1 #include <linux/bitops.h>
2 #include <linux/slab.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/version.h>
13 #include <linux/writeback.h>
14 #include "extent_map.h"
16 /* temporary define until extent_map moves out of btrfs */
17 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
18 unsigned long extra_flags,
19 void (*ctor)(void *, struct kmem_cache *,
22 static struct kmem_cache *extent_map_cache;
23 static struct kmem_cache *extent_state_cache;
24 static struct kmem_cache *extent_buffer_cache;
26 static LIST_HEAD(buffers);
27 static LIST_HEAD(states);
29 static spinlock_t state_lock = SPIN_LOCK_UNLOCKED;
30 #define BUFFER_LRU_MAX 64
36 struct rb_node rb_node;
39 struct extent_page_data {
41 struct extent_map_tree *tree;
42 get_extent_t *get_extent;
45 void __init extent_map_init(void)
47 extent_map_cache = btrfs_cache_create("extent_map",
48 sizeof(struct extent_map), 0,
50 extent_state_cache = btrfs_cache_create("extent_state",
51 sizeof(struct extent_state), 0,
53 extent_buffer_cache = btrfs_cache_create("extent_buffers",
54 sizeof(struct extent_buffer), 0,
58 void __exit extent_map_exit(void)
60 struct extent_state *state;
62 while (!list_empty(&states)) {
63 state = list_entry(states.next, struct extent_state, list);
64 printk("state leak: start %Lu end %Lu state %lu in tree %d refs %d\n", state->start, state->end, state->state, state->in_tree, atomic_read(&state->refs));
65 list_del(&state->list);
66 kmem_cache_free(extent_state_cache, state);
71 kmem_cache_destroy(extent_map_cache);
72 if (extent_state_cache)
73 kmem_cache_destroy(extent_state_cache);
74 if (extent_buffer_cache)
75 kmem_cache_destroy(extent_buffer_cache);
78 void extent_map_tree_init(struct extent_map_tree *tree,
79 struct address_space *mapping, gfp_t mask)
81 tree->map.rb_node = NULL;
82 tree->state.rb_node = NULL;
84 rwlock_init(&tree->lock);
85 spin_lock_init(&tree->lru_lock);
86 tree->mapping = mapping;
87 INIT_LIST_HEAD(&tree->buffer_lru);
90 EXPORT_SYMBOL(extent_map_tree_init);
92 void extent_map_tree_empty_lru(struct extent_map_tree *tree)
94 struct extent_buffer *eb;
95 while(!list_empty(&tree->buffer_lru)) {
96 eb = list_entry(tree->buffer_lru.next, struct extent_buffer,
99 free_extent_buffer(eb);
102 EXPORT_SYMBOL(extent_map_tree_empty_lru);
104 struct extent_map *alloc_extent_map(gfp_t mask)
106 struct extent_map *em;
107 em = kmem_cache_alloc(extent_map_cache, mask);
108 if (!em || IS_ERR(em))
111 atomic_set(&em->refs, 1);
114 EXPORT_SYMBOL(alloc_extent_map);
116 void free_extent_map(struct extent_map *em)
120 if (atomic_dec_and_test(&em->refs)) {
121 WARN_ON(em->in_tree);
122 kmem_cache_free(extent_map_cache, em);
125 EXPORT_SYMBOL(free_extent_map);
128 struct extent_state *alloc_extent_state(gfp_t mask)
130 struct extent_state *state;
133 state = kmem_cache_alloc(extent_state_cache, mask);
134 if (!state || IS_ERR(state))
140 spin_lock_irqsave(&state_lock, flags);
141 list_add(&state->list, &states);
142 spin_unlock_irqrestore(&state_lock, flags);
144 atomic_set(&state->refs, 1);
145 init_waitqueue_head(&state->wq);
148 EXPORT_SYMBOL(alloc_extent_state);
150 void free_extent_state(struct extent_state *state)
155 if (atomic_dec_and_test(&state->refs)) {
156 WARN_ON(state->in_tree);
157 spin_lock_irqsave(&state_lock, flags);
158 list_del(&state->list);
159 spin_unlock_irqrestore(&state_lock, flags);
160 kmem_cache_free(extent_state_cache, state);
163 EXPORT_SYMBOL(free_extent_state);
165 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
166 struct rb_node *node)
168 struct rb_node ** p = &root->rb_node;
169 struct rb_node * parent = NULL;
170 struct tree_entry *entry;
174 entry = rb_entry(parent, struct tree_entry, rb_node);
176 if (offset < entry->start)
178 else if (offset > entry->end)
184 entry = rb_entry(node, struct tree_entry, rb_node);
186 rb_link_node(node, parent, p);
187 rb_insert_color(node, root);
191 static struct rb_node *__tree_search(struct rb_root *root, u64 offset,
192 struct rb_node **prev_ret)
194 struct rb_node * n = root->rb_node;
195 struct rb_node *prev = NULL;
196 struct tree_entry *entry;
197 struct tree_entry *prev_entry = NULL;
200 entry = rb_entry(n, struct tree_entry, rb_node);
204 if (offset < entry->start)
206 else if (offset > entry->end)
213 while(prev && offset > prev_entry->end) {
214 prev = rb_next(prev);
215 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
221 static inline struct rb_node *tree_search(struct rb_root *root, u64 offset)
223 struct rb_node *prev;
225 ret = __tree_search(root, offset, &prev);
231 static int tree_delete(struct rb_root *root, u64 offset)
233 struct rb_node *node;
234 struct tree_entry *entry;
236 node = __tree_search(root, offset, NULL);
239 entry = rb_entry(node, struct tree_entry, rb_node);
241 rb_erase(node, root);
246 * add_extent_mapping tries a simple backward merge with existing
247 * mappings. The extent_map struct passed in will be inserted into
248 * the tree directly (no copies made, just a reference taken).
250 int add_extent_mapping(struct extent_map_tree *tree,
251 struct extent_map *em)
254 struct extent_map *prev = NULL;
257 write_lock_irq(&tree->lock);
258 rb = tree_insert(&tree->map, em->end, &em->rb_node);
260 prev = rb_entry(rb, struct extent_map, rb_node);
261 printk("found extent map %Lu %Lu on insert of %Lu %Lu\n", prev->start, prev->end, em->start, em->end);
265 atomic_inc(&em->refs);
266 if (em->start != 0) {
267 rb = rb_prev(&em->rb_node);
269 prev = rb_entry(rb, struct extent_map, rb_node);
270 if (prev && prev->end + 1 == em->start &&
271 ((em->block_start == EXTENT_MAP_HOLE &&
272 prev->block_start == EXTENT_MAP_HOLE) ||
273 (em->block_start == EXTENT_MAP_INLINE &&
274 prev->block_start == EXTENT_MAP_INLINE) ||
275 (em->block_start == EXTENT_MAP_DELALLOC &&
276 prev->block_start == EXTENT_MAP_DELALLOC) ||
277 (em->block_start < EXTENT_MAP_DELALLOC - 1 &&
278 em->block_start == prev->block_end + 1))) {
279 em->start = prev->start;
280 em->block_start = prev->block_start;
281 rb_erase(&prev->rb_node, &tree->map);
283 free_extent_map(prev);
287 write_unlock_irq(&tree->lock);
290 EXPORT_SYMBOL(add_extent_mapping);
293 * lookup_extent_mapping returns the first extent_map struct in the
294 * tree that intersects the [start, end] (inclusive) range. There may
295 * be additional objects in the tree that intersect, so check the object
296 * returned carefully to make sure you don't need additional lookups.
298 struct extent_map *lookup_extent_mapping(struct extent_map_tree *tree,
301 struct extent_map *em;
302 struct rb_node *rb_node;
304 read_lock_irq(&tree->lock);
305 rb_node = tree_search(&tree->map, start);
310 if (IS_ERR(rb_node)) {
311 em = ERR_PTR(PTR_ERR(rb_node));
314 em = rb_entry(rb_node, struct extent_map, rb_node);
315 if (em->end < start || em->start > end) {
319 atomic_inc(&em->refs);
321 read_unlock_irq(&tree->lock);
324 EXPORT_SYMBOL(lookup_extent_mapping);
327 * removes an extent_map struct from the tree. No reference counts are
328 * dropped, and no checks are done to see if the range is in use
330 int remove_extent_mapping(struct extent_map_tree *tree, struct extent_map *em)
334 write_lock_irq(&tree->lock);
335 ret = tree_delete(&tree->map, em->end);
336 write_unlock_irq(&tree->lock);
339 EXPORT_SYMBOL(remove_extent_mapping);
342 * utility function to look for merge candidates inside a given range.
