fs/btrfs/extent_io.c

   1 #include <linux/bitops.h>
   2 #include <linux/slab.h>
   3 #include <linux/bio.h>
   4 #include <linux/mm.h>
   5 #include <linux/gfp.h>
   6 #include <linux/pagemap.h>
   7 #include <linux/page-flags.h>
   8 #include <linux/module.h>
   9 #include <linux/spinlock.h>
  10 #include <linux/blkdev.h>
  11 #include <linux/swap.h>
  12 #include <linux/writeback.h>
  13 #include <linux/pagevec.h>
  14 #include "extent_io.h"
  15 #include "extent_map.h"
  16 #include "compat.h"
  17 #include "ctree.h"
  18 #include "btrfs_inode.h"
  19
  20 static struct kmem_cache *extent_state_cache;
  21 static struct kmem_cache *extent_buffer_cache;
  22
  23 static LIST_HEAD(buffers);
  24 static LIST_HEAD(states);
  25
  26 #define LEAK_DEBUG 0
  27 #if LEAK_DEBUG
  28 static DEFINE_SPINLOCK(leak_lock);
  29 #endif
  30
  31 #define BUFFER_LRU_MAX 64
  32
  33 struct tree_entry {
  34         u64 start;
  35         u64 end;
  36         struct rb_node rb_node;
  37 };
  38
  39 struct extent_page_data {
  40         struct bio *bio;
  41         struct extent_io_tree *tree;
  42         get_extent_t *get_extent;
  43
  44         /* tells writepage not to lock the state bits for this range
  45          * it still does the unlocking
  46          */
  47         unsigned int extent_locked:1;
  48
  49         /* tells the submit_bio code to use a WRITE_SYNC */
  50         unsigned int sync_io:1;
  51 };
  52
  53 int __init extent_io_init(void)
  54 {
  55         extent_state_cache = kmem_cache_create("extent_state",
  56                         sizeof(struct extent_state), 0,
  57                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
  58         if (!extent_state_cache)
  59                 return -ENOMEM;
  60
  61         extent_buffer_cache = kmem_cache_create("extent_buffers",
  62                         sizeof(struct extent_buffer), 0,
  63                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
  64         if (!extent_buffer_cache)
  65                 goto free_state_cache;
  66         return 0;
  67
  68 free_state_cache:
  69         kmem_cache_destroy(extent_state_cache);
  70         return -ENOMEM;
  71 }
  72
  73 void extent_io_exit(void)
  74 {
  75         struct extent_state *state;
  76         struct extent_buffer *eb;
  77
  78         while (!list_empty(&states)) {
  79                 state = list_entry(states.next, struct extent_state, leak_list);
  80                 printk(KERN_ERR "btrfs state leak: start %llu end %llu "
  81                        "state %lu in tree %p refs %d\n",
  82                        (unsigned long long)state->start,
  83                        (unsigned long long)state->end,
  84                        state->state, state->tree, atomic_read(&state->refs));
  85                 list_del(&state->leak_list);
  86                 kmem_cache_free(extent_state_cache, state);
  87
  88         }
  89
  90         while (!list_empty(&buffers)) {
  91                 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
  92                 printk(KERN_ERR "btrfs buffer leak start %llu len %lu "
  93                        "refs %d\n", (unsigned long long)eb->start,
  94                        eb->len, atomic_read(&eb->refs));
  95                 list_del(&eb->leak_list);
  96                 kmem_cache_free(extent_buffer_cache, eb);
  97         }
  98         if (extent_state_cache)
  99                 kmem_cache_destroy(extent_state_cache);
 100         if (extent_buffer_cache)
 101                 kmem_cache_destroy(extent_buffer_cache);
 102 }
 103
 104 void extent_io_tree_init(struct extent_io_tree *tree,
 105                           struct address_space *mapping, gfp_t mask)
 106 {
 107         tree->state.rb_node = NULL;
 108         tree->buffer.rb_node = NULL;
 109         tree->ops = NULL;
 110         tree->dirty_bytes = 0;
 111         spin_lock_init(&tree->lock);
 112         spin_lock_init(&tree->buffer_lock);
 113         tree->mapping = mapping;
 114 }
 115
 116 static struct extent_state *alloc_extent_state(gfp_t mask)
 117 {
 118         struct extent_state *state;
 119 #if LEAK_DEBUG
 120         unsigned long flags;
 121 #endif
 122
 123         state = kmem_cache_alloc(extent_state_cache, mask);
 124         if (!state)
 125                 return state;
 126         state->state = 0;
 127         state->private = 0;
 128         state->tree = NULL;
 129 #if LEAK_DEBUG
 130         spin_lock_irqsave(&leak_lock, flags);
 131         list_add(&state->leak_list, &states);
 132         spin_unlock_irqrestore(&leak_lock, flags);
 133 #endif
 134         atomic_set(&state->refs, 1);
 135         init_waitqueue_head(&state->wq);
 136         return state;
 137 }
 138
 139 static void free_extent_state(struct extent_state *state)
 140 {
 141         if (!state)
 142                 return;
 143         if (atomic_dec_and_test(&state->refs)) {
 144 #if LEAK_DEBUG
 145                 unsigned long flags;
 146 #endif
 147                 WARN_ON(state->tree);
 148 #if LEAK_DEBUG
 149                 spin_lock_irqsave(&leak_lock, flags);
 150                 list_del(&state->leak_list);
 151                 spin_unlock_irqrestore(&leak_lock, flags);
 152 #endif
 153                 kmem_cache_free(extent_state_cache, state);
 154         }
 155 }
 156
 157 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
 158                                    struct rb_node *node)
 159 {
 160         struct rb_node **p = &root->rb_node;
 161         struct rb_node *parent = NULL;
 162         struct tree_entry *entry;
 163
 164         while (*p) {
 165                 parent = *p;
 166                 entry = rb_entry(parent, struct tree_entry, rb_node);
 167
 168                 if (offset < entry->start)
 169                         p = &(*p)->rb_left;
 170                 else if (offset > entry->end)
 171                         p = &(*p)->rb_right;
 172                 else
 173                         return parent;
 174         }
 175
 176         entry = rb_entry(node, struct tree_entry, rb_node);
 177         rb_link_node(node, parent, p);
 178         rb_insert_color(node, root);
 179         return NULL;
 180 }
 181
 182 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
 183                                      struct rb_node **prev_ret,
 184                                      struct rb_node **next_ret)
 185 {
 186         struct rb_root *root = &tree->state;
 187         struct rb_node *n = root->rb_node;
 188         struct rb_node *prev = NULL;
 189         struct rb_node *orig_prev = NULL;
 190         struct tree_entry *entry;
 191         struct tree_entry *prev_entry = NULL;
 192
 193         while (n) {
 194                 entry = rb_entry(n, struct tree_entry, rb_node);
 195                 prev = n;
 196                 prev_entry = entry;
 197
 198                 if (offset < entry->start)
 199                         n = n->rb_left;
 200                 else if (offset > entry->end)
 201                         n = n->rb_right;
 202                 else
 203                         return n;
 204         }
 205
 206         if (prev_ret) {
 207                 orig_prev = prev;
 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                 prev = orig_prev;
 214         }
 215
 216         if (next_ret) {
 217                 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
 218                 while (prev && offset < prev_entry->start) {
 219                         prev = rb_prev(prev);
 220                         prev_entry = rb_entry(prev, struct tree_entry, rb_node);
 221                 }
 222                 *next_ret = prev;
 223         }
 224         return NULL;
 225 }
 226
 227 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
 228                                           u64 offset)
 229 {
 230         struct rb_node *prev = NULL;
 231         struct rb_node *ret;
 232
 233         ret = __etree_search(tree, offset, &prev, NULL);
 234         if (!ret)
 235                 return prev;
 236         return ret;
 237 }
 238
 239 static struct extent_buffer *buffer_tree_insert(struct extent_io_tree *tree,
 240                                           u64 offset, struct rb_node *node)
 241 {
 242         struct rb_root *root = &tree->buffer;
 243         struct rb_node **p = &root->rb_node;
 244         struct rb_node *parent = NULL;
 245         struct extent_buffer *eb;
 246
 247         while (*p) {
 248                 parent = *p;
 249                 eb = rb_entry(parent, struct extent_buffer, rb_node);
 250
 251                 if (offset < eb->start)
 252                         p = &(*p)->rb_left;
 253                 else if (offset > eb->start)
 254                         p = &(*p)->rb_right;
 255                 else
 256                         return eb;
 257         }
 258
 259         rb_link_node(node, parent, p);
 260         rb_insert_color(node, root);
 261         return NULL;
 262 }
 263
 264 static struct extent_buffer *buffer_search(struct extent_io_tree *tree,
 265                                            u64 offset)
 266 {
 267         struct rb_root *root = &tree->buffer;
 268         struct rb_node *n = root->rb_node;
 269         struct extent_buffer *eb;
 270
 271         while (n) {
 272                 eb = rb_entry(n, struct extent_buffer, rb_node);
 273                 if (offset < eb->start)
 274                         n = n->rb_left;
 275                 else if (offset > eb->start)
 276                         n = n->rb_right;
 277                 else
 278                         return eb;
 279         }
 280         return NULL;
 281 }
 282
 283 /*
 284  * utility function to look for merge candidates inside a given range.
 285  * Any extents with matching state are merged together into a single
 286  * extent in the tree.  Extents with EXTENT_IO in their state field
 287  * are not merged because the end_io handlers need to be able to do
 288  * operations on them without sleeping (or doing allocations/splits).
 289  *
 290  * This should be called with the tree lock held.
 291  */
 292 static int merge_state(struct extent_io_tree *tree,
 293                        struct extent_state *state)
 294 {
 295         struct extent_state *other;
 296         struct rb_node *other_node;
 297
 298         if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
 299                 return 0;
 300
 301         other_node = rb_prev(&state->rb_node);
 302         if (other_node) {
 303                 other = rb_entry(other_node, struct extent_state, rb_node);
 304                 if (other->end == state->start - 1 &&
 305                     other->state == state->state) {
 306                         state->start = other->start;
 307                         other->tree = NULL;
 308                         rb_erase(&other->rb_node, &tree->state);
 309                         free_extent_state(other);
 310                 }
 311         }
 312         other_node = rb_next(&state->rb_node);
 313         if (other_node) {
 314                 other = rb_entry(other_node, struct extent_state, rb_node);
 315                 if (other->start == state->end + 1 &&
 316                     other->state == state->state) {
 317                         other->start = state->start;
 318                         state->tree = NULL;
 319                         rb_erase(&state->rb_node, &tree->state);
 320                         free_extent_state(state);
 321                 }
 322         }
 323         return 0;
 324 }
 325
 326 static void set_state_cb(struct extent_io_tree *tree,
 327                          struct extent_state *state,
 328                          unsigned long bits)
 329 {
 330         if (tree->ops && tree->ops->set_bit_hook) {
 331                 tree->ops->set_bit_hook(tree->mapping->host, state->start,
 332                                         state->end, state->state, bits);
 333         }
 334 }
 335
 336 static void clear_state_cb(struct extent_io_tree *tree,
 337                            struct extent_state *state,
 338                            unsigned long bits)
 339 {
 340         if (tree->ops && tree->ops->clear_bit_hook) {
 341                 tree->ops->clear_bit_hook(tree->mapping->host, state->start,
 342                                           state->end, state->state, bits);
 343         }
 344 }
 345
 346 /*
 347  * insert an extent_state struct into the tree.  'bits' are set on the
 348  * struct before it is inserted.
 349  *
 350  * This may return -EEXIST if the extent is already there, in which case the
 351  * state struct is freed.
 352  *
 353  * The tree lock is not taken internally.  This is a utility function and
 354  * probably isn't what you want to call (see set/clear_extent_bit).
 355  */
 356 static int insert_state(struct extent_io_tree *tree,
 357                         struct extent_state *state, u64 start, u64 end,
 358                         int bits)
 359 {
 360         struct rb_node *node;
 361
 362         if (end < start) {
 363                 printk(KERN_ERR "btrfs end < start %llu %llu\n",
 364                        (unsigned long long)end,
 365                        (unsigned long long)start);
 366                 WARN_ON(1);
 367         }
 368         if (bits & EXTENT_DIRTY)
 369                 tree->dirty_bytes += end - start + 1;
 370         set_state_cb(tree, state, bits);
 371         state->state |= bits;
 372         state->start = start;
 373         state->end = end;
 374         node = tree_insert(&tree->state, end, &state->rb_node);
 375         if (node) {
 376                 struct extent_state *found;
 377                 found = rb_entry(node, struct extent_state, rb_node);
 378                 printk(KERN_ERR "btrfs found node %llu %llu on insert of "
 379                        "%llu %llu\n", (unsigned long long)found->start,
 380                        (unsigned long long)found->end,
 381                        (unsigned long long)start, (unsigned long long)end);
 382                 free_extent_state(state);
 383                 return -EEXIST;
 384         }
 385         state->tree = tree;
 386         merge_state(tree, state);
 387         return 0;
 388 }
 389
 390 /*
 391  * split a given extent state struct in two, inserting the preallocated
 392  * struct 'prealloc' as the newly created second half.  'split' indicates an
 393  * offset inside 'orig' where it should be split.
 394  *
 395  * Before calling,
 396  * the tree has 'orig' at [orig->start, orig->end].  After calling, there
 397  * are two extent state structs in the tree:
 398  * prealloc: [orig->start, split - 1]
 399  * orig: [ split, orig->end ]
 400  *
 401  * The tree locks are not taken by this function. They need to be held
 402  * by the caller.
 403  */
 404 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
 405                        struct extent_state *prealloc, u64 split)
 406 {
 407         struct rb_node *node;
 408         prealloc->start = orig->start;
 409         prealloc->end = split - 1;
 410         prealloc->state = orig->state;
 411         orig->start = split;
 412
 413         node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
 414         if (node) {
 415                 free_extent_state(prealloc);
 416                 return -EEXIST;
 417         }
 418         prealloc->tree = tree;
 419         return 0;
 420 }
 421
 422 /*
 423  * utility function to clear some bits in an extent state struct.
 424  * it will optionally wake up any one waiting on this state (wake == 1), or
 425  * forcibly remove the state from the tree (delete == 1).
 426  *
 427  * If no bits are set on the state struct after clearing things, the
 428  * struct is freed and removed from the tree
 429  */
 430 static int clear_state_bit(struct extent_io_tree *tree,
 431                             struct extent_state *state, int bits, int wake,
 432                             int delete)
 433 {
 434         int ret = state->state & bits;
 435
 436         if ((bits & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
 437                 u64 range = state->end - state->start + 1;
 438                 WARN_ON(range > tree->dirty_bytes);
 439                 tree->dirty_bytes -= range;
 440         }
 441         clear_state_cb(tree, state, bits);
 442         state->state &= ~bits;
 443         if (wake)
 444                 wake_up(&state->wq);
 445         if (delete || state->state == 0) {
 446                 if (state->tree) {
 447                         clear_state_cb(tree, state, state->state);
 448                         rb_erase(&state->rb_node, &tree->state);
 449                         state->tree = NULL;
 450                         free_extent_state(state);
 451                 } else {
 452                         WARN_ON(1);
 453                 }
 454         } else {
 455                 merge_state(tree, state);
 456         }
 457         return ret;
 458 }
 459
 460 /*
 461  * clear some bits on a range in the tree.  This may require splitting
 462  * or inserting elements in the tree, so the gfp mask is used to
 463  * indicate which allocations or sleeping are allowed.
