;
1077                 node = rb_next(node);
1078                 if (!node)
1079                         break;
1080                 total_bytes += state->end - state->start + 1;
1081                 if (total_bytes >= max_bytes)
1082                         break;
1083         }
1084 out:
1085         spin_unlock_irq(&tree->lock);
1086         return found;
1087 }
1088
1089 u64 count_range_bits(struct extent_io_tree *tree,
1090                      u64 *start, u64 search_end, u64 max_bytes,
1091                      unsigned long bits)
1092 {
1093         struct rb_node *node;
1094         struct extent_state *state;
1095         u64 cur_start = *start;
1096         u64 total_bytes = 0;
1097         int found = 0;
1098
1099         if (search_end <= cur_start) {
1100                 printk("search_end %Lu start %Lu\n", search_end, cur_start);
1101                 WARN_ON(1);
1102                 return 0;
1103         }
1104
1105         spin_lock_irq(&tree->lock);
1106         if (cur_start == 0 && bits == EXTENT_DIRTY) {
1107                 total_bytes = tree->dirty_bytes;
1108                 goto out;
1109         }
1110         /*
1111          * this search will find all the extents that end after
1112          * our range starts.
1113          */
1114         node = tree_search(tree, cur_start);
1115         if (!node) {
1116                 goto out;
1117         }
1118
1119         while(1) {
1120                 state = rb_entry(node, struct extent_state, rb_node);
1121                 if (state->start > search_end)
1122                         break;
1123                 if (state->end >= cur_start && (state->state & bits)) {
1124                         total_bytes += min(search_end, state->end) + 1 -
1125                                        max(cur_start, state->start);
1126                         if (total_bytes >= max_bytes)
1127                                 break;
1128                         if (!found) {
1129                                 *start = state->start;
1130                                 found = 1;
1131                         }
1132                 }
1133                 node = rb_next(node);
1134                 if (!node)
1135                         break;
1136         }
1137 out:
1138         spin_unlock_irq(&tree->lock);
1139         return total_bytes;
1140 }
1141 /*
1142  * helper function to lock both pages and extents in the tree.
1143  * pages must be locked first.
1144  */
1145 int lock_range(struct extent_io_tree *tree, u64 start, u64 end)
1146 {
1147         unsigned long index = start >> PAGE_CACHE_SHIFT;
1148         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1149         struct page *page;
1150         int err;
1151
1152         while (index <= end_index) {
1153                 page = grab_cache_page(tree->mapping, index);
1154                 if (!page) {
1155                         err = -ENOMEM;
1156                         goto failed;
1157                 }
1158                 if (IS_ERR(page)) {
1159                         err = PTR_ERR(page);
1160                         goto failed;
1161                 }
1162                 index++;
1163         }
1164         lock_extent(tree, start, end, GFP_NOFS);
1165         return 0;
1166
1167 failed:
1168         /*
1169          * we failed above in getting the page at 'index', so we undo here
1170          * up to but not including the page at 'index'
1171          */
1172         end_index = index;
1173         index = start >> PAGE_CACHE_SHIFT;
1174         while (index < end_index) {
1175                 page = find_get_page(tree->mapping, index);
1176                 unlock_page(page);
1177                 page_cache_release(page);
1178                 index++;
1179         }
1180         return err;
1181 }
1182 EXPORT_SYMBOL(lock_range);
1183
1184 /*
1185  * helper function to unlock both pages and extents in the tree.
1186  */
1187 int unlock_range(struct extent_io_tree *tree, u64 start, u64 end)
1188 {
1189         unsigned long index = start >> PAGE_CACHE_SHIFT;
1190         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1191         struct page *page;
1192
1193         while (index <= end_index) {
1194                 page = find_get_page(tree->mapping, index);
1195                 unlock_page(page);
1196                 page_cache_release(page);
1197                 index++;
1198         }
1199         unlock_extent(tree, start, end, GFP_NOFS);
1200         return 0;
1201 }
1202 EXPORT_SYMBOL(unlock_range);
1203
1204 int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1205 {
1206         struct rb_node *node;
1207         struct extent_state *state;
1208         int ret = 0;
1209
1210         spin_lock_irq(&tree->lock);
1211         /*
1212          * this search will find all the extents that end after
1213          * our range starts.
1214          */
1215         node = tree_search(tree, start);
1216         if (!node) {
1217                 ret = -ENOENT;
1218                 goto out;
1219         }
1220         state = rb_entry(node, struct extent_state, rb_node);
1221         if (state->start != start) {
1222                 ret = -ENOENT;
1223                 goto out;
1224         }
1225         state->private = private;
1226 out:
1227         spin_unlock_irq(&tree->lock);
1228         return ret;
1229 }
1230
1231 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1232 {
1233         struct rb_node *node;
1234         struct extent_state *state;
1235         int ret = 0;
1236
1237         spin_lock_irq(&tree->lock);
1238         /*
1239          * this search will find all the extents that end after
1240          * our range starts.
1241          */
1242         node = tree_search(tree, start);
1243         if (!node) {
1244                 ret = -ENOENT;
1245                 goto out;
1246         }
1247         state = rb_entry(node, struct extent_state, rb_node);
1248         if (state->start != start) {
1249                 ret = -ENOENT;
1250                 goto out;
1251         }
1252         *private = state->private;
1253 out:
1254         spin_unlock_irq(&tree->lock);
1255         return ret;
1256 }
1257
1258 /*
1259  * searches a range in the state tree for a given mask.
1260  * If 'filled' == 1, this returns 1 only if every extent in the tree
1261  * has the bits set.  Otherwise, 1 is returned if any bit in the
1262  * range is found set.
1263  */
1264 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1265                    int bits, int filled)
1266 {
1267         struct extent_state *state = NULL;
1268         struct rb_node *node;
1269         int bitset = 0;
1270         unsigned long flags;
1271
1272         spin_lock_irqsave(&tree->lock, flags);
1273         node = tree_search(tree, start);
1274         while (node && start <= end) {
1275                 state = rb_entry(node, struct extent_state, rb_node);
1276
1277                 if (filled && state->start > start) {
1278                         bitset = 0;
1279                         break;
1280                 }
1281
1282                 if (state->start > end)
1283                         break;
1284
1285                 if (state->state & bits) {
1286                         bitset = 1;
1287                         if (!filled)
1288                                 break;
1289                 } else if (filled) {
1290                         bitset = 0;
1291                         break;
1292                 }
1293                 start = state->end + 1;
1294                 if (start > end)
1295                         break;
1296                 node = rb_next(node);
1297                 if (!node) {
1298                         if (filled)
1299                                 bitset = 0;
1300                         break;
1301                 }
1302         }
1303         spin_unlock_irqrestore(&tree->lock, flags);
1304         return bitset;
1305 }
1306 EXPORT_SYMBOL(test_range_bit);
1307
1308 /*
1309  * helper function to set a given page up to date if all the
1310  * extents in the tree for that page are up to date
1311  */
1312 static int check_page_uptodate(struct extent_io_tree *tree,
1313                                struct page *page)
1314 {
1315         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1316         u64 end = start + PAGE_CACHE_SIZE - 1;
1317         if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1))
1318                 SetPageUptodate(page);
1319         return 0;
1320 }
1321
1322 /*
1323  * helper function to unlock a page if all the extents in the tree
1324  * for that page are unlocked
1325  */
1326 static int check_page_locked(struct extent_io_tree *tree,
1327                              struct page *page)
1328 {
1329         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1330         u64 end = start + PAGE_CACHE_SIZE - 1;
1331         if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0))
1332                 unlock_page(page);
1333         return 0;
1334 }
1335
1336 /*
1337  * helper function to end page writeback if all the extents
1338  * in the tree for that page are done with writeback
1339  */
1340 static int check_page_writeback(struct extent_io_tree *tree,
1341                              struct page *page)
1342 {
1343         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1344         u64 end = start + PAGE_CACHE_SIZE - 1;
1345         if (!test_range_bit(tree, start, end, EXTENT_WRITEBACK, 0))
1346                 end_page_writeback(page);
1347         return 0;
1348 }
1349
1350 /* lots and lots of room for performance fixes in the end_bio funcs */
1351
1352 /*
1353  * after a writepage IO is done, we need to:
1354  * clear the uptodate bits on error
1355  * clear the writeback bits in the extent tree for this IO
1356  * end_page_writeback if the page has no more pending IO
1357  *
1358  * Scheduling is not allowed, so the extent state tree is expected
1359  * to have one and only one object corresponding to this IO.