343 * Any extents with matching state are merged together into a single
344 * extent in the tree. Extents with EXTENT_IO in their state field
345 * are not merged because the end_io handlers need to be able to do
346 * operations on them without sleeping (or doing allocations/splits).
348 * This should be called with the tree lock held.
350 static int merge_state(struct extent_map_tree *tree,
351 struct extent_state *state)
353 struct extent_state *other;
354 struct rb_node *other_node;
356 if (state->state & EXTENT_IOBITS)
359 other_node = rb_prev(&state->rb_node);
361 other = rb_entry(other_node, struct extent_state, rb_node);
362 if (other->end == state->start - 1 &&
363 other->state == state->state) {
364 state->start = other->start;
366 rb_erase(&other->rb_node, &tree->state);
367 free_extent_state(other);
370 other_node = rb_next(&state->rb_node);
372 other = rb_entry(other_node, struct extent_state, rb_node);
373 if (other->start == state->end + 1 &&
374 other->state == state->state) {
375 other->start = state->start;
377 rb_erase(&state->rb_node, &tree->state);
378 free_extent_state(state);
385 * insert an extent_state struct into the tree. 'bits' are set on the
386 * struct before it is inserted.
388 * This may return -EEXIST if the extent is already there, in which case the
389 * state struct is freed.
391 * The tree lock is not taken internally. This is a utility function and
392 * probably isn't what you want to call (see set/clear_extent_bit).
394 static int insert_state(struct extent_map_tree *tree,
395 struct extent_state *state, u64 start, u64 end,
398 struct rb_node *node;
401 printk("end < start %Lu %Lu\n", end, start);
404 state->state |= bits;
405 state->start = start;
407 node = tree_insert(&tree->state, end, &state->rb_node);
409 struct extent_state *found;
410 found = rb_entry(node, struct extent_state, rb_node);
411 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, start, end);
412 free_extent_state(state);
415 merge_state(tree, state);
420 * split a given extent state struct in two, inserting the preallocated
421 * struct 'prealloc' as the newly created second half. 'split' indicates an
422 * offset inside 'orig' where it should be split.
425 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
426 * are two extent state structs in the tree:
427 * prealloc: [orig->start, split - 1]
428 * orig: [ split, orig->end ]
430 * The tree locks are not taken by this function. They need to be held
433 static int split_state(struct extent_map_tree *tree, struct extent_state *orig,
434 struct extent_state *prealloc, u64 split)
436 struct rb_node *node;
437 prealloc->start = orig->start;
438 prealloc->end = split - 1;
439 prealloc->state = orig->state;
442 node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
444 struct extent_state *found;
445 found = rb_entry(node, struct extent_state, rb_node);
446 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, prealloc->start, prealloc->end);
447 free_extent_state(prealloc);
454 * utility function to clear some bits in an extent state struct.
455 * it will optionally wake up any one waiting on this state (wake == 1), or
456 * forcibly remove the state from the tree (delete == 1).
458 * If no bits are set on the state struct after clearing things, the
459 * struct is freed and removed from the tree
461 static int clear_state_bit(struct extent_map_tree *tree,
462 struct extent_state *state, int bits, int wake,
465 int ret = state->state & bits;
466 state->state &= ~bits;
469 if (delete || state->state == 0) {
470 if (state->in_tree) {
471 rb_erase(&state->rb_node, &tree->state);
473 free_extent_state(state);
478 merge_state(tree, state);
484 * clear some bits on a range in the tree. This may require splitting
485 * or inserting elements in the tree, so the gfp mask is used to
486 * indicate which allocations or sleeping are allowed.
488 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
489 * the given range from the tree regardless of state (ie for truncate).
491 * the range [start, end] is inclusive.
493 * This takes the tree lock, and returns < 0 on error, > 0 if any of the
494 * bits were already set, or zero if none of the bits were already set.
496 int clear_extent_bit(struct extent_map_tree *tree, u64 start, u64 end,
497 int bits, int wake, int delete, gfp_t mask)
499 struct extent_state *state;
500 struct extent_state *prealloc = NULL;
501 struct rb_node *node;
507 if (!prealloc && (mask & __GFP_WAIT)) {
508 prealloc = alloc_extent_state(mask);
513 write_lock_irqsave(&tree->lock, flags);
515 * this search will find the extents that end after
518 node = tree_search(&tree->state, start);
521 state = rb_entry(node, struct extent_state, rb_node);
522 if (state->start > end)
524 WARN_ON(state->end < start);
527 * | ---- desired range ---- |
529 * | ------------- state -------------- |
531 * We need to split the extent we found, and may flip
532 * bits on second half.
534 * If the extent we found extends past our range, we
535 * just split and search again. It'll get split again
536 * the next time though.
538 * If the extent we found is inside our range, we clear
539 * the desired bit on it.
542 if (state->start < start) {
543 err = split_state(tree, state, prealloc, start);
544 BUG_ON(err == -EEXIST);
548 if (state->end <= end) {
549 start = state->end + 1;
550 set |= clear_state_bit(tree, state, bits,
553 start = state->start;
558 * | ---- desired range ---- |
560 * We need to split the extent, and clear the bit
563 if (state->start <= end && state->end > end) {
564 err = split_state(tree, state, prealloc, end + 1);
565 BUG_ON(err == -EEXIST);
569 set |= clear_state_bit(tree, prealloc, bits,
575 start = state->end + 1;
576 set |= clear_state_bit(tree, state, bits, wake, delete);
580 write_unlock_irqrestore(&tree->lock, flags);
582 free_extent_state(prealloc);
589 write_unlock_irqrestore(&tree->lock, flags);
590 if (mask & __GFP_WAIT)
594 EXPORT_SYMBOL(clear_extent_bit);
596 static int wait_on_state(struct extent_map_tree *tree,
597 struct extent_state *state)
600 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
601 read_unlock_irq(&tree->lock);
603 read_lock_irq(&tree->lock);
604 finish_wait(&state->wq, &wait);
609 * waits for one or more bits to clear on a range in the state tree.
610 * The range [start, end] is inclusive.
611 * The tree lock is taken by this function
613 int wait_extent_bit(struct extent_map_tree *tree, u64 start, u64 end, int bits)
615 struct extent_state *state;
616 struct rb_node *node;
618 read_lock_irq(&tree->lock);
622 * this search will find all the extents that end after
625 node = tree_search(&tree->state, start);
629 state = rb_entry(node, struct extent_state, rb_node);
631 if (state->start > end)
634 if (state->state & bits) {
635 start = state->start;
636 atomic_inc(&state->refs);
637 wait_on_state(tree, state);
638 free_extent_state(state);
641 start = state->end + 1;
646 if (need_resched()) {
647 read_unlock_irq(&tree->lock);
649 read_lock_irq(&tree->lock);
653 read_unlock_irq(&tree->lock);
656 EXPORT_SYMBOL(wait_extent_bit);
659 * set some bits on a range in the tree. This may require allocations
660 * or sleeping, so the gfp mask is used to indicate what is allowed.
662 * If 'exclusive' == 1, this will fail with -EEXIST if some part of the
663 * range already has the desired bits set. The start of the existing
664 * range is returned in failed_start in this case.
666 * [start, end] is inclusive
667 * This takes the tree lock.
669 int set_extent_bit(struct extent_map_tree *tree, u64 start, u64 end, int bits,
670 int exclusive, u64 *failed_start, gfp_t mask)
672 struct extent_state *state;
673 struct extent_state *prealloc = NULL;
674 struct rb_node *node;
681 if (!prealloc && (mask & __GFP_WAIT)) {
682 prealloc = alloc_extent_state(mask);
687 write_lock_irqsave(&tree->lock, flags);
689 * this search will find all the extents that end after
692 node = tree_search(&tree->state, start);
694 err = insert_state(tree, prealloc, start, end, bits);
696 BUG_ON(err == -EEXIST);
700 state = rb_entry(node, struct extent_state, rb_node);
701 last_start = state->start;
702 last_end = state->end;
705 * | ---- desired range ---- |
708 * Just lock what we found and keep going
710 if (state->start == start && state->end <= end) {
711 set = state->state & bits;
712 if (set && exclusive) {
713 *failed_start = state->start;
717 state->state |= bits;
718 start = state->end + 1;
719 merge_state(tree, state);
724 * | ---- desired range ---- |
727 * | ------------- state -------------- |
729 * We need to split the extent we found, and may flip bits on
732 * If the extent we found extends past our
733 * range, we just split and search again. It'll get split
734 * again the next time though.