 464  *
 465  * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
 466  * the given range from the tree regardless of state (ie for truncate).
 467  *
 468  * the range [start, end] is inclusive.
 469  *
 470  * This takes the tree lock, and returns < 0 on error, > 0 if any of the
 471  * bits were already set, or zero if none of the bits were already set.
 472  */
 473 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
 474                      int bits, int wake, int delete, gfp_t mask)
 475 {
 476         struct extent_state *state;
 477         struct extent_state *prealloc = NULL;
 478         struct rb_node *node;
 479         int err;
 480         int set = 0;
 481
 482 again:
 483         if (!prealloc && (mask & __GFP_WAIT)) {
 484                 prealloc = alloc_extent_state(mask);
 485                 if (!prealloc)
 486                         return -ENOMEM;
 487         }
 488
 489         spin_lock(&tree->lock);
 490         /*
 491          * this search will find the extents that end after
 492          * our range starts
 493          */
 494         node = tree_search(tree, start);
 495         if (!node)
 496                 goto out;
 497         state = rb_entry(node, struct extent_state, rb_node);
 498         if (state->start > end)
 499                 goto out;
 500         WARN_ON(state->end < start);
 501
 502         /*
 503          *     | ---- desired range ---- |
 504          *  | state | or
 505          *  | ------------- state -------------- |
 506          *
 507          * We need to split the extent we found, and may flip
 508          * bits on second half.
 509          *
 510          * If the extent we found extends past our range, we
 511          * just split and search again.  It'll get split again
 512          * the next time though.
 513          *
 514          * If the extent we found is inside our range, we clear
 515          * the desired bit on it.
 516          */
 517
 518         if (state->start < start) {
 519                 if (!prealloc)
 520                         prealloc = alloc_extent_state(GFP_ATOMIC);
 521                 err = split_state(tree, state, prealloc, start);
 522                 BUG_ON(err == -EEXIST);
 523                 prealloc = NULL;
 524                 if (err)
 525                         goto out;
 526                 if (state->end <= end) {
 527                         start = state->end + 1;
 528                         set |= clear_state_bit(tree, state, bits,
 529                                         wake, delete);
 530                 } else {
 531                         start = state->start;
 532                 }
 533                 goto search_again;
 534         }
 535         /*
 536          * | ---- desired range ---- |
 537          *                        | state |
 538          * We need to split the extent, and clear the bit
 539          * on the first half
 540          */
 541         if (state->start <= end && state->end > end) {
 542                 if (!prealloc)
 543                         prealloc = alloc_extent_state(GFP_ATOMIC);
 544                 err = split_state(tree, state, prealloc, end + 1);
 545                 BUG_ON(err == -EEXIST);
 546
 547                 if (wake)
 548                         wake_up(&state->wq);
 549                 set |= clear_state_bit(tree, prealloc, bits,
 550                                        wake, delete);
 551                 prealloc = NULL;
 552                 goto out;
 553         }
 554
 555         start = state->end + 1;
 556         set |= clear_state_bit(tree, state, bits, wake, delete);
 557         goto search_again;
 558
 559 out:
 560         spin_unlock(&tree->lock);
 561         if (prealloc)
 562                 free_extent_state(prealloc);
 563
 564         return set;
 565
 566 search_again:
 567         if (start > end)
 568                 goto out;
 569         spin_unlock(&tree->lock);
 570         if (mask & __GFP_WAIT)
 571                 cond_resched();
 572         goto again;
 573 }
 574
 575 static int wait_on_state(struct extent_io_tree *tree,
 576                          struct extent_state *state)
 577                 __releases(tree->lock)
 578                 __acquires(tree->lock)
 579 {
 580         DEFINE_WAIT(wait);
 581         prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
 582         spin_unlock(&tree->lock);
 583         schedule();
 584         spin_lock(&tree->lock);
 585         finish_wait(&state->wq, &wait);
 586         return 0;
 587 }
 588
 589 /*
 590  * waits for one or more bits to clear on a range in the state tree.
 591  * The range [start, end] is inclusive.
 592  * The tree lock is taken by this function
 593  */
 594 int wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits)
 595 {
 596         struct extent_state *state;
 597         struct rb_node *node;
 598
 599         spin_lock(&tree->lock);
 600 again:
 601         while (1) {
 602                 /*
 603                  * this search will find all the extents that end after
 604                  * our range starts
 605                  */
 606                 node = tree_search(tree, start);
 607                 if (!node)
 608                         break;
 609
 610                 state = rb_entry(node, struct extent_state, rb_node);
 611
 612                 if (state->start > end)
 613                         goto out;
 614
 615                 if (state->state & bits) {
 616                         start = state->start;
 617                         atomic_inc(&state->refs);
 618                         wait_on_state(tree, state);
 619                         free_extent_state(state);
 620                         goto again;
 621                 }
 622                 start = state->end + 1;
 623
 624                 if (start > end)
 625                         break;
 626
 627                 if (need_resched()) {
 628                         spin_unlock(&tree->lock);
 629                         cond_resched();
 630                         spin_lock(&tree->lock);
 631                 }
 632         }
 633 out:
 634         spin_unlock(&tree->lock);
 635         return 0;
 636 }
 637
 638 static void set_state_bits(struct extent_io_tree *tree,
 639                            struct extent_state *state,
 640                            int bits)
 641 {
 642         if ((bits & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
 643                 u64 range = state->end - state->start + 1;
 644                 tree->dirty_bytes += range;
 645         }
 646         set_state_cb(tree, state, bits);
 647         state->state |= bits;
 648 }
 649
 650 /*
 651  * set some bits on a range in the tree.  This may require allocations
 652  * or sleeping, so the gfp mask is used to indicate what is allowed.
 653  *
 654  * If 'exclusive' == 1, this will fail with -EEXIST if some part of the
 655  * range already has the desired bits set.  The start of the existing
 656  * range is returned in failed_start in this case.
 657  *
 658  * [start, end] is inclusive
 659  * This takes the tree lock.
 660  */
 661 static int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
 662                           int bits, int exclusive, u64 *failed_start,
 663                           gfp_t mask)
 664 {
 665         struct extent_state *state;
 666         struct extent_state *prealloc = NULL;
 667         struct rb_node *node;
 668         int err = 0;
 669         int set;
 670         u64 last_start;
 671         u64 last_end;
 672 again:
 673         if (!prealloc && (mask & __GFP_WAIT)) {
 674                 prealloc = alloc_extent_state(mask);
 675                 if (!prealloc)
 676                         return -ENOMEM;
 677         }
 678
 679         spin_lock(&tree->lock);
 680         /*
 681          * this search will find all the extents that end after
 682          * our range starts.
 683          */
 684         node = tree_search(tree, start);
 685         if (!node) {
 686                 err = insert_state(tree, prealloc, start, end, bits);
 687                 prealloc = NULL;
 688                 BUG_ON(err == -EEXIST);
 689                 goto out;
 690         }
 691
 692         state = rb_entry(node, struct extent_state, rb_node);
 693         last_start = state->start;
 694         last_end = state->end;
 695
 696         /*
 697          * | ---- desired range ---- |
 698          * | state |
 699          *
 700          * Just lock what we found and keep going
 701          */
 702         if (state->start == start && state->end <= end) {
 703                 set = state->state & bits;
 704                 if (set && exclusive) {
 705                         *failed_start = state->start;
 706                         err = -EEXIST;
 707                         goto out;
 708                 }
 709                 set_state_bits(tree, state, bits);
 710                 start = state->end + 1;
 711                 merge_state(tree, state);
 712                 goto search_again;
 713         }
 714
 715         /*
 716          *     | ---- desired range ---- |
 717          * | state |
 718          *   or
 719          * | ------------- state -------------- |
 720          *
 721          * We need to split the extent we found, and may flip bits on
 722          * second half.
 723          *
 724          * If the extent we found extends past our
 725          * range, we just split and search again.  It'll get split
 726          * again the next time though.
 727          *
 728          * If the extent we found is inside our range, we set the
 729          * desired bit on it.
 730          */
 731         if (state->start < start) {
 732                 set = state->state & bits;
 733                 if (exclusive && set) {
 734                         *failed_start = start;
 735                         err = -EEXIST;
 736                         goto out;
 737                 }
 738                 err = split_state(tree, state, prealloc, start);
 739                 BUG_ON(err == -EEXIST);
 740                 prealloc = NULL;
 741                 if (err)
 742                         goto out;
 743                 if (state->end <= end) {
 744                         set_state_bits(tree, state, bits);
 745                         start = state->end + 1;
 746                         merge_state(tree, state);
 747                 } else {
 748                         start = state->start;
 749                 }
 750                 goto search_again;
 751         }
 752         /*
 753          * | ---- desired range ---- |
 754          *     | state | or               | state |
 755          *
 756          * There's a hole, we need to insert something in it and
 757          * ignore the extent we found.
 758          */
 759         if (state->start > start) {
 760                 u64 this_end;
 761                 if (end < last_start)
 762                         this_end = end;
 763                 else
 764                         this_end = last_start - 1;
 765                 err = insert_state(tree, prealloc, start, this_end,
 766                                    bits);
 767                 prealloc = NULL;
 768                 BUG_ON(err == -EEXIST);
 769                 if (err)
 770                         goto out;
 771                 start = this_end + 1;
 772                 goto search_again;
 773         }
 774         /*
 775          * | ---- desired range ---- |
 776          *                        | state |
 777          * We need to split the extent, and set the bit
 778          * on the first half
 779          */
 780         if (state->start <= end && state->end > end) {
 781                 set = state->state & bits;
 782                 if (exclusive && set) {
 783                         *failed_start = start;
 784                         err = -EEXIST;
 785                         goto out;
 786                 }
 787                 err = split_state(tree, state, prealloc, end + 1);
 788                 BUG_ON(err == -EEXIST);
 789
 790                 set_state_bits(tree, prealloc, bits);
 791                 merge_state(tree, prealloc);
 792                 prealloc = NULL;
 793                 goto out;
 794         }
 795
 796         goto search_again;
 797
 798 out:
 799         spin_unlock(&tree->lock);
 800         if (prealloc)
 801                 free_extent_state(prealloc);
 802
 803         return err;
 804
 805 search_again:
 806         if (start > end)
 807                 goto out;
 808         spin_unlock(&tree->lock);
 809         if (mask & __GFP_WAIT)
 810                 cond_resched();
 811         goto again;
 812 }
 813
 814 /* wrappers around set/clear extent bit */
 815 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
 816                      gfp_t mask)
 817 {
 818         return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL,
 819                               mask);
 820 }
 821
 822 int set_extent_ordered(struct extent_io_tree *tree, u64 start, u64 end,
 823                        gfp_t mask)
 824 {
 825         return set_extent_bit(tree, start, end, EXTENT_ORDERED, 0, NULL, mask);
 826 }
 827
 828 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
 829                     int bits, gfp_t mask)
 830 {
 831         return set_extent_bit(tree, start, end, bits, 0, NULL,
 832                               mask);
 833 }
 834
 835 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
 836                       int bits, gfp_t mask)
 837 {
 838         return clear_extent_bit(tree, start, end, bits, 0, 0, mask);
 839 }
 840
 841 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
 842                      gfp_t mask)
 843 {
 844         return set_extent_bit(tree, start, end,
 845                               EXTENT_DELALLOC | EXTENT_DIRTY,
 846                               0, NULL, mask);
 847 }
 848
 849 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
 850                        gfp_t mask)
 851 {
 852         return clear_extent_bit(tree, start, end,
 853                                 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, mask);
 854 }
 855
 856 int clear_extent_ordered(struct extent_io_tree *tree, u64 start, u64 end,
 857                          gfp_t mask)
 858 {
 859         return clear_extent_bit(tree, start, end, EXTENT_ORDERED, 1, 0, mask);
 860 }
 861
 862 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
 863                      gfp_t mask)
 864 {
 865         return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL,
 866                               mask);
 867 }
 868
 869 static int clear_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
 870                        gfp_t mask)
 871 {
 872         return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0, mask);
 873 }
 874
 875 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
 876                         gfp_t mask)
 877 {
 878         return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, NULL,
 879                               mask);
 880 }
 881
 882 static int clear_extent_uptodate(struct extent_io_tree *tree, u64 start,
 883                                  u64 end, gfp_t mask)
 884 {
 885         return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, mask);
 886 }
 887
 888 static int set_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end,
 889                          gfp_t mask)
 890 {
 891         return set_extent_bit(tree, start, end, EXTENT_WRITEBACK,
 892                               0, NULL, mask);
 893 }
 894
 895 static int clear_extent_writeback(struct extent_io_tree *tree, u64 start,
 896                                   u64 end, gfp_t mask)
 897 {
 898         return clear_extent_bit(tree, start, end, EXTENT_WRITEBACK, 1, 0, mask);
 899 }
 900
 901 int wait_on_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end)
 902 {
 903         return wait_extent_bit(tree, start, end, EXTENT_WRITEBACK);
 904 }
 905
 906 /*
 907  * either insert or lock state struct between start and end use mask to tell
 908  * us if waiting is desired.
 909  */
 910 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
 911 {
 912         int err;
 913         u64 failed_start;
 914         while (1) {
 915                 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1,
 916                                      &failed_start, mask);
 917                 if (err == -EEXIST && (mask & __GFP_WAIT)) {
 918                         wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
 919                         start = failed_start;
 920                 } else {
 921                         break;
 922                 }
 923                 WARN_ON(start > end);
 924         }
 925         return err;
 926 }
 927
 928 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end,
 929                     gfp_t mask)
 930 {
 931         int err;
 932         u64 failed_start;
 933
 934         err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1,
 935                              &failed_start, mask);
 936         if (err == -EEXIST) {
 937                 if (failed_start > start)
 938                         clear_extent_bit(tree, start, failed_start - 1,
 939                                          EXTENT_LOCKED, 1, 0, mask);
 940                 return 0;
 941         }
 942         return 1;
 943 }
 944
 945 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end,
 946                   gfp_t mask)
 947 {
 948         return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, mask);
 949 }
 950
 951 /*
 952  * helper function to set pages and extents in the tree dirty
 953  */
 954 int set_range_dirty(struct extent_io_tree *tree, u64 start, u64 end)
 955 {
 956         unsigned long index = start >> PAGE_CACHE_SHIFT;
 957         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
 958         struct page *page;
 959
 960         while (index <= end_index) {
 961                 page = find_get_page(tree->mapping, index);
 962                 BUG_ON(!page);
 963                 __set_page_dirty_nobuffers(page);
 964                 page_cache_release(page);
 965                 index++;
 966         }
 967         set_extent_dirty(tree, start, end, GFP_NOFS);
 968         return 0;
 969 }
 970
 971 /*
 972  * helper function to set both pages and extents in the tree writeback
 973  */
 974 static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
 975 {
 976         unsigned long index = start >> PAGE_CACHE_SHIFT;
 977         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
 978         struct page *page;
 979
 980         while (index <= end_index) {
 981                 page = find_get_page(tree->mapping, index);
 982                 BUG_ON(!page);
 983                 set_page_writeback(page);
 984                 page_cache_release(page);
 985                 index++;
 986         }
 987         set_extent_writeback(tree, start, end, GFP_NOFS);
 988         return 0;
 989 }
 990
 991 /*
 992  * find the first offset in the io tree with 'bits' set. zero is
 993  * returned if we find something, and *start_ret and *end_ret are
 994  * set to reflect the state struct that was found.
