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

   1 #include <linux/bitops.h>
   2 #include <linux/slab.h>
   3 #include <linux/bio.h>
   4 #include <linux/mm.h>
   5 #include <linux/gfp.h>
   6 #include <linux/pagemap.h>
   7 #include <linux/page-flags.h>
   8 #include <linux/module.h>
   9 #include <linux/spinlock.h>
  10 #include <linux/blkdev.h>
  11 #include "extent_map.h"
  12
  13 /* temporary define until extent_map moves out of btrfs */
  14 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
  15                                        unsigned long extra_flags,
  16                                        void (*ctor)(void *, struct kmem_cache *,
  17                                                     unsigned long));
  18
  19 static struct kmem_cache *extent_map_cache;
  20 static struct kmem_cache *extent_state_cache;
  21 static struct kmem_cache *extent_buffer_cache;
  22
  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_bits(struct extent_map_tree *tree, u64 start, u64 end,
 823                     int bits, gfp_t mask)
 824 {
 825         return set_extent_bit(tree, start, end, bits, 0, NULL,
 826                               mask);
 827 }
 828 EXPORT_SYMBOL(set_extent_bits);
 829
 830 int clear_extent_bits(struct extent_map_tree *tree, u64 start, u64 end,
 831                       int bits, gfp_t mask)
 832 {
 833         return clear_extent_bit(tree, start, end, bits, 0, 0, mask);
 834 }
 835 EXPORT_SYMBOL(clear_extent_bits);
 836
 837 int set_extent_delalloc(struct extent_map_tree *tree, u64 start, u64 end,
 838                      gfp_t mask)
 839 {
 840         return set_extent_bit(tree, start, end,
 841                               EXTENT_DELALLOC | EXTENT_DIRTY, 0, NULL,
 842                               mask);
 843 }
 844 EXPORT_SYMBOL(set_extent_delalloc);
 845
 846 int clear_extent_dirty(struct extent_map_tree *tree, u64 start, u64 end,
 847                        gfp_t mask)
 848 {
 849         return clear_extent_bit(tree, start, end,
 850                                 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, mask);
 851 }
 852 EXPORT_SYMBOL(clear_extent_dirty);
 853
 854 int set_extent_new(struct extent_map_tree *tree, u64 start, u64 end,
 855                      gfp_t mask)
 856 {
 857         return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL,
 858                               mask);
 859 }
 860 EXPORT_SYMBOL(set_extent_new);
 861
 862 int clear_extent_new(struct extent_map_tree *tree, u64 start, u64 end,
 863                        gfp_t mask)
 864 {
 865         return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0, mask);
 866 }
 867 EXPORT_SYMBOL(clear_extent_new);
 868
 869 int set_extent_uptodate(struct extent_map_tree *tree, u64 start, u64 end,
 870                         gfp_t mask)
 871 {
 872         return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, NULL,
 873                               mask);
 874 }
 875 EXPORT_SYMBOL(set_extent_uptodate);
 876
 877 int clear_extent_uptodate(struct extent_map_tree *tree, u64 start, u64 end,
 878                           gfp_t mask)
 879 {
 880         return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, mask);
 881 }
 882 EXPORT_SYMBOL(clear_extent_uptodate);
 883
 884 int set_extent_writeback(struct extent_map_tree *tree, u64 start, u64 end,
 885                          gfp_t mask)
 886 {
 887         return set_extent_bit(tree, start, end, EXTENT_WRITEBACK,
 888                               0, NULL, mask);
 889 }
 890 EXPORT_SYMBOL(set_extent_writeback);
 891
 892 int clear_extent_writeback(struct extent_map_tree *tree, u64 start, u64 end,
 893                            gfp_t mask)
 894 {
 895         return clear_extent_bit(tree, start, end, EXTENT_WRITEBACK, 1, 0, mask);
 896 }
 897 EXPORT_SYMBOL(clear_extent_writeback);
 898
 899 int wait_on_extent_writeback(struct extent_map_tree *tree, u64 start, u64 end)
 900 {
 901         return wait_extent_bit(tree, start, end, EXTENT_WRITEBACK);
 902 }
 903 EXPORT_SYMBOL(wait_on_extent_writeback);
 904
 905 /*
 906  * locks a range in ascending order, waiting for any locked regions
 907  * it hits on the way.  [start,end] are inclusive, and this will sleep.
 908  */
 909 int lock_extent(struct extent_map_tree *tree, u64 start, u64 end, gfp_t mask)
 910 {
 911         int err;
 912         u64 failed_start;
 913         while (1) {
 914                 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1,
 915                                      &failed_start, mask);
 916                 if (err == -EEXIST && (mask & __GFP_WAIT)) {
 917                         wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
 918                         start = failed_start;
 919                 } else {
 920                         break;
 921                 }
 922                 WARN_ON(start > end);
 923         }
 924         return err;
 925 }
 926 EXPORT_SYMBOL(lock_extent);
 927
 928 int unlock_extent(struct extent_map_tree *tree, u64 start, u64 end,
 929                   gfp_t mask)
 930 {
 931         return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, mask);
 932 }
 933 EXPORT_SYMBOL(unlock_extent);
 934
 935 /*
 936  * helper function to set pages and extents in the tree dirty
 937  */
 938 int set_range_dirty(struct extent_map_tree *tree, u64 start, u64 end)
 939 {
 940         unsigned long index = start >> PAGE_CACHE_SHIFT;
 941         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
 942         struct page *page;
 943
 944         while (index <= end_index) {
 945                 page = find_get_page(tree->mapping, index);
 946                 BUG_ON(!page);
 947                 __set_page_dirty_nobuffers(page);
 948                 page_cache_release(page);
 949                 index++;
 950         }
 951         set_extent_dirty(tree, start, end, GFP_NOFS);
 952         return 0;
 953 }
 954 EXPORT_SYMBOL(set_range_dirty);
 955
 956 /*
 957  * helper function to set both pages and extents in the tree writeback
 958  */
 959 int set_range_writeback(struct extent_map_tree *tree, u64 start, u64 end)
 960 {
 961         unsigned long index = start >> PAGE_CACHE_SHIFT;
 962         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
 963         struct page *page;
 964
 965         while (index <= end_index) {
 966                 page = find_get_page(tree->mapping, index);
 967                 BUG_ON(!page);
 968                 set_page_writeback(page);
 969                 page_cache_release(page);
 970                 index++;
 971         }
 972         set_extent_writeback(tree, start, end, GFP_NOFS);
 973         return 0;
 974 }
 975 EXPORT_SYMBOL(set_range_writeback);
 976
 977 int find_first_extent_bit(struct extent_map_tree *tree, u64 start,
 978                           u64 *start_ret, u64 *end_ret, int bits)
 979 {
 980         struct rb_node *node;
 981         struct extent_state *state;
 982         int ret = 1;
 983
 984         write_lock_irq(&tree->lock);
 985         /*
 986          * this search will find all the extents that end after
 987          * our range starts.
 988          */
 989         node = tree_search(&tree->state, start);
 990         if (!node || IS_ERR(node)) {
 991                 goto out;
 992         }
 993
 994         while(1) {
 995                 state = rb_entry(node, struct extent_state, rb_node);
 996                 if (state->state & bits) {
 997                         *start_ret = state->start;
 998                         *end_ret = state->end;
 999                         ret = 0;
1000                         break;
1001                 }
1002                 node = rb_next(node);
1003                 if (!node)
1004                         break;
1005         }
1006 out:
1007         write_unlock_irq(&tree->lock);
1008         return ret;
1009 }
1010 EXPORT_SYMBOL(find_first_extent_bit);
1011
1012 u64 find_lock_delalloc_range(struct extent_map_tree *tree,
1013                              u64 start, u64 lock_start, u64 *end, u64 max_bytes)
1014 {
1015         struct rb_node *node;
1016         struct extent_state *state;
1017         u64 cur_start = start;
1018         u64 found = 0;
1019         u64 total_bytes = 0;
1020
1021         write_lock_irq(&tree->lock);
1022         /*
1023          * this search will find all the extents that end after
1024          * our range starts.
