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                 if (!found)
1029                         *end = (u64)-1;
1030                 goto out;
1031         }
1032
1033         while(1) {
1034                 state = rb_entry(node, struct extent_state, rb_node);
1035                 if (found && state->start != cur_start) {
1036                         goto out;
1037                 }
1038                 if (!(state->state & EXTENT_DELALLOC)) {
1039                         if (!found)
1040                                 *end = state->end;
1041                         goto out;
1042                 }
1043                 if (!found) {
1044                         struct extent_state *prev_state;
1045                         struct rb_node *prev_node = node;
1046                         while(1) {
1047                                 prev_node = rb_prev(prev_node);
1048                                 if (!prev_node)
1049                                         break;
1050                                 prev_state = rb_entry(prev_node,
1051                                                       struct extent_state,
1052                                                       rb_node);
1053                                 if (!(prev_state->state & EXTENT_DELALLOC))
1054                                         break;
1055                                 state = prev_state;
1056                                 node = prev_node;
1057                         }
1058                 }
1059                 if (state->state & EXTENT_LOCKED) {
1060                         DEFINE_WAIT(wait);
1061                         atomic_inc(&state->refs);
1062                         prepare_to_wait(&state->wq, &wait,
1063                                         TASK_UNINTERRUPTIBLE);
1064                         spin_unlock_irq(&tree->lock);
1065                         schedule();
1066                         spin_lock_irq(&tree->lock);
1067                         finish_wait(&state->wq, &wait);
1068                         free_extent_state(state);
1069                         goto search_again;
1070                 }
1071                 set_state_cb(tree, state, EXTENT_LOCKED);
1072                 state->state |= EXTENT_LOCKED;
1073                 if (!found)
1074                         *start = state->start;
1075                 found++;
1076                 *end = state->end;
1077                 cur_start = state->end + 1;
1078                 node = rb_next(node);
1079                 if (!node)
1080                         break;
1081                 total_bytes += state->end - state->start + 1;
1082                 if (total_bytes >= max_bytes)
1083                         break;
1084         }
1085 out:
1086         spin_unlock_irq(&tree->lock);
1087         return found;
1088 }
1089
1090 u64 count_range_bits(struct extent_io_tree *tree,
1091                      u64 *start, u64 search_end, u64 max_bytes,
1092                      unsigned long bits)
1093 {
1094         struct rb_node *node;
1095         struct extent_state *state;
1096         u64 cur_start = *start;
1097         u64 total_bytes = 0;
1098         int found = 0;
1099
1100         if (search_end <= cur_start) {
1101                 printk("search_end %Lu start %Lu\n", search_end, cur_start);
1102                 WARN_ON(1);
1103                 return 0;
1104         }
1105
1106         spin_lock_irq(&tree->lock);
1107         if (cur_start == 0 && bits == EXTENT_DIRTY) {
1108                 total_bytes = tree->dirty_bytes;
1109                 goto out;
1110         }
1111         /*
1112          * this search will find all the extents that end after
1113          * our range starts.
1114          */
1115         node = tree_search(tree, cur_start);
1116         if (!node) {
1117                 goto out;
1118         }
1119
1120         while(1) {
1121                 state = rb_entry(node, struct extent_state, rb_node);
1122                 if (state->start > search_end)
1123                         break;
1124                 if (state->end >= cur_start && (state->state & bits)) {
1125                         total_bytes += min(search_end, state->end) + 1 -
1126                                        max(cur_start, state->start);
1127                         if (total_bytes >= max_bytes)
1128                                 break;
1129                         if (!found) {
1130                                 *start = state->start;
1131                                 found = 1;
1132                         }
1133                 }
1134                 node = rb_next(node);
1135                 if (!node)
1136                         break;
1137         }
1138 out:
1139         spin_unlock_irq(&tree->lock);
1140         return total_bytes;
1141 }
1142 /*
1143  * helper function to lock both pages and extents in the tree.
1144  * pages must be locked first.
1145  */
1146 int lock_range(struct extent_io_tree *tree, u64 start, u64 end)
1147 {
1148         unsigned long index = start >> PAGE_CACHE_SHIFT;
1149         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1150         struct page *page;
1151         int err;
1152
1153         while (index <= end_index) {
1154                 page = grab_cache_page(tree->mapping, index);
1155                 if (!page) {
1156                         err = -ENOMEM;
1157                         goto failed;
1158                 }
1159                 if (IS_ERR(page)) {
1160                         err = PTR_ERR(page);
1161                         goto failed;
1162                 }
1163                 index++;
1164         }
1165         lock_extent(tree, start, end, GFP_NOFS);
1166         return 0;
1167
1168 failed:
1169         /*
1170          * we failed above in getting the page at 'index', so we undo here
1171          * up to but not including the page at 'index'
1172          */
1173         end_index = index;
1174         index = start >> PAGE_CACHE_SHIFT;
1175         while (index < end_index) {
1176                 page = find_get_page(tree->mapping, index);
1177                 unlock_page(page);
1178                 page_cache_release(page);
1179                 index++;
1180         }
1181         return err;
1182 }
1183 EXPORT_SYMBOL(lock_range);
1184
1185 /*
1186  * helper function to unlock both pages and extents in the tree.
1187  */
1188 int unlock_range(struct extent_io_tree *tree, u64 start, u64 end)
1189 {
1190         unsigned long index = start >> PAGE_CACHE_SHIFT;
1191         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1192         struct page *page;
1193
1194         while (index <= end_index) {
1195                 page = find_get_page(tree->mapping, index);
1196                 unlock_page(page);
1197                 page_cache_release(page);
1198                 index++;
1199         }
1200         unlock_extent(tree, start, end, GFP_NOFS);
1201         return 0;
1202 }
1203 EXPORT_SYMBOL(unlock_range);
1204
1205 int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1206 {
1207         struct rb_node *node;
1208         struct extent_state *state;
1209         int ret = 0;
1210
1211         spin_lock_irq(&tree->lock);
1212         /*
1213          * this search will find all the extents that end after
1214          * our range starts.
1215          */
1216         node = tree_search(tree, start);
1217         if (!node) {
1218                 ret = -ENOENT;
1219                 goto out;
1220         }
1221         state = rb_entry(node, struct extent_state, rb_node);
1222         if (state->start != start) {
1223                 ret = -ENOENT;
1224                 goto out;
1225         }
1226         state->private = private;
1227 out:
1228         spin_unlock_irq(&tree->lock);
1229         return ret;
1230 }
1231
1232 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1233 {
1234         struct rb_node *node;
1235         struct extent_state *state;
1236         int ret = 0;
1237
1238         spin_lock_irq(&tree->lock);
1239         /*
1240          * this search will find all the extents that end after
1241          * our range starts.
1242          */
1243         node = tree_search(tree, start);
1244         if (!node) {
1245                 ret = -ENOENT;
1246                 goto out;
1247         }
1248         state = rb_entry(node, struct extent_state, rb_node);
1249         if (state->start != start) {
1250                 ret = -ENOENT;
1251                 goto out;
1252         }
1253         *private = state->private;
1254 out:
1255         spin_unlock_irq(&tree->lock);
1256         return ret;
1257 }
1258
1259 /*
1260  * searches a range in the state tree for a given mask.
1261  * If 'filled' == 1, this returns 1 only if every extent in the tree
1262  * has the bits set.  Otherwise, 1 is returned if any bit in the
1263  * range is found set.
1264  */
1265 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1266                    int bits, int filled)
1267 {
1268         struct extent_state *state = NULL;
1269         struct rb_node *node;
1270         int bitset = 0;
1271         unsigned long flags;
1272
1273         spin_lock_irqsave(&tree->lock, flags);
1274         node = tree_search(tree, start);
1275         while (node && start <= end) {
1276                 state = rb_entry(node, struct extent_state, rb_node);
1277
1278                 if (filled && state->start > start) {
1279                         bitset = 0;
1280                         break;
1281                 }
1282
1283                 if (state->start > end)
1284                         break;
1285
1286                 if (state->state & bits) {
1287                         bitset = 1;
1288                         if (!filled)
1289                                 break;
1290                 } else if (filled) {
1291                         bitset = 0;
1292                         break;
1293                 }
1294                 start = state->end + 1;
1295                 if (start > end)
1296                         break;
1297                 node = rb_next(node);
1298                 if (!node) {
1299                         if (filled)
1300                                 bitset = 0;
1301                         break;
1302                 }
1303         }
1304         spin_unlock_irqrestore(&tree->lock, flags);
1305         return bitset;
1306 }
1307 EXPORT_SYMBOL(test_range_bit);
1308
1309 /*
1310  * helper function to set a given page up to date if all the
1311  * extents in the tree for that page are up to date
1312  */
1313 static int check_page_uptodate(struct extent_io_tree *tree,
1314                                struct page *page)
1315 {
1316         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1317         u64 end = start + PAGE_CACHE_SIZE - 1;
1318         if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1))
1319                 SetPageUptodate(page);
1320         return 0;
1321 }
1322
1323 /*
1324  * helper function to unlock a page if all the extents in the tree
1325  * for that page are unlocked
1326  */
1327 static int check_page_locked(struct extent_io_tree *tree,
1328                              struct page *page)
1329 {
1330         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1331         u64 end = start + PAGE_CACHE_SIZE - 1;
1332         if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0))
1333                 unlock_page(page);
1334         return 0;
1335 }
1336
1337 /*
1338  * helper function to end page writeback if all the extents
1339  * in the tree for that page are done with writeback
1340  */
1341 static int check_page_writeback(struct extent_io_tree *tree,
1342                              struct page *page)
1343 {
1344         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1345         u64 end = start + PAGE_CACHE_SIZE - 1;
1346         if (!test_range_bit(tree, start, end, EXTENT_WRITEBACK, 0))
1347                 end_page_writeback(page);
1348         return 0;
1349 }
1350
1351 /* lots and lots of room for performance fixes in the end_bio funcs */
1352
1353 /*
1354  * after a writepage IO is done, we need to:
1355  * clear the uptodate bits on error
1356  * clear the writeback bits in the extent tree for this IO
1357  * end_page_writeback if the page has no more pending IO
1358  *
1359  * Scheduling is not allowed, so the extent state tree is expected
1360  * to have one and only one object corresponding to this IO.