 995  *
 996  * If nothing was found, 1 is returned, < 0 on error
 997  */
 998 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
 999                           u64 *start_ret, u64 *end_ret, int bits)
1000 {
1001         struct rb_node *node;
1002         struct extent_state *state;
1003         int ret = 1;
1004
1005         spin_lock(&tree->lock);
1006         /*
1007          * this search will find all the extents that end after
1008          * our range starts.
1009          */
1010         node = tree_search(tree, start);
1011         if (!node)
1012                 goto out;
1013
1014         while (1) {
1015                 state = rb_entry(node, struct extent_state, rb_node);
1016                 if (state->end >= start && (state->state & bits)) {
1017                         *start_ret = state->start;
1018                         *end_ret = state->end;
1019                         ret = 0;
1020                         break;
1021                 }
1022                 node = rb_next(node);
1023                 if (!node)
1024                         break;
1025         }
1026 out:
1027         spin_unlock(&tree->lock);
1028         return ret;
1029 }
1030
1031 /* find the first state struct with 'bits' set after 'start', and
1032  * return it.  tree->lock must be held.  NULL will returned if
1033  * nothing was found after 'start'
1034  */
1035 struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree,
1036                                                  u64 start, int bits)
1037 {
1038         struct rb_node *node;
1039         struct extent_state *state;
1040
1041         /*
1042          * this search will find all the extents that end after
1043          * our range starts.
1044          */
1045         node = tree_search(tree, start);
1046         if (!node)
1047                 goto out;
1048
1049         while (1) {
1050                 state = rb_entry(node, struct extent_state, rb_node);
1051                 if (state->end >= start && (state->state & bits))
1052                         return state;
1053
1054                 node = rb_next(node);
1055                 if (!node)
1056                         break;
1057         }
1058 out:
1059         return NULL;
1060 }
1061
1062 /*
1063  * find a contiguous range of bytes in the file marked as delalloc, not
1064  * more than 'max_bytes'.  start and end are used to return the range,
1065  *
1066  * 1 is returned if we find something, 0 if nothing was in the tree
1067  */
1068 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1069                                         u64 *start, u64 *end, u64 max_bytes)
1070 {
1071         struct rb_node *node;
1072         struct extent_state *state;
1073         u64 cur_start = *start;
1074         u64 found = 0;
1075         u64 total_bytes = 0;
1076
1077         spin_lock(&tree->lock);
1078
1079         /*
1080          * this search will find all the extents that end after
1081          * our range starts.
1082          */
1083         node = tree_search(tree, cur_start);
1084         if (!node) {
1085                 if (!found)
1086                         *end = (u64)-1;
1087                 goto out;
1088         }
1089
1090         while (1) {
1091                 state = rb_entry(node, struct extent_state, rb_node);
1092                 if (found && (state->start != cur_start ||
1093                               (state->state & EXTENT_BOUNDARY))) {
1094                         goto out;
1095                 }
1096                 if (!(state->state & EXTENT_DELALLOC)) {
1097                         if (!found)
1098                                 *end = state->end;
1099                         goto out;
1100                 }
1101                 if (!found)
1102                         *start = state->start;
1103                 found++;
1104                 *end = state->end;
1105                 cur_start = state->end + 1;
1106                 node = rb_next(node);
1107                 if (!node)
1108                         break;
1109                 total_bytes += state->end - state->start + 1;
1110                 if (total_bytes >= max_bytes)
1111                         break;
1112         }
1113 out:
1114         spin_unlock(&tree->lock);
1115         return found;
1116 }
1117
1118 static noinline int __unlock_for_delalloc(struct inode *inode,
1119                                           struct page *locked_page,
1120                                           u64 start, u64 end)
1121 {
1122         int ret;
1123         struct page *pages[16];
1124         unsigned long index = start >> PAGE_CACHE_SHIFT;
1125         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1126         unsigned long nr_pages = end_index - index + 1;
1127         int i;
1128
1129         if (index == locked_page->index && end_index == index)
1130                 return 0;
1131
1132         while (nr_pages > 0) {
1133                 ret = find_get_pages_contig(inode->i_mapping, index,
1134                                      min_t(unsigned long, nr_pages,
1135                                      ARRAY_SIZE(pages)), pages);
1136                 for (i = 0; i < ret; i++) {
1137                         if (pages[i] != locked_page)
1138                                 unlock_page(pages[i]);
1139                         page_cache_release(pages[i]);
1140                 }
1141                 nr_pages -= ret;
1142                 index += ret;
1143                 cond_resched();
1144         }
1145         return 0;
1146 }
1147
1148 static noinline int lock_delalloc_pages(struct inode *inode,
1149                                         struct page *locked_page,
1150                                         u64 delalloc_start,
1151                                         u64 delalloc_end)
1152 {
1153         unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT;
1154         unsigned long start_index = index;
1155         unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT;
1156         unsigned long pages_locked = 0;
1157         struct page *pages[16];
1158         unsigned long nrpages;
1159         int ret;
1160         int i;
1161
1162         /* the caller is responsible for locking the start index */
1163         if (index == locked_page->index && index == end_index)
1164                 return 0;
1165
1166         /* skip the page at the start index */
1167         nrpages = end_index - index + 1;
1168         while (nrpages > 0) {
1169                 ret = find_get_pages_contig(inode->i_mapping, index,
1170                                      min_t(unsigned long,
1171                                      nrpages, ARRAY_SIZE(pages)), pages);
1172                 if (ret == 0) {
1173                         ret = -EAGAIN;
1174                         goto done;
1175                 }
1176                 /* now we have an array of pages, lock them all */
1177                 for (i = 0; i < ret; i++) {
1178                         /*
1179                          * the caller is taking responsibility for
1180                          * locked_page
1181                          */
1182                         if (pages[i] != locked_page) {
1183                                 lock_page(pages[i]);
1184                                 if (!PageDirty(pages[i]) ||
1185                                     pages[i]->mapping != inode->i_mapping) {
1186                                         ret = -EAGAIN;
1187                                         unlock_page(pages[i]);
1188                                         page_cache_release(pages[i]);
1189                                         goto done;
1190                                 }
1191                         }
1192                         page_cache_release(pages[i]);
1193                         pages_locked++;
1194                 }
1195                 nrpages -= ret;
1196                 index += ret;
1197                 cond_resched();
1198         }
1199         ret = 0;
1200 done:
1201         if (ret && pages_locked) {
1202                 __unlock_for_delalloc(inode, locked_page,
1203                               delalloc_start,
1204                               ((u64)(start_index + pages_locked - 1)) <<
1205                               PAGE_CACHE_SHIFT);
1206         }
1207         return ret;
1208 }
1209
1210 /*
1211  * find a contiguous range of bytes in the file marked as delalloc, not
1212  * more than 'max_bytes'.  start and end are used to return the range,
1213  *
1214  * 1 is returned if we find something, 0 if nothing was in the tree
1215  */
1216 static noinline u64 find_lock_delalloc_range(struct inode *inode,
1217                                              struct extent_io_tree *tree,
1218                                              struct page *locked_page,
1219                                              u64 *start, u64 *end,
1220                                              u64 max_bytes)
1221 {
1222         u64 delalloc_start;
1223         u64 delalloc_end;
1224         u64 found;
1225         int ret;
1226         int loops = 0;
1227
1228 again:
1229         /* step one, find a bunch of delalloc bytes starting at start */
1230         delalloc_start = *start;
1231         delalloc_end = 0;
1232         found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1233                                     max_bytes);
1234         if (!found || delalloc_end <= *start) {
1235                 *start = delalloc_start;
1236                 *end = delalloc_end;
1237                 return found;
1238         }
1239
1240         /*
1241          * start comes from the offset of locked_page.  We have to lock
1242          * pages in order, so we can't process delalloc bytes before
1243          * locked_page
1244          */
1245         if (delalloc_start < *start)
1246                 delalloc_start = *start;
1247
1248         /*
1249          * make sure to limit the number of pages we try to lock down
1250          * if we're looping.
1251          */
1252         if (delalloc_end + 1 - delalloc_start > max_bytes && loops)
1253                 delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1;
1254
1255         /* step two, lock all the pages after the page that has start */
1256         ret = lock_delalloc_pages(inode, locked_page,
1257                                   delalloc_start, delalloc_end);
1258         if (ret == -EAGAIN) {
1259                 /* some of the pages are gone, lets avoid looping by
1260                  * shortening the size of the delalloc range we're searching
1261                  */
1262                 if (!loops) {
1263                         unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1);
1264                         max_bytes = PAGE_CACHE_SIZE - offset;
1265                         loops = 1;
1266                         goto again;
1267                 } else {
1268                         found = 0;
1269                         goto out_failed;
1270                 }
1271         }
1272         BUG_ON(ret);
1273
1274         /* step three, lock the state bits for the whole range */
1275         lock_extent(tree, delalloc_start, delalloc_end, GFP_NOFS);
1276
1277         /* then test to make sure it is all still delalloc */
1278         ret = test_range_bit(tree, delalloc_start, delalloc_end,
1279                              EXTENT_DELALLOC, 1);
1280         if (!ret) {
1281                 unlock_extent(tree, delalloc_start, delalloc_end, GFP_NOFS);
1282                 __unlock_for_delalloc(inode, locked_page,
1283                               delalloc_start, delalloc_end);
1284                 cond_resched();
1285                 goto again;
1286         }
1287         *start = delalloc_start;
1288         *end = delalloc_end;
1289 out_failed:
1290         return found;
1291 }
1292
1293 int extent_clear_unlock_delalloc(struct inode *inode,
1294                                 struct extent_io_tree *tree,
1295                                 u64 start, u64 end, struct page *locked_page,
1296                                 int unlock_pages,
1297                                 int clear_unlock,
1298                                 int clear_delalloc, int clear_dirty,
1299                                 int set_writeback,
1300                                 int end_writeback)
1301 {
1302         int ret;
1303         struct page *pages[16];
1304         unsigned long index = start >> PAGE_CACHE_SHIFT;
1305         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1306         unsigned long nr_pages = end_index - index + 1;
1307         int i;
1308         int clear_bits = 0;
1309
1310         if (clear_unlock)
1311                 clear_bits |= EXTENT_LOCKED;
1312         if (clear_dirty)
1313                 clear_bits |= EXTENT_DIRTY;
1314
1315         if (clear_delalloc)
1316                 clear_bits |= EXTENT_DELALLOC;
1317
1318         clear_extent_bit(tree, start, end, clear_bits, 1, 0, GFP_NOFS);
1319         if (!(unlock_pages || clear_dirty || set_writeback || end_writeback))
1320                 return 0;
1321
1322         while (nr_pages > 0) {
1323                 ret = find_get_pages_contig(inode->i_mapping, index,
1324                                      min_t(unsigned long,
1325                                      nr_pages, ARRAY_SIZE(pages)), pages);
1326                 for (i = 0; i < ret; i++) {
1327                         if (pages[i] == locked_page) {
1328                                 page_cache_release(pages[i]);
1329                                 continue;
1330                         }
1331                         if (clear_dirty)
1332                                 clear_page_dirty_for_io(pages[i]);
1333                         if (set_writeback)
1334                                 set_page_writeback(pages[i]);
1335                         if (end_writeback)
1336                                 end_page_writeback(pages[i]);
1337                         if (unlock_pages)
1338                                 unlock_page(pages[i]);
1339                         page_cache_release(pages[i]);
1340                 }
1341                 nr_pages -= ret;
1342                 index += ret;
1343                 cond_resched();
1344         }
1345         return 0;
1346 }
1347
1348 /*
1349  * count the number of bytes in the tree that have a given bit(s)
1350  * set.  This can be fairly slow, except for EXTENT_DIRTY which is
1351  * cached.  The total number found is returned.
1352  */
1353 u64 count_range_bits(struct extent_io_tree *tree,
1354                      u64 *start, u64 search_end, u64 max_bytes,
1355                      unsigned long bits)
1356 {
1357         struct rb_node *node;
1358         struct extent_state *state;
1359         u64 cur_start = *start;
1360         u64 total_bytes = 0;
1361         int found = 0;
1362
1363         if (search_end <= cur_start) {
1364                 WARN_ON(1);
1365                 return 0;
1366         }
1367
1368         spin_lock(&tree->lock);
1369         if (cur_start == 0 && bits == EXTENT_DIRTY) {
1370                 total_bytes = tree->dirty_bytes;
1371                 goto out;
1372         }
1373         /*
1374          * this search will find all the extents that end after
1375          * our range starts.
1376          */
1377         node = tree_search(tree, cur_start);
1378         if (!node)
1379                 goto out;
1380
1381         while (1) {
1382                 state = rb_entry(node, struct extent_state, rb_node);
1383                 if (state->start > search_end)
1384                         break;
1385                 if (state->end >= cur_start && (state->state & bits)) {
1386                         total_bytes += min(search_end, state->end) + 1 -
1387                                        max(cur_start, state->start);
1388                         if (total_bytes >= max_bytes)
1389                                 break;
1390                         if (!found) {
1391                                 *start = state->start;
1392                                 found = 1;
1393                         }
1394                 }
1395                 node = rb_next(node);
1396                 if (!node)
1397                         break;
1398         }
1399 out:
1400         spin_unlock(&tree->lock);
1401         return total_bytes;
1402 }
1403
1404 /*
1405  * set the private field for a given byte offset in the tree.  If there isn't
1406  * an extent_state there already, this does nothing.
1407  */
1408 int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1409 {
1410         struct rb_node *node;
1411         struct extent_state *state;
1412         int ret = 0;
1413
1414         spin_lock(&tree->lock);
1415         /*
1416          * this search will find all the extents that end after
1417          * our range starts.
1418          */
1419         node = tree_search(tree, start);
1420         if (!node) {
1421                 ret = -ENOENT;
1422                 goto out;
1423         }
1424         state = rb_entry(node, struct extent_state, rb_node);
1425         if (state->start != start) {
1426                 ret = -ENOENT;
1427                 goto out;
1428         }
1429         state->private = private;
1430 out:
1431         spin_unlock(&tree->lock);
1432         return ret;
1433 }
1434
1435 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1436 {
1437         struct rb_node *node;
1438         struct extent_state *state;
1439         int ret = 0;
1440
1441         spin_lock(&tree->lock);
1442         /*
1443          * this search will find all the extents that end after
1444          * our range starts.
1445          */
1446         node = tree_search(tree, start);
1447         if (!node) {
1448                 ret = -ENOENT;
1449                 goto out;
1450         }
1451         state = rb_entry(node, struct extent_state, rb_node);
1452         if (state->start != start) {
1453                 ret = -ENOENT;
1454                 goto out;
1455         }
1456         *private = state->private;
1457 out:
1458         spin_unlock(&tree->lock);
1459         return ret;
1460 }
1461
1462 /*
1463  * searches a range in the state tree for a given mask.
