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 <linux/pagevec.h>
15 #include "extent_map.h"
17 /* temporary define until extent_map moves out of btrfs */
18 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
19 unsigned long extra_flags,
20 void (*ctor)(void *, struct kmem_cache *,
23 static struct kmem_cache *extent_map_cache;
24 static struct kmem_cache *extent_state_cache;
25 static struct kmem_cache *extent_buffer_cache;
27 static LIST_HEAD(buffers);
28 static LIST_HEAD(states);
30 static spinlock_t state_lock = SPIN_LOCK_UNLOCKED;
31 #define BUFFER_LRU_MAX 64
37 struct rb_node rb_node;
40 struct extent_page_data {
42 struct extent_map_tree *tree;
43 get_extent_t *get_extent;
46 int __init extent_map_init(void)
48 extent_map_cache = btrfs_cache_create("extent_map",
49 sizeof(struct extent_map), 0,
51 if (!extent_map_cache)
53 extent_state_cache = btrfs_cache_create("extent_state",
54 sizeof(struct extent_state), 0,
56 if (!extent_state_cache)
58 extent_buffer_cache = btrfs_cache_create("extent_buffers",
59 sizeof(struct extent_buffer), 0,
61 if (!extent_buffer_cache)
62 goto free_state_cache;
66 kmem_cache_destroy(extent_state_cache);
68 kmem_cache_destroy(extent_map_cache);
72 void extent_map_exit(void)
74 struct extent_state *state;
76 while (!list_empty(&states)) {
77 state = list_entry(states.next, struct extent_state, list);
78 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));
79 list_del(&state->list);
80 kmem_cache_free(extent_state_cache, state);
85 kmem_cache_destroy(extent_map_cache);
86 if (extent_state_cache)
87 kmem_cache_destroy(extent_state_cache);
88 if (extent_buffer_cache)
89 kmem_cache_destroy(extent_buffer_cache);
92 void extent_map_tree_init(struct extent_map_tree *tree,
93 struct address_space *mapping, gfp_t mask)
95 tree->map.rb_node = NULL;
96 tree->state.rb_node = NULL;
98 tree->dirty_bytes = 0;
99 rwlock_init(&tree->lock);
100 spin_lock_init(&tree->lru_lock);
101 tree->mapping = mapping;
102 INIT_LIST_HEAD(&tree->buffer_lru);
105 EXPORT_SYMBOL(extent_map_tree_init);
107 void extent_map_tree_empty_lru(struct extent_map_tree *tree)
109 struct extent_buffer *eb;
110 while(!list_empty(&tree->buffer_lru)) {
111 eb = list_entry(tree->buffer_lru.next, struct extent_buffer,
113 list_del_init(&eb->lru);
114 free_extent_buffer(eb);
117 EXPORT_SYMBOL(extent_map_tree_empty_lru);
119 struct extent_map *alloc_extent_map(gfp_t mask)
121 struct extent_map *em;
122 em = kmem_cache_alloc(extent_map_cache, mask);
123 if (!em || IS_ERR(em))
126 atomic_set(&em->refs, 1);
129 EXPORT_SYMBOL(alloc_extent_map);
131 void free_extent_map(struct extent_map *em)
135 if (atomic_dec_and_test(&em->refs)) {
136 WARN_ON(em->in_tree);
137 kmem_cache_free(extent_map_cache, em);
140 EXPORT_SYMBOL(free_extent_map);
143 struct extent_state *alloc_extent_state(gfp_t mask)
145 struct extent_state *state;
148 state = kmem_cache_alloc(extent_state_cache, mask);
149 if (!state || IS_ERR(state))
155 spin_lock_irqsave(&state_lock, flags);
156 list_add(&state->list, &states);
157 spin_unlock_irqrestore(&state_lock, flags);
159 atomic_set(&state->refs, 1);
160 init_waitqueue_head(&state->wq);
163 EXPORT_SYMBOL(alloc_extent_state);
165 void free_extent_state(struct extent_state *state)
