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
  23 static LIST_HEAD(extent_buffers);
  24 static LIST_HEAD(buffers);
  25 static LIST_HEAD(states);
  26
  27 static spinlock_t extent_buffers_lock;
  28 static spinlock_t state_lock = SPIN_LOCK_UNLOCKED;
  29 static int nr_extent_buffers;
  30 #define MAX_EXTENT_BUFFER_CACHE 128
  31
  32 struct tree_entry {
  33         u64 start;
  34         u64 end;
  35         int in_tree;
  36         struct rb_node rb_node;
  37 };
  38
  39 void __init extent_map_init(void)
  40 {
  41         extent_map_cache = btrfs_cache_create("extent_map",
  42                                             sizeof(struct extent_map), 0,
  43                                             NULL);
  44         extent_state_cache = btrfs_cache_create("extent_state",
  45                                             sizeof(struct extent_state), 0,
  46                                             NULL);
  47         extent_buffer_cache = btrfs_cache_create("extent_buffers",
  48                                             sizeof(struct extent_buffer), 0,
  49                                             NULL);
  50         spin_lock_init(&extent_buffers_lock);
  51 }
  52
  53 void __exit extent_map_exit(void)
  54 {
  55         struct extent_buffer *eb;
  56         struct extent_state *state;
  57
  58         while (!list_empty(&extent_buffers)) {
  59                 eb = list_entry(extent_buffers.next,
  60                                 struct extent_buffer, list);
  61                 list_del(&eb->list);
  62                 kmem_cache_free(extent_buffer_cache, eb);
  63         }
  64         while (!list_empty(&states)) {
  65                 state = list_entry(states.next, struct extent_state, list);
  66                 printk("state leak: start %Lu end %Lu state %lu in tree %d refs %d\n", state->start, state->end, state->state, state->in_tree, atomic_read(&state->refs));
  67                 list_del(&state->list);
  68                 kmem_cache_free(extent_state_cache, state);
  69
  70         }
  71         while (!list_empty(&buffers)) {
  72                 eb = list_entry(buffers.next,
  73                                 struct extent_buffer, leak_list);
  74                 printk("buffer leak start %Lu len %lu return %lX\n", eb->start, eb->len, eb->alloc_addr);
  75                 list_del(&eb->leak_list);
  76                 kmem_cache_free(extent_buffer_cache, eb);
  77         }
  78
  79
  80         if (extent_map_cache)
  81                 kmem_cache_destroy(extent_map_cache);
  82         if (extent_state_cache)
  83                 kmem_cache_destroy(extent_state_cache);
  84         if (extent_buffer_cache)
  85                 kmem_cache_destroy(extent_buffer_cache);
  86 }
  87
  88 void extent_map_tree_init(struct extent_map_tree *tree,
  89                           struct address_space *mapping, gfp_t mask)
  90 {
  91         tree->map.rb_node = NULL;
  92         tree->state.rb_node = NULL;
  93         tree->ops = NULL;
  94         rwlock_init(&tree->lock);
  95         tree->mapping = mapping;
  96 }
  97 EXPORT_SYMBOL(extent_map_tree_init);
  98
  99 struct extent_map *alloc_extent_map(gfp_t mask)
 100 {
 101         struct extent_map *em;
 102         em = kmem_cache_alloc(extent_map_cache, mask);
 103         if (!em || IS_ERR(em))
 104                 return em;
 105         em->in_tree = 0;
 106         atomic_set(&em->refs, 1);
 107         return em;
 108 }
 109 EXPORT_SYMBOL(alloc_extent_map);
 110
 111 void free_extent_map(struct extent_map *em)
 112 {
 113         if (!em)
 114                 return;
 115         if (atomic_dec_and_test(&em->refs)) {
 116                 WARN_ON(em->in_tree);
 117                 kmem_cache_free(extent_map_cache, em);
 118         }
 119 }
 120 EXPORT_SYMBOL(free_extent_map);
 121
 122
 123 struct extent_state *alloc_extent_state(gfp_t mask)
 124 {
 125         struct extent_state *state;
 126         unsigned long flags;
 127
 128         state = kmem_cache_alloc(extent_state_cache, mask);
 129         if (!state || IS_ERR(state))
 130                 return state;
 131         state->state = 0;
 132         state->in_tree = 0;
 133         state->private = 0;
 134
 135         spin_lock_irqsave(&state_lock, flags);
 136         list_add(&state->list, &states);
 137         spin_unlock_irqrestore(&state_lock, flags);
 138
 139         atomic_set(&state->refs, 1);
 140         init_waitqueue_head(&state->wq);
 141         return state;
 142 }
 143 EXPORT_SYMBOL(alloc_extent_state);
 144
 145 void free_extent_state(struct extent_state *state)
 146 {
 147         unsigned long flags;
 148         if (!state)
 149                 return;
 150         if (atomic_dec_and_test(&state->refs)) {
 151                 WARN_ON(state->in_tree);
 152                 spin_lock_irqsave(&state_lock, flags);
 153                 list_del(&state->list);
 154                 spin_unlock_irqrestore(&state_lock, flags);
 155                 kmem_cache_free(extent_state_cache, state);
 156         }
 157 }
 158 EXPORT_SYMBOL(free_extent_state);
 159
 160 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
 161                                    struct rb_node *node)
 162 {
 163         struct rb_node ** p = &root->rb_node;
 164         struct rb_node * parent = NULL;
 165         struct tree_entry *entry;
 166
 167         while(*p) {
 168                 parent = *p;
 169                 entry = rb_entry(parent, struct tree_entry, rb_node);
 170
 171                 if (offset < entry->start)
 172                         p = &(*p)->rb_left;
 173                 else if (offset > entry->end)
 174                         p = &(*p)->rb_right;
 175                 else
 176                         return parent;
 177         }
 178
 179         entry = rb_entry(node, struct tree_entry, rb_node);
 180         entry->in_tree = 1;
 181         rb_link_node(node, parent, p);
 182         rb_insert_color(node, root);
 183         return NULL;
 184 }
 185
 186 static struct rb_node *__tree_search(struct rb_root *root, u64 offset,
 187                                    struct rb_node **prev_ret)
 188 {
 189         struct rb_node * n = root->rb_node;
 190         struct rb_node *prev = NULL;
 191         struct tree_entry *entry;
 192         struct tree_entry *prev_entry = NULL;
 193
 194         while(n) {
 195                 entry = rb_entry(n, struct tree_entry, rb_node);
 196                 prev = n;
 197                 prev_entry = entry;
 198
 199                 if (offset < entry->start)
 200                         n = n->rb_left;
 201                 else if (offset > entry->end)
 202                         n = n->rb_right;
 203                 else
 204                         return n;
 205         }
 206         if (!prev_ret)
 207                 return NULL;
 208         while(prev && offset > prev_entry->end) {
 209                 prev = rb_next(prev);
 210                 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
 211         }
 212         *prev_ret = prev;
 213         return NULL;
 214 }
 215
 216 static inline struct rb_node *tree_search(struct rb_root *root, u64 offset)
 217 {
 218         struct rb_node *prev;
 219         struct rb_node *ret;
 220         ret = __tree_search(root, offset, &prev);
 221         if (!ret)
 222                 return prev;
 223         return ret;
 224 }
 225
 226 static int tree_delete(struct rb_root *root, u64 offset)
 227 {
 228         struct rb_node *node;
 229         struct tree_entry *entry;
 230
 231         node = __tree_search(root, offset, NULL);
 232         if (!node)
 233                 return -ENOENT;
 234         entry = rb_entry(node, struct tree_entry, rb_node);
 235         entry->in_tree = 0;
 236         rb_erase(node, root);
 237         return 0;
 238 }
 239
 240 /*
 241  * add_extent_mapping tries a simple backward merge with existing
 242  * mappings.  The extent_map struct passed in will be inserted into
 243  * the tree directly (no copies made, just a reference taken).
 244  */
 245 int add_extent_mapping(struct extent_map_tree *tree,
 246                        struct extent_map *em)
 247 {
 248         int ret = 0;
 249         struct extent_map *prev = NULL;
 250         struct rb_node *rb;
 251
 252         write_lock_irq(&tree->lock);
 253         rb = tree_insert(&tree->map, em->end, &em->rb_node);
 254         if (rb) {
 255                 prev = rb_entry(rb, struct extent_map, rb_node);
 256                 printk("found extent map %Lu %Lu on insert of %Lu %Lu\n", prev->start, prev->end, em->start, em->end);
 257                 ret = -EEXIST;
 258                 goto out;
 259         }
 260         atomic_inc(&em->refs);
 261         if (em->start != 0) {
 262                 rb = rb_prev(&em->rb_node);
 263                 if (rb)
 264                         prev = rb_entry(rb, struct extent_map, rb_node);
 265                 if (prev && prev->end + 1 == em->start &&
 266                     ((em->block_start == EXTENT_MAP_HOLE &&
 267                       prev->block_start == EXTENT_MAP_HOLE) ||
 268                              (em->block_start == prev->block_end + 1))) {
 269                         em->start = prev->start;
 270                         em->block_start = prev->block_start;
 271                         rb_erase(&prev->rb_node, &tree->map);
 272                         prev->in_tree = 0;
 273                         free_extent_map(prev);
 274                 }
 275          }
 276 out:
 277         write_unlock_irq(&tree->lock);
 278         return ret;
 279 }
 280 EXPORT_SYMBOL(add_extent_mapping);
 281
 282 /*
 283  * lookup_extent_mapping returns the first extent_map struct in the
 284  * tree that intersects the [start, end] (inclusive) range.  There may
 285  * be additional objects in the tree that intersect, so check the object
 286  * returned carefully to make sure you don't need additional lookups.
 287  */
 288 struct extent_map *lookup_extent_mapping(struct extent_map_tree *tree,
 289                                          u64 start, u64 end)
 290 {
 291         struct extent_map *em;
 292         struct rb_node *rb_node;
 293
 294         read_lock_irq(&tree->lock);
 295         rb_node = tree_search(&tree->map, start);
 296         if (!rb_node) {
 297                 em = NULL;
 298                 goto out;
 299         }
 300         if (IS_ERR(rb_node)) {
 301                 em = ERR_PTR(PTR_ERR(rb_node));
 302                 goto out;
 303         }
 304         em = rb_entry(rb_node, struct extent_map, rb_node);
 305         if (em->end < start || em->start > end) {
 306                 em = NULL;
 307                 goto out;
 308         }
 309         atomic_inc(&em->refs);
 310 out:
 311         read_unlock_irq(&tree->lock);
 312         return em;
 313 }
 314 EXPORT_SYMBOL(lookup_extent_mapping);
 315
 316 /*
 317  * removes an extent_map struct from the tree.  No reference counts are
 318  * dropped, and no checks are done to  see if the range is in use
 319  */
 320 int remove_extent_mapping(struct extent_map_tree *tree, struct extent_map *em)
 321 {
 322         int ret;
 323
 324         write_lock_irq(&tree->lock);
 325         ret = tree_delete(&tree->map, em->end);
 326         write_unlock_irq(&tree->lock);
 327         return ret;
 328 }
 329 EXPORT_SYMBOL(remove_extent_mapping);
 330
 331 /*
 332  * utility function to look for merge candidates inside a given range.