1464  * If 'filled' == 1, this returns 1 only if every extent in the tree
1465  * has the bits set.  Otherwise, 1 is returned if any bit in the
1466  * range is found set.
1467  */
1468 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1469                    int bits, int filled)
1470 {
1471         struct extent_state *state = NULL;
1472         struct rb_node *node;
1473         int bitset = 0;
1474
1475         spin_lock(&tree->lock);
1476         node = tree_search(tree, start);
1477         while (node && start <= end) {
1478                 state = rb_entry(node, struct extent_state, rb_node);
1479
1480                 if (filled && state->start > start) {
1481                         bitset = 0;
1482                         break;
1483                 }
1484
1485                 if (state->start > end)
1486                         break;
1487
1488                 if (state->state & bits) {
1489                         bitset = 1;
1490                         if (!filled)
1491                                 break;
1492                 } else if (filled) {
1493                         bitset = 0;
1494                         break;
1495                 }
1496                 start = state->end + 1;
1497                 if (start > end)
1498                         break;
1499                 node = rb_next(node);
1500                 if (!node) {
1501                         if (filled)
1502                                 bitset = 0;
1503                         break;
1504                 }
1505         }
1506         spin_unlock(&tree->lock);
1507         return bitset;
1508 }
1509
1510 /*
1511  * helper function to set a given page up to date if all the
1512  * extents in the tree for that page are up to date
1513  */
1514 static int check_page_uptodate(struct extent_io_tree *tree,
1515                                struct page *page)
1516 {
1517         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1518         u64 end = start + PAGE_CACHE_SIZE - 1;
1519         if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1))
1520                 SetPageUptodate(page);
1521         return 0;
1522 }
1523
1524 /*
1525  * helper function to unlock a page if all the extents in the tree
1526  * for that page are unlocked
1527  */
1528 static int check_page_locked(struct extent_io_tree *tree,
1529                              struct page *page)
1530 {
1531         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1532         u64 end = start + PAGE_CACHE_SIZE - 1;
1533         if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0))
1534                 unlock_page(page);
1535         return 0;
1536 }
1537
1538 /*
1539  * helper function to end page writeback if all the extents
1540  * in the tree for that page are done with writeback
1541  */
1542 static int check_page_writeback(struct extent_io_tree *tree,
1543                              struct page *page)
1544 {
1545         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1546         u64 end = start + PAGE_CACHE_SIZE - 1;
1547         if (!test_range_bit(tree, start, end, EXTENT_WRITEBACK, 0))
1548                 end_page_writeback(page);
1549         return 0;
1550 }
1551
1552 /* lots and lots of room for performance fixes in the end_bio funcs */
1553
1554 /*
1555  * after a writepage IO is done, we need to:
1556  * clear the uptodate bits on error
1557  * clear the writeback bits in the extent tree for this IO
1558  * end_page_writeback if the page has no more pending IO
1559  *
1560  * Scheduling is not allowed, so the extent state tree is expected
1561  * to have one and only one object corresponding to this IO.
1562  */
1563 static void end_bio_extent_writepage(struct bio *bio, int err)
1564 {
1565         int uptodate = err == 0;
1566         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1567         struct extent_io_tree *tree;
1568         u64 start;
1569         u64 end;
1570         int whole_page;
1571         int ret;
1572
1573         do {
1574                 struct page *page = bvec->bv_page;
1575                 tree = &BTRFS_I(page->mapping->host)->io_tree;
1576
1577                 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1578                          bvec->bv_offset;
1579                 end = start + bvec->bv_len - 1;
1580
1581                 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1582                         whole_page = 1;
1583                 else
1584                         whole_page = 0;
1585
1586                 if (--bvec >= bio->bi_io_vec)
1587                         prefetchw(&bvec->bv_page->flags);
1588                 if (tree->ops && tree->ops->writepage_end_io_hook) {
1589                         ret = tree->ops->writepage_end_io_hook(page, start,
1590                                                        end, NULL, uptodate);
1591                         if (ret)
1592                                 uptodate = 0;
1593                 }
1594
1595                 if (!uptodate && tree->ops &&
1596                     tree->ops->writepage_io_failed_hook) {
1597                         ret = tree->ops->writepage_io_failed_hook(bio, page,
1598                                                          start, end, NULL);
1599                         if (ret == 0) {
1600                                 uptodate = (err == 0);
1601                                 continue;
1602                         }
1603                 }
1604
1605                 if (!uptodate) {
1606                         clear_extent_uptodate(tree, start, end, GFP_ATOMIC);
1607                         ClearPageUptodate(page);
1608                         SetPageError(page);
1609                 }
1610
1611                 clear_extent_writeback(tree, start, end, GFP_ATOMIC);
1612
1613                 if (whole_page)
1614                         end_page_writeback(page);
1615                 else
1616                         check_page_writeback(tree, page);
1617         } while (bvec >= bio->bi_io_vec);
1618
1619         bio_put(bio);
1620 }
1621
1622 /*
1623  * after a readpage IO is done, we need to:
1624  * clear the uptodate bits on error
1625  * set the uptodate bits if things worked
1626  * set the page up to date if all extents in the tree are uptodate
1627  * clear the lock bit in the extent tree
1628  * unlock the page if there are no other extents locked for it
1629  *
1630  * Scheduling is not allowed, so the extent state tree is expected
1631  * to have one and only one object corresponding to this IO.
1632  */
1633 static void end_bio_extent_readpage(struct bio *bio, int err)
1634 {
1635         int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1636         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1637         struct extent_io_tree *tree;
1638         u64 start;
1639         u64 end;
1640         int whole_page;
1641         int ret;
1642
1643         if (err)
1644                 uptodate = 0;
1645
1646         do {
1647                 struct page *page = bvec->bv_page;
1648                 tree = &BTRFS_I(page->mapping->host)->io_tree;
1649
1650                 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1651                         bvec->bv_offset;
1652                 end = start + bvec->bv_len - 1;
1653
1654                 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1655                         whole_page = 1;
1656                 else
1657                         whole_page = 0;
1658
1659                 if (--bvec >= bio->bi_io_vec)
1660                         prefetchw(&bvec->bv_page->flags);
1661
1662                 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
1663                         ret = tree->ops->readpage_end_io_hook(page, start, end,
1664                                                               NULL);
1665                         if (ret)
1666                                 uptodate = 0;
1667                 }
1668                 if (!uptodate && tree->ops &&
1669                     tree->ops->readpage_io_failed_hook) {
1670                         ret = tree->ops->readpage_io_failed_hook(bio, page,
1671                                                          start, end, NULL);
1672                         if (ret == 0) {
1673                                 uptodate =
1674                                         test_bit(BIO_UPTODATE, &bio->bi_flags);
1675                                 if (err)
1676                                         uptodate = 0;
1677                                 continue;
1678                         }
1679                 }
1680
1681                 if (uptodate) {
1682                         set_extent_uptodate(tree, start, end,
1683                                             GFP_ATOMIC);
1684                 }
1685                 unlock_extent(tree, start, end, GFP_ATOMIC);
1686
1687                 if (whole_page) {
1688                         if (uptodate) {
1689                                 SetPageUptodate(page);
1690                         } else {
1691                                 ClearPageUptodate(page);
1692                                 SetPageError(page);
1693                         }
1694                         unlock_page(page);
1695                 } else {
1696                         if (uptodate) {
1697                                 check_page_uptodate(tree, page);
1698                         } else {
1699                                 ClearPageUptodate(page);
1700                                 SetPageError(page);
1701                         }
1702                         check_page_locked(tree, page);
1703                 }
1704         } while (bvec >= bio->bi_io_vec);
1705
1706         bio_put(bio);
1707 }
1708
1709 /*
1710  * IO done from prepare_write is pretty simple, we just unlock
1711  * the structs in the extent tree when done, and set the uptodate bits
1712  * as appropriate.
1713  */
1714 static void end_bio_extent_preparewrite(struct bio *bio, int err)
1715 {
1716         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1717         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1718         struct extent_io_tree *tree;
1719         u64 start;
1720         u64 end;
1721
1722         do {
1723                 struct page *page = bvec->bv_page;
1724                 tree = &BTRFS_I(page->mapping->host)->io_tree;
1725
1726                 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1727                         bvec->bv_offset;
1728                 end = start + bvec->bv_len - 1;
1729
1730                 if (--bvec >= bio->bi_io_vec)
1731                         prefetchw(&bvec->bv_page->flags);
1732
1733                 if (uptodate) {
1734                         set_extent_uptodate(tree, start, end, GFP_ATOMIC);
1735                 } else {
1736                         ClearPageUptodate(page);
1737                         SetPageError(page);
1738                 }
1739
1740                 unlock_extent(tree, start, end, GFP_ATOMIC);
1741
1742         } while (bvec >= bio->bi_io_vec);
1743
1744         bio_put(bio);
1745 }
1746
1747 static struct bio *
1748 extent_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
1749                  gfp_t gfp_flags)
1750 {
1751         struct bio *bio;
1752
1753         bio = bio_alloc(gfp_flags, nr_vecs);
1754
1755         if (bio == NULL && (current->flags & PF_MEMALLOC)) {
1756                 while (!bio && (nr_vecs /= 2))
1757                         bio = bio_alloc(gfp_flags, nr_vecs);
1758         }
1759
1760         if (bio) {
1761                 bio->bi_size = 0;
1762                 bio->bi_bdev = bdev;
1763                 bio->bi_sector = first_sector;
1764         }
1765         return bio;
1766 }
1767
1768 static int submit_one_bio(int rw, struct bio *bio, int mirror_num,
1769                           unsigned long bio_flags)
1770 {
1771         int ret = 0;
1772         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1773         struct page *page = bvec->bv_page;
1774         struct extent_io_tree *tree = bio->bi_private;
1775         u64 start;
1776         u64 end;
1777
1778         start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
1779         end = start + bvec->bv_len - 1;
1780
1781         bio->bi_private = NULL;
1782
1783         bio_get(bio);
1784
1785         if (tree->ops && tree->ops->submit_bio_hook)
1786                 tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
1787                                            mirror_num, bio_flags);
1788         else
1789                 submit_bio(rw, bio);
1790         if (bio_flagged(bio, BIO_EOPNOTSUPP))
1791                 ret = -EOPNOTSUPP;
1792         bio_put(bio);
1793         return ret;
1794 }
1795
1796 static int submit_extent_page(int rw, struct extent_io_tree *tree,
1797                               struct page *page, sector_t sector,
1798                               size_t size, unsigned long offset,
1799                               struct block_device *bdev,
1800                               struct bio **bio_ret,
1801                               unsigned long max_pages,
1802                               bio_end_io_t end_io_func,
1803                               int mirror_num,
1804                               unsigned long prev_bio_flags,
1805                               unsigned long bio_flags)
1806 {
1807         int ret = 0;
1808         struct bio *bio;
1809         int nr;
1810         int contig = 0;
1811         int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED;
1812         int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
1813         size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE);
1814
1815         if (bio_ret && *bio_ret) {
1816                 bio = *bio_ret;
1817                 if (old_compressed)
1818                         contig = bio->bi_sector == sector;
1819                 else
1820                         contig = bio->bi_sector + (bio->bi_size >> 9) ==
1821                                 sector;
1822
1823                 if (prev_bio_flags != bio_flags || !contig ||
1824                     (tree->ops && tree->ops->merge_bio_hook &&
1825                      tree->ops->merge_bio_hook(page, offset, page_size, bio,
1826                                                bio_flags)) ||
1827                     bio_add_page(bio, page, page_size, offset) < page_size) {
1828                         ret = submit_one_bio(rw, bio, mirror_num,
1829                                              prev_bio_flags);
1830                         bio = NULL;
1831                 } else {
1832                         return 0;
1833                 }
1834         }
1835         if (this_compressed)
1836                 nr = BIO_MAX_PAGES;
1837         else
1838                 nr = bio_get_nr_vecs(bdev);
1839
1840         bio = extent_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
1841
1842         bio_add_page(bio, page, page_size, offset);
1843         bio->bi_end_io = end_io_func;
1844         bio->bi_private = tree;
1845
1846         if (bio_ret)
1847                 *bio_ret = bio;
1848         else
1849                 ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
1850
1851         return ret;
1852 }
1853
1854 void set_page_extent_mapped(struct page *page)
1855 {
1856         if (!PagePrivate(page)) {
1857                 SetPagePrivate(page);
1858                 page_cache_get(page);
1859                 set_page_private(page, EXTENT_PAGE_PRIVATE);
1860         }
1861 }
1862
1863 static void set_page_extent_head(struct page *page, unsigned long len)
1864 {
1865         set_page_private(page, EXTENT_PAGE_PRIVATE_FIRST_PAGE | len << 2);
1866 }
1867
1868 /*
1869  * basic readpage implementation.  Locked extent state structs are inserted
1870  * into the tree that are removed when the IO is done (by the end_io
1871  * handlers)
1872  */
1873 static int __extent_read_full_page(struct extent_io_tree *tree,
1874                                    struct page *page,
1875                                    get_extent_t *get_extent,
1876                                    struct bio **bio, int mirror_num,
1877                                    unsigned long *bio_flags)
1878 {
1879         struct inode *inode = page->mapping->host;
1880         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1881         u64 page_end = start + PAGE_CACHE_SIZE - 1;
1882         u64 end;
1883         u64 cur = start;
1884         u64 extent_offset;
1885         u64 last_byte = i_size_read(inode);
1886         u64 block_start;
1887         u64 cur_end;
1888         sector_t sector;
1889         struct extent_map *em;
1890         struct block_device *bdev;
1891         int ret;
1892         int nr = 0;
1893         size_t page_offset = 0;
1894         size_t iosize;
1895         size_t disk_io_size;
1896         size_t blocksize = inode->i_sb->s_blocksize;
1897         unsigned long this_bio_flag = 0;
1898
1899         set_page_extent_mapped(page);
1900
1901         end = page_end;
1902         lock_extent(tree, start, end, GFP_NOFS);
1903
1904         if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
1905                 char *userpage;
1906                 size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
1907
1908                 if (zero_offset) {
1909                         iosize = PAGE_CACHE_SIZE - zero_offset;
1910                         userpage = kmap_atomic(page, KM_USER0);
1911                         memset(userpage + zero_offset, 0, iosize);
1912                         flush_dcache_page(page);
1913                         kunmap_atomic(userpage, KM_USER0);
1914                 }
1915         }
1916         while (cur <= end) {
1917                 if (cur >= last_byte) {
1918                         char *userpage;
1919                         iosize = PAGE_CACHE_SIZE - page_offset;
1920                         userpage = kmap_atomic(page, KM_USER0);
1921                         memset(userpage + page_offset, 0, iosize);
1922                         flush_dcache_page(page);
1923                         kunmap_atomic(userpage, KM_USER0);
1924                         set_extent_uptodate(tree, cur, cur + iosize - 1,
1925                                             GFP_NOFS);
1926                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1927                         break;
