SAFE public projects git trees. - safe/jmp/linux-2.6/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 "extent_map.h"
  12
  13 /* temporary define until extent_map moves out of btrfs */
  14 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
  15                                        unsigned long extra_flags,
  16                                        void (*ctor)(void *, struct kmem_cache *,
  17                                                     unsigned long));
  18
  19 static struct kmem_cache *extent_map_cache;
  20 static struct kmem_cache *extent_state_cache;
  21 static struct kmem_cache *extent_buffer_cache;
  22 static LIST_HEAD(extent_buffers);
  23 static spinlock_t extent_buffers_lock;
  24 static int nr_extent_buffers;
  25 #define MAX_EXTENT_BUFFER_CACHE 128
  26
  27 struct tree_entry {
  28         u64 start;
  29         u64 end;
  30         int in_tree;
  31         struct rb_node rb_node;
  32 };
  33
  34 void __init extent_map_init(void)
  35 {
  36         extent_map_cache = btrfs_cache_create("extent_map",
  37                                             sizeof(struct extent_map), 0,
  38                                             NULL);
  39         extent_state_cache = btrfs_cache_create("extent_state",
  40                                             sizeof(struct extent_state), 0,
  41                                             NULL);
  42         extent_buffer_cache = btrfs_cache_create("extent_buffers",
  43                                             sizeof(struct extent_buffer), 0,
  44                                             NULL);
  45         spin_lock_init(&extent_buffers_lock);
  46 }
  47
  48 void __exit extent_map_exit(void)
  49 {
  50         struct extent_buffer *eb;
  51
  52         while (!list_empty(&extent_buffers)) {
  53                 eb = list_entry(extent_buffers.next,
  54                                 struct extent_buffer, list);
  55                 list_del(&eb->list);
  56                 kmem_cache_free(extent_buffer_cache, eb);
  57         }
  58         if (extent_map_cache)
  59                 kmem_cache_destroy(extent_map_cache);
  60         if (extent_state_cache)
  61                 kmem_cache_destroy(extent_state_cache);
  62         if (extent_buffer_cache)
  63                 kmem_cache_destroy(extent_buffer_cache);
  64 }
  65
  66 void extent_map_tree_init(struct extent_map_tree *tree,
  67                           struct address_space *mapping, gfp_t mask)
  68 {
  69         tree->map.rb_node = NULL;
  70         tree->state.rb_node = NULL;
  71         tree->ops = NULL;
  72         rwlock_init(&tree->lock);
  73         tree->mapping = mapping;
  74 }
  75 EXPORT_SYMBOL(extent_map_tree_init);
  76
  77 struct extent_map *alloc_extent_map(gfp_t mask)
  78 {
  79         struct extent_map *em;
  80         em = kmem_cache_alloc(extent_map_cache, mask);
  81         if (!em || IS_ERR(em))
  82                 return em;
  83         em->in_tree = 0;
  84         atomic_set(&em->refs, 1);
  85         return em;
  86 }
  87 EXPORT_SYMBOL(alloc_extent_map);
  88
  89 void free_extent_map(struct extent_map *em)
  90 {
  91         if (!em)
  92                 return;
  93         if (atomic_dec_and_test(&em->refs)) {
  94                 WARN_ON(em->in_tree);
  95                 kmem_cache_free(extent_map_cache, em);
  96         }
  97 }
  98 EXPORT_SYMBOL(free_extent_map);
  99
 100
 101 struct extent_state *alloc_extent_state(gfp_t mask)
 102 {
 103         struct extent_state *state;
 104         state = kmem_cache_alloc(extent_state_cache, mask);
 105         if (!state || IS_ERR(state))
 106                 return state;
 107         state->state = 0;
 108         state->in_tree = 0;
 109         state->private = 0;
 110         atomic_set(&state->refs, 1);
 111         init_waitqueue_head(&state->wq);
 112         return state;
 113 }
 114 EXPORT_SYMBOL(alloc_extent_state);
 115
 116 void free_extent_state(struct extent_state *state)
 117 {
 118         if (!state)
 119                 return;
 120         if (atomic_dec_and_test(&state->refs)) {
 121                 WARN_ON(state->in_tree);
 122                 kmem_cache_free(extent_state_cache, state);
 123         }
 124 }
 125 EXPORT_SYMBOL(free_extent_state);
 126
 127 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
 128                                    struct rb_node *node)
 129 {
 130         struct rb_node ** p = &root->rb_node;
 131         struct rb_node * parent = NULL;
 132         struct tree_entry *entry;
 133
 134         while(*p) {
 135                 parent = *p;
 136                 entry = rb_entry(parent, struct tree_entry, rb_node);
 137
 138                 if (offset < entry->start)
 139                         p = &(*p)->rb_left;
 140                 else if (offset > entry->end)
 141                         p = &(*p)->rb_right;
 142                 else
 143                         return parent;
 144         }
 145
 146         entry = rb_entry(node, struct tree_entry, rb_node);
 147         entry->in_tree = 1;
 148         rb_link_node(node, parent, p);
 149         rb_insert_color(node, root);
 150         return NULL;
 151 }
 152
 153 static struct rb_node *__tree_search(struct rb_root *root, u64 offset,
 154                                    struct rb_node **prev_ret)
 155 {
 156         struct rb_node * n = root->rb_node;
 157         struct rb_node *prev = NULL;
 158         struct tree_entry *entry;
 159         struct tree_entry *prev_entry = NULL;
 160
 161         while(n) {
 162                 entry = rb_entry(n, struct tree_entry, rb_node);
 163                 prev = n;
 164                 prev_entry = entry;
 165
 166                 if (offset < entry->start)
 167                         n = n->rb_left;
 168                 else if (offset > entry->end)
 169                         n = n->rb_right;
 170                 else
 171                         return n;
 172         }
 173         if (!prev_ret)
 174                 return NULL;
 175         while(prev && offset > prev_entry->end) {
 176                 prev = rb_next(prev);
 177                 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
 178         }
 179         *prev_ret = prev;
 180         return NULL;
 181 }
 182
 183 static inline struct rb_node *tree_search(struct rb_root *root, u64 offset)
 184 {
 185         struct rb_node *prev;
 186         struct rb_node *ret;
 187         ret = __tree_search(root, offset, &prev);
 188         if (!ret)
 189                 return prev;
 190         return ret;
 191 }
 192
 193 static int tree_delete(struct rb_root *root, u64 offset)
 194 {
 195         struct rb_node *node;
 196         struct tree_entry *entry;
 197
 198         node = __tree_search(root, offset, NULL);
 199         if (!node)
 200                 return -ENOENT;
 201         entry = rb_entry(node, struct tree_entry, rb_node);
 202         entry->in_tree = 0;
 203         rb_erase(node, root);
 204         return 0;
 205 }
 206
 207 /*
 208  * add_extent_mapping tries a simple backward merge with existing
 209  * mappings.  The extent_map struct passed in will be inserted into
 210  * the tree directly (no copies made, just a reference taken).
 211  */
 212 int add_extent_mapping(struct extent_map_tree *tree,
 213                        struct extent_map *em)
 214 {
 215         int ret = 0;
 216         struct extent_map *prev = NULL;
 217         struct rb_node *rb;
 218
 219         write_lock_irq(&tree->lock);
 220         rb = tree_insert(&tree->map, em->end, &em->rb_node);
 221         if (rb) {
 222                 prev = rb_entry(rb, struct extent_map, rb_node);
 223                 printk("found extent map %Lu %Lu on insert of %Lu %Lu\n", prev->start, prev->end, em->start, em->end);
 224                 ret = -EEXIST;
 225                 goto out;
 226         }
 227         atomic_inc(&em->refs);
 228         if (em->start != 0) {
 229                 rb = rb_prev(&em->rb_node);
 230                 if (rb)
 231                         prev = rb_entry(rb, struct extent_map, rb_node);
 232                 if (prev && prev->end + 1 == em->start &&
 233                     ((em->block_start == EXTENT_MAP_HOLE &&
 234                       prev->block_start == EXTENT_MAP_HOLE) ||
 235                              (em->block_start == prev->block_end + 1))) {
 236                         em->start = prev->start;
 237                         em->block_start = prev->block_start;
 238                         rb_erase(&prev->rb_node, &tree->map);
 239                         prev->in_tree = 0;
 240                         free_extent_map(prev);
 241                 }
 242          }
 243 out:
 244         write_unlock_irq(&tree->lock);
 245         return ret;
 246 }
 247 EXPORT_SYMBOL(add_extent_mapping);
 248
 249 /*
 250  * lookup_extent_mapping returns the first extent_map struct in the
 251  * tree that intersects the [start, end] (inclusive) range.  There may
 252  * be additional objects in the tree that intersect, so check the object
 253  * returned carefully to make sure you don't need additional lookups.
 254  */
 255 struct extent_map *lookup_extent_mapping(struct extent_map_tree *tree,
 256                                          u64 start, u64 end)
 257 {
 258         struct extent_map *em;
 259         struct rb_node *rb_node;
 260
 261         read_lock_irq(&tree->lock);
 262         rb_node = tree_search(&tree->map, start);
 263         if (!rb_node) {
 264                 em = NULL;
 265                 goto out;
 266         }
 267         if (IS_ERR(rb_node)) {
 268                 em = ERR_PTR(PTR_ERR(rb_node));
 269                 goto out;
 270         }
 271         em = rb_entry(rb_node, struct extent_map, rb_node);
 272         if (em->end < start || em->start > end) {
 273                 em = NULL;
 274                 goto out;
 275         }
 276         atomic_inc(&em->refs);
 277 out:
 278         read_unlock_irq(&tree->lock);
 279         return em;
 280 }
 281 EXPORT_SYMBOL(lookup_extent_mapping);
 282
 283 /*
 284  * removes an extent_map struct from the tree.  No reference counts are
 285  * dropped, and no checks are done to  see if the range is in use
 286  */
 287 int remove_extent_mapping(struct extent_map_tree *tree, struct extent_map *em)
 288 {
 289         int ret;
 290
 291         write_lock_irq(&tree->lock);
 292         ret = tree_delete(&tree->map, em->end);
 293         write_unlock_irq(&tree->lock);
 294         return ret;
 295 }
 296 EXPORT_SYMBOL(remove_extent_mapping);
 297
 298 /*
 299  * utility function to look for merge candidates inside a given range.
 300  * Any extents with matching state are merged together into a single
 301  * extent in the tree.  Extents with EXTENT_IO in their state field
 302  * are not merged because the end_io handlers need to be able to do
 303  * operations on them without sleeping (or doing allocations/splits).
 304  *
 305  * This should be called with the tree lock held.