1360  */
1361 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1362 static void end_bio_extent_writepage(struct bio *bio, int err)
1363 #else
1364 static int end_bio_extent_writepage(struct bio *bio,
1365                                    unsigned int bytes_done, int err)
1366 #endif
1367 {
1368         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1369         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1370         struct extent_state *state = bio->bi_private;
1371         struct extent_io_tree *tree = state->tree;
1372         struct rb_node *node;
1373         u64 start;
1374         u64 end;
1375         u64 cur;
1376         int whole_page;
1377         unsigned long flags;
1378
1379 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1380         if (bio->bi_size)
1381                 return 1;
1382 #endif
1383         do {
1384                 struct page *page = bvec->bv_page;
1385                 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1386                          bvec->bv_offset;
1387                 end = start + bvec->bv_len - 1;
1388
1389                 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1390                         whole_page = 1;
1391                 else
1392                         whole_page = 0;
1393
1394                 if (--bvec >= bio->bi_io_vec)
1395                         prefetchw(&bvec->bv_page->flags);
1396
1397                 if (!uptodate) {
1398                         clear_extent_uptodate(tree, start, end, GFP_ATOMIC);
1399                         ClearPageUptodate(page);
1400                         SetPageError(page);
1401                 }
1402
1403                 if (tree->ops && tree->ops->writepage_end_io_hook) {
1404                         tree->ops->writepage_end_io_hook(page, start, end,
1405                                                          state);
1406                 }
1407
1408                 /*
1409                  * bios can get merged in funny ways, and so we need to
1410                  * be careful with the state variable.  We know the
1411                  * state won't be merged with others because it has
1412                  * WRITEBACK set, but we can't be sure each biovec is
1413                  * sequential in the file.  So, if our cached state
1414                  * doesn't match the expected end, search the tree
1415                  * for the correct one.
1416                  */
1417
1418                 spin_lock_irqsave(&tree->lock, flags);
1419                 if (!state || state->end != end) {
1420                         state = NULL;
1421                         node = __etree_search(tree, start, NULL, NULL);
1422                         if (node) {
1423                                 state = rb_entry(node, struct extent_state,
1424                                                  rb_node);
1425                                 if (state->end != end ||
1426                                     !(state->state & EXTENT_WRITEBACK))
1427                                         state = NULL;
1428                         }
1429                         if (!state) {
1430                                 spin_unlock_irqrestore(&tree->lock, flags);
1431                                 clear_extent_writeback(tree, start,
1432                                                        end, GFP_ATOMIC);
1433                                 goto next_io;
1434                         }
1435                 }
1436                 cur = end;
1437                 while(1) {
1438                         struct extent_state *clear = state;
1439                         cur = state->start;
1440                         node = rb_prev(&state->rb_node);
1441                         if (node) {
1442                                 state = rb_entry(node,
1443                                                  struct extent_state,
1444                                                  rb_node);
1445                         } else {
1446                                 state = NULL;
1447                         }
1448
1449                         clear_state_bit(tree, clear, EXTENT_WRITEBACK,
1450                                         1, 0);
1451                         if (cur == start)
1452                                 break;
1453                         if (cur < start) {
1454                                 WARN_ON(1);
1455                                 break;
1456                         }
1457                         if (!node)
1458                                 break;
1459                 }
1460                 /* before releasing the lock, make sure the next state
1461                  * variable has the expected bits set and corresponds
1462                  * to the correct offsets in the file
1463                  */
1464                 if (state && (state->end + 1 != start ||
1465                     !(state->state & EXTENT_WRITEBACK))) {
1466                         state = NULL;
1467                 }
1468                 spin_unlock_irqrestore(&tree->lock, flags);
1469 next_io:
1470
1471                 if (whole_page)
1472                         end_page_writeback(page);
1473                 else
1474                         check_page_writeback(tree, page);
1475         } while (bvec >= bio->bi_io_vec);
1476         bio_put(bio);
1477 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1478         return 0;
1479 #endif
1480 }
1481
1482 /*
1483  * after a readpage IO is done, we need to:
1484  * clear the uptodate bits on error
1485  * set the uptodate bits if things worked
1486  * set the page up to date if all extents in the tree are uptodate
1487  * clear the lock bit in the extent tree
1488  * unlock the page if there are no other extents locked for it
1489  *
1490  * Scheduling is not allowed, so the extent state tree is expected
1491  * to have one and only one object corresponding to this IO.
1492  */
1493 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1494 static void end_bio_extent_readpage(struct bio *bio, int err)
1495 #else
1496 static int end_bio_extent_readpage(struct bio *bio,
1497                                    unsigned int bytes_done, int err)
1498 #endif
1499 {
1500         int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1501         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1502         struct extent_state *state = bio->bi_private;
1503         struct extent_io_tree *tree = state->tree;
1504         struct rb_node *node;
1505         u64 start;
1506         u64 end;
1507         u64 cur;
1508         unsigned long flags;
1509         int whole_page;
1510         int ret;
1511
1512 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1513         if (bio->bi_size)
1514                 return 1;
1515 #endif
1516
1517         do {
1518                 struct page *page = bvec->bv_page;
1519                 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1520                         bvec->bv_offset;
1521                 end = start + bvec->bv_len - 1;
1522
1523                 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1524                         whole_page = 1;
1525                 else
1526                         whole_page = 0;
1527
1528                 if (--bvec >= bio->bi_io_vec)
1529                         prefetchw(&bvec->bv_page->flags);
1530
1531                 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
1532                         ret = tree->ops->readpage_end_io_hook(page, start, end,
1533                                                               state);
1534                         if (ret)
1535                                 uptodate = 0;
1536                 }
1537                 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(2,6,18)
2113
2114 /* Taken directly from 2.6.23 for 2.6.18 back port */
2115 typedef int (*writepage_t)(struct page *page, struct writeback_control *wbc,
2116                                 void *data);
2117
2118 /**
2119  * write_cache_pages - walk the list of dirty pages of the given address space
2120  * and write all of them.
2121  * @mapping: address space structure to write
2122  * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2123  * @writepage: function called for each page
2124  * @data: data passed to writepage function
2125  *
2126  * If a page is already under I/O, write_cache_pages() skips it, even
2127  * if it's dirty.  This is desirable behaviour for memory-cleaning writeback,
2128  * but it is INCORRECT for data-integrity system calls such as fsync().  fsync()
2129  * and msync() need to guarantee that all the data which was dirty at the time
2130  * the call was made get new I/O started against them.  If wbc->sync_mode is
2131  * WB_SYNC_ALL then we were called for data integrity and we must wait for
2132  * existing IO to complete.