1025          */
1026 search_again:
1027         node = tree_search(&tree->state, cur_start);
1028         if (!node || IS_ERR(node)) {
1029                 goto out;
1030         }
1031
1032         while(1) {
1033                 state = rb_entry(node, struct extent_state, rb_node);
1034                 if (state->start != cur_start) {
1035                         goto out;
1036                 }
1037                 if (!(state->state & EXTENT_DELALLOC)) {
1038                         goto out;
1039                 }
1040                 if (state->start >= lock_start) {
1041                         if (state->state & EXTENT_LOCKED) {
1042                                 DEFINE_WAIT(wait);
1043                                 atomic_inc(&state->refs);
1044                                 write_unlock_irq(&tree->lock);
1045                                 schedule();
1046                                 write_lock_irq(&tree->lock);
1047                                 finish_wait(&state->wq, &wait);
1048                                 free_extent_state(state);
1049                                 goto search_again;
1050                         }
1051                         state->state |= EXTENT_LOCKED;
1052                 }
1053                 found++;
1054                 *end = state->end;
1055                 cur_start = state->end + 1;
1056                 node = rb_next(node);
1057                 if (!node)
1058                         break;
1059                 total_bytes = state->end - state->start + 1;
1060                 if (total_bytes >= max_bytes)
1061                         break;
1062         }
1063 out:
1064         write_unlock_irq(&tree->lock);
1065         return found;
1066 }
1067
1068 /*
1069  * helper function to lock both pages and extents in the tree.
1070  * pages must be locked first.
1071  */
1072 int lock_range(struct extent_map_tree *tree, u64 start, u64 end)
1073 {
1074         unsigned long index = start >> PAGE_CACHE_SHIFT;
1075         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1076         struct page *page;
1077         int err;
1078
1079         while (index <= end_index) {
1080                 page = grab_cache_page(tree->mapping, index);
1081                 if (!page) {
1082                         err = -ENOMEM;
1083                         goto failed;
1084                 }
1085                 if (IS_ERR(page)) {
1086                         err = PTR_ERR(page);
1087                         goto failed;
1088                 }
1089                 index++;
1090         }
1091         lock_extent(tree, start, end, GFP_NOFS);
1092         return 0;
1093
1094 failed:
1095         /*
1096          * we failed above in getting the page at 'index', so we undo here
1097          * up to but not including the page at 'index'
1098          */
1099         end_index = index;
1100         index = start >> PAGE_CACHE_SHIFT;
1101         while (index < end_index) {
1102                 page = find_get_page(tree->mapping, index);
1103                 unlock_page(page);
1104                 page_cache_release(page);
1105                 index++;
1106         }
1107         return err;
1108 }
1109 EXPORT_SYMBOL(lock_range);
1110
1111 /*
1112  * helper function to unlock both pages and extents in the tree.
1113  */
1114 int unlock_range(struct extent_map_tree *tree, u64 start, u64 end)
1115 {
1116         unsigned long index = start >> PAGE_CACHE_SHIFT;
1117         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1118         struct page *page;
1119
1120         while (index <= end_index) {
1121                 page = find_get_page(tree->mapping, index);
1122                 unlock_page(page);
1123                 page_cache_release(page);
1124                 index++;
1125         }
1126         unlock_extent(tree, start, end, GFP_NOFS);
1127         return 0;
1128 }
1129 EXPORT_SYMBOL(unlock_range);
1130
1131 int set_state_private(struct extent_map_tree *tree, u64 start, u64 private)
1132 {
1133         struct rb_node *node;
1134         struct extent_state *state;
1135         int ret = 0;
1136
1137         write_lock_irq(&tree->lock);
1138         /*
1139          * this search will find all the extents that end after
1140          * our range starts.
1141          */
1142         node = tree_search(&tree->state, start);
1143         if (!node || IS_ERR(node)) {
1144                 ret = -ENOENT;
1145                 goto out;
1146         }
1147         state = rb_entry(node, struct extent_state, rb_node);
1148         if (state->start != start) {
1149                 ret = -ENOENT;
1150                 goto out;
1151         }
1152         state->private = private;
1153 out:
1154         write_unlock_irq(&tree->lock);
1155         return ret;
1156 }
1157
1158 int get_state_private(struct extent_map_tree *tree, u64 start, u64 *private)
1159 {
1160         struct rb_node *node;
1161         struct extent_state *state;
1162         int ret = 0;
1163
1164         read_lock_irq(&tree->lock);
1165         /*
1166          * this search will find all the extents that end after
1167          * our range starts.
1168          */
1169         node = tree_search(&tree->state, start);
1170         if (!node || IS_ERR(node)) {
1171                 ret = -ENOENT;
1172                 goto out;
1173         }
1174         state = rb_entry(node, struct extent_state, rb_node);
1175         if (state->start != start) {
1176                 ret = -ENOENT;
1177                 goto out;
1178         }
1179         *private = state->private;
1180 out:
1181         read_unlock_irq(&tree->lock);
1182         return ret;
1183 }
1184
1185 /*
1186  * searches a range in the state tree for a given mask.
1187  * If 'filled' == 1, this returns 1 only if ever extent in the tree
1188  * has the bits set.  Otherwise, 1 is returned if any bit in the
1189  * range is found set.
1190  */
1191 static int test_range_bit(struct extent_map_tree *tree, u64 start, u64 end,
1192                           int bits, int filled)
1193 {
1194         struct extent_state *state = NULL;
1195         struct rb_node *node;
1196         int bitset = 0;
1197
1198         read_lock_irq(&tree->lock);
1199         node = tree_search(&tree->state, start);
1200         while (node && start <= end) {
1201                 state = rb_entry(node, struct extent_state, rb_node);
1202                 if (state->start > end)
1203                         break;
1204
1205                 if (filled && state->start > start) {
1206                         bitset = 0;
1207                         break;
1208                 }
1209                 if (state->state & bits) {
1210                         bitset = 1;
1211                         if (!filled)
1212                                 break;
1213                 } else if (filled) {
1214                         bitset = 0;
1215                         break;
1216                 }
1217                 start = state->end + 1;
1218                 if (start > end)
1219                         break;
1220                 node = rb_next(node);
1221         }
1222         read_unlock_irq(&tree->lock);
1223         return bitset;
1224 }
1225
1226 /*
1227  * helper function to set a given page up to date if all the
1228  * extents in the tree for that page are up to date
1229  */
1230 static int check_page_uptodate(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_UPTODATE, 1))
1236                 SetPageUptodate(page);
1237         return 0;
1238 }
1239
1240 /*
1241  * helper function to unlock a page if all the extents in the tree
1242  * for that page are unlocked
1243  */
1244 static int check_page_locked(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_LOCKED, 0))
1250                 unlock_page(page);
1251         return 0;
1252 }
1253
1254 /*
1255  * helper function to end page writeback if all the extents
1256  * in the tree for that page are done with writeback
1257  */
1258 static int check_page_writeback(struct extent_map_tree *tree,
1259                              struct page *page)
1260 {
1261         u64 start = page->index << PAGE_CACHE_SHIFT;
1262         u64 end = start + PAGE_CACHE_SIZE - 1;
1263         if (!test_range_bit(tree, start, end, EXTENT_WRITEBACK, 0))
1264                 end_page_writeback(page);
1265         return 0;
1266 }
1267
1268 /* lots and lots of room for performance fixes in the end_bio funcs */
1269
1270 /*
1271  * after a writepage IO is done, we need to:
1272  * clear the uptodate bits on error
1273  * clear the writeback bits in the extent tree for this IO
1274  * end_page_writeback if the page has no more pending IO
1275  *
1276  * Scheduling is not allowed, so the extent state tree is expected
1277  * to have one and only one object corresponding to this IO.