 306  */
 307 static int merge_state(struct extent_map_tree *tree,
 308                        struct extent_state *state)
 309 {
 310         struct extent_state *other;
 311         struct rb_node *other_node;
 312
 313         if (state->state & EXTENT_IOBITS)
 314                 return 0;
 315
 316         other_node = rb_prev(&state->rb_node);
 317         if (other_node) {
 318                 other = rb_entry(other_node, struct extent_state, rb_node);
 319                 if (other->end == state->start - 1 &&
 320                     other->state == state->state) {
 321                         state->start = other->start;
 322                         other->in_tree = 0;
 323                         rb_erase(&other->rb_node, &tree->state);
 324                         free_extent_state(other);
 325                 }
 326         }
 327         other_node = rb_next(&state->rb_node);
 328         if (other_node) {
 329                 other = rb_entry(other_node, struct extent_state, rb_node);
 330                 if (other->start == state->end + 1 &&
 331                     other->state == state->state) {
 332                         other->start = state->start;
 333                         state->in_tree = 0;
 334                         rb_erase(&state->rb_node, &tree->state);
 335                         free_extent_state(state);
 336                 }
 337         }
 338         return 0;
 339 }
 340
 341 /*
 342  * insert an extent_state struct into the tree.  'bits' are set on the
 343  * struct before it is inserted.
 344  *
 345  * This may return -EEXIST if the extent is already there, in which case the
 346  * state struct is freed.
 347  *
 348  * The tree lock is not taken internally.  This is a utility function and
 349  * probably isn't what you want to call (see set/clear_extent_bit).
 350  */
 351 static int insert_state(struct extent_map_tree *tree,
 352                         struct extent_state *state, u64 start, u64 end,
 353                         int bits)
 354 {
 355         struct rb_node *node;
 356
 357         if (end < start) {
 358                 printk("end < start %Lu %Lu\n", end, start);
 359                 WARN_ON(1);
 360         }
 361         state->state |= bits;
 362         state->start = start;
 363         state->end = end;
 364         if ((end & 4095) == 0) {
 365                 printk("insert state %Lu %Lu strange end\n", start, end);
 366                 WARN_ON(1);
 367         }
 368         node = tree_insert(&tree->state, end, &state->rb_node);
 369         if (node) {
 370                 struct extent_state *found;
 371                 found = rb_entry(node, struct extent_state, rb_node);
 372                 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, start, end);
 373                 free_extent_state(state);
 374                 return -EEXIST;
 375         }
 376         merge_state(tree, state);
 377         return 0;
 378 }
 379
 380 /*
 381  * split a given extent state struct in two, inserting the preallocated
 382  * struct 'prealloc' as the newly created second half.  'split' indicates an
 383  * offset inside 'orig' where it should be split.
 384  *
 385  * Before calling,
 386  * the tree has 'orig' at [orig->start, orig->end].  After calling, there
 387  * are two extent state structs in the tree:
 388  * prealloc: [orig->start, split - 1]
 389  * orig: [ split, orig->end ]
 390  *
 391  * The tree locks are not taken by this function. They need to be held
 392  * by the caller.
 393  */
 394 static int split_state(struct extent_map_tree *tree, struct extent_state *orig,
 395                        struct extent_state *prealloc, u64 split)
 396 {
 397         struct rb_node *node;
 398         prealloc->start = orig->start;
 399         prealloc->end = split - 1;
 400         prealloc->state = orig->state;
 401         orig->start = split;
 402         if ((prealloc->end & 4095) == 0) {
 403                 printk("insert state %Lu %Lu strange end\n", prealloc->start,
 404                        prealloc->end);
 405                 WARN_ON(1);
 406         }
 407         node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
 408         if (node) {
 409                 struct extent_state *found;
 410                 found = rb_entry(node, struct extent_state, rb_node);
 411                 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, prealloc->start, prealloc->end);
 412                 free_extent_state(prealloc);
 413                 return -EEXIST;
 414         }
 415         return 0;
 416 }
 417
 418 /*
 419  * utility function to clear some bits in an extent state struct.
 420  * it will optionally wake up any one waiting on this state (wake == 1), or
 421  * forcibly remove the state from the tree (delete == 1).
 422  *
 423  * If no bits are set on the state struct after clearing things, the
 424  * struct is freed and removed from the tree
 425  */
 426 static int clear_state_bit(struct extent_map_tree *tree,
 427                             struct extent_state *state, int bits, int wake,
 428                             int delete)
 429 {
 430         int ret = state->state & bits;
 431         state->state &= ~bits;
 432         if (wake)
 433                 wake_up(&state->wq);
 434         if (delete || state->state == 0) {
 435                 if (state->in_tree) {
 436                         rb_erase(&state->rb_node, &tree->state);
 437                         state->in_tree = 0;
 438                         free_extent_state(state);
 439                 } else {
 440                         WARN_ON(1);
 441                 }
 442         } else {
 443                 merge_state(tree, state);
 444         }
 445         return ret;
 446 }
 447
 448 /*
 449  * clear some bits on a range in the tree.  This may require splitting
 450  * or inserting elements in the tree, so the gfp mask is used to
 451  * indicate which allocations or sleeping are allowed.
 452  *
 453  * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
 454  * the given range from the tree regardless of state (ie for truncate).
 455  *
 456  * the range [start, end] is inclusive.
 457  *
 458  * This takes the tree lock, and returns < 0 on error, > 0 if any of the
 459  * bits were already set, or zero if none of the bits were already set.
 460  */
 461 int clear_extent_bit(struct extent_map_tree *tree, u64 start, u64 end,
 462                      int bits, int wake, int delete, gfp_t mask)
 463 {
 464         struct extent_state *state;
 465         struct extent_state *prealloc = NULL;
 466         struct rb_node *node;
 467         unsigned long flags;
 468         int err;
 469         int set = 0;
 470
 471 again:
 472         if (!prealloc && (mask & __GFP_WAIT)) {
 473                 prealloc = alloc_extent_state(mask);
 474                 if (!prealloc)
 475                         return -ENOMEM;
 476         }
 477
 478         write_lock_irqsave(&tree->lock, flags);
 479         /*
 480          * this search will find the extents that end after
 481          * our range starts
 482          */
 483         node = tree_search(&tree->state, start);
 484         if (!node)
 485                 goto out;
 486         state = rb_entry(node, struct extent_state, rb_node);
 487         if (state->start > end)
 488                 goto out;
 489         WARN_ON(state->end < start);
 490
 491         /*
 492          *     | ---- desired range ---- |
 493          *  | state | or
 494          *  | ------------- state -------------- |
 495          *
 496          * We need to split the extent we found, and may flip
 497          * bits on second half.
 498          *
 499          * If the extent we found extends past our range, we
 500          * just split and search again.  It'll get split again
 501          * the next time though.
 502          *
 503          * If the extent we found is inside our range, we clear
 504          * the desired bit on it.
 505          */
 506
 507         if (state->start < start) {
 508                 err = split_state(tree, state, prealloc, start);
 509                 BUG_ON(err == -EEXIST);
 510                 prealloc = NULL;
 511                 if (err)
 512                         goto out;
 513                 if (state->end <= end) {
 514                         start = state->end + 1;
 515                         set |= clear_state_bit(tree, state, bits,
 516                                         wake, delete);
 517                 } else {
 518                         start = state->start;
 519                 }
 520                 goto search_again;
 521         }
 522         /*
 523          * | ---- desired range ---- |
 524          *                        | state |
 525          * We need to split the extent, and clear the bit
 526          * on the first half
 527          */
 528         if (state->start <= end && state->end > end) {
 529                 err = split_state(tree, state, prealloc, end + 1);
 530                 BUG_ON(err == -EEXIST);
 531
 532                 if (wake)
 533                         wake_up(&state->wq);
 534                 set |= clear_state_bit(tree, prealloc, bits,
 535                                        wake, delete);
 536                 prealloc = NULL;
 537                 goto out;
 538         }
 539
 540         start = state->end + 1;
 541         set |= clear_state_bit(tree, state, bits, wake, delete);
 542         goto search_again;
 543
 544 out:
 545         write_unlock_irqrestore(&tree->lock, flags);
 546         if (prealloc)
 547                 free_extent_state(prealloc);
 548
 549         return set;
 550
 551 search_again:
 552         if (start >= end)
 553                 goto out;
 554         write_unlock_irqrestore(&tree->lock, flags);
 555         if (mask & __GFP_WAIT)
 556                 cond_resched();
 557         goto again;
 558 }
 559 EXPORT_SYMBOL(clear_extent_bit);
 560
 561 static int wait_on_state(struct extent_map_tree *tree,
 562                          struct extent_state *state)
 563 {
 564         DEFINE_WAIT(wait);
 565         prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
 566         read_unlock_irq(&tree->lock);
 567         schedule();
 568         read_lock_irq(&tree->lock);
 569         finish_wait(&state->wq, &wait);
 570         return 0;
 571 }
 572
 573 /*
 574  * waits for one or more bits to clear on a range in the state tree.
 575  * The range [start, end] is inclusive.
 576  * The tree lock is taken by this function
 577  */
 578 int wait_extent_bit(struct extent_map_tree *tree, u64 start, u64 end, int bits)
 579 {
 580         struct extent_state *state;
 581         struct rb_node *node;
 582
 583         read_lock_irq(&tree->lock);
 584 again:
 585         while (1) {
 586                 /*
 587                  * this search will find all the extents that end after
 588                  * our range starts
 589                  */
 590                 node = tree_search(&tree->state, start);
 591                 if (!node)
 592                         break;
 593
 594                 state = rb_entry(node, struct extent_state, rb_node);
 595
 596                 if (state->start > end)
 597                         goto out;
 598
 599                 if (state->state & bits) {
 600                         start = state->start;
 601                         atomic_inc(&state->refs);
 602                         wait_on_state(tree, state);
 603                         free_extent_state(state);
 604                         goto again;
 605                 }
 606                 start = state->end + 1;
 607
 608                 if (start > end)
 609                         break;
 610
 611                 if (need_resched()) {
 612                         read_unlock_irq(&tree->lock);
 613                         cond_resched();
 614                         read_lock_irq(&tree->lock);
 615                 }
 616         }
 617 out:
 618         read_unlock_irq(&tree->lock);
 619         return 0;
 620 }
 621 EXPORT_SYMBOL(wait_extent_bit);
 622
 623 /*
 624  * set some bits on a range in the tree.  This may require allocations
 625  * or sleeping, so the gfp mask is used to indicate what is allowed.
 626  *
 627  * If 'exclusive' == 1, this will fail with -EEXIST if some part of the
 628  * range already has the desired bits set.  The start of the existing
 629  * range is returned in failed_start in this case.
 630  *
 631  * [start, end] is inclusive
 632  * This takes the tree lock.
 633  */
 634 int set_extent_bit(struct extent_map_tree *tree, u64 start, u64 end, int bits,
 635                    int exclusive, u64 *failed_start, gfp_t mask)
 636 {
 637         struct extent_state *state;
 638         struct extent_state *prealloc = NULL;
 639         struct rb_node *node;
 640         unsigned long flags;
 641         int err = 0;
 642         int set;
 643         u64 last_start;
 644         u64 last_end;
 645 again:
 646         if (!prealloc && (mask & __GFP_WAIT)) {
 647                 prealloc = alloc_extent_state(mask);
 648                 if (!prealloc)
 649                         return -ENOMEM;
 650         }
 651
 652         write_lock_irqsave(&tree->lock, flags);
 653         /*
 654          * this search will find all the extents that end after
 655          * our range starts.
