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