1278  */
1279 static int end_bio_extent_writepage(struct bio *bio,
1280                                    unsigned int bytes_done, int err)
1281 {
1282         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1283         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1284         struct extent_map_tree *tree = bio->bi_private;
1285         u64 start;
1286         u64 end;
1287         int whole_page;
1288
1289         if (bio->bi_size)
1290                 return 1;
1291
1292         do {
1293                 struct page *page = bvec->bv_page;
1294                 start = (page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
1295                 end = start + bvec->bv_len - 1;
1296
1297                 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1298                         whole_page = 1;
1299                 else
1300                         whole_page = 0;
1301
1302                 if (--bvec >= bio->bi_io_vec)
1303                         prefetchw(&bvec->bv_page->flags);
1304
1305                 if (!uptodate) {
1306                         clear_extent_uptodate(tree, start, end, GFP_ATOMIC);
1307                         ClearPageUptodate(page);
1308                         SetPageError(page);
1309                 }
1310                 clear_extent_writeback(tree, start, end, GFP_ATOMIC);
1311
1312                 if (whole_page)
1313                         end_page_writeback(page);
1314                 else
1315                         check_page_writeback(tree, page);
1316                 if (tree->ops && tree->ops->writepage_end_io_hook)
1317                         tree->ops->writepage_end_io_hook(page, start, end);
1318         } while (bvec >= bio->bi_io_vec);
1319
1320         bio_put(bio);
1321         return 0;
1322 }
1323
1324 /*
1325  * after a readpage IO is done, we need to:
1326  * clear the uptodate bits on error
1327  * set the uptodate bits if things worked
1328  * set the page up to date if all extents in the tree are uptodate
1329  * clear the lock bit in the extent tree
1330  * unlock the page if there are no other extents locked for it
1331  *
1332  * Scheduling is not allowed, so the extent state tree is expected
1333  * to have one and only one object corresponding to this IO.
1334  */
1335 static int end_bio_extent_readpage(struct bio *bio,
1336                                    unsigned int bytes_done, int err)
1337 {
1338         int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1339         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1340         struct extent_map_tree *tree = bio->bi_private;
1341         u64 start;
1342         u64 end;
1343         int whole_page;
1344         int ret;
1345
1346         if (bio->bi_size)
1347                 return 1;
1348
1349         do {
1350                 struct page *page = bvec->bv_page;
1351                 start = (page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
1352                 end = start + bvec->bv_len - 1;
1353
1354                 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1355                         whole_page = 1;
1356                 else
1357                         whole_page = 0;
1358
1359                 if (--bvec >= bio->bi_io_vec)
1360                         prefetchw(&bvec->bv_page->flags);
1361
1362                 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
1363                         ret = tree->ops->readpage_end_io_hook(page, start, end);
1364                         if (ret)
1365                                 uptodate = 0;
1366                 }
1367                 if (uptodate) {
1368                         set_extent_uptodate(tree, start, end, GFP_ATOMIC);
1369                         if (whole_page)
1370                                 SetPageUptodate(page);
1371                         else
1372                                 check_page_uptodate(tree, page);
1373                 } else {
1374                         ClearPageUptodate(page);
1375                         SetPageError(page);
1376                 }
1377
1378                 unlock_extent(tree, start, end, GFP_ATOMIC);
1379
1380                 if (whole_page)
1381                         unlock_page(page);
1382                 else
1383                         check_page_locked(tree, page);
1384         } while (bvec >= bio->bi_io_vec);
1385
1386         bio_put(bio);
1387         return 0;
1388 }
1389
1390 /*
1391  * IO done from prepare_write is pretty simple, we just unlock
1392  * the structs in the extent tree when done, and set the uptodate bits
1393  * as appropriate.
1394  */
1395 static int end_bio_extent_preparewrite(struct bio *bio,
1396                                        unsigned int bytes_done, int err)
1397 {
1398         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1399         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1400         struct extent_map_tree *tree = bio->bi_private;
1401         u64 start;
1402         u64 end;
1403
1404         if (bio->bi_size)
1405                 return 1;
1406
1407         do {
1408                 struct page *page = bvec->bv_page;
1409                 start = (page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
1410                 end = start + bvec->bv_len - 1;
1411
1412                 if (--bvec >= bio->bi_io_vec)
1413                         prefetchw(&bvec->bv_page->flags);
1414
1415                 if (uptodate) {
1416                         set_extent_uptodate(tree, start, end, GFP_ATOMIC);
1417                 } else {
1418                         ClearPageUptodate(page);
1419                         SetPageError(page);
1420                 }
1421
1422                 unlock_extent(tree, start, end, GFP_ATOMIC);
1423
1424         } while (bvec >= bio->bi_io_vec);
1425
1426         bio_put(bio);
1427         return 0;
1428 }
1429
1430 static int submit_extent_page(int rw, struct extent_map_tree *tree,
1431                               struct page *page, sector_t sector,
1432                               size_t size, unsigned long offset,
1433                               struct block_device *bdev,
1434                               bio_end_io_t end_io_func)
1435 {
1436         struct bio *bio;
1437         int ret = 0;
1438
1439         bio = bio_alloc(GFP_NOIO, 1);
1440
1441         bio->bi_sector = sector;
1442         bio->bi_bdev = bdev;
1443         bio->bi_io_vec[0].bv_page = page;
1444         bio->bi_io_vec[0].bv_len = size;
1445         bio->bi_io_vec[0].bv_offset = offset;
1446
1447         bio->bi_vcnt = 1;
1448         bio->bi_idx = 0;
1449         bio->bi_size = size;
1450
1451         bio->bi_end_io = end_io_func;
1452         bio->bi_private = tree;
1453
1454         bio_get(bio);
1455         submit_bio(rw, bio);
1456
1457         if (bio_flagged(bio, BIO_EOPNOTSUPP))
1458                 ret = -EOPNOTSUPP;
1459
1460         bio_put(bio);
1461         return ret;
1462 }
1463
1464 void set_page_extent_mapped(struct page *page)
1465 {
1466         if (!PagePrivate(page)) {
1467                 SetPagePrivate(page);
1468                 WARN_ON(!page->mapping->a_ops->invalidatepage);
1469                 set_page_private(page, 1);
1470                 page_cache_get(page);
1471         }
1472 }
1473
1474 /*
1475  * basic readpage implementation.  Locked extent state structs are inserted
1476  * into the tree that are removed when the IO is done (by the end_io
1477  * handlers)
1478  */
1479 int extent_read_full_page(struct extent_map_tree *tree, struct page *page,
1480                           get_extent_t *get_extent)
1481 {
1482         struct inode *inode = page->mapping->host;
1483         u64 start = page->index << PAGE_CACHE_SHIFT;
1484         u64 page_end = start + PAGE_CACHE_SIZE - 1;
1485         u64 end;
1486         u64 cur = start;
1487         u64 extent_offset;
1488         u64 last_byte = i_size_read(inode);
1489         u64 block_start;
1490         u64 cur_end;
1491         sector_t sector;
1492         struct extent_map *em;
1493         struct block_device *bdev;
1494         int ret;
1495         int nr = 0;
1496         size_t page_offset = 0;
1497         size_t iosize;
1498         size_t blocksize = inode->i_sb->s_blocksize;
1499
1500         set_page_extent_mapped(page);
1501
1502         end = page_end;
1503         lock_extent(tree, start, end, GFP_NOFS);
1504
1505         while (cur <= end) {
1506                 if (cur >= last_byte) {
1507                         iosize = PAGE_CACHE_SIZE - page_offset;
1508                         zero_user_page(page, page_offset, iosize, KM_USER0);
1509                         set_extent_uptodate(tree, cur, cur + iosize - 1,
1510                                             GFP_NOFS);