 656          */
 657         node = tree_search(&tree->state, start);
 658         if (!node) {
 659                 err = insert_state(tree, prealloc, start, end, bits);
 660                 prealloc = NULL;
 661                 BUG_ON(err == -EEXIST);
 662                 goto out;
 663         }
 664
 665         state = rb_entry(node, struct extent_state, rb_node);
 666         last_start = state->start;
 667         last_end = state->end;
 668
 669         /*
 670          * | ---- desired range ---- |
 671          * | state |
 672          *
 673          * Just lock what we found and keep going
 674          */
 675         if (state->start == start && state->end <= end) {
 676                 set = state->state & bits;
 677                 if (set && exclusive) {
 678                         *failed_start = state->start;
 679                         err = -EEXIST;
 680                         goto out;
 681                 }
 682                 state->state |= bits;
 683                 start = state->end + 1;
 684                 merge_state(tree, state);
 685                 goto search_again;
 686         }
 687
 688         /*
 689          *     | ---- desired range ---- |
 690          * | state |
 691          *   or
 692          * | ------------- state -------------- |
 693          *
 694          * We need to split the extent we found, and may flip bits on
 695          * second half.
 696          *
 697          * If the extent we found extends past our
 698          * range, we just split and search again.  It'll get split
 699          * again the next time though.
 700          *
 701          * If the extent we found is inside our range, we set the
 702          * desired bit on it.
 703          */
 704         if (state->start < start) {
 705                 set = state->state & bits;
 706                 if (exclusive && set) {
 707                         *failed_start = start;
 708                         err = -EEXIST;
 709                         goto out;
 710                 }
 711                 err = split_state(tree, state, prealloc, start);
 712                 BUG_ON(err == -EEXIST);
 713                 prealloc = NULL;
 714                 if (err)
 715                         goto out;
 716                 if (state->end <= end) {
 717                         state->state |= bits;
 718                         start = state->end + 1;
 719                         merge_state(tree, state);
 720                 } else {
 721                         start = state->start;
 722                 }
 723                 goto search_again;
 724         }
 725         /*
 726          * | ---- desired range ---- |
 727          *     | state | or               | state |
 728          *
 729          * There's a hole, we need to insert something in it and
 730          * ignore the extent we found.
 731          */
 732         if (state->start > start) {
 733                 u64 this_end;
 734                 if (end < last_start)
 735                         this_end = end;
 736                 else
 737                         this_end = last_start -1;
 738                 err = insert_state(tree, prealloc, start, this_end,
 739                                    bits);
 740                 prealloc = NULL;
 741                 BUG_ON(err == -EEXIST);
 742                 if (err)
 743                         goto out;
 744                 start = this_end + 1;
 745                 goto search_again;
 746         }
 747         /*
 748          * | ---- desired range ---- |
 749          *                        | state |
 750          * We need to split the extent, and set the bit
 751          * on the first half
 752          */
 753         if (state->start <= end && state->end > end) {
 754                 set = state->state & bits;
 755                 if (exclusive && set) {
 756                         *failed_start = start;
 757                         err = -EEXIST;
 758                         goto out;
 759                 }
 760                 err = split_state(tree, state, prealloc, end + 1);
 761                 BUG_ON(err == -EEXIST);
 762
 763                 prealloc->state |= bits;
 764                 merge_state(tree, prealloc);
 765                 prealloc = NULL;
 766                 goto out;
 767         }
 768
 769         goto search_again;
 770
 771 out:
 772         write_unlock_irqrestore(&tree->lock, flags);
 773         if (prealloc)
 774                 free_extent_state(prealloc);
 775
 776         return err;
 777
 778 search_again:
 779         if (start > end)
 780                 goto out;
 781         write_unlock_irqrestore(&tree->lock, flags);
 782         if (mask & __GFP_WAIT)
 783                 cond_resched();
 784         goto again;
 785 }
 786 EXPORT_SYMBOL(set_extent_bit);
 787
 788 /* wrappers around set/clear extent bit */
 789 int set_extent_dirty(struct extent_map_tree *tree, u64 start, u64 end,
 790                      gfp_t mask)
 791 {
 792         return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL,
 793                               mask);
 794 }
 795 EXPORT_SYMBOL(set_extent_dirty);
 796
 797 int set_extent_delalloc(struct extent_map_tree *tree, u64 start, u64 end,
 798                      gfp_t mask)
 799 {
 800         return set_extent_bit(tree, start, end,
 801                               EXTENT_DELALLOC | EXTENT_DIRTY, 0, NULL,
 802                               mask);
 803 }
 804 EXPORT_SYMBOL(set_extent_delalloc);
 805
 806 int clear_extent_dirty(struct extent_map_tree *tree, u64 start, u64 end,
 807                        gfp_t mask)
 808 {
 809         return clear_extent_bit(tree, start, end,
 810                                 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, mask);
 811 }
 812 EXPORT_SYMBOL(clear_extent_dirty);
 813
 814 int set_extent_new(struct extent_map_tree *tree, u64 start, u64 end,
 815                      gfp_t mask)
 816 {
 817         return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL,
 818                               mask);
 819 }
 820 EXPORT_SYMBOL(set_extent_new);
 821
 822 int clear_extent_new(struct extent_map_tree *tree, u64 start, u64 end,
 823                        gfp_t mask)
 824 {
 825         return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0, mask);
 826 }
 827 EXPORT_SYMBOL(clear_extent_new);
 828
 829 int set_extent_uptodate(struct extent_map_tree *tree, u64 start, u64 end,
 830                         gfp_t mask)
 831 {
 832         return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, NULL,
 833                               mask);
 834 }
 835 EXPORT_SYMBOL(set_extent_uptodate);
 836
 837 int clear_extent_uptodate(struct extent_map_tree *tree, u64 start, u64 end,
 838                           gfp_t mask)
 839 {
 840         return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, mask);
 841 }
 842 EXPORT_SYMBOL(clear_extent_uptodate);
 843
 844 int set_extent_writeback(struct extent_map_tree *tree, u64 start, u64 end,
 845                          gfp_t mask)
 846 {
 847         return set_extent_bit(tree, start, end, EXTENT_WRITEBACK,
 848                               0, NULL, mask);
 849 }
 850 EXPORT_SYMBOL(set_extent_writeback);
 851
 852 int clear_extent_writeback(struct extent_map_tree *tree, u64 start, u64 end,
 853                            gfp_t mask)
 854 {
 855         return clear_extent_bit(tree, start, end, EXTENT_WRITEBACK, 1, 0, mask);
 856 }
 857 EXPORT_SYMBOL(clear_extent_writeback);
 858
 859 int wait_on_extent_writeback(struct extent_map_tree *tree, u64 start, u64 end)
 860 {
 861         return wait_extent_bit(tree, start, end, EXTENT_WRITEBACK);
 862 }
 863 EXPORT_SYMBOL(wait_on_extent_writeback);
 864
 865 /*
 866  * locks a range in ascending order, waiting for any locked regions
 867  * it hits on the way.  [start,end] are inclusive, and this will sleep.
 868  */
 869 int lock_extent(struct extent_map_tree *tree, u64 start, u64 end, gfp_t mask)
 870 {
 871         int err;
 872         u64 failed_start;
 873         while (1) {
 874                 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1,
 875                                      &failed_start, mask);
 876                 if (err == -EEXIST && (mask & __GFP_WAIT)) {
 877                         wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
 878                         start = failed_start;
 879                 } else {
 880                         break;
 881                 }
 882                 WARN_ON(start > end);
 883         }
 884         return err;
 885 }
 886 EXPORT_SYMBOL(lock_extent);
 887
 888 int unlock_extent(struct extent_map_tree *tree, u64 start, u64 end,
 889                   gfp_t mask)
 890 {
 891         return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, mask);
 892 }
 893 EXPORT_SYMBOL(unlock_extent);
 894
 895 /*
 896  * helper function to set pages and extents in the tree dirty
 897  */
 898 int set_range_dirty(struct extent_map_tree *tree, u64 start, u64 end)
 899 {
 900         unsigned long index = start >> PAGE_CACHE_SHIFT;
 901         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
 902         struct page *page;
 903
 904         while (index <= end_index) {
 905                 page = find_get_page(tree->mapping, index);
 906                 BUG_ON(!page);
 907                 __set_page_dirty_nobuffers(page);
 908                 page_cache_release(page);
 909                 index++;
 910         }
 911         set_extent_dirty(tree, start, end, GFP_NOFS);
 912         return 0;
 913 }
 914 EXPORT_SYMBOL(set_range_dirty);
 915
 916 /*
 917  * helper function to set both pages and extents in the tree writeback
 918  */
 919 int set_range_writeback(struct extent_map_tree *tree, u64 start, u64 end)
 920 {
 921         unsigned long index = start >> PAGE_CACHE_SHIFT;
 922         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
 923         struct page *page;
 924
 925         while (index <= end_index) {
 926                 page = find_get_page(tree->mapping, index);
 927                 BUG_ON(!page);
 928                 set_page_writeback(page);
 929                 page_cache_release(page);
 930                 index++;
 931         }
 932         set_extent_writeback(tree, start, end, GFP_NOFS);
 933         return 0;
 934 }
 935 EXPORT_SYMBOL(set_range_writeback);
 936
 937 int find_first_extent_bit(struct extent_map_tree *tree, u64 start,
 938                           u64 *start_ret, u64 *end_ret, int bits)
 939 {
 940         struct rb_node *node;
 941         struct extent_state *state;
 942         int ret = 1;
 943
 944         write_lock_irq(&tree->lock);
 945         /*
 946          * this search will find all the extents that end after
 947          * our range starts.
 948          */
 949         node = tree_search(&tree->state, start);
 950         if (!node || IS_ERR(node)) {
 951                 goto out;
 952         }
 953
 954         while(1) {
 955                 state = rb_entry(node, struct extent_state, rb_node);
 956                 if (state->state & bits) {
 957                         *start_ret = state->start;
 958                         *end_ret = state->end;
 959                         ret = 0;
 960                 }
 961                 node = rb_next(node);
 962                 if (!node)
 963                         break;
 964         }
 965 out:
 966         write_unlock_irq(&tree->lock);
 967         return ret;
 968 }
 969 EXPORT_SYMBOL(find_first_extent_bit);
 970
 971 u64 find_lock_delalloc_range(struct extent_map_tree *tree,
 972                              u64 start, u64 lock_start, u64 *end, u64 max_bytes)
 973 {
 974         struct rb_node *node;
 975         struct extent_state *state;
 976         u64 cur_start = start;
 977         u64 found = 0;
 978         u64 total_bytes = 0;
 979
 980         write_lock_irq(&tree->lock);
 981         /*
 982          * this search will find all the extents that end after
 983          * our range starts.
