git.oblomov.eu Git - linux-2.6/blob - 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 /* temporary define until extent_map moves out of btrfs */
  21 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
  22                                        unsigned long extra_flags,
  23                                        void (*ctor)(void *, struct kmem_cache *,
  24                                                     unsigned long));
  25
  26 static struct kmem_cache *extent_state_cache;
  27 static struct kmem_cache *extent_buffer_cache;
  28
  29 static LIST_HEAD(buffers);
  30 static LIST_HEAD(states);
  31
  32 #define LEAK_DEBUG 0
  33 #ifdef LEAK_DEBUG
  34 static DEFINE_SPINLOCK(leak_lock);
  35 #endif
  36
  37 #define BUFFER_LRU_MAX 64
  38
  39 struct tree_entry {
  40         u64 start;
  41         u64 end;
  42         struct rb_node rb_node;
  43 };
  44
  45 struct extent_page_data {
  46         struct bio *bio;
  47         struct extent_io_tree *tree;
  48         get_extent_t *get_extent;
  49
  50         /* tells writepage not to lock the state bits for this range
  51          * it still does the unlocking
  52          */
  53         int extent_locked;
  54 };
  55
  56 int __init extent_io_init(void)
  57 {
  58         extent_state_cache = btrfs_cache_create("extent_state",
  59                                             sizeof(struct extent_state), 0,
  60                                             NULL);
  61         if (!extent_state_cache)
  62                 return -ENOMEM;
  63
  64         extent_buffer_cache = btrfs_cache_create("extent_buffers",
  65                                             sizeof(struct extent_buffer), 0,
  66                                             NULL);
  67         if (!extent_buffer_cache)
  68                 goto free_state_cache;
  69         return 0;
  70
  71 free_state_cache:
  72         kmem_cache_destroy(extent_state_cache);
  73         return -ENOMEM;
  74 }
  75
  76 void extent_io_exit(void)
  77 {
  78         struct extent_state *state;
  79         struct extent_buffer *eb;
  80
  81         while (!list_empty(&states)) {
  82                 state = list_entry(states.next, struct extent_state, leak_list);
  83                 printk(KERN_ERR "btrfs state leak: start %llu end %llu "
  84                        "state %lu in tree %p refs %d\n",
  85                        (unsigned long long)state->start,
  86                        (unsigned long long)state->end,
  87                        state->state, state->tree, atomic_read(&state->refs));
  88                 list_del(&state->leak_list);
  89                 kmem_cache_free(extent_state_cache, state);
  90
  91         }
  92
  93         while (!list_empty(&buffers)) {
  94                 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
  95                 printk(KERN_ERR "btrfs buffer leak start %llu len %lu "
  96                        "refs %d\n", (unsigned long long)eb->start,
  97                        eb->len, atomic_read(&eb->refs));
  98                 list_del(&eb->leak_list);
  99                 kmem_cache_free(extent_buffer_cache, eb);
 100         }
 101         if (extent_state_cache)
 102                 kmem_cache_destroy(extent_state_cache);
 103         if (extent_buffer_cache)
 104                 kmem_cache_destroy(extent_buffer_cache);
 105 }
 106
 107 void extent_io_tree_init(struct extent_io_tree *tree,
 108                           struct address_space *mapping, gfp_t mask)
 109 {
 110         tree->state.rb_node = NULL;
 111         tree->buffer.rb_node = NULL;
 112         tree->ops = NULL;
 113         tree->dirty_bytes = 0;
 114         spin_lock_init(&tree->lock);
 115         spin_lock_init(&tree->buffer_lock);
 116         tree->mapping = mapping;
 117 }
 118
 119 static struct extent_state *alloc_extent_state(gfp_t mask)
 120 {
 121         struct extent_state *state;
 122 #ifdef LEAK_DEBUG
 123         unsigned long flags;
 124 #endif
 125
 126         state = kmem_cache_alloc(extent_state_cache, mask);
 127         if (!state)
 128                 return state;
 129         state->state = 0;
 130         state->private = 0;
 131         state->tree = NULL;
 132 #ifdef LEAK_DEBUG
 133         spin_lock_irqsave(&leak_lock, flags);
 134         list_add(&state->leak_list, &states);
 135         spin_unlock_irqrestore(&leak_lock, flags);
 136 #endif
 137         atomic_set(&state->refs, 1);
 138         init_waitqueue_head(&state->wq);
 139         return state;
 140 }
 141
 142 static void free_extent_state(struct extent_state *state)
 143 {
 144         if (!state)
 145                 return;
 146         if (atomic_dec_and_test(&state->refs)) {
 147 #ifdef LEAK_DEBUG
 148                 unsigned long flags;
 149 #endif
 150                 WARN_ON(state->tree);
 151 #ifdef LEAK_DEBUG
 152                 spin_lock_irqsave(&leak_lock, flags);
 153                 list_del(&state->leak_list);
 154                 spin_unlock_irqrestore(&leak_lock, flags);
 155 #endif
 156                 kmem_cache_free(extent_state_cache, state);
 157         }
 158 }
 159
 160 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
 161                                    struct rb_node *node)
 162 {
 163         struct rb_node **p = &root->rb_node;
 164         struct rb_node *parent = NULL;
 165         struct tree_entry *entry;
 166
 167         while (*p) {
 168                 parent = *p;
 169                 entry = rb_entry(parent, struct tree_entry, rb_node);
 170
 171                 if (offset < entry->start)
 172                         p = &(*p)->rb_left;
 173                 else if (offset > entry->end)
 174                         p = &(*p)->rb_right;
 175                 else
 176                         return parent;
 177         }
 178
 179         entry = rb_entry(node, struct tree_entry, rb_node);
 180         rb_link_node(node, parent, p);
 181         rb_insert_color(node, root);
 182         return NULL;
 183 }
 184
 185 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
 186                                      struct rb_node **prev_ret,
 187                                      struct rb_node **next_ret)
 188 {
 189         struct rb_root *root = &tree->state;
 190         struct rb_node *n = root->rb_node;
 191         struct rb_node *prev = NULL;
 192         struct rb_node *orig_prev = NULL;
 193         struct tree_entry *entry;
 194         struct tree_entry *prev_entry = NULL;
 195
 196         while (n) {
 197                 entry = rb_entry(n, struct tree_entry, rb_node);
 198                 prev = n;
 199                 prev_entry = entry;
 200
 201                 if (offset < entry->start)
 202                         n = n->rb_left;
 203                 else if (offset > entry->end)
 204                         n = n->rb_right;
 205                 else
 206                         return n;
 207         }
 208
 209         if (prev_ret) {
 210                 orig_prev = prev;
 211                 while (prev && offset > prev_entry->end) {
 212                         prev = rb_next(prev);
 213                         prev_entry = rb_entry(prev, struct tree_entry, rb_node);
 214                 }
 215                 *prev_ret = prev;
 216                 prev = orig_prev;
 217         }
 218
 219         if (next_ret) {
 220                 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
 221                 while (prev && offset < prev_entry->start) {
 222                         prev = rb_prev(prev);
 223                         prev_entry = rb_entry(prev, struct tree_entry, rb_node);
 224                 }
 225                 *next_ret = prev;
 226         }
 227         return NULL;
 228 }
 229
 230 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
 231                                           u64 offset)
 232 {
 233         struct rb_node *prev = NULL;
 234         struct rb_node *ret;
 235
 236         ret = __etree_search(tree, offset, &prev, NULL);
 237         if (!ret)
 238                 return prev;
 239         return ret;
 240 }
 241
 242 static struct extent_buffer *buffer_tree_insert(struct extent_io_tree *tree,
 243                                           u64 offset, struct rb_node *node)
 244 {
 245         struct rb_root *root = &tree->buffer;
 246         struct rb_node **p = &root->rb_node;
 247         struct rb_node *parent = NULL;
 248         struct extent_buffer *eb;
 249
 250         while (*p) {
 251                 parent = *p;
 252                 eb = rb_entry(parent, struct extent_buffer, rb_node);
 253
 254                 if (offset < eb->start)
 255                         p = &(*p)->rb_left;
 256                 else if (offset > eb->start)
 257                         p = &(*p)->rb_right;
 258                 else
 259                         return eb;
 260         }
 261
 262         rb_link_node(node, parent, p);
 263         rb_insert_color(node, root);
 264         return NULL;
 265 }
 266
 267 static struct extent_buffer *buffer_search(struct extent_io_tree *tree,
 268                                            u64 offset)
 269 {
 270         struct rb_root *root = &tree->buffer;
 271         struct rb_node *n = root->rb_node;
 272         struct extent_buffer *eb;
 273
 274         while (n) {
 275                 eb = rb_entry(n, struct extent_buffer, rb_node);
 276                 if (offset < eb->start)
 277                         n = n->rb_left;
 278                 else if (offset > eb->start)
 279                         n = n->rb_right;
 280                 else
 281                         return eb;
 282         }
 283         return NULL;
 284 }
 285
 286 /*
 287  * utility function to look for merge candidates inside a given range.
 288  * Any extents with matching state are merged together into a single
 289  * extent in the tree.  Extents with EXTENT_IO in their state field
 290  * are not merged because the end_io handlers need to be able to do
 291  * operations on them without sleeping (or doing allocations/splits).
 292  *
 293  * This should be called with the tree lock held.
 294  */
 295 static int merge_state(struct extent_io_tree *tree,
 296                        struct extent_state *state)
 297 {
 298         struct extent_state *other;
 299         struct rb_node *other_node;
 300
 301         if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
 302                 return 0;
 303
 304         other_node = rb_prev(&state->rb_node);
 305         if (other_node) {
 306                 other = rb_entry(other_node, struct extent_state, rb_node);
 307                 if (other->end == state->start - 1 &&
 308                     other->state == state->state) {
 309                         state->start = other->start;
 310                         other->tree = NULL;
 311                         rb_erase(&other->rb_node, &tree->state);
 312                         free_extent_state(other);
 313                 }
 314         }
 315         other_node = rb_next(&state->rb_node);
 316         if (other_node) {
 317                 other = rb_entry(other_node, struct extent_state, rb_node);
 318                 if (other->start == state->end + 1 &&
 319                     other->state == state->state) {
 320                         other->start = state->start;
 321                         state->tree = NULL;
 322                         rb_erase(&state->rb_node, &tree->state);
 323                         free_extent_state(state);
 324                 }
 325         }
 326         return 0;
 327 }
 328
 329 static void set_state_cb(struct extent_io_tree *tree,
 330                          struct extent_state *state,
 331                          unsigned long bits)
 332 {
 333         if (tree->ops && tree->ops->set_bit_hook) {
 334                 tree->ops->set_bit_hook(tree->mapping->host, state->start,
 335                                         state->end, state->state, bits);
 336         }
 337 }
 338
 339 static void clear_state_cb(struct extent_io_tree *tree,
 340                            struct extent_state *state,
 341                            unsigned long bits)
 342 {
 343         if (tree->ops && tree->ops->clear_bit_hook) {
 344                 tree->ops->clear_bit_hook(tree->mapping->host, state->start,
 345                                           state->end, state->state, bits);
 346         }
 347 }
 348
 349 /*
 350  * insert an extent_state struct into the tree.  'bits' are set on the
 351  * struct before it is inserted.
 352  *
 353  * This may return -EEXIST if the extent is already there, in which case the
 354  * state struct is freed.
 355  *
 356  * The tree lock is not taken internally.  This is a utility function and
 357  * probably isn't what you want to call (see set/clear_extent_bit).
 358  */
 359 static int insert_state(struct extent_io_tree *tree,
 360                         struct extent_state *state, u64 start, u64 end,
 361                         int bits)
 362 {
 363         struct rb_node *node;
 364
 365         if (end < start) {
 366                 printk(KERN_ERR "btrfs end < start %llu %llu\n",
 367                        (unsigned long long)end,
 368                        (unsigned long long)start);
 369                 WARN_ON(1);
 370         }
 371         if (bits & EXTENT_DIRTY)
 372                 tree->dirty_bytes += end - start + 1;
 373         set_state_cb(tree, state, bits);
 374         state->state |= bits;
 375         state->start = start;
 376         state->end = end;
 377         node = tree_insert(&tree->state, end, &state->rb_node);
 378         if (node) {
 379                 struct extent_state *found;
 380                 found = rb_entry(node, struct extent_state, rb_node);
 381                 printk(KERN_ERR "btrfs found node %llu %llu on insert of "
 382                        "%llu %llu\n", (unsigned long long)found->start,
 383                        (unsigned long long)found->end,
 384                        (unsigned long long)start, (unsigned long long)end);
 385                 free_extent_state(state);
 386                 return -EEXIST;
 387         }
 388         state->tree = tree;
 389         merge_state(tree, state);
 390         return 0;
 391 }
 392
 393 /*
 394  * split a given extent state struct in two, inserting the preallocated
 395  * struct 'prealloc' as the newly created second half.  'split' indicates an
 396  * offset inside 'orig' where it should be split.
 397  *
 398  * Before calling,
 399  * the tree has 'orig' at [orig->start, orig->end].  After calling, there
 400  * are two extent state structs in the tree:
 401  * prealloc: [orig->start, split - 1]
 402  * orig: [ split, orig->end ]
 403  *
 404  * The tree locks are not taken by this function. They need to be held
 405  * by the caller.
 406  */
 407 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
 408                        struct extent_state *prealloc, u64 split)
 409 {
 410         struct rb_node *node;
 411         prealloc->start = orig->start;
 412         prealloc->end = split - 1;
 413         prealloc->state = orig->state;
 414         orig->start = split;
 415
 416         node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
 417         if (node) {
 418                 struct extent_state *found;
 419                 found = rb_entry(node, struct extent_state, rb_node);
 420                 free_extent_state(prealloc);
 421                 return -EEXIST;
 422         }
 423         prealloc->tree = tree;
 424         return 0;
 425 }
 426
 427 /*
 428  * utility function to clear some bits in an extent state struct.
 429  * it will optionally wake up any one waiting on this state (wake == 1), or
 430  * forcibly remove the state from the tree (delete == 1).
 431  *
 432  * If no bits are set on the state struct after clearing things, the
 433  * struct is freed and removed from the tree
 434  */
 435 static int clear_state_bit(struct extent_io_tree *tree,
 436                             struct extent_state *state, int bits, int wake,
 437                             int delete)
 438 {
 439         int ret = state->state & bits;
 440
 441         if ((bits & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
 442                 u64 range = state->end - state->start + 1;
 443                 WARN_ON(range > tree->dirty_bytes);
 444                 tree->dirty_bytes -= range;
 445         }
 446         clear_state_cb(tree, state, bits);
 447         state->state &= ~bits;
 448         if (wake)
 449                 wake_up(&state->wq);
 450         if (delete || state->state == 0) {
 451                 if (state->tree) {
 452                         clear_state_cb(tree, state, state->state);
 453                         rb_erase(&state->rb_node, &tree->state);
 454                         state->tree = NULL;
 455                         free_extent_state(state);
 456                 } else {
 457                         WARN_ON(1);
 458                 }
 459         } else {
 460                 merge_state(tree, state);
 461         }
 462         return ret;
 463 }
 464
 465 /*
 466  * clear some bits on a range in the tree.  This may require splitting
 467  * or inserting elements in the tree, so the gfp mask is used to
 468  * indicate which allocations or sleeping are allowed.