 333  * Any extents with matching state are merged together into a single
 334  * extent in the tree.  Extents with EXTENT_IO in their state field
 335  * are not merged because the end_io handlers need to be able to do
 336  * operations on them without sleeping (or doing allocations/splits).
 337  *
 338  * This should be called with the tree lock held.
 339  */
 340 static int merge_state(struct extent_map_tree *tree,
 341                        struct extent_state *state)
 342 {
 343         struct extent_state *other;
 344         struct rb_node *other_node;
 345
 346         if (state->state & EXTENT_IOBITS)
 347                 return 0;
 348
 349         other_node = rb_prev(&state->rb_node);
 350         if (other_node) {
 351                 other = rb_entry(other_node, struct extent_state, rb_node);
 352                 if (other->end == state->start - 1 &&
 353                     other->state == state->state) {
 354                         state->start = other->start;
 355                         other->in_tree = 0;
 356                         rb_erase(&other->rb_node, &tree->state);
 357                         free_extent_state(other);
 358                 }
 359         }
 360         other_node = rb_next(&state->rb_node);
 361         if (other_node) {
 362                 other = rb_entry(other_node, struct extent_state, rb_node);
 363                 if (other->start == state->end + 1 &&
 364                     other->state == state->state) {
 365                         other->start = state->start;
 366                         state->in_tree = 0;
 367                         rb_erase(&state->rb_node, &tree->state);
 368                         free_extent_state(state);
 369                 }
 370         }
 371         return 0;
 372 }
 373
 374 /*
 375  * insert an extent_state struct into the tree.  'bits' are set on the
 376  * struct before it is inserted.
 377  *
 378  * This may return -EEXIST if the extent is already there, in which case the
 379  * state struct is freed.
 380  *
 381  * The tree lock is not taken internally.  This is a utility function and
 382  * probably isn't what you want to call (see set/clear_extent_bit).
 383  */
 384 static int insert_state(struct extent_map_tree *tree,
 385                         struct extent_state *state, u64 start, u64 end,
 386                         int bits)
 387 {
 388         struct rb_node *node;
 389
 390         if (end < start) {
 391                 printk("end < start %Lu %Lu\n", end, start);
 392                 WARN_ON(1);
 393         }
 394         state->state |= bits;
 395         state->start = start;
 396         state->end = end;
 397         node = tree_insert(&tree->state, end, &state->rb_node);
 398         if (node) {
 399                 struct extent_state *found;
 400                 found = rb_entry(node, struct extent_state, rb_node);
 401                 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, start, end);
 402                 free_extent_state(state);
 403                 return -EEXIST;
 404         }
 405         merge_state(tree, state);
 406         return 0;
 407 }
 408
 409 /*
 410  * split a given extent state struct in two, inserting the preallocated
 411  * struct 'prealloc' as the newly created second half.  'split' indicates an
 412  * offset inside 'orig' where it should be split.
 413  *
 414  * Before calling,
 415  * the tree has 'orig' at [orig->start, orig->end].  After calling, there
 416  * are two extent state structs in the tree:
 417  * prealloc: [orig->start, split - 1]
 418  * orig: [ split, orig->end ]
 419  *
 420  * The tree locks are not taken by this function. They need to be held
 421  * by the caller.
 422  */
 423 static int split_state(struct extent_map_tree *tree, struct extent_state *orig,
 424                        struct extent_state *prealloc, u64 split)
 425 {
 426         struct rb_node *node;
 427         prealloc->start = orig->start;
 428         prealloc->end = split - 1;
 429         prealloc->state = orig->state;
 430         orig->start = split;
 431
 432         node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
 433         if (node) {
 434                 struct extent_state *found;
 435                 found = rb_entry(node, struct extent_state, rb_node);
 436                 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, prealloc->start, prealloc->end);
 437                 free_extent_state(prealloc);
 438                 return -EEXIST;
 439         }
 440         return 0;
 441 }
 442
 443 /*
 444  * utility function to clear some bits in an extent state struct.
 445  * it will optionally wake up any one waiting on this state (wake == 1), or
 446  * forcibly remove the state from the tree (delete == 1).
 447  *
 448  * If no bits are set on the state struct after clearing things, the
 449  * struct is freed and removed from the tree
 450  */
 451 static int clear_state_bit(struct extent_map_tree *tree,
 452                             struct extent_state *state, int bits, int wake,
 453                             int delete)
 454 {
 455         int ret = state->state & bits;
 456         state->state &= ~bits;
 457         if (wake)
 458                 wake_up(&state->wq);
 459         if (delete || state->state == 0) {
 460                 if (state->in_tree) {
 461                         rb_erase(&state->rb_node, &tree->state);
 462                         state->in_tree = 0;
 463                         free_extent_state(state);
 464                 } else {
 465                         WARN_ON(1);
 466                 }
 467         } else {
 468                 merge_state(tree, state);
 469         }
 470         return ret;
 471 }
 472
 473 /*
 474  * clear some bits on a range in the tree.  This may require splitting
 475  * or inserting elements in the tree, so the gfp mask is used to
 476  * indicate which allocations or sleeping are allowed.
 477  *
 478  * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
 479  * the given range from the tree regardless of state (ie for truncate).
 480  *
 481  * the range [start, end] is inclusive.
 482  *
 483  * This takes the tree lock, and returns < 0 on error, > 0 if any of the
 484  * bits were already set, or zero if none of the bits were already set.
 485  */
 486 int clear_extent_bit(struct extent_map_tree *tree, u64 start, u64 end,
 487                      int bits, int wake, int delete, gfp_t mask)
 488 {
 489         struct extent_state *state;
 490         struct extent_state *prealloc = NULL;
 491         struct rb_node *node;
 492         unsigned long flags;
 493         int err;
 494         int set = 0;
 495
 496 again:
 497         if (!prealloc && (mask & __GFP_WAIT)) {
 498                 prealloc = alloc_extent_state(mask);
 499                 if (!prealloc)
 500                         return -ENOMEM;
 501         }
 502
 503         write_lock_irqsave(&tree->lock, flags);
 504         /*
 505          * this search will find the extents that end after
 506          * our range starts
 507          */
 508         node = tree_search(&tree->state, start);
 509         if (!node)
 510                 goto out;
 511         state = rb_entry(node, struct extent_state, rb_node);
 512         if (state->start > end)
 513                 goto out;
 514         WARN_ON(state->end < start);
 515
 516         /*
 517          *     | ---- desired range ---- |
 518          *  | state | or
 519          *  | ------------- state -------------- |
 520          *
 521          * We need to split the extent we found, and may flip
 522          * bits on second half.
 523          *
 524          * If the extent we found extends past our range, we
 525          * just split and search again.  It'll get split again
 526          * the next time though.
 527          *
 528          * If the extent we found is inside our range, we clear
 529          * the desired bit on it.
 530          */
 531
 532         if (state->start < start) {
 533                 err = split_state(tree, state, prealloc, start);
 534                 BUG_ON(err == -EEXIST);
 535                 prealloc = NULL;
 536                 if (err)
 537                         goto out;
 538                 if (state->end <= end) {
 539                         start = state->end + 1;
 540                         set |= clear_state_bit(tree, state, bits,
 541                                         wake, delete);
 542                 } else {
 543                         start = state->start;
 544                 }
 545                 goto search_again;
 546         }
 547         /*
 548          * | ---- desired range ---- |
 549          *                        | state |
 550          * We need to split the extent, and clear the bit
 551          * on the first half
 552          */
 553         if (state->start <= end && state->end > end) {
 554                 err = split_state(tree, state, prealloc, end + 1);
 555                 BUG_ON(err == -EEXIST);
 556
 557                 if (wake)
 558                         wake_up(&state->wq);
 559                 set |= clear_state_bit(tree, prealloc, bits,
 560                                        wake, delete);
 561                 prealloc = NULL;
 562                 goto out;
 563         }
 564
 565         start = state->end + 1;
 566         set |= clear_state_bit(tree, state, bits, wake, delete);
 567         goto search_again;
 568
 569 out:
 570         write_unlock_irqrestore(&tree->lock, flags);
 571         if (prealloc)
 572                 free_extent_state(prealloc);
 573
 574         return set;
 575
 576 search_again:
 577         if (start >= end)
 578                 goto out;
 579         write_unlock_irqrestore(&tree->lock, flags);
 580         if (mask & __GFP_WAIT)
 581                 cond_resched();
 582         goto again;
 583 }
 584 EXPORT_SYMBOL(clear_extent_bit);
 585
 586 static int wait_on_state(struct extent_map_tree *tree,
 587                          struct extent_state *state)
 588 {
 589         DEFINE_WAIT(wait);
 590         prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
 591         read_unlock_irq(&tree->lock);
 592         schedule();
 593         read_lock_irq(&tree->lock);
 594         finish_wait(&state->wq, &wait);
 595         return 0;
 596 }
 597
 598 /*
 599  * waits for one or more bits to clear on a range in the state tree.
 600  * The range [start, end] is inclusive.
 601  * The tree lock is taken by this function
 602  */
 603 int wait_extent_bit(struct extent_map_tree *tree, u64 start, u64 end, int bits)
 604 {
 605         struct extent_state *state;
 606         struct rb_node *node;
 607
 608         read_lock_irq(&tree->lock);
 609 again:
 610         while (1) {
 611                 /*
 612                  * this search will find all the extents that end after
 613                  * our range starts
 614                  */
 615                 node = tree_search(&tree->state, start);
 616                 if (!node)
 617                         break;
 618
 619                 state = rb_entry(node, struct extent_state, rb_node);
 620
 621                 if (state->start > end)
 622                         goto out;
 623
 624                 if (state->state & bits) {
 625                         start = state->start;
 626                         atomic_inc(&state->refs);
 627                         wait_on_state(tree, state);
 628                         free_extent_state(state);
 629                         goto again;
 630                 }
 631                 start = state->end + 1;
 632
 633                 if (start > end)
 634                         break;
 635
 636                 if (need_resched()) {
 637                         read_unlock_irq(&tree->lock);
 638                         cond_resched();
 639                         read_lock_irq(&tree->lock);
 640                 }
 641         }
 642 out:
 643         read_unlock_irq(&tree->lock);
 644         return 0;
 645 }
 646 EXPORT_SYMBOL(wait_extent_bit);
 647
 648 /*
 649  * set some bits on a range in the tree.  This may require allocations
 650  * or sleeping, so the gfp mask is used to indicate what is allowed.
 651  *
 652  * If 'exclusive' == 1, this will fail with -EEXIST if some part of the
 653  * range already has the desired bits set.  The start of the existing
 654  * range is returned in failed_start in this case.
 655  *
 656  * [start, end] is inclusive
 657  * This takes the tree lock.