1361  */
1362 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1363 static void end_bio_extent_writepage(struct bio *bio, int err)
1364 #else
1365 static int end_bio_extent_writepage(struct bio *bio,
1366                                    unsigned int bytes_done, int err)
1367 #endif
1368 {
1369         int uptodate = err == 0;
1370         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1371         struct extent_state *state = bio->bi_private;
1372         struct extent_io_tree *tree = state->tree;
1373         struct rb_node *node;
1374         u64 start;
1375         u64 end;
1376         u64 cur;
1377         int whole_page;
1378         int ret;
1379         unsigned long flags;
1380
1381 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1382         if (bio->bi_size)
1383                 return 1;
1384 #endif
1385         do {
1386                 struct page *page = bvec->bv_page;
1387                 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1388                          bvec->bv_offset;
1389                 end = start + bvec->bv_len - 1;
1390
1391                 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1392                         whole_page = 1;
1393                 else
1394                         whole_page = 0;
1395
1396                 if (--bvec >= bio->bi_io_vec)
1397                         prefetchw(&bvec->bv_page->flags);
1398
1399                 if (tree->ops && tree->ops->writepage_end_io_hook) {
1400                         ret = tree->ops->writepage_end_io_hook(page, start,
1401                                                        end, state);
1402                         if (ret)
1403                                 uptodate = 0;
1404                 }
1405
1406                 if (!uptodate && tree->ops &&
1407                     tree->ops->writepage_io_failed_hook) {
1408                         ret = tree->ops->writepage_io_failed_hook(bio, page,
1409                                                          start, end, state);
1410                         if (ret == 0) {
1411                                 state = NULL;
1412                                 uptodate = (err == 0);
1413                                 continue;
1414                         }
1415                 }
1416
1417                 if (!uptodate) {
1418                         clear_extent_uptodate(tree, start, end, GFP_ATOMIC);
1419                         ClearPageUptodate(page);
1420                         SetPageError(page);
1421                 }
1422
1423                 /*
1424                  * bios can get merged in funny ways, and so we need to
1425                  * be careful with the state variable.  We know the
1426                  * state won't be merged with others because it has
1427                  * WRITEBACK set, but we can't be sure each biovec is
1428                  * sequential in the file.  So, if our cached state
1429                  * doesn't match the expected end, search the tree
1430                  * for the correct one.
1431                  */
1432
1433                 spin_lock_irqsave(&tree->lock, flags);
1434                 if (!state || state->end != end) {
1435                         state = NULL;
1436                         node = __etree_search(tree, start, NULL, NULL);
1437                         if (node) {
1438                                 state = rb_entry(node, struct extent_state,
1439                                                  rb_node);
1440                                 if (state->end != end ||
1441                                     !(state->state & EXTENT_WRITEBACK))
1442                                         state = NULL;
1443                         }
1444                         if (!state) {
1445                                 spin_unlock_irqrestore(&tree->lock, flags);
1446                                 clear_extent_writeback(tree, start,
1447                                                        end, GFP_ATOMIC);
1448                                 goto next_io;
1449                         }
1450                 }
1451                 cur = end;
1452                 while(1) {
1453                         struct extent_state *clear = state;
1454                         cur = state->start;
1455                         node = rb_prev(&state->rb_node);
1456                         if (node) {
1457                                 state = rb_entry(node,
1458                                                  struct extent_state,
1459                                                  rb_node);
1460                         } else {
1461                                 state = NULL;
1462                         }
1463
1464                         clear_state_bit(tree, clear, EXTENT_WRITEBACK,
1465                                         1, 0);
1466                         if (cur == start)
1467                                 break;
1468                         if (cur < start) {
1469                                 WARN_ON(1);
1470                                 break;
1471                         }
1472                         if (!node)
1473                                 break;
1474                 }
1475                 /* before releasing the lock, make sure the next state
1476                  * variable has the expected bits set and corresponds
1477                  * to the correct offsets in the file
1478                  */
1479                 if (state && (state->end + 1 != start ||
1480                     !(state->state & EXTENT_WRITEBACK))) {
1481                         state = NULL;
1482                 }
1483                 spin_unlock_irqrestore(&tree->lock, flags);
1484 next_io:
1485
1486                 if (whole_page)
1487                         end_page_writeback(page);
1488                 else
1489                         check_page_writeback(tree, page);
1490         } while (bvec >= bio->bi_io_vec);
1491         bio_put(bio);
1492 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1493         return 0;
1494 #endif
1495 }
1496
1497 /*
1498  * after a readpage IO is done, we need to:
1499  * clear the uptodate bits on error
1500  * set the uptodate bits if things worked
1501  * set the page up to date if all extents in the tree are uptodate
1502  * clear the lock bit in the extent tree
1503  * unlock the page if there are no other extents locked for it
1504  *
1505  * Scheduling is not allowed, so the extent state tree is expected
1506  * to have one and only one object corresponding to this IO.
1507  */
1508 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1509 static void end_bio_extent_readpage(struct bio *bio, int err)
1510 #else
1511 static int end_bio_extent_readpage(struct bio *bio,
1512                                    unsigned int bytes_done, int err)
1513 #endif
1514 {
1515         int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1516         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1517         struct extent_state *state = bio->bi_private;
1518         struct extent_io_tree *tree = state->tree;
1519         struct rb_node *node;
1520         u64 start;
1521         u64 end;
1522         u64 cur;
1523         unsigned long flags;
1524         int whole_page;
1525         int ret;
1526
1527 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1528         if (bio->bi_size)
1529                 return 1;
1530 #endif
1531
1532         do {
1533                 struct page *page = bvec->bv_page;
1534                 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1535                         bvec->bv_offset;
1536                 end = start + bvec->bv_len - 1;
1537
1538                 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1539                         whole_page = 1;
1540                 else
1541                         whole_page = 0;
1542
1543                 if (--bvec >= bio->bi_io_vec)
1544                         prefetchw(&bvec->bv_page->flags);
1545
1546                 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
1547                         ret = tree->ops->readpage_end_io_hook(page, start, end,
1548                                                               state);
1549                         if (ret)
1550                                 uptodate = 0;
1551                 }
1552                 if (!uptodate && tree->ops &&
1553                     tree->ops->readpage_io_failed_hook) {
1554                         ret = tree->ops->readpage_io_failed_hook(bio, page,
1555                                                          start, end, state);
1556                         if (ret == 0) {
1557                                 state = NULL;
1558                                 uptodate =
1559                                         test_bit(BIO_UPTODATE, &bio->bi_flags);
1560                                 continue;
1561                         }
1562                 }
1563
1564                 spin_lock_irqsave(&tree->lock, flags);
1565                 if (!state || state->end != end) {
1566                         state = NULL;
1567                         node = __etree_search(tree, start, NULL, NULL);
1568                         if (node) {
1569                                 state = rb_entry(node, struct extent_state,
1570                                                  rb_node);
1571                                 if (state->end != end ||
1572                                     !(state->state & EXTENT_LOCKED))
1573                                         state = NULL;
1574                         }
1575                         if (!state) {
1576                                 spin_unlock_irqrestore(&tree->lock, flags);
1577                                 if (uptodate)
1578                                         set_extent_uptodate(tree, start, end,
1579                                                             GFP_ATOMIC);
1580                                 unlock_extent(tree, start, end, GFP_ATOMIC);
1581                                 goto next_io;
1582                         }
1583                 }
1584
1585                 cur = end;
1586                 while(1) {
1587                         struct extent_state *clear = state;
1588                         cur = state->start;
1589                         node = rb_prev(&state->rb_node);
1590                         if (node) {
1591                                 state = rb_entry(node,
1592                                          struct extent_state,
1593                                          rb_node);
1594                         } else {
1595                                 state = NULL;
1596                         }
1597                         if (uptodate) {
1598                                 set_state_cb(tree, clear, EXTENT_UPTODATE);
1599                                 clear->state |= EXTENT_UPTODATE;
1600                         }
1601                         clear_state_bit(tree, clear, EXTENT_LOCKED,
1602                                         1, 0);
1603                         if (cur == start)
1604                                 break;
1605                         if (cur < start) {
1606                                 WARN_ON(1);
1607                                 break;
1608                         }
1609                         if (!node)
1610                                 break;
1611                 }
1612                 /* before releasing the lock, make sure the next state
1613                  * variable has the expected bits set and corresponds
1614                  * to the correct offsets in the file
1615                  */
1616                 if (state && (state->end + 1 != start ||
1617                     !(state->state & EXTENT_LOCKED))) {
1618                         state = NULL;
1619                 }
1620                 spin_unlock_irqrestore(&tree->lock, flags);
1621 next_io:
1622                 if (whole_page) {
1623                         if (uptodate) {
1624                                 SetPageUptodate(page);
1625                         } else {
1626                                 ClearPageUptodate(page);
1627                                 SetPageError(page);
1628                         }
1629                         unlock_page(page);
1630                 } else {
1631                         if (uptodate) {
1632                                 check_page_uptodate(tree, page);
1633                         } else {
1634                                 ClearPageUptodate(page);
1635                                 SetPageError(page);
1636                         }
1637                         check_page_locked(tree, page);
1638                 }
1639         } while (bvec >= bio->bi_io_vec);
1640
1641         bio_put(bio);
1642 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1643         return 0;
1644 #endif
1645 }
1646
1647 /*
1648  * IO done from prepare_write is pretty simple, we just unlock
1649  * the structs in the extent tree when done, and set the uptodate bits
1650  * as appropriate.