 469  *
 470  * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
 471  * the given range from the tree regardless of state (ie for truncate).
 472  *
 473  * the range [start, end] is inclusive.
 474  *
 475  * This takes the tree lock, and returns < 0 on error, > 0 if any of the
 476  * bits were already set, or zero if none of the bits were already set.
 477  */
 478 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
 479                      int bits, int wake, int delete, gfp_t mask)
 480 {
 481         struct extent_state *state;
 482         struct extent_state *prealloc = NULL;
 483         struct rb_node *node;
 484         int err;
 485         int set = 0;
 486
 487 again:
 488         if (!prealloc && (mask & __GFP_WAIT)) {
 489                 prealloc = alloc_extent_state(mask);
 490                 if (!prealloc)
 491                         return -ENOMEM;
 492         }
 493
 494         spin_lock(&tree->lock);
 495         /*
 496          * this search will find the extents that end after
 497          * our range starts
 498          */
 499         node = tree_search(tree, start);
 500         if (!node)
 501                 goto out;
 502         state = rb_entry(node, struct extent_state, rb_node);
 503         if (state->start > end)
 504                 goto out;
 505         WARN_ON(state->end < start);
 506
 507         /*
 508          *     | ---- desired range ---- |
 509          *  | state | or
 510          *  | ------------- state -------------- |
 511          *
 512          * We need to split the extent we found, and may flip
 513          * bits on second half.
 514          *
 515          * If the extent we found extends past our range, we
 516          * just split and search again.  It'll get split again
 517          * the next time though.
 518          *
 519          * If the extent we found is inside our range, we clear
 520          * the desired bit on it.
 521          */
 522
 523         if (state->start < start) {
 524                 if (!prealloc)
 525                         prealloc = alloc_extent_state(GFP_ATOMIC);
 526                 err = split_state(tree, state, prealloc, start);
 527                 BUG_ON(err == -EEXIST);
 528                 prealloc = NULL;
 529                 if (err)
 530                         goto out;
 531                 if (state->end <= end) {
 532                         start = state->end + 1;
 533                         set |= clear_state_bit(tree, state, bits,
 534                                         wake, delete);
 535                 } else {
 536                         start = state->start;
 537                 }
 538                 goto search_again;
 539         }
 540         /*
 541          * | ---- desired range ---- |
 542          *                        | state |
 543          * We need to split the extent, and clear the bit
 544          * on the first half
 545          */
 546         if (state->start <= end && state->end > end) {
 547                 if (!prealloc)
 548                         prealloc = alloc_extent_state(GFP_ATOMIC);
 549                 err = split_state(tree, state, prealloc, end + 1);
 550                 BUG_ON(err == -EEXIST);
 551
 552                 if (wake)
 553                         wake_up(&state->wq);
 554                 set |= clear_state_bit(tree, prealloc, bits,
 555                                        wake, delete);
 556                 prealloc = NULL;
 557                 goto out;
 558         }
 559
 560         start = state->end + 1;
 561         set |= clear_state_bit(tree, state, bits, wake, delete);
 562         goto search_again;
 563
 564 out:
 565         spin_unlock(&tree->lock);
 566         if (prealloc)
 567                 free_extent_state(prealloc);
 568
 569         return set;
 570
 571 search_again:
 572         if (start > end)
 573                 goto out;
 574         spin_unlock(&tree->lock);
 575         if (mask & __GFP_WAIT)
 576                 cond_resched();
 577         goto again;
 578 }
 579
 580 static int wait_on_state(struct extent_io_tree *tree,
 581                          struct extent_state *state)
 582                 __releases(tree->lock)
 583                 __acquires(tree->lock)
 584 {
 585         DEFINE_WAIT(wait);
 586         prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
 587         spin_unlock(&tree->lock);
 588         schedule();
 589         spin_lock(&tree->lock);
 590         finish_wait(&state->wq, &wait);
 591         return 0;
 592 }
 593
 594 /*
 595  * waits for one or more bits to clear on a range in the state tree.
 596  * The range [start, end] is inclusive.
 597  * The tree lock is taken by this function
 598  */
 599 int wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits)
 600 {
 601         struct extent_state *state;
 602         struct rb_node *node;
 603
 604         spin_lock(&tree->lock);
 605 again:
 606         while (1) {
 607                 /*
 608                  * this search will find all the extents that end after
 609                  * our range starts
 610                  */
 611                 node = tree_search(tree, start);
 612                 if (!node)
 613                         break;
 614
 615                 state = rb_entry(node, struct extent_state, rb_node);
 616
 617                 if (state->start > end)
 618                         goto out;
 619
 620                 if (state->state & bits) {
 621                         start = state->start;
 622                         atomic_inc(&state->refs);
 623                         wait_on_state(tree, state);
 624                         free_extent_state(state);
 625                         goto again;
 626                 }
 627                 start = state->end + 1;
 628
 629                 if (start > end)
 630                         break;
 631
 632                 if (need_resched()) {
 633                         spin_unlock(&tree->lock);
 634                         cond_resched();
 635                         spin_lock(&tree->lock);
 636                 }
 637         }
 638 out:
 639         spin_unlock(&tree->lock);
 640         return 0;
 641 }
 642
 643 static void set_state_bits(struct extent_io_tree *tree,
 644                            struct extent_state *state,
 645                            int bits)
 646 {
 647         if ((bits & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
 648                 u64 range = state->end - state->start + 1;
 649                 tree->dirty_bytes += range;
 650         }
 651         set_state_cb(tree, state, bits);
 652         state->state |= bits;
 653 }
 654
 655 /*
 656  * set some bits on a range in the tree.  This may require allocations
 657  * or sleeping, so the gfp mask is used to indicate what is allowed.
 658  *
 659  * If 'exclusive' == 1, this will fail with -EEXIST if some part of the
 660  * range already has the desired bits set.  The start of the existing
 661  * range is returned in failed_start in this case.
 662  *
 663  * [start, end] is inclusive
 664  * This takes the tree lock.
 665  */
 666 static int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
 667                           int bits, int exclusive, u64 *failed_start,
 668                           gfp_t mask)
 669 {
 670         struct extent_state *state;
 671         struct extent_state *prealloc = NULL;
 672         struct rb_node *node;
 673         int err = 0;
 674         int set;
 675         u64 last_start;
 676         u64 last_end;
 677 again:
 678         if (!prealloc && (mask & __GFP_WAIT)) {
 679                 prealloc = alloc_extent_state(mask);
 680                 if (!prealloc)
 681                         return -ENOMEM;
 682         }
 683
 684         spin_lock(&tree->lock);
 685         /*
 686          * this search will find all the extents that end after
 687          * our range starts.
 688          */
 689         node = tree_search(tree, start);
 690         if (!node) {
 691                 err = insert_state(tree, prealloc, start, end, bits);
 692                 prealloc = NULL;
 693                 BUG_ON(err == -EEXIST);
 694                 goto out;
 695         }
 696
 697         state = rb_entry(node, struct extent_state, rb_node);
 698         last_start = state->start;
 699         last_end = state->end;
 700
 701         /*
 702          * | ---- desired range ---- |
 703          * | state |
 704          *
 705          * Just lock what we found and keep going
 706          */
 707         if (state->start == start && state->end <= end) {
 708                 set = state->state & bits;
 709                 if (set && exclusive) {
 710                         *failed_start = state->start;
 711                         err = -EEXIST;
 712                         goto out;
 713                 }
 714                 set_state_bits(tree, state, bits);
 715                 start = state->end + 1;
 716                 merge_state(tree, state);
 717                 goto search_again;
 718         }
 719
 720         /*
 721          *     | ---- desired range ---- |
 722          * | state |
 723          *   or
 724          * | ------------- state -------------- |
 725          *
 726          * We need to split the extent we found, and may flip bits on
 727          * second half.
 728          *
 729          * If the extent we found extends past our
 730          * range, we just split and search again.  It'll get split
 731          * again the next time though.
 732          *
 733          * If the extent we found is inside our range, we set the
 734          * desired bit on it.
 735          */
 736         if (state->start < start) {
 737                 set = state->state & bits;
 738                 if (exclusive && set) {
 739                         *failed_start = start;
 740                         err = -EEXIST;
 741                         goto out;
 742                 }
 743                 err = split_state(tree, state, prealloc, start);
 744                 BUG_ON(err == -EEXIST);
 745                 prealloc = NULL;
 746                 if (err)
 747                         goto out;
 748                 if (state->end <= end) {
 749                         set_state_bits(tree, state, bits);
 750                         start = state->end + 1;
 751                         merge_state(tree, state);
 752                 } else {
 753                         start = state->start;
 754                 }
 755                 goto search_again;
 756         }
 757         /*
 758          * | ---- desired range ---- |
 759          *     | state | or               | state |
 760          *
 761          * There's a hole, we need to insert something in it and
 762          * ignore the extent we found.
 763          */
 764         if (state->start > start) {
 765                 u64 this_end;
 766                 if (end < last_start)
 767                         this_end = end;
 768                 else
 769                         this_end = last_start - 1;
 770                 err = insert_state(tree, prealloc, start, this_end,
 771                                    bits);
 772                 prealloc = NULL;
 773                 BUG_ON(err == -EEXIST);
 774                 if (err)
 775                         goto out;
 776                 start = this_end + 1;
 777                 goto search_again;
 778         }
 779         /*
 780          * | ---- desired range ---- |
 781          *                        | state |
 782          * We need to split the extent, and set the bit
 783          * on the first half
 784          */
 785         if (state->start <= end && state->end > end) {
 786                 set = state->state & bits;
 787                 if (exclusive && set) {
 788                         *failed_start = start;
 789                         err = -EEXIST;
 790                         goto out;
 791                 }
 792                 err = split_state(tree, state, prealloc, end + 1);
 793                 BUG_ON(err == -EEXIST);
 794
 795                 set_state_bits(tree, prealloc, bits);
 796                 merge_state(tree, prealloc);
 797                 prealloc = NULL;
 798                 goto out;
 799         }
 800
 801         goto search_again;
 802
 803 out:
 804         spin_unlock(&tree->lock);
 805         if (prealloc)
 806                 free_extent_state(prealloc);
 807
 808         return err;
 809
 810 search_again:
 811         if (start > end)
 812                 goto out;
 813         spin_unlock(&tree->lock);
 814         if (mask & __GFP_WAIT)
 815                 cond_resched();
 816         goto again;
 817 }
 818
 819 /* wrappers around set/clear extent bit */
 820 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
 821                      gfp_t mask)
 822 {
 823         return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL,
 824                               mask);
 825 }
 826
 827 int set_extent_ordered(struct extent_io_tree *tree, u64 start, u64 end,
 828                        gfp_t mask)
 829 {
 830         return set_extent_bit(tree, start, end, EXTENT_ORDERED, 0, NULL, mask);
 831 }
 832
 833 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
 834                     int bits, gfp_t mask)
 835 {
 836         return set_extent_bit(tree, start, end, bits, 0, NULL,
 837                               mask);
 838 }
 839
 840 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
 841                       int bits, gfp_t mask)
 842 {
 843         return clear_extent_bit(tree, start, end, bits, 0, 0, mask);
 844 }
 845
 846 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
 847                      gfp_t mask)
 848 {
 849         return set_extent_bit(tree, start, end,
 850                               EXTENT_DELALLOC | EXTENT_DIRTY,
 851                               0, NULL, mask);
 852 }
 853
 854 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
 855                        gfp_t mask)
 856 {
 857         return clear_extent_bit(tree, start, end,
 858                                 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, mask);
 859 }
 860
 861 int clear_extent_ordered(struct extent_io_tree *tree, u64 start, u64 end,
 862                          gfp_t mask)
 863 {
 864         return clear_extent_bit(tree, start, end, EXTENT_ORDERED, 1, 0, mask);
 865 }
 866
 867 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
 868                      gfp_t mask)
 869 {
 870         return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL,
 871                               mask);
 872 }
 873
 874 static int clear_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
 875                        gfp_t mask)
 876 {
 877         return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0, mask);
 878 }
 879
 880 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
 881                         gfp_t mask)
 882 {
 883         return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, NULL,
 884                               mask);
 885 }
 886
 887 static int clear_extent_uptodate(struct extent_io_tree *tree, u64 start,
 888                                  u64 end, gfp_t mask)
 889 {
 890         return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, mask);
 891 }
 892
 893 static int set_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end,
 894                          gfp_t mask)
 895 {
 896         return set_extent_bit(tree, start, end, EXTENT_WRITEBACK,
 897                               0, NULL, mask);
 898 }
 899
 900 static int clear_extent_writeback(struct extent_io_tree *tree, u64 start,
 901                                   u64 end, gfp_t mask)
 902 {
 903         return clear_extent_bit(tree, start, end, EXTENT_WRITEBACK, 1, 0, mask);
 904 }
 905
 906 int wait_on_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end)
 907 {
 908         return wait_extent_bit(tree, start, end, EXTENT_WRITEBACK);
 909 }
 910
 911 /*
 912  * either insert or lock state struct between start and end use mask to tell
 913  * us if waiting is desired.
 914  */
 915 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
 916 {
 917         int err;
 918         u64 failed_start;
 919         while (1) {
 920                 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1,
 921                                      &failed_start, mask);
 922                 if (err == -EEXIST && (mask & __GFP_WAIT)) {
 923                         wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
 924                         start = failed_start;
 925                 } else {
 926                         break;
 927                 }
 928                 WARN_ON(start > end);
 929         }
 930         return err;
 931 }
 932
 933 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end,
 934                     gfp_t mask)
 935 {
 936         int err;
 937         u64 failed_start;
 938
 939         err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1,
 940                              &failed_start, mask);
 941         if (err == -EEXIST) {
 942                 if (failed_start > start)
 943                         clear_extent_bit(tree, start, failed_start - 1,
 944                                          EXTENT_LOCKED, 1, 0, mask);
 945                 return 0;
 946         }
 947         return 1;
 948 }
 949
 950 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end,
 951                   gfp_t mask)
 952 {
 953         return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, mask);
 954 }
 955
 956 /*
 957  * helper function to set pages and extents in the tree dirty
 958  */
 959 int set_range_dirty(struct extent_io_tree *tree, u64 start, u64 end)
 960 {
 961         unsigned long index = start >> PAGE_CACHE_SHIFT;
 962         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
 963         struct page *page;
 964
 965         while (index <= end_index) {
 966                 page = find_get_page(tree->mapping, index);
 967                 BUG_ON(!page);
 968                 __set_page_dirty_nobuffers(page);
 969                 page_cache_release(page);
 970                 index++;
 971         }
 972         set_extent_dirty(tree, start, end, GFP_NOFS);
 973         return 0;
 974 }
 975
 976 /*
 977  * helper function to set both pages and extents in the tree writeback
 978  */
 979 static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
 980 {
 981         unsigned long index = start >> PAGE_CACHE_SHIFT;
 982         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
 983         struct page *page;
 984
 985         while (index <= end_index) {
 986                 page = find_get_page(tree->mapping, index);
 987                 BUG_ON(!page);
 988                 set_page_writeback(page);
 989                 page_cache_release(page);
 990                 index++;
 991         }
 992         set_extent_writeback(tree, start, end, GFP_NOFS);
 993         return 0;
 994 }
 995
 996 /*
 997  * find the first offset in the io tree with 'bits' set. zero is
 998  * returned if we find something, and *start_ret and *end_ret are
 999  * set to reflect the state struct that was found.