1511                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1512                         break;
1513                 }
1514                 em = get_extent(inode, page, page_offset, cur, end, 0);
1515                 if (IS_ERR(em) || !em) {
1516                         SetPageError(page);
1517                         unlock_extent(tree, cur, end, GFP_NOFS);
1518                         break;
1519                 }
1520
1521                 extent_offset = cur - em->start;
1522                 BUG_ON(em->end < cur);
1523                 BUG_ON(end < cur);
1524
1525                 iosize = min(em->end - cur, end - cur) + 1;
1526                 cur_end = min(em->end, end);
1527                 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
1528                 sector = (em->block_start + extent_offset) >> 9;
1529                 bdev = em->bdev;
1530                 block_start = em->block_start;
1531                 free_extent_map(em);
1532                 em = NULL;
1533
1534                 /* we've found a hole, just zero and go on */
1535                 if (block_start == EXTENT_MAP_HOLE) {
1536                         zero_user_page(page, page_offset, iosize, KM_USER0);
1537                         set_extent_uptodate(tree, cur, cur + iosize - 1,
1538                                             GFP_NOFS);
1539                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1540                         cur = cur + iosize;
1541                         page_offset += iosize;
1542                         continue;
1543                 }
1544                 /* the get_extent function already copied into the page */
1545                 if (test_range_bit(tree, cur, cur_end, EXTENT_UPTODATE, 1)) {
1546                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1547                         cur = cur + iosize;
1548                         page_offset += iosize;
1549                         continue;
1550                 }
1551
1552                 ret = 0;
1553                 if (tree->ops && tree->ops->readpage_io_hook) {
1554                         ret = tree->ops->readpage_io_hook(page, cur,
1555                                                           cur + iosize - 1);
1556                 }
1557                 if (!ret) {
1558                         ret = submit_extent_page(READ, tree, page,
1559                                                  sector, iosize, page_offset,
1560                                                  bdev, end_bio_extent_readpage);
1561                 }
1562                 if (ret)
1563                         SetPageError(page);
1564                 cur = cur + iosize;
1565                 page_offset += iosize;
1566                 nr++;
1567         }
1568         if (!nr) {
1569                 if (!PageError(page))
1570                         SetPageUptodate(page);
1571                 unlock_page(page);
1572         }
1573         return 0;
1574 }
1575 EXPORT_SYMBOL(extent_read_full_page);
1576
1577 /*
1578  * the writepage semantics are similar to regular writepage.  extent
1579  * records are inserted to lock ranges in the tree, and as dirty areas
1580  * are found, they are marked writeback.  Then the lock bits are removed
1581  * and the end_io handler clears the writeback ranges
1582  */
1583 int extent_write_full_page(struct extent_map_tree *tree, struct page *page,
1584                           get_extent_t *get_extent,
1585                           struct writeback_control *wbc)
1586 {
1587         struct inode *inode = page->mapping->host;
1588         u64 start = page->index << PAGE_CACHE_SHIFT;
1589         u64 page_end = start + PAGE_CACHE_SIZE - 1;
1590         u64 end;
1591         u64 cur = start;
1592         u64 extent_offset;
1593         u64 last_byte = i_size_read(inode);
1594         u64 block_start;
1595         sector_t sector;
1596         struct extent_map *em;
1597         struct block_device *bdev;
1598         int ret;
1599         int nr = 0;
1600         size_t page_offset = 0;
1601         size_t iosize;
1602         size_t blocksize;
1603         loff_t i_size = i_size_read(inode);
1604         unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
1605         u64 nr_delalloc;
1606         u64 delalloc_end;
1607
1608         WARN_ON(!PageLocked(page));
1609         if (page->index > end_index) {
1610                 clear_extent_dirty(tree, start, page_end, GFP_NOFS);
1611                 unlock_page(page);
1612                 return 0;
1613         }
1614
1615         if (page->index == end_index) {
1616                 size_t offset = i_size & (PAGE_CACHE_SIZE - 1);
1617                 zero_user_page(page, offset,
1618                                PAGE_CACHE_SIZE - offset, KM_USER0);
1619         }
1620
1621         set_page_extent_mapped(page);
1622
1623         lock_extent(tree, start, page_end, GFP_NOFS);
1624         nr_delalloc = find_lock_delalloc_range(tree, start, page_end + 1,
1625                                                &delalloc_end,
1626                                                128 * 1024 * 1024);
1627         if (nr_delalloc) {
1628                 tree->ops->fill_delalloc(inode, start, delalloc_end);
1629                 if (delalloc_end >= page_end + 1) {
1630                         clear_extent_bit(tree, page_end + 1, delalloc_end,
1631                                          EXTENT_LOCKED | EXTENT_DELALLOC,
1632                                          1, 0, GFP_NOFS);
1633                 }
1634                 clear_extent_bit(tree, start, page_end, EXTENT_DELALLOC,
1635                                  0, 0, GFP_NOFS);
1636                 if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0)) {
1637                         printk("found delalloc bits after clear extent_bit\n");
1638                 }
1639         } else if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0)) {
1640                 printk("found delalloc bits after find_delalloc_range returns 0\n");
1641         }
1642
1643         end = page_end;
1644         if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0)) {
1645                 printk("found delalloc bits after lock_extent\n");
1646         }
1647
1648         if (last_byte <= start) {
1649                 clear_extent_dirty(tree, start, page_end, GFP_NOFS);
1650                 goto done;
1651         }
1652
1653         set_extent_uptodate(tree, start, page_end, GFP_NOFS);
1654         blocksize = inode->i_sb->s_blocksize;
1655
1656         while (cur <= end) {
1657                 if (cur >= last_byte) {
1658                         clear_extent_dirty(tree, cur, page_end, GFP_NOFS);
1659                         break;
1660                 }
1661                 em = get_extent(inode, page, page_offset, cur, end, 0);
1662                 if (IS_ERR(em) || !em) {
1663                         SetPageError(page);
1664                         break;
1665                 }
1666
1667                 extent_offset = cur - em->start;
1668                 BUG_ON(em->end < cur);
1669                 BUG_ON(end < cur);
1670                 iosize = min(em->end - cur, end - cur) + 1;
1671                 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
1672                 sector = (em->block_start + extent_offset) >> 9;
1673                 bdev = em->bdev;
1674                 block_start = em->block_start;
1675                 free_extent_map(em);
1676                 em = NULL;
1677
1678                 if (block_start == EXTENT_MAP_HOLE ||
1679                     block_start == EXTENT_MAP_INLINE) {
1680                         clear_extent_dirty(tree, cur,
1681                                            cur + iosize - 1, GFP_NOFS);
1682                         cur = cur + iosize;
1683                         page_offset += iosize;
1684                         continue;
1685                 }
1686
1687                 /* leave this out until we have a page_mkwrite call */
1688                 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
1689                                    EXTENT_DIRTY, 0)) {
1690                         cur = cur + iosize;
1691                         page_offset += iosize;
1692                         continue;
1693                 }
1694                 clear_extent_dirty(tree, cur, cur + iosize - 1, GFP_NOFS);
1695                 if (tree->ops && tree->ops->writepage_io_hook) {
1696                         ret = tree->ops->writepage_io_hook(page, cur,
1697                                                 cur + iosize - 1);
1698                 } else {
1699                         ret = 0;
1700                 }
1701                 if (ret)
1702                         SetPageError(page);
1703                 else {