736 * If the extent we found is inside our range, we set the
739 if (state->start < start) {
740 set = state->state & bits;
741 if (exclusive && set) {
742 *failed_start = start;
746 err = split_state(tree, state, prealloc, start);
747 BUG_ON(err == -EEXIST);
751 if (state->end <= end) {
752 state->state |= bits;
753 start = state->end + 1;
754 merge_state(tree, state);
756 start = state->start;
761 * | ---- desired range ---- |
762 * | state | or | state |
764 * There's a hole, we need to insert something in it and
765 * ignore the extent we found.
767 if (state->start > start) {
769 if (end < last_start)
772 this_end = last_start -1;
773 err = insert_state(tree, prealloc, start, this_end,
776 BUG_ON(err == -EEXIST);
779 start = this_end + 1;
783 * | ---- desired range ---- |
785 * We need to split the extent, and set the bit
788 if (state->start <= end && state->end > end) {
789 set = state->state & bits;
790 if (exclusive && set) {
791 *failed_start = start;
795 err = split_state(tree, state, prealloc, end + 1);
796 BUG_ON(err == -EEXIST);
798 prealloc->state |= bits;
799 merge_state(tree, prealloc);
807 write_unlock_irqrestore(&tree->lock, flags);
809 free_extent_state(prealloc);
816 write_unlock_irqrestore(&tree->lock, flags);
817 if (mask & __GFP_WAIT)
821 EXPORT_SYMBOL(set_extent_bit);
823 /* wrappers around set/clear extent bit */
824 int set_extent_dirty(struct extent_map_tree *tree, u64 start, u64 end,
827 return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL,
830 EXPORT_SYMBOL(set_extent_dirty);
832 int set_extent_bits(struct extent_map_tree *tree, u64 start, u64 end,
833 int bits, gfp_t mask)
835 return set_extent_bit(tree, start, end, bits, 0, NULL,
838 EXPORT_SYMBOL(set_extent_bits);
840 int clear_extent_bits(struct extent_map_tree *tree, u64 start, u64 end,
841 int bits, gfp_t mask)
843 return clear_extent_bit(tree, start, end, bits, 0, 0, mask);
845 EXPORT_SYMBOL(clear_extent_bits);
847 int set_extent_delalloc(struct extent_map_tree *tree, u64 start, u64 end,
850 return set_extent_bit(tree, start, end,
851 EXTENT_DELALLOC | EXTENT_DIRTY, 0, NULL,
854 EXPORT_SYMBOL(set_extent_delalloc);
856 int clear_extent_dirty(struct extent_map_tree *tree, u64 start, u64 end,
859 return clear_extent_bit(tree, start, end,
860 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, mask);
862 EXPORT_SYMBOL(clear_extent_dirty);
864 int set_extent_new(struct extent_map_tree *tree, u64 start, u64 end,
867 return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL,
870 EXPORT_SYMBOL(set_extent_new);
872 int clear_extent_new(struct extent_map_tree *tree, u64 start, u64 end,
875 return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0, mask);
877 EXPORT_SYMBOL(clear_extent_new);
879 int set_extent_uptodate(struct extent_map_tree *tree, u64 start, u64 end,
882 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, NULL,
885 EXPORT_SYMBOL(set_extent_uptodate);
887 int clear_extent_uptodate(struct extent_map_tree *tree, u64 start, u64 end,
890 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, mask);
892 EXPORT_SYMBOL(clear_extent_uptodate);
894 int set_extent_writeback(struct extent_map_tree *tree, u64 start, u64 end,
897 return set_extent_bit(tree, start, end, EXTENT_WRITEBACK,
900 EXPORT_SYMBOL(set_extent_writeback);
902 int clear_extent_writeback(struct extent_map_tree *tree, u64 start, u64 end,
905 return clear_extent_bit(tree, start, end, EXTENT_WRITEBACK, 1, 0, mask);
907 EXPORT_SYMBOL(clear_extent_writeback);
909 int wait_on_extent_writeback(struct extent_map_tree *tree, u64 start, u64 end)
911 return wait_extent_bit(tree, start, end, EXTENT_WRITEBACK);
913 EXPORT_SYMBOL(wait_on_extent_writeback);
916 * locks a range in ascending order, waiting for any locked regions
917 * it hits on the way. [start,end] are inclusive, and this will sleep.
919 int lock_extent(struct extent_map_tree *tree, u64 start, u64 end, gfp_t mask)
924 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1,
925 &failed_start, mask);
926 if (err == -EEXIST && (mask & __GFP_WAIT)) {
927 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
928 start = failed_start;
932 WARN_ON(start > end);
936 EXPORT_SYMBOL(lock_extent);
938 int unlock_extent(struct extent_map_tree *tree, u64 start, u64 end,
941 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, mask);
943 EXPORT_SYMBOL(unlock_extent);
946 * helper function to set pages and extents in the tree dirty
948 int set_range_dirty(struct extent_map_tree *tree, u64 start, u64 end)
950 unsigned long index = start >> PAGE_CACHE_SHIFT;
951 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
954 while (index <= end_index) {
955 page = find_get_page(tree->mapping, index);
957 __set_page_dirty_nobuffers(page);
958 page_cache_release(page);
961 set_extent_dirty(tree, start, end, GFP_NOFS);
964 EXPORT_SYMBOL(set_range_dirty);
967 * helper function to set both pages and extents in the tree writeback
969 int set_range_writeback(struct extent_map_tree *tree, u64 start, u64 end)
971 unsigned long index = start >> PAGE_CACHE_SHIFT;
972 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
975 while (index <= end_index) {
976 page = find_get_page(tree->mapping, index);
978 set_page_writeback(page);
979 page_cache_release(page);
982 set_extent_writeback(tree, start, end, GFP_NOFS);
985 EXPORT_SYMBOL(set_range_writeback);
987 int find_first_extent_bit(struct extent_map_tree *tree, u64 start,
988 u64 *start_ret, u64 *end_ret, int bits)
990 struct rb_node *node;
991 struct extent_state *state;
994 read_lock_irq(&tree->lock);
996 * this search will find all the extents that end after
999 node = tree_search(&tree->state, start);
1000 if (!node || IS_ERR(node)) {
1005 state = rb_entry(node, struct extent_state, rb_node);
1006 if (state->end >= start && (state->state & bits)) {
1007 *start_ret = state->start;
1008 *end_ret = state->end;
1012 node = rb_next(node);
1017 read_unlock_irq(&tree->lock);
1020 EXPORT_SYMBOL(find_first_extent_bit);
1022 u64 find_lock_delalloc_range(struct extent_map_tree *tree,
1023 u64 start, u64 lock_start, u64 *end, u64 max_bytes)
1025 struct rb_node *node;
1026 struct extent_state *state;
1027 u64 cur_start = start;
1029 u64 total_bytes = 0;
1031 write_lock_irq(&tree->lock);
1033 * this search will find all the extents that end after
1037 node = tree_search(&tree->state, cur_start);
1038 if (!node || IS_ERR(node)) {
1043 state = rb_entry(node, struct extent_state, rb_node);
1044 if (state->start != cur_start) {
1047 if (!(state->state & EXTENT_DELALLOC)) {
1050 if (state->start >= lock_start) {
1051 if (state->state & EXTENT_LOCKED) {
1053 atomic_inc(&state->refs);
1054 prepare_to_wait(&state->wq, &wait,
1055 TASK_UNINTERRUPTIBLE);
1056 write_unlock_irq(&tree->lock);
1058 write_lock_irq(&tree->lock);
1059 finish_wait(&state->wq, &wait);
1060 free_extent_state(state);
1063 state->state |= EXTENT_LOCKED;
1067 cur_start = state->end + 1;
1068 node = rb_next(node);
1071 total_bytes += state->end - state->start + 1;
1072 if (total_bytes >= max_bytes)
1076 write_unlock_irq(&tree->lock);
1081 * helper function to lock both pages and extents in the tree.
1082 * pages must be locked first.