1000  *
1001  * If nothing was found, 1 is returned, < 0 on error
1002  */
1003 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1004                           u64 *start_ret, u64 *end_ret, int bits)
1005 {
1006         struct rb_node *node;
1007         struct extent_state *state;
1008         int ret = 1;
1009
1010         spin_lock(&tree->lock);
1011         /*
1012          * this search will find all the extents that end after
1013          * our range starts.
1014          */
1015         node = tree_search(tree, start);
1016         if (!node)
1017                 goto out;
1018
1019         while (1) {
1020                 state = rb_entry(node, struct extent_state, rb_node);
1021                 if (state->end >= start && (state->state & bits)) {
1022                         *start_ret = state->start;
1023                         *end_ret = state->end;
1024                         ret = 0;
1025                         break;
1026                 }
1027                 node = rb_next(node);
1028                 if (!node)
1029                         break;
1030         }
1031 out:
1032         spin_unlock(&tree->lock);
1033         return ret;
1034 }
1035
1036 /* find the first state struct with 'bits' set after 'start', and
1037  * return it.  tree->lock must be held.  NULL will returned if
1038  * nothing was found after 'start'
1039  */
1040 struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree,
1041                                                  u64 start, int bits)
1042 {
1043         struct rb_node *node;
1044         struct extent_state *state;
1045
1046         /*
1047          * this search will find all the extents that end after
1048          * our range starts.
1049          */
1050         node = tree_search(tree, start);
1051         if (!node)
1052                 goto out;
1053
1054         while (1) {
1055                 state = rb_entry(node, struct extent_state, rb_node);
1056                 if (state->end >= start && (state->state & bits))
1057                         return state;
1058
1059                 node = rb_next(node);
1060                 if (!node)
1061                         break;
1062         }
1063 out:
1064         return NULL;
1065 }
1066
1067 /*
1068  * find a contiguous range of bytes in the file marked as delalloc, not
1069  * more than 'max_bytes'.  start and end are used to return the range,
1070  *
1071  * 1 is returned if we find something, 0 if nothing was in the tree
1072  */
1073 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1074                                         u64 *start, u64 *end, u64 max_bytes)
1075 {
1076         struct rb_node *node;
1077         struct extent_state *state;
1078         u64 cur_start = *start;
1079         u64 found = 0;
1080         u64 total_bytes = 0;
1081
1082         spin_lock(&tree->lock);
1083
1084         /*
1085          * this search will find all the extents that end after
1086          * our range starts.
1087          */
1088         node = tree_search(tree, cur_start);
1089         if (!node) {
1090                 if (!found)
1091                         *end = (u64)-1;
1092                 goto out;
1093         }
1094
1095         while (1) {
1096                 state = rb_entry(node, struct extent_state, rb_node);
1097                 if (found && (state->start != cur_start ||
1098                               (state->state & EXTENT_BOUNDARY))) {
1099                         goto out;
1100                 }
1101                 if (!(state->state & EXTENT_DELALLOC)) {
1102                         if (!found)
1103                                 *end = state->end;
1104                         goto out;
1105                 }
1106                 if (!found)
1107                         *start = state->start;
1108                 found++;
1109                 *end = state->end;
1110                 cur_start = state->end + 1;
1111                 node = rb_next(node);
1112                 if (!node)
1113                         break;
1114                 total_bytes += state->end - state->start + 1;
1115                 if (total_bytes >= max_bytes)
1116                         break;
1117         }
1118 out:
1119         spin_unlock(&tree->lock);
1120         return found;
1121 }
1122
1123 static noinline int __unlock_for_delalloc(struct inode *inode,
1124                                           struct page *locked_page,
1125                                           u64 start, u64 end)
1126 {
1127         int ret;
1128         struct page *pages[16];
1129         unsigned long index = start >> PAGE_CACHE_SHIFT;
1130         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1131         unsigned long nr_pages = end_index - index + 1;
1132         int i;
1133
1134         if (index == locked_page->index && end_index == index)
1135                 return 0;
1136
1137         while (nr_pages > 0) {
1138                 ret = find_get_pages_contig(inode->i_mapping, index,
1139                                      min_t(unsigned long, nr_pages,
1140                                      ARRAY_SIZE(pages)), pages);
1141                 for (i = 0; i < ret; i++) {
1142                         if (pages[i] != locked_page)
1143                                 unlock_page(pages[i]);
1144                         page_cache_release(pages[i]);
1145                 }
1146                 nr_pages -= ret;
1147                 index += ret;
1148                 cond_resched();
1149         }
1150         return 0;
1151 }
1152
1153 static noinline int lock_delalloc_pages(struct inode *inode,
1154                                         struct page *locked_page,
1155                                         u64 delalloc_start,
1156                                         u64 delalloc_end)
1157 {
1158         unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT;
1159         unsigned long start_index = index;
1160         unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT;
1161         unsigned long pages_locked = 0;
1162         struct page *pages[16];
1163         unsigned long nrpages;
1164         int ret;
1165         int i;
1166
1167         /* the caller is responsible for locking the start index */
1168         if (index == locked_page->index && index == end_index)
1169                 return 0;
1170
1171         /* skip the page at the start index */
1172         nrpages = end_index - index + 1;
1173         while (nrpages > 0) {
1174                 ret = find_get_pages_contig(inode->i_mapping, index,
1175                                      min_t(unsigned long,
1176                                      nrpages, ARRAY_SIZE(pages)), pages);
1177                 if (ret == 0) {
1178                         ret = -EAGAIN;
1179                         goto done;
1180                 }
1181                 /* now we have an array of pages, lock them all */
1182                 for (i = 0; i < ret; i++) {
1183                         /*
1184                          * the caller is taking responsibility for
1185                          * locked_page
1186                          */
1187                         if (pages[i] != locked_page) {
1188                                 lock_page(pages[i]);
1189                                 if (!PageDirty(pages[i]) ||
1190                                     pages[i]->mapping != inode->i_mapping) {
1191                                         ret = -EAGAIN;
1192                                         unlock_page(pages[i]);
1193                                         page_cache_release(pages[i]);
1194                                         goto done;
1195                                 }
1196                         }
1197                         page_cache_release(pages[i]);
1198                         pages_locked++;
1199                 }
1200                 nrpages -= ret;
1201                 index += ret;
1202                 cond_resched();
1203         }
1204         ret = 0;
1205 done:
1206         if (ret && pages_locked) {
1207                 __unlock_for_delalloc(inode, locked_page,
1208                               delalloc_start,
1209                               ((u64)(start_index + pages_locked - 1)) <<
1210                               PAGE_CACHE_SHIFT);
1211         }
1212         return ret;
1213 }
1214
1215 /*
1216  * find a contiguous range of bytes in the file marked as delalloc, not
1217  * more than 'max_bytes'.  start and end are used to return the range,
1218  *
1219  * 1 is returned if we find something, 0 if nothing was in the tree
1220  */
1221 static noinline u64 find_lock_delalloc_range(struct inode *inode,
1222                                              struct extent_io_tree *tree,
1223                                              struct page *locked_page,
1224                                              u64 *start, u64 *end,
1225                                              u64 max_bytes)
1226 {
1227         u64 delalloc_start;
1228         u64 delalloc_end;
1229         u64 found;
1230         int ret;
1231         int loops = 0;
1232
1233 again:
1234         /* step one, find a bunch of delalloc bytes starting at start */
1235         delalloc_start = *start;
1236         delalloc_end = 0;
1237         found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1238                                     max_bytes);
1239         if (!found || delalloc_end <= *start) {
1240                 *start = delalloc_start;
1241                 *end = delalloc_end;
1242                 return found;
1243         }
1244
1245         /*
1246          * start comes from the offset of locked_page.  We have to lock
1247          * pages in order, so we can't process delalloc bytes before
1248          * locked_page
1249          */
1250         if (delalloc_start < *start)
1251                 delalloc_start = *start;
1252
1253         /*
1254          * make sure to limit the number of pages we try to lock down
1255          * if we're looping.
1256          */
1257         if (delalloc_end + 1 - delalloc_start > max_bytes && loops)
1258                 delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1;
1259
1260         /* step two, lock all the pages after the page that has start */
1261         ret = lock_delalloc_pages(inode, locked_page,
1262                                   delalloc_start, delalloc_end);
1263         if (ret == -EAGAIN) {
1264                 /* some of the pages are gone, lets avoid looping by
1265                  * shortening the size of the delalloc range we're searching
1266                  */
1267                 if (!loops) {
1268                         unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1);
1269                         max_bytes = PAGE_CACHE_SIZE - offset;
1270                         loops = 1;
1271                         goto again;
1272                 } else {
1273                         found = 0;
1274                         goto out_failed;
1275                 }
1276         }
1277         BUG_ON(ret);
1278
1279         /* step three, lock the state bits for the whole range */
1280         lock_extent(tree, delalloc_start, delalloc_end, GFP_NOFS);
1281
1282         /* then test to make sure it is all still delalloc */
1283         ret = test_range_bit(tree, delalloc_start, delalloc_end,
1284                              EXTENT_DELALLOC, 1);
1285         if (!ret) {
1286                 unlock_extent(tree, delalloc_start, delalloc_end, GFP_NOFS);
1287                 __unlock_for_delalloc(inode, locked_page,
1288                               delalloc_start, delalloc_end);
1289                 cond_resched();
1290                 goto again;
1291         }
1292         *start = delalloc_start;
1293         *end = delalloc_end;
1294 out_failed:
1295         return found;
1296 }
1297
1298 int extent_clear_unlock_delalloc(struct inode *inode,
1299                                 struct extent_io_tree *tree,
1300                                 u64 start, u64 end, struct page *locked_page,
1301                                 int unlock_pages,
1302                                 int clear_unlock,
1303                                 int clear_delalloc, int clear_dirty,
1304                                 int set_writeback,
1305                                 int end_writeback)
1306 {
1307         int ret;
1308         struct page *pages[16];
1309         unsigned long index = start >> PAGE_CACHE_SHIFT;
1310         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1311         unsigned long nr_pages = end_index - index + 1;
1312         int i;
1313         int clear_bits = 0;
1314
1315         if (clear_unlock)
1316                 clear_bits |= EXTENT_LOCKED;
1317         if (clear_dirty)
1318                 clear_bits |= EXTENT_DIRTY;
1319
1320         if (clear_delalloc)
1321                 clear_bits |= EXTENT_DELALLOC;
1322
1323         clear_extent_bit(tree, start, end, clear_bits, 1, 0, GFP_NOFS);
1324         if (!(unlock_pages || clear_dirty || set_writeback || end_writeback))
1325                 return 0;
1326
1327         while (nr_pages > 0) {
1328                 ret = find_get_pages_contig(inode->i_mapping, index,
1329                                      min_t(unsigned long,
1330                                      nr_pages, ARRAY_SIZE(pages)), pages);
1331                 for (i = 0; i < ret; i++) {
1332                         if (pages[i] == locked_page) {
1333                                 page_cache_release(pages[i]);
1334                                 continue;
1335                         }
1336                         if (clear_dirty)
1337                                 clear_page_dirty_for_io(pages[i]);
1338                         if (set_writeback)
1339                                 set_page_writeback(pages[i]);
1340                         if (end_writeback)
1341                                 end_page_writeback(pages[i]);
1342                         if (unlock_pages)
1343                                 unlock_page(pages[i]);
1344                         page_cache_release(pages[i]);
1345                 }
1346                 nr_pages -= ret;
1347                 index += ret;
1348                 cond_resched();
1349         }
1350         return 0;
1351 }
1352
1353 /*
1354  * count the number of bytes in the tree that have a given bit(s)
1355  * set.  This can be fairly slow, except for EXTENT_DIRTY which is
1356  * cached.  The total number found is returned.
1357  */
1358 u64 count_range_bits(struct extent_io_tree *tree,
1359                      u64 *start, u64 search_end, u64 max_bytes,
1360                      unsigned long bits)
1361 {
1362         struct rb_node *node;
1363         struct extent_state *state;
1364         u64 cur_start = *start;
1365         u64 total_bytes = 0;
1366         int found = 0;
1367
1368         if (search_end <= cur_start) {
1369                 WARN_ON(1);
1370                 return 0;
1371         }
1372
1373         spin_lock(&tree->lock);
1374         if (cur_start == 0 && bits == EXTENT_DIRTY) {
1375                 total_bytes = tree->dirty_bytes;
1376                 goto out;
1377         }
1378         /*
1379          * this search will find all the extents that end after
1380          * our range starts.
1381          */
1382         node = tree_search(tree, cur_start);
1383         if (!node)
1384                 goto out;
1385
1386         while (1) {
1387                 state = rb_entry(node, struct extent_state, rb_node);
1388                 if (state->start > search_end)
1389                         break;
1390                 if (state->end >= cur_start && (state->state & bits)) {
1391                         total_bytes += min(search_end, state->end) + 1 -
1392                                        max(cur_start, state->start);
1393                         if (total_bytes >= max_bytes)
1394                                 break;
1395                         if (!found) {
1396                                 *start = state->start;
1397                                 found = 1;
1398                         }
1399                 }
1400                 node = rb_next(node);
1401                 if (!node)
1402                         break;
1403         }
1404 out:
1405         spin_unlock(&tree->lock);
1406         return total_bytes;
1407 }
1408
1409 #if 0
1410 /*
1411  * helper function to lock both pages and extents in the tree.
1412  * pages must be locked first.
1413  */
1414 static int lock_range(struct extent_io_tree *tree, u64 start, u64 end)
1415 {
1416         unsigned long index = start >> PAGE_CACHE_SHIFT;
1417         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1418         struct page *page;
1419         int err;
1420
1421         while (index <= end_index) {
1422                 page = grab_cache_page(tree->mapping, index);
1423                 if (!page) {
1424                         err = -ENOMEM;
1425                         goto failed;
1426                 }
1427                 if (IS_ERR(page)) {
1428                         err = PTR_ERR(page);
1429                         goto failed;
1430                 }
1431                 index++;
1432         }
1433         lock_extent(tree, start, end, GFP_NOFS);
1434         return 0;
1435
1436 failed:
1437         /*
1438          * we failed above in getting the page at 'index', so we undo here
1439          * up to but not including the page at 'index'
1440          */
1441         end_index = index;
1442         index = start >> PAGE_CACHE_SHIFT;
1443         while (index < end_index) {
1444                 page = find_get_page(tree->mapping, index);
1445                 unlock_page(page);
1446                 page_cache_release(page);
1447                 index++;
1448         }
1449         return err;
1450 }
1451
1452 /*
1453  * helper function to unlock both pages and extents in the tree.