1651  */
1652 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1653 static void end_bio_extent_preparewrite(struct bio *bio, int err)
1654 #else
1655 static int end_bio_extent_preparewrite(struct bio *bio,
1656                                        unsigned int bytes_done, int err)
1657 #endif
1658 {
1659         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1660         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1661         struct extent_state *state = bio->bi_private;
1662         struct extent_io_tree *tree = state->tree;
1663         u64 start;
1664         u64 end;
1665
1666 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1667         if (bio->bi_size)
1668                 return 1;
1669 #endif
1670
1671         do {
1672                 struct page *page = bvec->bv_page;
1673                 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1674                         bvec->bv_offset;
1675                 end = start + bvec->bv_len - 1;
1676
1677                 if (--bvec >= bio->bi_io_vec)
1678                         prefetchw(&bvec->bv_page->flags);
1679
1680                 if (uptodate) {
1681                         set_extent_uptodate(tree, start, end, GFP_ATOMIC);
1682                 } else {
1683                         ClearPageUptodate(page);
1684                         SetPageError(page);
1685                 }
1686
1687                 unlock_extent(tree, start, end, GFP_ATOMIC);
1688
1689         } while (bvec >= bio->bi_io_vec);
1690
1691         bio_put(bio);
1692 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1693         return 0;
1694 #endif
1695 }
1696
1697 static struct bio *
1698 extent_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
1699                  gfp_t gfp_flags)
1700 {
1701         struct bio *bio;
1702
1703         bio = bio_alloc(gfp_flags, nr_vecs);
1704
1705         if (bio == NULL && (current->flags & PF_MEMALLOC)) {
1706                 while (!bio && (nr_vecs /= 2))
1707                         bio = bio_alloc(gfp_flags, nr_vecs);
1708         }
1709
1710         if (bio) {
1711                 bio->bi_size = 0;
1712                 bio->bi_bdev = bdev;
1713                 bio->bi_sector = first_sector;
1714         }
1715         return bio;
1716 }
1717
1718 static int submit_one_bio(int rw, struct bio *bio, int mirror_num)
1719 {
1720         int ret = 0;
1721         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1722         struct page *page = bvec->bv_page;
1723         struct extent_io_tree *tree = bio->bi_private;
1724         struct rb_node *node;
1725         struct extent_state *state;
1726         u64 start;
1727         u64 end;
1728
1729         start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
1730         end = start + bvec->bv_len - 1;
1731
1732         spin_lock_irq(&tree->lock);
1733         node = __etree_search(tree, start, NULL, NULL);
1734         BUG_ON(!node);
1735         state = rb_entry(node, struct extent_state, rb_node);
1736         while(state->end < end) {
1737                 node = rb_next(node);
1738                 state = rb_entry(node, struct extent_state, rb_node);
1739         }
1740         BUG_ON(state->end != end);
1741         spin_unlock_irq(&tree->lock);
1742
1743         bio->bi_private = state;
1744
1745         bio_get(bio);
1746
1747         if (tree->ops && tree->ops->submit_bio_hook)
1748                 tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
1749                                            mirror_num);
1750         else
1751                 submit_bio(rw, bio);
1752         if (bio_flagged(bio, BIO_EOPNOTSUPP))
1753                 ret = -EOPNOTSUPP;
1754         bio_put(bio);
1755         return ret;
1756 }
1757
1758 static int submit_extent_page(int rw, struct extent_io_tree *tree,
1759                               struct page *page, sector_t sector,
1760                               size_t size, unsigned long offset,
1761                               struct block_device *bdev,
1762                               struct bio **bio_ret,
1763                               unsigned long max_pages,
1764                               bio_end_io_t end_io_func,
1765                               int mirror_num)
1766 {
1767         int ret = 0;
1768         struct bio *bio;
1769         int nr;
1770
1771         if (bio_ret && *bio_ret) {
1772                 bio = *bio_ret;
1773                 if (bio->bi_sector + (bio->bi_size >> 9) != sector ||
1774                     (tree->ops && tree->ops->merge_bio_hook &&
1775                      tree->ops->merge_bio_hook(page, offset, size, bio)) ||
1776                     bio_add_page(bio, page, size, offset) < size) {
1777                         ret = submit_one_bio(rw, bio, mirror_num);
1778                         bio = NULL;
1779                 } else {
1780                         return 0;
1781                 }
1782         }
1783         nr = bio_get_nr_vecs(bdev);
1784         bio = extent_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
1785         if (!bio) {
1786                 printk("failed to allocate bio nr %d\n", nr);
1787         }
1788
1789
1790         bio_add_page(bio, page, size, offset);
1791         bio->bi_end_io = end_io_func;
1792         bio->bi_private = tree;
1793
1794         if (bio_ret) {
1795                 *bio_ret = bio;
1796         } else {
1797                 ret = submit_one_bio(rw, bio, mirror_num);
1798         }
1799
1800         return ret;
1801 }
1802
1803 void set_page_extent_mapped(struct page *page)
1804 {
1805         if (!PagePrivate(page)) {
1806                 SetPagePrivate(page);
1807                 WARN_ON(!page->mapping->a_ops->invalidatepage);
1808                 set_page_private(page, EXTENT_PAGE_PRIVATE);
1809                 page_cache_get(page);
1810         }
1811 }
1812
1813 void set_page_extent_head(struct page *page, unsigned long len)
1814 {
1815         set_page_private(page, EXTENT_PAGE_PRIVATE_FIRST_PAGE | len << 2);
1816 }
1817
1818 /*
1819  * basic readpage implementation.  Locked extent state structs are inserted
1820  * into the tree that are removed when the IO is done (by the end_io
1821  * handlers)
1822  */
1823 static int __extent_read_full_page(struct extent_io_tree *tree,
1824                                    struct page *page,
1825                                    get_extent_t *get_extent,
1826                                    struct bio **bio, int mirror_num)
1827 {
1828         struct inode *inode = page->mapping->host;
1829         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1830         u64 page_end = start + PAGE_CACHE_SIZE - 1;
1831         u64 end;
1832         u64 cur = start;
1833         u64 extent_offset;
1834         u64 last_byte = i_size_read(inode);
1835         u64 block_start;
1836         u64 cur_end;
1837         sector_t sector;
1838         struct extent_map *em;
1839         struct block_device *bdev;
1840         int ret;
1841         int nr = 0;
1842         size_t page_offset = 0;
1843         size_t iosize;
1844         size_t blocksize = inode->i_sb->s_blocksize;
1845
1846         set_page_extent_mapped(page);
1847
1848         end = page_end;
1849         lock_extent(tree, start, end, GFP_NOFS);
1850
1851         while (cur <= end) {
1852                 if (cur >= last_byte) {
1853                         char *userpage;
1854                         iosize = PAGE_CACHE_SIZE - page_offset;
1855                         userpage = kmap_atomic(page, KM_USER0);
1856                         memset(userpage + page_offset, 0, iosize);
1857                         flush_dcache_page(page);
1858                         kunmap_atomic(userpage, KM_USER0);
1859                         set_extent_uptodate(tree, cur, cur + iosize - 1,
1860                                             GFP_NOFS);
1861                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1862                         break;
1863                 }
1864                 em = get_extent(inode, page, page_offset, cur,
1865                                 end - cur + 1, 0);
1866                 if (IS_ERR(em) || !em) {
1867                         SetPageError(page);
1868                         unlock_extent(tree, cur, end, GFP_NOFS);
1869                         break;
1870                 }
1871
1872                 extent_offset = cur - em->start;
1873                 BUG_ON(extent_map_end(em) <= cur);
1874                 BUG_ON(end < cur);
1875
1876                 iosize = min(extent_map_end(em) - cur, end - cur + 1);
1877                 cur_end = min(extent_map_end(em) - 1, end);
1878                 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
1879                 sector = (em->block_start + extent_offset) >> 9;
1880                 bdev = em->bdev;
1881                 block_start = em->block_start;
1882                 free_extent_map(em);
1883                 em = NULL;
1884
1885                 /* we've found a hole, just zero and go on */
1886                 if (block_start == EXTENT_MAP_HOLE) {
1887                         char *userpage;
1888                         userpage = kmap_atomic(page, KM_USER0);
1889                         memset(userpage + page_offset, 0, iosize);
1890                         flush_dcache_page(page);
1891                         kunmap_atomic(userpage, KM_USER0);
1892
1893                         set_extent_uptodate(tree, cur, cur + iosize - 1,
1894                                             GFP_NOFS);
1895                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1896                         cur = cur + iosize;
1897                         page_offset += iosize;
1898                         continue;
1899                 }
1900                 /* the get_extent function already copied into the page */
1901                 if (test_range_bit(tree, cur, cur_end, EXTENT_UPTODATE, 1)) {
1902                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1903                         cur = cur + iosize;
1904                         page_offset += iosize;
1905                         continue;
1906                 }
1907                 /* we have an inline extent but it didn't get marked up
1908                  * to date.  Error out
1909                  */
1910                 if (block_start == EXTENT_MAP_INLINE) {
1911                         SetPageError(page);
1912                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1913                         cur = cur + iosize;
1914                         page_offset += iosize;
1915                         continue;
1916                 }
1917
1918                 ret = 0;
1919                 if (tree->ops && tree->ops->readpage_io_hook) {
1920                         ret = tree->ops->readpage_io_hook(page, cur,
1921                                                           cur + iosize - 1);
1922                 }
1923                 if (!ret) {
1924                         unsigned long nr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
1925                         nr -= page->index;
1926                         ret = submit_extent_page(READ, tree, page,
1927                                          sector, iosize, page_offset,
1928                                          bdev, bio, nr,
1929                                          end_bio_extent_readpage, mirror_num);
1930                 }
1931                 if (ret)
1932                         SetPageError(page);
1933                 cur = cur + iosize;
1934                 page_offset += iosize;
1935                 nr++;
1936         }
1937         if (!nr) {
1938                 if (!PageError(page))
1939                         SetPageUptodate(page);
1940                 unlock_page(page);
1941         }
1942         return 0;
1943 }
1944
1945 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
1946                             get_extent_t *get_extent)
1947 {
1948         struct bio *bio = NULL;
1949         int ret;
1950
1951         ret = __extent_read_full_page(tree, page, get_extent, &bio, 0);
1952         if (bio)
1953                 submit_one_bio(READ, bio, 0);
1954         return ret;
1955 }
1956 EXPORT_SYMBOL(extent_read_full_page);
1957
1958 /*
1959  * the writepage semantics are similar to regular writepage.  extent
1960  * records are inserted to lock ranges in the tree, and as dirty areas
1961  * are found, they are marked writeback.  Then the lock bits are removed
1962  * and the end_io handler clears the writeback ranges
1963  */
1964 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
1965                               void *data)
1966 {
1967         struct inode *inode = page->mapping->host;
1968         struct extent_page_data *epd = data;
1969         struct extent_io_tree *tree = epd->tree;
1970         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1971         u64 delalloc_start;
1972         u64 page_end = start + PAGE_CACHE_SIZE - 1;
1973         u64 end;
1974         u64 cur = start;
1975         u64 extent_offset;
1976         u64 last_byte = i_size_read(inode);
1977         u64 block_start;
1978         u64 iosize;
1979         sector_t sector;
1980         struct extent_map *em;
1981         struct block_device *bdev;
1982         int ret;
1983         int nr = 0;