170 if (atomic_dec_and_test(&state->refs)) {
171 WARN_ON(state->in_tree);
172 spin_lock_irqsave(&state_lock, flags);
173 list_del(&state->list);
174 spin_unlock_irqrestore(&state_lock, flags);
175 kmem_cache_free(extent_state_cache, state);
178 EXPORT_SYMBOL(free_extent_state);
180 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
181 struct rb_node *node)
183 struct rb_node ** p = &root->rb_node;
184 struct rb_node * parent = NULL;
185 struct tree_entry *entry;
189 entry = rb_entry(parent, struct tree_entry, rb_node);
191 if (offset < entry->start)
193 else if (offset > entry->end)
199 entry = rb_entry(node, struct tree_entry, rb_node);
201 rb_link_node(node, parent, p);
202 rb_insert_color(node, root);
206 static struct rb_node *__tree_search(struct rb_root *root, u64 offset,
207 struct rb_node **prev_ret)
209 struct rb_node * n = root->rb_node;
210 struct rb_node *prev = NULL;
211 struct tree_entry *entry;
212 struct tree_entry *prev_entry = NULL;
215 entry = rb_entry(n, struct tree_entry, rb_node);
219 if (offset < entry->start)
221 else if (offset > entry->end)
228 while(prev && offset > prev_entry->end) {
229 prev = rb_next(prev);
230 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
236 static inline struct rb_node *tree_search(struct rb_root *root, u64 offset)
238 struct rb_node *prev;
240 ret = __tree_search(root, offset, &prev);
246 static int tree_delete(struct rb_root *root, u64 offset)
248 struct rb_node *node;
249 struct tree_entry *entry;
251 node = __tree_search(root, offset, NULL);
254 entry = rb_entry(node, struct tree_entry, rb_node);
256 rb_erase(node, root);
261 * add_extent_mapping tries a simple backward merge with existing
262 * mappings. The extent_map struct passed in will be inserted into
263 * the tree directly (no copies made, just a reference taken).
265 int add_extent_mapping(struct extent_map_tree *tree,
266 struct extent_map *em)
269 struct extent_map *prev = NULL;
272 write_lock_irq(&tree->lock);
273 rb = tree_insert(&tree->map, em->end, &em->rb_node);
275 prev = rb_entry(rb, struct extent_map, rb_node);
276 printk("found extent map %Lu %Lu on insert of %Lu %Lu\n", prev->start, prev->end, em->start, em->end);
280 atomic_inc(&em->refs);
281 if (em->start != 0) {
282 rb = rb_prev(&em->rb_node);
284 prev = rb_entry(rb, struct extent_map, rb_node);
285 if (prev && prev->end + 1 == em->start &&
286 ((em->block_start == EXTENT_MAP_HOLE &&
287 prev->block_start == EXTENT_MAP_HOLE) ||
288 (em->block_start == EXTENT_MAP_INLINE &&
289 prev->block_start == EXTENT_MAP_INLINE) ||
290 (em->block_start == EXTENT_MAP_DELALLOC &&
291 prev->block_start == EXTENT_MAP_DELALLOC) ||
292 (em->block_start < EXTENT_MAP_DELALLOC - 1 &&
293 em->block_start == prev->block_end + 1))) {
294 em->start = prev->start;
295 em->block_start = prev->block_start;
296 rb_erase(&prev->rb_node, &tree->map);
298 free_extent_map(prev);
302 write_unlock_irq(&tree->lock);
305 EXPORT_SYMBOL(add_extent_mapping);
308 * lookup_extent_mapping returns the first extent_map struct in the
309 * tree that intersects the [start, end] (inclusive) range. There may
310 * be additional objects in the tree that intersect, so check the object
311 * returned carefully to make sure you don't need additional lookups.