1084 int lock_range(struct extent_map_tree *tree, u64 start, u64 end)
1086 unsigned long index = start >> PAGE_CACHE_SHIFT;
1087 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1091 while (index <= end_index) {
1092 page = grab_cache_page(tree->mapping, index);
1098 err = PTR_ERR(page);
1103 lock_extent(tree, start, end, GFP_NOFS);
1108 * we failed above in getting the page at 'index', so we undo here
1109 * up to but not including the page at 'index'
1112 index = start >> PAGE_CACHE_SHIFT;
1113 while (index < end_index) {
1114 page = find_get_page(tree->mapping, index);
1116 page_cache_release(page);
1121 EXPORT_SYMBOL(lock_range);
1124 * helper function to unlock both pages and extents in the tree.
1126 int unlock_range(struct extent_map_tree *tree, u64 start, u64 end)
1128 unsigned long index = start >> PAGE_CACHE_SHIFT;
1129 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1132 while (index <= end_index) {
1133 page = find_get_page(tree->mapping, index);
1135 page_cache_release(page);
1138 unlock_extent(tree, start, end, GFP_NOFS);
1141 EXPORT_SYMBOL(unlock_range);
1143 int set_state_private(struct extent_map_tree *tree, u64 start, u64 private)
1145 struct rb_node *node;
1146 struct extent_state *state;
1149 write_lock_irq(&tree->lock);
1151 * this search will find all the extents that end after
1154 node = tree_search(&tree->state, start);
1155 if (!node || IS_ERR(node)) {
1159 state = rb_entry(node, struct extent_state, rb_node);
1160 if (state->start != start) {
1164 state->private = private;
1166 write_unlock_irq(&tree->lock);
1170 int get_state_private(struct extent_map_tree *tree, u64 start, u64 *private)
1172 struct rb_node *node;
1173 struct extent_state *state;
1176 read_lock_irq(&tree->lock);
1178 * this search will find all the extents that end after
1181 node = tree_search(&tree->state, start);
1182 if (!node || IS_ERR(node)) {
1186 state = rb_entry(node, struct extent_state, rb_node);
1187 if (state->start != start) {
1191 *private = state->private;
1193 read_unlock_irq(&tree->lock);
1198 * searches a range in the state tree for a given mask.
1199 * If 'filled' == 1, this returns 1 only if ever extent in the tree
1200 * has the bits set. Otherwise, 1 is returned if any bit in the
1201 * range is found set.
1203 int test_range_bit(struct extent_map_tree *tree, u64 start, u64 end,
1204 int bits, int filled)
1206 struct extent_state *state = NULL;
1207 struct rb_node *node;
1210 read_lock_irq(&tree->lock);
1211 node = tree_search(&tree->state, start);
1212 while (node && start <= end) {
1213 state = rb_entry(node, struct extent_state, rb_node);
1214 if (state->start > end)
1217 if (filled && state->start > start) {
1221 if (state->state & bits) {
1225 } else if (filled) {
1229 start = state->end + 1;
1232 node = rb_next(node);
1234 read_unlock_irq(&tree->lock);
1237 EXPORT_SYMBOL(test_range_bit);
1240 * helper function to set a given page up to date if all the
1241 * extents in the tree for that page are up to date
1243 static int check_page_uptodate(struct extent_map_tree *tree,
1246 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1247 u64 end = start + PAGE_CACHE_SIZE - 1;
1248 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1))
1249 SetPageUptodate(page);
1254 * helper function to unlock a page if all the extents in the tree
1255 * for that page are unlocked
1257 static int check_page_locked(struct extent_map_tree *tree,
1260 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1261 u64 end = start + PAGE_CACHE_SIZE - 1;
1262 if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0))
1268 * helper function to end page writeback if all the extents
1269 * in the tree for that page are done with writeback
1271 static int check_page_writeback(struct extent_map_tree *tree,
1274 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1275 u64 end = start + PAGE_CACHE_SIZE - 1;
1276 if (!test_range_bit(tree, start, end, EXTENT_WRITEBACK, 0))
1277 end_page_writeback(page);
1281 /* lots and lots of room for performance fixes in the end_bio funcs */
1284 * after a writepage IO is done, we need to:
1285 * clear the uptodate bits on error
1286 * clear the writeback bits in the extent tree for this IO
1287 * end_page_writeback if the page has no more pending IO
1289 * Scheduling is not allowed, so the extent state tree is expected
1290 * to have one and only one object corresponding to this IO.
1292 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1293 static void end_bio_extent_writepage(struct bio *bio, int err)
1295 static int end_bio_extent_writepage(struct bio *bio,
1296 unsigned int bytes_done, int err)
1299 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1300 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1301 struct extent_map_tree *tree = bio->bi_private;
1306 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1312 struct page *page = bvec->bv_page;
1313 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1315 end = start + bvec->bv_len - 1;
1317 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1322 if (--bvec >= bio->bi_io_vec)
1323 prefetchw(&bvec->bv_page->flags);
1326 clear_extent_uptodate(tree, start, end, GFP_ATOMIC);
1327 ClearPageUptodate(page);
1330 clear_extent_writeback(tree, start, end, GFP_ATOMIC);
1333 end_page_writeback(page);
1335 check_page_writeback(tree, page);
1336 if (tree->ops && tree->ops->writepage_end_io_hook)
1337 tree->ops->writepage_end_io_hook(page, start, end);
1338 } while (bvec >= bio->bi_io_vec);
1341 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1347 * after a readpage IO is done, we need to:
1348 * clear the uptodate bits on error
1349 * set the uptodate bits if things worked
1350 * set the page up to date if all extents in the tree are uptodate
1351 * clear the lock bit in the extent tree
1352 * unlock the page if there are no other extents locked for it
1354 * Scheduling is not allowed, so the extent state tree is expected
1355 * to have one and only one object corresponding to this IO.
1357 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1358 static void end_bio_extent_readpage(struct bio *bio, int err)
1360 static int end_bio_extent_readpage(struct bio *bio,
1361 unsigned int bytes_done, int err)
1364 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1365 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1366 struct extent_map_tree *tree = bio->bi_private;
1372 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1378 struct page *page = bvec->bv_page;
1379 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1381 end = start + bvec->bv_len - 1;
1383 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1388 if (--bvec >= bio->bi_io_vec)
1389 prefetchw(&bvec->bv_page->flags);
1391 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
1392 ret = tree->ops->readpage_end_io_hook(page, start, end);
1397 set_extent_uptodate(tree, start, end, GFP_ATOMIC);
1399 SetPageUptodate(page);
1401 check_page_uptodate(tree, page);
1403 ClearPageUptodate(page);
1407 unlock_extent(tree, start, end, GFP_ATOMIC);
1412 check_page_locked(tree, page);
1413 } while (bvec >= bio->bi_io_vec);
1416 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1422 * IO done from prepare_write is pretty simple, we just unlock
1423 * the structs in the extent tree when done, and set the uptodate bits
1426 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1427 static void end_bio_extent_preparewrite(struct bio *bio, int err)
1429 static int end_bio_extent_preparewrite(struct bio *bio,
1430 unsigned int bytes_done, int err)
1433 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1434 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1435 struct extent_map_tree *tree = bio->bi_private;