 658  */
 659 int set_extent_bit(struct extent_map_tree *tree, u64 start, u64 end, int bits,
 660                    int exclusive, u64 *failed_start, gfp_t mask)
 661 {
 662         struct extent_state *state;
 663         struct extent_state *prealloc = NULL;
 664         struct rb_node *node;
 665         unsigned long flags;
 666         int err = 0;
 667         int set;
 668         u64 last_start;
 669         u64 last_end;
 670 again:
 671         if (!prealloc && (mask & __GFP_WAIT)) {
 672                 prealloc = alloc_extent_state(mask);
 673                 if (!prealloc)
 674                         return -ENOMEM;
 675         }
 676
 677         write_lock_irqsave(&tree->lock, flags);
 678         /*
 679          * this search will find all the extents that end after
 680          * our range starts.
 681          */
 682         node = tree_search(&tree->state, start);
 683         if (!node) {
 684                 err = insert_state(tree, prealloc, start, end, bits);
 685                 prealloc = NULL;
 686                 BUG_ON(err == -EEXIST);
 687                 goto out;
 688         }
 689
 690         state = rb_entry(node, struct extent_state, rb_node);
 691         last_start = state->start;
 692         last_end = state->end;
 693
 694         /*
 695          * | ---- desired range ---- |
 696          * | state |
 697          *
 698          * Just lock what we found and keep going
 699          */
 700         if (state->start == start && state->end <= end) {
 701                 set = state->state & bits;
 702                 if (set && exclusive) {
 703                         *failed_start = state->start;
 704                         err = -EEXIST;
 705                         goto out;
 706                 }
 707                 state->state |= bits;
 708                 start = state->end + 1;
 709                 merge_state(tree, state);
 710                 goto search_again;
 711         }
 712
 713         /*
 714          *     | ---- desired range ---- |
 715          * | state |
 716          *   or
 717          * | ------------- state -------------- |
 718          *
 719          * We need to split the extent we found, and may flip bits on
 720          * second half.
 721          *
 722          * If the extent we found extends past our
 723          * range, we just split and search again.  It'll get split
 724          * again the next time though.
 725          *
 726          * If the extent we found is inside our range, we set the
 727          * desired bit on it.
 728          */
 729         if (state->start < start) {
 730                 set = state->state & bits;
 731                 if (exclusive && set) {
 732                         *failed_start = start;
 733                         err = -EEXIST;
 734                         goto out;
 735                 }
 736                 err = split_state(tree, state, prealloc, start);
 737                 BUG_ON(err == -EEXIST);
 738                 prealloc = NULL;
 739                 if (err)
 740                         goto out;
 741                 if (state->end <= end) {
 742                         state->state |= bits;
 743                         start = state->end + 1;
 744                         merge_state(tree, state);
 745                 } else {
 746                         start = state->start;
 747                 }
 748                 goto search_again;
 749         }
 750         /*
 751          * | ---- desired range ---- |
 752          *     | state | or               | state |
 753          *
 754          * There's a hole, we need to insert something in it and
 755          * ignore the extent we found.
 756          */
 757         if (state->start > start) {
 758                 u64 this_end;
 759                 if (end < last_start)
 760                         this_end = end;
 761                 else
 762                         this_end = last_start -1;
 763                 err = insert_state(tree, prealloc, start, this_end,
 764                                    bits);
 765                 prealloc = NULL;
 766                 BUG_ON(err == -EEXIST);
 767                 if (err)
 768                         goto out;
 769                 start = this_end + 1;
 770                 goto search_again;
 771         }
 772         /*
 773          * | ---- desired range ---- |
 774          *                        | state |
 775          * We need to split the extent, and set the bit
 776          * on the first half
 777          */
 778         if (state->start <= end && state->end > end) {
 779                 set = state->state & bits;
 780                 if (exclusive && set) {
 781                         *failed_start = start;
 782                         err = -EEXIST;
 783                         goto out;
 784                 }
 785                 err = split_state(tree, state, prealloc, end + 1);
 786                 BUG_ON(err == -EEXIST);
 787
 788                 prealloc->state |= bits;
 789                 merge_state(tree, prealloc);
 790                 prealloc = NULL;
 791                 goto out;
 792         }
 793
 794         goto search_again;
 795
 796 out:
 797         write_unlock_irqrestore(&tree->lock, flags);
 798         if (prealloc)
 799                 free_extent_state(prealloc);
 800
 801         return err;
 802
 803 search_again:
 804         if (start > end)
 805                 goto out;
 806         write_unlock_irqrestore(&tree->lock, flags);
 807         if (mask & __GFP_WAIT)
 808                 cond_resched();
 809         goto again;
 810 }
 811 EXPORT_SYMBOL(set_extent_bit);
 812
 813 /* wrappers around set/clear extent bit */
 814 int set_extent_dirty(struct extent_map_tree *tree, u64 start, u64 end,
 815                      gfp_t mask)
 816 {
 817         return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL,
 818                               mask);
 819 }
 820 EXPORT_SYMBOL(set_extent_dirty);
 821
 822 int set_extent_delalloc(struct extent_map_tree *tree, u64 start, u64 end,
 823                      gfp_t mask)
 824 {
 825         return set_extent_bit(tree, start, end,
 826                               EXTENT_DELALLOC | EXTENT_DIRTY, 0, NULL,
 827                               mask);
 828 }
 829 EXPORT_SYMBOL(set_extent_delalloc);
 830
 831 int clear_extent_dirty(struct extent_map_tree *tree, u64 start, u64 end,
 832                        gfp_t mask)
 833 {
 834         return clear_extent_bit(tree, start, end,
 835                                 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, mask);
 836 }
 837 EXPORT_SYMBOL(clear_extent_dirty);
 838
 839 int set_extent_new(struct extent_map_tree *tree, u64 start, u64 end,
 840                      gfp_t mask)
 841 {
 842         return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL,
 843                               mask);
 844 }
 845 EXPORT_SYMBOL(set_extent_new);
 846
 847 int clear_extent_new(struct extent_map_tree *tree, u64 start, u64 end,
 848                        gfp_t mask)
 849 {
 850         return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0, mask);
 851 }
 852 EXPORT_SYMBOL(clear_extent_new);
 853
 854 int set_extent_uptodate(struct extent_map_tree *tree, u64 start, u64 end,
 855                         gfp_t mask)
 856 {
 857         return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, NULL,
 858                               mask);
 859 }
 860 EXPORT_SYMBOL(set_extent_uptodate);
 861
 862 int clear_extent_uptodate(struct extent_map_tree *tree, u64 start, u64 end,
 863                           gfp_t mask)
 864 {
 865         return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, mask);
 866 }
 867 EXPORT_SYMBOL(clear_extent_uptodate);
 868
 869 int set_extent_writeback(struct extent_map_tree *tree, u64 start, u64 end,
 870                          gfp_t mask)
 871 {
 872         return set_extent_bit(tree, start, end, EXTENT_WRITEBACK,
 873                               0, NULL, mask);
 874 }
 875 EXPORT_SYMBOL(set_extent_writeback);
 876
 877 int clear_extent_writeback(struct extent_map_tree *tree, u64 start, u64 end,
 878                            gfp_t mask)
 879 {
 880         return clear_extent_bit(tree, start, end, EXTENT_WRITEBACK, 1, 0, mask);
 881 }
 882 EXPORT_SYMBOL(clear_extent_writeback);
 883
 884 int wait_on_extent_writeback(struct extent_map_tree *tree, u64 start, u64 end)
 885 {
 886         return wait_extent_bit(tree, start, end, EXTENT_WRITEBACK);
 887 }
 888 EXPORT_SYMBOL(wait_on_extent_writeback);
 889
 890 /*
 891  * locks a range in ascending order, waiting for any locked regions
 892  * it hits on the way.  [start,end] are inclusive, and this will sleep.
 893  */
 894 int lock_extent(struct extent_map_tree *tree, u64 start, u64 end, gfp_t mask)
 895 {
 896         int err;
 897         u64 failed_start;
 898         while (1) {
 899                 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1,
 900                                      &failed_start, mask);
 901                 if (err == -EEXIST && (mask & __GFP_WAIT)) {
 902                         wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
 903                         start = failed_start;
 904                 } else {
 905                         break;
 906                 }
 907                 WARN_ON(start > end);
 908         }
 909         return err;
 910 }
 911 EXPORT_SYMBOL(lock_extent);
 912
 913 int unlock_extent(struct extent_map_tree *tree, u64 start, u64 end,
 914                   gfp_t mask)
 915 {
 916         return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, mask);
 917 }
 918 EXPORT_SYMBOL(unlock_extent);
 919
 920 /*
 921  * helper function to set pages and extents in the tree dirty
 922  */
 923 int set_range_dirty(struct extent_map_tree *tree, u64 start, u64 end)
 924 {
 925         unsigned long index = start >> PAGE_CACHE_SHIFT;
 926         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
 927         struct page *page;
 928
 929         while (index <= end_index) {
 930                 page = find_get_page(tree->mapping, index);
 931                 BUG_ON(!page);
 932                 __set_page_dirty_nobuffers(page);
 933                 page_cache_release(page);
 934                 index++;
 935         }
 936         set_extent_dirty(tree, start, end, GFP_NOFS);
 937         return 0;
 938 }
 939 EXPORT_SYMBOL(set_range_dirty);
 940
 941 /*
 942  * helper function to set both pages and extents in the tree writeback
 943  */
 944 int set_range_writeback(struct extent_map_tree *tree, u64 start, u64 end)
 945 {
 946         unsigned long index = start >> PAGE_CACHE_SHIFT;
 947         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
 948         struct page *page;
 949
 950         while (index <= end_index) {
 951                 page = find_get_page(tree->mapping, index);
 952                 BUG_ON(!page);
 953                 set_page_writeback(page);
 954                 page_cache_release(page);
 955                 index++;
 956         }
 957         set_extent_writeback(tree, start, end, GFP_NOFS);
 958         return 0;
 959 }
 960 EXPORT_SYMBOL(set_range_writeback);
 961
 962 int find_first_extent_bit(struct extent_map_tree *tree, u64 start,
 963                           u64 *start_ret, u64 *end_ret, int bits)
 964 {
 965         struct rb_node *node;
 966         struct extent_state *state;
 967         int ret = 1;
 968
 969         write_lock_irq(&tree->lock);
 970         /*
 971          * this search will find all the extents that end after
 972          * our range starts.
 973          */
 974         node = tree_search(&tree->state, start);
 975         if (!node || IS_ERR(node)) {
 976                 goto out;
 977         }
 978
 979         while(1) {
 980                 state = rb_entry(node, struct extent_state, rb_node);
 981                 if (state->state & bits) {
 982                         *start_ret = state->start;
 983                         *end_ret = state->end;
 984                         ret = 0;
 985                         break;
 986                 }
 987                 node = rb_next(node);
 988                 if (!node)
 989                         break;
 990         }
 991 out:
 992         write_unlock_irq(&tree->lock);
 993         return ret;
 994 }
 995 EXPORT_SYMBOL(find_first_extent_bit);
 996
 997 u64 find_lock_delalloc_range(struct extent_map_tree *tree,
 998                              u64 start, u64 lock_start, u64 *end, u64 max_bytes)
 999 {
1000         struct rb_node *node;
1001         struct extent_state *state;
1002         u64 cur_start = start;
1003         u64 found = 0;
1004         u64 total_bytes = 0;
1005
1006         write_lock_irq(&tree->lock);
1007         /*
1008          * this search will find all the extents that end after
1009          * our range starts.