1704                         set_range_writeback(tree, cur, cur + iosize - 1);
1705                         ret = submit_extent_page(WRITE, tree, page, sector,
1706                                                  iosize, page_offset, bdev,
1707                                                  end_bio_extent_writepage);
1708                         if (ret)
1709                                 SetPageError(page);
1710                 }
1711                 cur = cur + iosize;
1712                 page_offset += iosize;
1713                 nr++;
1714         }
1715 done:
1716         unlock_extent(tree, start, page_end, GFP_NOFS);
1717         unlock_page(page);
1718         return 0;
1719 }
1720 EXPORT_SYMBOL(extent_write_full_page);
1721
1722 /*
1723  * basic invalidatepage code, this waits on any locked or writeback
1724  * ranges corresponding to the page, and then deletes any extent state
1725  * records from the tree
1726  */
1727 int extent_invalidatepage(struct extent_map_tree *tree,
1728                           struct page *page, unsigned long offset)
1729 {
1730         u64 start = (page->index << PAGE_CACHE_SHIFT);
1731         u64 end = start + PAGE_CACHE_SIZE - 1;
1732         size_t blocksize = page->mapping->host->i_sb->s_blocksize;
1733
1734         start += (offset + blocksize -1) & ~(blocksize - 1);
1735         if (start > end)
1736                 return 0;
1737
1738         lock_extent(tree, start, end, GFP_NOFS);
1739         wait_on_extent_writeback(tree, start, end);
1740         clear_extent_bit(tree, start, end,
1741                          EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC,
1742                          1, 1, GFP_NOFS);
1743         return 0;
1744 }
1745 EXPORT_SYMBOL(extent_invalidatepage);
1746
1747 /*
1748  * simple commit_write call, set_range_dirty is used to mark both
1749  * the pages and the extent records as dirty
1750  */
1751 int extent_commit_write(struct extent_map_tree *tree,
1752                         struct inode *inode, struct page *page,
1753                         unsigned from, unsigned to)
1754 {
1755         loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1756
1757         set_page_extent_mapped(page);
1758         set_page_dirty(page);
1759
1760         if (pos > inode->i_size) {
1761                 i_size_write(inode, pos);
1762                 mark_inode_dirty(inode);
1763         }
1764         return 0;
1765 }
1766 EXPORT_SYMBOL(extent_commit_write);
1767
1768 int extent_prepare_write(struct extent_map_tree *tree,
1769                          struct inode *inode, struct page *page,
1770                          unsigned from, unsigned to, get_extent_t *get_extent)
1771 {
1772         u64 page_start = page->index << PAGE_CACHE_SHIFT;
1773         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
1774         u64 block_start;
1775         u64 orig_block_start;
1776         u64 block_end;
1777         u64 cur_end;
1778         struct extent_map *em;
1779         unsigned blocksize = 1 << inode->i_blkbits;
1780         size_t page_offset = 0;
1781         size_t block_off_start;
1782         size_t block_off_end;
1783         int err = 0;
1784         int iocount = 0;
1785         int ret = 0;
1786         int isnew;
1787
1788         set_page_extent_mapped(page);
1789
1790         block_start = (page_start + from) & ~((u64)blocksize - 1);
1791         block_end = (page_start + to - 1) | (blocksize - 1);
1792         orig_block_start = block_start;
1793
1794         lock_extent(tree, page_start, page_end, GFP_NOFS);
1795         while(block_start <= block_end) {
1796                 em = get_extent(inode, page, page_offset, block_start,
1797                                 block_end, 1);
1798                 if (IS_ERR(em) || !em) {
1799                         goto err;
1800                 }
1801                 cur_end = min(block_end, em->end);
1802                 block_off_start = block_start & (PAGE_CACHE_SIZE - 1);
1803                 block_off_end = block_off_start + blocksize;
1804                 isnew = clear_extent_new(tree, block_start, cur_end, GFP_NOFS);
1805
1806                 if (!PageUptodate(page) && isnew &&
1807                     (block_off_end > to || block_off_start < from)) {
1808                         void *kaddr;
1809
1810                         kaddr = kmap_atomic(page, KM_USER0);
1811                         if (block_off_end > to)
1812                                 memset(kaddr + to, 0, block_off_end - to);
1813                         if (block_off_start < from)
1814                                 memset(kaddr + block_off_start, 0,
1815                                        from - block_off_start);
1816                         flush_dcache_page(page);
1817                         kunmap_atomic(kaddr, KM_USER0);
1818                 }
1819                 if (!isnew && !PageUptodate(page) &&
1820                     (block_off_end > to || block_off_start < from) &&
1821                     !test_range_bit(tree, block_start, cur_end,
1822                                     EXTENT_UPTODATE, 1)) {
1823                         u64 sector;
1824                         u64 extent_offset = block_start - em->start;
1825                         size_t iosize;
1826                         sector = (em->block_start + extent_offset) >> 9;
1827                         iosize = (cur_end - block_start + blocksize - 1) &
1828                                 ~((u64)blocksize - 1);
1829                         /*
1830                          * we've already got the extent locked, but we
1831                          * need to split the state such that our end_bio
1832                          * handler can clear the lock.
1833                          */
1834                         set_extent_bit(tree, block_start,
1835                                        block_start + iosize - 1,
1836                                        EXTENT_LOCKED, 0, NULL, GFP_NOFS);
1837                         ret = submit_extent_page(READ, tree, page,
1838                                          sector, iosize, page_offset, em->bdev,
1839                                          end_bio_extent_preparewrite);
1840                         iocount++;
1841                         block_start = block_start + iosize;
1842                 } else {
1843                         set_extent_uptodate(tree, block_start, cur_end,
1844                                             GFP_NOFS);
1845                         unlock_extent(tree, block_start, cur_end, GFP_NOFS);
1846                         block_start = cur_end + 1;
1847                 }
1848                 page_offset = block_start & (PAGE_CACHE_SIZE - 1);
1849                 free_extent_map(em);
1850         }
1851         if (iocount) {
1852                 wait_extent_bit(tree, orig_block_start,
1853                                 block_end, EXTENT_LOCKED);
1854         }
1855         check_page_uptodate(tree, page);
1856 err:
1857         /* FIXME, zero out newly allocated blocks on error */
1858         return err;
1859 }
1860 EXPORT_SYMBOL(extent_prepare_write);
1861
1862 /*
1863  * a helper for releasepage.  As long as there are no locked extents
1864  * in the range corresponding to the page, both state records and extent
1865  * map records are removed
1866  */
1867 int try_release_extent_mapping(struct extent_map_tree *tree, struct page *page)
1868 {
1869         struct extent_map *em;
1870         u64 start = page->index << PAGE_CACHE_SHIFT;
1871         u64 end = start + PAGE_CACHE_SIZE - 1;
1872         u64 orig_start = start;
1873         int ret = 1;
1874
1875         while (start <= end) {
1876                 em = lookup_extent_mapping(tree, start, end);
1877                 if (!em || IS_ERR(em))
1878                         break;
1879                 if (!test_range_bit(tree, em->start, em->end,
1880                                     EXTENT_LOCKED, 0)) {
1881                         remove_extent_mapping(tree, em);
1882                         /* once for the rb tree */
1883                         free_extent_map(em);
1884                 }
1885                 start = em->end + 1;
1886                 /* once for us */
1887                 free_extent_map(em);
1888         }
1889         if (test_range_bit(tree, orig_start, end, EXTENT_LOCKED, 0))
1890                 ret = 0;
1891         else
1892                 clear_extent_bit(tree, orig_start, end, EXTENT_UPTODATE,
1893                                  1, 1, GFP_NOFS);
1894         return ret;
1895 }
1896 EXPORT_SYMBOL(try_release_extent_mapping);
1897
1898 sector_t extent_bmap(struct address_space *mapping, sector_t iblock,
1899                 get_extent_t *get_extent)
1900 {
1901         struct inode *inode = mapping->host;