1984         size_t page_offset = 0;
1985         size_t blocksize;
1986         loff_t i_size = i_size_read(inode);
1987         unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
1988         u64 nr_delalloc;
1989         u64 delalloc_end;
1990
1991         WARN_ON(!PageLocked(page));
1992         if (page->index > end_index) {
1993                 clear_extent_dirty(tree, start, page_end, GFP_NOFS);
1994                 unlock_page(page);
1995                 return 0;
1996         }
1997
1998         if (page->index == end_index) {
1999                 char *userpage;
2000
2001                 size_t offset = i_size & (PAGE_CACHE_SIZE - 1);
2002
2003                 userpage = kmap_atomic(page, KM_USER0);
2004                 memset(userpage + offset, 0, PAGE_CACHE_SIZE - offset);
2005                 flush_dcache_page(page);
2006                 kunmap_atomic(userpage, KM_USER0);
2007         }
2008
2009         set_page_extent_mapped(page);
2010
2011         delalloc_start = start;
2012         delalloc_end = 0;
2013         while(delalloc_end < page_end) {
2014                 nr_delalloc = find_lock_delalloc_range(tree, &delalloc_start,
2015                                                        &delalloc_end,
2016                                                        128 * 1024 * 1024);
2017                 if (nr_delalloc == 0) {
2018                         delalloc_start = delalloc_end + 1;
2019                         continue;
2020                 }
2021                 tree->ops->fill_delalloc(inode, delalloc_start,
2022                                          delalloc_end);
2023                 clear_extent_bit(tree, delalloc_start,
2024                                  delalloc_end,
2025                                  EXTENT_LOCKED | EXTENT_DELALLOC,
2026                                  1, 0, GFP_NOFS);
2027                 delalloc_start = delalloc_end + 1;
2028         }
2029         lock_extent(tree, start, page_end, GFP_NOFS);
2030
2031         end = page_end;
2032         if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0)) {
2033                 printk("found delalloc bits after lock_extent\n");
2034         }
2035
2036         if (last_byte <= start) {
2037                 clear_extent_dirty(tree, start, page_end, GFP_NOFS);
2038                 goto done;
2039         }
2040
2041         set_extent_uptodate(tree, start, page_end, GFP_NOFS);
2042         blocksize = inode->i_sb->s_blocksize;
2043
2044         while (cur <= end) {
2045                 if (cur >= last_byte) {
2046                         clear_extent_dirty(tree, cur, page_end, GFP_NOFS);
2047                         break;
2048                 }
2049                 em = epd->get_extent(inode, page, page_offset, cur,
2050                                      end - cur + 1, 1);
2051                 if (IS_ERR(em) || !em) {
2052                         SetPageError(page);
2053                         break;
2054                 }
2055
2056                 extent_offset = cur - em->start;
2057                 BUG_ON(extent_map_end(em) <= cur);
2058                 BUG_ON(end < cur);
2059                 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2060                 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2061                 sector = (em->block_start + extent_offset) >> 9;
2062                 bdev = em->bdev;
2063                 block_start = em->block_start;
2064                 free_extent_map(em);
2065                 em = NULL;
2066
2067                 if (block_start == EXTENT_MAP_HOLE ||
2068                     block_start == EXTENT_MAP_INLINE) {
2069                         clear_extent_dirty(tree, cur,
2070                                            cur + iosize - 1, GFP_NOFS);
2071                         cur = cur + iosize;
2072                         page_offset += iosize;
2073                         continue;
2074                 }
2075
2076                 /* leave this out until we have a page_mkwrite call */
2077                 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
2078                                    EXTENT_DIRTY, 0)) {
2079                         cur = cur + iosize;
2080                         page_offset += iosize;
2081                         continue;
2082                 }
2083                 clear_extent_dirty(tree, cur, cur + iosize - 1, GFP_NOFS);
2084                 if (tree->ops && tree->ops->writepage_io_hook) {
2085                         ret = tree->ops->writepage_io_hook(page, cur,
2086                                                 cur + iosize - 1);
2087                 } else {
2088                         ret = 0;
2089                 }
2090                 if (ret) {
2091                         SetPageError(page);
2092                 } else {
2093                         unsigned long max_nr = end_index + 1;
2094                         set_range_writeback(tree, cur, cur + iosize - 1);
2095                         if (!PageWriteback(page)) {
2096                                 printk("warning page %lu not writeback, "
2097                                        "cur %llu end %llu\n", page->index,
2098                                        (unsigned long long)cur,
2099                                        (unsigned long long)end);
2100                         }
2101
2102                         ret = submit_extent_page(WRITE, tree, page, sector,
2103                                                  iosize, page_offset, bdev,
2104                                                  &epd->bio, max_nr,
2105                                                  end_bio_extent_writepage, 0);
2106                         if (ret)
2107                                 SetPageError(page);
2108                 }
2109                 cur = cur + iosize;
2110                 page_offset += iosize;
2111                 nr++;
2112         }
2113 done:
2114         if (nr == 0) {
2115                 /* make sure the mapping tag for page dirty gets cleared */
2116                 set_page_writeback(page);
2117                 end_page_writeback(page);
2118         }
2119         unlock_extent(tree, start, page_end, GFP_NOFS);
2120         unlock_page(page);
2121         return 0;
2122 }
2123
2124 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,22)
2125 /* Taken directly from 2.6.23 for 2.6.18 back port */
2126 typedef int (*writepage_t)(struct page *page, struct writeback_control *wbc,
2127                                 void *data);
2128
2129 /**
2130  * write_cache_pages - walk the list of dirty pages of the given address space
2131  * and write all of them.
2132  * @mapping: address space structure to write
2133  * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2134  * @writepage: function called for each page
2135  * @data: data passed to writepage function
2136  *
2137  * If a page is already under I/O, write_cache_pages() skips it, even
2138  * if it's dirty.  This is desirable behaviour for memory-cleaning writeback,
2139  * but it is INCORRECT for data-integrity system calls such as fsync().  fsync()
2140  * and msync() need to guarantee that all the data which was dirty at the time
2141  * the call was made get new I/O started against them.  If wbc->sync_mode is
2142  * WB_SYNC_ALL then we were called for data integrity and we must wait for
2143  * existing IO to complete.
2144  */
2145 static int write_cache_pages(struct address_space *mapping,
2146                       struct writeback_control *wbc, writepage_t writepage,
2147                       void *data)
2148 {
2149         struct backing_dev_info *bdi = mapping->backing_dev_info;
2150         int ret = 0;
2151         int done = 0;
2152         struct pagevec pvec;
2153         int nr_pages;
2154         pgoff_t index;
2155         pgoff_t end;            /* Inclusive */
2156         int scanned = 0;
2157         int range_whole = 0;
2158
2159         if (wbc->nonblocking && bdi_write_congested(bdi)) {
2160                 wbc->encountered_congestion = 1;
2161                 return 0;
2162         }
2163
2164         pagevec_init(&pvec, 0);
2165         if (wbc->range_cyclic) {
2166                 index = mapping->writeback_index; /* Start from prev offset */
2167                 end = -1;
2168         } else {
2169                 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2170                 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2171                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2172                         range_whole = 1;
2173                 scanned = 1;
2174         }
2175 retry:
2176         while (!done && (index <= end) &&
2177                (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
2178                                               PAGECACHE_TAG_DIRTY,
2179                                               min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
2180                 unsigned i;
2181
2182                 scanned = 1;
2183                 for (i = 0; i < nr_pages; i++) {
2184                         struct page *page = pvec.pages[i];
2185
2186                         /*
2187                          * At this point we hold neither mapping->tree_lock nor
2188                          * lock on the page itself: the page may be truncated or
2189                          * invalidated (changing page->mapping to NULL), or even
2190                          * swizzled back from swapper_space to tmpfs file
2191                          * mapping
2192                          */
2193                         lock_page(page);
2194
2195                         if (unlikely(page->mapping != mapping)) {
2196                                 unlock_page(page);
2197                                 continue;
2198                         }
2199
2200                         if (!wbc->range_cyclic && page->index > end) {
2201                                 done = 1;
2202                                 unlock_page(page);
2203                                 continue;
2204                         }
2205
2206                         if (wbc->sync_mode != WB_SYNC_NONE)
2207                                 wait_on_page_writeback(page);
2208
2209                         if (PageWriteback(page) ||
2210                             !clear_page_dirty_for_io(page)) {
2211                                 unlock_page(page);
2212                                 continue;
2213                         }
2214
2215                         ret = (*writepage)(page, wbc, data);
2216
2217                         if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
2218                                 unlock_page(page);
2219                                 ret = 0;
2220                         }
2221                         if (ret || (--(wbc->nr_to_write) <= 0))
2222                                 done = 1;
2223                         if (wbc->nonblocking && bdi_write_congested(bdi)) {
2224                                 wbc->encountered_congestion = 1;
2225                                 done = 1;
2226                         }
2227                 }
2228                 pagevec_release(&pvec);
2229                 cond_resched();
2230         }
2231         if (!scanned && !done) {
2232                 /*
2233                  * We hit the last page and there is more work to be done: wrap
2234                  * back to the start of the file
2235                  */
2236                 scanned = 1;
2237                 index = 0;
2238                 goto retry;
2239         }
2240         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2241                 mapping->writeback_index = index;
2242         return ret;
2243 }
2244 #endif
2245
2246 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
2247                           get_extent_t *get_extent,
2248                           struct writeback_control *wbc)
2249 {
2250         int ret;
2251         struct address_space *mapping = page->mapping;
2252         struct extent_page_data epd = {
2253                 .bio = NULL,
2254                 .tree = tree,
2255                 .get_extent = get_extent,
2256         };
2257         struct writeback_control wbc_writepages = {
2258                 .bdi            = wbc->bdi,
2259                 .sync_mode      = WB_SYNC_NONE,
2260                 .older_than_this = NULL,
2261                 .nr_to_write    = 64,
2262                 .range_start    = page_offset(page) + PAGE_CACHE_SIZE,
2263                 .range_end      = (loff_t)-1,
2264         };
2265
2266
2267         ret = __extent_writepage(page, wbc, &epd);
2268
2269         write_cache_pages(mapping, &wbc_writepages, __extent_writepage, &epd);
2270         if (epd.bio) {
2271                 submit_one_bio(WRITE, epd.bio, 0);
2272         }
2273         return ret;
2274 }
2275 EXPORT_SYMBOL(extent_write_full_page);
2276
2277
2278 int extent_writepages(struct extent_io_tree *tree,
2279                       struct address_space *mapping,
2280                       get_extent_t *get_extent,
2281                       struct writeback_control *wbc)
2282 {
2283         int ret = 0;
2284         struct extent_page_data epd = {
2285                 .bio = NULL,
2286                 .tree = tree,
2287                 .get_extent = get_extent,