1010          */
1011 search_again:
1012         node = tree_search(&tree->state, cur_start);
1013         if (!node || IS_ERR(node)) {
1014                 goto out;
1015         }
1016
1017         while(1) {
1018                 state = rb_entry(node, struct extent_state, rb_node);
1019                 if (state->start != cur_start) {
1020                         goto out;
1021                 }
1022                 if (!(state->state & EXTENT_DELALLOC)) {
1023                         goto out;
1024                 }
1025                 if (state->start >= lock_start) {
1026                         if (state->state & EXTENT_LOCKED) {
1027                                 DEFINE_WAIT(wait);
1028                                 atomic_inc(&state->refs);
1029                                 write_unlock_irq(&tree->lock);
1030                                 schedule();
1031                                 write_lock_irq(&tree->lock);
1032                                 finish_wait(&state->wq, &wait);
1033                                 free_extent_state(state);
1034                                 goto search_again;
1035                         }
1036                         state->state |= EXTENT_LOCKED;
1037                 }
1038                 found++;
1039                 *end = state->end;
1040                 cur_start = state->end + 1;
1041                 node = rb_next(node);
1042                 if (!node)
1043                         break;
1044                 total_bytes = state->end - state->start + 1;
1045                 if (total_bytes >= max_bytes)
1046                         break;
1047         }
1048 out:
1049         write_unlock_irq(&tree->lock);
1050         return found;
1051 }
1052
1053 /*
1054  * helper function to lock both pages and extents in the tree.
1055  * pages must be locked first.
1056  */
1057 int lock_range(struct extent_map_tree *tree, u64 start, u64 end)
1058 {
1059         unsigned long index = start >> PAGE_CACHE_SHIFT;
1060         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1061         struct page *page;
1062         int err;
1063
1064         while (index <= end_index) {
1065                 page = grab_cache_page(tree->mapping, index);
1066                 if (!page) {
1067                         err = -ENOMEM;
1068                         goto failed;
1069                 }
1070                 if (IS_ERR(page)) {
1071                         err = PTR_ERR(page);
1072                         goto failed;
1073                 }
1074                 index++;
1075         }
1076         lock_extent(tree, start, end, GFP_NOFS);
1077         return 0;
1078
1079 failed:
1080         /*
1081          * we failed above in getting the page at 'index', so we undo here
1082          * up to but not including the page at 'index'
1083          */
1084         end_index = index;
1085         index = start >> PAGE_CACHE_SHIFT;
1086         while (index < end_index) {
1087                 page = find_get_page(tree->mapping, index);
1088                 unlock_page(page);
1089                 page_cache_release(page);
1090                 index++;
1091         }
1092         return err;
1093 }
1094 EXPORT_SYMBOL(lock_range);
1095
1096 /*
1097  * helper function to unlock both pages and extents in the tree.
1098  */
1099 int unlock_range(struct extent_map_tree *tree, u64 start, u64 end)
1100 {
1101         unsigned long index = start >> PAGE_CACHE_SHIFT;
1102         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1103         struct page *page;
1104
1105         while (index <= end_index) {
1106                 page = find_get_page(tree->mapping, index);
1107                 unlock_page(page);
1108                 page_cache_release(page);
1109                 index++;
1110         }
1111         unlock_extent(tree, start, end, GFP_NOFS);
1112         return 0;
1113 }
1114 EXPORT_SYMBOL(unlock_range);
1115
1116 int set_state_private(struct extent_map_tree *tree, u64 start, u64 private)
1117 {
1118         struct rb_node *node;
1119         struct extent_state *state;
1120         int ret = 0;
1121
1122         write_lock_irq(&tree->lock);
1123         /*
1124          * this search will find all the extents that end after
1125          * our range starts.
1126          */
1127         node = tree_search(&tree->state, start);
1128         if (!node || IS_ERR(node)) {
1129                 ret = -ENOENT;
1130                 goto out;
1131         }
1132         state = rb_entry(node, struct extent_state, rb_node);
1133         if (state->start != start) {
1134                 ret = -ENOENT;
1135                 goto out;
1136         }
1137         state->private = private;
1138 out:
1139         write_unlock_irq(&tree->lock);
1140         return ret;
1141
1142 }
1143
1144 int get_state_private(struct extent_map_tree *tree, u64 start, u64 *private)
1145 {
1146         struct rb_node *node;
1147         struct extent_state *state;
1148         int ret = 0;
1149
1150         read_lock_irq(&tree->lock);
1151         /*
1152          * this search will find all the extents that end after
1153          * our range starts.
1154          */
1155         node = tree_search(&tree->state, start);
1156         if (!node || IS_ERR(node)) {
1157                 ret = -ENOENT;
1158                 goto out;
1159         }
1160         state = rb_entry(node, struct extent_state, rb_node);
1161         if (state->start != start) {
1162                 ret = -ENOENT;
1163                 goto out;
1164         }
1165         *private = state->private;
1166 out:
1167         read_unlock_irq(&tree->lock);
1168         return ret;
1169 }
1170
1171 /*
1172  * searches a range in the state tree for a given mask.
1173  * If 'filled' == 1, this returns 1 only if ever extent in the tree
1174  * has the bits set.  Otherwise, 1 is returned if any bit in the
1175  * range is found set.
1176  */
1177 static int test_range_bit(struct extent_map_tree *tree, u64 start, u64 end,
1178                           int bits, int filled)
1179 {
1180         struct extent_state *state = NULL;
1181         struct rb_node *node;
1182         int bitset = 0;
1183
1184         read_lock_irq(&tree->lock);
1185         node = tree_search(&tree->state, start);
1186         while (node && start <= end) {
1187                 state = rb_entry(node, struct extent_state, rb_node);
1188                 if (state->start > end)
1189                         break;
1190
1191                 if (filled && state->start > start) {
1192                         bitset = 0;
1193                         break;
1194                 }
1195                 if (state->state & bits) {
1196                         bitset = 1;
1197                         if (!filled)
1198                                 break;
1199                 } else if (filled) {
1200                         bitset = 0;
1201                         break;
1202                 }
1203                 start = state->end + 1;
1204                 if (start > end)
1205                         break;
1206                 node = rb_next(node);
1207         }
1208         read_unlock_irq(&tree->lock);
1209         return bitset;
1210 }
1211
1212 /*
1213  * helper function to set a given page up to date if all the
1214  * extents in the tree for that page are up to date
1215  */
1216 static int check_page_uptodate(struct extent_map_tree *tree,
1217                                struct page *page)
1218 {
1219         u64 start = page->index << PAGE_CACHE_SHIFT;
1220         u64 end = start + PAGE_CACHE_SIZE - 1;
1221         if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1))
1222                 SetPageUptodate(page);
1223         return 0;
1224 }
1225
1226 /*
1227  * helper function to unlock a page if all the extents in the tree
1228  * for that page are unlocked
1229  */
1230 static int check_page_locked(struct extent_map_tree *tree,
1231                              struct page *page)
1232 {
1233         u64 start = page->index << PAGE_CACHE_SHIFT;
1234         u64 end = start + PAGE_CACHE_SIZE - 1;
1235         if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0))
1236                 unlock_page(page);
1237         return 0;
1238 }
1239
1240 /*
1241  * helper function to end page writeback if all the extents
1242  * in the tree for that page are done with writeback
1243  */
1244 static int check_page_writeback(struct extent_map_tree *tree,
1245                              struct page *page)
1246 {
1247         u64 start = page->index << PAGE_CACHE_SHIFT;
1248         u64 end = start + PAGE_CACHE_SIZE - 1;
1249         if (!test_range_bit(tree, start, end, EXTENT_WRITEBACK, 0))
1250                 end_page_writeback(page);
1251         return 0;
1252 }
1253
1254 /* lots and lots of room for performance fixes in the end_bio funcs */
1255
1256 /*
1257  * after a writepage IO is done, we need to:
1258  * clear the uptodate bits on error
1259  * clear the writeback bits in the extent tree for this IO
1260  * end_page_writeback if the page has no more pending IO
1261  *
1262  * Scheduling is not allowed, so the extent state tree is expected
1263  * to have one and only one object corresponding to this IO.
1264  */
1265 static int end_bio_extent_writepage(struct bio *bio,
1266                                    unsigned int bytes_done, int err)
1267 {
1268         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1269         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1270         struct extent_map_tree *tree = bio->bi_private;
1271         u64 start;
1272         u64 end;
1273         int whole_page;
1274
1275         if (bio->bi_size)
1276                 return 1;
1277
1278         do {
1279                 struct page *page = bvec->bv_page;
1280                 start = (page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
1281                 end = start + bvec->bv_len - 1;
1282
1283                 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1284                         whole_page = 1;
1285                 else
1286                         whole_page = 0;
1287
1288                 if (--bvec >= bio->bi_io_vec)
1289                         prefetchw(&bvec->bv_page->flags);
1290
1291                 if (!uptodate) {
1292                         clear_extent_uptodate(tree, start, end, GFP_ATOMIC);
1293                         ClearPageUptodate(page);
1294                         SetPageError(page);
1295                 }
1296                 clear_extent_writeback(tree, start, end, GFP_ATOMIC);
1297
1298                 if (whole_page)
1299                         end_page_writeback(page);
1300                 else
1301                         check_page_writeback(tree, page);
1302                 if (tree->ops && tree->ops->writepage_end_io_hook)
1303                         tree->ops->writepage_end_io_hook(page, start, end);
1304         } while (bvec >= bio->bi_io_vec);
1305
1306         bio_put(bio);
1307         return 0;
1308 }
1309
1310 /*
1311  * after a readpage IO is done, we need to:
1312  * clear the uptodate bits on error
1313  * set the uptodate bits if things worked
1314  * set the page up to date if all extents in the tree are uptodate
1315  * clear the lock bit in the extent tree
1316  * unlock the page if there are no other extents locked for it
1317  *
1318  * Scheduling is not allowed, so the extent state tree is expected
1319  * to have one and only one object corresponding to this IO.
1320  */
1321 static int end_bio_extent_readpage(struct bio *bio,
1322                                    unsigned int bytes_done, int err)
1323 {
1324         int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1325         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1326         struct extent_map_tree *tree = bio->bi_private;
1327         u64 start;
1328         u64 end;
1329         int whole_page;
1330         int ret;
1331
1332         if (bio->bi_size)
1333                 return 1;
1334
1335         do {
1336                 struct page *page = bvec->bv_page;
1337                 start = (page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
1338                 end = start + bvec->bv_len - 1;
1339
1340                 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1341                         whole_page = 1;
1342                 else
1343                         whole_page = 0;
1344
1345                 if (--bvec >= bio->bi_io_vec)
1346                         prefetchw(&bvec->bv_page->flags);
1347
1348                 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
1349                         ret = tree->ops->readpage_end_io_hook(page, start, end);
1350                         if (ret)
1351                                 uptodate = 0;
1352                 }
1353                 if (uptodate) {
1354                         set_extent_uptodate(tree, start, end, GFP_ATOMIC);
1355                         if (whole_page)
1356                                 SetPageUptodate(page);
1357                         else
1358                                 check_page_uptodate(tree, page);
1359                 } else {
1360                         ClearPageUptodate(page);
1361                         SetPageError(page);
1362                 }
1363
1364                 unlock_extent(tree, start, end, GFP_ATOMIC);
1365
1366                 if (whole_page)
1367                         unlock_page(page);
1368                 else
1369                         check_page_locked(tree, page);
1370         } while (bvec >= bio->bi_io_vec);
1371
1372         bio_put(bio);
1373         return 0;
1374 }
1375
1376 /*
1377  * IO done from prepare_write is pretty simple, we just unlock
1378  * the structs in the extent tree when done, and set the uptodate bits
1379  * as appropriate.