1928                 }
1929                 em = get_extent(inode, page, page_offset, cur,
1930                                 end - cur + 1, 0);
1931                 if (IS_ERR(em) || !em) {
1932                         SetPageError(page);
1933                         unlock_extent(tree, cur, end, GFP_NOFS);
1934                         break;
1935                 }
1936                 extent_offset = cur - em->start;
1937                 BUG_ON(extent_map_end(em) <= cur);
1938                 BUG_ON(end < cur);
1939
1940                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
1941                         this_bio_flag = EXTENT_BIO_COMPRESSED;
1942
1943                 iosize = min(extent_map_end(em) - cur, end - cur + 1);
1944                 cur_end = min(extent_map_end(em) - 1, end);
1945                 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
1946                 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
1947                         disk_io_size = em->block_len;
1948                         sector = em->block_start >> 9;
1949                 } else {
1950                         sector = (em->block_start + extent_offset) >> 9;
1951                         disk_io_size = iosize;
1952                 }
1953                 bdev = em->bdev;
1954                 block_start = em->block_start;
1955                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
1956                         block_start = EXTENT_MAP_HOLE;
1957                 free_extent_map(em);
1958                 em = NULL;
1959
1960                 /* we've found a hole, just zero and go on */
1961                 if (block_start == EXTENT_MAP_HOLE) {
1962                         char *userpage;
1963                         userpage = kmap_atomic(page, KM_USER0);
1964                         memset(userpage + page_offset, 0, iosize);
1965                         flush_dcache_page(page);
1966                         kunmap_atomic(userpage, KM_USER0);
1967
1968                         set_extent_uptodate(tree, cur, cur + iosize - 1,
1969                                             GFP_NOFS);
1970                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1971                         cur = cur + iosize;
1972                         page_offset += iosize;
1973                         continue;
1974                 }
1975                 /* the get_extent function already copied into the page */
1976                 if (test_range_bit(tree, cur, cur_end, EXTENT_UPTODATE, 1)) {
1977                         check_page_uptodate(tree, page);
1978                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1979                         cur = cur + iosize;
1980                         page_offset += iosize;
1981                         continue;
1982                 }
1983                 /* we have an inline extent but it didn't get marked up
1984                  * to date.  Error out
1985                  */
1986                 if (block_start == EXTENT_MAP_INLINE) {
1987                         SetPageError(page);
1988                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1989                         cur = cur + iosize;
1990                         page_offset += iosize;
1991                         continue;
1992                 }
1993
1994                 ret = 0;
1995                 if (tree->ops && tree->ops->readpage_io_hook) {
1996                         ret = tree->ops->readpage_io_hook(page, cur,
1997                                                           cur + iosize - 1);
1998                 }
1999                 if (!ret) {
2000                         unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
2001                         pnr -= page->index;
2002                         ret = submit_extent_page(READ, tree, page,
2003                                          sector, disk_io_size, page_offset,
2004                                          bdev, bio, pnr,
2005                                          end_bio_extent_readpage, mirror_num,
2006                                          *bio_flags,
2007                                          this_bio_flag);
2008                         nr++;
2009                         *bio_flags = this_bio_flag;
2010                 }
2011                 if (ret)
2012                         SetPageError(page);
2013                 cur = cur + iosize;
2014                 page_offset += iosize;
2015         }
2016         if (!nr) {
2017                 if (!PageError(page))
2018                         SetPageUptodate(page);
2019                 unlock_page(page);
2020         }
2021         return 0;
2022 }
2023
2024 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
2025                             get_extent_t *get_extent)
2026 {
2027         struct bio *bio = NULL;
2028         unsigned long bio_flags = 0;
2029         int ret;
2030
2031         ret = __extent_read_full_page(tree, page, get_extent, &bio, 0,
2032                                       &bio_flags);
2033         if (bio)
2034                 submit_one_bio(READ, bio, 0, bio_flags);
2035         return ret;
2036 }
2037
2038 static noinline void update_nr_written(struct page *page,
2039                                       struct writeback_control *wbc,
2040                                       unsigned long nr_written)
2041 {
2042         wbc->nr_to_write -= nr_written;
2043         if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
2044             wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
2045                 page->mapping->writeback_index = page->index + nr_written;
2046 }
2047
2048 /*
2049  * the writepage semantics are similar to regular writepage.  extent
2050  * records are inserted to lock ranges in the tree, and as dirty areas
2051  * are found, they are marked writeback.  Then the lock bits are removed
2052  * and the end_io handler clears the writeback ranges
2053  */
2054 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2055                               void *data)
2056 {
2057         struct inode *inode = page->mapping->host;
2058         struct extent_page_data *epd = data;
2059         struct extent_io_tree *tree = epd->tree;
2060         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2061         u64 delalloc_start;
2062         u64 page_end = start + PAGE_CACHE_SIZE - 1;
2063         u64 end;
2064         u64 cur = start;
2065         u64 extent_offset;
2066         u64 last_byte = i_size_read(inode);
2067         u64 block_start;
2068         u64 iosize;
2069         u64 unlock_start;
2070         sector_t sector;
2071         struct extent_map *em;
2072         struct block_device *bdev;
2073         int ret;
2074         int nr = 0;
2075         size_t pg_offset = 0;
2076         size_t blocksize;
2077         loff_t i_size = i_size_read(inode);
2078         unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
2079         u64 nr_delalloc;
2080         u64 delalloc_end;
2081         int page_started;
2082         int compressed;
2083         int write_flags;
2084         unsigned long nr_written = 0;
2085
2086         if (wbc->sync_mode == WB_SYNC_ALL)
2087                 write_flags = WRITE_SYNC_PLUG;
2088         else
2089                 write_flags = WRITE;
2090
2091         WARN_ON(!PageLocked(page));
2092         pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
2093         if (page->index > end_index ||
2094            (page->index == end_index && !pg_offset)) {
2095                 page->mapping->a_ops->invalidatepage(page, 0);
2096                 unlock_page(page);
2097                 return 0;
2098         }
2099
2100         if (page->index == end_index) {
2101                 char *userpage;
2102
2103                 userpage = kmap_atomic(page, KM_USER0);
2104                 memset(userpage + pg_offset, 0,
2105                        PAGE_CACHE_SIZE - pg_offset);
2106                 kunmap_atomic(userpage, KM_USER0);
2107                 flush_dcache_page(page);
2108         }
2109         pg_offset = 0;
2110
2111         set_page_extent_mapped(page);
2112
2113         delalloc_start = start;
2114         delalloc_end = 0;
2115         page_started = 0;
2116         if (!epd->extent_locked) {
2117                 /*
2118                  * make sure the wbc mapping index is at least updated
2119                  * to this page.
2120                  */
2121                 update_nr_written(page, wbc, 0);
2122
2123                 while (delalloc_end < page_end) {
2124                         nr_delalloc = find_lock_delalloc_range(inode, tree,
2125                                                        page,
2126                                                        &delalloc_start,
2127                                                        &delalloc_end,
2128                                                        128 * 1024 * 1024);
2129                         if (nr_delalloc == 0) {
2130                                 delalloc_start = delalloc_end + 1;
2131                                 continue;
2132                         }
2133                         tree->ops->fill_delalloc(inode, page, delalloc_start,
2134                                                  delalloc_end, &page_started,
2135                                                  &nr_written);
2136                         delalloc_start = delalloc_end + 1;
2137                 }
2138
2139                 /* did the fill delalloc function already unlock and start
2140                  * the IO?
2141                  */
2142                 if (page_started) {
2143                         ret = 0;
2144                         /*
2145                          * we've unlocked the page, so we can't update
2146                          * the mapping's writeback index, just update
2147                          * nr_to_write.
2148                          */
2149                         wbc->nr_to_write -= nr_written;
2150                         goto done_unlocked;
2151                 }
2152         }
2153         lock_extent(tree, start, page_end, GFP_NOFS);
2154
2155         unlock_start = start;
2156
2157         if (tree->ops && tree->ops->writepage_start_hook) {
2158                 ret = tree->ops->writepage_start_hook(page, start,
2159                                                       page_end);
2160                 if (ret == -EAGAIN) {
2161                         unlock_extent(tree, start, page_end, GFP_NOFS);
2162                         redirty_page_for_writepage(wbc, page);
2163                         update_nr_written(page, wbc, nr_written);
2164                         unlock_page(page);
2165                         ret = 0;
2166                         goto done_unlocked;
2167                 }
2168         }
2169
2170         /*
2171          * we don't want to touch the inode after unlocking the page,
2172          * so we update the mapping writeback index now
2173          */
2174         update_nr_written(page, wbc, nr_written + 1);
2175
2176         end = page_end;
2177         if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0))
2178                 printk(KERN_ERR "btrfs delalloc bits after lock_extent\n");
2179
2180         if (last_byte <= start) {
2181                 clear_extent_dirty(tree, start, page_end, GFP_NOFS);
2182                 unlock_extent(tree, start, page_end, GFP_NOFS);
2183                 if (tree->ops && tree->ops->writepage_end_io_hook)
2184                         tree->ops->writepage_end_io_hook(page, start,
2185                                                          page_end, NULL, 1);
2186                 unlock_start = page_end + 1;
2187                 goto done;
2188         }
2189
2190         set_extent_uptodate(tree, start, page_end, GFP_NOFS);
2191         blocksize = inode->i_sb->s_blocksize;
2192
2193         while (cur <= end) {
2194                 if (cur >= last_byte) {
2195                         clear_extent_dirty(tree, cur, page_end, GFP_NOFS);
2196                         unlock_extent(tree, unlock_start, page_end, GFP_NOFS);
2197                         if (tree->ops && tree->ops->writepage_end_io_hook)
2198                                 tree->ops->writepage_end_io_hook(page, cur,
2199                                                          page_end, NULL, 1);
2200                         unlock_start = page_end + 1;
2201                         break;
2202                 }
2203                 em = epd->get_extent(inode, page, pg_offset, cur,
2204                                      end - cur + 1, 1);
2205                 if (IS_ERR(em) || !em) {
2206                         SetPageError(page);
2207                         break;
2208                 }
2209
2210                 extent_offset = cur - em->start;
2211                 BUG_ON(extent_map_end(em) <= cur);
2212                 BUG_ON(end < cur);
2213                 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2214                 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2215                 sector = (em->block_start + extent_offset) >> 9;
2216                 bdev = em->bdev;
2217                 block_start = em->block_start;
2218                 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
2219                 free_extent_map(em);
2220                 em = NULL;
2221
2222                 /*
2223                  * compressed and inline extents are written through other
2224                  * paths in the FS
2225                  */
2226                 if (compressed || block_start == EXTENT_MAP_HOLE ||
2227                     block_start == EXTENT_MAP_INLINE) {
2228                         clear_extent_dirty(tree, cur,
2229                                            cur + iosize - 1, GFP_NOFS);
2230
2231                         unlock_extent(tree, unlock_start, cur + iosize - 1,
2232                                       GFP_NOFS);
2233
2234                         /*
2235                          * end_io notification does not happen here for
2236                          * compressed extents
2237                          */
2238                         if (!compressed && tree->ops &&
2239                             tree->ops->writepage_end_io_hook)
2240                                 tree->ops->writepage_end_io_hook(page, cur,
2241                                                          cur + iosize - 1,
2242                                                          NULL, 1);
2243                         else if (compressed) {
2244                                 /* we don't want to end_page_writeback on
2245                                  * a compressed extent.  this happens
2246                                  * elsewhere
2247                                  */
2248                                 nr++;
2249                         }
2250
2251                         cur += iosize;
2252                         pg_offset += iosize;
2253                         unlock_start = cur;
2254                         continue;
2255                 }
2256                 /* leave this out until we have a page_mkwrite call */
2257                 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
2258                                    EXTENT_DIRTY, 0)) {
2259                         cur = cur + iosize;
2260                         pg_offset += iosize;
2261                         continue;
2262                 }
2263
2264                 clear_extent_dirty(tree, cur, cur + iosize - 1, GFP_NOFS);
2265                 if (tree->ops && tree->ops->writepage_io_hook) {
2266                         ret = tree->ops->writepage_io_hook(page, cur,
2267                                                 cur + iosize - 1);
2268                 } else {
2269                         ret = 0;
2270                 }
2271                 if (ret) {
2272                         SetPageError(page);
2273                 } else {
2274                         unsigned long max_nr = end_index + 1;
2275
2276                         set_range_writeback(tree, cur, cur + iosize - 1);
2277                         if (!PageWriteback(page)) {
2278                                 printk(KERN_ERR "btrfs warning page %lu not "
2279                                        "writeback, cur %llu end %llu\n",
2280                                        page->index, (unsigned long long)cur,
2281                                        (unsigned long long)end);
2282                         }
2283
2284                         ret = submit_extent_page(write_flags, tree, page,
2285                                                  sector, iosize, pg_offset,
2286                                                  bdev, &epd->bio, max_nr,
2287                                                  end_bio_extent_writepage,
2288                                                  0, 0, 0);
2289                         if (ret)
2290                                 SetPageError(page);
2291                 }
2292                 cur = cur + iosize;
2293                 pg_offset += iosize;
2294                 nr++;
2295         }
2296 done:
2297         if (nr == 0) {
2298                 /* make sure the mapping tag for page dirty gets cleared */
2299                 set_page_writeback(page);
2300                 end_page_writeback(page);
2301         }
2302         if (unlock_start <= page_end)
2303                 unlock_extent(tree, unlock_start, page_end, GFP_NOFS);
2304         unlock_page(page);
2305
2306 done_unlocked:
2307
2308         return 0;
2309 }
2310
2311 /**
2312  * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
2313  * @mapping: address space structure to write
2314  * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2315  * @writepage: function called for each page
2316  * @data: data passed to writepage function
2317  *
2318  * If a page is already under I/O, write_cache_pages() skips it, even
2319  * if it's dirty.  This is desirable behaviour for memory-cleaning writeback,
2320  * but it is INCORRECT for data-integrity system calls such as fsync().  fsync()
2321  * and msync() need to guarantee that all the data which was dirty at the time
2322  * the call was made get new I/O started against them.  If wbc->sync_mode is
2323  * WB_SYNC_ALL then we were called for data integrity and we must wait for
2324  * existing IO to complete.