1902         u64 start = iblock << inode->i_blkbits;
1903         u64 end = start + (1 << inode->i_blkbits) - 1;
1904         struct extent_map *em;
1905
1906         em = get_extent(inode, NULL, 0, start, end, 0);
1907         if (!em || IS_ERR(em))
1908                 return 0;
1909
1910         if (em->block_start == EXTENT_MAP_INLINE ||
1911             em->block_start == EXTENT_MAP_HOLE)
1912                 return 0;
1913
1914         return (em->block_start + start - em->start) >> inode->i_blkbits;
1915 }
1916
1917 static struct extent_buffer *__alloc_extent_buffer(gfp_t mask)
1918 {
1919         struct extent_buffer *eb = NULL;
1920
1921         spin_lock(&extent_buffers_lock);
1922         if (!list_empty(&extent_buffers)) {
1923                 eb = list_entry(extent_buffers.next, struct extent_buffer,
1924                                 list);
1925                 list_del(&eb->list);
1926                 WARN_ON(nr_extent_buffers == 0);
1927                 nr_extent_buffers--;
1928         }
1929         spin_unlock(&extent_buffers_lock);
1930
1931         if (eb) {
1932                 memset(eb, 0, sizeof(*eb));
1933         } else {
1934                 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
1935         }
1936         spin_lock(&extent_buffers_lock);
1937         list_add(&eb->leak_list, &buffers);
1938         spin_unlock(&extent_buffers_lock);
1939
1940         return eb;
1941 }
1942
1943 static void __free_extent_buffer(struct extent_buffer *eb)
1944 {
1945
1946         spin_lock(&extent_buffers_lock);
1947         list_del_init(&eb->leak_list);
1948         spin_unlock(&extent_buffers_lock);
1949
1950         if (nr_extent_buffers >= MAX_EXTENT_BUFFER_CACHE) {
1951                 kmem_cache_free(extent_buffer_cache, eb);
1952         } else {
1953                 spin_lock(&extent_buffers_lock);
1954                 list_add(&eb->list, &extent_buffers);
1955                 nr_extent_buffers++;
1956                 spin_unlock(&extent_buffers_lock);
1957         }
1958 }
1959
1960 static inline struct page *extent_buffer_page(struct extent_buffer *eb, int i)
1961 {
1962         struct page *p;
1963         if (i == 0)
1964                 return eb->first_page;
1965         i += eb->start >> PAGE_CACHE_SHIFT;
1966         p = find_get_page(eb->first_page->mapping, i);
1967         page_cache_release(p);
1968         return p;
1969 }
1970
1971 struct extent_buffer *alloc_extent_buffer(struct extent_map_tree *tree,
1972                                           u64 start, unsigned long len,
1973                                           gfp_t mask)
1974 {
1975         unsigned long num_pages = ((start + len - 1) >> PAGE_CACHE_SHIFT) -
1976                                   (start >> PAGE_CACHE_SHIFT) + 1;
1977         unsigned long i;
1978         unsigned long index = start >> PAGE_CACHE_SHIFT;
1979         struct extent_buffer *eb;
1980         struct page *p;
1981         struct address_space *mapping = tree->mapping;
1982         int uptodate = 0;
1983
1984         eb = __alloc_extent_buffer(mask);
1985         if (!eb || IS_ERR(eb))
1986                 return NULL;
1987
1988         eb->alloc_addr = __builtin_return_address(0);
1989         eb->start = start;
1990         eb->len = len;
1991         atomic_set(&eb->refs, 1);
1992
1993         for (i = 0; i < num_pages; i++, index++) {
1994                 p = find_or_create_page(mapping, index, mask | __GFP_HIGHMEM);
1995                 if (!p) {
1996                         /* make sure the free only frees the pages we've
1997                          * grabbed a reference on
1998                          */
1999                         eb->len = i << PAGE_CACHE_SHIFT;
2000                         eb->start &= ~((u64)PAGE_CACHE_SIZE - 1);
2001                         goto fail;
2002                 }
2003                 set_page_extent_mapped(p);
2004                 if (i == 0)
2005                         eb->first_page = p;
2006                 if (!PageUptodate(p))
2007                         uptodate = 0;
2008                 unlock_page(p);
2009         }
2010         if (uptodate)
2011                 eb->flags |= EXTENT_UPTODATE;
2012         return eb;
2013 fail:
2014         free_extent_buffer(eb);
2015         return NULL;
2016 }
2017 EXPORT_SYMBOL(alloc_extent_buffer);
2018
2019 struct extent_buffer *find_extent_buffer(struct extent_map_tree *tree,
2020                                          u64 start, unsigned long len,
2021                                           gfp_t mask)
2022 {
2023         unsigned long num_pages = ((start + len - 1) >> PAGE_CACHE_SHIFT) -
2024                                   (start >> PAGE_CACHE_SHIFT) + 1;
2025         unsigned long i;
2026         unsigned long index = start >> PAGE_CACHE_SHIFT;
2027         struct extent_buffer *eb;
2028         struct page *p;
2029         struct address_space *mapping = tree->mapping;
2030
2031         eb = __alloc_extent_buffer(mask);
2032         if (!eb || IS_ERR(eb))
2033                 return NULL;
2034
2035         eb->alloc_addr = __builtin_return_address(0);
2036         eb->start = start;
2037         eb->len = len;
2038         atomic_set(&eb->refs, 1);
2039
2040         for (i = 0; i < num_pages; i++, index++) {
2041                 p = find_get_page(mapping, index);
2042                 if (!p) {
2043                         /* make sure the free only frees the pages we've
2044                          * grabbed a reference on
2045                          */
2046                         eb->len = i << PAGE_CACHE_SHIFT;
2047                         eb->start &= ~((u64)PAGE_CACHE_SIZE - 1);
2048                         goto fail;
2049                 }
2050                 set_page_extent_mapped(p);
2051                 if (i == 0)
2052                         eb->first_page = p;
2053         }
2054         return eb;
2055 fail:
2056         free_extent_buffer(eb);
2057         return NULL;
2058 }
2059 EXPORT_SYMBOL(find_extent_buffer);
2060
2061 void free_extent_buffer(struct extent_buffer *eb)
2062 {
2063         unsigned long i;
2064         unsigned long num_pages;
2065
2066         if (!eb)
2067                 return;
2068
2069         if (!atomic_dec_and_test(&eb->refs))
2070                 return;
2071
2072         num_pages = ((eb->start + eb->len - 1) >> PAGE_CACHE_SHIFT) -
2073                 (eb->start >> PAGE_CACHE_SHIFT) + 1;
2074
2075         if (eb->first_page)
2076                 page_cache_release(eb->first_page);
2077         for (i = 1; i < num_pages; i++) {
2078                 page_cache_release(extent_buffer_page(eb, i));
2079         }
2080         __free_extent_buffer(eb);
2081 }
2082 EXPORT_SYMBOL(free_extent_buffer);
2083
2084 int clear_extent_buffer_dirty(struct extent_map_tree *tree,
2085                               struct extent_buffer *eb)
2086 {
2087         int set;
2088         unsigned long i;
2089         unsigned long num_pages;
2090         struct page *page;
2091
2092         u64 start = eb->start;
2093         u64 end = start + eb->len - 1;
2094
2095         set = clear_extent_dirty(tree, start, end, GFP_NOFS);
2096         num_pages = ((eb->start + eb->len - 1) >> PAGE_CACHE_SHIFT) -
2097                 (eb->start >> PAGE_CACHE_SHIFT) + 1;
2098
2099         for (i = 0; i < num_pages; i++) {
2100                 page = extent_buffer_page(eb, i);
2101                 lock_page(page);
2102                 /*
2103                  * if we're on the last page or the first page and the
2104                  * block isn't aligned on a page boundary, do extra checks
2105                  * to make sure we don't clean page that is partially dirty
2106                  */
2107                 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
2108                     ((i == num_pages - 1) &&
2109                      ((eb->start + eb->len - 1) & (PAGE_CACHE_SIZE - 1)))) {
2110                         start = page->index << PAGE_CACHE_SHIFT;
2111                         end  = start + PAGE_CACHE_SIZE - 1;
2112                         if (test_range_bit(tree, start, end,
2113                                            EXTENT_DIRTY, 0)) {
2114                                 unlock_page(page);
2115                                 continue;
2116                         }
2117                 }
2118                 clear_page_dirty_for_io(page);
2119                 unlock_page(page);
2120         }
2121         return 0;
2122 }
2123 EXPORT_SYMBOL(clear_extent_buffer_dirty);
2124
2125 int wait_on_extent_buffer_writeback(struct extent_map_tree *tree,
2126                                     struct extent_buffer *eb)
2127 {
2128         return wait_on_extent_writeback(tree, eb->start,