2288         };
2289
2290         ret = write_cache_pages(mapping, wbc, __extent_writepage, &epd);
2291         if (epd.bio) {
2292                 submit_one_bio(WRITE, epd.bio, 0);
2293         }
2294         return ret;
2295 }
2296 EXPORT_SYMBOL(extent_writepages);
2297
2298 int extent_readpages(struct extent_io_tree *tree,
2299                      struct address_space *mapping,
2300                      struct list_head *pages, unsigned nr_pages,
2301                      get_extent_t get_extent)
2302 {
2303         struct bio *bio = NULL;
2304         unsigned page_idx;
2305         struct pagevec pvec;
2306
2307         pagevec_init(&pvec, 0);
2308         for (page_idx = 0; page_idx < nr_pages; page_idx++) {
2309                 struct page *page = list_entry(pages->prev, struct page, lru);
2310
2311                 prefetchw(&page->flags);
2312                 list_del(&page->lru);
2313                 /*
2314                  * what we want to do here is call add_to_page_cache_lru,
2315                  * but that isn't exported, so we reproduce it here
2316                  */
2317                 if (!add_to_page_cache(page, mapping,
2318                                         page->index, GFP_KERNEL)) {
2319
2320                         /* open coding of lru_cache_add, also not exported */
2321                         page_cache_get(page);
2322                         if (!pagevec_add(&pvec, page))
2323                                 __pagevec_lru_add(&pvec);
2324                         __extent_read_full_page(tree, page, get_extent,
2325                                                 &bio, 0);
2326                 }
2327                 page_cache_release(page);
2328         }
2329         if (pagevec_count(&pvec))
2330                 __pagevec_lru_add(&pvec);
2331         BUG_ON(!list_empty(pages));
2332         if (bio)
2333                 submit_one_bio(READ, bio, 0);
2334         return 0;
2335 }
2336 EXPORT_SYMBOL(extent_readpages);
2337
2338 /*
2339  * basic invalidatepage code, this waits on any locked or writeback
2340  * ranges corresponding to the page, and then deletes any extent state
2341  * records from the tree
2342  */
2343 int extent_invalidatepage(struct extent_io_tree *tree,
2344                           struct page *page, unsigned long offset)
2345 {
2346         u64 start = ((u64)page->index << PAGE_CACHE_SHIFT);
2347         u64 end = start + PAGE_CACHE_SIZE - 1;
2348         size_t blocksize = page->mapping->host->i_sb->s_blocksize;
2349
2350         start += (offset + blocksize -1) & ~(blocksize - 1);
2351         if (start > end)
2352                 return 0;
2353
2354         lock_extent(tree, start, end, GFP_NOFS);
2355         wait_on_extent_writeback(tree, start, end);
2356         clear_extent_bit(tree, start, end,
2357                          EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC,
2358                          1, 1, GFP_NOFS);
2359         return 0;
2360 }
2361 EXPORT_SYMBOL(extent_invalidatepage);
2362
2363 /*
2364  * simple commit_write call, set_range_dirty is used to mark both
2365  * the pages and the extent records as dirty
2366  */
2367 int extent_commit_write(struct extent_io_tree *tree,
2368                         struct inode *inode, struct page *page,
2369                         unsigned from, unsigned to)
2370 {
2371         loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
2372
2373         set_page_extent_mapped(page);
2374         set_page_dirty(page);
2375
2376         if (pos > inode->i_size) {
2377                 i_size_write(inode, pos);
2378                 mark_inode_dirty(inode);
2379         }
2380         return 0;
2381 }
2382 EXPORT_SYMBOL(extent_commit_write);
2383
2384 int extent_prepare_write(struct extent_io_tree *tree,
2385                          struct inode *inode, struct page *page,
2386                          unsigned from, unsigned to, get_extent_t *get_extent)
2387 {
2388         u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2389         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
2390         u64 block_start;
2391         u64 orig_block_start;
2392         u64 block_end;
2393         u64 cur_end;
2394         struct extent_map *em;
2395         unsigned blocksize = 1 << inode->i_blkbits;
2396         size_t page_offset = 0;
2397         size_t block_off_start;
2398         size_t block_off_end;
2399         int err = 0;
2400         int iocount = 0;
2401         int ret = 0;
2402         int isnew;
2403
2404         set_page_extent_mapped(page);
2405
2406         block_start = (page_start + from) & ~((u64)blocksize - 1);
2407         block_end = (page_start + to - 1) | (blocksize - 1);
2408         orig_block_start = block_start;
2409
2410         lock_extent(tree, page_start, page_end, GFP_NOFS);
2411         while(block_start <= block_end) {
2412                 em = get_extent(inode, page, page_offset, block_start,
2413                                 block_end - block_start + 1, 1);
2414                 if (IS_ERR(em) || !em) {
2415                         goto err;
2416                 }
2417                 cur_end = min(block_end, extent_map_end(em) - 1);
2418                 block_off_start = block_start & (PAGE_CACHE_SIZE - 1);
2419                 block_off_end = block_off_start + blocksize;
2420                 isnew = clear_extent_new(tree, block_start, cur_end, GFP_NOFS);
2421
2422                 if (!PageUptodate(page) && isnew &&
2423                     (block_off_end > to || block_off_start < from)) {
2424                         void *kaddr;
2425
2426                         kaddr = kmap_atomic(page, KM_USER0);
2427                         if (block_off_end > to)
2428                                 memset(kaddr + to, 0, block_off_end - to);
2429                         if (block_off_start < from)
2430                                 memset(kaddr + block_off_start, 0,
2431                                        from - block_off_start);
2432                         flush_dcache_page(page);
2433                         kunmap_atomic(kaddr, KM_USER0);
2434                 }
2435                 if ((em->block_start != EXTENT_MAP_HOLE &&
2436                      em->block_start != EXTENT_MAP_INLINE) &&
2437                     !isnew && !PageUptodate(page) &&
2438                     (block_off_end > to || block_off_start < from) &&
2439                     !test_range_bit(tree, block_start, cur_end,
2440                                     EXTENT_UPTODATE, 1)) {
2441                         u64 sector;
2442                         u64 extent_offset = block_start - em->start;
2443                         size_t iosize;
2444                         sector = (em->block_start + extent_offset) >> 9;
2445                         iosize = (cur_end - block_start + blocksize) &
2446                                 ~((u64)blocksize - 1);
2447                         /*
2448                          * we've already got the extent locked, but we
2449                          * need to split the state such that our end_bio
2450                          * handler can clear the lock.
2451                          */
2452                         set_extent_bit(tree, block_start,
2453                                        block_start + iosize - 1,
2454                                        EXTENT_LOCKED, 0, NULL, GFP_NOFS);
2455                         ret = submit_extent_page(READ, tree, page,
2456                                          sector, iosize, page_offset, em->bdev,
2457                                          NULL, 1,
2458                                          end_bio_extent_preparewrite, 0);
2459                         iocount++;
2460                         block_start = block_start + iosize;
2461                 } else {
2462                         set_extent_uptodate(tree, block_start, cur_end,
2463                                             GFP_NOFS);
2464                         unlock_extent(tree, block_start, cur_end, GFP_NOFS);
2465                         block_start = cur_end + 1;
2466                 }
2467                 page_offset = block_start & (PAGE_CACHE_SIZE - 1);
2468                 free_extent_map(em);
2469         }
2470         if (iocount) {
2471                 wait_extent_bit(tree, orig_block_start,
2472                                 block_end, EXTENT_LOCKED);
2473         }
2474         check_page_uptodate(tree, page);
2475 err:
2476         /* FIXME, zero out newly allocated blocks on error */
2477         return err;
2478 }
2479 EXPORT_SYMBOL(extent_prepare_write);
2480
2481 /*
2482  * a helper for releasepage, this tests for areas of the page that
2483  * are locked or under IO and drops the related state bits if it is safe
2484  * to drop the page.
2485  */
2486 int try_release_extent_state(struct extent_map_tree *map,
2487                              struct extent_io_tree *tree, struct page *page,
2488                              gfp_t mask)
2489 {
2490         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2491         u64 end = start + PAGE_CACHE_SIZE - 1;
2492         int ret = 1;
2493
2494         if (test_range_bit(tree, start, end, EXTENT_IOBITS, 0))
2495                 ret = 0;
2496         else {
2497                 if ((mask & GFP_NOFS) == GFP_NOFS)
2498                         mask = GFP_NOFS;
2499                 clear_extent_bit(tree, start, end, EXTENT_UPTODATE,
2500                                  1, 1, mask);
2501         }
2502         return ret;
2503 }
2504 EXPORT_SYMBOL(try_release_extent_state);
2505
2506 /*
2507  * a helper for releasepage.  As long as there are no locked extents
2508  * in the range corresponding to the page, both state records and extent
2509  * map records are removed
2510  */
2511 int try_release_extent_mapping(struct extent_map_tree *map,
2512                                struct extent_io_tree *tree, struct page *page,
2513                                gfp_t mask)
2514 {
2515         struct extent_map *em;
2516         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2517         u64 end = start + PAGE_CACHE_SIZE - 1;
2518
2519         if ((mask & __GFP_WAIT) &&
2520             page->mapping->host->i_size > 16 * 1024 * 1024) {
2521                 u64 len;
2522                 while (start <= end) {
2523                         len = end - start + 1;
2524                         spin_lock(&map->lock);
2525                         em = lookup_extent_mapping(map, start, len);
2526                         if (!em || IS_ERR(em)) {
2527                                 spin_unlock(&map->lock);
2528                                 break;
2529                         }
2530                         if (em->start != start) {
2531                                 spin_unlock(&map->lock);
2532                                 free_extent_map(em);
2533                                 break;
2534                         }
2535                         if (!test_range_bit(tree, em->start,
2536                                             extent_map_end(em) - 1,
2537                                             EXTENT_LOCKED, 0)) {
2538                                 remove_extent_mapping(map, em);
2539                                 /* once for the rb tree */
2540                                 free_extent_map(em);
2541                         }
2542                         start = extent_map_end(em);
2543                         spin_unlock(&map->lock);
2544
2545                         /* once for us */
2546                         free_extent_map(em);
2547                 }
2548         }
2549         return try_release_extent_state(map, tree, page, mask);
2550 }
2551 EXPORT_SYMBOL(try_release_extent_mapping);
2552
2553 sector_t extent_bmap(struct address_space *mapping, sector_t iblock,
2554                 get_extent_t *get_extent)
2555 {
2556         struct inode *inode = mapping->host;
2557         u64 start = iblock << inode->i_blkbits;
2558         sector_t sector = 0;
2559         struct extent_map *em;
2560
2561         em = get_extent(inode, NULL, 0, start, (1 << inode->i_blkbits), 0);
2562         if (!em || IS_ERR(em))
2563                 return 0;
2564
2565         if (em->block_start == EXTENT_MAP_INLINE ||
2566             em->block_start == EXTENT_MAP_HOLE)
2567                 goto out;
2568
2569         sector = (em->block_start + start - em->start) >> inode->i_blkbits;
2570 out:
2571         free_extent_map(em);
2572         return sector;
2573 }
2574
2575 static int add_lru(struct extent_io_tree *tree, struct extent_buffer *eb)
2576 {
2577         if (list_empty(&eb->lru)) {
2578                 extent_buffer_get(eb);
2579                 list_add(&eb->lru, &tree->buffer_lru);
2580                 tree->lru_size++;
2581                 if (tree->lru_size >= BUFFER_LRU_MAX) {
2582                         struct extent_buffer *rm;
2583                         rm = list_entry(tree->buffer_lru.prev,
2584                                         struct extent_buffer, lru);
2585                         tree->lru_size--;
2586                         list_del_init(&rm->lru);
2587                         free_extent_buffer(rm);
2588                 }
2589         } else
2590                 list_move(&eb->lru, &tree->buffer_lru);
2591         return 0;
2592 }