1380  */
1381 static int end_bio_extent_preparewrite(struct bio *bio,
1382                                        unsigned int bytes_done, int err)
1383 {
1384         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1385         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1386         struct extent_map_tree *tree = bio->bi_private;
1387         u64 start;
1388         u64 end;
1389
1390         if (bio->bi_size)
1391                 return 1;
1392
1393         do {
1394                 struct page *page = bvec->bv_page;
1395                 start = (page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
1396                 end = start + bvec->bv_len - 1;
1397
1398                 if (--bvec >= bio->bi_io_vec)
1399                         prefetchw(&bvec->bv_page->flags);
1400
1401                 if (uptodate) {
1402                         set_extent_uptodate(tree, start, end, GFP_ATOMIC);
1403                 } else {
1404                         ClearPageUptodate(page);
1405                         SetPageError(page);
1406                 }
1407
1408                 unlock_extent(tree, start, end, GFP_ATOMIC);
1409
1410         } while (bvec >= bio->bi_io_vec);
1411
1412         bio_put(bio);
1413         return 0;
1414 }
1415
1416 static int submit_extent_page(int rw, struct extent_map_tree *tree,
1417                               struct page *page, sector_t sector,
1418                               size_t size, unsigned long offset,
1419                               struct block_device *bdev,
1420                               bio_end_io_t end_io_func)
1421 {
1422         struct bio *bio;
1423         int ret = 0;
1424
1425         bio = bio_alloc(GFP_NOIO, 1);
1426
1427         bio->bi_sector = sector;
1428         bio->bi_bdev = bdev;
1429         bio->bi_io_vec[0].bv_page = page;
1430         bio->bi_io_vec[0].bv_len = size;
1431         bio->bi_io_vec[0].bv_offset = offset;
1432
1433         bio->bi_vcnt = 1;
1434         bio->bi_idx = 0;
1435         bio->bi_size = size;
1436
1437         bio->bi_end_io = end_io_func;
1438         bio->bi_private = tree;
1439
1440         bio_get(bio);
1441         submit_bio(rw, bio);
1442
1443         if (bio_flagged(bio, BIO_EOPNOTSUPP))
1444                 ret = -EOPNOTSUPP;
1445
1446         bio_put(bio);
1447         return ret;
1448 }
1449
1450 void set_page_extent_mapped(struct page *page)
1451 {
1452         if (!PagePrivate(page)) {
1453                 SetPagePrivate(page);
1454                 WARN_ON(!page->mapping->a_ops->invalidatepage);
1455                 set_page_private(page, 1);
1456                 page_cache_get(page);
1457         }
1458 }
1459
1460 /*
1461  * basic readpage implementation.  Locked extent state structs are inserted
1462  * into the tree that are removed when the IO is done (by the end_io
1463  * handlers)
1464  */
1465 int extent_read_full_page(struct extent_map_tree *tree, struct page *page,
1466                           get_extent_t *get_extent)
1467 {
1468         struct inode *inode = page->mapping->host;
1469         u64 start = page->index << PAGE_CACHE_SHIFT;
1470         u64 page_end = start + PAGE_CACHE_SIZE - 1;
1471         u64 end;
1472         u64 cur = start;
1473         u64 extent_offset;
1474         u64 last_byte = i_size_read(inode);
1475         u64 block_start;
1476         u64 cur_end;
1477         sector_t sector;
1478         struct extent_map *em;
1479         struct block_device *bdev;
1480         int ret;
1481         int nr = 0;
1482         size_t page_offset = 0;
1483         size_t iosize;
1484         size_t blocksize = inode->i_sb->s_blocksize;
1485
1486         set_page_extent_mapped(page);
1487
1488         end = page_end;
1489         lock_extent(tree, start, end, GFP_NOFS);
1490
1491         while (cur <= end) {
1492                 if (cur >= last_byte) {
1493                         iosize = PAGE_CACHE_SIZE - page_offset;
1494                         zero_user_page(page, page_offset, iosize, KM_USER0);
1495                         set_extent_uptodate(tree, cur, cur + iosize - 1,
1496                                             GFP_NOFS);
1497                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1498                         break;
1499                 }
1500                 em = get_extent(inode, page, page_offset, cur, end, 0);
1501                 if (IS_ERR(em) || !em) {
1502                         SetPageError(page);
1503                         unlock_extent(tree, cur, end, GFP_NOFS);
1504                         break;
1505                 }
1506
1507                 extent_offset = cur - em->start;
1508                 BUG_ON(em->end < cur);
1509                 BUG_ON(end < cur);
1510
1511                 iosize = min(em->end - cur, end - cur) + 1;
1512                 cur_end = min(em->end, end);
1513                 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
1514                 sector = (em->block_start + extent_offset) >> 9;
1515                 bdev = em->bdev;
1516                 block_start = em->block_start;
1517                 free_extent_map(em);
1518                 em = NULL;
1519
1520                 /* we've found a hole, just zero and go on */
1521                 if (block_start == EXTENT_MAP_HOLE) {
1522                         zero_user_page(page, page_offset, iosize, KM_USER0);
1523                         set_extent_uptodate(tree, cur, cur + iosize - 1,
1524                                             GFP_NOFS);
1525                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1526                         cur = cur + iosize;
1527                         page_offset += iosize;
1528                         continue;
1529                 }
1530                 /* the get_extent function already copied into the page */
1531                 if (test_range_bit(tree, cur, cur_end, EXTENT_UPTODATE, 1)) {
1532                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1533                         cur = cur + iosize;
1534                         page_offset += iosize;
1535                         continue;
1536                 }
1537
1538                 ret = 0;
1539                 if (tree->ops && tree->ops->readpage_io_hook) {
1540                         ret = tree->ops->readpage_io_hook(page, cur,
1541                                                           cur + iosize - 1);
1542                 }
1543                 if (!ret) {
1544                         ret = submit_extent_page(READ, tree, page,
1545                                                  sector, iosize, page_offset,
1546                                                  bdev, end_bio_extent_readpage);
1547                 }
1548                 if (ret)
1549                         SetPageError(page);
1550                 cur = cur + iosize;
1551                 page_offset += iosize;
1552                 nr++;
1553         }
1554         if (!nr) {
1555                 if (!PageError(page))
1556                         SetPageUptodate(page);
1557                 unlock_page(page);
1558         }
1559         return 0;
1560 }
1561 EXPORT_SYMBOL(extent_read_full_page);
1562
1563 /*
1564  * the writepage semantics are similar to regular writepage.  extent
1565  * records are inserted to lock ranges in the tree, and as dirty areas
1566  * are found, they are marked writeback.  Then the lock bits are removed
1567  * and the end_io handler clears the writeback ranges
1568  */
1569 int extent_write_full_page(struct extent_map_tree *tree, struct page *page,
1570                           get_extent_t *get_extent,
1571                           struct writeback_control *wbc)
1572 {
1573         struct inode *inode = page->mapping->host;
1574         u64 start = page->index << PAGE_CACHE_SHIFT;
1575         u64 page_end = start + PAGE_CACHE_SIZE - 1;
1576         u64 end;
1577         u64 cur = start;
1578         u64 extent_offset;
1579         u64 last_byte = i_size_read(inode);
1580         u64 block_start;
1581         sector_t sector;
1582         struct extent_map *em;
1583         struct block_device *bdev;
1584         int ret;
1585         int nr = 0;
1586         size_t page_offset = 0;
1587         size_t iosize;
1588         size_t blocksize;
1589         loff_t i_size = i_size_read(inode);
1590         unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
1591         u64 nr_delalloc;
1592         u64 delalloc_end;
1593
1594         WARN_ON(!PageLocked(page));
1595         if (page->index > end_index) {
1596                 clear_extent_dirty(tree, start, page_end, GFP_NOFS);
1597                 unlock_page(page);
1598                 return 0;
1599         }
1600
1601         if (page->index == end_index) {
1602                 size_t offset = i_size & (PAGE_CACHE_SIZE - 1);
1603                 zero_user_page(page, offset,
1604                                PAGE_CACHE_SIZE - offset, KM_USER0);
1605         }
1606
1607         set_page_extent_mapped(page);
1608
1609         lock_extent(tree, start, page_end, GFP_NOFS);
1610         nr_delalloc = find_lock_delalloc_range(tree, start, page_end + 1,
1611                                                &delalloc_end,
1612                                                128 * 1024 * 1024);
1613         if (nr_delalloc) {
1614                 tree->ops->fill_delalloc(inode, start, delalloc_end);
1615                 if (delalloc_end >= page_end + 1) {
1616                         clear_extent_bit(tree, page_end + 1, delalloc_end,
1617                                          EXTENT_LOCKED | EXTENT_DELALLOC,
1618                                          1, 0, GFP_NOFS);
1619                 }
1620                 clear_extent_bit(tree, start, page_end, EXTENT_DELALLOC,
1621                                  0, 0, GFP_NOFS);
1622                 if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0)) {
1623                         printk("found delalloc bits after clear extent_bit\n");
1624                 }
1625         } else if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0)) {
1626                 printk("found delalloc bits after find_delalloc_range returns 0\n");
1627         }
1628
1629         end = page_end;
1630         if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0)) {
1631                 printk("found delalloc bits after lock_extent\n");
1632         }
1633
1634         if (last_byte <= start) {
1635                 clear_extent_dirty(tree, start, page_end, GFP_NOFS);
1636                 goto done;
1637         }
1638
1639         set_extent_uptodate(tree, start, page_end, GFP_NOFS);
1640         blocksize = inode->i_sb->s_blocksize;
1641
1642         while (cur <= end) {
1643                 if (cur >= last_byte) {
1644                         clear_extent_dirty(tree, cur, page_end, GFP_NOFS);
1645                         break;
1646                 }
1647                 em = get_extent(inode, page, page_offset, cur, end, 0);
1648                 if (IS_ERR(em) || !em) {
1649                         SetPageError(page);
1650                         break;
1651                 }
1652
1653                 extent_offset = cur - em->start;
1654                 BUG_ON(em->end < cur);
1655                 BUG_ON(end < cur);
1656                 iosize = min(em->end - cur, end - cur) + 1;
1657                 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
1658                 sector = (em->block_start + extent_offset) >> 9;
1659                 bdev = em->bdev;
1660                 block_start = em->block_start;
1661                 free_extent_map(em);
1662                 em = NULL;
1663
1664                 if (block_start == EXTENT_MAP_HOLE ||
1665                     block_start == EXTENT_MAP_INLINE) {
1666                         clear_extent_dirty(tree, cur,
1667                                            cur + iosize - 1, GFP_NOFS);
1668                         cur = cur + iosize;
1669                         page_offset += iosize;