2133  */
2134 static int write_cache_pages(struct address_space *mapping,
2135                       struct writeback_control *wbc, writepage_t writepage,
2136                       void *data)
2137 {
2138         struct backing_dev_info *bdi = mapping->backing_dev_info;
2139         int ret = 0;
2140         int done = 0;
2141         struct pagevec pvec;
2142         int nr_pages;
2143         pgoff_t index;
2144         pgoff_t end;            /* Inclusive */
2145         int scanned = 0;
2146         int range_whole = 0;
2147
2148         if (wbc->nonblocking && bdi_write_congested(bdi)) {
2149                 wbc->encountered_congestion = 1;
2150                 return 0;
2151         }
2152
2153         pagevec_init(&pvec, 0);
2154         if (wbc->range_cyclic) {
2155                 index = mapping->writeback_index; /* Start from prev offset */
2156                 end = -1;
2157         } else {
2158                 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2159                 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2160                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2161                         range_whole = 1;
2162                 scanned = 1;
2163         }
2164 retry:
2165         while (!done && (index <= end) &&
2166                (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
2167                                               PAGECACHE_TAG_DIRTY,
2168                                               min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
2169                 unsigned i;
2170
2171                 scanned = 1;
2172                 for (i = 0; i < nr_pages; i++) {
2173                         struct page *page = pvec.pages[i];
2174
2175                         /*
2176                          * At this point we hold neither mapping->tree_lock nor
2177                          * lock on the page itself: the page may be truncated or
2178                          * invalidated (changing page->mapping to NULL), or even
2179                          * swizzled back from swapper_space to tmpfs file
2180                          * mapping
2181                          */
2182                         lock_page(page);
2183
2184                         if (unlikely(page->mapping != mapping)) {
2185                                 unlock_page(page);
2186                                 continue;
2187                         }
2188
2189                         if (!wbc->range_cyclic && page->index > end) {
2190                                 done = 1;
2191                                 unlock_page(page);
2192                                 continue;
2193                         }
2194
2195                         if (wbc->sync_mode != WB_SYNC_NONE)
2196                                 wait_on_page_writeback(page);
2197
2198                         if (PageWriteback(page) ||
2199                             !clear_page_dirty_for_io(page)) {
2200                                 unlock_page(page);
2201                                 continue;
2202                         }
2203
2204                         ret = (*writepage)(page, wbc, data);
2205
2206                         if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
2207                                 unlock_page(page);
2208                                 ret = 0;
2209                         }
2210                         if (ret || (--(wbc->nr_to_write) <= 0))
2211                                 done = 1;
2212                         if (wbc->nonblocking && bdi_write_congested(bdi)) {
2213                                 wbc->encountered_congestion = 1;
2214                                 done = 1;
2215                         }
2216                 }
2217                 pagevec_release(&pvec);
2218                 cond_resched();
2219         }
2220         if (!scanned && !done) {
2221                 /*
2222                  * We hit the last page and there is more work to be done: wrap
2223                  * back to the start of the file
2224                  */
2225                 scanned = 1;
2226                 index = 0;
2227                 goto retry;
2228         }
2229         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2230                 mapping->writeback_index = index;
2231         return ret;
2232 }
2233 #endif
2234
2235 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
2236                           get_extent_t *get_extent,
2237                           struct writeback_control *wbc)
2238 {
2239         int ret;
2240         struct address_space *mapping = page->mapping;
2241         struct extent_page_data epd = {
2242                 .bio = NULL,
2243                 .tree = tree,
2244                 .get_extent = get_extent,
2245         };
2246         struct writeback_control wbc_writepages = {
2247                 .bdi            = wbc->bdi,
2248                 .sync_mode      = WB_SYNC_NONE,
2249                 .older_than_this = NULL,
2250                 .nr_to_write    = 64,
2251                 .range_start    = page_offset(page) + PAGE_CACHE_SIZE,
2252                 .range_end      = (loff_t)-1,
2253         };
2254
2255
2256         ret = __extent_writepage(page, wbc, &epd);
2257
2258         write_cache_pages(mapping, &wbc_writepages, __extent_writepage, &epd);
2259         if (epd.bio) {
2260                 submit_one_bio(WRITE, epd.bio, 0);
2261         }
2262         return ret;
2263 }
2264 EXPORT_SYMBOL(extent_write_full_page);
2265
2266
2267 int extent_writepages(struct extent_io_tree *tree,
2268                       struct address_space *mapping,
2269                       get_extent_t *get_extent,
2270                       struct writeback_control *wbc)
2271 {
2272         int ret = 0;
2273         struct extent_page_data epd = {
2274                 .bio = NULL,
2275                 .tree = tree,
2276                 .get_extent = get_extent,
2277         };
2278
2279         ret = write_cache_pages(mapping, wbc, __extent_writepage, &epd);
2280         if (epd.bio) {
2281                 submit_one_bio(WRITE, epd.bio, 0);
2282         }
2283         return ret;
2284 }
2285 EXPORT_SYMBOL(extent_writepages);
2286
2287 int extent_readpages(struct extent_io_tree *tree,
2288                      struct address_space *mapping,
2289                      struct list_head *pages, unsigned nr_pages,
2290                      get_extent_t get_extent)
2291 {
2292         struct bio *bio = NULL;
2293         unsigned page_idx;
2294         struct pagevec pvec;
2295
2296         pagevec_init(&pvec, 0);
2297         for (page_idx = 0; page_idx < nr_pages; page_idx++) {
2298                 struct page *page = list_entry(pages->prev, struct page, lru);
2299
2300                 prefetchw(&page->flags);
2301                 list_del(&page->lru);
2302                 /*
2303                  * what we want to do here is call add_to_page_cache_lru,
2304                  * but that isn't exported, so we reproduce it here
2305                  */
2306                 if (!add_to_page_cache(page, mapping,
2307                                         page->index, GFP_KERNEL)) {
2308
2309                         /* open coding of lru_cache_add, also not exported */
2310                         page_cache_get(page);
2311                         if (!pagevec_add(&pvec, page))
2312                                 __pagevec_lru_add(&pvec);
2313                         __extent_read_full_page(tree, page, get_extent,
2314                                                 &bio, 0);
2315                 }
2316                 page_cache_release(page);
2317         }
2318         if (pagevec_count(&pvec))
2319                 __pagevec_lru_add(&pvec);
2320         BUG_ON(!list_empty(pages));
2321         if (bio)
2322                 submit_one_bio(READ, bio, 0);
2323         return 0;
2324 }
2325 EXPORT_SYMBOL(extent_readpages);
2326
2327 /*
2328  * basic invalidatepage code, this waits on any locked or writeback
2329  * ranges corresponding to the page, and then deletes any extent state
2330  * records from the tree
2331  */
2332 int extent_invalidatepage(struct extent_io_tree *tree,
2333                           struct page *page, unsigned long offset)
2334 {
2335         u64 start = ((u64)page->index << PAGE_CACHE_SHIFT);
2336         u64 end = start + PAGE_CACHE_SIZE - 1;
2337         size_t blocksize = page->mapping->host->i_sb->s_blocksize;
2338
2339         start += (offset + blocksize -1) & ~(blocksize - 1);
2340         if (start > end)
2341                 return 0;
2342
2343         lock_extent(tree, start, end, GFP_NOFS);
2344         wait_on_extent_writeback(tree, start, end);
2345         clear_extent_bit(tree, start, end,
2346                          EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC,
2347                          1, 1, GFP_NOFS);
2348         return 0;
2349 }
2350 EXPORT_SYMBOL(extent_invalidatepage);