 984          */
 985 search_again:
 986         node = tree_search(&tree->state, cur_start);
 987         if (!node || IS_ERR(node)) {
 988                 goto out;
 989         }
 990
 991         while(1) {
 992                 state = rb_entry(node, struct extent_state, rb_node);
 993                 if (state->start != cur_start) {
 994                         goto out;
 995                 }
 996                 if (!(state->state & EXTENT_DELALLOC)) {
 997                         goto out;
 998                 }
 999                 if (state->start >= lock_start) {
1000                         if (state->state & EXTENT_LOCKED) {
1001                                 DEFINE_WAIT(wait);
1002                                 atomic_inc(&state->refs);
1003                                 write_unlock_irq(&tree->lock);
1004                                 schedule();
1005                                 write_lock_irq(&tree->lock);
1006                                 finish_wait(&state->wq, &wait);
1007                                 free_extent_state(state);
1008                                 goto search_again;
1009                         }
1010                         state->state |= EXTENT_LOCKED;
1011                 }
1012                 found++;
1013                 *end = state->end;
1014                 cur_start = state->end + 1;
1015                 node = rb_next(node);
1016                 if (!node)
1017                         break;
1018                 total_bytes = state->end - state->start + 1;
1019                 if (total_bytes >= max_bytes)
1020                         break;
1021         }
1022 out:
1023         write_unlock_irq(&tree->lock);
1024         return found;
1025 }
1026
1027 /*
1028  * helper function to lock both pages and extents in the tree.
1029  * pages must be locked first.
1030  */
1031 int lock_range(struct extent_map_tree *tree, u64 start, u64 end)
1032 {
1033         unsigned long index = start >> PAGE_CACHE_SHIFT;
1034         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1035         struct page *page;
1036         int err;
1037
1038         while (index <= end_index) {
1039                 page = grab_cache_page(tree->mapping, index);
1040                 if (!page) {
1041                         err = -ENOMEM;
1042                         goto failed;
1043                 }
1044                 if (IS_ERR(page)) {
1045                         err = PTR_ERR(page);
1046                         goto failed;
1047                 }
1048                 index++;
1049         }
1050         lock_extent(tree, start, end, GFP_NOFS);
1051         return 0;
1052
1053 failed:
1054         /*
1055          * we failed above in getting the page at 'index', so we undo here
1056          * up to but not including the page at 'index'
1057          */
1058         end_index = index;
1059         index = start >> PAGE_CACHE_SHIFT;
1060         while (index < end_index) {
1061                 page = find_get_page(tree->mapping, index);
1062                 unlock_page(page);
1063                 page_cache_release(page);
1064                 index++;
1065         }
1066         return err;
1067 }
1068 EXPORT_SYMBOL(lock_range);
1069
1070 /*
1071  * helper function to unlock both pages and extents in the tree.
1072  */
1073 int unlock_range(struct extent_map_tree *tree, u64 start, u64 end)
1074 {
1075         unsigned long index = start >> PAGE_CACHE_SHIFT;
1076         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1077         struct page *page;
1078
1079         while (index <= end_index) {
1080                 page = find_get_page(tree->mapping, index);
1081                 unlock_page(page);
1082                 page_cache_release(page);
1083                 index++;
1084         }
1085         unlock_extent(tree, start, end, GFP_NOFS);
1086         return 0;
1087 }
1088 EXPORT_SYMBOL(unlock_range);
1089
1090 int set_state_private(struct extent_map_tree *tree, u64 start, u64 private)
1091 {
1092         struct rb_node *node;
1093         struct extent_state *state;
1094         int ret = 0;
1095
1096         write_lock_irq(&tree->lock);
1097         /*
1098          * this search will find all the extents that end after
1099          * our range starts.
1100          */
1101         node = tree_search(&tree->state, start);
1102         if (!node || IS_ERR(node)) {
1103                 ret = -ENOENT;
1104                 goto out;
1105         }
1106         state = rb_entry(node, struct extent_state, rb_node);
1107         if (state->start != start) {
1108                 ret = -ENOENT;
1109                 goto out;
1110         }
1111         state->private = private;
1112 out:
1113         write_unlock_irq(&tree->lock);
1114         return ret;
1115
1116 }
1117
1118 int get_state_private(struct extent_map_tree *tree, u64 start, u64 *private)
1119 {
1120         struct rb_node *node;
1121         struct extent_state *state;
1122         int ret = 0;
1123
1124         read_lock_irq(&tree->lock);
1125         /*
1126          * this search will find all the extents that end after
1127          * our range starts.
1128          */
1129         node = tree_search(&tree->state, start);
1130         if (!node || IS_ERR(node)) {
1131                 ret = -ENOENT;
1132                 goto out;
1133         }
1134         state = rb_entry(node, struct extent_state, rb_node);
1135         if (state->start != start) {
1136                 ret = -ENOENT;
1137                 goto out;
1138         }
1139         *private = state->private;
1140 out:
1141         read_unlock_irq(&tree->lock);
1142         return ret;
1143 }
1144
1145 /*
1146  * searches a range in the state tree for a given mask.
1147  * If 'filled' == 1, this returns 1 only if ever extent in the tree
1148  * has the bits set.  Otherwise, 1 is returned if any bit in the
1149  * range is found set.
1150  */
1151 static int test_range_bit(struct extent_map_tree *tree, u64 start, u64 end,
1152                           int bits, int filled)
1153 {
1154         struct extent_state *state = NULL;
1155         struct rb_node *node;
1156         int bitset = 0;
1157
1158         read_lock_irq(&tree->lock);
1159         node = tree_search(&tree->state, start);
1160         while (node && start <= end) {
1161                 state = rb_entry(node, struct extent_state, rb_node);
1162                 if (state->start > end)
1163                         break;
1164
1165                 if (filled && state->start > start) {
1166                         bitset = 0;
1167                         break;
1168                 }
1169                 if (state->state & bits) {
1170                         bitset = 1;
1171                         if (!filled)
1172                                 break;
1173                 } else if (filled) {
1174                         bitset = 0;
1175                         break;
1176                 }
1177                 start = state->end + 1;
1178                 if (start > end)
1179                         break;
1180                 node = rb_next(node);
1181         }
1182         read_unlock_irq(&tree->lock);
1183         return bitset;
1184 }
1185
1186 /*
1187  * helper function to set a given page up to date if all the
1188  * extents in the tree for that page are up to date
1189  */
1190 static int check_page_uptodate(struct extent_map_tree *tree,
1191                                struct page *page)
1192 {
1193         u64 start = page->index << PAGE_CACHE_SHIFT;
1194         u64 end = start + PAGE_CACHE_SIZE - 1;
1195         if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1))
1196                 SetPageUptodate(page);
1197         return 0;
1198 }
1199
1200 /*
1201  * helper function to unlock a page if all the extents in the tree
1202  * for that page are unlocked
1203  */
1204 static int check_page_locked(struct extent_map_tree *tree,
1205                              struct page *page)
1206 {
1207         u64 start = page->index << PAGE_CACHE_SHIFT;
1208         u64 end = start + PAGE_CACHE_SIZE - 1;
1209         if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0))
1210                 unlock_page(page);
1211         return 0;
1212 }
1213
1214 /*
1215  * helper function to end page writeback if all the extents
1216  * in the tree for that page are done with writeback
1217  */
1218 static int check_page_writeback(struct extent_map_tree *tree,
1219                              struct page *page)
1220 {
1221         u64 start = page->index << PAGE_CACHE_SHIFT;
1222         u64 end = start + PAGE_CACHE_SIZE - 1;
1223         if (!test_range_bit(tree, start, end, EXTENT_WRITEBACK, 0))
1224                 end_page_writeback(page);
1225         return 0;
1226 }
1227
1228 /* lots and lots of room for performance fixes in the end_bio funcs */
1229
1230 /*
1231  * after a writepage IO is done, we need to:
1232  * clear the uptodate bits on error
1233  * clear the writeback bits in the extent tree for this IO
1234  * end_page_writeback if the page has no more pending IO
1235  *
1236  * Scheduling is not allowed, so the extent state tree is expected
1237  * to have one and only one object corresponding to this IO.
1238  */
1239 static int end_bio_extent_writepage(struct bio *bio,
1240                                    unsigned int bytes_done, int err)
1241 {
1242         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1243         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1244         struct extent_map_tree *tree = bio->bi_private;
1245         u64 start;
1246         u64 end;
1247         int whole_page;
1248
1249         if (bio->bi_size)
1250                 return 1;
1251
1252         do {
1253                 struct page *page = bvec->bv_page;
1254                 start = (page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
1255                 end = start + bvec->bv_len - 1;
1256
1257                 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1258                         whole_page = 1;
1259                 else
1260                         whole_page = 0;
1261
1262                 if (--bvec >= bio->bi_io_vec)
1263                         prefetchw(&bvec->bv_page->flags);
1264
1265                 if (!uptodate) {
1266                         clear_extent_uptodate(tree, start, end, GFP_ATOMIC);
1267                         ClearPageUptodate(page);
1268                         SetPageError(page);
1269                 }
1270                 clear_extent_writeback(tree, start, end, GFP_ATOMIC);
1271
1272                 if (whole_page)
1273                         end_page_writeback(page);
1274                 else
1275                         check_page_writeback(tree, page);
1276                 if (tree->ops && tree->ops->writepage_end_io_hook)
1277                         tree->ops->writepage_end_io_hook(page, start, end);
1278         } while (bvec >= bio->bi_io_vec);
1279
1280         bio_put(bio);
1281         return 0;
1282 }
1283
1284 /*
1285  * after a readpage IO is done, we need to:
1286  * clear the uptodate bits on error
1287  * set the uptodate bits if things worked
1288  * set the page up to date if all extents in the tree are uptodate
1289  * clear the lock bit in the extent tree
1290  * unlock the page if there are no other extents locked for it
1291  *
1292  * Scheduling is not allowed, so the extent state tree is expected
1293  * to have one and only one object corresponding to this IO.
1294  */
1295 static int end_bio_extent_readpage(struct bio *bio,
1296                                    unsigned int bytes_done, int err)
1297 {
1298         int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1299         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1300         struct extent_map_tree *tree = bio->bi_private;
1301         u64 start;
1302         u64 end;
1303         int whole_page;
1304         int ret;
1305
1306         if (bio->bi_size)
1307                 return 1;
1308
1309         do {
1310                 struct page *page = bvec->bv_page;
1311                 start = (page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
1312                 end = start + bvec->bv_len - 1;
1313
1314                 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1315                         whole_page = 1;
1316                 else
1317                         whole_page = 0;
1318
1319                 if (--bvec >= bio->bi_io_vec)
1320                         prefetchw(&bvec->bv_page->flags);
1321
1322                 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
1323                         ret = tree->ops->readpage_end_io_hook(page, start, end);
1324                         if (ret)
1325                                 uptodate = 0;
1326                 }
1327                 if (uptodate) {
1328                         set_extent_uptodate(tree, start, end, GFP_ATOMIC);
1329                         if (whole_page)
1330                                 SetPageUptodate(page);
1331                         else
1332                                 check_page_uptodate(tree, page);
1333                 } else {
1334                         ClearPageUptodate(page);
1335                         SetPageError(page);
1336                 }
1337
1338                 unlock_extent(tree, start, end, GFP_ATOMIC);
1339
1340                 if (whole_page)
1341                         unlock_page(page);
1342                 else
1343                         check_page_locked(tree, page);
1344         } while (bvec >= bio->bi_io_vec);
1345
1346         bio_put(bio);
1347         return 0;
1348 }
1349
1350 /*
1351  * IO done from prepare_write is pretty simple, we just unlock
1352  * the structs in the extent tree when done, and set the uptodate bits
1353  * as appropriate.