2129                                         eb->start + eb->len - 1);
2130 }
2131 EXPORT_SYMBOL(wait_on_extent_buffer_writeback);
2132
2133 int set_extent_buffer_dirty(struct extent_map_tree *tree,
2134                              struct extent_buffer *eb)
2135 {
2136         return set_range_dirty(tree, eb->start, eb->start + eb->len - 1);
2137 }
2138 EXPORT_SYMBOL(set_extent_buffer_dirty);
2139
2140 int set_extent_buffer_uptodate(struct extent_map_tree *tree,
2141                                 struct extent_buffer *eb)
2142 {
2143         unsigned long i;
2144         struct page *page;
2145         unsigned long num_pages;
2146
2147         num_pages = ((eb->start + eb->len - 1) >> PAGE_CACHE_SHIFT) -
2148                 (eb->start >> PAGE_CACHE_SHIFT) + 1;
2149
2150         set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
2151                             GFP_NOFS);
2152         for (i = 0; i < num_pages; i++) {
2153                 page = extent_buffer_page(eb, i);
2154                 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
2155                     ((i == num_pages - 1) &&
2156                      ((eb->start + eb->len - 1) & (PAGE_CACHE_SIZE - 1)))) {
2157                         check_page_uptodate(tree, page);
2158                         continue;
2159                 }
2160                 SetPageUptodate(page);
2161         }
2162         return 0;
2163 }
2164 EXPORT_SYMBOL(set_extent_buffer_uptodate);
2165
2166 int extent_buffer_uptodate(struct extent_map_tree *tree,
2167                              struct extent_buffer *eb)
2168 {
2169         if (eb->flags & EXTENT_UPTODATE)
2170                 return 1;
2171         return test_range_bit(tree, eb->start, eb->start + eb->len - 1,
2172                            EXTENT_UPTODATE, 1);
2173 }
2174 EXPORT_SYMBOL(extent_buffer_uptodate);
2175
2176 int read_extent_buffer_pages(struct extent_map_tree *tree,
2177                              struct extent_buffer *eb, int wait)
2178 {
2179         unsigned long i;
2180         struct page *page;
2181         int err;
2182         int ret = 0;
2183         unsigned long num_pages;
2184
2185         if (eb->flags & EXTENT_UPTODATE)
2186                 return 0;
2187
2188         if (test_range_bit(tree, eb->start, eb->start + eb->len - 1,
2189                            EXTENT_UPTODATE, 1)) {
2190                 return 0;
2191         }
2192
2193         num_pages = ((eb->start + eb->len - 1) >> PAGE_CACHE_SHIFT) -
2194                 (eb->start >> PAGE_CACHE_SHIFT) + 1;
2195         for (i = 0; i < num_pages; i++) {
2196                 page = extent_buffer_page(eb, i);
2197                 if (PageUptodate(page)) {
2198                         continue;
2199                 }
2200                 if (!wait) {
2201                         if (TestSetPageLocked(page)) {
2202                                 continue;
2203                         }
2204                 } else {
2205                         lock_page(page);
2206                 }
2207                 if (!PageUptodate(page)) {
2208                         err = page->mapping->a_ops->readpage(NULL, page);
2209                         if (err) {
2210                                 ret = err;
2211                         }
2212                 } else {
2213                         unlock_page(page);
2214                 }
2215         }
2216
2217         if (ret || !wait) {
2218                 return ret;
2219         }
2220
2221         for (i = 0; i < num_pages; i++) {
2222                 page = extent_buffer_page(eb, i);
2223                 wait_on_page_locked(page);
2224                 if (!PageUptodate(page)) {
2225                         ret = -EIO;
2226                 }
2227         }
2228         eb->flags |= EXTENT_UPTODATE;
2229         return ret;
2230 }
2231 EXPORT_SYMBOL(read_extent_buffer_pages);
2232
2233 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
2234                         unsigned long start,
2235                         unsigned long len)
2236 {
2237         size_t cur;
2238         size_t offset;
2239         struct page *page;
2240         char *kaddr;
2241         char *dst = (char *)dstv;
2242         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
2243         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
2244
2245         WARN_ON(start > eb->len);
2246         WARN_ON(start + len > eb->start + eb->len);
2247
2248         offset = start & ((unsigned long)PAGE_CACHE_SIZE - 1);
2249         if (i == 0)
2250                 offset += start_offset;
2251
2252         while(len > 0) {
2253                 page = extent_buffer_page(eb, i);
2254                 WARN_ON(!PageUptodate(page));
2255
2256                 cur = min(len, (PAGE_CACHE_SIZE - offset));
2257                 kaddr = kmap_atomic(page, KM_USER0);
2258                 memcpy(dst, kaddr + offset, cur);
2259                 kunmap_atomic(kaddr, KM_USER0);
2260
2261                 dst += cur;
2262                 len -= cur;
2263                 offset = 0;
2264                 i++;
2265         }
2266 }
2267 EXPORT_SYMBOL(read_extent_buffer);
2268
2269 int map_extent_buffer(struct extent_buffer *eb, unsigned long start,
2270                       unsigned long min_len,
2271                       char **token, char **map,
2272                       unsigned long *map_start,
2273                       unsigned long *map_len, int km)
2274 {
2275         size_t offset = start & (PAGE_CACHE_SIZE - 1);
2276         char *kaddr;
2277         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
2278         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
2279         unsigned long end_i = (start_offset + start + min_len) >>
2280                                 PAGE_CACHE_SHIFT;
2281
2282         if (i != end_i)
2283                 return -EINVAL;
2284
2285         WARN_ON(start > eb->len);
2286
2287         if (i == 0) {
2288                 offset = start_offset;
2289                 *map_start = 0;
2290         } else {
2291                 *map_start = (i << PAGE_CACHE_SHIFT) - start_offset;
2292         }
2293
2294         kaddr = kmap_atomic(extent_buffer_page(eb, i), km);
2295         *token = kaddr;
2296         *map = kaddr + offset;
2297         *map_len = PAGE_CACHE_SIZE - offset;
2298         return 0;
2299 }
2300 EXPORT_SYMBOL(map_extent_buffer);
2301
2302 void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km)
2303 {
2304         kunmap_atomic(token, km);
2305 }
2306 EXPORT_SYMBOL(unmap_extent_buffer);
2307
2308 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
2309                           unsigned long start,
2310                           unsigned long len)
2311 {
2312         size_t cur;
2313         size_t offset;
2314         struct page *page;
2315         char *kaddr;
2316         char *ptr = (char *)ptrv;
2317         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
2318         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
2319         int ret = 0;
2320
2321         WARN_ON(start > eb->len);
2322         WARN_ON(start + len > eb->start + eb->len);
2323
2324         offset = start & ((unsigned long)PAGE_CACHE_SIZE - 1);
2325         if (i == 0)
2326                 offset += start_offset;
2327
2328         while(len > 0) {
2329                 page = extent_buffer_page(eb, i);
2330                 WARN_ON(!PageUptodate(page));
2331
2332                 cur = min(len, (PAGE_CACHE_SIZE - offset));
2333
2334                 kaddr = kmap_atomic(page, KM_USER0);
2335                 ret = memcmp(ptr, kaddr + offset, cur);
2336                 kunmap_atomic(kaddr, KM_USER0);
2337                 if (ret)
2338                         break;
2339
2340                 ptr += cur;
2341                 len -= cur;
2342                 offset = 0;
2343                 i++;
2344         }
2345         return ret;
2346 }
2347 EXPORT_SYMBOL(memcmp_extent_buffer);
2348
2349 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
2350                          unsigned long start, unsigned long len)
2351 {
2352         size_t cur;
2353         size_t offset;
2354         struct page *page;
2355         char *kaddr;
2356         char *src = (char *)srcv;
2357         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
2358         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
2359
2360         WARN_ON(start > eb->len);
2361         WARN_ON(start + len > eb->start + eb->len);
2362
2363         offset = start & ((unsigned long)PAGE_CACHE_SIZE - 1);
2364         if (i == 0)
2365                 offset += start_offset;
2366
2367         while(len > 0) {
2368                 page = extent_buffer_page(eb, i);
2369                 WARN_ON(!PageUptodate(page));
2370
2371                 cur = min(len, PAGE_CACHE_SIZE - offset);