2351
2352 /*
2353  * simple commit_write call, set_range_dirty is used to mark both
2354  * the pages and the extent records as dirty
2355  */
2356 int extent_commit_write(struct extent_io_tree *tree,
2357                         struct inode *inode, struct page *page,
2358                         unsigned from, unsigned to)
2359 {
2360         loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
2361
2362         set_page_extent_mapped(page);
2363         set_page_dirty(page);
2364
2365         if (pos > inode->i_size) {
2366                 i_size_write(inode, pos);
2367                 mark_inode_dirty(inode);
2368         }
2369         return 0;
2370 }
2371 EXPORT_SYMBOL(extent_commit_write);
2372
2373 int extent_prepare_write(struct extent_io_tree *tree,
2374                          struct inode *inode, struct page *page,
2375                          unsigned from, unsigned to, get_extent_t *get_extent)
2376 {
2377         u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2378         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
2379         u64 block_start;
2380         u64 orig_block_start;
2381         u64 block_end;
2382         u64 cur_end;
2383         struct extent_map *em;
2384         unsigned blocksize = 1 << inode->i_blkbits;
2385         size_t page_offset = 0;
2386         size_t block_off_start;
2387         size_t block_off_end;
2388         int err = 0;
2389         int iocount = 0;
2390         int ret = 0;
2391         int isnew;
2392
2393         set_page_extent_mapped(page);
2394
2395         block_start = (page_start + from) & ~((u64)blocksize - 1);
2396         block_end = (page_start + to - 1) | (blocksize - 1);
2397         orig_block_start = block_start;
2398
2399         lock_extent(tree, page_start, page_end, GFP_NOFS);
2400         while(block_start <= block_end) {
2401                 em = get_extent(inode, page, page_offset, block_start,
2402                                 block_end - block_start + 1, 1);
2403                 if (IS_ERR(em) || !em) {
2404                         goto err;
2405                 }
2406                 cur_end = min(block_end, extent_map_end(em) - 1);
2407                 block_off_start = block_start & (PAGE_CACHE_SIZE - 1);
2408                 block_off_end = block_off_start + blocksize;
2409                 isnew = clear_extent_new(tree, block_start, cur_end, GFP_NOFS);
2410
2411                 if (!PageUptodate(page) && isnew &&
2412                     (block_off_end > to || block_off_start < from)) {
2413                         void *kaddr;
2414
2415                         kaddr = kmap_atomic(page, KM_USER0);
2416                         if (block_off_end > to)
2417                                 memset(kaddr + to, 0, block_off_end - to);
2418                         if (block_off_start < from)
2419                                 memset(kaddr + block_off_start, 0,
2420                                        from - block_off_start);
2421                         flush_dcache_page(page);
2422                         kunmap_atomic(kaddr, KM_USER0);
2423                 }
2424                 if ((em->block_start != EXTENT_MAP_HOLE &&
2425                      em->block_start != EXTENT_MAP_INLINE) &&
2426                     !isnew && !PageUptodate(page) &&
2427                     (block_off_end > to || block_off_start < from) &&
2428                     !test_range_bit(tree, block_start, cur_end,
2429                                     EXTENT_UPTODATE, 1)) {
2430                         u64 sector;
2431                         u64 extent_offset = block_start - em->start;
2432                         size_t iosize;
2433                         sector = (em->block_start + extent_offset) >> 9;
2434                         iosize = (cur_end - block_start + blocksize) &
2435                                 ~((u64)blocksize - 1);
2436                         /*
2437                          * we've already got the extent locked, but we
2438                          * need to split the state such that our end_bio
2439                          * handler can clear the lock.
2440                          */
2441                         set_extent_bit(tree, block_start,
2442                                        block_start + iosize - 1,
2443                                        EXTENT_LOCKED, 0, NULL, GFP_NOFS);
2444                         ret = submit_extent_page(READ, tree, page,
2445                                          sector, iosize, page_offset, em->bdev,
2446                                          NULL, 1,
2447                                          end_bio_extent_preparewrite, 0);
2448                         iocount++;
2449                         block_start = block_start + iosize;
2450                 } else {
2451                         set_extent_uptodate(tree, block_start, cur_end,
2452                                             GFP_NOFS);
2453                         unlock_extent(tree, block_start, cur_end, GFP_NOFS);
2454                         block_start = cur_end + 1;
2455                 }
2456                 page_offset = block_start & (PAGE_CACHE_SIZE - 1);
2457                 free_extent_map(em);
2458         }
2459         if (iocount) {
2460                 wait_extent_bit(tree, orig_block_start,
2461                                 block_end, EXTENT_LOCKED);
2462         }
2463         check_page_uptodate(tree, page);
2464 err:
2465         /* FIXME, zero out newly allocated blocks on error */
2466         return err;
2467 }
2468 EXPORT_SYMBOL(extent_prepare_write);
2469
2470 /*
2471  * a helper for releasepage.  As long as there are no locked extents
2472  * in the range corresponding to the page, both state records and extent
2473  * map records are removed
2474  */
2475 int try_release_extent_mapping(struct extent_map_tree *map,
2476                                struct extent_io_tree *tree, struct page *page,
2477                                gfp_t mask)
2478 {
2479         struct extent_map *em;
2480         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2481         u64 end = start + PAGE_CACHE_SIZE - 1;
2482         u64 orig_start = start;
2483         int ret = 1;
2484         if ((mask & __GFP_WAIT) &&
2485             page->mapping->host->i_size > 16 * 1024 * 1024) {
2486                 u64 len;
2487                 while (start <= end) {
2488                         len = end - start + 1;
2489                         spin_lock(&map->lock);
2490                         em = lookup_extent_mapping(map, start, len);
2491                         if (!em || IS_ERR(em)) {
2492                                 spin_unlock(&map->lock);
2493                                 break;
2494                         }
2495                         if (em->start != start) {
2496                                 spin_unlock(&map->lock);
2497                                 free_extent_map(em);
2498                                 break;
2499                         }
2500                         if (!test_range_bit(tree, em->start,
2501                                             extent_map_end(em) - 1,
2502                                             EXTENT_LOCKED, 0)) {
2503                                 remove_extent_mapping(map, em);
2504                                 /* once for the rb tree */
2505                                 free_extent_map(em);
2506                         }
2507                         start = extent_map_end(em);
2508                         spin_unlock(&map->lock);
2509
2510                         /* once for us */
2511                         free_extent_map(em);
2512                 }
2513         }
2514         if (test_range_bit(tree, orig_start, end, EXTENT_IOBITS, 0))
2515                 ret = 0;
2516         else {
2517                 if ((mask & GFP_NOFS) == GFP_NOFS)
2518                         mask = GFP_NOFS;
2519                 clear_extent_bit(tree, orig_start, end, EXTENT_UPTODATE,
2520                                  1, 1, mask);
2521         }
2522         return ret;
2523 }
2524 EXPORT_SYMBOL(try_release_extent_mapping);
2525
2526 sector_t extent_bmap(struct address_space *mapping, sector_t iblock,
2527                 get_extent_t *get_extent)
2528 {
2529         struct inode *inode = mapping->host;
2530         u64 start = iblock << inode->i_blkbits;
2531         sector_t sector = 0;
2532         struct extent_map *em;
2533
2534         em = get_extent(inode, NULL, 0, start, (1 << inode->i_blkbits), 0);
2535         if (!em || IS_ERR(em))
2536                 return 0;
2537
2538         if (em->block_start == EXTENT_MAP_INLINE ||
2539             em->block_start == EXTENT_MAP_HOLE)
2540                 goto out;
2541
2542         sector = (em->block_start + start - em->start) >> inode->i_blkbits;
2543 out:
2544         free_extent_map(em);
2545         return sector;
2546 }
2547