1670                         continue;
1671                 }
1672
1673                 /* leave this out until we have a page_mkwrite call */
1674                 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
1675                                    EXTENT_DIRTY, 0)) {
1676                         cur = cur + iosize;
1677                         page_offset += iosize;
1678                         continue;
1679                 }
1680                 clear_extent_dirty(tree, cur, cur + iosize - 1, GFP_NOFS);
1681                 if (tree->ops && tree->ops->writepage_io_hook) {
1682                         ret = tree->ops->writepage_io_hook(page, cur,
1683                                                 cur + iosize - 1);
1684                 } else {
1685                         ret = 0;
1686                 }
1687                 if (ret)
1688                         SetPageError(page);
1689                 else {
1690                         set_range_writeback(tree, cur, cur + iosize - 1);
1691                         ret = submit_extent_page(WRITE, tree, page, sector,
1692                                                  iosize, page_offset, bdev,
1693                                                  end_bio_extent_writepage);
1694                         if (ret)
1695                                 SetPageError(page);
1696                 }
1697                 cur = cur + iosize;
1698                 page_offset += iosize;
1699                 nr++;
1700         }
1701 done:
1702         unlock_extent(tree, start, page_end, GFP_NOFS);
1703         unlock_page(page);
1704         return 0;
1705 }
1706 EXPORT_SYMBOL(extent_write_full_page);
1707
1708 /*
1709  * basic invalidatepage code, this waits on any locked or writeback
1710  * ranges corresponding to the page, and then deletes any extent state
1711  * records from the tree
1712  */
1713 int extent_invalidatepage(struct extent_map_tree *tree,
1714                           struct page *page, unsigned long offset)
1715 {
1716         u64 start = (page->index << PAGE_CACHE_SHIFT);
1717         u64 end = start + PAGE_CACHE_SIZE - 1;
1718         size_t blocksize = page->mapping->host->i_sb->s_blocksize;
1719
1720         start += (offset + blocksize -1) & ~(blocksize - 1);
1721         if (start > end)
1722                 return 0;
1723
1724         lock_extent(tree, start, end, GFP_NOFS);
1725         wait_on_extent_writeback(tree, start, end);
1726         clear_extent_bit(tree, start, end,
1727                          EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC,
1728                          1, 1, GFP_NOFS);
1729         return 0;
1730 }
1731 EXPORT_SYMBOL(extent_invalidatepage);
1732
1733 /*
1734  * simple commit_write call, set_range_dirty is used to mark both
1735  * the pages and the extent records as dirty
1736  */
1737 int extent_commit_write(struct extent_map_tree *tree,
1738                         struct inode *inode, struct page *page,
1739                         unsigned from, unsigned to)
1740 {
1741         loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1742
1743         set_page_extent_mapped(page);
1744         set_page_dirty(page);
1745
1746         if (pos > inode->i_size) {
1747                 i_size_write(inode, pos);
1748                 mark_inode_dirty(inode);
1749         }
1750         return 0;
1751 }
1752 EXPORT_SYMBOL(extent_commit_write);
1753
1754 int extent_prepare_write(struct extent_map_tree *tree,
1755                          struct inode *inode, struct page *page,
1756                          unsigned from, unsigned to, get_extent_t *get_extent)
1757 {
1758         u64 page_start = page->index << PAGE_CACHE_SHIFT;
1759         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
1760         u64 block_start;
1761         u64 orig_block_start;
1762         u64 block_end;
1763         u64 cur_end;
1764         struct extent_map *em;
1765         unsigned blocksize = 1 << inode->i_blkbits;
1766         size_t page_offset = 0;
1767         size_t block_off_start;
1768         size_t block_off_end;
1769         int err = 0;
1770         int iocount = 0;
1771         int ret = 0;
1772         int isnew;
1773
1774         set_page_extent_mapped(page);
1775
1776         block_start = (page_start + from) & ~((u64)blocksize - 1);
1777         block_end = (page_start + to - 1) | (blocksize - 1);
1778         orig_block_start = block_start;
1779
1780         lock_extent(tree, page_start, page_end, GFP_NOFS);
1781         while(block_start <= block_end) {
1782                 em = get_extent(inode, page, page_offset, block_start,
1783                                 block_end, 1);
1784                 if (IS_ERR(em) || !em) {
1785                         goto err;
1786                 }
1787                 cur_end = min(block_end, em->end);
1788                 block_off_start = block_start & (PAGE_CACHE_SIZE - 1);
1789                 block_off_end = block_off_start + blocksize;
1790                 isnew = clear_extent_new(tree, block_start, cur_end, GFP_NOFS);
1791
1792                 if (!PageUptodate(page) && isnew &&
1793                     (block_off_end > to || block_off_start < from)) {
1794                         void *kaddr;
1795
1796                         kaddr = kmap_atomic(page, KM_USER0);
1797                         if (block_off_end > to)
1798                                 memset(kaddr + to, 0, block_off_end - to);
1799                         if (block_off_start < from)
1800                                 memset(kaddr + block_off_start, 0,
1801                                        from - block_off_start);
1802                         flush_dcache_page(page);
1803                         kunmap_atomic(kaddr, KM_USER0);
1804                 }
1805                 if (!isnew && !PageUptodate(page) &&
1806                     (block_off_end > to || block_off_start < from) &&
1807                     !test_range_bit(tree, block_start, cur_end,
1808                                     EXTENT_UPTODATE, 1)) {
1809                         u64 sector;
1810                         u64 extent_offset = block_start - em->start;
1811                         size_t iosize;
1812                         sector = (em->block_start + extent_offset) >> 9;
1813                         iosize = (cur_end - block_start + blocksize - 1) &
1814                                 ~((u64)blocksize - 1);
1815                         /*
1816                          * we've already got the extent locked, but we
1817                          * need to split the state such that our end_bio
1818                          * handler can clear the lock.
1819                          */
1820                         set_extent_bit(tree, block_start,
1821                                        block_start + iosize - 1,
1822                                        EXTENT_LOCKED, 0, NULL, GFP_NOFS);
1823                         ret = submit_extent_page(READ, tree, page,
1824                                          sector, iosize, page_offset, em->bdev,
1825                                          end_bio_extent_preparewrite);
1826                         iocount++;
1827                         block_start = block_start + iosize;
1828                 } else {
1829                         set_extent_uptodate(tree, block_start, cur_end,
1830                                             GFP_NOFS);
1831                         unlock_extent(tree, block_start, cur_end, GFP_NOFS);
1832                         block_start = cur_end + 1;
1833                 }
1834                 page_offset = block_start & (PAGE_CACHE_SIZE - 1);
1835                 free_extent_map(em);
1836         }
1837         if (iocount) {
1838                 wait_extent_bit(tree, orig_block_start,
1839                                 block_end, EXTENT_LOCKED);
1840         }
1841         check_page_uptodate(tree, page);
1842 err:
1843         /* FIXME, zero out newly allocated blocks on error */
1844         return err;
1845 }
1846 EXPORT_SYMBOL(extent_prepare_write);
1847
1848 /*
1849  * a helper for releasepage.  As long as there are no locked extents
1850  * in the range corresponding to the page, both state records and extent
1851  * map records are removed
1852  */
1853 int try_release_extent_mapping(struct extent_map_tree *tree, struct page *page)
1854 {
1855         struct extent_map *em;
1856         u64 start = page->index << PAGE_CACHE_SHIFT;
1857         u64 end = start + PAGE_CACHE_SIZE - 1;
1858         u64 orig_start = start;
1859         int ret = 1;
1860
1861         while (start <= end) {
1862                 em = lookup_extent_mapping(tree, start, end);
1863                 if (!em || IS_ERR(em))
1864                         break;
1865                 if (!test_range_bit(tree, em->start, em->end,
1866                                     EXTENT_LOCKED, 0)) {
1867                         remove_extent_mapping(tree, em);
1868                         /* once for the rb tree */
1869                         free_extent_map(em);
1870                 }
1871                 start = em->end + 1;
1872                 /* once for us */
1873                 free_extent_map(em);
1874         }
1875         if (test_range_bit(tree, orig_start, end, EXTENT_LOCKED, 0))
1876                 ret = 0;
1877         else
1878                 clear_extent_bit(tree, orig_start, end, EXTENT_UPTODATE,
1879                                  1, 1, GFP_NOFS);
1880         return ret;
1881 }
1882 EXPORT_SYMBOL(try_release_extent_mapping);
1883
1884 sector_t extent_bmap(struct address_space *mapping, sector_t iblock,
1885                 get_extent_t *get_extent)
1886 {
1887         struct inode *inode = mapping->host;
1888         u64 start = iblock << inode->i_blkbits;
1889         u64 end = start + (1 << inode->i_blkbits) - 1;
1890         struct extent_map *em;
1891
1892         em = get_extent(inode, NULL, 0, start, end, 0);
1893         if (!em || IS_ERR(em))
1894                 return 0;
1895
1896         if (em->block_start == EXTENT_MAP_INLINE ||
1897             em->block_start == EXTENT_MAP_HOLE)
1898                 return 0;
1899
1900         return (em->block_start + start - em->start) >> inode->i_blkbits;
1901 }
1902
1903 static struct extent_buffer *__alloc_extent_buffer(gfp_t mask)
1904 {
1905         struct extent_buffer *eb = NULL;
1906
1907         spin_lock(&extent_buffers_lock);
1908         if (!list_empty(&extent_buffers)) {
1909                 eb = list_entry(extent_buffers.next, struct extent_buffer,
1910                                 list);
1911                 list_del(&eb->list);
1912                 WARN_ON(nr_extent_buffers == 0);
1913                 nr_extent_buffers--;
1914         }
1915         spin_unlock(&extent_buffers_lock);
1916
1917         if (eb) {
1918                 memset(eb, 0, sizeof(*eb));
1919         } else {
1920                 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
1921         }
1922         spin_lock(&extent_buffers_lock);
1923         list_add(&eb->leak_list, &buffers);
1924         spin_unlock(&extent_buffers_lock);
1925
1926         return eb;
1927 }
1928
1929 static void __free_extent_buffer(struct extent_buffer *eb)
1930 {
1931
1932         spin_lock(&extent_buffers_lock);
1933         list_del_init(&eb->leak_list);
1934         spin_unlock(&extent_buffers_lock);
1935
1936         if (nr_extent_buffers >= MAX_EXTENT_BUFFER_CACHE) {
1937                 kmem_cache_free(extent_buffer_cache, eb);
1938         } else {
1939                 spin_lock(&extent_buffers_lock);
1940                 list_add(&eb->list, &extent_buffers);
1941                 nr_extent_buffers++;
1942                 spin_unlock(&extent_buffers_lock);
1943         }
1944 }
1945
1946 static inline struct page *extent_buffer_page(struct extent_buffer *eb, int i)
1947 {
1948         struct page *p;
1949         if (i == 0)
1950                 return eb->first_page;
1951         i += eb->start >> PAGE_CACHE_SHIFT;