1439 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1445 struct page *page = bvec->bv_page;
1446 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1448 end = start + bvec->bv_len - 1;
1450 if (--bvec >= bio->bi_io_vec)
1451 prefetchw(&bvec->bv_page->flags);
1454 set_extent_uptodate(tree, start, end, GFP_ATOMIC);
1456 ClearPageUptodate(page);
1460 unlock_extent(tree, start, end, GFP_ATOMIC);
1462 } while (bvec >= bio->bi_io_vec);
1465 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1471 extent_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
1476 bio = bio_alloc(gfp_flags, nr_vecs);
1478 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
1479 while (!bio && (nr_vecs /= 2))
1480 bio = bio_alloc(gfp_flags, nr_vecs);
1484 bio->bi_bdev = bdev;
1485 bio->bi_sector = first_sector;
1490 static int submit_one_bio(int rw, struct bio *bio)
1494 submit_bio(rw, bio);
1495 if (bio_flagged(bio, BIO_EOPNOTSUPP))
1501 static int submit_extent_page(int rw, struct extent_map_tree *tree,
1502 struct page *page, sector_t sector,
1503 size_t size, unsigned long offset,
1504 struct block_device *bdev,
1505 struct bio **bio_ret,
1507 bio_end_io_t end_io_func)
1513 if (bio_ret && *bio_ret) {
1515 if (bio->bi_sector + (bio->bi_size >> 9) != sector ||
1516 bio_add_page(bio, page, size, offset) < size) {
1517 ret = submit_one_bio(rw, bio);
1523 nr = min(max_pages, bio_get_nr_vecs(bdev));
1524 bio = extent_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
1526 printk("failed to allocate bio nr %d\n", nr);
1528 bio_add_page(bio, page, size, offset);
1529 bio->bi_end_io = end_io_func;
1530 bio->bi_private = tree;
1534 ret = submit_one_bio(rw, bio);
1540 void set_page_extent_mapped(struct page *page)
1542 if (!PagePrivate(page)) {
1543 SetPagePrivate(page);
1544 WARN_ON(!page->mapping->a_ops->invalidatepage);
1545 set_page_private(page, EXTENT_PAGE_PRIVATE);
1546 page_cache_get(page);
1551 * basic readpage implementation. Locked extent state structs are inserted
1552 * into the tree that are removed when the IO is done (by the end_io
1555 int extent_read_full_page(struct extent_map_tree *tree, struct page *page,
1556 get_extent_t *get_extent)
1558 struct inode *inode = page->mapping->host;
1559 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1560 u64 page_end = start + PAGE_CACHE_SIZE - 1;
1564 u64 last_byte = i_size_read(inode);
1568 struct extent_map *em;
1569 struct block_device *bdev;
1572 size_t page_offset = 0;
1574 size_t blocksize = inode->i_sb->s_blocksize;
1576 set_page_extent_mapped(page);
1579 lock_extent(tree, start, end, GFP_NOFS);
1581 while (cur <= end) {
1582 if (cur >= last_byte) {
1583 iosize = PAGE_CACHE_SIZE - page_offset;
1584 zero_user_page(page, page_offset, iosize, KM_USER0);
1585 set_extent_uptodate(tree, cur, cur + iosize - 1,
1587 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1590 em = get_extent(inode, page, page_offset, cur, end, 0);
1591 if (IS_ERR(em) || !em) {
1593 unlock_extent(tree, cur, end, GFP_NOFS);
1597 extent_offset = cur - em->start;
1598 BUG_ON(em->end < cur);
1601 iosize = min(em->end - cur, end - cur) + 1;
1602 cur_end = min(em->end, end);
1603 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
1604 sector = (em->block_start + extent_offset) >> 9;
1606 block_start = em->block_start;
1607 free_extent_map(em);
1610 /* we've found a hole, just zero and go on */
1611 if (block_start == EXTENT_MAP_HOLE) {
1612 zero_user_page(page, page_offset, iosize, KM_USER0);
1613 set_extent_uptodate(tree, cur, cur + iosize - 1,
1615 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1617 page_offset += iosize;
1620 /* the get_extent function already copied into the page */
1621 if (test_range_bit(tree, cur, cur_end, EXTENT_UPTODATE, 1)) {
1622 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1624 page_offset += iosize;
1629 if (tree->ops && tree->ops->readpage_io_hook) {
1630 ret = tree->ops->readpage_io_hook(page, cur,
1634 ret = submit_extent_page(READ, tree, page,
1635 sector, iosize, page_offset,
1637 end_bio_extent_readpage);
1642 page_offset += iosize;
1646 if (!PageError(page))
1647 SetPageUptodate(page);
1652 EXPORT_SYMBOL(extent_read_full_page);
1655 * the writepage semantics are similar to regular writepage. extent
1656 * records are inserted to lock ranges in the tree, and as dirty areas
1657 * are found, they are marked writeback. Then the lock bits are removed
1658 * and the end_io handler clears the writeback ranges
1660 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
1663 struct inode *inode = page->mapping->host;
1664 struct extent_page_data *epd = data;
1665 struct extent_map_tree *tree = epd->tree;
1666 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1667 u64 page_end = start + PAGE_CACHE_SIZE - 1;
1671 u64 last_byte = i_size_read(inode);
1675 struct extent_map *em;
1676 struct block_device *bdev;
1679 size_t page_offset = 0;
1681 loff_t i_size = i_size_read(inode);
1682 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
1686 WARN_ON(!PageLocked(page));
1687 if (page->index > end_index) {
1688 clear_extent_dirty(tree, start, page_end, GFP_NOFS);
1693 if (page->index == end_index) {
1694 size_t offset = i_size & (PAGE_CACHE_SIZE - 1);
1695 zero_user_page(page, offset,
1696 PAGE_CACHE_SIZE - offset, KM_USER0);
1699 set_page_extent_mapped(page);
1701 lock_extent(tree, start, page_end, GFP_NOFS);
1702 nr_delalloc = find_lock_delalloc_range(tree, start, page_end + 1,
1706 tree->ops->fill_delalloc(inode, start, delalloc_end);
1707 if (delalloc_end >= page_end + 1) {
1708 clear_extent_bit(tree, page_end + 1, delalloc_end,
1709 EXTENT_LOCKED | EXTENT_DELALLOC,
1712 clear_extent_bit(tree, start, page_end, EXTENT_DELALLOC,
1714 if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0)) {
1715 printk("found delalloc bits after clear extent_bit\n");
1717 } else if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0)) {
1718 printk("found delalloc bits after find_delalloc_range returns 0\n");
1722 if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0)) {
1723 printk("found delalloc bits after lock_extent\n");
1726 if (last_byte <= start) {
1727 clear_extent_dirty(tree, start, page_end, GFP_NOFS);
1731 set_extent_uptodate(tree, start, page_end, GFP_NOFS);
1732 blocksize = inode->i_sb->s_blocksize;
1734 while (cur <= end) {
1735 if (cur >= last_byte) {
1736 clear_extent_dirty(tree, cur, page_end, GFP_NOFS);
1739 em = epd->get_extent(inode, page, page_offset, cur, end, 1);
1740 if (IS_ERR(em) || !em) {
1745 extent_offset = cur - em->start;
1746 BUG_ON(em->end < cur);
1748 iosize = min(em->end - cur, end - cur) + 1;
1749 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
1750 sector = (em->block_start + extent_offset) >> 9;
1752 block_start = em->block_start;
1753 free_extent_map(em);
1756 if (block_start == EXTENT_MAP_HOLE ||
1757 block_start == EXTENT_MAP_INLINE) {
1758 clear_extent_dirty(tree, cur,
1759 cur + iosize - 1, GFP_NOFS);
1761 page_offset += iosize;
1765 /* leave this out until we have a page_mkwrite call */
1766 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
1769 page_offset += iosize;
1772 clear_extent_dirty(tree, cur, cur + iosize - 1, GFP_NOFS);
1773 if (tree->ops && tree->ops->writepage_io_hook) {
1774 ret = tree->ops->writepage_io_hook(page, cur,
1782 unsigned long nr = end_index + 1;
1783 set_range_writeback(tree, cur, cur + iosize - 1);
1785 ret = submit_extent_page(WRITE, tree, page, sector,
1786 iosize, page_offset, bdev,
1788 end_bio_extent_writepage);
1793 page_offset += iosize;
1797 unlock_extent(tree, start, page_end, GFP_NOFS);