2325  */
2326 static int extent_write_cache_pages(struct extent_io_tree *tree,
2327                              struct address_space *mapping,
2328                              struct writeback_control *wbc,
2329                              writepage_t writepage, void *data,
2330                              void (*flush_fn)(void *))
2331 {
2332         struct backing_dev_info *bdi = mapping->backing_dev_info;
2333         int ret = 0;
2334         int done = 0;
2335         struct pagevec pvec;
2336         int nr_pages;
2337         pgoff_t index;
2338         pgoff_t end;            /* Inclusive */
2339         int scanned = 0;
2340         int range_whole = 0;
2341
2342         pagevec_init(&pvec, 0);
2343         if (wbc->range_cyclic) {
2344                 index = mapping->writeback_index; /* Start from prev offset */
2345                 end = -1;
2346         } else {
2347                 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2348                 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2349                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2350                         range_whole = 1;
2351                 scanned = 1;
2352         }
2353 retry:
2354         while (!done && (index <= end) &&
2355                (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
2356                               PAGECACHE_TAG_DIRTY, min(end - index,
2357                                   (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
2358                 unsigned i;
2359
2360                 scanned = 1;
2361                 for (i = 0; i < nr_pages; i++) {
2362                         struct page *page = pvec.pages[i];
2363
2364                         /*
2365                          * At this point we hold neither mapping->tree_lock nor
2366                          * lock on the page itself: the page may be truncated or
2367                          * invalidated (changing page->mapping to NULL), or even
2368                          * swizzled back from swapper_space to tmpfs file
2369                          * mapping
2370                          */
2371                         if (tree->ops && tree->ops->write_cache_pages_lock_hook)
2372                                 tree->ops->write_cache_pages_lock_hook(page);
2373                         else
2374                                 lock_page(page);
2375
2376                         if (unlikely(page->mapping != mapping)) {
2377                                 unlock_page(page);
2378                                 continue;
2379                         }
2380
2381                         if (!wbc->range_cyclic && page->index > end) {
2382                                 done = 1;
2383                                 unlock_page(page);
2384                                 continue;
2385                         }
2386
2387                         if (wbc->sync_mode != WB_SYNC_NONE) {
2388                                 if (PageWriteback(page))
2389                                         flush_fn(data);
2390                                 wait_on_page_writeback(page);
2391                         }
2392
2393                         if (PageWriteback(page) ||
2394                             !clear_page_dirty_for_io(page)) {
2395                                 unlock_page(page);
2396                                 continue;
2397                         }
2398
2399                         ret = (*writepage)(page, wbc, data);
2400
2401                         if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
2402                                 unlock_page(page);
2403                                 ret = 0;
2404                         }
2405                         if (ret || wbc->nr_to_write <= 0)
2406                                 done = 1;
2407                         if (wbc->nonblocking && bdi_write_congested(bdi)) {
2408                                 wbc->encountered_congestion = 1;
2409                                 done = 1;
2410                         }
2411                 }
2412                 pagevec_release(&pvec);
2413                 cond_resched();
2414         }
2415         if (!scanned && !done) {
2416                 /*
2417                  * We hit the last page and there is more work to be done: wrap
2418                  * back to the start of the file
2419                  */
2420                 scanned = 1;
2421                 index = 0;
2422                 goto retry;
2423         }
2424         return ret;
2425 }
2426
2427 static void flush_epd_write_bio(struct extent_page_data *epd)
2428 {
2429         if (epd->bio) {
2430                 if (epd->sync_io)
2431                         submit_one_bio(WRITE_SYNC, epd->bio, 0, 0);
2432                 else
2433                         submit_one_bio(WRITE, epd->bio, 0, 0);
2434                 epd->bio = NULL;
2435         }
2436 }
2437
2438 static noinline void flush_write_bio(void *data)
2439 {
2440         struct extent_page_data *epd = data;
2441         flush_epd_write_bio(epd);
2442 }
2443
2444 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
2445                           get_extent_t *get_extent,
2446                           struct writeback_control *wbc)
2447 {
2448         int ret;
2449         struct address_space *mapping = page->mapping;
2450         struct extent_page_data epd = {
2451                 .bio = NULL,
2452                 .tree = tree,
2453                 .get_extent = get_extent,
2454                 .extent_locked = 0,
2455                 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
2456         };
2457         struct writeback_control wbc_writepages = {
2458                 .bdi            = wbc->bdi,
2459                 .sync_mode      = wbc->sync_mode,
2460                 .older_than_this = NULL,
2461                 .nr_to_write    = 64,
2462                 .range_start    = page_offset(page) + PAGE_CACHE_SIZE,
2463                 .range_end      = (loff_t)-1,
2464         };
2465
2466         ret = __extent_writepage(page, wbc, &epd);
2467
2468         extent_write_cache_pages(tree, mapping, &wbc_writepages,
2469                                  __extent_writepage, &epd, flush_write_bio);
2470         flush_epd_write_bio(&epd);
2471         return ret;
2472 }
2473
2474 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
2475                               u64 start, u64 end, get_extent_t *get_extent,
2476                               int mode)
2477 {
2478         int ret = 0;
2479         struct address_space *mapping = inode->i_mapping;
2480         struct page *page;
2481         unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
2482                 PAGE_CACHE_SHIFT;
2483
2484         struct extent_page_data epd = {
2485                 .bio = NULL,
2486                 .tree = tree,
2487                 .get_extent = get_extent,
2488                 .extent_locked = 1,
2489                 .sync_io = mode == WB_SYNC_ALL,
2490         };
2491         struct writeback_control wbc_writepages = {
2492                 .bdi            = inode->i_mapping->backing_dev_info,
2493                 .sync_mode      = mode,
2494                 .older_than_this = NULL,
2495                 .nr_to_write    = nr_pages * 2,
2496                 .range_start    = start,
2497                 .range_end      = end + 1,
2498         };
2499
2500         while (start <= end) {
2501                 page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
2502                 if (clear_page_dirty_for_io(page))
2503                         ret = __extent_writepage(page, &wbc_writepages, &epd);
2504                 else {
2505                         if (tree->ops && tree->ops->writepage_end_io_hook)
2506                                 tree->ops->writepage_end_io_hook(page, start,
2507                                                  start + PAGE_CACHE_SIZE - 1,
2508                                                  NULL, 1);
2509                         unlock_page(page);
2510                 }
2511                 page_cache_release(page);
2512                 start += PAGE_CACHE_SIZE;
2513         }
2514
2515         flush_epd_write_bio(&epd);
2516         return ret;
2517 }
2518
2519 int extent_writepages(struct extent_io_tree *tree,
2520                       struct address_space *mapping,
2521                       get_extent_t *get_extent,
2522                       struct writeback_control *wbc)
2523 {
2524         int ret = 0;
2525         struct extent_page_data epd = {
2526                 .bio = NULL,
2527                 .tree = tree,
2528                 .get_extent = get_extent,
2529                 .extent_locked = 0,
2530                 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
2531         };
2532
2533         ret = extent_write_cache_pages(tree, mapping, wbc,
2534                                        __extent_writepage, &epd,
2535                                        flush_write_bio);
2536         flush_epd_write_bio(&epd);
2537         return ret;
2538 }
2539
2540 int extent_readpages(struct extent_io_tree *tree,
2541                      struct address_space *mapping,
2542                      struct list_head *pages, unsigned nr_pages,
2543                      get_extent_t get_extent)
2544 {
2545         struct bio *bio = NULL;
2546         unsigned page_idx;
2547         struct pagevec pvec;
2548         unsigned long bio_flags = 0;
2549
2550         pagevec_init(&pvec, 0);
2551         for (page_idx = 0; page_idx < nr_pages; page_idx++) {
2552                 struct page *page = list_entry(pages->prev, struct page, lru);
2553
2554                 prefetchw(&page->flags);
2555                 list_del(&page->lru);
2556                 /*
2557                  * what we want to do here is call add_to_page_cache_lru,
2558                  * but that isn't exported, so we reproduce it here
2559                  */
2560                 if (!add_to_page_cache(page, mapping,
2561                                         page->index, GFP_KERNEL)) {
2562
2563                         /* open coding of lru_cache_add, also not exported */
2564                         page_cache_get(page);
2565                         if (!pagevec_add(&pvec, page))
2566                                 __pagevec_lru_add_file(&pvec);
2567                         __extent_read_full_page(tree, page, get_extent,
2568                                                 &bio, 0, &bio_flags);
2569                 }
2570                 page_cache_release(page);
2571         }
2572         if (pagevec_count(&pvec))
2573                 __pagevec_lru_add_file(&pvec);
2574         BUG_ON(!list_empty(pages));
2575         if (bio)
2576                 submit_one_bio(READ, bio, 0, bio_flags);
2577         return 0;
2578 }
2579
2580 /*
2581  * basic invalidatepage code, this waits on any locked or writeback
2582  * ranges corresponding to the page, and then deletes any extent state
2583  * records from the tree
2584  */
2585 int extent_invalidatepage(struct extent_io_tree *tree,
2586                           struct page *page, unsigned long offset)
2587 {
2588         u64 start = ((u64)page->index << PAGE_CACHE_SHIFT);
2589         u64 end = start + PAGE_CACHE_SIZE - 1;
2590         size_t blocksize = page->mapping->host->i_sb->s_blocksize;
2591
2592         start += (offset + blocksize - 1) & ~(blocksize - 1);
2593         if (start > end)
2594                 return 0;
2595
2596         lock_extent(tree, start, end, GFP_NOFS);
2597         wait_on_extent_writeback(tree, start, end);
2598         clear_extent_bit(tree, start, end,
2599                          EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC,
2600                          1, 1, GFP_NOFS);
2601         return 0;
2602 }
2603
2604 /*
2605  * simple commit_write call, set_range_dirty is used to mark both
2606  * the pages and the extent records as dirty
2607  */
2608 int extent_commit_write(struct extent_io_tree *tree,
2609                         struct inode *inode, struct page *page,
2610                         unsigned from, unsigned to)
2611 {
2612         loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
2613
2614         set_page_extent_mapped(page);
2615         set_page_dirty(page);
2616
2617         if (pos > inode->i_size) {
2618                 i_size_write(inode, pos);
2619                 mark_inode_dirty(inode);
2620         }
2621         return 0;
2622 }
2623
2624 int extent_prepare_write(struct extent_io_tree *tree,
2625                          struct inode *inode, struct page *page,
2626                          unsigned from, unsigned to, get_extent_t *get_extent)
2627 {
2628         u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2629         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
2630         u64 block_start;
2631         u64 orig_block_start;
2632         u64 block_end;
2633         u64 cur_end;
2634         struct extent_map *em;
2635         unsigned blocksize = 1 << inode->i_blkbits;
2636         size_t page_offset = 0;
2637         size_t block_off_start;
2638         size_t block_off_end;
2639         int err = 0;
2640         int iocount = 0;
2641         int ret = 0;
2642         int isnew;
2643
2644         set_page_extent_mapped(page);
2645
2646         block_start = (page_start + from) & ~((u64)blocksize - 1);
2647         block_end = (page_start + to - 1) | (blocksize - 1);
2648         orig_block_start = block_start;
2649
2650         lock_extent(tree, page_start, page_end, GFP_NOFS);
2651         while (block_start <= block_end) {
2652                 em = get_extent(inode, page, page_offset, block_start,
2653                                 block_end - block_start + 1, 1);
2654                 if (IS_ERR(em) || !em)
2655                         goto err;
2656
2657                 cur_end = min(block_end, extent_map_end(em) - 1);
2658                 block_off_start = block_start & (PAGE_CACHE_SIZE - 1);
2659                 block_off_end = block_off_start + blocksize;
2660                 isnew = clear_extent_new(tree, block_start, cur_end, GFP_NOFS);
2661
2662                 if (!PageUptodate(page) && isnew &&
2663                     (block_off_end > to || block_off_start < from)) {
2664                         void *kaddr;
2665
2666                         kaddr = kmap_atomic(page, KM_USER0);
2667                         if (block_off_end > to)
2668                                 memset(kaddr + to, 0, block_off_end - to);
2669                         if (block_off_start < from)
2670                                 memset(kaddr + block_off_start, 0,
2671                                        from - block_off_start);
2672                         flush_dcache_page(page);
2673                         kunmap_atomic(kaddr, KM_USER0);
2674                 }
2675                 if ((em->block_start != EXTENT_MAP_HOLE &&
2676                      em->block_start != EXTENT_MAP_INLINE) &&
2677                     !isnew && !PageUptodate(page) &&
2678                     (block_off_end > to || block_off_start < from) &&
2679                     !test_range_bit(tree, block_start, cur_end,
2680                                     EXTENT_UPTODATE, 1)) {
2681                         u64 sector;
2682                         u64 extent_offset = block_start - em->start;
2683                         size_t iosize;
2684                         sector = (em->block_start + extent_offset) >> 9;
2685                         iosize = (cur_end - block_start + blocksize) &
2686                                 ~((u64)blocksize - 1);
2687                         /*
2688                          * we've already got the extent locked, but we
2689                          * need to split the state such that our end_bio
2690                          * handler can clear the lock.
2691                          */
2692                         set_extent_bit(tree, block_start,
2693                                        block_start + iosize - 1,
2694                                        EXTENT_LOCKED, 0, NULL, GFP_NOFS);
2695                         ret = submit_extent_page(READ, tree, page,
2696                                          sector, iosize, page_offset, em->bdev,
2697                                          NULL, 1,
2698                                          end_bio_extent_preparewrite, 0,
2699                                          0, 0);
2700                         iocount++;
2701                         block_start = block_start + iosize;
2702                 } else {
2703                         set_extent_uptodate(tree, block_start, cur_end,
2704                                             GFP_NOFS);
2705                         unlock_extent(tree, block_start, cur_end, GFP_NOFS);
2706                         block_start = cur_end + 1;
2707                 }
2708                 page_offset = block_start & (PAGE_CACHE_SIZE - 1);
2709                 free_extent_map(em);
2710         }
2711         if (iocount) {
2712                 wait_extent_bit(tree, orig_block_start,
2713                                 block_end, EXTENT_LOCKED);
2714         }
2715         check_page_uptodate(tree, page);
2716 err:
2717         /* FIXME, zero out newly allocated blocks on error */
2718         return err;
2719 }
2720
2721 /*
2722  * a helper for releasepage, this tests for areas of the page that
2723  * are locked or under IO and drops the related state bits if it is safe
2724  * to drop the page.