2593 static struct extent_buffer *find_lru(struct extent_io_tree *tree,
2594                                       u64 start, unsigned long len)
2595 {
2596         struct list_head *lru = &tree->buffer_lru;
2597         struct list_head *cur = lru->next;
2598         struct extent_buffer *eb;
2599
2600         if (list_empty(lru))
2601                 return NULL;
2602
2603         do {
2604                 eb = list_entry(cur, struct extent_buffer, lru);
2605                 if (eb->start == start && eb->len == len) {
2606                         extent_buffer_get(eb);
2607                         return eb;
2608                 }
2609                 cur = cur->next;
2610         } while (cur != lru);
2611         return NULL;
2612 }
2613
2614 static inline unsigned long num_extent_pages(u64 start, u64 len)
2615 {
2616         return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
2617                 (start >> PAGE_CACHE_SHIFT);
2618 }
2619
2620 static inline struct page *extent_buffer_page(struct extent_buffer *eb,
2621                                               unsigned long i)
2622 {
2623         struct page *p;
2624         struct address_space *mapping;
2625
2626         if (i == 0)
2627                 return eb->first_page;
2628         i += eb->start >> PAGE_CACHE_SHIFT;
2629         mapping = eb->first_page->mapping;
2630         read_lock_irq(&mapping->tree_lock);
2631         p = radix_tree_lookup(&mapping->page_tree, i);
2632         read_unlock_irq(&mapping->tree_lock);
2633         return p;
2634 }
2635
2636 int release_extent_buffer_tail_pages(struct extent_buffer *eb)
2637 {
2638         unsigned long num_pages = num_extent_pages(eb->start, eb->len);
2639         struct page *page;
2640         unsigned long i;
2641
2642         if (num_pages == 1)
2643                 return 0;
2644         for (i = 1; i < num_pages; i++) {
2645                 page = extent_buffer_page(eb, i);
2646                 page_cache_release(page);
2647         }
2648         return 0;
2649 }
2650
2651
2652 int invalidate_extent_lru(struct extent_io_tree *tree, u64 start,
2653                           unsigned long len)
2654 {
2655         struct list_head *lru = &tree->buffer_lru;
2656         struct list_head *cur = lru->next;
2657         struct extent_buffer *eb;
2658         int found = 0;
2659
2660         spin_lock(&tree->lru_lock);
2661         if (list_empty(lru))
2662                 goto out;
2663
2664         do {
2665                 eb = list_entry(cur, struct extent_buffer, lru);
2666                 if (eb->start <= start && eb->start + eb->len > start) {
2667                         eb->flags &= ~EXTENT_UPTODATE;
2668                 }
2669                 cur = cur->next;
2670         } while (cur != lru);
2671 out:
2672         spin_unlock(&tree->lru_lock);
2673         return found;
2674 }
2675
2676 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
2677                                                    u64 start,
2678                                                    unsigned long len,
2679                                                    gfp_t mask)
2680 {
2681         struct extent_buffer *eb = NULL;
2682         unsigned long flags;
2683
2684         spin_lock(&tree->lru_lock);
2685         eb = find_lru(tree, start, len);
2686         spin_unlock(&tree->lru_lock);
2687         if (eb) {
2688                 return eb;
2689         }
2690
2691         eb = kmem_cache_zalloc(extent_buffer_cache, mask);
2692         INIT_LIST_HEAD(&eb->lru);
2693         eb->start = start;
2694         eb->len = len;
2695         spin_lock_irqsave(&leak_lock, flags);
2696         list_add(&eb->leak_list, &buffers);
2697         spin_unlock_irqrestore(&leak_lock, flags);
2698         atomic_set(&eb->refs, 1);
2699
2700         return eb;
2701 }
2702
2703 static void __free_extent_buffer(struct extent_buffer *eb)
2704 {
2705         unsigned long flags;
2706         spin_lock_irqsave(&leak_lock, flags);
2707         list_del(&eb->leak_list);
2708         spin_unlock_irqrestore(&leak_lock, flags);
2709         kmem_cache_free(extent_buffer_cache, eb);
2710 }
2711
2712 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
2713                                           u64 start, unsigned long len,
2714                                           struct page *page0,
2715                                           gfp_t mask)
2716 {
2717         unsigned long num_pages = num_extent_pages(start, len);
2718         unsigned long i;
2719         unsigned long index = start >> PAGE_CACHE_SHIFT;
2720         struct extent_buffer *eb;
2721         struct page *p;
2722         struct address_space *mapping = tree->mapping;
2723         int uptodate = 1;
2724
2725         eb = __alloc_extent_buffer(tree, start, len, mask);
2726         if (!eb)
2727                 return NULL;
2728
2729         if (eb->flags & EXTENT_BUFFER_FILLED)
2730                 goto lru_add;
2731
2732         if (page0) {
2733                 eb->first_page = page0;
2734                 i = 1;
2735                 index++;
2736                 page_cache_get(page0);
2737                 mark_page_accessed(page0);
2738                 set_page_extent_mapped(page0);
2739                 set_page_extent_head(page0, len);
2740                 uptodate = PageUptodate(page0);
2741         } else {
2742                 i = 0;
2743         }
2744         for (; i < num_pages; i++, index++) {
2745                 p = find_or_create_page(mapping, index, mask | __GFP_HIGHMEM);
2746                 if (!p) {
2747                         WARN_ON(1);
2748                         goto fail;
2749                 }
2750                 set_page_extent_mapped(p);
2751                 mark_page_accessed(p);
2752                 if (i == 0) {
2753                         eb->first_page = p;
2754                         set_page_extent_head(p, len);
2755                 } else {
2756                         set_page_private(p, EXTENT_PAGE_PRIVATE);
2757                 }
2758                 if (!PageUptodate(p))
2759                         uptodate = 0;
2760                 unlock_page(p);
2761         }
2762         if (uptodate)
2763                 eb->flags |= EXTENT_UPTODATE;
2764         eb->flags |= EXTENT_BUFFER_FILLED;
2765
2766 lru_add:
2767         spin_lock(&tree->lru_lock);
2768         add_lru(tree, eb);
2769         spin_unlock(&tree->lru_lock);
2770         return eb;
2771
2772 fail:
2773         spin_lock(&tree->lru_lock);
2774         list_del_init(&eb->lru);
2775         spin_unlock(&tree->lru_lock);
2776         if (!atomic_dec_and_test(&eb->refs))
2777                 return NULL;
2778         for (index = 1; index < i; index++) {
2779                 page_cache_release(extent_buffer_page(eb, index));
2780         }
2781         if (i > 0)
2782                 page_cache_release(extent_buffer_page(eb, 0));
2783         __free_extent_buffer(eb);
2784         return NULL;
2785 }
2786 EXPORT_SYMBOL(alloc_extent_buffer);
2787
2788 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
2789                                          u64 start, unsigned long len,
2790                                           gfp_t mask)
2791 {
2792         unsigned long num_pages = num_extent_pages(start, len);
2793         unsigned long i;
2794         unsigned long index = start >> PAGE_CACHE_SHIFT;
2795         struct extent_buffer *eb;
2796         struct page *p;
2797         struct address_space *mapping = tree->mapping;
2798         int uptodate = 1;
2799
2800         eb = __alloc_extent_buffer(tree, start, len, mask);
2801         if (!eb)
2802                 return NULL;
2803
2804         if (eb->flags & EXTENT_BUFFER_FILLED)
2805                 goto lru_add;
2806
2807         for (i = 0; i < num_pages; i++, index++) {
2808                 p = find_lock_page(mapping, index);
2809                 if (!p) {
2810                         goto fail;
2811                 }
2812                 set_page_extent_mapped(p);
2813                 mark_page_accessed(p);
2814
2815                 if (i == 0) {
2816                         eb->first_page = p;
2817                         set_page_extent_head(p, len);
2818                 } else {
2819                         set_page_private(p, EXTENT_PAGE_PRIVATE);
2820                 }
2821
2822                 if (!PageUptodate(p))
2823                         uptodate = 0;
2824                 unlock_page(p);
2825         }
2826         if (uptodate)
2827                 eb->flags |= EXTENT_UPTODATE;
2828         eb->flags |= EXTENT_BUFFER_FILLED;
2829
2830 lru_add:
2831         spin_lock(&tree->lru_lock);
2832         add_lru(tree, eb);
2833         spin_unlock(&tree->lru_lock);
2834         return eb;
2835 fail:
2836         spin_lock(&tree->lru_lock);
2837         list_del_init(&eb->lru);
2838         spin_unlock(&tree->lru_lock);
2839         if (!atomic_dec_and_test(&eb->refs))
2840                 return NULL;
2841         for (index = 1; index < i; index++) {
2842                 page_cache_release(extent_buffer_page(eb, index));
2843         }
2844         if (i > 0)
2845                 page_cache_release(extent_buffer_page(eb, 0));
2846         __free_extent_buffer(eb);
2847         return NULL;
2848 }
2849 EXPORT_SYMBOL(find_extent_buffer);
2850
2851 void free_extent_buffer(struct extent_buffer *eb)
2852 {
2853         unsigned long i;
2854         unsigned long num_pages;
2855
2856         if (!eb)
2857                 return;
2858
2859         if (!atomic_dec_and_test(&eb->refs))
2860                 return;
2861
2862         WARN_ON(!list_empty(&eb->lru));
2863         num_pages = num_extent_pages(eb->start, eb->len);
2864
2865         for (i = 1; i < num_pages; i++) {
2866                 page_cache_release(extent_buffer_page(eb, i));
2867         }
2868         page_cache_release(extent_buffer_page(eb, 0));
2869         __free_extent_buffer(eb);
2870 }
2871 EXPORT_SYMBOL(free_extent_buffer);
2872
2873 int clear_extent_buffer_dirty(struct extent_io_tree *tree,
2874                               struct extent_buffer *eb)
2875 {
2876         int set;
2877         unsigned long i;
2878         unsigned long num_pages;
2879         struct page *page;
2880
2881         u64 start = eb->start;
2882         u64 end = start + eb->len - 1;
2883
2884         set = clear_extent_dirty(tree, start, end, GFP_NOFS);
2885         num_pages = num_extent_pages(eb->start, eb->len);
2886
2887         for (i = 0; i < num_pages; i++) {
2888                 page = extent_buffer_page(eb, i);
2889                 lock_page(page);
2890                 if (i == 0)
2891                         set_page_extent_head(page, eb->len);
2892                 else
2893                         set_page_private(page, EXTENT_PAGE_PRIVATE);
2894
2895                 /*
2896                  * if we're on the last page or the first page and the
2897                  * block isn't aligned on a page boundary, do extra checks
2898                  * to make sure we don't clean page that is partially dirty
2899                  */
2900                 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
2901                     ((i == num_pages - 1) &&
2902                      ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
2903                         start = (u64)page->index << PAGE_CACHE_SHIFT;
2904                         end  = start + PAGE_CACHE_SIZE - 1;
2905                         if (test_range_bit(tree, start, end,
2906                                            EXTENT_DIRTY, 0)) {
2907                                 unlock_page(page);
2908                                 continue;
2909                         }
2910                 }
2911                 clear_page_dirty_for_io(page);
2912                 read_lock_irq(&page->mapping->tree_lock);
2913                 if (!PageDirty(page)) {
2914                         radix_tree_tag_clear(&page->mapping->page_tree,
2915                                                 page_index(page),
2916                                                 PAGECACHE_TAG_DIRTY);
2917                 }
2918                 read_unlock_irq(&page->mapping->tree_lock);
2919                 unlock_page(page);
2920         }
2921         return 0;
2922 }
2923 EXPORT_SYMBOL(clear_extent_buffer_dirty);
2924
2925 int wait_on_extent_buffer_writeback(struct extent_io_tree *tree,
2926                                     struct extent_buffer *eb)
2927 {
2928         return wait_on_extent_writeback(tree, eb->start,
2929                                         eb->start + eb->len - 1);
2930 }
2931 EXPORT_SYMBOL(wait_on_extent_buffer_writeback);
2932
2933 int set_extent_buffer_dirty(struct extent_io_tree *tree,
2934                              struct extent_buffer *eb)
2935 {
2936         unsigned long i;
2937         unsigned long num_pages;
2938
2939         num_pages = num_extent_pages(eb->start, eb->len);
2940         for (i = 0; i < num_pages; i++) {
2941                 struct page *page = extent_buffer_page(eb, i);
2942                 /* writepage may need to do something special for the
2943                  * first page, we have to make sure page->private is
2944                  * properly set.  releasepage may drop page->private
2945                  * on us if the page isn't already dirty.