1802 int extent_write_full_page(struct extent_map_tree *tree, struct page *page,
1803 get_extent_t *get_extent,
1804 struct writeback_control *wbc)
1807 struct extent_page_data epd = {
1810 .get_extent = get_extent,
1813 ret = __extent_writepage(page, wbc, &epd);
1815 submit_one_bio(WRITE, epd.bio);
1818 EXPORT_SYMBOL(extent_write_full_page);
1820 int extent_writepages(struct extent_map_tree *tree,
1821 struct address_space *mapping,
1822 get_extent_t *get_extent,
1823 struct writeback_control *wbc)
1826 struct extent_page_data epd = {
1829 .get_extent = get_extent,
1832 ret = write_cache_pages(mapping, wbc, __extent_writepage, &epd);
1834 submit_one_bio(WRITE, epd.bio);
1837 EXPORT_SYMBOL(extent_writepages);
1840 * basic invalidatepage code, this waits on any locked or writeback
1841 * ranges corresponding to the page, and then deletes any extent state
1842 * records from the tree
1844 int extent_invalidatepage(struct extent_map_tree *tree,
1845 struct page *page, unsigned long offset)
1847 u64 start = ((u64)page->index << PAGE_CACHE_SHIFT);
1848 u64 end = start + PAGE_CACHE_SIZE - 1;
1849 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
1851 start += (offset + blocksize -1) & ~(blocksize - 1);
1855 lock_extent(tree, start, end, GFP_NOFS);
1856 wait_on_extent_writeback(tree, start, end);
1857 clear_extent_bit(tree, start, end,
1858 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC,
1862 EXPORT_SYMBOL(extent_invalidatepage);
1865 * simple commit_write call, set_range_dirty is used to mark both
1866 * the pages and the extent records as dirty
1868 int extent_commit_write(struct extent_map_tree *tree,
1869 struct inode *inode, struct page *page,
1870 unsigned from, unsigned to)
1872 loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1874 set_page_extent_mapped(page);
1875 set_page_dirty(page);
1877 if (pos > inode->i_size) {
1878 i_size_write(inode, pos);
1879 mark_inode_dirty(inode);
1883 EXPORT_SYMBOL(extent_commit_write);
1885 int extent_prepare_write(struct extent_map_tree *tree,
1886 struct inode *inode, struct page *page,
1887 unsigned from, unsigned to, get_extent_t *get_extent)
1889 u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
1890 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
1892 u64 orig_block_start;
1895 struct extent_map *em;
1896 unsigned blocksize = 1 << inode->i_blkbits;
1897 size_t page_offset = 0;
1898 size_t block_off_start;
1899 size_t block_off_end;
1905 set_page_extent_mapped(page);
1907 block_start = (page_start + from) & ~((u64)blocksize - 1);
1908 block_end = (page_start + to - 1) | (blocksize - 1);
1909 orig_block_start = block_start;
1911 lock_extent(tree, page_start, page_end, GFP_NOFS);
1912 while(block_start <= block_end) {
1913 em = get_extent(inode, page, page_offset, block_start,
1915 if (IS_ERR(em) || !em) {
1918 cur_end = min(block_end, em->end);
1919 block_off_start = block_start & (PAGE_CACHE_SIZE - 1);
1920 block_off_end = block_off_start + blocksize;
1921 isnew = clear_extent_new(tree, block_start, cur_end, GFP_NOFS);
1923 if (!PageUptodate(page) && isnew &&
1924 (block_off_end > to || block_off_start < from)) {
1927 kaddr = kmap_atomic(page, KM_USER0);
1928 if (block_off_end > to)
1929 memset(kaddr + to, 0, block_off_end - to);
1930 if (block_off_start < from)
1931 memset(kaddr + block_off_start, 0,
1932 from - block_off_start);
1933 flush_dcache_page(page);
1934 kunmap_atomic(kaddr, KM_USER0);
1936 if (!isnew && !PageUptodate(page) &&
1937 (block_off_end > to || block_off_start < from) &&
1938 !test_range_bit(tree, block_start, cur_end,
1939 EXTENT_UPTODATE, 1)) {
1941 u64 extent_offset = block_start - em->start;
1943 sector = (em->block_start + extent_offset) >> 9;
1944 iosize = (cur_end - block_start + blocksize - 1) &
1945 ~((u64)blocksize - 1);
1947 * we've already got the extent locked, but we
1948 * need to split the state such that our end_bio
1949 * handler can clear the lock.
1951 set_extent_bit(tree, block_start,
1952 block_start + iosize - 1,
1953 EXTENT_LOCKED, 0, NULL, GFP_NOFS);
1954 ret = submit_extent_page(READ, tree, page,
1955 sector, iosize, page_offset, em->bdev,
1957 end_bio_extent_preparewrite);
1959 block_start = block_start + iosize;
1961 set_extent_uptodate(tree, block_start, cur_end,
1963 unlock_extent(tree, block_start, cur_end, GFP_NOFS);
1964 block_start = cur_end + 1;
1966 page_offset = block_start & (PAGE_CACHE_SIZE - 1);
1967 free_extent_map(em);
1970 wait_extent_bit(tree, orig_block_start,
1971 block_end, EXTENT_LOCKED);
1973 check_page_uptodate(tree, page);
1975 /* FIXME, zero out newly allocated blocks on error */
1978 EXPORT_SYMBOL(extent_prepare_write);
1981 * a helper for releasepage. As long as there are no locked extents
1982 * in the range corresponding to the page, both state records and extent
1983 * map records are removed
1985 int try_release_extent_mapping(struct extent_map_tree *tree, struct page *page)
1987 struct extent_map *em;
1988 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1989 u64 end = start + PAGE_CACHE_SIZE - 1;
1990 u64 orig_start = start;
1993 while (start <= end) {
1994 em = lookup_extent_mapping(tree, start, end);
1995 if (!em || IS_ERR(em))
1997 if (!test_range_bit(tree, em->start, em->end,
1998 EXTENT_LOCKED, 0)) {
1999 remove_extent_mapping(tree, em);
2000 /* once for the rb tree */
2001 free_extent_map(em);
2003 start = em->end + 1;
2005 free_extent_map(em);
2007 if (test_range_bit(tree, orig_start, end, EXTENT_LOCKED, 0))
2010 clear_extent_bit(tree, orig_start, end, EXTENT_UPTODATE,
2014 EXPORT_SYMBOL(try_release_extent_mapping);
2016 sector_t extent_bmap(struct address_space *mapping, sector_t iblock,
2017 get_extent_t *get_extent)
2019 struct inode *inode = mapping->host;
2020 u64 start = iblock << inode->i_blkbits;
2021 u64 end = start + (1 << inode->i_blkbits) - 1;
2022 sector_t sector = 0;
2023 struct extent_map *em;
2025 em = get_extent(inode, NULL, 0, start, end, 0);
2026 if (!em || IS_ERR(em))
2029 if (em->block_start == EXTENT_MAP_INLINE ||
2030 em->block_start == EXTENT_MAP_HOLE)
2033 sector = (em->block_start + start - em->start) >> inode->i_blkbits;
2035 free_extent_map(em);
2039 static int add_lru(struct extent_map_tree *tree, struct extent_buffer *eb)
2041 if (list_empty(&eb->lru)) {
2042 extent_buffer_get(eb);
2043 list_add(&eb->lru, &tree->buffer_lru);
2045 if (tree->lru_size >= BUFFER_LRU_MAX) {
2046 struct extent_buffer *rm;
2047 rm = list_entry(tree->buffer_lru.prev,
2048 struct extent_buffer, lru);
2051 free_extent_buffer(rm);
2054 list_move(&eb->lru, &tree->buffer_lru);
2057 static struct extent_buffer *find_lru(struct extent_map_tree *tree,
2058 u64 start, unsigned long len)
2060 struct list_head *lru = &tree->buffer_lru;
2061 struct list_head *cur = lru->next;
2062 struct extent_buffer *eb;
2064 if (list_empty(lru))
2068 eb = list_entry(cur, struct extent_buffer, lru);
2069 if (eb->start == start && eb->len == len) {
2070 extent_buffer_get(eb);
2074 } while (cur != lru);
2078 static inline unsigned long num_extent_pages(u64 start, u64 len)
2080 return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
2081 (start >> PAGE_CACHE_SHIFT);
2084 static inline struct page *extent_buffer_page(struct extent_buffer *eb,
2088 struct address_space *mapping;
2091 return eb->first_page;
2092 i += eb->start >> PAGE_CACHE_SHIFT;
2093 mapping = eb->first_page->mapping;
2094 read_lock_irq(&mapping->tree_lock);
2095 p = radix_tree_lookup(&mapping->page_tree, i);
2096 read_unlock_irq(&mapping->tree_lock);
2100 static struct extent_buffer *__alloc_extent_buffer(struct extent_map_tree *tree,
2105 struct extent_buffer *eb = NULL;
2107 spin_lock(&tree->lru_lock);