1454  */
1455 static int unlock_range(struct extent_io_tree *tree, u64 start, u64 end)
1456 {
1457         unsigned long index = start >> PAGE_CACHE_SHIFT;
1458         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1459         struct page *page;
1460
1461         while (index <= end_index) {
1462                 page = find_get_page(tree->mapping, index);
1463                 unlock_page(page);
1464                 page_cache_release(page);
1465                 index++;
1466         }
1467         unlock_extent(tree, start, end, GFP_NOFS);
1468         return 0;
1469 }
1470 #endif
1471
1472 /*
1473  * set the private field for a given byte offset in the tree.  If there isn't
1474  * an extent_state there already, this does nothing.
1475  */
1476 int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1477 {
1478         struct rb_node *node;
1479         struct extent_state *state;
1480         int ret = 0;
1481
1482         spin_lock(&tree->lock);
1483         /*
1484          * this search will find all the extents that end after
1485          * our range starts.
1486          */
1487         node = tree_search(tree, start);
1488         if (!node) {
1489                 ret = -ENOENT;
1490                 goto out;
1491         }
1492         state = rb_entry(node, struct extent_state, rb_node);
1493         if (state->start != start) {
1494                 ret = -ENOENT;
1495                 goto out;
1496         }
1497         state->private = private;
1498 out:
1499         spin_unlock(&tree->lock);
1500         return ret;
1501 }
1502
1503 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1504 {
1505         struct rb_node *node;
1506         struct extent_state *state;
1507         int ret = 0;
1508
1509         spin_lock(&tree->lock);
1510         /*
1511          * this search will find all the extents that end after
1512          * our range starts.
1513          */
1514         node = tree_search(tree, start);
1515         if (!node) {
1516                 ret = -ENOENT;
1517                 goto out;
1518         }
1519         state = rb_entry(node, struct extent_state, rb_node);
1520         if (state->start != start) {
1521                 ret = -ENOENT;
1522                 goto out;
1523         }
1524         *private = state->private;
1525 out:
1526         spin_unlock(&tree->lock);
1527         return ret;
1528 }
1529
1530 /*
1531  * searches a range in the state tree for a given mask.
1532  * If 'filled' == 1, this returns 1 only if every extent in the tree
1533  * has the bits set.  Otherwise, 1 is returned if any bit in the
1534  * range is found set.
1535  */
1536 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1537                    int bits, int filled)
1538 {
1539         struct extent_state *state = NULL;
1540         struct rb_node *node;
1541         int bitset = 0;
1542
1543         spin_lock(&tree->lock);
1544         node = tree_search(tree, start);
1545         while (node && start <= end) {
1546                 state = rb_entry(node, struct extent_state, rb_node);
1547
1548                 if (filled && state->start > start) {
1549                         bitset = 0;
1550                         break;
1551                 }
1552
1553                 if (state->start > end)
1554                         break;
1555
1556                 if (state->state & bits) {
1557                         bitset = 1;
1558                         if (!filled)
1559                                 break;
1560                 } else if (filled) {
1561                         bitset = 0;
1562                         break;
1563                 }
1564                 start = state->end + 1;
1565                 if (start > end)
1566                         break;
1567                 node = rb_next(node);
1568                 if (!node) {
1569                         if (filled)
1570                                 bitset = 0;
1571                         break;
1572                 }
1573         }
1574         spin_unlock(&tree->lock);
1575         return bitset;
1576 }
1577
1578 /*
1579  * helper function to set a given page up to date if all the
1580  * extents in the tree for that page are up to date
1581  */
1582 static int check_page_uptodate(struct extent_io_tree *tree,
1583                                struct page *page)
1584 {
1585         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1586         u64 end = start + PAGE_CACHE_SIZE - 1;
1587         if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1))
1588                 SetPageUptodate(page);
1589         return 0;
1590 }
1591
1592 /*
1593  * helper function to unlock a page if all the extents in the tree
1594  * for that page are unlocked
1595  */
1596 static int check_page_locked(struct extent_io_tree *tree,
1597                              struct page *page)
1598 {
1599         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1600         u64 end = start + PAGE_CACHE_SIZE - 1;
1601         if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0))
1602                 unlock_page(page);
1603         return 0;
1604 }
1605
1606 /*
1607  * helper function to end page writeback if all the extents
1608  * in the tree for that page are done with writeback
1609  */
1610 static int check_page_writeback(struct extent_io_tree *tree,
1611                              struct page *page)
1612 {
1613         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1614         u64 end = start + PAGE_CACHE_SIZE - 1;
1615         if (!test_range_bit(tree, start, end, EXTENT_WRITEBACK, 0))
1616                 end_page_writeback(page);
1617         return 0;
1618 }
1619
1620 /* lots and lots of room for performance fixes in the end_bio funcs */
1621
1622 /*
1623  * after a writepage IO is done, we need to:
1624  * clear the uptodate bits on error
1625  * clear the writeback bits in the extent tree for this IO
1626  * end_page_writeback if the page has no more pending IO
1627  *
1628  * Scheduling is not allowed, so the extent state tree is expected
1629  * to have one and only one object corresponding to this IO.
1630  */
1631 static void end_bio_extent_writepage(struct bio *bio, int err)
1632 {
1633         int uptodate = err == 0;
1634         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1635         struct extent_io_tree *tree;
1636         u64 start;
1637         u64 end;
1638         int whole_page;
1639         int ret;
1640
1641         do {
1642                 struct page *page = bvec->bv_page;
1643                 tree = &BTRFS_I(page->mapping->host)->io_tree;
1644
1645                 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1646                          bvec->bv_offset;
1647                 end = start + bvec->bv_len - 1;
1648
1649                 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1650                         whole_page = 1;
1651                 else
1652                         whole_page = 0;
1653
1654                 if (--bvec >= bio->bi_io_vec)
1655                         prefetchw(&bvec->bv_page->flags);
1656                 if (tree->ops && tree->ops->writepage_end_io_hook) {
1657                         ret = tree->ops->writepage_end_io_hook(page, start,
1658                                                        end, NULL, uptodate);
1659                         if (ret)
1660                                 uptodate = 0;
1661                 }
1662
1663                 if (!uptodate && tree->ops &&
1664                     tree->ops->writepage_io_failed_hook) {
1665                         ret = tree->ops->writepage_io_failed_hook(bio, page,
1666                                                          start, end, NULL);
1667                         if (ret == 0) {
1668                                 uptodate = (err == 0);
1669                                 continue;
1670                         }
1671                 }
1672
1673                 if (!uptodate) {
1674                         clear_extent_uptodate(tree, start, end, GFP_ATOMIC);
1675                         ClearPageUptodate(page);
1676                         SetPageError(page);
1677                 }
1678
1679                 clear_extent_writeback(tree, start, end, GFP_ATOMIC);
1680
1681                 if (whole_page)
1682                         end_page_writeback(page);
1683                 else
1684                         check_page_writeback(tree, page);
1685         } while (bvec >= bio->bi_io_vec);
1686
1687         bio_put(bio);
1688 }
1689
1690 /*
1691  * after a readpage IO is done, we need to:
1692  * clear the uptodate bits on error
1693  * set the uptodate bits if things worked
1694  * set the page up to date if all extents in the tree are uptodate
1695  * clear the lock bit in the extent tree
1696  * unlock the page if there are no other extents locked for it
1697  *
1698  * Scheduling is not allowed, so the extent state tree is expected
1699  * to have one and only one object corresponding to this IO.
1700  */
1701 static void end_bio_extent_readpage(struct bio *bio, int err)
1702 {
1703         int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1704         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1705         struct extent_io_tree *tree;
1706         u64 start;
1707         u64 end;
1708         int whole_page;
1709         int ret;
1710
1711         if (err)
1712                 uptodate = 0;
1713
1714         do {
1715                 struct page *page = bvec->bv_page;
1716                 tree = &BTRFS_I(page->mapping->host)->io_tree;
1717
1718                 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1719                         bvec->bv_offset;
1720                 end = start + bvec->bv_len - 1;
1721
1722                 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1723                         whole_page = 1;
1724                 else
1725                         whole_page = 0;
1726
1727                 if (--bvec >= bio->bi_io_vec)
1728                         prefetchw(&bvec->bv_page->flags);
1729
1730                 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
1731                         ret = tree->ops->readpage_end_io_hook(page, start, end,
1732                                                               NULL);
1733                         if (ret)
1734                                 uptodate = 0;
1735                 }
1736                 if (!uptodate && tree->ops &&
1737                     tree->ops->readpage_io_failed_hook) {
1738                         ret = tree->ops->readpage_io_failed_hook(bio, page,
1739                                                          start, end, NULL);
1740                         if (ret == 0) {
1741                                 uptodate =
1742                                         test_bit(BIO_UPTODATE, &bio->bi_flags);
1743                                 if (err)
1744                                         uptodate = 0;
1745                                 continue;
1746                         }
1747                 }
1748
1749                 if (uptodate) {
1750                         set_extent_uptodate(tree, start, end,
1751                                             GFP_ATOMIC);
1752                 }
1753                 unlock_extent(tree, start, end, GFP_ATOMIC);
1754
1755                 if (whole_page) {
1756                         if (uptodate) {
1757                                 SetPageUptodate(page);
1758                         } else {
1759                                 ClearPageUptodate(page);
1760                                 SetPageError(page);
1761                         }
1762                         unlock_page(page);
1763                 } else {
1764                         if (uptodate) {
1765                                 check_page_uptodate(tree, page);
1766                         } else {
1767                                 ClearPageUptodate(page);
1768                                 SetPageError(page);
1769                         }
1770                         check_page_locked(tree, page);
1771                 }
1772         } while (bvec >= bio->bi_io_vec);
1773
1774         bio_put(bio);
1775 }
1776
1777 /*
1778  * IO done from prepare_write is pretty simple, we just unlock
1779  * the structs in the extent tree when done, and set the uptodate bits
1780  * as appropriate.
1781  */
1782 static void end_bio_extent_preparewrite(struct bio *bio, int err)
1783 {
1784         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1785         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1786         struct extent_io_tree *tree;
1787         u64 start;
1788         u64 end;
1789
1790         do {
1791                 struct page *page = bvec->bv_page;
1792                 tree = &BTRFS_I(page->mapping->host)->io_tree;
1793
1794                 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1795                         bvec->bv_offset;
1796                 end = start + bvec->bv_len - 1;
1797
1798                 if (--bvec >= bio->bi_io_vec)
1799                         prefetchw(&bvec->bv_page->flags);
1800
1801                 if (uptodate) {
1802                         set_extent_uptodate(tree, start, end, GFP_ATOMIC);
1803                 } else {
1804                         ClearPageUptodate(page);
1805                         SetPageError(page);
1806                 }
1807
1808                 unlock_extent(tree, start, end, GFP_ATOMIC);
1809
1810         } while (bvec >= bio->bi_io_vec);
1811
1812         bio_put(bio);
1813 }
1814
1815 static struct bio *
1816 extent_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
1817                  gfp_t gfp_flags)
1818 {
1819         struct bio *bio;
1820
1821         bio = bio_alloc(gfp_flags, nr_vecs);
1822
1823         if (bio == NULL && (current->flags & PF_MEMALLOC)) {
1824                 while (!bio && (nr_vecs /= 2))
1825                         bio = bio_alloc(gfp_flags, nr_vecs);
1826         }
1827
1828         if (bio) {
1829                 bio->bi_size = 0;
1830                 bio->bi_bdev = bdev;
1831                 bio->bi_sector = first_sector;
1832         }
1833         return bio;
1834 }
1835
1836 static int submit_one_bio(int rw, struct bio *bio, int mirror_num,
1837                           unsigned long bio_flags)
1838 {
1839         int ret = 0;
1840         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1841         struct page *page = bvec->bv_page;
1842         struct extent_io_tree *tree = bio->bi_private;
1843         u64 start;
1844         u64 end;
1845
1846         start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
1847         end = start + bvec->bv_len - 1;
1848
1849         bio->bi_private = NULL;
1850
1851         bio_get(bio);
1852
1853         if (tree->ops && tree->ops->submit_bio_hook)
1854                 tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
1855                                            mirror_num, bio_flags);
1856         else
1857                 submit_bio(rw, bio);
1858         if (bio_flagged(bio, BIO_EOPNOTSUPP))
1859                 ret = -EOPNOTSUPP;
1860         bio_put(bio);
1861         return ret;
1862 }
1863
1864 static int submit_extent_page(int rw, struct extent_io_tree *tree,
1865                               struct page *page, sector_t sector,
1866                               size_t size, unsigned long offset,
1867                               struct block_device *bdev,
1868                               struct bio **bio_ret,
1869                               unsigned long max_pages,
1870                               bio_end_io_t end_io_func,
1871                               int mirror_num,
1872                               unsigned long prev_bio_flags,
1873                               unsigned long bio_flags)
1874 {
1875         int ret = 0;
1876         struct bio *bio;
1877         int nr;
1878         int contig = 0;
1879         int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED;
1880         int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
1881         size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE);
1882
1883         if (bio_ret && *bio_ret) {
1884                 bio = *bio_ret;
1885                 if (old_compressed)
1886                         contig = bio->bi_sector == sector;
1887                 else
1888                         contig = bio->bi_sector + (bio->bi_size >> 9) ==
1889                                 sector;
1890
1891                 if (prev_bio_flags != bio_flags || !contig ||
1892                     (tree->ops && tree->ops->merge_bio_hook &&
1893                      tree->ops->merge_bio_hook(page, offset, page_size, bio,
1894                                                bio_flags)) ||
1895                     bio_add_page(bio, page, page_size, offset) < page_size) {
1896                         ret = submit_one_bio(rw, bio, mirror_num,
1897                                              prev_bio_flags);
1898                         bio = NULL;
1899                 } else {
1900                         return 0;
1901                 }
1902         }
1903         if (this_compressed)
1904                 nr = BIO_MAX_PAGES;
1905         else
1906                 nr = bio_get_nr_vecs(bdev);
1907
1908         bio = extent_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
1909
1910         bio_add_page(bio, page, page_size, offset);
1911         bio->bi_end_io = end_io_func;
1912         bio->bi_private = tree;
1913
1914         if (bio_ret)
1915                 *bio_ret = bio;
1916         else
1917                 ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
1918
1919         return ret;
1920 }
1921
1922 void set_page_extent_mapped(struct page *page)
1923 {
1924         if (!PagePrivate(page)) {
1925                 SetPagePrivate(page);
1926                 page_cache_get(page);
1927                 set_page_private(page, EXTENT_PAGE_PRIVATE);
1928         }
1929 }
1930
1931 static void set_page_extent_head(struct page *page, unsigned long len)
1932 {
1933         set_page_private(page, EXTENT_PAGE_PRIVATE_FIRST_PAGE | len << 2);
1934 }
1935
1936 /*
1937  * basic readpage implementation.  Locked extent state structs are inserted
1938  * into the tree that are removed when the IO is done (by the end_io
1939  * handlers)
1940  */
1941 static int __extent_read_full_page(struct extent_io_tree *tree,
1942                                    struct page *page,
1943                                    get_extent_t *get_extent,
1944                                    struct bio **bio, int mirror_num,
1945                                    unsigned long *bio_flags)
1946 {
1947         struct inode *inode = page->mapping->host;
1948         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1949         u64 page_end = start + PAGE_CACHE_SIZE - 1;
1950         u64 end;
1951         u64 cur = start;
1952         u64 extent_offset;
1953         u64 last_byte = i_size_read(inode);