2946                  */
2947                 if (i == 0) {
2948                         lock_page(page);
2949                         set_page_extent_head(page, eb->len);
2950                 } else if (PagePrivate(page) &&
2951                            page->private != EXTENT_PAGE_PRIVATE) {
2952                         lock_page(page);
2953                         set_page_extent_mapped(page);
2954                         unlock_page(page);
2955                 }
2956                 __set_page_dirty_nobuffers(extent_buffer_page(eb, i));
2957                 if (i == 0)
2958                         unlock_page(page);
2959         }
2960         return set_extent_dirty(tree, eb->start,
2961                                 eb->start + eb->len - 1, GFP_NOFS);
2962 }
2963 EXPORT_SYMBOL(set_extent_buffer_dirty);
2964
2965 int clear_extent_buffer_uptodate(struct extent_io_tree *tree,
2966                                 struct extent_buffer *eb)
2967 {
2968         unsigned long i;
2969         struct page *page;
2970         unsigned long num_pages;
2971
2972         num_pages = num_extent_pages(eb->start, eb->len);
2973         eb->flags &= ~EXTENT_UPTODATE;
2974
2975         clear_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
2976                               GFP_NOFS);
2977         for (i = 0; i < num_pages; i++) {
2978                 page = extent_buffer_page(eb, i);
2979                 ClearPageUptodate(page);
2980         }
2981         return 0;
2982 }
2983
2984 int set_extent_buffer_uptodate(struct extent_io_tree *tree,
2985                                 struct extent_buffer *eb)
2986 {
2987         unsigned long i;
2988         struct page *page;
2989         unsigned long num_pages;
2990
2991         num_pages = num_extent_pages(eb->start, eb->len);
2992
2993         set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
2994                             GFP_NOFS);
2995         for (i = 0; i < num_pages; i++) {
2996                 page = extent_buffer_page(eb, i);
2997                 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
2998                     ((i == num_pages - 1) &&
2999                      ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
3000                         check_page_uptodate(tree, page);
3001                         continue;
3002                 }
3003                 SetPageUptodate(page);
3004         }
3005         return 0;
3006 }
3007 EXPORT_SYMBOL(set_extent_buffer_uptodate);
3008
3009 int extent_range_uptodate(struct extent_io_tree *tree,
3010                           u64 start, u64 end)
3011 {
3012         struct page *page;
3013         int ret;
3014         int pg_uptodate = 1;
3015         int uptodate;
3016         unsigned long index;
3017
3018         ret = test_range_bit(tree, start, end, EXTENT_UPTODATE, 1);
3019         if (ret)
3020                 return 1;
3021         while(start <= end) {
3022                 index = start >> PAGE_CACHE_SHIFT;
3023                 page = find_get_page(tree->mapping, index);
3024                 uptodate = PageUptodate(page);
3025                 page_cache_release(page);
3026                 if (!uptodate) {
3027                         pg_uptodate = 0;
3028                         break;
3029                 }
3030                 start += PAGE_CACHE_SIZE;
3031         }
3032         return pg_uptodate;
3033 }
3034
3035 int extent_buffer_uptodate(struct extent_io_tree *tree,
3036                            struct extent_buffer *eb)
3037 {
3038         int ret = 0;
3039         unsigned long num_pages;
3040         unsigned long i;
3041         struct page *page;
3042         int pg_uptodate = 1;
3043
3044         if (eb->flags & EXTENT_UPTODATE)
3045                 return 1;
3046
3047         ret = test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3048                            EXTENT_UPTODATE, 1);
3049         if (ret)
3050                 return ret;
3051
3052         num_pages = num_extent_pages(eb->start, eb->len);
3053         for (i = 0; i < num_pages; i++) {
3054                 page = extent_buffer_page(eb, i);
3055                 if (!PageUptodate(page)) {
3056                         pg_uptodate = 0;
3057                         break;
3058                 }
3059         }
3060         return pg_uptodate;
3061 }
3062 EXPORT_SYMBOL(extent_buffer_uptodate);
3063
3064 int read_extent_buffer_pages(struct extent_io_tree *tree,
3065                              struct extent_buffer *eb,
3066                              u64 start, int wait,
3067                              get_extent_t *get_extent, int mirror_num)
3068 {
3069         unsigned long i;
3070         unsigned long start_i;
3071         struct page *page;
3072         int err;
3073         int ret = 0;
3074         int locked_pages = 0;
3075         int all_uptodate = 1;
3076         int inc_all_pages = 0;
3077         unsigned long num_pages;
3078         struct bio *bio = NULL;
3079
3080         if (eb->flags & EXTENT_UPTODATE)
3081                 return 0;
3082
3083         if (test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3084                            EXTENT_UPTODATE, 1)) {
3085                 return 0;
3086         }
3087
3088         if (start) {
3089                 WARN_ON(start < eb->start);
3090                 start_i = (start >> PAGE_CACHE_SHIFT) -
3091                         (eb->start >> PAGE_CACHE_SHIFT);
3092         } else {
3093                 start_i = 0;
3094         }
3095
3096         num_pages = num_extent_pages(eb->start, eb->len);
3097         for (i = start_i; i < num_pages; i++) {
3098                 page = extent_buffer_page(eb, i);
3099                 if (!wait) {
3100                         if (TestSetPageLocked(page))
3101                                 goto unlock_exit;
3102                 } else {
3103                         lock_page(page);
3104                 }
3105                 locked_pages++;
3106                 if (!PageUptodate(page)) {
3107                         all_uptodate = 0;
3108                 }
3109         }
3110         if (all_uptodate) {
3111                 if (start_i == 0)
3112                         eb->flags |= EXTENT_UPTODATE;
3113                 goto unlock_exit;
3114         }
3115
3116         for (i = start_i; i < num_pages; i++) {
3117                 page = extent_buffer_page(eb, i);
3118                 if (inc_all_pages)
3119                         page_cache_get(page);
3120                 if (!PageUptodate(page)) {
3121                         if (start_i == 0)
3122                                 inc_all_pages = 1;
3123                         ClearPageError(page);
3124                         err = __extent_read_full_page(tree, page,
3125                                                       get_extent, &bio,
3126                                                       mirror_num);
3127                         if (err) {
3128                                 ret = err;
3129                         }
3130                 } else {
3131                         unlock_page(page);
3132                 }
3133         }
3134
3135         if (bio)
3136                 submit_one_bio(READ, bio, mirror_num);
3137
3138         if (ret || !wait) {
3139                 return ret;
3140         }
3141         for (i = start_i; i < num_pages; i++) {
3142                 page = extent_buffer_page(eb, i);
3143                 wait_on_page_locked(page);
3144                 if (!PageUptodate(page)) {
3145                         ret = -EIO;
3146                 }
3147         }
3148         if (!ret)
3149                 eb->flags |= EXTENT_UPTODATE;
3150         return ret;
3151
3152 unlock_exit:
3153         i = start_i;
3154         while(locked_pages > 0) {
3155                 page = extent_buffer_page(eb, i);
3156                 i++;
3157                 unlock_page(page);
3158                 locked_pages--;
3159         }
3160         return ret;
3161 }
3162 EXPORT_SYMBOL(read_extent_buffer_pages);
3163
3164 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
3165                         unsigned long start,
3166                         unsigned long len)
3167 {
3168         size_t cur;
3169         size_t offset;
3170         struct page *page;
3171         char *kaddr;
3172         char *dst = (char *)dstv;
3173         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3174         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3175
3176         WARN_ON(start > eb->len);
3177         WARN_ON(start + len > eb->start + eb->len);
3178
3179         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3180
3181         while(len > 0) {
3182                 page = extent_buffer_page(eb, i);
3183
3184                 cur = min(len, (PAGE_CACHE_SIZE - offset));
3185                 kaddr = kmap_atomic(page, KM_USER1);
3186                 memcpy(dst, kaddr + offset, cur);
3187                 kunmap_atomic(kaddr, KM_USER1);
3188
3189                 dst += cur;
3190                 len -= cur;
3191                 offset = 0;
3192                 i++;
3193         }
3194 }
3195 EXPORT_SYMBOL(read_extent_buffer);
3196
3197 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
3198                                unsigned long min_len, char **token, char **map,
3199                                unsigned long *map_start,
3200                                unsigned long *map_len, int km)
3201 {
3202         size_t offset = start & (PAGE_CACHE_SIZE - 1);
3203         char *kaddr;
3204         struct page *p;
3205         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3206         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3207         unsigned long end_i = (start_offset + start + min_len - 1) >>
3208                 PAGE_CACHE_SHIFT;
3209
3210         if (i != end_i)
3211                 return -EINVAL;
3212
3213         if (i == 0) {
3214                 offset = start_offset;
3215                 *map_start = 0;
3216         } else {
3217                 offset = 0;
3218                 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
3219         }
3220         if (start + min_len > eb->len) {
3221 printk("bad mapping eb start %Lu len %lu, wanted %lu %lu\n", eb->start, eb->len, start, min_len);
3222                 WARN_ON(1);
3223         }
3224
3225         p = extent_buffer_page(eb, i);
3226         kaddr = kmap_atomic(p, km);
3227         *token = kaddr;
3228         *map = kaddr + offset;
3229         *map_len = PAGE_CACHE_SIZE - offset;
3230         return 0;
3231 }
3232 EXPORT_SYMBOL(map_private_extent_buffer);
3233
3234 int map_extent_buffer(struct extent_buffer *eb, unsigned long start,
3235                       unsigned long min_len,
3236                       char **token, char **map,
3237                       unsigned long *map_start,
3238                       unsigned long *map_len, int km)
3239 {
3240         int err;
3241         int save = 0;
3242         if (eb->map_token) {
3243                 unmap_extent_buffer(eb, eb->map_token, km);