1354  */
1355 static int end_bio_extent_preparewrite(struct bio *bio,
1356                                        unsigned int bytes_done, int err)
1357 {
1358         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1359         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1360         struct extent_map_tree *tree = bio->bi_private;
1361         u64 start;
1362         u64 end;
1363
1364         if (bio->bi_size)
1365                 return 1;
1366
1367         do {
1368                 struct page *page = bvec->bv_page;
1369                 start = (page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
1370                 end = start + bvec->bv_len - 1;
1371
1372                 if (--bvec >= bio->bi_io_vec)
1373                         prefetchw(&bvec->bv_page->flags);
1374
1375                 if (uptodate) {
1376                         set_extent_uptodate(tree, start, end, GFP_ATOMIC);
1377                 } else {
1378                         ClearPageUptodate(page);
1379                         SetPageError(page);
1380                 }
1381
1382                 unlock_extent(tree, start, end, GFP_ATOMIC);
1383
1384         } while (bvec >= bio->bi_io_vec);
1385
1386         bio_put(bio);
1387         return 0;
1388 }
1389
1390 static int submit_extent_page(int rw, struct extent_map_tree *tree,
1391                               struct page *page, sector_t sector,
1392                               size_t size, unsigned long offset,
1393                               struct block_device *bdev,
1394                               bio_end_io_t end_io_func)
1395 {
1396         struct bio *bio;
1397         int ret = 0;
1398
1399         bio = bio_alloc(GFP_NOIO, 1);
1400
1401         bio->bi_sector = sector;
1402         bio->bi_bdev = bdev;
1403         bio->bi_io_vec[0].bv_page = page;
1404         bio->bi_io_vec[0].bv_len = size;
1405         bio->bi_io_vec[0].bv_offset = offset;
1406
1407         bio->bi_vcnt = 1;
1408         bio->bi_idx = 0;
1409         bio->bi_size = size;
1410
1411         bio->bi_end_io = end_io_func;
1412         bio->bi_private = tree;
1413
1414         bio_get(bio);
1415         submit_bio(rw, bio);
1416
1417         if (bio_flagged(bio, BIO_EOPNOTSUPP))
1418                 ret = -EOPNOTSUPP;
1419
1420         bio_put(bio);
1421         return ret;
1422 }
1423
1424 void set_page_extent_mapped(struct page *page)
1425 {
1426         if (!PagePrivate(page)) {
1427                 SetPagePrivate(page);
1428                 WARN_ON(!page->mapping->a_ops->invalidatepage);
1429                 set_page_private(page, 1);
1430                 page_cache_get(page);
1431         }
1432 }
1433
1434 /*
1435  * basic readpage implementation.  Locked extent state structs are inserted
1436  * into the tree that are removed when the IO is done (by the end_io
1437  * handlers)
1438  */
1439 int extent_read_full_page(struct extent_map_tree *tree, struct page *page,
1440                           get_extent_t *get_extent)
1441 {
1442         struct inode *inode = page->mapping->host;
1443         u64 start = page->index << PAGE_CACHE_SHIFT;
1444         u64 page_end = start + PAGE_CACHE_SIZE - 1;
1445         u64 end;
1446         u64 cur = start;
1447         u64 extent_offset;
1448         u64 last_byte = i_size_read(inode);
1449         u64 block_start;
1450         u64 cur_end;
1451         sector_t sector;
1452         struct extent_map *em;
1453         struct block_device *bdev;
1454         int ret;
1455         int nr = 0;
1456         size_t page_offset = 0;
1457         size_t iosize;
1458         size_t blocksize = inode->i_sb->s_blocksize;
1459
1460         set_page_extent_mapped(page);
1461
1462         end = page_end;
1463         lock_extent(tree, start, end, GFP_NOFS);
1464
1465         while (cur <= end) {
1466                 if (cur >= last_byte) {
1467                         iosize = PAGE_CACHE_SIZE - page_offset;
1468                         zero_user_page(page, page_offset, iosize, KM_USER0);
1469                         set_extent_uptodate(tree, cur, cur + iosize - 1,
1470                                             GFP_NOFS);
1471                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1472                         break;
1473                 }
1474                 em = get_extent(inode, page, page_offset, cur, end, 0);
1475                 if (IS_ERR(em) || !em) {
1476                         SetPageError(page);
1477                         unlock_extent(tree, cur, end, GFP_NOFS);
1478                         break;
1479                 }
1480
1481                 extent_offset = cur - em->start;
1482                 BUG_ON(em->end < cur);
1483                 BUG_ON(end < cur);
1484
1485                 iosize = min(em->end - cur, end - cur) + 1;
1486                 cur_end = min(em->end, end);
1487                 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
1488                 sector = (em->block_start + extent_offset) >> 9;
1489                 bdev = em->bdev;
1490                 block_start = em->block_start;
1491                 free_extent_map(em);
1492                 em = NULL;
1493
1494                 /* we've found a hole, just zero and go on */
1495                 if (block_start == EXTENT_MAP_HOLE) {
1496                         zero_user_page(page, page_offset, iosize, KM_USER0);
1497                         set_extent_uptodate(tree, cur, cur + iosize - 1,
1498                                             GFP_NOFS);
1499                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1500                         cur = cur + iosize;
1501                         page_offset += iosize;
1502                         continue;
1503                 }
1504                 /* the get_extent function already copied into the page */
1505                 if (test_range_bit(tree, cur, cur_end, EXTENT_UPTODATE, 1)) {
1506                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1507                         cur = cur + iosize;
1508                         page_offset += iosize;
1509                         continue;
1510                 }
1511
1512                 ret = 0;
1513                 if (tree->ops && tree->ops->readpage_io_hook) {
1514                         ret = tree->ops->readpage_io_hook(page, cur,
1515                                                           cur + iosize - 1);
1516                 }
1517                 if (!ret) {
1518                         ret = submit_extent_page(READ, tree, page,
1519                                                  sector, iosize, page_offset,
1520                                                  bdev, end_bio_extent_readpage);
1521                 }
1522                 if (ret)
1523                         SetPageError(page);
1524                 cur = cur + iosize;
1525                 page_offset += iosize;
1526                 nr++;
1527         }
1528         if (!nr) {
1529                 if (!PageError(page))
1530                         SetPageUptodate(page);
1531                 unlock_page(page);
1532         }
1533         return 0;
1534 }
1535 EXPORT_SYMBOL(extent_read_full_page);
1536
1537 /*
1538  * the writepage semantics are similar to regular writepage.  extent
1539  * records are inserted to lock ranges in the tree, and as dirty areas
1540  * are found, they are marked writeback.  Then the lock bits are removed
1541  * and the end_io handler clears the writeback ranges
1542  */
1543 int extent_write_full_page(struct extent_map_tree *tree, struct page *page,
1544                           get_extent_t *get_extent,
1545                           struct writeback_control *wbc)
1546 {
1547         struct inode *inode = page->mapping->host;
1548         u64 start = page->index << PAGE_CACHE_SHIFT;
1549         u64 page_end = start + PAGE_CACHE_SIZE - 1;
1550         u64 end;
1551         u64 cur = start;
1552         u64 extent_offset;
1553         u64 last_byte = i_size_read(inode);
1554         u64 block_start;
1555         sector_t sector;
1556         struct extent_map *em;
1557         struct block_device *bdev;
1558         int ret;
1559         int nr = 0;
1560         size_t page_offset = 0;
1561         size_t iosize;
1562         size_t blocksize;
1563         loff_t i_size = i_size_read(inode);
1564         unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
1565         u64 nr_delalloc;
1566         u64 delalloc_end;
1567
1568         WARN_ON(!PageLocked(page));
1569         if (page->index > end_index) {
1570                 clear_extent_dirty(tree, start, page_end, GFP_NOFS);
1571                 unlock_page(page);
1572                 return 0;
1573         }
1574
1575         if (page->index == end_index) {
1576                 size_t offset = i_size & (PAGE_CACHE_SIZE - 1);
1577                 zero_user_page(page, offset,
1578                                PAGE_CACHE_SIZE - offset, KM_USER0);
1579         }
1580
1581         set_page_extent_mapped(page);
1582
1583         lock_extent(tree, start, page_end, GFP_NOFS);
1584         nr_delalloc = find_lock_delalloc_range(tree, start, page_end + 1,
1585                                                &delalloc_end,
1586                                                128 * 1024 * 1024);
1587         if (nr_delalloc) {
1588                 tree->ops->fill_delalloc(inode, start, delalloc_end);
1589                 if (delalloc_end >= page_end + 1) {
1590                         clear_extent_bit(tree, page_end + 1, delalloc_end,
1591                                          EXTENT_LOCKED | EXTENT_DELALLOC,
1592                                          1, 0, GFP_NOFS);
1593                 }
1594                 clear_extent_bit(tree, start, page_end, EXTENT_DELALLOC,
1595                                  0, 0, GFP_NOFS);
1596                 if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0)) {
1597                         printk("found delalloc bits after clear extent_bit\n");
1598                 }
1599         } else if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0)) {
1600                 printk("found delalloc bits after find_delalloc_range returns 0\n");
1601         }
1602
1603         end = page_end;
1604         if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0)) {
1605                 printk("found delalloc bits after lock_extent\n");
1606         }
1607
1608         if (last_byte <= start) {
1609                 clear_extent_dirty(tree, start, page_end, GFP_NOFS);
1610                 goto done;
1611         }
1612
1613         set_extent_uptodate(tree, start, page_end, GFP_NOFS);
1614         blocksize = inode->i_sb->s_blocksize;
1615
1616         while (cur <= end) {
1617                 if (cur >= last_byte) {
1618                         clear_extent_dirty(tree, cur, page_end, GFP_NOFS);
1619                         break;
1620                 }
1621                 em = get_extent(inode, page, page_offset, cur, end, 0);
1622                 if (IS_ERR(em) || !em) {
1623                         SetPageError(page);
1624                         break;
1625                 }
1626
1627                 extent_offset = cur - em->start;
1628                 BUG_ON(em->end < cur);
1629                 BUG_ON(end < cur);
1630                 iosize = min(em->end - cur, end - cur) + 1;
1631                 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
1632                 sector = (em->block_start + extent_offset) >> 9;
1633                 bdev = em->bdev;
1634                 block_start = em->block_start;
1635                 free_extent_map(em);
1636                 em = NULL;
1637
1638                 if (block_start == EXTENT_MAP_HOLE ||
1639                     block_start == EXTENT_MAP_INLINE) {
1640                         clear_extent_dirty(tree, cur,
1641                                            cur + iosize - 1, GFP_NOFS);