1954         u64 block_start;
1955         u64 cur_end;
1956         sector_t sector;
1957         struct extent_map *em;
1958         struct block_device *bdev;
1959         int ret;
1960         int nr = 0;
1961         size_t page_offset = 0;
1962         size_t iosize;
1963         size_t disk_io_size;
1964         size_t blocksize = inode->i_sb->s_blocksize;
1965         unsigned long this_bio_flag = 0;
1966
1967         set_page_extent_mapped(page);
1968
1969         end = page_end;
1970         lock_extent(tree, start, end, GFP_NOFS);
1971
1972         if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
1973                 char *userpage;
1974                 size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
1975
1976                 if (zero_offset) {
1977                         iosize = PAGE_CACHE_SIZE - zero_offset;
1978                         userpage = kmap_atomic(page, KM_USER0);
1979                         memset(userpage + zero_offset, 0, iosize);
1980                         flush_dcache_page(page);
1981                         kunmap_atomic(userpage, KM_USER0);
1982                 }
1983         }
1984         while (cur <= end) {
1985                 if (cur >= last_byte) {
1986                         char *userpage;
1987                         iosize = PAGE_CACHE_SIZE - page_offset;
1988                         userpage = kmap_atomic(page, KM_USER0);
1989                         memset(userpage + page_offset, 0, iosize);
1990                         flush_dcache_page(page);
1991                         kunmap_atomic(userpage, KM_USER0);
1992                         set_extent_uptodate(tree, cur, cur + iosize - 1,
1993                                             GFP_NOFS);
1994                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1995                         break;
1996                 }
1997                 em = get_extent(inode, page, page_offset, cur,
1998                                 end - cur + 1, 0);
1999                 if (IS_ERR(em) || !em) {
2000                         SetPageError(page);
2001                         unlock_extent(tree, cur, end, GFP_NOFS);
2002                         break;
2003                 }
2004                 extent_offset = cur - em->start;
2005                 BUG_ON(extent_map_end(em) <= cur);
2006                 BUG_ON(end < cur);
2007
2008                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
2009                         this_bio_flag = EXTENT_BIO_COMPRESSED;
2010
2011                 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2012                 cur_end = min(extent_map_end(em) - 1, end);
2013                 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2014                 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2015                         disk_io_size = em->block_len;
2016                         sector = em->block_start >> 9;
2017                 } else {
2018                         sector = (em->block_start + extent_offset) >> 9;
2019                         disk_io_size = iosize;
2020                 }
2021                 bdev = em->bdev;
2022                 block_start = em->block_start;
2023                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2024                         block_start = EXTENT_MAP_HOLE;
2025                 free_extent_map(em);
2026                 em = NULL;
2027
2028                 /* we've found a hole, just zero and go on */
2029                 if (block_start == EXTENT_MAP_HOLE) {
2030                         char *userpage;
2031                         userpage = kmap_atomic(page, KM_USER0);
2032                         memset(userpage + page_offset, 0, iosize);
2033                         flush_dcache_page(page);
2034                         kunmap_atomic(userpage, KM_USER0);
2035
2036                         set_extent_uptodate(tree, cur, cur + iosize - 1,
2037                                             GFP_NOFS);
2038                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
2039                         cur = cur + iosize;
2040                         page_offset += iosize;
2041                         continue;
2042                 }
2043                 /* the get_extent function already copied into the page */
2044                 if (test_range_bit(tree, cur, cur_end, EXTENT_UPTODATE, 1)) {
2045                         check_page_uptodate(tree, page);
2046                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
2047                         cur = cur + iosize;
2048                         page_offset += iosize;
2049                         continue;
2050                 }
2051                 /* we have an inline extent but it didn't get marked up
2052                  * to date.  Error out
2053                  */
2054                 if (block_start == EXTENT_MAP_INLINE) {
2055                         SetPageError(page);
2056                         unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
2057                         cur = cur + iosize;
2058                         page_offset += iosize;
2059                         continue;
2060                 }
2061
2062                 ret = 0;
2063                 if (tree->ops && tree->ops->readpage_io_hook) {
2064                         ret = tree->ops->readpage_io_hook(page, cur,
2065                                                           cur + iosize - 1);
2066                 }
2067                 if (!ret) {
2068                         unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
2069                         pnr -= page->index;
2070                         ret = submit_extent_page(READ, tree, page,
2071                                          sector, disk_io_size, page_offset,
2072                                          bdev, bio, pnr,
2073                                          end_bio_extent_readpage, mirror_num,
2074                                          *bio_flags,
2075                                          this_bio_flag);
2076                         nr++;
2077                         *bio_flags = this_bio_flag;
2078                 }
2079                 if (ret)
2080                         SetPageError(page);
2081                 cur = cur + iosize;
2082                 page_offset += iosize;
2083         }
2084         if (!nr) {
2085                 if (!PageError(page))
2086                         SetPageUptodate(page);
2087                 unlock_page(page);
2088         }
2089         return 0;
2090 }
2091
2092 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
2093                             get_extent_t *get_extent)
2094 {
2095         struct bio *bio = NULL;
2096         unsigned long bio_flags = 0;
2097         int ret;
2098
2099         ret = __extent_read_full_page(tree, page, get_extent, &bio, 0,
2100                                       &bio_flags);
2101         if (bio)
2102                 submit_one_bio(READ, bio, 0, bio_flags);
2103         return ret;
2104 }
2105
2106 /*
2107  * the writepage semantics are similar to regular writepage.  extent
2108  * records are inserted to lock ranges in the tree, and as dirty areas
2109  * are found, they are marked writeback.  Then the lock bits are removed
2110  * and the end_io handler clears the writeback ranges
2111  */
2112 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2113                               void *data)
2114 {
2115         struct inode *inode = page->mapping->host;
2116         struct extent_page_data *epd = data;
2117         struct extent_io_tree *tree = epd->tree;
2118         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2119         u64 delalloc_start;
2120         u64 page_end = start + PAGE_CACHE_SIZE - 1;
2121         u64 end;
2122         u64 cur = start;
2123         u64 extent_offset;
2124         u64 last_byte = i_size_read(inode);
2125         u64 block_start;
2126         u64 iosize;
2127         u64 unlock_start;
2128         sector_t sector;
2129         struct extent_map *em;
2130         struct block_device *bdev;
2131         int ret;
2132         int nr = 0;
2133         size_t pg_offset = 0;
2134         size_t blocksize;
2135         loff_t i_size = i_size_read(inode);
2136         unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
2137         u64 nr_delalloc;
2138         u64 delalloc_end;
2139         int page_started;
2140         int compressed;
2141         unsigned long nr_written = 0;
2142
2143         WARN_ON(!PageLocked(page));
2144         pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
2145         if (page->index > end_index ||
2146            (page->index == end_index && !pg_offset)) {
2147                 page->mapping->a_ops->invalidatepage(page, 0);
2148                 unlock_page(page);
2149                 return 0;
2150         }
2151
2152         if (page->index == end_index) {
2153                 char *userpage;
2154
2155                 userpage = kmap_atomic(page, KM_USER0);
2156                 memset(userpage + pg_offset, 0,
2157                        PAGE_CACHE_SIZE - pg_offset);
2158                 kunmap_atomic(userpage, KM_USER0);
2159                 flush_dcache_page(page);
2160         }
2161         pg_offset = 0;
2162
2163         set_page_extent_mapped(page);
2164
2165         delalloc_start = start;
2166         delalloc_end = 0;
2167         page_started = 0;
2168         if (!epd->extent_locked) {
2169                 while (delalloc_end < page_end) {
2170                         nr_delalloc = find_lock_delalloc_range(inode, tree,
2171                                                        page,
2172                                                        &delalloc_start,
2173                                                        &delalloc_end,
2174                                                        128 * 1024 * 1024);
2175                         if (nr_delalloc == 0) {
2176                                 delalloc_start = delalloc_end + 1;
2177                                 continue;
2178                         }
2179                         tree->ops->fill_delalloc(inode, page, delalloc_start,
2180                                                  delalloc_end, &page_started,
2181                                                  &nr_written);
2182                         delalloc_start = delalloc_end + 1;
2183                 }
2184
2185                 /* did the fill delalloc function already unlock and start
2186                  * the IO?
2187                  */
2188                 if (page_started) {
2189                         ret = 0;
2190                         goto update_nr_written;
2191                 }
2192         }
2193         lock_extent(tree, start, page_end, GFP_NOFS);
2194
2195         unlock_start = start;
2196
2197         if (tree->ops && tree->ops->writepage_start_hook) {
2198                 ret = tree->ops->writepage_start_hook(page, start,
2199                                                       page_end);
2200                 if (ret == -EAGAIN) {
2201                         unlock_extent(tree, start, page_end, GFP_NOFS);
2202                         redirty_page_for_writepage(wbc, page);
2203                         unlock_page(page);
2204                         ret = 0;
2205                         goto update_nr_written;
2206                 }
2207         }
2208
2209         nr_written++;
2210
2211         end = page_end;
2212         if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0))
2213                 printk(KERN_ERR "btrfs delalloc bits after lock_extent\n");
2214
2215         if (last_byte <= start) {
2216                 clear_extent_dirty(tree, start, page_end, GFP_NOFS);
2217                 unlock_extent(tree, start, page_end, GFP_NOFS);
2218                 if (tree->ops && tree->ops->writepage_end_io_hook)
2219                         tree->ops->writepage_end_io_hook(page, start,
2220                                                          page_end, NULL, 1);
2221                 unlock_start = page_end + 1;
2222                 goto done;
2223         }
2224
2225         set_extent_uptodate(tree, start, page_end, GFP_NOFS);
2226         blocksize = inode->i_sb->s_blocksize;
2227
2228         while (cur <= end) {
2229                 if (cur >= last_byte) {
2230                         clear_extent_dirty(tree, cur, page_end, GFP_NOFS);
2231                         unlock_extent(tree, unlock_start, page_end, GFP_NOFS);
2232                         if (tree->ops && tree->ops->writepage_end_io_hook)
2233                                 tree->ops->writepage_end_io_hook(page, cur,
2234                                                          page_end, NULL, 1);
2235                         unlock_start = page_end + 1;
2236                         break;
2237                 }
2238                 em = epd->get_extent(inode, page, pg_offset, cur,
2239                                      end - cur + 1, 1);
2240                 if (IS_ERR(em) || !em) {
2241                         SetPageError(page);
2242                         break;
2243                 }
2244
2245                 extent_offset = cur - em->start;
2246                 BUG_ON(extent_map_end(em) <= cur);
2247                 BUG_ON(end < cur);
2248                 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2249                 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2250                 sector = (em->block_start + extent_offset) >> 9;
2251                 bdev = em->bdev;
2252                 block_start = em->block_start;
2253                 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
2254                 free_extent_map(em);
2255                 em = NULL;
2256
2257                 /*
2258                  * compressed and inline extents are written through other
2259                  * paths in the FS
2260                  */
2261                 if (compressed || block_start == EXTENT_MAP_HOLE ||
2262                     block_start == EXTENT_MAP_INLINE) {
2263                         clear_extent_dirty(tree, cur,
2264                                            cur + iosize - 1, GFP_NOFS);
2265
2266                         unlock_extent(tree, unlock_start, cur + iosize - 1,
2267                                       GFP_NOFS);
2268
2269                         /*
2270                          * end_io notification does not happen here for
2271                          * compressed extents
2272                          */
2273                         if (!compressed && tree->ops &&
2274                             tree->ops->writepage_end_io_hook)
2275                                 tree->ops->writepage_end_io_hook(page, cur,
2276                                                          cur + iosize - 1,
2277                                                          NULL, 1);
2278                         else if (compressed) {
2279                                 /* we don't want to end_page_writeback on
2280                                  * a compressed extent.  this happens
2281                                  * elsewhere
2282                                  */
2283                                 nr++;
2284                         }
2285
2286                         cur += iosize;
2287                         pg_offset += iosize;
2288                         unlock_start = cur;
2289                         continue;
2290                 }
2291                 /* leave this out until we have a page_mkwrite call */
2292                 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
2293                                    EXTENT_DIRTY, 0)) {
2294                         cur = cur + iosize;
2295                         pg_offset += iosize;
2296                         continue;
2297                 }
2298
2299                 clear_extent_dirty(tree, cur, cur + iosize - 1, GFP_NOFS);
2300                 if (tree->ops && tree->ops->writepage_io_hook) {
2301                         ret = tree->ops->writepage_io_hook(page, cur,
2302                                                 cur + iosize - 1);
2303                 } else {
2304                         ret = 0;
2305                 }
2306                 if (ret) {
2307                         SetPageError(page);
2308                 } else {
2309                         unsigned long max_nr = end_index + 1;
2310
2311                         set_range_writeback(tree, cur, cur + iosize - 1);
2312                         if (!PageWriteback(page)) {
2313                                 printk(KERN_ERR "btrfs warning page %lu not "
2314                                        "writeback, cur %llu end %llu\n",
2315                                        page->index, (unsigned long long)cur,
2316                                        (unsigned long long)end);
2317                         }
2318
2319                         ret = submit_extent_page(WRITE, tree, page, sector,
2320                                                  iosize, pg_offset, bdev,
2321                                                  &epd->bio, max_nr,
2322                                                  end_bio_extent_writepage,
2323                                                  0, 0, 0);
2324                         if (ret)
2325                                 SetPageError(page);
2326                 }
2327                 cur = cur + iosize;
2328                 pg_offset += iosize;
2329                 nr++;
2330         }
2331 done:
2332         if (nr == 0) {
2333                 /* make sure the mapping tag for page dirty gets cleared */
2334                 set_page_writeback(page);
2335                 end_page_writeback(page);
2336         }
2337         if (unlock_start <= page_end)
2338                 unlock_extent(tree, unlock_start, page_end, GFP_NOFS);
2339         unlock_page(page);
2340
2341 update_nr_written:
2342         wbc->nr_to_write -= nr_written;
2343         if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
2344             wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
2345                 page->mapping->writeback_index = page->index + nr_written;
2346         return 0;
2347 }
2348
2349 /**
2350  * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
2351  * @mapping: address space structure to write
2352  * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2353  * @writepage: function called for each page
2354  * @data: data passed to writepage function
2355  *
2356  * If a page is already under I/O, write_cache_pages() skips it, even
2357  * if it's dirty.  This is desirable behaviour for memory-cleaning writeback,
2358  * but it is INCORRECT for data-integrity system calls such as fsync().  fsync()
2359  * and msync() need to guarantee that all the data which was dirty at the time
2360  * the call was made get new I/O started against them.  If wbc->sync_mode is
2361  * WB_SYNC_ALL then we were called for data integrity and we must wait for
2362  * existing IO to complete.