3244                 eb->map_token = NULL;
3245                 save = 1;
3246         }
3247         err = map_private_extent_buffer(eb, start, min_len, token, map,
3248                                        map_start, map_len, km);
3249         if (!err && save) {
3250                 eb->map_token = *token;
3251                 eb->kaddr = *map;
3252                 eb->map_start = *map_start;
3253                 eb->map_len = *map_len;
3254         }
3255         return err;
3256 }
3257 EXPORT_SYMBOL(map_extent_buffer);
3258
3259 void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km)
3260 {
3261         kunmap_atomic(token, km);
3262 }
3263 EXPORT_SYMBOL(unmap_extent_buffer);
3264
3265 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
3266                           unsigned long start,
3267                           unsigned long len)
3268 {
3269         size_t cur;
3270         size_t offset;
3271         struct page *page;
3272         char *kaddr;
3273         char *ptr = (char *)ptrv;
3274         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3275         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3276         int ret = 0;
3277
3278         WARN_ON(start > eb->len);
3279         WARN_ON(start + len > eb->start + eb->len);
3280
3281         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3282
3283         while(len > 0) {
3284                 page = extent_buffer_page(eb, i);
3285
3286                 cur = min(len, (PAGE_CACHE_SIZE - offset));
3287
3288                 kaddr = kmap_atomic(page, KM_USER0);
3289                 ret = memcmp(ptr, kaddr + offset, cur);
3290                 kunmap_atomic(kaddr, KM_USER0);
3291                 if (ret)
3292                         break;
3293
3294                 ptr += cur;
3295                 len -= cur;
3296                 offset = 0;
3297                 i++;
3298         }
3299         return ret;
3300 }
3301 EXPORT_SYMBOL(memcmp_extent_buffer);
3302
3303 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
3304                          unsigned long start, unsigned long len)
3305 {
3306         size_t cur;
3307         size_t offset;
3308         struct page *page;
3309         char *kaddr;
3310         char *src = (char *)srcv;
3311         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3312         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3313
3314         WARN_ON(start > eb->len);
3315         WARN_ON(start + len > eb->start + eb->len);
3316
3317         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3318
3319         while(len > 0) {
3320                 page = extent_buffer_page(eb, i);
3321                 WARN_ON(!PageUptodate(page));
3322
3323                 cur = min(len, PAGE_CACHE_SIZE - offset);
3324                 kaddr = kmap_atomic(page, KM_USER1);
3325                 memcpy(kaddr + offset, src, cur);
3326                 kunmap_atomic(kaddr, KM_USER1);
3327
3328                 src += cur;
3329                 len -= cur;
3330                 offset = 0;
3331                 i++;
3332         }
3333 }
3334 EXPORT_SYMBOL(write_extent_buffer);
3335
3336 void memset_extent_buffer(struct extent_buffer *eb, char c,
3337                           unsigned long start, unsigned long len)
3338 {
3339         size_t cur;
3340         size_t offset;
3341         struct page *page;
3342         char *kaddr;
3343         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3344         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3345
3346         WARN_ON(start > eb->len);
3347         WARN_ON(start + len > eb->start + eb->len);
3348
3349         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3350
3351         while(len > 0) {
3352                 page = extent_buffer_page(eb, i);
3353                 WARN_ON(!PageUptodate(page));
3354
3355                 cur = min(len, PAGE_CACHE_SIZE - offset);
3356                 kaddr = kmap_atomic(page, KM_USER0);
3357                 memset(kaddr + offset, c, cur);
3358                 kunmap_atomic(kaddr, KM_USER0);
3359
3360                 len -= cur;
3361                 offset = 0;
3362                 i++;
3363         }
3364 }
3365 EXPORT_SYMBOL(memset_extent_buffer);
3366
3367 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
3368                         unsigned long dst_offset, unsigned long src_offset,
3369                         unsigned long len)
3370 {
3371         u64 dst_len = dst->len;
3372         size_t cur;
3373         size_t offset;
3374         struct page *page;
3375         char *kaddr;
3376         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3377         unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3378
3379         WARN_ON(src->len != dst_len);
3380
3381         offset = (start_offset + dst_offset) &
3382                 ((unsigned long)PAGE_CACHE_SIZE - 1);
3383
3384         while(len > 0) {
3385                 page = extent_buffer_page(dst, i);
3386                 WARN_ON(!PageUptodate(page));
3387
3388                 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
3389
3390                 kaddr = kmap_atomic(page, KM_USER0);
3391                 read_extent_buffer(src, kaddr + offset, src_offset, cur);
3392                 kunmap_atomic(kaddr, KM_USER0);
3393
3394                 src_offset += cur;
3395                 len -= cur;
3396                 offset = 0;
3397                 i++;
3398         }
3399 }
3400 EXPORT_SYMBOL(copy_extent_buffer);
3401
3402 static void move_pages(struct page *dst_page, struct page *src_page,
3403                        unsigned long dst_off, unsigned long src_off,
3404                        unsigned long len)
3405 {
3406         char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3407         if (dst_page == src_page) {
3408                 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
3409         } else {
3410                 char *src_kaddr = kmap_atomic(src_page, KM_USER1);
3411                 char *p = dst_kaddr + dst_off + len;
3412                 char *s = src_kaddr + src_off + len;
3413
3414                 while (len--)
3415                         *--p = *--s;
3416
3417                 kunmap_atomic(src_kaddr, KM_USER1);
3418         }
3419         kunmap_atomic(dst_kaddr, KM_USER0);
3420 }
3421
3422 static void copy_pages(struct page *dst_page, struct page *src_page,
3423                        unsigned long dst_off, unsigned long src_off,
3424                        unsigned long len)
3425 {
3426         char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3427         char *src_kaddr;
3428
3429         if (dst_page != src_page)
3430                 src_kaddr = kmap_atomic(src_page, KM_USER1);
3431         else
3432                 src_kaddr = dst_kaddr;
3433
3434         memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
3435         kunmap_atomic(dst_kaddr, KM_USER0);
3436         if (dst_page != src_page)
3437                 kunmap_atomic(src_kaddr, KM_USER1);
3438 }
3439
3440 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3441                            unsigned long src_offset, unsigned long len)
3442 {
3443         size_t cur;
3444         size_t dst_off_in_page;
3445         size_t src_off_in_page;
3446         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3447         unsigned long dst_i;
3448         unsigned long src_i;
3449
3450         if (src_offset + len > dst->len) {
3451                 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
3452                        src_offset, len, dst->len);
3453                 BUG_ON(1);
3454         }
3455         if (dst_offset + len > dst->len) {
3456                 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
3457                        dst_offset, len, dst->len);
3458                 BUG_ON(1);
3459         }
3460
3461         while(len > 0) {
3462                 dst_off_in_page = (start_offset + dst_offset) &
3463                         ((unsigned long)PAGE_CACHE_SIZE - 1);
3464                 src_off_in_page = (start_offset + src_offset) &
3465                         ((unsigned long)PAGE_CACHE_SIZE - 1);
3466
3467                 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3468                 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
3469
3470                 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
3471                                                src_off_in_page));
3472                 cur = min_t(unsigned long, cur,
3473                         (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
3474
3475                 copy_pages(extent_buffer_page(dst, dst_i),
3476                            extent_buffer_page(dst, src_i),
3477                            dst_off_in_page, src_off_in_page, cur);
3478
3479                 src_offset += cur;
3480                 dst_offset += cur;
3481                 len -= cur;
3482         }
3483 }
3484 EXPORT_SYMBOL(memcpy_extent_buffer);
3485
3486 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3487                            unsigned long src_offset, unsigned long len)
3488 {
3489         size_t cur;
3490         size_t dst_off_in_page;
3491         size_t src_off_in_page;
3492         unsigned long dst_end = dst_offset + len - 1;
3493         unsigned long src_end = src_offset + len - 1;
3494         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3495         unsigned long dst_i;
3496         unsigned long src_i;
3497
3498         if (src_offset + len > dst->len) {
3499                 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
3500                        src_offset, len, dst->len);
3501                 BUG_ON(1);
3502         }
3503         if (dst_offset + len > dst->len) {
3504                 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
3505                        dst_offset, len, dst->len);
3506                 BUG_ON(1);
3507         }
3508         if (dst_offset < src_offset) {
3509                 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
3510                 return;
3511         }
3512         while(len > 0) {
3513                 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
3514                 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
3515
3516                 dst_off_in_page = (start_offset + dst_end) &
3517                         ((unsigned long)PAGE_CACHE_SIZE - 1);
3518                 src_off_in_page = (start_offset + src_end) &
3519                         ((unsigned long)PAGE_CACHE_SIZE - 1);
3520
3521                 cur = min_t(unsigned long, len, src_off_in_page + 1);
3522                 cur = min(cur, dst_off_in_page + 1);
3523                 move_pages(extent_buffer_page(dst, dst_i),
3524                            extent_buffer_page(dst, src_i),
3525                            dst_off_in_page - cur + 1,
3526                            src_off_in_page - cur + 1, cur);
3527
3528                 dst_end -= cur;
3529                 src_end -= cur;
3530                 len -= cur;
3531         }
3532 }
3533 EXPORT_SYMBOL(memmove_extent_buffer);