2372                 kaddr = kmap_atomic(page, KM_USER0);
2373                 memcpy(kaddr + offset, src, cur);
2374                 kunmap_atomic(kaddr, KM_USER0);
2375
2376                 src += cur;
2377                 len -= cur;
2378                 offset = 0;
2379                 i++;
2380         }
2381 }
2382 EXPORT_SYMBOL(write_extent_buffer);
2383
2384 void memset_extent_buffer(struct extent_buffer *eb, char c,
2385                           unsigned long start, unsigned long len)
2386 {
2387         size_t cur;
2388         size_t offset;
2389         struct page *page;
2390         char *kaddr;
2391         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
2392         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
2393
2394         WARN_ON(start > eb->len);
2395         WARN_ON(start + len > eb->start + eb->len);
2396
2397         offset = start & ((unsigned long)PAGE_CACHE_SIZE - 1);
2398         if (i == 0)
2399                 offset += start_offset;
2400
2401         while(len > 0) {
2402                 page = extent_buffer_page(eb, i);
2403                 WARN_ON(!PageUptodate(page));
2404
2405                 cur = min(len, PAGE_CACHE_SIZE - offset);
2406                 kaddr = kmap_atomic(page, KM_USER0);
2407                 memset(kaddr + offset, c, cur);
2408                 kunmap_atomic(kaddr, KM_USER0);
2409
2410                 len -= cur;
2411                 offset = 0;
2412                 i++;
2413         }
2414 }
2415 EXPORT_SYMBOL(memset_extent_buffer);
2416
2417 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
2418                         unsigned long dst_offset, unsigned long src_offset,
2419                         unsigned long len)
2420 {
2421         u64 dst_len = dst->len;
2422         size_t cur;
2423         size_t offset;
2424         struct page *page;
2425         char *kaddr;
2426         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
2427         unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
2428
2429         WARN_ON(src->len != dst_len);
2430
2431         offset = dst_offset & ((unsigned long)PAGE_CACHE_SIZE - 1);
2432         if (i == 0)
2433                 offset += start_offset;
2434
2435         while(len > 0) {
2436                 page = extent_buffer_page(dst, i);
2437                 WARN_ON(!PageUptodate(page));
2438
2439                 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
2440
2441                 kaddr = kmap_atomic(page, KM_USER1);
2442                 read_extent_buffer(src, kaddr + offset, src_offset, cur);
2443                 kunmap_atomic(kaddr, KM_USER1);
2444
2445                 src_offset += cur;
2446                 len -= cur;
2447                 offset = 0;
2448                 i++;
2449         }
2450 }
2451 EXPORT_SYMBOL(copy_extent_buffer);
2452
2453 static void move_pages(struct page *dst_page, struct page *src_page,
2454                        unsigned long dst_off, unsigned long src_off,
2455                        unsigned long len)
2456 {
2457         char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
2458         if (dst_page == src_page) {
2459                 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
2460         } else {
2461                 char *src_kaddr = kmap_atomic(src_page, KM_USER1);
2462                 char *p = dst_kaddr + dst_off + len;
2463                 char *s = src_kaddr + src_off + len;
2464
2465                 while (len--)
2466                         *--p = *--s;
2467
2468                 kunmap_atomic(src_kaddr, KM_USER1);
2469         }
2470         kunmap_atomic(dst_kaddr, KM_USER0);
2471 }
2472
2473 static void copy_pages(struct page *dst_page, struct page *src_page,
2474                        unsigned long dst_off, unsigned long src_off,
2475                        unsigned long len)
2476 {
2477         char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
2478         char *src_kaddr;
2479
2480         if (dst_page != src_page)
2481                 src_kaddr = kmap_atomic(src_page, KM_USER1);
2482         else
2483                 src_kaddr = dst_kaddr;
2484
2485         memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
2486         kunmap_atomic(dst_kaddr, KM_USER0);
2487         if (dst_page != src_page)
2488                 kunmap_atomic(src_kaddr, KM_USER1);
2489 }
2490
2491 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
2492                            unsigned long src_offset, unsigned long len)
2493 {
2494         size_t cur;
2495         size_t dst_off_in_page;
2496         size_t src_off_in_page;
2497         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
2498         unsigned long dst_i;
2499         unsigned long src_i;
2500
2501         if (src_offset + len > dst->len) {
2502                 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
2503                        src_offset, len, dst->len);
2504                 BUG_ON(1);
2505         }
2506         if (dst_offset + len > dst->len) {
2507                 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
2508                        dst_offset, len, dst->len);
2509                 BUG_ON(1);
2510         }
2511
2512         while(len > 0) {
2513                 dst_off_in_page = dst_offset &
2514                         ((unsigned long)PAGE_CACHE_SIZE - 1);
2515                 src_off_in_page = src_offset &
2516                         ((unsigned long)PAGE_CACHE_SIZE - 1);
2517
2518                 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
2519                 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
2520
2521                 if (src_i == 0)
2522                         src_off_in_page += start_offset;
2523                 if (dst_i == 0)
2524                         dst_off_in_page += start_offset;
2525
2526                 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
2527                                                src_off_in_page));
2528                 cur = min(cur, (unsigned long)(PAGE_CACHE_SIZE -
2529                                                dst_off_in_page));
2530
2531                 copy_pages(extent_buffer_page(dst, dst_i),
2532                            extent_buffer_page(dst, src_i),
2533                            dst_off_in_page, src_off_in_page, cur);
2534
2535                 src_offset += cur;
2536                 dst_offset += cur;
2537                 len -= cur;
2538         }
2539 }
2540 EXPORT_SYMBOL(memcpy_extent_buffer);
2541
2542 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
2543                            unsigned long src_offset, unsigned long len)
2544 {
2545         size_t cur;
2546         size_t dst_off_in_page;
2547         size_t src_off_in_page;
2548         unsigned long dst_end = dst_offset + len - 1;
2549         unsigned long src_end = src_offset + len - 1;
2550         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
2551         unsigned long dst_i;
2552         unsigned long src_i;
2553
2554         if (src_offset + len > dst->len) {
2555                 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
2556                        src_offset, len, dst->len);
2557                 BUG_ON(1);
2558         }
2559         if (dst_offset + len > dst->len) {
2560                 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
2561                        dst_offset, len, dst->len);
2562                 BUG_ON(1);
2563         }
2564         if (dst_offset < src_offset) {
2565                 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
2566                 return;
2567         }
2568         while(len > 0) {
2569                 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
2570                 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
2571
2572                 dst_off_in_page = dst_end &
2573                         ((unsigned long)PAGE_CACHE_SIZE - 1);
2574                 src_off_in_page = src_end &
2575                         ((unsigned long)PAGE_CACHE_SIZE - 1);
2576
2577                 if (src_i == 0)
2578                         src_off_in_page += start_offset;
2579                 if (dst_i == 0)
2580                         dst_off_in_page += start_offset;
2581
2582                 cur = min(len, src_off_in_page + 1);
2583                 cur = min(cur, dst_off_in_page + 1);
2584
2585                 move_pages(extent_buffer_page(dst, dst_i),
2586                            extent_buffer_page(dst, src_i),
2587                            dst_off_in_page - cur + 1,
2588                            src_off_in_page - cur + 1, cur);
2589
2590                 dst_end -= cur - 1;
2591                 src_end -= cur - 1;
2592                 len -= cur;
2593         }
2594 }
2595 EXPORT_SYMBOL(memmove_extent_buffer);