313 struct extent_map *lookup_extent_mapping(struct extent_map_tree *tree,
316 struct extent_map *em;
317 struct rb_node *rb_node;
319 read_lock_irq(&tree->lock);
320 rb_node = tree_search(&tree->map, start);
325 if (IS_ERR(rb_node)) {
326 em = ERR_PTR(PTR_ERR(rb_node));
329 em = rb_entry(rb_node, struct extent_map, rb_node);
330 if (em->end < start || em->start > end) {
334 atomic_inc(&em->refs);
336 read_unlock_irq(&tree->lock);
339 EXPORT_SYMBOL(lookup_extent_mapping);
342 * removes an extent_map struct from the tree. No reference counts are
343 * dropped, and no checks are done to see if the range is in use
345 int remove_extent_mapping(struct extent_map_tree *tree, struct extent_map *em)
349 write_lock_irq(&tree->lock);
350 ret = tree_delete(&tree->map, em->end);
351 write_unlock_irq(&tree->lock);
354 EXPORT_SYMBOL(remove_extent_mapping);
357 * utility function to look for merge candidates inside a given range.
358 * Any extents with matching state are merged together into a single
359 * extent in the tree. Extents with EXTENT_IO in their state field
360 * are not merged because the end_io handlers need to be able to do
361 * operations on them without sleeping (or doing allocations/splits).
363 * This should be called with the tree lock held.
365 static int merge_state(struct extent_map_tree *tree,
366 struct extent_state *state)
368 struct extent_state *other;
369 struct rb_node *other_node;
371 if (state->state & EXTENT_IOBITS)
374 other_node = rb_prev(&state->rb_node);
376 other = rb_entry(other_node, struct extent_state, rb_node);
377 if (other->end == state->start - 1 &&
378 other->state == state->state) {
379 state->start = other->start;
381 rb_erase(&other->rb_node, &tree->state);
382 free_extent_state(other);
385 other_node = rb_next(&state->rb_node);
387 other = rb_entry(other_node, struct extent_state, rb_node);
388 if (other->start == state->end + 1 &&
389 other->state == state->state) {
390 other->start = state->start;
392 rb_erase(&state->rb_node, &tree->state);
393 free_extent_state(state);
400 * insert an extent_state struct into the tree. 'bits' are set on the
401 * struct before it is inserted.
403 * This may return -EEXIST if the extent is already there, in which case the
404 * state struct is freed.
406 * The tree lock is not taken internally. This is a utility function and
407 * probably isn't what you want to call (see set/clear_extent_bit).
409 static int insert_state(struct extent_map_tree *tree,
410 struct extent_state *state, u64 start, u64 end,
413 struct rb_node *node;
416 printk("end < start %Lu %Lu\n", end, start);
419 if (bits & EXTENT_DIRTY)
420 tree->dirty_bytes += end - start + 1;
421 state->state |= bits;
422 state->start = start;
424 node = tree_insert(&tree->state, end, &state->rb_node);
426 struct extent_state *found;
427 found = rb_entry(node, struct extent_state, rb_node);
428 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, start, end);
429 free_extent_state(state);
432 merge_state(tree, state);
437 * split a given extent state struct in two, inserting the preallocated
438 * struct 'prealloc' as the newly created second half. 'split' indicates an
439 * offset inside 'orig' where it should be split.
442 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
443 * are two extent state structs in the tree:
444 * prealloc: [orig->start, split - 1]
445 * orig: [ split, orig->end ]
447 * The tree locks are not taken by this function. They need to be held
450 static int split_state(struct extent_map_tree *tree, struct extent_state *orig,
451 struct extent_state *prealloc, u64 split)
453 struct rb_node *node;
454 prealloc->start = orig->start;
455 prealloc->end = split - 1;
456 prealloc->state = orig->state;
459 node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
461 struct extent_state *found;
462 found = rb_entry(node, struct extent_state, rb_node);
463 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, prealloc->start, prealloc->end);
464 free_extent_state(prealloc);
471 * utility function to clear some bits in an extent state struct.
472 * it will optionally wake up any one waiting on this state (wake == 1), or
473 * forcibly remove the state from the tree (delete == 1).