2548 static int add_lru(struct extent_io_tree *tree, struct extent_buffer *eb)
2549 {
2550         if (list_empty(&eb->lru)) {
2551                 extent_buffer_get(eb);
2552                 list_add(&eb->lru, &tree->buffer_lru);
2553                 tree->lru_size++;
2554                 if (tree->lru_size >= BUFFER_LRU_MAX) {
2555                         struct extent_buffer *rm;
2556                         rm = list_entry(tree->buffer_lru.prev,
2557                                         struct extent_buffer, lru);
2558                         tree->lru_size--;
2559                         list_del_init(&rm->lru);
2560                         free_extent_buffer(rm);
2561                 }
2562         } else
2563                 list_move(&eb->lru, &tree->buffer_lru);
2564         return 0;
2565 }
2566 static struct extent_buffer *find_lru(struct extent_io_tree *tree,
2567                                       u64 start, unsigned long len)
2568 {
2569         struct list_head *lru = &tree->buffer_lru;
2570         struct list_head *cur = lru->next;
2571         struct extent_buffer *eb;
2572
2573         if (list_empty(lru))
2574                 return NULL;
2575
2576         do {
2577                 eb = list_entry(cur, struct extent_buffer, lru);
2578                 if (eb->start == start && eb->len == len) {
2579                         extent_buffer_get(eb);
2580                         return eb;
2581                 }
2582                 cur = cur->next;
2583         } while (cur != lru);
2584         return NULL;
2585 }
2586
2587 static inline unsigned long num_extent_pages(u64 start, u64 len)
2588 {
2589         return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
2590                 (start >> PAGE_CACHE_SHIFT);
2591 }
2592
2593 static inline struct page *extent_buffer_page(struct extent_buffer *eb,
2594                                               unsigned long i)
2595 {
2596         struct page *p;
2597         struct address_space *mapping;
2598
2599         if (i == 0)
2600                 return eb->first_page;
2601         i += eb->start >> PAGE_CACHE_SHIFT;
2602         mapping = eb->first_page->mapping;
2603         read_lock_irq(&mapping->tree_lock);
2604         p = radix_tree_lookup(&mapping->page_tree, i);
2605         read_unlock_irq(&mapping->tree_lock);
2606         return p;
2607 }
2608
2609 int release_extent_buffer_tail_pages(struct extent_buffer *eb)
2610 {
2611         unsigned long num_pages = num_extent_pages(eb->start, eb->len);
2612         struct page *page;
2613         unsigned long i;
2614
2615         if (num_pages == 1)
2616                 return 0;
2617         for (i = 1; i < num_pages; i++) {
2618                 page = extent_buffer_page(eb, i);
2619                 page_cache_release(page);
2620         }
2621         return 0;
2622 }
2623
2624
2625 int invalidate_extent_lru(struct extent_io_tree *tree, u64 start,
2626                           unsigned long len)
2627 {
2628         struct list_head *lru = &tree->buffer_lru;
2629         struct list_head *cur = lru->next;
2630         struct extent_buffer *eb;
2631         int found = 0;
2632
2633         spin_lock(&tree->lru_lock);
2634         if (list_empty(lru))
2635                 goto out;
2636
2637         do {
2638                 eb = list_entry(cur, struct extent_buffer, lru);
2639                 if (eb->start <= start && eb->start + eb->len > start) {
2640                         eb->flags &= ~EXTENT_UPTODATE;
2641                 }
2642                 cur = cur->next;
2643         } while (cur != lru);
2644 out:
2645         spin_unlock(&tree->lru_lock);
2646         return found;
2647 }
2648
2649 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
2650                                                    u64 start,
2651                                                    unsigned long len,
2652                                                    gfp_t mask)
2653 {
2654         struct extent_buffer *eb = NULL;
2655         unsigned long flags;
2656
2657         spin_lock(&tree->lru_lock);
2658         eb = find_lru(tree, start, len);
2659         spin_unlock(&tree->lru_lock);
2660         if (eb) {
2661                 return eb;
2662         }
2663
2664         eb = kmem_cache_zalloc(extent_buffer_cache, mask);
2665         INIT_LIST_HEAD(&eb->lru);
2666         eb->start = start;
2667         eb->len = len;
2668         spin_lock_irqsave(&leak_lock, flags);
2669         list_add(&eb->leak_list, &buffers);
2670         spin_unlock_irqrestore(&leak_lock, flags);
2671         atomic_set(&eb->refs, 1);
2672
2673         return eb;
2674 }
2675
2676 static void __free_extent_buffer(struct extent_buffer *eb)
2677 {
2678         unsigned long flags;
2679         spin_lock_irqsave(&leak_lock, flags);
2680         list_del(&eb->leak_list);
2681         spin_unlock_irqrestore(&leak_lock, flags);
2682         kmem_cache_free(extent_buffer_cache, eb);
2683 }
2684
2685 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
2686                                           u64 start, unsigned long len,
2687                                           struct page *page0,
2688                                           gfp_t mask)
2689 {
2690         unsigned long num_pages = num_extent_pages(start, len);
2691         unsigned long i;
2692         unsigned long index = start >> PAGE_CACHE_SHIFT;
2693         struct extent_buffer *eb;
2694         struct page *p;
2695         struct address_space *mapping = tree->mapping;
2696         int uptodate = 1;
2697
2698         eb = __alloc_extent_buffer(tree, start, len, mask);
2699         if (!eb)
2700                 return NULL;
2701
2702         if (eb->flags & EXTENT_BUFFER_FILLED)
2703                 goto lru_add;
2704
2705         if (page0) {
2706                 eb->first_page = page0;
2707                 i = 1;
2708                 index++;
2709                 page_cache_get(page0);
2710                 mark_page_accessed(page0);
2711                 set_page_extent_mapped(page0);
2712                 set_page_extent_head(page0, len);
2713                 uptodate = PageUptodate(page0);
2714         } else {
2715                 i = 0;
2716         }
2717         for (; i < num_pages; i++, index++) {
2718                 p = find_or_create_page(mapping, index, mask | __GFP_HIGHMEM);
2719                 if (!p) {
2720                         WARN_ON(1);
2721                         goto fail;
2722                 }
2723                 set_page_extent_mapped(p);
2724                 mark_page_accessed(p);
2725                 if (i == 0) {
2726                         eb->first_page = p;
2727                         set_page_extent_head(p, len);
2728                 } else {
2729                         set_page_private(p, EXTENT_PAGE_PRIVATE);
2730                 }
2731                 if (!PageUptodate(p))
2732                         uptodate = 0;
2733                 unlock_page(p);
2734         }
2735         if (uptodate)
2736                 eb->flags |= EXTENT_UPTODATE;
2737         eb->flags |= EXTENT_BUFFER_FILLED;
2738
2739 lru_add:
2740         spin_lock(&tree->lru_lock);
2741         add_lru(tree, eb);
2742         spin_unlock(&tree->lru_lock);
2743         return eb;
2744
2745 fail:
2746         spin_lock(&tree->lru_lock);
2747         list_del_init(&eb->lru);
2748         spin_unlock(&tree->lru_lock);
2749         if (!atomic_dec_and_test(&eb->refs))
2750                 return NULL;
2751         for (index = 1; index < i; index++) {
2752                 page_cache_release(extent_buffer_page(eb, index));
2753         }
2754         if (i > 0)
2755                 page_cache_release(extent_buffer_page(eb, 0));
2756         __free_extent_buffer(eb);
2757         return NULL;
2758 }
2759 EXPORT_SYMBOL(alloc_extent_buffer);
2760
2761 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
2762                                          u64 start, unsigned long len,
2763                                           gfp_t mask)
2764 {
2765         unsigned long num_pages = num_extent_pages(start, len);
2766         unsigned long i;
2767         unsigned long index = start >> PAGE_CACHE_SHIFT;
2768         struct extent_buffer *eb;
2769         struct page *p;
2770         struct address_space *mapping = tree->mapping;
2771         int uptodate = 1;
2772
2773         eb = __alloc_extent_buffer(tree, start, len, mask);
2774         if (!eb)
2775                 return NULL;
2776
2777         if (eb->flags & EXTENT_BUFFER_FILLED)
2778                 goto lru_add;
2779
2780         for (i = 0; i < num_pages; i++, index++) {
2781                 p = find_lock_page(mapping, index);
2782                 if (!p) {
2783                         goto fail;
2784                 }
2785                 set_page_extent_mapped(p);
2786                 mark_page_accessed(p);
2787
2788                 if (i == 0) {
2789                         eb->first_page = p;