2363  */
2364 static int extent_write_cache_pages(struct extent_io_tree *tree,
2365                              struct address_space *mapping,
2366                              struct writeback_control *wbc,
2367                              writepage_t writepage, void *data,
2368                              void (*flush_fn)(void *))
2369 {
2370         struct backing_dev_info *bdi = mapping->backing_dev_info;
2371         int ret = 0;
2372         int done = 0;
2373         struct pagevec pvec;
2374         int nr_pages;
2375         pgoff_t index;
2376         pgoff_t end;            /* Inclusive */
2377         int scanned = 0;
2378         int range_whole = 0;
2379
2380         if (wbc->nonblocking && bdi_write_congested(bdi)) {
2381                 wbc->encountered_congestion = 1;
2382                 return 0;
2383         }
2384
2385         pagevec_init(&pvec, 0);
2386         if (wbc->range_cyclic) {
2387                 index = mapping->writeback_index; /* Start from prev offset */
2388                 end = -1;
2389         } else {
2390                 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2391                 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2392                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2393                         range_whole = 1;
2394                 scanned = 1;
2395         }
2396 retry:
2397         while (!done && (index <= end) &&
2398                (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
2399                               PAGECACHE_TAG_DIRTY, min(end - index,
2400                                   (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
2401                 unsigned i;
2402
2403                 scanned = 1;
2404                 for (i = 0; i < nr_pages; i++) {
2405                         struct page *page = pvec.pages[i];
2406
2407                         /*
2408                          * At this point we hold neither mapping->tree_lock nor
2409                          * lock on the page itself: the page may be truncated or
2410                          * invalidated (changing page->mapping to NULL), or even
2411                          * swizzled back from swapper_space to tmpfs file
2412                          * mapping
2413                          */
2414                         if (tree->ops && tree->ops->write_cache_pages_lock_hook)
2415                                 tree->ops->write_cache_pages_lock_hook(page);
2416                         else
2417                                 lock_page(page);
2418
2419                         if (unlikely(page->mapping != mapping)) {
2420                                 unlock_page(page);
2421                                 continue;
2422                         }
2423
2424                         if (!wbc->range_cyclic && page->index > end) {
2425                                 done = 1;
2426                                 unlock_page(page);
2427                                 continue;
2428                         }
2429
2430                         if (wbc->sync_mode != WB_SYNC_NONE) {
2431                                 if (PageWriteback(page))
2432                                         flush_fn(data);
2433                                 wait_on_page_writeback(page);
2434                         }
2435
2436                         if (PageWriteback(page) ||
2437                             !clear_page_dirty_for_io(page)) {
2438                                 unlock_page(page);
2439                                 continue;
2440                         }
2441
2442                         ret = (*writepage)(page, wbc, data);
2443
2444                         if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
2445                                 unlock_page(page);
2446                                 ret = 0;
2447                         }
2448                         if (ret || wbc->nr_to_write <= 0)
2449                                 done = 1;
2450                         if (wbc->nonblocking && bdi_write_congested(bdi)) {
2451                                 wbc->encountered_congestion = 1;
2452                                 done = 1;
2453                         }
2454                 }
2455                 pagevec_release(&pvec);
2456                 cond_resched();
2457         }
2458         if (!scanned && !done) {
2459                 /*
2460                  * We hit the last page and there is more work to be done: wrap
2461                  * back to the start of the file
2462                  */
2463                 scanned = 1;
2464                 index = 0;
2465                 goto retry;
2466         }
2467         return ret;
2468 }
2469
2470 static noinline void flush_write_bio(void *data)
2471 {
2472         struct extent_page_data *epd = data;
2473         if (epd->bio) {
2474                 submit_one_bio(WRITE, epd->bio, 0, 0);
2475                 epd->bio = NULL;
2476         }
2477 }
2478
2479 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
2480                           get_extent_t *get_extent,
2481                           struct writeback_control *wbc)
2482 {
2483         int ret;
2484         struct address_space *mapping = page->mapping;
2485         struct extent_page_data epd = {
2486                 .bio = NULL,
2487                 .tree = tree,
2488                 .get_extent = get_extent,
2489                 .extent_locked = 0,
2490         };
2491         struct writeback_control wbc_writepages = {
2492                 .bdi            = wbc->bdi,
2493                 .sync_mode      = WB_SYNC_NONE,
2494                 .older_than_this = NULL,
2495                 .nr_to_write    = 64,
2496                 .range_start    = page_offset(page) + PAGE_CACHE_SIZE,
2497                 .range_end      = (loff_t)-1,
2498         };
2499
2500
2501         ret = __extent_writepage(page, wbc, &epd);
2502
2503         extent_write_cache_pages(tree, mapping, &wbc_writepages,
2504                                  __extent_writepage, &epd, flush_write_bio);
2505         if (epd.bio)
2506                 submit_one_bio(WRITE, epd.bio, 0, 0);
2507         return ret;
2508 }
2509
2510 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
2511                               u64 start, u64 end, get_extent_t *get_extent,
2512                               int mode)
2513 {
2514         int ret = 0;
2515         struct address_space *mapping = inode->i_mapping;
2516         struct page *page;
2517         unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
2518                 PAGE_CACHE_SHIFT;
2519
2520         struct extent_page_data epd = {
2521                 .bio = NULL,
2522                 .tree = tree,
2523                 .get_extent = get_extent,
2524                 .extent_locked = 1,
2525         };
2526         struct writeback_control wbc_writepages = {
2527                 .bdi            = inode->i_mapping->backing_dev_info,
2528                 .sync_mode      = mode,
2529                 .older_than_this = NULL,
2530                 .nr_to_write    = nr_pages * 2,
2531                 .range_start    = start,
2532                 .range_end      = end + 1,
2533         };
2534
2535         while (start <= end) {
2536                 page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
2537                 if (clear_page_dirty_for_io(page))
2538                         ret = __extent_writepage(page, &wbc_writepages, &epd);
2539                 else {
2540                         if (tree->ops && tree->ops->writepage_end_io_hook)
2541                                 tree->ops->writepage_end_io_hook(page, start,
2542                                                  start + PAGE_CACHE_SIZE - 1,
2543                                                  NULL, 1);
2544                         unlock_page(page);
2545                 }
2546                 page_cache_release(page);
2547                 start += PAGE_CACHE_SIZE;
2548         }
2549
2550         if (epd.bio)
2551                 submit_one_bio(WRITE, epd.bio, 0, 0);
2552         return ret;
2553 }
2554
2555 int extent_writepages(struct extent_io_tree *tree,
2556                       struct address_space *mapping,
2557                       get_extent_t *get_extent,
2558                       struct writeback_control *wbc)
2559 {
2560         int ret = 0;
2561         struct extent_page_data epd = {
2562                 .bio = NULL,
2563                 .tree = tree,
2564                 .get_extent = get_extent,
2565                 .extent_locked = 0,
2566         };
2567
2568         ret = extent_write_cache_pages(tree, mapping, wbc,
2569                                        __extent_writepage, &epd,
2570                                        flush_write_bio);
2571         if (epd.bio)
2572                 submit_one_bio(WRITE, epd.bio, 0, 0);
2573         return ret;
2574 }
2575
2576 int extent_readpages(struct extent_io_tree *tree,
2577                      struct address_space *mapping,
2578                      struct list_head *pages, unsigned nr_pages,
2579                      get_extent_t get_extent)
2580 {
2581         struct bio *bio = NULL;
2582         unsigned page_idx;
2583         struct pagevec pvec;
2584         unsigned long bio_flags = 0;
2585
2586         pagevec_init(&pvec, 0);
2587         for (page_idx = 0; page_idx < nr_pages; page_idx++) {
2588                 struct page *page = list_entry(pages->prev, struct page, lru);
2589
2590                 prefetchw(&page->flags);
2591                 list_del(&page->lru);
2592                 /*
2593                  * what we want to do here is call add_to_page_cache_lru,
2594                  * but that isn't exported, so we reproduce it here
2595                  */
2596                 if (!add_to_page_cache(page, mapping,
2597                                         page->index, GFP_KERNEL)) {
2598
2599                         /* open coding of lru_cache_add, also not exported */
2600                         page_cache_get(page);
2601                         if (!pagevec_add(&pvec, page))
2602                                 __pagevec_lru_add_file(&pvec);
2603                         __extent_read_full_page(tree, page, get_extent,
2604                                                 &bio, 0, &bio_flags);
2605                 }
2606                 page_cache_release(page);
2607         }
2608         if (pagevec_count(&pvec))
2609                 __pagevec_lru_add_file(&pvec);
2610         BUG_ON(!list_empty(pages));
2611         if (bio)
2612                 submit_one_bio(READ, bio, 0, bio_flags);
2613         return 0;
2614 }
2615
2616 /*
2617  * basic invalidatepage code, this waits on any locked or writeback
2618  * ranges corresponding to the page, and then deletes any extent state
2619  * records from the tree
2620  */
2621 int extent_invalidatepage(struct extent_io_tree *tree,
2622                           struct page *page, unsigned long offset)
2623 {
2624         u64 start = ((u64)page->index << PAGE_CACHE_SHIFT);
2625         u64 end = start + PAGE_CACHE_SIZE - 1;
2626         size_t blocksize = page->mapping->host->i_sb->s_blocksize;
2627
2628         start += (offset + blocksize - 1) & ~(blocksize - 1);
2629         if (start > end)
2630                 return 0;
2631
2632         lock_extent(tree, start, end, GFP_NOFS);
2633         wait_on_extent_writeback(tree, start, end);
2634         clear_extent_bit(tree, start, end,
2635                          EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC,
2636                          1, 1, GFP_NOFS);
2637         return 0;
2638 }
2639
2640 /*
2641  * simple commit_write call, set_range_dirty is used to mark both
2642  * the pages and the extent records as dirty
2643  */
2644 int extent_commit_write(struct extent_io_tree *tree,
2645                         struct inode *inode, struct page *page,
2646                         unsigned from, unsigned to)
2647 {
2648         loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
2649
2650         set_page_extent_mapped(page);
2651         set_page_dirty(page);
2652
2653         if (pos > inode->i_size) {
2654                 i_size_write(inode, pos);
2655                 mark_inode_dirty(inode);
2656         }
2657         return 0;
2658 }
2659
2660 int extent_prepare_write(struct extent_io_tree *tree,
2661                          struct inode *inode, struct page *page,
2662                          unsigned from, unsigned to, get_extent_t *get_extent)
2663 {
2664         u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2665         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
2666         u64 block_start;
2667         u64 orig_block_start;
2668         u64 block_end;
2669         u64 cur_end;
2670         struct extent_map *em;
2671         unsigned blocksize = 1 << inode->i_blkbits;
2672         size_t page_offset = 0;
2673         size_t block_off_start;
2674         size_t block_off_end;
2675         int err = 0;
2676         int iocount = 0;
2677         int ret = 0;
2678         int isnew;
2679
2680         set_page_extent_mapped(page);
2681
2682         block_start = (page_start + from) & ~((u64)blocksize - 1);
2683         block_end = (page_start + to - 1) | (blocksize - 1);
2684         orig_block_start = block_start;
2685
2686         lock_extent(tree, page_start, page_end, GFP_NOFS);
2687         while (block_start <= block_end) {
2688                 em = get_extent(inode, page, page_offset, block_start,
2689                                 block_end - block_start + 1, 1);
2690                 if (IS_ERR(em) || !em)
2691                         goto err;
2692
2693                 cur_end = min(block_end, extent_map_end(em) - 1);
2694                 block_off_start = block_start & (PAGE_CACHE_SIZE - 1);
2695                 block_off_end = block_off_start + blocksize;
2696                 isnew = clear_extent_new(tree, block_start, cur_end, GFP_NOFS);
2697
2698                 if (!PageUptodate(page) && isnew &&
2699                     (block_off_end > to || block_off_start < from)) {
2700                         void *kaddr;
2701
2702                         kaddr = kmap_atomic(page, KM_USER0);
2703                         if (block_off_end > to)
2704                                 memset(kaddr + to, 0, block_off_end - to);
2705                         if (block_off_start < from)
2706                                 memset(kaddr + block_off_start, 0,
2707                                        from - block_off_start);
2708                         flush_dcache_page(page);
2709                         kunmap_atomic(kaddr, KM_USER0);
2710                 }
2711                 if ((em->block_start != EXTENT_MAP_HOLE &&
2712                      em->block_start != EXTENT_MAP_INLINE) &&
2713                     !isnew && !PageUptodate(page) &&
2714                     (block_off_end > to || block_off_start < from) &&
2715                     !test_range_bit(tree, block_start, cur_end,
2716                                     EXTENT_UPTODATE, 1)) {
2717                         u64 sector;
2718                         u64 extent_offset = block_start - em->start;
2719                         size_t iosize;
2720                         sector = (em->block_start + extent_offset) >> 9;
2721                         iosize = (cur_end - block_start + blocksize) &
2722                                 ~((u64)blocksize - 1);
2723                         /*
2724                          * we've already got the extent locked, but we
2725                          * need to split the state such that our end_bio
2726                          * handler can clear the lock.
2727                          */
2728                         set_extent_bit(tree, block_start,
2729                                        block_start + iosize - 1,
2730                                        EXTENT_LOCKED, 0, NULL, GFP_NOFS);
2731                         ret = submit_extent_page(READ, tree, page,
2732                                          sector, iosize, page_offset, em->bdev,
2733                                          NULL, 1,
2734                                          end_bio_extent_preparewrite, 0,
2735                                          0, 0);
2736                         iocount++;
2737                         block_start = block_start + iosize;
2738                 } else {
2739                         set_extent_uptodate(tree, block_start, cur_end,
2740                                             GFP_NOFS);
2741                         unlock_extent(tree, block_start, cur_end, GFP_NOFS);
2742                         block_start = cur_end + 1;
2743                 }
2744                 page_offset = block_start & (PAGE_CACHE_SIZE - 1);
2745                 free_extent_map(em);
2746         }
2747         if (iocount) {
2748                 wait_extent_bit(tree, orig_block_start,
2749                                 block_end, EXTENT_LOCKED);
2750         }
2751         check_page_uptodate(tree, page);
2752 err:
2753         /* FIXME, zero out newly allocated blocks on error */
2754         return err;
2755 }
2756
2757 /*
2758  * a helper for releasepage, this tests for areas of the page that
2759  * are locked or under IO and drops the related state bits if it is safe
2760  * to drop the page.