475 * If no bits are set on the state struct after clearing things, the
476 * struct is freed and removed from the tree
478 static int clear_state_bit(struct extent_map_tree *tree,
479 struct extent_state *state, int bits, int wake,
482 int ret = state->state & bits;
484 if ((bits & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
485 u64 range = state->end - state->start + 1;
486 WARN_ON(range > tree->dirty_bytes);
487 tree->dirty_bytes -= range;
489 state->state &= ~bits;
492 if (delete || state->state == 0) {
493 if (state->in_tree) {
494 rb_erase(&state->rb_node, &tree->state);
496 free_extent_state(state);
501 merge_state(tree, state);
507 * clear some bits on a range in the tree. This may require splitting
508 * or inserting elements in the tree, so the gfp mask is used to
509 * indicate which allocations or sleeping are allowed.
511 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
512 * the given range from the tree regardless of state (ie for truncate).
514 * the range [start, end] is inclusive.
516 * This takes the tree lock, and returns < 0 on error, > 0 if any of the
517 * bits were already set, or zero if none of the bits were already set.
519 int clear_extent_bit(struct extent_map_tree *tree, u64 start, u64 end,
520 int bits, int wake, int delete, gfp_t mask)
522 struct extent_state *state;
523 struct extent_state *prealloc = NULL;
524 struct rb_node *node;
530 if (!prealloc && (mask & __GFP_WAIT)) {
531 prealloc = alloc_extent_state(mask);
536 write_lock_irqsave(&tree->lock, flags);
538 * this search will find the extents that end after
541 node = tree_search(&tree->state, start);
544 state = rb_entry(node, struct extent_state, rb_node);
545 if (state->start > end)
547 WARN_ON(state->end < start);
550 * | ---- desired range ---- |
552 * | ------------- state -------------- |
554 * We need to split the extent we found, and may flip
555 * bits on second half.
557 * If the extent we found extends past our range, we
558 * just split and search again. It'll get split again
559 * the next time though.
561 * If the extent we found is inside our range, we clear
562 * the desired bit on it.
565 if (state->start < start) {
566 err = split_state(tree, state, prealloc, start);
567 BUG_ON(err == -EEXIST);
571 if (state->end <= end) {
572 start = state->end + 1;
573 set |= clear_state_bit(tree, state, bits,
576 start = state->start;
581 * | ---- desired range ---- |
583 * We need to split the extent, and clear the bit
586 if (state->start <= end && state->end > end) {
587 err = split_state(tree, state, prealloc, end + 1);
588 BUG_ON(err == -EEXIST);
592 set |= clear_state_bit(tree, prealloc, bits,
598 start = state->end + 1;
599 set |= clear_state_bit(tree, state, bits, wake, delete);
603 write_unlock_irqrestore(&tree->lock, flags);
605 free_extent_state(prealloc);
612 write_unlock_irqrestore(&tree->lock, flags);
613 if (mask & __GFP_WAIT)
617 EXPORT_SYMBOL(clear_extent_bit);
619 static int wait_on_state(struct extent_map_tree *tree,
620 struct extent_state *state)
623 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
624 read_unlock_irq(&tree->lock);
626 read_lock_irq(&tree->lock);
627 finish_wait(&state->wq, &wait);
632 * waits for one or more bits to clear on a range in the state tree.
633 * The range [start, end] is inclusive.