2790                         set_page_extent_head(p, len);
2791                 } else {
2792                         set_page_private(p, EXTENT_PAGE_PRIVATE);
2793                 }
2794
2795                 if (!PageUptodate(p))
2796                         uptodate = 0;
2797                 unlock_page(p);
2798         }
2799         if (uptodate)
2800                 eb->flags |= EXTENT_UPTODATE;
2801         eb->flags |= EXTENT_BUFFER_FILLED;
2802
2803 lru_add:
2804         spin_lock(&tree->lru_lock);
2805         add_lru(tree, eb);
2806         spin_unlock(&tree->lru_lock);
2807         return eb;
2808 fail:
2809         spin_lock(&tree->lru_lock);
2810         list_del_init(&eb->lru);
2811         spin_unlock(&tree->lru_lock);
2812         if (!atomic_dec_and_test(&eb->refs))
2813                 return NULL;
2814         for (index = 1; index < i; index++) {
2815                 page_cache_release(extent_buffer_page(eb, index));
2816         }
2817         if (i > 0)
2818                 page_cache_release(extent_buffer_page(eb, 0));
2819         __free_extent_buffer(eb);
2820         return NULL;
2821 }
2822 EXPORT_SYMBOL(find_extent_buffer);
2823
2824 void free_extent_buffer(struct extent_buffer *eb)
2825 {
2826         unsigned long i;
2827         unsigned long num_pages;
2828
2829         if (!eb)
2830                 return;
2831
2832         if (!atomic_dec_and_test(&eb->refs))
2833                 return;
2834
2835         WARN_ON(!list_empty(&eb->lru));
2836         num_pages = num_extent_pages(eb->start, eb->len);
2837
2838         for (i = 1; i < num_pages; i++) {
2839                 page_cache_release(extent_buffer_page(eb, i));
2840         }
2841         page_cache_release(extent_buffer_page(eb, 0));
2842         __free_extent_buffer(eb);
2843 }
2844 EXPORT_SYMBOL(free_extent_buffer);
2845
2846 int clear_extent_buffer_dirty(struct extent_io_tree *tree,
2847                               struct extent_buffer *eb)
2848 {
2849         int set;
2850         unsigned long i;
2851         unsigned long num_pages;
2852         struct page *page;
2853
2854         u64 start = eb->start;
2855         u64 end = start + eb->len - 1;
2856
2857         set = clear_extent_dirty(tree, start, end, GFP_NOFS);
2858         num_pages = num_extent_pages(eb->start, eb->len);
2859
2860         for (i = 0; i < num_pages; i++) {
2861                 page = extent_buffer_page(eb, i);
2862                 lock_page(page);
2863                 if (i == 0)
2864                         set_page_extent_head(page, eb->len);
2865                 else
2866                         set_page_private(page, EXTENT_PAGE_PRIVATE);
2867
2868                 /*
2869                  * if we're on the last page or the first page and the
2870                  * block isn't aligned on a page boundary, do extra checks
2871                  * to make sure we don't clean page that is partially dirty
2872                  */
2873                 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
2874                     ((i == num_pages - 1) &&
2875                      ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
2876                         start = (u64)page->index << PAGE_CACHE_SHIFT;
2877                         end  = start + PAGE_CACHE_SIZE - 1;
2878                         if (test_range_bit(tree, start, end,
2879                                            EXTENT_DIRTY, 0)) {
2880                                 unlock_page(page);
2881                                 continue;
2882                         }
2883                 }
2884                 clear_page_dirty_for_io(page);
2885                 read_lock_irq(&page->mapping->tree_lock);
2886                 if (!PageDirty(page)) {
2887                         radix_tree_tag_clear(&page->mapping->page_tree,
2888                                                 page_index(page),
2889                                                 PAGECACHE_TAG_DIRTY);
2890                 }
2891                 read_unlock_irq(&page->mapping->tree_lock);
2892                 unlock_page(page);
2893         }
2894         return 0;
2895 }
2896 EXPORT_SYMBOL(clear_extent_buffer_dirty);
2897
2898 int wait_on_extent_buffer_writeback(struct extent_io_tree *tree,
2899                                     struct extent_buffer *eb)
2900 {
2901         return wait_on_extent_writeback(tree, eb->start,
2902                                         eb->start + eb->len - 1);
2903 }
2904 EXPORT_SYMBOL(wait_on_extent_buffer_writeback);
2905
2906 int set_extent_buffer_dirty(struct extent_io_tree *tree,
2907                              struct extent_buffer *eb)
2908 {
2909         unsigned long i;
2910         unsigned long num_pages;
2911
2912         num_pages = num_extent_pages(eb->start, eb->len);
2913         for (i = 0; i < num_pages; i++) {
2914                 struct page *page = extent_buffer_page(eb, i);
2915                 /* writepage may need to do something special for the
2916                  * first page, we have to make sure page->private is
2917                  * properly set.  releasepage may drop page->private
2918                  * on us if the page isn't already dirty.
2919                  */
2920                 if (i == 0) {
2921                         lock_page(page);
2922                         set_page_extent_head(page, eb->len);
2923                 } else if (PagePrivate(page) &&
2924                            page->private != EXTENT_PAGE_PRIVATE) {
2925                         lock_page(page);
2926                         set_page_extent_mapped(page);
2927                         unlock_page(page);
2928                 }
2929                 __set_page_dirty_nobuffers(extent_buffer_page(eb, i));
2930                 if (i == 0)
2931                         unlock_page(page);
2932         }
2933         return set_extent_dirty(tree, eb->start,
2934                                 eb->start + eb->len - 1, GFP_NOFS);
2935 }
2936 EXPORT_SYMBOL(set_extent_buffer_dirty);
2937
2938 int set_extent_buffer_uptodate(struct extent_io_tree *tree,
2939                                 struct extent_buffer *eb)
2940 {
2941         unsigned long i;
2942         struct page *page;
2943         unsigned long num_pages;
2944
2945         num_pages = num_extent_pages(eb->start, eb->len);
2946
2947         set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
2948                             GFP_NOFS);
2949         for (i = 0; i < num_pages; i++) {
2950                 page = extent_buffer_page(eb, i);
2951                 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
2952                     ((i == num_pages - 1) &&
2953                      ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
2954                         check_page_uptodate(tree, page);
2955                         continue;
2956                 }
2957                 SetPageUptodate(page);
2958         }
2959         return 0;
2960 }
2961 EXPORT_SYMBOL(set_extent_buffer_uptodate);
2962
2963 int extent_range_uptodate(struct extent_io_tree *tree,
2964                           u64 start, u64 end)
2965 {
2966         struct page *page;
2967         int ret;
2968         int pg_uptodate = 1;
2969         int uptodate;
2970         unsigned long index;
2971
2972         ret = test_range_bit(tree, start, end, EXTENT_UPTODATE, 1);
2973         if (ret)
2974                 return 1;
2975         while(start <= end) {
2976                 index = start >> PAGE_CACHE_SHIFT;
2977                 page = find_get_page(tree->mapping, index);
2978                 uptodate = PageUptodate(page);
2979                 page_cache_release(page);
2980                 if (!uptodate) {
2981                         pg_uptodate = 0;
2982                         break;
2983                 }
2984                 start += PAGE_CACHE_SIZE;
2985         }
2986         return pg_uptodate;
2987 }
2988
2989 int extent_buffer_uptodate(struct extent_io_tree *tree,
2990                            struct extent_buffer *eb)
2991 {
2992         int ret = 0;
2993         int ret2;
2994         unsigned long num_pages;
2995         unsigned long i;
2996         struct page *page;
2997         int pg_uptodate = 1;
2998
2999         if (eb->flags & EXTENT_UPTODATE)
3000                 ret = 1;
3001
3002         ret2  = test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3003                            EXTENT_UPTODATE, 1);
3004
3005         num_pages = num_extent_pages(eb->start, eb->len);
3006         for (i = 0; i < num_pages; i++) {
3007                 page = extent_buffer_page(eb, i);
3008                 if (!PageUptodate(page)) {
3009                         pg_uptodate = 0;
3010                         break;
3011                 }
3012         }
3013         if ((ret || ret2) && !pg_uptodate) {
3014 printk("uptodate error2 eb %Lu ret %d ret2 %d pg_uptodate %d\n", eb->start, ret, ret2, pg_uptodate);
3015                 WARN_ON(1);