1642                         cur = cur + iosize;
1643                         page_offset += iosize;
1644                         continue;
1645                 }
1646
1647                 /* leave this out until we have a page_mkwrite call */
1648                 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
1649                                    EXTENT_DIRTY, 0)) {
1650                         cur = cur + iosize;
1651                         page_offset += iosize;
1652                         continue;
1653                 }
1654                 clear_extent_dirty(tree, cur, cur + iosize - 1, GFP_NOFS);
1655                 if (tree->ops && tree->ops->writepage_io_hook) {
1656                         ret = tree->ops->writepage_io_hook(page, cur,
1657                                                 cur + iosize - 1);
1658                 } else {
1659                         ret = 0;
1660                 }
1661                 if (ret)
1662                         SetPageError(page);
1663                 else {
1664                         set_range_writeback(tree, cur, cur + iosize - 1);
1665                         ret = submit_extent_page(WRITE, tree, page, sector,
1666                                                  iosize, page_offset, bdev,
1667                                                  end_bio_extent_writepage);
1668                         if (ret)
1669                                 SetPageError(page);
1670                 }
1671                 cur = cur + iosize;
1672                 page_offset += iosize;
1673                 nr++;
1674         }
1675 done:
1676         unlock_extent(tree, start, page_end, GFP_NOFS);
1677         unlock_page(page);
1678         return 0;
1679 }
1680 EXPORT_SYMBOL(extent_write_full_page);
1681
1682 /*
1683  * basic invalidatepage code, this waits on any locked or writeback
1684  * ranges corresponding to the page, and then deletes any extent state
1685  * records from the tree
1686  */
1687 int extent_invalidatepage(struct extent_map_tree *tree,
1688                           struct page *page, unsigned long offset)
1689 {
1690         u64 start = (page->index << PAGE_CACHE_SHIFT);
1691         u64 end = start + PAGE_CACHE_SIZE - 1;
1692         size_t blocksize = page->mapping->host->i_sb->s_blocksize;
1693
1694         start += (offset + blocksize -1) & ~(blocksize - 1);
1695         if (start > end)
1696                 return 0;
1697
1698         lock_extent(tree, start, end, GFP_NOFS);
1699         wait_on_extent_writeback(tree, start, end);
1700         clear_extent_bit(tree, start, end,
1701                          EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC,
1702                          1, 1, GFP_NOFS);
1703         return 0;
1704 }
1705 EXPORT_SYMBOL(extent_invalidatepage);
1706
1707 /*
1708  * simple commit_write call, set_range_dirty is used to mark both
1709  * the pages and the extent records as dirty
1710  */
1711 int extent_commit_write(struct extent_map_tree *tree,
1712                         struct inode *inode, struct page *page,
1713                         unsigned from, unsigned to)
1714 {
1715         loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1716
1717         set_page_extent_mapped(page);
1718         set_page_dirty(page);
1719
1720         if (pos > inode->i_size) {
1721                 i_size_write(inode, pos);
1722                 mark_inode_dirty(inode);
1723         }
1724         return 0;
1725 }
1726 EXPORT_SYMBOL(extent_commit_write);
1727
1728 int extent_prepare_write(struct extent_map_tree *tree,
1729                          struct inode *inode, struct page *page,
1730                          unsigned from, unsigned to, get_extent_t *get_extent)
1731 {
1732         u64 page_start = page->index << PAGE_CACHE_SHIFT;
1733         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
1734         u64 block_start;
1735         u64 orig_block_start;
1736         u64 block_end;
1737         u64 cur_end;
1738         struct extent_map *em;
1739         unsigned blocksize = 1 << inode->i_blkbits;
1740         size_t page_offset = 0;
1741         size_t block_off_start;
1742         size_t block_off_end;
1743         int err = 0;
1744         int iocount = 0;
1745         int ret = 0;
1746         int isnew;
1747
1748         set_page_extent_mapped(page);
1749
1750         block_start = (page_start + from) & ~((u64)blocksize - 1);
1751         block_end = (page_start + to - 1) | (blocksize - 1);
1752         orig_block_start = block_start;
1753
1754         lock_extent(tree, page_start, page_end, GFP_NOFS);
1755         while(block_start <= block_end) {
1756                 em = get_extent(inode, page, page_offset, block_start,
1757                                 block_end, 1);
1758                 if (IS_ERR(em) || !em) {
1759                         goto err;
1760                 }
1761                 cur_end = min(block_end, em->end);
1762                 block_off_start = block_start & (PAGE_CACHE_SIZE - 1);
1763                 block_off_end = block_off_start + blocksize;
1764                 isnew = clear_extent_new(tree, block_start, cur_end, GFP_NOFS);
1765
1766                 if (!PageUptodate(page) && isnew &&
1767                     (block_off_end > to || block_off_start < from)) {
1768                         void *kaddr;
1769
1770                         kaddr = kmap_atomic(page, KM_USER0);
1771                         if (block_off_end > to)
1772                                 memset(kaddr + to, 0, block_off_end - to);
1773                         if (block_off_start < from)
1774                                 memset(kaddr + block_off_start, 0,
1775                                        from - block_off_start);
1776                         flush_dcache_page(page);
1777                         kunmap_atomic(kaddr, KM_USER0);
1778                 }
1779                 if (!isnew && !PageUptodate(page) &&
1780                     (block_off_end > to || block_off_start < from) &&
1781                     !test_range_bit(tree, block_start, cur_end,
1782                                     EXTENT_UPTODATE, 1)) {
1783                         u64 sector;
1784                         u64 extent_offset = block_start - em->start;
1785                         size_t iosize;
1786                         sector = (em->block_start + extent_offset) >> 9;
1787                         iosize = (cur_end - block_start + blocksize - 1) &
1788                                 ~((u64)blocksize - 1);
1789                         /*
1790                          * we've already got the extent locked, but we
1791                          * need to split the state such that our end_bio
1792                          * handler can clear the lock.
1793                          */
1794                         set_extent_bit(tree, block_start,
1795                                        block_start + iosize - 1,
1796                                        EXTENT_LOCKED, 0, NULL, GFP_NOFS);
1797                         ret = submit_extent_page(READ, tree, page,
1798                                          sector, iosize, page_offset, em->bdev,
1799                                          end_bio_extent_preparewrite);
1800                         iocount++;
1801                         block_start = block_start + iosize;
1802                 } else {
1803                         set_extent_uptodate(tree, block_start, cur_end,
1804                                             GFP_NOFS);
1805                         unlock_extent(tree, block_start, cur_end, GFP_NOFS);
1806                         block_start = cur_end + 1;
1807                 }
1808                 page_offset = block_start & (PAGE_CACHE_SIZE - 1);
1809                 free_extent_map(em);
1810         }
1811         if (iocount) {
1812                 wait_extent_bit(tree, orig_block_start,
1813                                 block_end, EXTENT_LOCKED);
1814         }
1815         check_page_uptodate(tree, page);
1816 err:
1817         /* FIXME, zero out newly allocated blocks on error */
1818         return err;
1819 }
1820 EXPORT_SYMBOL(extent_prepare_write);
1821
1822 /*
1823  * a helper for releasepage.  As long as there are no locked extents
1824  * in the range corresponding to the page, both state records and extent
1825  * map records are removed
1826  */
1827 int try_release_extent_mapping(struct extent_map_tree *tree, struct page *page)
1828 {
1829         struct extent_map *em;
1830         u64 start = page->index << PAGE_CACHE_SHIFT;
1831         u64 end = start + PAGE_CACHE_SIZE - 1;
1832         u64 orig_start = start;
1833         int ret = 1;
1834
1835         while (start <= end) {
1836                 em = lookup_extent_mapping(tree, start, end);
1837                 if (!em || IS_ERR(em))
1838                         break;
1839                 if (!test_range_bit(tree, em->start, em->end,
1840                                     EXTENT_LOCKED, 0)) {
1841                         remove_extent_mapping(tree, em);
1842                         /* once for the rb tree */
1843                         free_extent_map(em);
1844                 }
1845                 start = em->end + 1;
1846                 /* once for us */
1847                 free_extent_map(em);
1848         }
1849         if (test_range_bit(tree, orig_start, end, EXTENT_LOCKED, 0))
1850                 ret = 0;
1851         else
1852                 clear_extent_bit(tree, orig_start, end, EXTENT_UPTODATE,
1853                                  1, 1, GFP_NOFS);
1854         return ret;
1855 }
1856 EXPORT_SYMBOL(try_release_extent_mapping);
1857
1858 sector_t extent_bmap(struct address_space *mapping, sector_t iblock,
1859                 get_extent_t *get_extent)
1860 {
1861         struct inode *inode = mapping->host;
1862         u64 start = iblock << inode->i_blkbits;
1863         u64 end = start + (1 << inode->i_blkbits) - 1;
1864         struct extent_map *em;
1865
1866         em = get_extent(inode, NULL, 0, start, end, 0);
1867         if (!em || IS_ERR(em))
1868                 return 0;
1869
1870         if (em->block_start == EXTENT_MAP_INLINE ||
1871             em->block_start == EXTENT_MAP_HOLE)
1872                 return 0;
1873
1874         return (em->block_start + start - em->start) >> inode->i_blkbits;
1875 }
1876
1877 static struct extent_buffer *__alloc_extent_buffer(gfp_t mask)
1878 {
1879         struct extent_buffer *eb = NULL;
1880         spin_lock(&extent_buffers_lock);
1881         if (!list_empty(&extent_buffers)) {
1882                 eb = list_entry(extent_buffers.next, struct extent_buffer,
1883                                 list);
1884                 list_del(&eb->list);
1885                 WARN_ON(nr_extent_buffers == 0);
1886                 nr_extent_buffers--;
1887         }
1888         spin_unlock(&extent_buffers_lock);
1889         if (eb) {
1890                 memset(eb, 0, sizeof(*eb));
1891                 return eb;
1892         }
1893         return kmem_cache_zalloc(extent_buffer_cache, mask);
1894 }
1895
1896 static void __free_extent_buffer(struct extent_buffer *eb)
1897 {
1898         if (nr_extent_buffers >= MAX_EXTENT_BUFFER_CACHE) {
1899                 kmem_cache_free(extent_buffer_cache, eb);
1900         } else {
1901                 spin_lock(&extent_buffers_lock);
1902                 list_add(&eb->list, &extent_buffers);
1903                 nr_extent_buffers++;
1904                 spin_unlock(&extent_buffers_lock);
1905         }
1906 }
1907
1908 static inline struct page *extent_buffer_page(struct extent_buffer *eb, int i)
1909 {
1910         struct page *p;
1911         if (i == 0)
1912                 return eb->first_page;
1913         i += eb->start >> PAGE_CACHE_SHIFT;
1914         p = find_get_page(eb->first_page->mapping, i);
1915         page_cache_release(p);