634 * The tree lock is taken by this function
636 int wait_extent_bit(struct extent_map_tree *tree, u64 start, u64 end, int bits)
638 struct extent_state *state;
639 struct rb_node *node;
641 read_lock_irq(&tree->lock);
645 * this search will find all the extents that end after
648 node = tree_search(&tree->state, start);
652 state = rb_entry(node, struct extent_state, rb_node);
654 if (state->start > end)
657 if (state->state & bits) {
658 start = state->start;
659 atomic_inc(&state->refs);
660 wait_on_state(tree, state);
661 free_extent_state(state);
664 start = state->end + 1;
669 if (need_resched()) {
670 read_unlock_irq(&tree->lock);
672 read_lock_irq(&tree->lock);
676 read_unlock_irq(&tree->lock);
679 EXPORT_SYMBOL(wait_extent_bit);
681 static void set_state_bits(struct extent_map_tree *tree,
682 struct extent_state *state,
685 if ((bits & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
686 u64 range = state->end - state->start + 1;
687 tree->dirty_bytes += range;
689 state->state |= bits;
693 * set some bits on a range in the tree. This may require allocations
694 * or sleeping, so the gfp mask is used to indicate what is allowed.
696 * If 'exclusive' == 1, this will fail with -EEXIST if some part of the
697 * range already has the desired bits set. The start of the existing
698 * range is returned in failed_start in this case.
700 * [start, end] is inclusive
701 * This takes the tree lock.
703 int set_extent_bit(struct extent_map_tree *tree, u64 start, u64 end, int bits,
704 int exclusive, u64 *failed_start, gfp_t mask)
706 struct extent_state *state;
707 struct extent_state *prealloc = NULL;
708 struct rb_node *node;
715 if (!prealloc && (mask & __GFP_WAIT)) {
716 prealloc = alloc_extent_state(mask);
721 write_lock_irqsave(&tree->lock, flags);
723 * this search will find all the extents that end after
726 node = tree_search(&tree->state, start);
728 err = insert_state(tree, prealloc, start, end, bits);
730 BUG_ON(err == -EEXIST);
734 state = rb_entry(node, struct extent_state, rb_node);
735 last_start = state->start;
736 last_end = state->end;
739 * | ---- desired range ---- |
742 * Just lock what we found and keep going
744 if (state->start == start && state->end <= end) {
745 set = state->state & bits;
746 if (set && exclusive) {
747 *failed_start = state->start;
751 set_state_bits(tree, state, bits);
752 start = state->end + 1;
753 merge_state(tree, state);
758 * | ---- desired range ---- |
761 * | ------------- state -------------- |
763 * We need to split the extent we found, and may flip bits on
766 * If the extent we found extends past our
767 * range, we just split and search again. It'll get split
768 * again the next time though.
770 * If the extent we found is inside our range, we set the
773 if (state->start < start) {
774 set = state->state & bits;
775 if (exclusive && set) {
776 *failed_start = start;
780 err = split_state(tree, state, prealloc, start);
781 BUG_ON(err == -EEXIST);
785 if (state->end <= end) {
786 set_state_bits(tree, state, bits);
787 start = state->end + 1;
788 merge_state(tree, state);
790 start = state->start;
795 * | ---- desired range ---- |
796 * | state | or | state |
798 * There's a hole, we need to insert something in it and
799 * ignore the extent we found.