2761  */
2762 int try_release_extent_state(struct extent_map_tree *map,
2763                              struct extent_io_tree *tree, struct page *page,
2764                              gfp_t mask)
2765 {
2766         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2767         u64 end = start + PAGE_CACHE_SIZE - 1;
2768         int ret = 1;
2769
2770         if (test_range_bit(tree, start, end,
2771                            EXTENT_IOBITS | EXTENT_ORDERED, 0))
2772                 ret = 0;
2773         else {
2774                 if ((mask & GFP_NOFS) == GFP_NOFS)
2775                         mask = GFP_NOFS;
2776                 clear_extent_bit(tree, start, end, EXTENT_UPTODATE,
2777                                  1, 1, mask);
2778         }
2779         return ret;
2780 }
2781
2782 /*
2783  * a helper for releasepage.  As long as there are no locked extents
2784  * in the range corresponding to the page, both state records and extent
2785  * map records are removed
2786  */
2787 int try_release_extent_mapping(struct extent_map_tree *map,
2788                                struct extent_io_tree *tree, struct page *page,
2789                                gfp_t mask)
2790 {
2791         struct extent_map *em;
2792         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2793         u64 end = start + PAGE_CACHE_SIZE - 1;
2794
2795         if ((mask & __GFP_WAIT) &&
2796             page->mapping->host->i_size > 16 * 1024 * 1024) {
2797                 u64 len;
2798                 while (start <= end) {
2799                         len = end - start + 1;
2800                         spin_lock(&map->lock);
2801                         em = lookup_extent_mapping(map, start, len);
2802                         if (!em || IS_ERR(em)) {
2803                                 spin_unlock(&map->lock);
2804                                 break;
2805                         }
2806                         if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
2807                             em->start != start) {
2808                                 spin_unlock(&map->lock);
2809                                 free_extent_map(em);
2810                                 break;
2811                         }
2812                         if (!test_range_bit(tree, em->start,
2813                                             extent_map_end(em) - 1,
2814                                             EXTENT_LOCKED | EXTENT_WRITEBACK |
2815                                             EXTENT_ORDERED,
2816                                             0)) {
2817                                 remove_extent_mapping(map, em);
2818                                 /* once for the rb tree */
2819                                 free_extent_map(em);
2820                         }
2821                         start = extent_map_end(em);
2822                         spin_unlock(&map->lock);
2823
2824                         /* once for us */
2825                         free_extent_map(em);
2826                 }
2827         }
2828         return try_release_extent_state(map, tree, page, mask);
2829 }
2830
2831 sector_t extent_bmap(struct address_space *mapping, sector_t iblock,
2832                 get_extent_t *get_extent)
2833 {
2834         struct inode *inode = mapping->host;
2835         u64 start = iblock << inode->i_blkbits;
2836         sector_t sector = 0;
2837         size_t blksize = (1 << inode->i_blkbits);
2838         struct extent_map *em;
2839
2840         lock_extent(&BTRFS_I(inode)->io_tree, start, start + blksize - 1,
2841                     GFP_NOFS);
2842         em = get_extent(inode, NULL, 0, start, blksize, 0);
2843         unlock_extent(&BTRFS_I(inode)->io_tree, start, start + blksize - 1,
2844                       GFP_NOFS);
2845         if (!em || IS_ERR(em))
2846                 return 0;
2847
2848         if (em->block_start > EXTENT_MAP_LAST_BYTE)
2849                 goto out;
2850
2851         sector = (em->block_start + start - em->start) >> inode->i_blkbits;
2852 out:
2853         free_extent_map(em);
2854         return sector;
2855 }
2856
2857 static inline struct page *extent_buffer_page(struct extent_buffer *eb,
2858                                               unsigned long i)
2859 {
2860         struct page *p;
2861         struct address_space *mapping;
2862
2863         if (i == 0)
2864                 return eb->first_page;
2865         i += eb->start >> PAGE_CACHE_SHIFT;
2866         mapping = eb->first_page->mapping;
2867         if (!mapping)
2868                 return NULL;
2869
2870         /*
2871          * extent_buffer_page is only called after pinning the page
2872          * by increasing the reference count.  So we know the page must
2873          * be in the radix tree.
2874          */
2875         rcu_read_lock();
2876         p = radix_tree_lookup(&mapping->page_tree, i);
2877         rcu_read_unlock();
2878
2879         return p;
2880 }
2881
2882 static inline unsigned long num_extent_pages(u64 start, u64 len)
2883 {
2884         return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
2885                 (start >> PAGE_CACHE_SHIFT);
2886 }
2887
2888 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
2889                                                    u64 start,
2890                                                    unsigned long len,
2891                                                    gfp_t mask)
2892 {
2893         struct extent_buffer *eb = NULL;
2894 #ifdef LEAK_DEBUG
2895         unsigned long flags;
2896 #endif
2897
2898         eb = kmem_cache_zalloc(extent_buffer_cache, mask);
2899         eb->start = start;
2900         eb->len = len;
2901         mutex_init(&eb->mutex);
2902 #ifdef LEAK_DEBUG
2903         spin_lock_irqsave(&leak_lock, flags);
2904         list_add(&eb->leak_list, &buffers);
2905         spin_unlock_irqrestore(&leak_lock, flags);
2906 #endif
2907         atomic_set(&eb->refs, 1);
2908
2909         return eb;
2910 }
2911
2912 static void __free_extent_buffer(struct extent_buffer *eb)
2913 {
2914 #ifdef LEAK_DEBUG
2915         unsigned long flags;
2916         spin_lock_irqsave(&leak_lock, flags);
2917         list_del(&eb->leak_list);
2918         spin_unlock_irqrestore(&leak_lock, flags);
2919 #endif
2920         kmem_cache_free(extent_buffer_cache, eb);
2921 }
2922
2923 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
2924                                           u64 start, unsigned long len,
2925                                           struct page *page0,
2926                                           gfp_t mask)
2927 {
2928         unsigned long num_pages = num_extent_pages(start, len);
2929         unsigned long i;
2930         unsigned long index = start >> PAGE_CACHE_SHIFT;
2931         struct extent_buffer *eb;
2932         struct extent_buffer *exists = NULL;
2933         struct page *p;
2934         struct address_space *mapping = tree->mapping;
2935         int uptodate = 1;
2936
2937         spin_lock(&tree->buffer_lock);
2938         eb = buffer_search(tree, start);
2939         if (eb) {
2940                 atomic_inc(&eb->refs);
2941                 spin_unlock(&tree->buffer_lock);
2942                 mark_page_accessed(eb->first_page);
2943                 return eb;
2944         }
2945         spin_unlock(&tree->buffer_lock);
2946
2947         eb = __alloc_extent_buffer(tree, start, len, mask);
2948         if (!eb)
2949                 return NULL;
2950
2951         if (page0) {
2952                 eb->first_page = page0;
2953                 i = 1;
2954                 index++;
2955                 page_cache_get(page0);
2956                 mark_page_accessed(page0);
2957                 set_page_extent_mapped(page0);
2958                 set_page_extent_head(page0, len);
2959                 uptodate = PageUptodate(page0);
2960         } else {
2961                 i = 0;
2962         }
2963         for (; i < num_pages; i++, index++) {
2964                 p = find_or_create_page(mapping, index, mask | __GFP_HIGHMEM);
2965                 if (!p) {
2966                         WARN_ON(1);
2967                         goto free_eb;
2968                 }
2969                 set_page_extent_mapped(p);
2970                 mark_page_accessed(p);
2971                 if (i == 0) {
2972                         eb->first_page = p;
2973                         set_page_extent_head(p, len);
2974                 } else {
2975                         set_page_private(p, EXTENT_PAGE_PRIVATE);
2976                 }
2977                 if (!PageUptodate(p))
2978                         uptodate = 0;
2979                 unlock_page(p);
2980         }
2981         if (uptodate)
2982                 eb->flags |= EXTENT_UPTODATE;
2983         eb->flags |= EXTENT_BUFFER_FILLED;
2984
2985         spin_lock(&tree->buffer_lock);
2986         exists = buffer_tree_insert(tree, start, &eb->rb_node);
2987         if (exists) {
2988                 /* add one reference for the caller */
2989                 atomic_inc(&exists->refs);
2990                 spin_unlock(&tree->buffer_lock);
2991                 goto free_eb;
2992         }
2993         spin_unlock(&tree->buffer_lock);
2994
2995         /* add one reference for the tree */
2996         atomic_inc(&eb->refs);
2997         return eb;
2998
2999 free_eb:
3000         if (!atomic_dec_and_test(&eb->refs))
3001                 return exists;
3002         for (index = 1; index < i; index++)
3003                 page_cache_release(extent_buffer_page(eb, index));
3004         page_cache_release(extent_buffer_page(eb, 0));
3005         __free_extent_buffer(eb);
3006         return exists;
3007 }
3008
3009 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
3010                                          u64 start, unsigned long len,
3011                                           gfp_t mask)
3012 {
3013         struct extent_buffer *eb;
3014
3015         spin_lock(&tree->buffer_lock);
3016         eb = buffer_search(tree, start);
3017         if (eb)
3018                 atomic_inc(&eb->refs);
3019         spin_unlock(&tree->buffer_lock);
3020
3021         if (eb)
3022                 mark_page_accessed(eb->first_page);
3023
3024         return eb;
3025 }
3026
3027 void free_extent_buffer(struct extent_buffer *eb)
3028 {
3029         if (!eb)
3030                 return;
3031
3032         if (!atomic_dec_and_test(&eb->refs))
3033                 return;
3034
3035         WARN_ON(1);
3036 }
3037
3038 int clear_extent_buffer_dirty(struct extent_io_tree *tree,
3039                               struct extent_buffer *eb)
3040 {
3041         int set;
3042         unsigned long i;
3043         unsigned long num_pages;
3044         struct page *page;
3045
3046         u64 start = eb->start;
3047         u64 end = start + eb->len - 1;
3048
3049         set = clear_extent_dirty(tree, start, end, GFP_NOFS);
3050         num_pages = num_extent_pages(eb->start, eb->len);
3051
3052         for (i = 0; i < num_pages; i++) {
3053                 page = extent_buffer_page(eb, i);
3054                 if (!set && !PageDirty(page))
3055                         continue;
3056
3057                 lock_page(page);
3058                 if (i == 0)
3059                         set_page_extent_head(page, eb->len);
3060                 else
3061                         set_page_private(page, EXTENT_PAGE_PRIVATE);
3062
3063                 /*
3064                  * if we're on the last page or the first page and the
3065                  * block isn't aligned on a page boundary, do extra checks
3066                  * to make sure we don't clean page that is partially dirty
3067                  */
3068                 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
3069                     ((i == num_pages - 1) &&
3070                      ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
3071                         start = (u64)page->index << PAGE_CACHE_SHIFT;
3072                         end  = start + PAGE_CACHE_SIZE - 1;
3073                         if (test_range_bit(tree, start, end,
3074                                            EXTENT_DIRTY, 0)) {
3075                                 unlock_page(page);
3076                                 continue;
3077                         }
3078                 }
3079                 clear_page_dirty_for_io(page);
3080                 spin_lock_irq(&page->mapping->tree_lock);
3081                 if (!PageDirty(page)) {
3082                         radix_tree_tag_clear(&page->mapping->page_tree,
3083                                                 page_index(page),
3084                                                 PAGECACHE_TAG_DIRTY);
3085                 }
3086                 spin_unlock_irq(&page->mapping->tree_lock);
3087                 unlock_page(page);
3088         }
3089         return 0;
3090 }
3091
3092 int wait_on_extent_buffer_writeback(struct extent_io_tree *tree,
3093                                     struct extent_buffer *eb)
3094 {
3095         return wait_on_extent_writeback(tree, eb->start,
3096                                         eb->start + eb->len - 1);
3097 }
3098
3099 int set_extent_buffer_dirty(struct extent_io_tree *tree,
3100                              struct extent_buffer *eb)
3101 {
3102         unsigned long i;
3103         unsigned long num_pages;
3104
3105         num_pages = num_extent_pages(eb->start, eb->len);
3106         for (i = 0; i < num_pages; i++) {
3107                 struct page *page = extent_buffer_page(eb, i);
3108                 /* writepage may need to do something special for the
3109                  * first page, we have to make sure page->private is
3110                  * properly set.  releasepage may drop page->private
3111                  * on us if the page isn't already dirty.