1916         return p;
1917 }
1918
1919 struct extent_buffer *alloc_extent_buffer(struct extent_map_tree *tree,
1920                                           u64 start, unsigned long len,
1921                                           gfp_t mask)
1922 {
1923         unsigned long num_pages = ((start + len - 1) >> PAGE_CACHE_SHIFT) -
1924                                   (start >> PAGE_CACHE_SHIFT) + 1;
1925         unsigned long i;
1926         unsigned long index = start >> PAGE_CACHE_SHIFT;
1927         struct extent_buffer *eb;
1928         struct page *p;
1929         struct address_space *mapping = tree->mapping;
1930         int uptodate = 0;
1931
1932         eb = __alloc_extent_buffer(mask);
1933         if (!eb || IS_ERR(eb))
1934                 return NULL;
1935
1936         eb->start = start;
1937         eb->len = len;
1938         atomic_set(&eb->refs, 1);
1939
1940         for (i = 0; i < num_pages; i++, index++) {
1941                 p = find_or_create_page(mapping, index, mask | __GFP_HIGHMEM);
1942                 if (!p) {
1943                         /* make sure the free only frees the pages we've
1944                          * grabbed a reference on
1945                          */
1946                         eb->len = i << PAGE_CACHE_SHIFT;
1947                         eb->start &= ~((u64)PAGE_CACHE_SIZE - 1);
1948                         goto fail;
1949                 }
1950                 if (i == 0)
1951                         eb->first_page = p;
1952                 if (!PageUptodate(p))
1953                         uptodate = 0;
1954                 unlock_page(p);
1955         }
1956         if (uptodate)
1957                 eb->flags |= EXTENT_UPTODATE;
1958         return eb;
1959 fail:
1960         free_extent_buffer(eb);
1961         return NULL;
1962 }
1963 EXPORT_SYMBOL(alloc_extent_buffer);
1964
1965 struct extent_buffer *find_extent_buffer(struct extent_map_tree *tree,
1966                                          u64 start, unsigned long len,
1967                                           gfp_t mask)
1968 {
1969         unsigned long num_pages = ((start + len - 1) >> PAGE_CACHE_SHIFT) -
1970                                   (start >> PAGE_CACHE_SHIFT) + 1;
1971         unsigned long i;
1972         unsigned long index = start >> PAGE_CACHE_SHIFT;
1973         struct extent_buffer *eb;
1974         struct page *p;
1975         struct address_space *mapping = tree->mapping;
1976
1977         eb = __alloc_extent_buffer(mask);
1978         if (!eb || IS_ERR(eb))
1979                 return NULL;
1980
1981         eb->start = start;
1982         eb->len = len;
1983         atomic_set(&eb->refs, 1);
1984
1985         for (i = 0; i < num_pages; i++, index++) {
1986                 p = find_get_page(mapping, index);
1987                 if (!p) {
1988                         /* make sure the free only frees the pages we've
1989                          * grabbed a reference on
1990                          */
1991                         eb->len = i << PAGE_CACHE_SHIFT;
1992                         eb->start &= ~((u64)PAGE_CACHE_SIZE - 1);
1993                         goto fail;
1994                 }
1995                 if (i == 0)
1996                         eb->first_page = p;
1997         }
1998         return eb;
1999 fail:
2000         free_extent_buffer(eb);
2001         return NULL;
2002 }
2003 EXPORT_SYMBOL(find_extent_buffer);
2004
2005 void free_extent_buffer(struct extent_buffer *eb)
2006 {
2007         unsigned long i;
2008         unsigned long num_pages;
2009
2010         if (!eb)
2011                 return;
2012
2013         if (!atomic_dec_and_test(&eb->refs))
2014                 return;
2015
2016         num_pages = ((eb->start + eb->len - 1) >> PAGE_CACHE_SHIFT) -
2017                 (eb->start >> PAGE_CACHE_SHIFT) + 1;
2018
2019         if (eb->first_page)
2020                 page_cache_release(eb->first_page);
2021         for (i = 1; i < num_pages; i++) {
2022                 page_cache_release(extent_buffer_page(eb, i));
2023         }
2024         __free_extent_buffer(eb);
2025 }
2026 EXPORT_SYMBOL(free_extent_buffer);
2027
2028 int clear_extent_buffer_dirty(struct extent_map_tree *tree,
2029                               struct extent_buffer *eb)
2030 {
2031         int set;
2032         unsigned long i;
2033         unsigned long num_pages;
2034         struct page *page;
2035
2036         u64 start = eb->start;
2037         u64 end = start + eb->len - 1;
2038
2039         set = clear_extent_dirty(tree, start, end, GFP_NOFS);
2040         num_pages = ((eb->start + eb->len - 1) >> PAGE_CACHE_SHIFT) -
2041                 (eb->start >> PAGE_CACHE_SHIFT) + 1;
2042
2043         for (i = 0; i < num_pages; i++) {
2044                 page = extent_buffer_page(eb, i);
2045                 lock_page(page);
2046                 /*
2047                  * if we're on the last page or the first page and the
2048                  * block isn't aligned on a page boundary, do extra checks
2049                  * to make sure we don't clean page that is partially dirty
2050                  */
2051                 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
2052                     ((i == num_pages - 1) &&
2053                      ((eb->start + eb->len - 1) & (PAGE_CACHE_SIZE - 1)))) {
2054                         start = page->index << PAGE_CACHE_SHIFT;
2055                         end  = start + PAGE_CACHE_SIZE - 1;
2056                         if (test_range_bit(tree, start, end,
2057                                            EXTENT_DIRTY, 0)) {
2058                                 unlock_page(page);
2059                                 continue;
2060                         }
2061                 }
2062                 clear_page_dirty_for_io(page);
2063                 unlock_page(page);
2064         }
2065         return 0;
2066 }
2067 EXPORT_SYMBOL(clear_extent_buffer_dirty);
2068
2069 int wait_on_extent_buffer_writeback(struct extent_map_tree *tree,
2070                                     struct extent_buffer *eb)
2071 {
2072         return wait_on_extent_writeback(tree, eb->start,
2073                                         eb->start + eb->len - 1);
2074 }
2075 EXPORT_SYMBOL(wait_on_extent_buffer_writeback);
2076
2077 int set_extent_buffer_dirty(struct extent_map_tree *tree,
2078                              struct extent_buffer *eb)
2079 {
2080         return set_range_dirty(tree, eb->start, eb->start + eb->len - 1);
2081 }
2082 EXPORT_SYMBOL(set_extent_buffer_dirty);
2083
2084 int set_extent_buffer_uptodate(struct extent_map_tree *tree,
2085                                 struct extent_buffer *eb)
2086 {
2087         unsigned long i;
2088         struct page *page;
2089         unsigned long num_pages;
2090
2091         num_pages = ((eb->start + eb->len - 1) >> PAGE_CACHE_SHIFT) -
2092                 (eb->start >> PAGE_CACHE_SHIFT) + 1;
2093
2094         set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
2095                             GFP_NOFS);
2096         for (i = 0; i < num_pages; i++) {
2097                 page = extent_buffer_page(eb, i);
2098                 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
2099                     ((i == num_pages - 1) &&
2100                      ((eb->start + eb->len - 1) & (PAGE_CACHE_SIZE - 1)))) {
2101                         check_page_uptodate(tree, page);
2102                         continue;
2103                 }
2104                 SetPageUptodate(page);
2105         }
2106         return 0;
2107 }
2108 EXPORT_SYMBOL(set_extent_buffer_uptodate);
2109
2110 int extent_buffer_uptodate(struct extent_map_tree *tree,
2111                              struct extent_buffer *eb)
2112 {
2113         if (eb->flags & EXTENT_UPTODATE)
2114                 return 1;
2115         return test_range_bit(tree, eb->start, eb->start + eb->len - 1,
2116                            EXTENT_UPTODATE, 1);
2117 }
2118 EXPORT_SYMBOL(extent_buffer_uptodate);
2119
2120 int read_extent_buffer_pages(struct extent_map_tree *tree,
2121                              struct extent_buffer *eb, int wait)
2122 {
2123         unsigned long i;
2124         struct page *page;
2125         int err;
2126         int ret = 0;
2127         unsigned long num_pages;
2128
2129         if (eb->flags & EXTENT_UPTODATE)
2130                 return 0;
2131
2132         if (test_range_bit(tree, eb->start, eb->start + eb->len - 1,
2133                            EXTENT_UPTODATE, 1)) {
2134                 return 0;
2135         }
2136
2137         num_pages = ((eb->start + eb->len - 1) >> PAGE_CACHE_SHIFT) -
2138                 (eb->start >> PAGE_CACHE_SHIFT) + 1;
2139         for (i = 0; i < num_pages; i++) {
2140                 page = extent_buffer_page(eb, i);
2141                 if (PageUptodate(page)) {
2142                         continue;
2143                 }
2144                 if (!wait) {
2145                         if (TestSetPageLocked(page)) {
2146                                 continue;
2147                         }
2148                 } else {
2149                         lock_page(page);
2150                 }
2151                 if (!PageUptodate(page)) {
2152                         err = page->mapping->a_ops->readpage(NULL, page);
2153                         if (err) {
2154                                 ret = err;
2155                         }
2156                 } else {
2157                         unlock_page(page);
2158                 }
2159         }
2160
2161         if (ret || !wait) {
2162                 return ret;
2163         }
2164
2165         for (i = 0; i < num_pages; i++) {
2166                 page = extent_buffer_page(eb, i);
2167                 wait_on_page_locked(page);
2168                 if (!PageUptodate(page)) {
2169                         ret = -EIO;
2170                 }
2171         }
2172         eb->flags |= EXTENT_UPTODATE;
2173         return ret;
2174 }
2175 EXPORT_SYMBOL(read_extent_buffer_pages);
2176
2177 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
2178                         unsigned long start,
2179                         unsigned long len)
2180 {
2181         size_t cur;
2182         size_t offset;
2183         struct page *page;
2184         char *kaddr;
2185         char *dst = (char *)dstv;
2186         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
2187         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
2188
2189         WARN_ON(start > eb->len);
2190         WARN_ON(start + len > eb->start + eb->len);
2191
2192         offset = start & ((unsigned long)PAGE_CACHE_SIZE - 1);
2193         if (i == 0)
2194                 offset += start_offset;
2195
2196         while(len > 0) {
2197                 page = extent_buffer_page(eb, i);
2198                 WARN_ON(!PageUptodate(page));
2199
2200                 cur = min(len, (PAGE_CACHE_SIZE - offset));
2201                 kaddr = kmap_atomic(page, KM_USER0);
2202                 memcpy(dst, kaddr + offset, cur);
2203                 kunmap_atomic(kaddr, KM_USER0);
2204
2205                 dst += cur;
2206                 len -= cur;
2207                 offset = 0;
2208                 i++;
2209         }
2210 }
2211 EXPORT_SYMBOL(read_extent_buffer);
2212
2213 int map_extent_buffer(struct extent_buffer *eb, unsigned long start,
2214                       unsigned long min_len,
2215                       char **token, char **map,
2216                       unsigned long *map_start,
2217                       unsigned long *map_len, int km)
2218 {
2219         size_t offset = start & (PAGE_CACHE_SIZE - 1);
2220         char *kaddr;
2221         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
2222         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
2223         unsigned long end_i = (start_offset + start + min_len) >>