2725  */
2726 int try_release_extent_state(struct extent_map_tree *map,
2727                              struct extent_io_tree *tree, struct page *page,
2728                              gfp_t mask)
2729 {
2730         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2731         u64 end = start + PAGE_CACHE_SIZE - 1;
2732         int ret = 1;
2733
2734         if (test_range_bit(tree, start, end,
2735                            EXTENT_IOBITS | EXTENT_ORDERED, 0))
2736                 ret = 0;
2737         else {
2738                 if ((mask & GFP_NOFS) == GFP_NOFS)
2739                         mask = GFP_NOFS;
2740                 clear_extent_bit(tree, start, end, EXTENT_UPTODATE,
2741                                  1, 1, mask);
2742         }
2743         return ret;
2744 }
2745
2746 /*
2747  * a helper for releasepage.  As long as there are no locked extents
2748  * in the range corresponding to the page, both state records and extent
2749  * map records are removed
2750  */
2751 int try_release_extent_mapping(struct extent_map_tree *map,
2752                                struct extent_io_tree *tree, struct page *page,
2753                                gfp_t mask)
2754 {
2755         struct extent_map *em;
2756         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2757         u64 end = start + PAGE_CACHE_SIZE - 1;
2758
2759         if ((mask & __GFP_WAIT) &&
2760             page->mapping->host->i_size > 16 * 1024 * 1024) {
2761                 u64 len;
2762                 while (start <= end) {
2763                         len = end - start + 1;
2764                         spin_lock(&map->lock);
2765                         em = lookup_extent_mapping(map, start, len);
2766                         if (!em || IS_ERR(em)) {
2767                                 spin_unlock(&map->lock);
2768                                 break;
2769                         }
2770                         if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
2771                             em->start != start) {
2772                                 spin_unlock(&map->lock);
2773                                 free_extent_map(em);
2774                                 break;
2775                         }
2776                         if (!test_range_bit(tree, em->start,
2777                                             extent_map_end(em) - 1,
2778                                             EXTENT_LOCKED | EXTENT_WRITEBACK |
2779                                             EXTENT_ORDERED,
2780                                             0)) {
2781                                 remove_extent_mapping(map, em);
2782                                 /* once for the rb tree */
2783                                 free_extent_map(em);
2784                         }
2785                         start = extent_map_end(em);
2786                         spin_unlock(&map->lock);
2787
2788                         /* once for us */
2789                         free_extent_map(em);
2790                 }
2791         }
2792         return try_release_extent_state(map, tree, page, mask);
2793 }
2794
2795 sector_t extent_bmap(struct address_space *mapping, sector_t iblock,
2796                 get_extent_t *get_extent)
2797 {
2798         struct inode *inode = mapping->host;
2799         u64 start = iblock << inode->i_blkbits;
2800         sector_t sector = 0;
2801         size_t blksize = (1 << inode->i_blkbits);
2802         struct extent_map *em;
2803
2804         lock_extent(&BTRFS_I(inode)->io_tree, start, start + blksize - 1,
2805                     GFP_NOFS);
2806         em = get_extent(inode, NULL, 0, start, blksize, 0);
2807         unlock_extent(&BTRFS_I(inode)->io_tree, start, start + blksize - 1,
2808                       GFP_NOFS);
2809         if (!em || IS_ERR(em))
2810                 return 0;
2811
2812         if (em->block_start > EXTENT_MAP_LAST_BYTE)
2813                 goto out;
2814
2815         sector = (em->block_start + start - em->start) >> inode->i_blkbits;
2816 out:
2817         free_extent_map(em);
2818         return sector;
2819 }
2820
2821 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
2822                 __u64 start, __u64 len, get_extent_t *get_extent)
2823 {
2824         int ret;
2825         u64 off = start;
2826         u64 max = start + len;
2827         u32 flags = 0;
2828         u64 disko = 0;
2829         struct extent_map *em = NULL;
2830         int end = 0;
2831         u64 em_start = 0, em_len = 0;
2832         unsigned long emflags;
2833         ret = 0;
2834
2835         if (len == 0)
2836                 return -EINVAL;
2837
2838         lock_extent(&BTRFS_I(inode)->io_tree, start, start + len,
2839                 GFP_NOFS);
2840         em = get_extent(inode, NULL, 0, off, max - off, 0);
2841         if (!em)
2842                 goto out;
2843         if (IS_ERR(em)) {
2844                 ret = PTR_ERR(em);
2845                 goto out;
2846         }
2847         while (!end) {
2848                 off = em->start + em->len;
2849                 if (off >= max)
2850                         end = 1;
2851
2852                 em_start = em->start;
2853                 em_len = em->len;
2854
2855                 disko = 0;
2856                 flags = 0;
2857
2858                 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
2859                         end = 1;
2860                         flags |= FIEMAP_EXTENT_LAST;
2861                 } else if (em->block_start == EXTENT_MAP_HOLE) {
2862                         flags |= FIEMAP_EXTENT_UNWRITTEN;
2863                 } else if (em->block_start == EXTENT_MAP_INLINE) {
2864                         flags |= (FIEMAP_EXTENT_DATA_INLINE |
2865                                   FIEMAP_EXTENT_NOT_ALIGNED);
2866                 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
2867                         flags |= (FIEMAP_EXTENT_DELALLOC |
2868                                   FIEMAP_EXTENT_UNKNOWN);
2869                 } else {
2870                         disko = em->block_start;
2871                 }
2872                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
2873                         flags |= FIEMAP_EXTENT_ENCODED;
2874
2875                 emflags = em->flags;
2876                 free_extent_map(em);
2877                 em = NULL;
2878
2879                 if (!end) {
2880                         em = get_extent(inode, NULL, 0, off, max - off, 0);
2881                         if (!em)
2882                                 goto out;
2883                         if (IS_ERR(em)) {
2884                                 ret = PTR_ERR(em);
2885                                 goto out;
2886                         }
2887                         emflags = em->flags;
2888                 }
2889                 if (test_bit(EXTENT_FLAG_VACANCY, &emflags)) {
2890                         flags |= FIEMAP_EXTENT_LAST;
2891                         end = 1;
2892                 }
2893
2894                 ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
2895                                         em_len, flags);
2896                 if (ret)
2897                         goto out_free;
2898         }
2899 out_free:
2900         free_extent_map(em);
2901 out:
2902         unlock_extent(&BTRFS_I(inode)->io_tree, start, start + len,
2903                         GFP_NOFS);
2904         return ret;
2905 }
2906
2907 static inline struct page *extent_buffer_page(struct extent_buffer *eb,
2908                                               unsigned long i)
2909 {
2910         struct page *p;
2911         struct address_space *mapping;
2912
2913         if (i == 0)
2914                 return eb->first_page;
2915         i += eb->start >> PAGE_CACHE_SHIFT;
2916         mapping = eb->first_page->mapping;
2917         if (!mapping)
2918                 return NULL;
2919
2920         /*
2921          * extent_buffer_page is only called after pinning the page
2922          * by increasing the reference count.  So we know the page must
2923          * be in the radix tree.
2924          */
2925         rcu_read_lock();
2926         p = radix_tree_lookup(&mapping->page_tree, i);
2927         rcu_read_unlock();
2928
2929         return p;
2930 }
2931
2932 static inline unsigned long num_extent_pages(u64 start, u64 len)
2933 {
2934         return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
2935                 (start >> PAGE_CACHE_SHIFT);
2936 }
2937
2938 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
2939                                                    u64 start,
2940                                                    unsigned long len,
2941                                                    gfp_t mask)
2942 {
2943         struct extent_buffer *eb = NULL;
2944 #if LEAK_DEBUG
2945         unsigned long flags;
2946 #endif
2947
2948         eb = kmem_cache_zalloc(extent_buffer_cache, mask);
2949         eb->start = start;
2950         eb->len = len;
2951         spin_lock_init(&eb->lock);
2952         init_waitqueue_head(&eb->lock_wq);
2953
2954 #if LEAK_DEBUG
2955         spin_lock_irqsave(&leak_lock, flags);
2956         list_add(&eb->leak_list, &buffers);
2957         spin_unlock_irqrestore(&leak_lock, flags);
2958 #endif
2959         atomic_set(&eb->refs, 1);
2960
2961         return eb;
2962 }
2963
2964 static void __free_extent_buffer(struct extent_buffer *eb)
2965 {
2966 #if LEAK_DEBUG
2967         unsigned long flags;
2968         spin_lock_irqsave(&leak_lock, flags);
2969         list_del(&eb->leak_list);
2970         spin_unlock_irqrestore(&leak_lock, flags);
2971 #endif
2972         kmem_cache_free(extent_buffer_cache, eb);
2973 }
2974
2975 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
2976                                           u64 start, unsigned long len,
2977                                           struct page *page0,
2978                                           gfp_t mask)
2979 {
2980         unsigned long num_pages = num_extent_pages(start, len);
2981         unsigned long i;
2982         unsigned long index = start >> PAGE_CACHE_SHIFT;
2983         struct extent_buffer *eb;
2984         struct extent_buffer *exists = NULL;
2985         struct page *p;
2986         struct address_space *mapping = tree->mapping;
2987         int uptodate = 1;
2988
2989         spin_lock(&tree->buffer_lock);
2990         eb = buffer_search(tree, start);
2991         if (eb) {
2992                 atomic_inc(&eb->refs);
2993                 spin_unlock(&tree->buffer_lock);
2994                 mark_page_accessed(eb->first_page);
2995                 return eb;
2996         }
2997         spin_unlock(&tree->buffer_lock);
2998
2999         eb = __alloc_extent_buffer(tree, start, len, mask);
3000         if (!eb)
3001                 return NULL;
3002
3003         if (page0) {
3004                 eb->first_page = page0;
3005                 i = 1;
3006                 index++;
3007                 page_cache_get(page0);
3008                 mark_page_accessed(page0);
3009                 set_page_extent_mapped(page0);
3010                 set_page_extent_head(page0, len);
3011                 uptodate = PageUptodate(page0);
3012         } else {
3013                 i = 0;
3014         }
3015         for (; i < num_pages; i++, index++) {
3016                 p = find_or_create_page(mapping, index, mask | __GFP_HIGHMEM);
3017                 if (!p) {
3018                         WARN_ON(1);
3019                         goto free_eb;
3020                 }
3021                 set_page_extent_mapped(p);
3022                 mark_page_accessed(p);
3023                 if (i == 0) {
3024                         eb->first_page = p;
3025                         set_page_extent_head(p, len);
3026                 } else {
3027                         set_page_private(p, EXTENT_PAGE_PRIVATE);
3028                 }
3029                 if (!PageUptodate(p))
3030                         uptodate = 0;
3031                 unlock_page(p);
3032         }
3033         if (uptodate)
3034                 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3035
3036         spin_lock(&tree->buffer_lock);
3037         exists = buffer_tree_insert(tree, start, &eb->rb_node);
3038         if (exists) {
3039                 /* add one reference for the caller */
3040                 atomic_inc(&exists->refs);
3041                 spin_unlock(&tree->buffer_lock);
3042                 goto free_eb;
3043         }
3044         spin_unlock(&tree->buffer_lock);
3045
3046         /* add one reference for the tree */
3047         atomic_inc(&eb->refs);
3048         return eb;
3049
3050 free_eb:
3051         if (!atomic_dec_and_test(&eb->refs))
3052                 return exists;
3053         for (index = 1; index < i; index++)
3054                 page_cache_release(extent_buffer_page(eb, index));
3055         page_cache_release(extent_buffer_page(eb, 0));
3056         __free_extent_buffer(eb);
3057         return exists;
3058 }
3059
3060 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
3061                                          u64 start, unsigned long len,
3062                                           gfp_t mask)
3063 {
3064         struct extent_buffer *eb;
3065
3066         spin_lock(&tree->buffer_lock);
3067         eb = buffer_search(tree, start);
3068         if (eb)
3069                 atomic_inc(&eb->refs);
3070         spin_unlock(&tree->buffer_lock);
3071
3072         if (eb)
3073                 mark_page_accessed(eb->first_page);
3074
3075         return eb;
3076 }
3077
3078 void free_extent_buffer(struct extent_buffer *eb)
3079 {
3080         if (!eb)
3081                 return;
3082
3083         if (!atomic_dec_and_test(&eb->refs))
3084                 return;
3085
3086         WARN_ON(1);
3087 }
3088
3089 int clear_extent_buffer_dirty(struct extent_io_tree *tree,
3090                               struct extent_buffer *eb)
3091 {
3092         unsigned long i;
3093         unsigned long num_pages;
3094         struct page *page;
3095
3096         num_pages = num_extent_pages(eb->start, eb->len);
3097
3098         for (i = 0; i < num_pages; i++) {
3099                 page = extent_buffer_page(eb, i);
3100                 if (!PageDirty(page))
3101                         continue;
3102
3103                 lock_page(page);
3104                 if (i == 0)
3105                         set_page_extent_head(page, eb->len);
3106                 else
3107                         set_page_private(page, EXTENT_PAGE_PRIVATE);
3108
3109                 clear_page_dirty_for_io(page);
3110                 spin_lock_irq(&page->mapping->tree_lock);
3111                 if (!PageDirty(page)) {
3112                         radix_tree_tag_clear(&page->mapping->page_tree,
3113                                                 page_index(page),
3114                                                 PAGECACHE_TAG_DIRTY);
3115                 }
3116                 spin_unlock_irq(&page->mapping->tree_lock);
3117                 unlock_page(page);
3118         }
3119         return 0;
3120 }
3121
3122 int wait_on_extent_buffer_writeback(struct extent_io_tree *tree,
3123                                     struct extent_buffer *eb)
3124 {
3125         return wait_on_extent_writeback(tree, eb->start,
3126                                         eb->start + eb->len - 1);
3127 }
3128
3129 int set_extent_buffer_dirty(struct extent_io_tree *tree,
3130                              struct extent_buffer *eb)
3131 {
3132         unsigned long i;
3133         unsigned long num_pages;
3134         int was_dirty = 0;
3135
3136         was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
3137         num_pages = num_extent_pages(eb->start, eb->len);
3138         for (i = 0; i < num_pages; i++)
3139                 __set_page_dirty_nobuffers(extent_buffer_page(eb, i));
3140         return was_dirty;
3141 }
3142
3143 int clear_extent_buffer_uptodate(struct extent_io_tree *tree,
3144                                 struct extent_buffer *eb)
3145 {
3146         unsigned long i;
3147         struct page *page;
3148         unsigned long num_pages;
3149
3150         num_pages = num_extent_pages(eb->start, eb->len);
3151         clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3152
3153         clear_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
3154                               GFP_NOFS);
3155         for (i = 0; i < num_pages; i++) {
3156                 page = extent_buffer_page(eb, i);
3157                 if (page)
3158                         ClearPageUptodate(page);
3159         }
3160         return 0;
3161 }
3162
3163 int set_extent_buffer_uptodate(struct extent_io_tree *tree,
3164                                 struct extent_buffer *eb)
3165 {
3166         unsigned long i;
3167         struct page *page;
3168         unsigned long num_pages;
3169
3170         num_pages = num_extent_pages(eb->start, eb->len);
3171
3172         set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
3173                             GFP_NOFS);
3174         for (i = 0; i < num_pages; i++) {
3175                 page = extent_buffer_page(eb, i);
3176                 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
3177                     ((i == num_pages - 1) &&
3178                      ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
3179                         check_page_uptodate(tree, page);
3180                         continue;
3181                 }
3182                 SetPageUptodate(page);
3183         }
3184         return 0;
3185 }
3186
3187 int extent_range_uptodate(struct extent_io_tree *tree,
3188                           u64 start, u64 end)
3189 {
3190         struct page *page;
3191         int ret;
3192         int pg_uptodate = 1;
3193         int uptodate;
3194         unsigned long index;
3195
3196         ret = test_range_bit(tree, start, end, EXTENT_UPTODATE, 1);
3197         if (ret)
3198                 return 1;
3199         while (start <= end) {
3200                 index = start >> PAGE_CACHE_SHIFT;
3201                 page = find_get_page(tree->mapping, index);
3202                 uptodate = PageUptodate(page);
3203                 page_cache_release(page);
3204                 if (!uptodate) {
3205                         pg_uptodate = 0;
3206                         break;
3207                 }
3208                 start += PAGE_CACHE_SIZE;
3209         }
3210         return pg_uptodate;
3211 }
3212
3213 int extent_buffer_uptodate(struct extent_io_tree *tree,
3214                            struct extent_buffer *eb)
3215 {
3216         int ret = 0;
3217         unsigned long num_pages;
3218         unsigned long i;
3219         struct page *page;
3220         int pg_uptodate = 1;
3221
3222         if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
3223                 return 1;
3224
3225         ret = test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3226                            EXTENT_UPTODATE, 1);
3227         if (ret)