801 if (state->start > start) {
803 if (end < last_start)
806 this_end = last_start -1;
807 err = insert_state(tree, prealloc, start, this_end,
810 BUG_ON(err == -EEXIST);
813 start = this_end + 1;
817 * | ---- desired range ---- |
819 * We need to split the extent, and set the bit
822 if (state->start <= end && state->end > end) {
823 set = state->state & bits;
824 if (exclusive && set) {
825 *failed_start = start;
829 err = split_state(tree, state, prealloc, end + 1);
830 BUG_ON(err == -EEXIST);
832 set_state_bits(tree, prealloc, bits);
833 merge_state(tree, prealloc);
841 write_unlock_irqrestore(&tree->lock, flags);
843 free_extent_state(prealloc);
850 write_unlock_irqrestore(&tree->lock, flags);
851 if (mask & __GFP_WAIT)
855 EXPORT_SYMBOL(set_extent_bit);
857 /* wrappers around set/clear extent bit */
858 int set_extent_dirty(struct extent_map_tree *tree, u64 start, u64 end,
861 return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL,
864 EXPORT_SYMBOL(set_extent_dirty);
866 int set_extent_bits(struct extent_map_tree *tree, u64 start, u64 end,
867 int bits, gfp_t mask)
869 return set_extent_bit(tree, start, end, bits, 0, NULL,
872 EXPORT_SYMBOL(set_extent_bits);
874 int clear_extent_bits(struct extent_map_tree *tree, u64 start, u64 end,
875 int bits, gfp_t mask)
877 return clear_extent_bit(tree, start, end, bits, 0, 0, mask);
879 EXPORT_SYMBOL(clear_extent_bits);
881 int set_extent_delalloc(struct extent_map_tree *tree, u64 start, u64 end,
884 return set_extent_bit(tree, start, end,
885 EXTENT_DELALLOC | EXTENT_DIRTY, 0, NULL,
888 EXPORT_SYMBOL(set_extent_delalloc);
890 int clear_extent_dirty(struct extent_map_tree *tree, u64 start, u64 end,
893 return clear_extent_bit(tree, start, end,
894 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, mask);
896 EXPORT_SYMBOL(clear_extent_dirty);
898 int set_extent_new(struct extent_map_tree *tree, u64 start, u64 end,
901 return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL,
904 EXPORT_SYMBOL(set_extent_new);
906 int clear_extent_new(struct extent_map_tree *tree, u64 start, u64 end,
909 return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0, mask);
911 EXPORT_SYMBOL(clear_extent_new);
913 int set_extent_uptodate(struct extent_map_tree *tree, u64 start, u64 end,
916 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, NULL,
919 EXPORT_SYMBOL(set_extent_uptodate);
921 int clear_extent_uptodate(struct extent_map_tree *tree, u64 start, u64 end,
924 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, mask);
926 EXPORT_SYMBOL(clear_extent_uptodate);
928 int set_extent_writeback(struct extent_map_tree *tree, u64 start, u64 end,
931 return set_extent_bit(tree, start, end, EXTENT_WRITEBACK,
934 EXPORT_SYMBOL(set_extent_writeback);
936 int clear_extent_writeback(struct extent_map_tree *tree, u64 start, u64 end,
939 return clear_extent_bit(tree, start, end, EXTENT_WRITEBACK, 1, 0, mask);
941 EXPORT_SYMBOL(clear_extent_writeback);
943 int wait_on_extent_writeback(struct extent_map_tree *tree, u64 start, u64 end)
945 return wait_extent_bit(tree, start, end, EXTENT_WRITEBACK);
947 EXPORT_SYMBOL(wait_on_extent_writeback);
950 * locks a range in ascending order, waiting for any locked regions
951 * it hits on the way. [start,end] are inclusive, and this will sleep.
953 int lock_extent(struct extent_map_tree *tree, u64 start, u64 end, gfp_t mask)
958 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1,
959 &failed_start, mask);
960 if (err == -EEXIST && (mask & __GFP_WAIT)) {
961 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
962 start = failed_start;
966 WARN_ON(start > end);
970 EXPORT_SYMBOL(lock_extent);
972 int unlock_extent(struct extent_map_tree *tree, u64 start, u64 end,
975 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, mask);
977 EXPORT_SYMBOL(unlock_extent);
980 * helper function to set pages and extents in the tree dirty
982 int set_range_dirty(struct extent_map_tree *tree, u64 start, u64 end)