1952         p = find_get_page(eb->first_page->mapping, i);
1953         page_cache_release(p);
1954         return p;
1955 }
1956
1957 struct extent_buffer *alloc_extent_buffer(struct extent_map_tree *tree,
1958                                           u64 start, unsigned long len,
1959                                           gfp_t mask)
1960 {
1961         unsigned long num_pages = ((start + len - 1) >> PAGE_CACHE_SHIFT) -
1962                                   (start >> PAGE_CACHE_SHIFT) + 1;
1963         unsigned long i;
1964         unsigned long index = start >> PAGE_CACHE_SHIFT;
1965         struct extent_buffer *eb;
1966         struct page *p;
1967         struct address_space *mapping = tree->mapping;
1968         int uptodate = 0;
1969
1970         eb = __alloc_extent_buffer(mask);
1971         if (!eb || IS_ERR(eb))
1972                 return NULL;
1973
1974         eb->alloc_addr = __builtin_return_address(0);
1975         eb->start = start;
1976         eb->len = len;
1977         atomic_set(&eb->refs, 1);
1978
1979         for (i = 0; i < num_pages; i++, index++) {
1980                 p = find_or_create_page(mapping, index, mask | __GFP_HIGHMEM);
1981                 if (!p) {
1982                         /* make sure the free only frees the pages we've
1983                          * grabbed a reference on
1984                          */
1985                         eb->len = i << PAGE_CACHE_SHIFT;
1986                         eb->start &= ~((u64)PAGE_CACHE_SIZE - 1);
1987                         goto fail;
1988                 }
1989                 set_page_extent_mapped(p);
1990                 if (i == 0)
1991                         eb->first_page = p;
1992                 if (!PageUptodate(p))
1993                         uptodate = 0;
1994                 unlock_page(p);
1995         }
1996         if (uptodate)
1997                 eb->flags |= EXTENT_UPTODATE;
1998         return eb;
1999 fail:
2000         free_extent_buffer(eb);
2001         return NULL;
2002 }
2003 EXPORT_SYMBOL(alloc_extent_buffer);
2004
2005 struct extent_buffer *find_extent_buffer(struct extent_map_tree *tree,
2006                                          u64 start, unsigned long len,
2007                                           gfp_t mask)
2008 {
2009         unsigned long num_pages = ((start + len - 1) >> PAGE_CACHE_SHIFT) -
2010                                   (start >> PAGE_CACHE_SHIFT) + 1;
2011         unsigned long i;
2012         unsigned long index = start >> PAGE_CACHE_SHIFT;
2013         struct extent_buffer *eb;
2014         struct page *p;
2015         struct address_space *mapping = tree->mapping;
2016
2017         eb = __alloc_extent_buffer(mask);
2018         if (!eb || IS_ERR(eb))
2019                 return NULL;
2020
2021         eb->alloc_addr = __builtin_return_address(0);
2022         eb->start = start;
2023         eb->len = len;
2024         atomic_set(&eb->refs, 1);
2025
2026         for (i = 0; i < num_pages; i++, index++) {
2027                 p = find_get_page(mapping, index);
2028                 if (!p) {
2029                         /* make sure the free only frees the pages we've
2030                          * grabbed a reference on
2031                          */
2032                         eb->len = i << PAGE_CACHE_SHIFT;
2033                         eb->start &= ~((u64)PAGE_CACHE_SIZE - 1);
2034                         goto fail;
2035                 }
2036                 set_page_extent_mapped(p);
2037                 if (i == 0)
2038                         eb->first_page = p;
2039         }
2040         return eb;
2041 fail:
2042         free_extent_buffer(eb);
2043         return NULL;
2044 }
2045 EXPORT_SYMBOL(find_extent_buffer);
2046
2047 void free_extent_buffer(struct extent_buffer *eb)
2048 {
2049         unsigned long i;
2050         unsigned long num_pages;
2051
2052         if (!eb)
2053                 return;
2054
2055         if (!atomic_dec_and_test(&eb->refs))
2056                 return;
2057
2058         num_pages = ((eb->start + eb->len - 1) >> PAGE_CACHE_SHIFT) -
2059                 (eb->start >> PAGE_CACHE_SHIFT) + 1;
2060
2061         if (eb->first_page)
2062                 page_cache_release(eb->first_page);
2063         for (i = 1; i < num_pages; i++) {
2064                 page_cache_release(extent_buffer_page(eb, i));
2065         }
2066         __free_extent_buffer(eb);
2067 }
2068 EXPORT_SYMBOL(free_extent_buffer);
2069
2070 int clear_extent_buffer_dirty(struct extent_map_tree *tree,
2071                               struct extent_buffer *eb)
2072 {
2073         int set;
2074         unsigned long i;
2075         unsigned long num_pages;
2076         struct page *page;
2077
2078         u64 start = eb->start;
2079         u64 end = start + eb->len - 1;
2080
2081         set = clear_extent_dirty(tree, start, end, GFP_NOFS);
2082         num_pages = ((eb->start + eb->len - 1) >> PAGE_CACHE_SHIFT) -
2083                 (eb->start >> PAGE_CACHE_SHIFT) + 1;
2084
2085         for (i = 0; i < num_pages; i++) {
2086                 page = extent_buffer_page(eb, i);
2087                 lock_page(page);
2088                 /*
2089                  * if we're on the last page or the first page and the
2090                  * block isn't aligned on a page boundary, do extra checks
2091                  * to make sure we don't clean page that is partially dirty
2092                  */
2093                 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
2094                     ((i == num_pages - 1) &&
2095                      ((eb->start + eb->len - 1) & (PAGE_CACHE_SIZE - 1)))) {
2096                         start = page->index << PAGE_CACHE_SHIFT;
2097                         end  = start + PAGE_CACHE_SIZE - 1;
2098                         if (test_range_bit(tree, start, end,
2099                                            EXTENT_DIRTY, 0)) {
2100                                 unlock_page(page);
2101                                 continue;
2102                         }
2103                 }
2104                 clear_page_dirty_for_io(page);
2105                 unlock_page(page);
2106         }
2107         return 0;
2108 }
2109 EXPORT_SYMBOL(clear_extent_buffer_dirty);
2110
2111 int wait_on_extent_buffer_writeback(struct extent_map_tree *tree,
2112                                     struct extent_buffer *eb)
2113 {
2114         return wait_on_extent_writeback(tree, eb->start,
2115                                         eb->start + eb->len - 1);
2116 }
2117 EXPORT_SYMBOL(wait_on_extent_buffer_writeback);
2118
2119 int set_extent_buffer_dirty(struct extent_map_tree *tree,
2120                              struct extent_buffer *eb)
2121 {
2122         return set_range_dirty(tree, eb->start, eb->start + eb->len - 1);
2123 }
2124 EXPORT_SYMBOL(set_extent_buffer_dirty);
2125
2126 int set_extent_buffer_uptodate(struct extent_map_tree *tree,
2127                                 struct extent_buffer *eb)
2128 {
2129         unsigned long i;
2130         struct page *page;
2131         unsigned long num_pages;
2132
2133         num_pages = ((eb->start + eb->len - 1) >> PAGE_CACHE_SHIFT) -
2134                 (eb->start >> PAGE_CACHE_SHIFT) + 1;
2135
2136         set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
2137                             GFP_NOFS);
2138         for (i = 0; i < num_pages; i++) {
2139                 page = extent_buffer_page(eb, i);
2140                 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
2141                     ((i == num_pages - 1) &&
2142                      ((eb->start + eb->len - 1) & (PAGE_CACHE_SIZE - 1)))) {
2143                         check_page_uptodate(tree, page);
2144                         continue;
2145                 }
2146                 SetPageUptodate(page);
2147         }
2148         return 0;
2149 }
2150 EXPORT_SYMBOL(set_extent_buffer_uptodate);
2151
2152 int extent_buffer_uptodate(struct extent_map_tree *tree,
2153                              struct extent_buffer *eb)
2154 {
2155         if (eb->flags & EXTENT_UPTODATE)
2156                 return 1;
2157         return test_range_bit(tree, eb->start, eb->start + eb->len - 1,
2158                            EXTENT_UPTODATE, 1);
2159 }
2160 EXPORT_SYMBOL(extent_buffer_uptodate);
2161
2162 int read_extent_buffer_pages(struct extent_map_tree *tree,
2163                              struct extent_buffer *eb, int wait)
2164 {
2165         unsigned long i;
2166         struct page *page;
2167         int err;
2168         int ret = 0;
2169         unsigned long num_pages;
2170
2171         if (eb->flags & EXTENT_UPTODATE)
2172                 return 0;
2173
2174         if (test_range_bit(tree, eb->start, eb->start + eb->len - 1,
2175                            EXTENT_UPTODATE, 1)) {
2176                 return 0;
2177         }
2178
2179         num_pages = ((eb->start + eb->len - 1) >> PAGE_CACHE_SHIFT) -
2180                 (eb->start >> PAGE_CACHE_SHIFT) + 1;
2181         for (i = 0; i < num_pages; i++) {
2182                 page = extent_buffer_page(eb, i);
2183                 if (PageUptodate(page)) {
2184                         continue;
2185                 }
2186                 if (!wait) {
2187                         if (TestSetPageLocked(page)) {
2188                                 continue;
2189                         }
2190                 } else {
2191                         lock_page(page);
2192                 }
2193                 if (!PageUptodate(page)) {
2194                         err = page->mapping->a_ops->readpage(NULL, page);
2195                         if (err) {
2196                                 ret = err;
2197                         }
2198                 } else {
2199                         unlock_page(page);
2200                 }
2201         }
2202
2203         if (ret || !wait) {
2204                 return ret;
2205         }
2206
2207         for (i = 0; i < num_pages; i++) {
2208                 page = extent_buffer_page(eb, i);
2209                 wait_on_page_locked(page);
2210                 if (!PageUptodate(page)) {
2211                         ret = -EIO;
2212                 }
2213         }
2214         eb->flags |= EXTENT_UPTODATE;
2215         return ret;
2216 }
2217 EXPORT_SYMBOL(read_extent_buffer_pages);
2218
2219 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
2220                         unsigned long start,
2221                         unsigned long len)
2222 {
2223         size_t cur;
2224         size_t offset;
2225         struct page *page;
2226         char *kaddr;
2227         char *dst = (char *)dstv;
2228         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
2229         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
2230
2231         WARN_ON(start > eb->len);
2232         WARN_ON(start + len > eb->start + eb->len);
2233
2234         offset = start & ((unsigned long)PAGE_CACHE_SIZE - 1);
2235         if (i == 0)
2236                 offset += start_offset;
2237
2238         while(len > 0) {
2239                 page = extent_buffer_page(eb, i);
2240                 WARN_ON(!PageUptodate(page));
2241
2242                 cur = min(len, (PAGE_CACHE_SIZE - offset));
2243                 kaddr = kmap_atomic(page, KM_USER0);
2244                 memcpy(dst, kaddr + offset, cur);
2245                 kunmap_atomic(kaddr, KM_USER0);
2246
2247                 dst += cur;
2248                 len -= cur;
2249                 offset = 0;
2250                 i++;
2251         }
2252 }
2253 EXPORT_SYMBOL(read_extent_buffer);
2254
2255 int map_extent_buffer(struct extent_buffer *eb, unsigned long start,
2256                       unsigned long min_len,
2257                       char **token, char **map,
2258                       unsigned long *map_start,
2259                       unsigned long *map_len, int km)
2260 {
2261         size_t offset = start & (PAGE_CACHE_SIZE - 1);
2262         char *kaddr;
2263         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