2108 eb = find_lru(tree, start, len);
2112 spin_unlock(&tree->lru_lock);
2115 memset(eb, 0, sizeof(*eb));
2117 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
2119 INIT_LIST_HEAD(&eb->lru);
2122 atomic_set(&eb->refs, 1);
2124 spin_lock(&tree->lru_lock);
2127 spin_unlock(&tree->lru_lock);
2131 static void __free_extent_buffer(struct extent_buffer *eb)
2133 kmem_cache_free(extent_buffer_cache, eb);
2136 struct extent_buffer *alloc_extent_buffer(struct extent_map_tree *tree,
2137 u64 start, unsigned long len,
2141 unsigned long num_pages = num_extent_pages(start, len);
2143 unsigned long index = start >> PAGE_CACHE_SHIFT;
2144 struct extent_buffer *eb;
2146 struct address_space *mapping = tree->mapping;
2149 eb = __alloc_extent_buffer(tree, start, len, mask);
2150 if (!eb || IS_ERR(eb))
2153 if (eb->flags & EXTENT_BUFFER_FILLED)
2157 eb->first_page = page0;
2160 page_cache_get(page0);
2161 mark_page_accessed(page0);
2162 set_page_extent_mapped(page0);
2163 set_page_private(page0, EXTENT_PAGE_PRIVATE_FIRST_PAGE |
2168 for (; i < num_pages; i++, index++) {
2169 p = find_or_create_page(mapping, index, mask | __GFP_HIGHMEM);
2172 /* make sure the free only frees the pages we've
2173 * grabbed a reference on
2175 eb->len = i << PAGE_CACHE_SHIFT;
2176 eb->start &= ~((u64)PAGE_CACHE_SIZE - 1);
2179 set_page_extent_mapped(p);
2180 mark_page_accessed(p);
2183 set_page_private(p, EXTENT_PAGE_PRIVATE_FIRST_PAGE |
2186 set_page_private(p, EXTENT_PAGE_PRIVATE);
2188 if (!PageUptodate(p))
2193 eb->flags |= EXTENT_UPTODATE;
2194 eb->flags |= EXTENT_BUFFER_FILLED;
2197 free_extent_buffer(eb);
2200 EXPORT_SYMBOL(alloc_extent_buffer);
2202 struct extent_buffer *find_extent_buffer(struct extent_map_tree *tree,
2203 u64 start, unsigned long len,
2206 unsigned long num_pages = num_extent_pages(start, len);
2207 unsigned long i; unsigned long index = start >> PAGE_CACHE_SHIFT;
2208 struct extent_buffer *eb;
2210 struct address_space *mapping = tree->mapping;
2213 eb = __alloc_extent_buffer(tree, start, len, mask);
2214 if (!eb || IS_ERR(eb))
2217 if (eb->flags & EXTENT_BUFFER_FILLED)
2220 for (i = 0; i < num_pages; i++, index++) {
2221 p = find_lock_page(mapping, index);
2223 /* make sure the free only frees the pages we've
2224 * grabbed a reference on
2226 eb->len = i << PAGE_CACHE_SHIFT;
2227 eb->start &= ~((u64)PAGE_CACHE_SIZE - 1);
2230 set_page_extent_mapped(p);
2231 mark_page_accessed(p);
2235 set_page_private(p, EXTENT_PAGE_PRIVATE_FIRST_PAGE |
2238 set_page_private(p, EXTENT_PAGE_PRIVATE);
2241 if (!PageUptodate(p))
2246 eb->flags |= EXTENT_UPTODATE;
2247 eb->flags |= EXTENT_BUFFER_FILLED;
2250 free_extent_buffer(eb);
2253 EXPORT_SYMBOL(find_extent_buffer);
2255 void free_extent_buffer(struct extent_buffer *eb)
2258 unsigned long num_pages;
2263 if (!atomic_dec_and_test(&eb->refs))
2266 num_pages = num_extent_pages(eb->start, eb->len);
2268 for (i = 0; i < num_pages; i++) {
2269 page_cache_release(extent_buffer_page(eb, i));
2271 __free_extent_buffer(eb);
2273 EXPORT_SYMBOL(free_extent_buffer);
2275 int clear_extent_buffer_dirty(struct extent_map_tree *tree,
2276 struct extent_buffer *eb)
2280 unsigned long num_pages;
2283 u64 start = eb->start;
2284 u64 end = start + eb->len - 1;
2286 set = clear_extent_dirty(tree, start, end, GFP_NOFS);
2287 num_pages = num_extent_pages(eb->start, eb->len);
2289 for (i = 0; i < num_pages; i++) {
2290 page = extent_buffer_page(eb, i);
2293 * if we're on the last page or the first page and the
2294 * block isn't aligned on a page boundary, do extra checks
2295 * to make sure we don't clean page that is partially dirty
2297 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
2298 ((i == num_pages - 1) &&
2299 ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
2300 start = (u64)page->index << PAGE_CACHE_SHIFT;
2301 end = start + PAGE_CACHE_SIZE - 1;
2302 if (test_range_bit(tree, start, end,
2308 clear_page_dirty_for_io(page);
2313 EXPORT_SYMBOL(clear_extent_buffer_dirty);
2315 int wait_on_extent_buffer_writeback(struct extent_map_tree *tree,
2316 struct extent_buffer *eb)
2318 return wait_on_extent_writeback(tree, eb->start,
2319 eb->start + eb->len - 1);
2321 EXPORT_SYMBOL(wait_on_extent_buffer_writeback);
2323 int set_extent_buffer_dirty(struct extent_map_tree *tree,
2324 struct extent_buffer *eb)
2327 unsigned long num_pages;
2329 num_pages = num_extent_pages(eb->start, eb->len);
2330 for (i = 0; i < num_pages; i++) {
2331 struct page *page = extent_buffer_page(eb, i);
2332 /* writepage may need to do something special for the
2333 * first page, we have to make sure page->private is
2334 * properly set. releasepage may drop page->private
2335 * on us if the page isn't already dirty.
2339 set_page_private(page,
2340 EXTENT_PAGE_PRIVATE_FIRST_PAGE |
2343 __set_page_dirty_nobuffers(extent_buffer_page(eb, i));
2347 return set_extent_dirty(tree, eb->start,
2348 eb->start + eb->len - 1, GFP_NOFS);
2350 EXPORT_SYMBOL(set_extent_buffer_dirty);
2352 int set_extent_buffer_uptodate(struct extent_map_tree *tree,
2353 struct extent_buffer *eb)
2357 unsigned long num_pages;
2359 num_pages = num_extent_pages(eb->start, eb->len);
2361 set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
2363 for (i = 0; i < num_pages; i++) {
2364 page = extent_buffer_page(eb, i);
2365 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
2366 ((i == num_pages - 1) &&
2367 ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
2368 check_page_uptodate(tree, page);
2371 SetPageUptodate(page);
2375 EXPORT_SYMBOL(set_extent_buffer_uptodate);
2377 int extent_buffer_uptodate(struct extent_map_tree *tree,
2378 struct extent_buffer *eb)
2380 if (eb->flags & EXTENT_UPTODATE)
2382 return test_range_bit(tree, eb->start, eb->start + eb->len - 1,
2383 EXTENT_UPTODATE, 1);
2385 EXPORT_SYMBOL(extent_buffer_uptodate);
2387 int read_extent_buffer_pages(struct extent_map_tree *tree,
2388 struct extent_buffer *eb,
2393 unsigned long start_i;
2397 unsigned long num_pages;
2399 if (eb->flags & EXTENT_UPTODATE)
2402 if (0 && test_range_bit(tree, eb->start, eb->start + eb->len - 1,
2403 EXTENT_UPTODATE, 1)) {
2407 WARN_ON(start < eb->start);
2408 start_i = (start >> PAGE_CACHE_SHIFT) -
2409 (eb->start >> PAGE_CACHE_SHIFT);
2414 num_pages = num_extent_pages(eb->start, eb->len);
2415 for (i = start_i; i < num_pages; i++) {
2416 page = extent_buffer_page(eb, i);
2417 if (PageUptodate(page)) {
2421 if (TestSetPageLocked(page)) {
2427 if (!PageUptodate(page)) {
2428 err = page->mapping->a_ops->readpage(NULL, page);
2441 for (i = start_i; i < num_pages; i++) {
2442 page = extent_buffer_page(eb, i);
2443 wait_on_page_locked(page);
2444 if (!PageUptodate(page)) {
2449 eb->flags |= EXTENT_UPTODATE;
2452 EXPORT_SYMBOL(read_extent_buffer_pages);
2454 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
2455 unsigned long start,
2462 char *dst = (char *)dstv;
2463 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
2464 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
2465 unsigned long num_pages = num_extent_pages(eb->start, eb->len);
2467 WARN_ON(start > eb->len);
2468 WARN_ON(start + len > eb->start + eb->len);
2470 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
2473 page = extent_buffer_page(eb, i);
2474 if (!PageUptodate(page)) {
2475 printk("page %lu not up to date i %lu, total %lu, len %lu\n", page->index, i, num_pages, eb->len);
2478 WARN_ON(!PageUptodate(page));
2480 cur = min(len, (PAGE_CACHE_SIZE - offset));
2481 kaddr = kmap_atomic(page, KM_USER1);
2482 memcpy(dst, kaddr + offset, cur);
2483 kunmap_atomic(kaddr, KM_USER1);
2491 EXPORT_SYMBOL(read_extent_buffer);
2493 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