3228                 return ret;
3229
3230         num_pages = num_extent_pages(eb->start, eb->len);
3231         for (i = 0; i < num_pages; i++) {
3232                 page = extent_buffer_page(eb, i);
3233                 if (!PageUptodate(page)) {
3234                         pg_uptodate = 0;
3235                         break;
3236                 }
3237         }
3238         return pg_uptodate;
3239 }
3240
3241 int read_extent_buffer_pages(struct extent_io_tree *tree,
3242                              struct extent_buffer *eb,
3243                              u64 start, int wait,
3244                              get_extent_t *get_extent, int mirror_num)
3245 {
3246         unsigned long i;
3247         unsigned long start_i;
3248         struct page *page;
3249         int err;
3250         int ret = 0;
3251         int locked_pages = 0;
3252         int all_uptodate = 1;
3253         int inc_all_pages = 0;
3254         unsigned long num_pages;
3255         struct bio *bio = NULL;
3256         unsigned long bio_flags = 0;
3257
3258         if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
3259                 return 0;
3260
3261         if (test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3262                            EXTENT_UPTODATE, 1)) {
3263                 return 0;
3264         }
3265
3266         if (start) {
3267                 WARN_ON(start < eb->start);
3268                 start_i = (start >> PAGE_CACHE_SHIFT) -
3269                         (eb->start >> PAGE_CACHE_SHIFT);
3270         } else {
3271                 start_i = 0;
3272         }
3273
3274         num_pages = num_extent_pages(eb->start, eb->len);
3275         for (i = start_i; i < num_pages; i++) {
3276                 page = extent_buffer_page(eb, i);
3277                 if (!wait) {
3278                         if (!trylock_page(page))
3279                                 goto unlock_exit;
3280                 } else {
3281                         lock_page(page);
3282                 }
3283                 locked_pages++;
3284                 if (!PageUptodate(page))
3285                         all_uptodate = 0;
3286         }
3287         if (all_uptodate) {
3288                 if (start_i == 0)
3289                         set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3290                 goto unlock_exit;
3291         }
3292
3293         for (i = start_i; i < num_pages; i++) {
3294                 page = extent_buffer_page(eb, i);
3295                 if (inc_all_pages)
3296                         page_cache_get(page);
3297                 if (!PageUptodate(page)) {
3298                         if (start_i == 0)
3299                                 inc_all_pages = 1;
3300                         ClearPageError(page);
3301                         err = __extent_read_full_page(tree, page,
3302                                                       get_extent, &bio,
3303                                                       mirror_num, &bio_flags);
3304                         if (err)
3305                                 ret = err;
3306                 } else {
3307                         unlock_page(page);
3308                 }
3309         }
3310
3311         if (bio)
3312                 submit_one_bio(READ, bio, mirror_num, bio_flags);
3313
3314         if (ret || !wait)
3315                 return ret;
3316
3317         for (i = start_i; i < num_pages; i++) {
3318                 page = extent_buffer_page(eb, i);
3319                 wait_on_page_locked(page);
3320                 if (!PageUptodate(page))
3321                         ret = -EIO;
3322         }
3323
3324         if (!ret)
3325                 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3326         return ret;
3327
3328 unlock_exit:
3329         i = start_i;
3330         while (locked_pages > 0) {
3331                 page = extent_buffer_page(eb, i);
3332                 i++;
3333                 unlock_page(page);
3334                 locked_pages--;
3335         }
3336         return ret;
3337 }
3338
3339 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
3340                         unsigned long start,
3341                         unsigned long len)
3342 {
3343         size_t cur;
3344         size_t offset;
3345         struct page *page;
3346         char *kaddr;
3347         char *dst = (char *)dstv;
3348         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3349         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3350
3351         WARN_ON(start > eb->len);
3352         WARN_ON(start + len > eb->start + eb->len);
3353
3354         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3355
3356         while (len > 0) {
3357                 page = extent_buffer_page(eb, i);
3358
3359                 cur = min(len, (PAGE_CACHE_SIZE - offset));
3360                 kaddr = kmap_atomic(page, KM_USER1);
3361                 memcpy(dst, kaddr + offset, cur);
3362                 kunmap_atomic(kaddr, KM_USER1);
3363
3364                 dst += cur;
3365                 len -= cur;
3366                 offset = 0;
3367                 i++;
3368         }
3369 }
3370
3371 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
3372                                unsigned long min_len, char **token, char **map,
3373                                unsigned long *map_start,
3374                                unsigned long *map_len, int km)
3375 {
3376         size_t offset = start & (PAGE_CACHE_SIZE - 1);
3377         char *kaddr;
3378         struct page *p;
3379         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3380         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3381         unsigned long end_i = (start_offset + start + min_len - 1) >>
3382                 PAGE_CACHE_SHIFT;
3383
3384         if (i != end_i)
3385                 return -EINVAL;
3386
3387         if (i == 0) {
3388                 offset = start_offset;
3389                 *map_start = 0;
3390         } else {
3391                 offset = 0;
3392                 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
3393         }
3394
3395         if (start + min_len > eb->len) {
3396                 printk(KERN_ERR "btrfs bad mapping eb start %llu len %lu, "
3397                        "wanted %lu %lu\n", (unsigned long long)eb->start,
3398                        eb->len, start, min_len);
3399                 WARN_ON(1);
3400         }
3401
3402         p = extent_buffer_page(eb, i);
3403         kaddr = kmap_atomic(p, km);
3404         *token = kaddr;
3405         *map = kaddr + offset;
3406         *map_len = PAGE_CACHE_SIZE - offset;
3407         return 0;
3408 }
3409
3410 int map_extent_buffer(struct extent_buffer *eb, unsigned long start,
3411                       unsigned long min_len,
3412                       char **token, char **map,
3413                       unsigned long *map_start,
3414                       unsigned long *map_len, int km)
3415 {
3416         int err;
3417         int save = 0;
3418         if (eb->map_token) {
3419                 unmap_extent_buffer(eb, eb->map_token, km);
3420                 eb->map_token = NULL;
3421                 save = 1;
3422         }
3423         err = map_private_extent_buffer(eb, start, min_len, token, map,
3424                                        map_start, map_len, km);
3425         if (!err && save) {
3426                 eb->map_token = *token;
3427                 eb->kaddr = *map;
3428                 eb->map_start = *map_start;
3429                 eb->map_len = *map_len;
3430         }
3431         return err;
3432 }
3433
3434 void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km)
3435 {
3436         kunmap_atomic(token, km);
3437 }
3438
3439 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
3440                           unsigned long start,
3441                           unsigned long len)
3442 {
3443         size_t cur;
3444         size_t offset;
3445         struct page *page;
3446         char *kaddr;
3447         char *ptr = (char *)ptrv;
3448         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3449         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3450         int ret = 0;
3451
3452         WARN_ON(start > eb->len);
3453         WARN_ON(start + len > eb->start + eb->len);
3454
3455         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3456
3457         while (len > 0) {
3458                 page = extent_buffer_page(eb, i);
3459
3460                 cur = min(len, (PAGE_CACHE_SIZE - offset));
3461
3462                 kaddr = kmap_atomic(page, KM_USER0);
3463                 ret = memcmp(ptr, kaddr + offset, cur);
3464                 kunmap_atomic(kaddr, KM_USER0);
3465                 if (ret)
3466                         break;
3467
3468                 ptr += cur;
3469                 len -= cur;
3470                 offset = 0;
3471                 i++;
3472         }
3473         return ret;
3474 }
3475
3476 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
3477                          unsigned long start, unsigned long len)
3478 {
3479         size_t cur;
3480         size_t offset;
3481         struct page *page;
3482         char *kaddr;
3483         char *src = (char *)srcv;
3484         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3485         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3486
3487         WARN_ON(start > eb->len);
3488         WARN_ON(start + len > eb->start + eb->len);
3489
3490         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3491
3492         while (len > 0) {
3493                 page = extent_buffer_page(eb, i);
3494                 WARN_ON(!PageUptodate(page));
3495
3496                 cur = min(len, PAGE_CACHE_SIZE - offset);
3497                 kaddr = kmap_atomic(page, KM_USER1);
3498                 memcpy(kaddr + offset, src, cur);
3499                 kunmap_atomic(kaddr, KM_USER1);
3500
3501                 src += cur;
3502                 len -= cur;
3503                 offset = 0;
3504                 i++;
3505         }
3506 }
3507
3508 void memset_extent_buffer(struct extent_buffer *eb, char c,
3509                           unsigned long start, unsigned long len)
3510 {
3511         size_t cur;
3512         size_t offset;
3513         struct page *page;
3514         char *kaddr;
3515         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3516         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3517
3518         WARN_ON(start > eb->len);
3519         WARN_ON(start + len > eb->start + eb->len);
3520
3521         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3522
3523         while (len > 0) {
3524                 page = extent_buffer_page(eb, i);
3525                 WARN_ON(!PageUptodate(page));
3526
3527                 cur = min(len, PAGE_CACHE_SIZE - offset);
3528                 kaddr = kmap_atomic(page, KM_USER0);
3529                 memset(kaddr + offset, c, cur);
3530                 kunmap_atomic(kaddr, KM_USER0);
3531
3532                 len -= cur;
3533                 offset = 0;
3534                 i++;
3535         }
3536 }
3537
3538 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
3539                         unsigned long dst_offset, unsigned long src_offset,
3540                         unsigned long len)
3541 {
3542         u64 dst_len = dst->len;
3543         size_t cur;
3544         size_t offset;
3545         struct page *page;
3546         char *kaddr;
3547         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3548         unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3549
3550         WARN_ON(src->len != dst_len);
3551
3552         offset = (start_offset + dst_offset) &
3553                 ((unsigned long)PAGE_CACHE_SIZE - 1);
3554
3555         while (len > 0) {
3556                 page = extent_buffer_page(dst, i);
3557                 WARN_ON(!PageUptodate(page));
3558
3559                 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
3560
3561                 kaddr = kmap_atomic(page, KM_USER0);
3562                 read_extent_buffer(src, kaddr + offset, src_offset, cur);
3563                 kunmap_atomic(kaddr, KM_USER0);
3564
3565                 src_offset += cur;
3566                 len -= cur;
3567                 offset = 0;
3568                 i++;
3569         }
3570 }
3571
3572 static void move_pages(struct page *dst_page, struct page *src_page,
3573                        unsigned long dst_off, unsigned long src_off,
3574                        unsigned long len)
3575 {
3576         char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3577         if (dst_page == src_page) {
3578                 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
3579         } else {
3580                 char *src_kaddr = kmap_atomic(src_page, KM_USER1);
3581                 char *p = dst_kaddr + dst_off + len;
3582                 char *s = src_kaddr + src_off + len;
3583
3584                 while (len--)
3585                         *--p = *--s;
3586
3587                 kunmap_atomic(src_kaddr, KM_USER1);
3588         }
3589         kunmap_atomic(dst_kaddr, KM_USER0);
3590 }
3591
3592 static void copy_pages(struct page *dst_page, struct page *src_page,
3593                        unsigned long dst_off, unsigned long src_off,
3594                        unsigned long len)
3595 {
3596         char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3597         char *src_kaddr;
3598
3599         if (dst_page != src_page)
3600                 src_kaddr = kmap_atomic(src_page, KM_USER1);
3601         else
3602                 src_kaddr = dst_kaddr;
3603
3604         memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
3605         kunmap_atomic(dst_kaddr, KM_USER0);
3606         if (dst_page != src_page)
3607                 kunmap_atomic(src_kaddr, KM_USER1);
3608 }
3609
3610 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3611                            unsigned long src_offset, unsigned long len)
3612 {
3613         size_t cur;
3614         size_t dst_off_in_page;
3615         size_t src_off_in_page;
3616         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3617         unsigned long dst_i;
3618         unsigned long src_i;
3619
3620         if (src_offset + len > dst->len) {
3621                 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
3622                        "len %lu dst len %lu\n", src_offset, len, dst->len);
3623                 BUG_ON(1);
3624         }
3625         if (dst_offset + len > dst->len) {
3626                 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
3627                        "len %lu dst len %lu\n", dst_offset, len, dst->len);
3628                 BUG_ON(1);
3629         }
3630
3631         while (len > 0) {
3632                 dst_off_in_page = (start_offset + dst_offset) &
3633                         ((unsigned long)PAGE_CACHE_SIZE - 1);
3634                 src_off_in_page = (start_offset + src_offset) &
3635                         ((unsigned long)PAGE_CACHE_SIZE - 1);
3636
3637                 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3638                 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
3639
3640                 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
3641                                                src_off_in_page));
3642                 cur = min_t(unsigned long, cur,
3643                         (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
3644
3645                 copy_pages(extent_buffer_page(dst, dst_i),
3646                            extent_buffer_page(dst, src_i),
3647                            dst_off_in_page, src_off_in_page, cur);
3648
3649                 src_offset += cur;
3650                 dst_offset += cur;
3651                 len -= cur;
3652         }
3653 }
3654
3655 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3656                            unsigned long src_offset, unsigned long len)
3657 {
3658         size_t cur;
3659         size_t dst_off_in_page;
3660         size_t src_off_in_page;
3661         unsigned long dst_end = dst_offset + len - 1;
3662         unsigned long src_end = src_offset + len - 1;
3663         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3664         unsigned long dst_i;
3665         unsigned long src_i;
3666
3667         if (src_offset + len > dst->len) {
3668                 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
3669                        "len %lu len %lu\n", src_offset, len, dst->len);
3670                 BUG_ON(1);
3671         }
3672         if (dst_offset + len > dst->len) {
3673                 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
3674                        "len %lu len %lu\n", dst_offset, len, dst->len);
3675                 BUG_ON(1);
3676         }
3677         if (dst_offset < src_offset) {
3678                 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
3679                 return;
3680         }
3681         while (len > 0) {
3682                 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
3683                 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
3684
3685                 dst_off_in_page = (start_offset + dst_end) &
3686                         ((unsigned long)PAGE_CACHE_SIZE - 1);
3687                 src_off_in_page = (start_offset + src_end) &
3688                         ((unsigned long)PAGE_CACHE_SIZE - 1);
3689
3690                 cur = min_t(unsigned long, len, src_off_in_page + 1);
3691                 cur = min(cur, dst_off_in_page + 1);
3692                 move_pages(extent_buffer_page(dst, dst_i),
3693                            extent_buffer_page(dst, src_i),
3694                            dst_off_in_page - cur + 1,
3695                            src_off_in_page - cur + 1, cur);
3696
3697                 dst_end -= cur;
3698                 src_end -= cur;
3699                 len -= cur;
3700         }
3701 }
3702
3703 int try_release_extent_buffer(struct extent_io_tree *tree, struct page *page)
3704 {
3705         u64 start = page_offset(page);
3706         struct extent_buffer *eb;
3707         int ret = 1;
3708         unsigned long i;
3709         unsigned long num_pages;
3710
3711         spin_lock(&tree->buffer_lock);
3712         eb = buffer_search(tree, start);
3713         if (!eb)
3714                 goto out;
3715
3716         if (atomic_read(&eb->refs) > 1) {
3717                 ret = 0;
3718                 goto out;
3719         }
3720         if (test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3721                 ret = 0;
3722                 goto out;
3723         }
3724         /* at this point we can safely release the extent buffer */
3725         num_pages = num_extent_pages(eb->start, eb->len);
3726         for (i = 0; i < num_pages; i++)
3727                 page_cache_release(extent_buffer_page(eb, i));
3728         rb_erase(&eb->rb_node, &tree->buffer);
3729         __free_extent_buffer(eb);
3730 out:
3731         spin_unlock(&tree->buffer_lock);
3732         return ret;
3733 }