2224                                 PAGE_CACHE_SHIFT;
2225
2226         if (i != end_i)
2227                 return -EINVAL;
2228
2229         WARN_ON(start > eb->len);
2230
2231         if (i == 0) {
2232                 offset = start_offset;
2233                 *map_start = 0;
2234         } else {
2235                 *map_start = (i << PAGE_CACHE_SHIFT) - start_offset;
2236         }
2237
2238         kaddr = kmap_atomic(extent_buffer_page(eb, i), km);
2239         *token = kaddr;
2240         *map = kaddr + offset;
2241         *map_len = PAGE_CACHE_SIZE - offset;
2242         return 0;
2243 }
2244 EXPORT_SYMBOL(map_extent_buffer);
2245
2246 void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km)
2247 {
2248         kunmap_atomic(token, km);
2249 }
2250 EXPORT_SYMBOL(unmap_extent_buffer);
2251
2252 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
2253                           unsigned long start,
2254                           unsigned long len)
2255 {
2256         size_t cur;
2257         size_t offset;
2258         struct page *page;
2259         char *kaddr;
2260         char *ptr = (char *)ptrv;
2261         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
2262         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
2263         int ret = 0;
2264
2265         WARN_ON(start > eb->len);
2266         WARN_ON(start + len > eb->start + eb->len);
2267
2268         offset = start & ((unsigned long)PAGE_CACHE_SIZE - 1);
2269         if (i == 0)
2270                 offset += start_offset;
2271
2272         while(len > 0) {
2273                 page = extent_buffer_page(eb, i);
2274                 WARN_ON(!PageUptodate(page));
2275
2276                 cur = min(len, (PAGE_CACHE_SIZE - offset));
2277
2278                 kaddr = kmap_atomic(page, KM_USER0);
2279                 ret = memcmp(ptr, kaddr + offset, cur);
2280                 kunmap_atomic(kaddr, KM_USER0);
2281                 if (ret)
2282                         break;
2283
2284                 ptr += cur;
2285                 len -= cur;
2286                 offset = 0;
2287                 i++;
2288         }
2289         return ret;
2290 }
2291 EXPORT_SYMBOL(memcmp_extent_buffer);
2292
2293 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
2294                          unsigned long start, unsigned long len)
2295 {
2296         size_t cur;
2297         size_t offset;
2298         struct page *page;
2299         char *kaddr;
2300         char *src = (char *)srcv;
2301         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
2302         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
2303
2304         WARN_ON(start > eb->len);
2305         WARN_ON(start + len > eb->start + eb->len);
2306
2307         offset = start & ((unsigned long)PAGE_CACHE_SIZE - 1);
2308         if (i == 0)
2309                 offset += start_offset;
2310
2311         while(len > 0) {
2312                 page = extent_buffer_page(eb, i);
2313                 WARN_ON(!PageUptodate(page));
2314
2315                 cur = min(len, PAGE_CACHE_SIZE - offset);
2316                 kaddr = kmap_atomic(page, KM_USER0);
2317                 memcpy(kaddr + offset, src, cur);
2318                 kunmap_atomic(kaddr, KM_USER0);
2319
2320                 src += cur;
2321                 len -= cur;
2322                 offset = 0;
2323                 i++;
2324         }
2325 }
2326 EXPORT_SYMBOL(write_extent_buffer);
2327
2328 void memset_extent_buffer(struct extent_buffer *eb, char c,
2329                           unsigned long start, unsigned long len)
2330 {
2331         size_t cur;
2332         size_t offset;
2333         struct page *page;
2334         char *kaddr;
2335         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
2336         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
2337
2338         WARN_ON(start > eb->len);
2339         WARN_ON(start + len > eb->start + eb->len);
2340
2341         offset = start & ((unsigned long)PAGE_CACHE_SIZE - 1);
2342         if (i == 0)
2343                 offset += start_offset;
2344
2345         while(len > 0) {
2346                 page = extent_buffer_page(eb, i);
2347                 WARN_ON(!PageUptodate(page));
2348
2349                 cur = min(len, PAGE_CACHE_SIZE - offset);
2350                 kaddr = kmap_atomic(page, KM_USER0);
2351                 memset(kaddr + offset, c, cur);
2352                 kunmap_atomic(kaddr, KM_USER0);
2353
2354                 len -= cur;
2355                 offset = 0;
2356                 i++;
2357         }
2358 }
2359 EXPORT_SYMBOL(memset_extent_buffer);
2360
2361 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
2362                         unsigned long dst_offset, unsigned long src_offset,
2363                         unsigned long len)
2364 {
2365         u64 dst_len = dst->len;
2366         size_t cur;
2367         size_t offset;
2368         struct page *page;
2369         char *kaddr;
2370         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
2371         unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
2372
2373         WARN_ON(src->len != dst_len);
2374
2375         offset = dst_offset & ((unsigned long)PAGE_CACHE_SIZE - 1);
2376         if (i == 0)
2377                 offset += start_offset;
2378
2379         while(len > 0) {
2380                 page = extent_buffer_page(dst, i);
2381                 WARN_ON(!PageUptodate(page));
2382
2383                 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
2384
2385                 kaddr = kmap_atomic(page, KM_USER1);
2386                 read_extent_buffer(src, kaddr + offset, src_offset, cur);
2387                 kunmap_atomic(kaddr, KM_USER1);
2388
2389                 src_offset += cur;
2390                 len -= cur;
2391                 offset = 0;
2392                 i++;
2393         }
2394 }
2395 EXPORT_SYMBOL(copy_extent_buffer);
2396
2397 static void move_pages(struct page *dst_page, struct page *src_page,
2398                        unsigned long dst_off, unsigned long src_off,
2399                        unsigned long len)
2400 {
2401         char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
2402         if (dst_page == src_page) {
2403                 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
2404         } else {
2405                 char *src_kaddr = kmap_atomic(src_page, KM_USER1);
2406                 char *p = dst_kaddr + dst_off + len;
2407                 char *s = src_kaddr + src_off + len;
2408
2409                 while (len--)
2410                         *--p = *--s;
2411
2412                 kunmap_atomic(src_kaddr, KM_USER1);
2413         }
2414         kunmap_atomic(dst_kaddr, KM_USER0);
2415 }
2416
2417 static void copy_pages(struct page *dst_page, struct page *src_page,
2418                        unsigned long dst_off, unsigned long src_off,
2419                        unsigned long len)
2420 {
2421         char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
2422         char *src_kaddr;
2423
2424         if (dst_page != src_page)
2425                 src_kaddr = kmap_atomic(src_page, KM_USER1);
2426         else
2427                 src_kaddr = dst_kaddr;
2428
2429         memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
2430         kunmap_atomic(dst_kaddr, KM_USER0);
2431         if (dst_page != src_page)
2432                 kunmap_atomic(src_kaddr, KM_USER1);
2433 }
2434
2435 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
2436                            unsigned long src_offset, unsigned long len)
2437 {
2438         size_t cur;
2439         size_t dst_off_in_page;
2440         size_t src_off_in_page;
2441         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
2442         unsigned long dst_i;
2443         unsigned long src_i;
2444
2445         if (src_offset + len > dst->len) {
2446                 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
2447                        src_offset, len, dst->len);
2448                 BUG_ON(1);
2449         }
2450         if (dst_offset + len > dst->len) {
2451                 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
2452                        dst_offset, len, dst->len);
2453                 BUG_ON(1);
2454         }
2455
2456         while(len > 0) {
2457                 dst_off_in_page = dst_offset &
2458                         ((unsigned long)PAGE_CACHE_SIZE - 1);
2459                 src_off_in_page = src_offset &
2460                         ((unsigned long)PAGE_CACHE_SIZE - 1);
2461
2462                 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
2463                 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
2464
2465                 if (src_i == 0)
2466                         src_off_in_page += start_offset;
2467                 if (dst_i == 0)
2468                         dst_off_in_page += start_offset;
2469
2470                 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
2471                                                src_off_in_page));
2472                 cur = min(cur, (unsigned long)(PAGE_CACHE_SIZE -
2473                                                dst_off_in_page));
2474
2475                 copy_pages(extent_buffer_page(dst, dst_i),
2476                            extent_buffer_page(dst, src_i),
2477                            dst_off_in_page, src_off_in_page, cur);
2478
2479                 src_offset += cur;
2480                 dst_offset += cur;
2481                 len -= cur;
2482         }
2483 }
2484 EXPORT_SYMBOL(memcpy_extent_buffer);
2485
2486 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
2487                            unsigned long src_offset, unsigned long len)
2488 {
2489         size_t cur;
2490         size_t dst_off_in_page;
2491         size_t src_off_in_page;
2492         unsigned long dst_end = dst_offset + len - 1;
2493         unsigned long src_end = src_offset + len - 1;
2494         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
2495         unsigned long dst_i;
2496         unsigned long src_i;
2497
2498         if (src_offset + len > dst->len) {
2499                 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
2500                        src_offset, len, dst->len);
2501                 BUG_ON(1);
2502         }
2503         if (dst_offset + len > dst->len) {
2504                 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
2505                        dst_offset, len, dst->len);
2506                 BUG_ON(1);
2507         }
2508         if (dst_offset < src_offset) {
2509                 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
2510                 return;
2511         }
2512         while(len > 0) {
2513                 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
2514                 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
2515
2516                 dst_off_in_page = dst_end &
2517                         ((unsigned long)PAGE_CACHE_SIZE - 1);
2518                 src_off_in_page = src_end &
2519                         ((unsigned long)PAGE_CACHE_SIZE - 1);
2520
2521                 if (src_i == 0)
2522                         src_off_in_page += start_offset;
2523                 if (dst_i == 0)
2524                         dst_off_in_page += start_offset;
2525
2526                 cur = min(len, src_off_in_page + 1);
2527                 cur = min(cur, dst_off_in_page + 1);
2528
2529                 move_pages(extent_buffer_page(dst, dst_i),
2530                            extent_buffer_page(dst, src_i),
2531                            dst_off_in_page - cur + 1,
2532                            src_off_in_page - cur + 1, cur);
2533
2534                 dst_end -= cur - 1;
2535                 src_end -= cur - 1;
2536                 len -= cur;
2537         }
2538 }
2539 EXPORT_SYMBOL(memmove_extent_buffer);