2494 unsigned long min_len, char **token, char **map,
2495 unsigned long *map_start,
2496 unsigned long *map_len, int km)
2498 size_t offset = start & (PAGE_CACHE_SIZE - 1);
2501 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
2502 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
2503 unsigned long end_i = (start_offset + start + min_len - 1) >>
2510 offset = start_offset;
2514 *map_start = (i << PAGE_CACHE_SHIFT) - start_offset;
2516 if (start + min_len > eb->len) {
2517 printk("bad mapping eb start %Lu len %lu, wanted %lu %lu\n", eb->start, eb->len, start, min_len);
2521 p = extent_buffer_page(eb, i);
2522 WARN_ON(!PageUptodate(p));
2523 kaddr = kmap_atomic(p, km);
2525 *map = kaddr + offset;
2526 *map_len = PAGE_CACHE_SIZE - offset;
2529 EXPORT_SYMBOL(map_private_extent_buffer);
2531 int map_extent_buffer(struct extent_buffer *eb, unsigned long start,
2532 unsigned long min_len,
2533 char **token, char **map,
2534 unsigned long *map_start,
2535 unsigned long *map_len, int km)
2539 if (eb->map_token) {
2540 unmap_extent_buffer(eb, eb->map_token, km);
2541 eb->map_token = NULL;
2544 err = map_private_extent_buffer(eb, start, min_len, token, map,
2545 map_start, map_len, km);
2547 eb->map_token = *token;
2549 eb->map_start = *map_start;
2550 eb->map_len = *map_len;
2554 EXPORT_SYMBOL(map_extent_buffer);
2556 void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km)
2558 kunmap_atomic(token, km);
2560 EXPORT_SYMBOL(unmap_extent_buffer);
2562 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
2563 unsigned long start,
2570 char *ptr = (char *)ptrv;
2571 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
2572 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
2575 WARN_ON(start > eb->len);
2576 WARN_ON(start + len > eb->start + eb->len);
2578 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
2581 page = extent_buffer_page(eb, i);
2582 WARN_ON(!PageUptodate(page));
2584 cur = min(len, (PAGE_CACHE_SIZE - offset));
2586 kaddr = kmap_atomic(page, KM_USER0);
2587 ret = memcmp(ptr, kaddr + offset, cur);
2588 kunmap_atomic(kaddr, KM_USER0);
2599 EXPORT_SYMBOL(memcmp_extent_buffer);
2601 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
2602 unsigned long start, unsigned long len)
2608 char *src = (char *)srcv;
2609 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
2610 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
2612 WARN_ON(start > eb->len);
2613 WARN_ON(start + len > eb->start + eb->len);
2615 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
2618 page = extent_buffer_page(eb, i);
2619 WARN_ON(!PageUptodate(page));
2621 cur = min(len, PAGE_CACHE_SIZE - offset);
2622 kaddr = kmap_atomic(page, KM_USER1);
2623 memcpy(kaddr + offset, src, cur);
2624 kunmap_atomic(kaddr, KM_USER1);
2632 EXPORT_SYMBOL(write_extent_buffer);
2634 void memset_extent_buffer(struct extent_buffer *eb, char c,
2635 unsigned long start, unsigned long len)
2641 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
2642 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
2644 WARN_ON(start > eb->len);
2645 WARN_ON(start + len > eb->start + eb->len);
2647 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
2650 page = extent_buffer_page(eb, i);
2651 WARN_ON(!PageUptodate(page));
2653 cur = min(len, PAGE_CACHE_SIZE - offset);
2654 kaddr = kmap_atomic(page, KM_USER0);
2655 memset(kaddr + offset, c, cur);
2656 kunmap_atomic(kaddr, KM_USER0);
2663 EXPORT_SYMBOL(memset_extent_buffer);
2665 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
2666 unsigned long dst_offset, unsigned long src_offset,
2669 u64 dst_len = dst->len;
2674 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
2675 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
2677 WARN_ON(src->len != dst_len);
2679 offset = (start_offset + dst_offset) &
2680 ((unsigned long)PAGE_CACHE_SIZE - 1);
2683 page = extent_buffer_page(dst, i);
2684 WARN_ON(!PageUptodate(page));
2686 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
2688 kaddr = kmap_atomic(page, KM_USER0);
2689 read_extent_buffer(src, kaddr + offset, src_offset, cur);
2690 kunmap_atomic(kaddr, KM_USER0);
2698 EXPORT_SYMBOL(copy_extent_buffer);
2700 static void move_pages(struct page *dst_page, struct page *src_page,
2701 unsigned long dst_off, unsigned long src_off,
2704 char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
2705 if (dst_page == src_page) {
2706 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
2708 char *src_kaddr = kmap_atomic(src_page, KM_USER1);
2709 char *p = dst_kaddr + dst_off + len;
2710 char *s = src_kaddr + src_off + len;
2715 kunmap_atomic(src_kaddr, KM_USER1);
2717 kunmap_atomic(dst_kaddr, KM_USER0);
2720 static void copy_pages(struct page *dst_page, struct page *src_page,
2721 unsigned long dst_off, unsigned long src_off,
2724 char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
2727 if (dst_page != src_page)
2728 src_kaddr = kmap_atomic(src_page, KM_USER1);
2730 src_kaddr = dst_kaddr;
2732 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
2733 kunmap_atomic(dst_kaddr, KM_USER0);
2734 if (dst_page != src_page)
2735 kunmap_atomic(src_kaddr, KM_USER1);
2738 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
2739 unsigned long src_offset, unsigned long len)
2742 size_t dst_off_in_page;
2743 size_t src_off_in_page;
2744 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
2745 unsigned long dst_i;
2746 unsigned long src_i;
2748 if (src_offset + len > dst->len) {
2749 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
2750 src_offset, len, dst->len);
2753 if (dst_offset + len > dst->len) {
2754 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
2755 dst_offset, len, dst->len);
2760 dst_off_in_page = (start_offset + dst_offset) &
2761 ((unsigned long)PAGE_CACHE_SIZE - 1);
2762 src_off_in_page = (start_offset + src_offset) &
2763 ((unsigned long)PAGE_CACHE_SIZE - 1);
2765 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
2766 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
2768 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
2770 cur = min_t(unsigned long, cur,
2771 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
2773 copy_pages(extent_buffer_page(dst, dst_i),
2774 extent_buffer_page(dst, src_i),
2775 dst_off_in_page, src_off_in_page, cur);
2782 EXPORT_SYMBOL(memcpy_extent_buffer);
2784 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
2785 unsigned long src_offset, unsigned long len)
2788 size_t dst_off_in_page;
2789 size_t src_off_in_page;
2790 unsigned long dst_end = dst_offset + len - 1;
2791 unsigned long src_end = src_offset + len - 1;
2792 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
2793 unsigned long dst_i;
2794 unsigned long src_i;
2796 if (src_offset + len > dst->len) {
2797 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
2798 src_offset, len, dst->len);
2801 if (dst_offset + len > dst->len) {
2802 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
2803 dst_offset, len, dst->len);
2806 if (dst_offset < src_offset) {
2807 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
2811 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
2812 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
2814 dst_off_in_page = (start_offset + dst_end) &
2815 ((unsigned long)PAGE_CACHE_SIZE - 1);
2816 src_off_in_page = (start_offset + src_end) &
2817 ((unsigned long)PAGE_CACHE_SIZE - 1);
2819 cur = min_t(unsigned long, len, src_off_in_page + 1);
2820 cur = min(cur, dst_off_in_page + 1);
2821 move_pages(extent_buffer_page(dst, dst_i),
2822 extent_buffer_page(dst, src_i),
2823 dst_off_in_page - cur + 1,
2824 src_off_in_page - cur + 1, cur);
2831 EXPORT_SYMBOL(memmove_extent_buffer);