2264         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
2265         unsigned long end_i = (start_offset + start + min_len) >>
2266                                 PAGE_CACHE_SHIFT;
2267
2268         if (i != end_i)
2269                 return -EINVAL;
2270
2271         WARN_ON(start > eb->len);
2272
2273         if (i == 0) {
2274                 offset = start_offset;
2275                 *map_start = 0;
2276         } else {
2277                 *map_start = (i << PAGE_CACHE_SHIFT) - start_offset;
2278         }
2279
2280         kaddr = kmap_atomic(extent_buffer_page(eb, i), km);
2281         *token = kaddr;
2282         *map = kaddr + offset;
2283         *map_len = PAGE_CACHE_SIZE - offset;
2284         return 0;
2285 }
2286 EXPORT_SYMBOL(map_extent_buffer);
2287
2288 void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km)
2289 {
2290         kunmap_atomic(token, km);
2291 }
2292 EXPORT_SYMBOL(unmap_extent_buffer);
2293
2294 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
2295                           unsigned long start,
2296                           unsigned long len)
2297 {
2298         size_t cur;
2299         size_t offset;
2300         struct page *page;
2301         char *kaddr;
2302         char *ptr = (char *)ptrv;
2303         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
2304         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
2305         int ret = 0;
2306
2307         WARN_ON(start > eb->len);
2308         WARN_ON(start + len > eb->start + eb->len);
2309
2310         offset = start & ((unsigned long)PAGE_CACHE_SIZE - 1);
2311         if (i == 0)
2312                 offset += start_offset;
2313
2314         while(len > 0) {
2315                 page = extent_buffer_page(eb, i);
2316                 WARN_ON(!PageUptodate(page));
2317
2318                 cur = min(len, (PAGE_CACHE_SIZE - offset));
2319
2320                 kaddr = kmap_atomic(page, KM_USER0);
2321                 ret = memcmp(ptr, kaddr + offset, cur);
2322                 kunmap_atomic(kaddr, KM_USER0);
2323                 if (ret)
2324                         break;
2325
2326                 ptr += cur;
2327                 len -= cur;
2328                 offset = 0;
2329                 i++;
2330         }
2331         return ret;
2332 }
2333 EXPORT_SYMBOL(memcmp_extent_buffer);
2334
2335 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
2336                          unsigned long start, unsigned long len)
2337 {
2338         size_t cur;
2339         size_t offset;
2340         struct page *page;
2341         char *kaddr;
2342         char *src = (char *)srcv;
2343         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
2344         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
2345
2346         WARN_ON(start > eb->len);
2347         WARN_ON(start + len > eb->start + eb->len);
2348
2349         offset = start & ((unsigned long)PAGE_CACHE_SIZE - 1);
2350         if (i == 0)
2351                 offset += start_offset;
2352
2353         while(len > 0) {
2354                 page = extent_buffer_page(eb, i);
2355                 WARN_ON(!PageUptodate(page));
2356
2357                 cur = min(len, PAGE_CACHE_SIZE - offset);
2358                 kaddr = kmap_atomic(page, KM_USER0);
2359                 memcpy(kaddr + offset, src, cur);
2360                 kunmap_atomic(kaddr, KM_USER0);
2361
2362                 src += cur;
2363                 len -= cur;
2364                 offset = 0;
2365                 i++;
2366         }
2367 }
2368 EXPORT_SYMBOL(write_extent_buffer);
2369
2370 void memset_extent_buffer(struct extent_buffer *eb, char c,
2371                           unsigned long start, unsigned long len)
2372 {
2373         size_t cur;
2374         size_t offset;
2375         struct page *page;
2376         char *kaddr;
2377         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
2378         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
2379
2380         WARN_ON(start > eb->len);
2381         WARN_ON(start + len > eb->start + eb->len);
2382
2383         offset = start & ((unsigned long)PAGE_CACHE_SIZE - 1);
2384         if (i == 0)
2385                 offset += start_offset;
2386
2387         while(len > 0) {
2388                 page = extent_buffer_page(eb, i);
2389                 WARN_ON(!PageUptodate(page));
2390
2391                 cur = min(len, PAGE_CACHE_SIZE - offset);
2392                 kaddr = kmap_atomic(page, KM_USER0);
2393                 memset(kaddr + offset, c, cur);
2394                 kunmap_atomic(kaddr, KM_USER0);
2395
2396                 len -= cur;
2397                 offset = 0;
2398                 i++;
2399         }
2400 }
2401 EXPORT_SYMBOL(memset_extent_buffer);
2402
2403 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
2404                         unsigned long dst_offset, unsigned long src_offset,
2405                         unsigned long len)
2406 {
2407         u64 dst_len = dst->len;
2408         size_t cur;
2409         size_t offset;
2410         struct page *page;
2411         char *kaddr;
2412         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
2413         unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
2414
2415         WARN_ON(src->len != dst_len);
2416
2417         offset = dst_offset & ((unsigned long)PAGE_CACHE_SIZE - 1);
2418         if (i == 0)
2419                 offset += start_offset;
2420
2421         while(len > 0) {
2422                 page = extent_buffer_page(dst, i);
2423                 WARN_ON(!PageUptodate(page));
2424
2425                 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
2426
2427                 kaddr = kmap_atomic(page, KM_USER1);
2428                 read_extent_buffer(src, kaddr + offset, src_offset, cur);
2429                 kunmap_atomic(kaddr, KM_USER1);
2430
2431                 src_offset += cur;
2432                 len -= cur;
2433                 offset = 0;
2434                 i++;
2435         }
2436 }
2437 EXPORT_SYMBOL(copy_extent_buffer);
2438
2439 static void move_pages(struct page *dst_page, struct page *src_page,
2440                        unsigned long dst_off, unsigned long src_off,
2441                        unsigned long len)
2442 {
2443         char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
2444         if (dst_page == src_page) {
2445                 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
2446         } else {
2447                 char *src_kaddr = kmap_atomic(src_page, KM_USER1);
2448                 char *p = dst_kaddr + dst_off + len;
2449                 char *s = src_kaddr + src_off + len;
2450
2451                 while (len--)
2452                         *--p = *--s;
2453
2454                 kunmap_atomic(src_kaddr, KM_USER1);
2455         }
2456         kunmap_atomic(dst_kaddr, KM_USER0);
2457 }
2458
2459 static void copy_pages(struct page *dst_page, struct page *src_page,
2460                        unsigned long dst_off, unsigned long src_off,
2461                        unsigned long len)
2462 {
2463         char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
2464         char *src_kaddr;
2465
2466         if (dst_page != src_page)
2467                 src_kaddr = kmap_atomic(src_page, KM_USER1);
2468         else
2469                 src_kaddr = dst_kaddr;
2470
2471         memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
2472         kunmap_atomic(dst_kaddr, KM_USER0);
2473         if (dst_page != src_page)
2474                 kunmap_atomic(src_kaddr, KM_USER1);
2475 }
2476
2477 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
2478                            unsigned long src_offset, unsigned long len)
2479 {
2480         size_t cur;
2481         size_t dst_off_in_page;
2482         size_t src_off_in_page;
2483         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
2484         unsigned long dst_i;
2485         unsigned long src_i;
2486
2487         if (src_offset + len > dst->len) {
2488                 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
2489                        src_offset, len, dst->len);
2490                 BUG_ON(1);
2491         }
2492         if (dst_offset + len > dst->len) {
2493                 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
2494                        dst_offset, len, dst->len);
2495                 BUG_ON(1);
2496         }
2497
2498         while(len > 0) {
2499                 dst_off_in_page = dst_offset &
2500                         ((unsigned long)PAGE_CACHE_SIZE - 1);
2501                 src_off_in_page = src_offset &
2502                         ((unsigned long)PAGE_CACHE_SIZE - 1);
2503
2504                 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
2505                 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
2506
2507                 if (src_i == 0)
2508                         src_off_in_page += start_offset;
2509                 if (dst_i == 0)
2510                         dst_off_in_page += start_offset;
2511
2512                 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
2513                                                src_off_in_page));
2514                 cur = min(cur, (unsigned long)(PAGE_CACHE_SIZE -
2515                                                dst_off_in_page));
2516
2517                 copy_pages(extent_buffer_page(dst, dst_i),
2518                            extent_buffer_page(dst, src_i),
2519                            dst_off_in_page, src_off_in_page, cur);
2520
2521                 src_offset += cur;
2522                 dst_offset += cur;
2523                 len -= cur;
2524         }
2525 }
2526 EXPORT_SYMBOL(memcpy_extent_buffer);
2527
2528 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
2529                            unsigned long src_offset, unsigned long len)
2530 {
2531         size_t cur;
2532         size_t dst_off_in_page;
2533         size_t src_off_in_page;
2534         unsigned long dst_end = dst_offset + len - 1;
2535         unsigned long src_end = src_offset + len - 1;
2536         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
2537         unsigned long dst_i;
2538         unsigned long src_i;
2539
2540         if (src_offset + len > dst->len) {
2541                 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
2542                        src_offset, len, dst->len);
2543                 BUG_ON(1);
2544         }
2545         if (dst_offset + len > dst->len) {
2546                 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
2547                        dst_offset, len, dst->len);
2548                 BUG_ON(1);
2549         }
2550         if (dst_offset < src_offset) {
2551                 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
2552                 return;
2553         }
2554         while(len > 0) {
2555                 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
2556                 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
2557
2558                 dst_off_in_page = dst_end &
2559                         ((unsigned long)PAGE_CACHE_SIZE - 1);
2560                 src_off_in_page = src_end &
2561                         ((unsigned long)PAGE_CACHE_SIZE - 1);
2562
2563                 if (src_i == 0)
2564                         src_off_in_page += start_offset;
2565                 if (dst_i == 0)
2566                         dst_off_in_page += start_offset;
2567
2568                 cur = min(len, src_off_in_page + 1);
2569                 cur = min(cur, dst_off_in_page + 1);
2570
2571                 move_pages(extent_buffer_page(dst, dst_i),
2572                            extent_buffer_page(dst, src_i),
2573                            dst_off_in_page - cur + 1,
2574                            src_off_in_page - cur + 1, cur);
2575
2576                 dst_end -= cur - 1;
2577                 src_end -= cur - 1;
2578                 len -= cur;
2579         }
2580 }
2581 EXPORT_SYMBOL(memmove_extent_buffer);