3016         }
3017         return (ret || ret2);
3018 }
3019 EXPORT_SYMBOL(extent_buffer_uptodate);
3020
3021 int read_extent_buffer_pages(struct extent_io_tree *tree,
3022                              struct extent_buffer *eb,
3023                              u64 start, int wait,
3024                              get_extent_t *get_extent, int mirror_num)
3025 {
3026         unsigned long i;
3027         unsigned long start_i;
3028         struct page *page;
3029         int err;
3030         int ret = 0;
3031         int locked_pages = 0;
3032         int all_uptodate = 1;
3033         int inc_all_pages = 0;
3034         unsigned long num_pages;
3035         struct bio *bio = NULL;
3036
3037         if (eb->flags & EXTENT_UPTODATE)
3038                 return 0;
3039
3040         if (test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3041                            EXTENT_UPTODATE, 1)) {
3042                 return 0;
3043         }
3044
3045         if (start) {
3046                 WARN_ON(start < eb->start);
3047                 start_i = (start >> PAGE_CACHE_SHIFT) -
3048                         (eb->start >> PAGE_CACHE_SHIFT);
3049         } else {
3050                 start_i = 0;
3051         }
3052
3053         num_pages = num_extent_pages(eb->start, eb->len);
3054         for (i = start_i; i < num_pages; i++) {
3055                 page = extent_buffer_page(eb, i);
3056                 if (!wait) {
3057                         if (TestSetPageLocked(page))
3058                                 goto unlock_exit;
3059                 } else {
3060                         lock_page(page);
3061                 }
3062                 locked_pages++;
3063                 if (!PageUptodate(page)) {
3064                         all_uptodate = 0;
3065                 }
3066         }
3067         if (all_uptodate) {
3068                 if (start_i == 0)
3069                         eb->flags |= EXTENT_UPTODATE;
3070                 goto unlock_exit;
3071         }
3072
3073         for (i = start_i; i < num_pages; i++) {
3074                 page = extent_buffer_page(eb, i);
3075                 if (inc_all_pages)
3076                         page_cache_get(page);
3077                 if (!PageUptodate(page)) {
3078                         if (start_i == 0)
3079                                 inc_all_pages = 1;
3080                         ClearPageError(page);
3081                         err = __extent_read_full_page(tree, page,
3082                                                       get_extent, &bio,
3083                                                       mirror_num);
3084                         if (err) {
3085                                 ret = err;
3086                         }
3087                 } else {
3088                         unlock_page(page);
3089                 }
3090         }
3091
3092         if (bio)
3093                 submit_one_bio(READ, bio, mirror_num);
3094
3095         if (ret || !wait) {
3096                 return ret;
3097         }
3098         for (i = start_i; i < num_pages; i++) {
3099                 page = extent_buffer_page(eb, i);
3100                 wait_on_page_locked(page);
3101                 if (!PageUptodate(page)) {
3102                         ret = -EIO;
3103                 }
3104         }
3105         if (!ret)
3106                 eb->flags |= EXTENT_UPTODATE;
3107         return ret;
3108
3109 unlock_exit:
3110         i = start_i;
3111         while(locked_pages > 0) {
3112                 page = extent_buffer_page(eb, i);
3113                 i++;
3114                 unlock_page(page);
3115                 locked_pages--;
3116         }
3117         return ret;
3118 }
3119 EXPORT_SYMBOL(read_extent_buffer_pages);
3120
3121 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
3122                         unsigned long start,
3123                         unsigned long len)
3124 {
3125         size_t cur;
3126         size_t offset;
3127         struct page *page;
3128         char *kaddr;
3129         char *dst = (char *)dstv;
3130         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3131         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3132
3133         WARN_ON(start > eb->len);
3134         WARN_ON(start + len > eb->start + eb->len);
3135
3136         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3137
3138         while(len > 0) {
3139                 page = extent_buffer_page(eb, i);
3140
3141                 cur = min(len, (PAGE_CACHE_SIZE - offset));
3142                 kaddr = kmap_atomic(page, KM_USER1);
3143                 memcpy(dst, kaddr + offset, cur);
3144                 kunmap_atomic(kaddr, KM_USER1);
3145
3146                 dst += cur;
3147                 len -= cur;
3148                 offset = 0;
3149                 i++;
3150         }
3151 }
3152 EXPORT_SYMBOL(read_extent_buffer);
3153
3154 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
3155                                unsigned long min_len, char **token, char **map,
3156                                unsigned long *map_start,
3157                                unsigned long *map_len, int km)
3158 {
3159         size_t offset = start & (PAGE_CACHE_SIZE - 1);
3160         char *kaddr;
3161         struct page *p;
3162         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3163         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3164         unsigned long end_i = (start_offset + start + min_len - 1) >>
3165                 PAGE_CACHE_SHIFT;
3166
3167         if (i != end_i)
3168                 return -EINVAL;
3169
3170         if (i == 0) {
3171                 offset = start_offset;
3172                 *map_start = 0;
3173         } else {
3174                 offset = 0;
3175                 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
3176         }
3177         if (start + min_len > eb->len) {
3178 printk("bad mapping eb start %Lu len %lu, wanted %lu %lu\n", eb->start, eb->len, start, min_len);
3179                 WARN_ON(1);
3180         }
3181
3182         p = extent_buffer_page(eb, i);
3183         kaddr = kmap_atomic(p, km);
3184         *token = kaddr;
3185         *map = kaddr + offset;
3186         *map_len = PAGE_CACHE_SIZE - offset;
3187         return 0;
3188 }
3189 EXPORT_SYMBOL(map_private_extent_buffer);
3190
3191 int map_extent_buffer(struct extent_buffer *eb, unsigned long start,
3192                       unsigned long min_len,
3193                       char **token, char **map,
3194                       unsigned long *map_start,
3195                       unsigned long *map_len, int km)
3196 {
3197         int err;
3198         int save = 0;
3199         if (eb->map_token) {
3200                 unmap_extent_buffer(eb, eb->map_token, km);
3201                 eb->map_token = NULL;
3202                 save = 1;
3203         }
3204         err = map_private_extent_buffer(eb, start, min_len, token, map,
3205                                        map_start, map_len, km);
3206         if (!err && save) {
3207                 eb->map_token = *token;
3208                 eb->kaddr = *map;
3209                 eb->map_start = *map_start;
3210                 eb->map_len = *map_len;
3211         }
3212         return err;
3213 }
3214 EXPORT_SYMBOL(map_extent_buffer);
3215
3216 void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km)
3217 {
3218         kunmap_atomic(token, km);
3219 }
3220 EXPORT_SYMBOL(unmap_extent_buffer);
3221
3222 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
3223                           unsigned long start,
3224                           unsigned long len)
3225 {
3226         size_t cur;
3227         size_t offset;
3228         struct page *page;
3229         char *kaddr;
3230         char *ptr = (char *)ptrv;
3231         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3232         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3233         int ret = 0;
3234
3235         WARN_ON(start > eb->len);
3236         WARN_ON(start + len > eb->start + eb->len);
3237
3238         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3239
3240         while(len > 0) {
3241                 page = extent_buffer_page(eb, i);
3242
3243                 cur = min(len, (PAGE_CACHE_SIZE - offset));
3244
3245                 kaddr = kmap_atomic(page, KM_USER0);
3246                 ret = memcmp(ptr, kaddr + offset, cur);
3247                 kunmap_atomic(kaddr, KM_USER0);
3248                 if (ret)
3249                         break;
3250
3251                 ptr += cur;
3252                 len -= cur;
3253                 offset = 0;
3254                 i++;
3255         }
3256         return ret;
3257 }
3258 EXPORT_SYMBOL(memcmp_extent_buffer);
3259
3260 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
3261                          unsigned long start, unsigned long len)
3262 {
3263         size_t cur;
3264         size_t offset;