3112                  */
3113                 lock_page(page);
3114                 if (i == 0) {
3115                         set_page_extent_head(page, eb->len);
3116                 } else if (PagePrivate(page) &&
3117                            page->private != EXTENT_PAGE_PRIVATE) {
3118                         set_page_extent_mapped(page);
3119                 }
3120                 __set_page_dirty_nobuffers(extent_buffer_page(eb, i));
3121                 set_extent_dirty(tree, page_offset(page),
3122                                  page_offset(page) + PAGE_CACHE_SIZE - 1,
3123                                  GFP_NOFS);
3124                 unlock_page(page);
3125         }
3126         return 0;
3127 }
3128
3129 int clear_extent_buffer_uptodate(struct extent_io_tree *tree,
3130                                 struct extent_buffer *eb)
3131 {
3132         unsigned long i;
3133         struct page *page;
3134         unsigned long num_pages;
3135
3136         num_pages = num_extent_pages(eb->start, eb->len);
3137         eb->flags &= ~EXTENT_UPTODATE;
3138
3139         clear_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
3140                               GFP_NOFS);
3141         for (i = 0; i < num_pages; i++) {
3142                 page = extent_buffer_page(eb, i);
3143                 if (page)
3144                         ClearPageUptodate(page);
3145         }
3146         return 0;
3147 }
3148
3149 int set_extent_buffer_uptodate(struct extent_io_tree *tree,
3150                                 struct extent_buffer *eb)
3151 {
3152         unsigned long i;
3153         struct page *page;
3154         unsigned long num_pages;
3155
3156         num_pages = num_extent_pages(eb->start, eb->len);
3157
3158         set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
3159                             GFP_NOFS);
3160         for (i = 0; i < num_pages; i++) {
3161                 page = extent_buffer_page(eb, i);
3162                 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
3163                     ((i == num_pages - 1) &&
3164                      ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
3165                         check_page_uptodate(tree, page);
3166                         continue;
3167                 }
3168                 SetPageUptodate(page);
3169         }
3170         return 0;
3171 }
3172
3173 int extent_range_uptodate(struct extent_io_tree *tree,
3174                           u64 start, u64 end)
3175 {
3176         struct page *page;
3177         int ret;
3178         int pg_uptodate = 1;
3179         int uptodate;
3180         unsigned long index;
3181
3182         ret = test_range_bit(tree, start, end, EXTENT_UPTODATE, 1);
3183         if (ret)
3184                 return 1;
3185         while (start <= end) {
3186                 index = start >> PAGE_CACHE_SHIFT;
3187                 page = find_get_page(tree->mapping, index);
3188                 uptodate = PageUptodate(page);
3189                 page_cache_release(page);
3190                 if (!uptodate) {
3191                         pg_uptodate = 0;
3192                         break;
3193                 }
3194                 start += PAGE_CACHE_SIZE;
3195         }
3196         return pg_uptodate;
3197 }
3198
3199 int extent_buffer_uptodate(struct extent_io_tree *tree,
3200                            struct extent_buffer *eb)
3201 {
3202         int ret = 0;
3203         unsigned long num_pages;
3204         unsigned long i;
3205         struct page *page;
3206         int pg_uptodate = 1;
3207
3208         if (eb->flags & EXTENT_UPTODATE)
3209                 return 1;
3210
3211         ret = test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3212                            EXTENT_UPTODATE, 1);
3213         if (ret)
3214                 return ret;
3215
3216         num_pages = num_extent_pages(eb->start, eb->len);
3217         for (i = 0; i < num_pages; i++) {
3218                 page = extent_buffer_page(eb, i);
3219                 if (!PageUptodate(page)) {
3220                         pg_uptodate = 0;
3221                         break;
3222                 }
3223         }
3224         return pg_uptodate;
3225 }
3226
3227 int read_extent_buffer_pages(struct extent_io_tree *tree,
3228                              struct extent_buffer *eb,
3229                              u64 start, int wait,
3230                              get_extent_t *get_extent, int mirror_num)
3231 {
3232         unsigned long i;
3233         unsigned long start_i;
3234         struct page *page;
3235         int err;
3236         int ret = 0;
3237         int locked_pages = 0;
3238         int all_uptodate = 1;
3239         int inc_all_pages = 0;
3240         unsigned long num_pages;
3241         struct bio *bio = NULL;
3242         unsigned long bio_flags = 0;
3243
3244         if (eb->flags & EXTENT_UPTODATE)
3245                 return 0;
3246
3247         if (test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3248                            EXTENT_UPTODATE, 1)) {
3249                 return 0;
3250         }
3251
3252         if (start) {
3253                 WARN_ON(start < eb->start);
3254                 start_i = (start >> PAGE_CACHE_SHIFT) -
3255                         (eb->start >> PAGE_CACHE_SHIFT);
3256         } else {
3257                 start_i = 0;
3258         }
3259
3260         num_pages = num_extent_pages(eb->start, eb->len);
3261         for (i = start_i; i < num_pages; i++) {
3262                 page = extent_buffer_page(eb, i);
3263                 if (!wait) {
3264                         if (!trylock_page(page))
3265                                 goto unlock_exit;
3266                 } else {
3267                         lock_page(page);
3268                 }
3269                 locked_pages++;
3270                 if (!PageUptodate(page))
3271                         all_uptodate = 0;
3272         }
3273         if (all_uptodate) {
3274                 if (start_i == 0)
3275                         eb->flags |= EXTENT_UPTODATE;
3276                 goto unlock_exit;
3277         }
3278
3279         for (i = start_i; i < num_pages; i++) {
3280                 page = extent_buffer_page(eb, i);
3281                 if (inc_all_pages)
3282                         page_cache_get(page);
3283                 if (!PageUptodate(page)) {
3284                         if (start_i == 0)
3285                                 inc_all_pages = 1;
3286                         ClearPageError(page);
3287                         err = __extent_read_full_page(tree, page,
3288                                                       get_extent, &bio,
3289                                                       mirror_num, &bio_flags);
3290                         if (err)
3291                                 ret = err;
3292                 } else {
3293                         unlock_page(page);
3294                 }
3295         }
3296
3297         if (bio)
3298                 submit_one_bio(READ, bio, mirror_num, bio_flags);
3299
3300         if (ret || !wait)
3301                 return ret;
3302
3303         for (i = start_i; i < num_pages; i++) {
3304                 page = extent_buffer_page(eb, i);
3305                 wait_on_page_locked(page);
3306                 if (!PageUptodate(page))
3307                         ret = -EIO;
3308         }
3309
3310         if (!ret)
3311                 eb->flags |= EXTENT_UPTODATE;
3312         return ret;
3313
3314 unlock_exit:
3315         i = start_i;
3316         while (locked_pages > 0) {
3317                 page = extent_buffer_page(eb, i);
3318                 i++;
3319                 unlock_page(page);
3320                 locked_pages--;
3321         }
3322         return ret;
3323 }
3324
3325 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
3326                         unsigned long start,
3327                         unsigned long len)
3328 {
3329         size_t cur;
3330         size_t offset;
3331         struct page *page;
3332         char *kaddr;
3333         char *dst = (char *)dstv;
3334         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3335         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3336
3337         WARN_ON(start > eb->len);
3338         WARN_ON(start + len > eb->start + eb->len);
3339
3340         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3341
3342         while (len > 0) {
3343                 page = extent_buffer_page(eb, i);
3344
3345                 cur = min(len, (PAGE_CACHE_SIZE - offset));
3346                 kaddr = kmap_atomic(page, KM_USER1);
3347                 memcpy(dst, kaddr + offset, cur);
3348                 kunmap_atomic(kaddr, KM_USER1);
3349
3350                 dst += cur;
3351                 len -= cur;
3352                 offset = 0;
3353                 i++;
3354         }
3355 }
3356
3357 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
3358                                unsigned long min_len, char **token, char **map,
3359                                unsigned long *map_start,
3360                                unsigned long *map_len, int km)
3361 {
3362         size_t offset = start & (PAGE_CACHE_SIZE - 1);
3363         char *kaddr;
3364         struct page *p;
3365         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3366         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3367         unsigned long end_i = (start_offset + start + min_len - 1) >>
3368                 PAGE_CACHE_SHIFT;
3369
3370         if (i != end_i)
3371                 return -EINVAL;
3372
3373         if (i == 0) {
3374                 offset = start_offset;
3375                 *map_start = 0;
3376         } else {
3377                 offset = 0;
3378                 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
3379         }
3380
3381         if (start + min_len > eb->len) {
3382                 printk(KERN_ERR "btrfs bad mapping eb start %llu len %lu, "
3383                        "wanted %lu %lu\n", (unsigned long long)eb->start,
3384                        eb->len, start, min_len);
3385                 WARN_ON(1);
3386         }
3387
3388         p = extent_buffer_page(eb, i);
3389         kaddr = kmap_atomic(p, km);
3390         *token = kaddr;
3391         *map = kaddr + offset;
3392         *map_len = PAGE_CACHE_SIZE - offset;
3393         return 0;
3394 }
3395
3396 int map_extent_buffer(struct extent_buffer *eb, unsigned long start,
3397                       unsigned long min_len,
3398                       char **token, char **map,
3399                       unsigned long *map_start,
3400                       unsigned long *map_len, int km)
3401 {
3402         int err;
3403         int save = 0;
3404         if (eb->map_token) {
3405                 unmap_extent_buffer(eb, eb->map_token, km);
3406                 eb->map_token = NULL;
3407                 save = 1;
3408                 WARN_ON(!mutex_is_locked(&eb->mutex));
3409         }
3410         err = map_private_extent_buffer(eb, start, min_len, token, map,
3411                                        map_start, map_len, km);
3412         if (!err && save) {
3413                 eb->map_token = *token;
3414                 eb->kaddr = *map;
3415                 eb->map_start = *map_start;
3416                 eb->map_len = *map_len;
3417         }
3418         return err;
3419 }
3420
3421 void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km)
3422 {
3423         kunmap_atomic(token, km);
3424 }
3425
3426 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
3427                           unsigned long start,
3428                           unsigned long len)
3429 {
3430         size_t cur;
3431         size_t offset;
3432         struct page *page;
3433         char *kaddr;
3434         char *ptr = (char *)ptrv;
3435         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3436         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3437         int ret = 0;
3438
3439         WARN_ON(start > eb->len);
3440         WARN_ON(start + len > eb->start + eb->len);
3441
3442         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3443
3444         while (len > 0) {
3445                 page = extent_buffer_page(eb, i);
3446
3447                 cur = min(len, (PAGE_CACHE_SIZE - offset));
3448
3449                 kaddr = kmap_atomic(page, KM_USER0);
3450                 ret = memcmp(ptr, kaddr + offset, cur);
3451                 kunmap_atomic(kaddr, KM_USER0);
3452                 if (ret)
3453                         break;
3454
3455                 ptr += cur;
3456                 len -= cur;
3457                 offset = 0;
3458                 i++;
3459         }
3460         return ret;
3461 }
3462
3463 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
3464                          unsigned long start, unsigned long len)
3465 {
3466         size_t cur;
3467         size_t offset;
3468         struct page *page;
3469         char *kaddr;
3470         char *src = (char *)srcv;
3471         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3472         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3473
3474         WARN_ON(start > eb->len);
3475         WARN_ON(start + len > eb->start + eb->len);
3476
3477         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3478
3479         while (len > 0) {
3480                 page = extent_buffer_page(eb, i);
3481                 WARN_ON(!PageUptodate(page));
3482
3483                 cur = min(len, PAGE_CACHE_SIZE - offset);
3484                 kaddr = kmap_atomic(page, KM_USER1);
3485                 memcpy(kaddr + offset, src, cur);
3486                 kunmap_atomic(kaddr, KM_USER1);
3487
3488                 src += cur;
3489                 len -= cur;
3490                 offset = 0;
3491                 i++;
3492         }
3493 }
3494
3495 void memset_extent_buffer(struct extent_buffer *eb, char c,
3496                           unsigned long start, unsigned long len)
3497 {
3498         size_t cur;
3499         size_t offset;
3500         struct page *page;
3501         char *kaddr;
3502         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3503         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3504
3505         WARN_ON(start > eb->len);
3506         WARN_ON(start + len > eb->start + eb->len);
3507
3508         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3509
3510         while (len > 0) {
3511                 page = extent_buffer_page(eb, i);
3512                 WARN_ON(!PageUptodate(page));
3513
3514                 cur = min(len, PAGE_CACHE_SIZE - offset);
3515                 kaddr = kmap_atomic(page, KM_USER0);
3516                 memset(kaddr + offset, c, cur);
3517                 kunmap_atomic(kaddr, KM_USER0);
3518
3519                 len -= cur;
3520                 offset = 0;
3521                 i++;
3522         }
3523 }
3524
3525 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
3526                         unsigned long dst_offset, unsigned long src_offset,
3527                         unsigned long len)
3528 {
3529         u64 dst_len = dst->len;
3530         size_t cur;
3531         size_t offset;
3532         struct page *page;
3533         char *kaddr;
3534         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3535         unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3536
3537         WARN_ON(src->len != dst_len);
3538
3539         offset = (start_offset + dst_offset) &
3540                 ((unsigned long)PAGE_CACHE_SIZE - 1);
3541
3542         while (len > 0) {
3543                 page = extent_buffer_page(dst, i);
3544                 WARN_ON(!PageUptodate(page));
3545
3546                 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
3547
3548                 kaddr = kmap_atomic(page, KM_USER0);
3549                 read_extent_buffer(src, kaddr + offset, src_offset, cur);
3550                 kunmap_atomic(kaddr, KM_USER0);
3551
3552                 src_offset += cur;
3553                 len -= cur;
3554                 offset = 0;
3555                 i++;
3556         }
3557 }
3558
3559 static void move_pages(struct page *dst_page, struct page *src_page,
3560                        unsigned long dst_off, unsigned long src_off,
3561                        unsigned long len)
3562 {
3563         char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3564         if (dst_page == src_page) {
3565                 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
3566         } else {
3567                 char *src_kaddr = kmap_atomic(src_page, KM_USER1);
3568                 char *p = dst_kaddr + dst_off + len;
3569                 char *s = src_kaddr + src_off + len;
3570
3571                 while (len--)
3572                         *--p = *--s;
3573
3574                 kunmap_atomic(src_kaddr, KM_USER1);
3575         }
3576         kunmap_atomic(dst_kaddr, KM_USER0);
3577 }
3578
3579 static void copy_pages(struct page *dst_page, struct page *src_page,
3580                        unsigned long dst_off, unsigned long src_off,
3581                        unsigned long len)
3582 {
3583         char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3584         char *src_kaddr;
3585
3586         if (dst_page != src_page)
3587                 src_kaddr = kmap_atomic(src_page, KM_USER1);
3588         else
3589                 src_kaddr = dst_kaddr;
3590
3591         memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
3592         kunmap_atomic(dst_kaddr, KM_USER0);
3593         if (dst_page != src_page)
3594                 kunmap_atomic(src_kaddr, KM_USER1);
3595 }
3596
3597 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3598                            unsigned long src_offset, unsigned long len)
3599 {
3600         size_t cur;
3601         size_t dst_off_in_page;
3602         size_t src_off_in_page;
3603         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3604         unsigned long dst_i;
3605         unsigned long src_i;
3606
3607         if (src_offset + len > dst->len) {
3608                 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
3609                        "len %lu dst len %lu\n", src_offset, len, dst->len);
3610                 BUG_ON(1);
3611         }
3612         if (dst_offset + len > dst->len) {
3613                 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
3614                        "len %lu dst len %lu\n", dst_offset, len, dst->len);
3615                 BUG_ON(1);
3616         }
3617
3618         while (len > 0) {
3619                 dst_off_in_page = (start_offset + dst_offset) &
3620                         ((unsigned long)PAGE_CACHE_SIZE - 1);
3621                 src_off_in_page = (start_offset + src_offset) &
3622                         ((unsigned long)PAGE_CACHE_SIZE - 1);
3623
3624                 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3625                 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
3626
3627                 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
3628                                                src_off_in_page));
3629                 cur = min_t(unsigned long, cur,
3630                         (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
3631
3632                 copy_pages(extent_buffer_page(dst, dst_i),
3633                            extent_buffer_page(dst, src_i),
3634                            dst_off_in_page, src_off_in_page, cur);
3635
3636                 src_offset += cur;
3637                 dst_offset += cur;
3638                 len -= cur;
3639         }
3640 }
3641
3642 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3643                            unsigned long src_offset, unsigned long len)
3644 {
3645         size_t cur;
3646         size_t dst_off_in_page;
3647         size_t src_off_in_page;
3648         unsigned long dst_end = dst_offset + len - 1;
3649         unsigned long src_end = src_offset + len - 1;
3650         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3651         unsigned long dst_i;
3652         unsigned long src_i;
3653
3654         if (src_offset + len > dst->len) {
3655                 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
3656                        "len %lu len %lu\n", src_offset, len, dst->len);
3657                 BUG_ON(1);
3658         }
3659         if (dst_offset + len > dst->len) {
3660                 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
3661                        "len %lu len %lu\n", dst_offset, len, dst->len);
3662                 BUG_ON(1);
3663         }
3664         if (dst_offset < src_offset) {
3665                 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
3666                 return;
3667         }
3668         while (len > 0) {
3669                 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
3670                 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
3671
3672                 dst_off_in_page = (start_offset + dst_end) &
3673                         ((unsigned long)PAGE_CACHE_SIZE - 1);
3674                 src_off_in_page = (start_offset + src_end) &
3675                         ((unsigned long)PAGE_CACHE_SIZE - 1);
3676
3677                 cur = min_t(unsigned long, len, src_off_in_page + 1);
3678                 cur = min(cur, dst_off_in_page + 1);
3679                 move_pages(extent_buffer_page(dst, dst_i),
3680                            extent_buffer_page(dst, src_i),
3681                            dst_off_in_page - cur + 1,
3682                            src_off_in_page - cur + 1, cur);
3683
3684                 dst_end -= cur;
3685                 src_end -= cur;
3686                 len -= cur;
3687         }
3688 }
3689
3690 int try_release_extent_buffer(struct extent_io_tree *tree, struct page *page)
3691 {
3692         u64 start = page_offset(page);
3693         struct extent_buffer *eb;
3694         int ret = 1;
3695         unsigned long i;
3696         unsigned long num_pages;
3697
3698         spin_lock(&tree->buffer_lock);
3699         eb = buffer_search(tree, start);
3700         if (!eb)
3701                 goto out;
3702
3703         if (atomic_read(&eb->refs) > 1) {
3704                 ret = 0;
3705                 goto out;
3706         }
3707         /* at this point we can safely release the extent buffer */
3708         num_pages = num_extent_pages(eb->start, eb->len);
3709         for (i = 0; i < num_pages; i++)
3710                 page_cache_release(extent_buffer_page(eb, i));
3711         rb_erase(&eb->rb_node, &tree->buffer);
3712         __free_extent_buffer(eb);
3713 out:
3714         spin_unlock(&tree->buffer_lock);
3715         return ret;
3716 }