3265         struct page *page;
3266         char *kaddr;
3267         char *src = (char *)srcv;
3268         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3269         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3270
3271         WARN_ON(start > eb->len);
3272         WARN_ON(start + len > eb->start + eb->len);
3273
3274         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3275
3276         while(len > 0) {
3277                 page = extent_buffer_page(eb, i);
3278                 WARN_ON(!PageUptodate(page));
3279
3280                 cur = min(len, PAGE_CACHE_SIZE - offset);
3281                 kaddr = kmap_atomic(page, KM_USER1);
3282                 memcpy(kaddr + offset, src, cur);
3283                 kunmap_atomic(kaddr, KM_USER1);
3284
3285                 src += cur;
3286                 len -= cur;
3287                 offset = 0;
3288                 i++;
3289         }
3290 }
3291 EXPORT_SYMBOL(write_extent_buffer);
3292
3293 void memset_extent_buffer(struct extent_buffer *eb, char c,
3294                           unsigned long start, unsigned long len)
3295 {
3296         size_t cur;
3297         size_t offset;
3298         struct page *page;
3299         char *kaddr;
3300         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3301         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3302
3303         WARN_ON(start > eb->len);
3304         WARN_ON(start + len > eb->start + eb->len);
3305
3306         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3307
3308         while(len > 0) {
3309                 page = extent_buffer_page(eb, i);
3310                 WARN_ON(!PageUptodate(page));
3311
3312                 cur = min(len, PAGE_CACHE_SIZE - offset);
3313                 kaddr = kmap_atomic(page, KM_USER0);
3314                 memset(kaddr + offset, c, cur);
3315                 kunmap_atomic(kaddr, KM_USER0);
3316
3317                 len -= cur;
3318                 offset = 0;
3319                 i++;
3320         }
3321 }
3322 EXPORT_SYMBOL(memset_extent_buffer);
3323
3324 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
3325                         unsigned long dst_offset, unsigned long src_offset,
3326                         unsigned long len)
3327 {
3328         u64 dst_len = dst->len;
3329         size_t cur;
3330         size_t offset;
3331         struct page *page;
3332         char *kaddr;
3333         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3334         unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3335
3336         WARN_ON(src->len != dst_len);
3337
3338         offset = (start_offset + dst_offset) &
3339                 ((unsigned long)PAGE_CACHE_SIZE - 1);
3340
3341         while(len > 0) {
3342                 page = extent_buffer_page(dst, i);
3343                 WARN_ON(!PageUptodate(page));
3344
3345                 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
3346
3347                 kaddr = kmap_atomic(page, KM_USER0);
3348                 read_extent_buffer(src, kaddr + offset, src_offset, cur);
3349                 kunmap_atomic(kaddr, KM_USER0);
3350
3351                 src_offset += cur;
3352                 len -= cur;
3353                 offset = 0;
3354                 i++;
3355         }
3356 }
3357 EXPORT_SYMBOL(copy_extent_buffer);
3358
3359 static void move_pages(struct page *dst_page, struct page *src_page,
3360                        unsigned long dst_off, unsigned long src_off,
3361                        unsigned long len)
3362 {
3363         char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3364         if (dst_page == src_page) {
3365                 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
3366         } else {
3367                 char *src_kaddr = kmap_atomic(src_page, KM_USER1);
3368                 char *p = dst_kaddr + dst_off + len;
3369                 char *s = src_kaddr + src_off + len;
3370
3371                 while (len--)
3372                         *--p = *--s;
3373
3374                 kunmap_atomic(src_kaddr, KM_USER1);
3375         }
3376         kunmap_atomic(dst_kaddr, KM_USER0);
3377 }
3378
3379 static void copy_pages(struct page *dst_page, struct page *src_page,
3380                        unsigned long dst_off, unsigned long src_off,
3381                        unsigned long len)
3382 {
3383         char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3384         char *src_kaddr;
3385
3386         if (dst_page != src_page)
3387                 src_kaddr = kmap_atomic(src_page, KM_USER1);
3388         else
3389                 src_kaddr = dst_kaddr;
3390
3391         memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
3392         kunmap_atomic(dst_kaddr, KM_USER0);
3393         if (dst_page != src_page)
3394                 kunmap_atomic(src_kaddr, KM_USER1);
3395 }
3396
3397 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3398                            unsigned long src_offset, unsigned long len)
3399 {
3400         size_t cur;
3401         size_t dst_off_in_page;
3402         size_t src_off_in_page;
3403         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3404         unsigned long dst_i;
3405         unsigned long src_i;
3406
3407         if (src_offset + len > dst->len) {
3408                 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
3409                        src_offset, len, dst->len);
3410                 BUG_ON(1);
3411         }
3412         if (dst_offset + len > dst->len) {
3413                 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
3414                        dst_offset, len, dst->len);
3415                 BUG_ON(1);
3416         }
3417
3418         while(len > 0) {
3419                 dst_off_in_page = (start_offset + dst_offset) &
3420                         ((unsigned long)PAGE_CACHE_SIZE - 1);
3421                 src_off_in_page = (start_offset + src_offset) &
3422                         ((unsigned long)PAGE_CACHE_SIZE - 1);
3423
3424                 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3425                 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
3426
3427                 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
3428                                                src_off_in_page));
3429                 cur = min_t(unsigned long, cur,
3430                         (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
3431
3432                 copy_pages(extent_buffer_page(dst, dst_i),
3433                            extent_buffer_page(dst, src_i),
3434                            dst_off_in_page, src_off_in_page, cur);
3435
3436                 src_offset += cur;
3437                 dst_offset += cur;
3438                 len -= cur;
3439         }
3440 }
3441 EXPORT_SYMBOL(memcpy_extent_buffer);
3442
3443 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3444                            unsigned long src_offset, unsigned long len)
3445 {
3446         size_t cur;
3447         size_t dst_off_in_page;
3448         size_t src_off_in_page;
3449         unsigned long dst_end = dst_offset + len - 1;
3450         unsigned long src_end = src_offset + len - 1;
3451         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3452         unsigned long dst_i;
3453         unsigned long src_i;
3454
3455         if (src_offset + len > dst->len) {
3456                 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
3457                        src_offset, len, dst->len);
3458                 BUG_ON(1);
3459         }
3460         if (dst_offset + len > dst->len) {
3461                 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
3462                        dst_offset, len, dst->len);
3463                 BUG_ON(1);
3464         }
3465         if (dst_offset < src_offset) {
3466                 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
3467                 return;
3468         }
3469         while(len > 0) {
3470                 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
3471                 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
3472
3473                 dst_off_in_page = (start_offset + dst_end) &
3474                         ((unsigned long)PAGE_CACHE_SIZE - 1);
3475                 src_off_in_page = (start_offset + src_end) &
3476                         ((unsigned long)PAGE_CACHE_SIZE - 1);
3477
3478                 cur = min_t(unsigned long, len, src_off_in_page + 1);
3479                 cur = min(cur, dst_off_in_page + 1);
3480                 move_pages(extent_buffer_page(dst, dst_i),
3481                            extent_buffer_page(dst, src_i),
3482                            dst_off_in_page - cur + 1,
3483                            src_off_in_page - cur + 1, cur);
3484
3485                 dst_end -= cur;
3486                 src_end -= cur;
3487                 len -= cur;
3488         }
3489 }
3490 EXPORT_SYMBOL(memmove_extent_buffer);