984 unsigned long index = start >> PAGE_CACHE_SHIFT;
985 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
988 while (index <= end_index) {
989 page = find_get_page(tree->mapping, index);
991 __set_page_dirty_nobuffers(page);
992 page_cache_release(page);
995 set_extent_dirty(tree, start, end, GFP_NOFS);
998 EXPORT_SYMBOL(set_range_dirty);
1001 * helper function to set both pages and extents in the tree writeback
1003 int set_range_writeback(struct extent_map_tree *tree, u64 start, u64 end)
1005 unsigned long index = start >> PAGE_CACHE_SHIFT;
1006 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1009 while (index <= end_index) {
1010 page = find_get_page(tree->mapping, index);
1012 set_page_writeback(page);
1013 page_cache_release(page);
1016 set_extent_writeback(tree, start, end, GFP_NOFS);
1019 EXPORT_SYMBOL(set_range_writeback);
1021 int find_first_extent_bit(struct extent_map_tree *tree, u64 start,
1022 u64 *start_ret, u64 *end_ret, int bits)
1024 struct rb_node *node;
1025 struct extent_state *state;
1028 read_lock_irq(&tree->lock);
1030 * this search will find all the extents that end after
1033 node = tree_search(&tree->state, start);
1034 if (!node || IS_ERR(node)) {
1039 state = rb_entry(node, struct extent_state, rb_node);
1040 if (state->end >= start && (state->state & bits)) {
1041 *start_ret = state->start;
1042 *end_ret = state->end;
1046 node = rb_next(node);
1051 read_unlock_irq(&tree->lock);
1054 EXPORT_SYMBOL(find_first_extent_bit);
1056 u64 find_lock_delalloc_range(struct extent_map_tree *tree,
1057 u64 *start, u64 *end, u64 max_bytes)
1059 struct rb_node *node;
1060 struct extent_state *state;
1061 u64 cur_start = *start;
1063 u64 total_bytes = 0;
1065 write_lock_irq(&tree->lock);
1067 * this search will find all the extents that end after
1071 node = tree_search(&tree->state, cur_start);
1072 if (!node || IS_ERR(node)) {
1078 state = rb_entry(node, struct extent_state, rb_node);
1079 if (found && state->start != cur_start) {
1082 if (!(state->state & EXTENT_DELALLOC)) {
1088 struct extent_state *prev_state;
1089 struct rb_node *prev_node = node;
1091 prev_node = rb_prev(prev_node);
1094 prev_state = rb_entry(prev_node,
1095 struct extent_state,
1097 if (!(prev_state->state & EXTENT_DELALLOC))
1103 if (state->state & EXTENT_LOCKED) {
1105 atomic_inc(&state->refs);
1106 prepare_to_wait(&state->wq, &wait,
1107 TASK_UNINTERRUPTIBLE);
1108 write_unlock_irq(&tree->lock);
1110 write_lock_irq(&tree->lock);
1111 finish_wait(&state->wq, &wait);
1112 free_extent_state(state);
1115 state->state |= EXTENT_LOCKED;
1117 *start = state->start;
1120 cur_start = state->end + 1;
1121 node = rb_next(node);
1124 total_bytes += state->end - state->start + 1;
1125 if (total_bytes >= max_bytes)
1129 write_unlock_irq(&tree->lock);
1133 u64 count_range_bits(struct extent_map_tree *tree,
1134 u64 *start, u64 search_end, u64 max_bytes,
1137 struct rb_node *node;
1138 struct extent_state *state;
1139 u64 cur_start = *start;
1140 u64 total_bytes = 0;
1143 if (search_end <= cur_start) {
1144 printk("search_end %Lu start %Lu\n", search_end, cur_start);
1149 write_lock_irq(&tree->lock);
1150 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1151 total_bytes = tree->dirty_bytes;
1155 * this search will find all the extents that end after
1158 node = tree_search(&tree->state, cur_start);
1159 if (!node || IS_ERR(node)) {
1164 state = rb_entry(node, struct extent_state, rb_node);
1165 if (state->start > search_end)
1167 if (state->end >= cur_start && (state->state & bits)) {
1168 total_bytes += min(search_end, state->end) + 1 -
1169 max(cur_start, state->start);
1170 if (total_bytes >= max_bytes)
1173 *start = state->start;
1177 node = rb_next(node);
1182 write_unlock_irq(&tree->lock);
1186 * helper function to lock both pages and extents in the tree.
1187 * pages must be locked first.
1189 int lock_range(struct extent_map_tree *tree, u64 start, u64 end)
1191 unsigned long index = start >> PAGE_CACHE_SHIFT;
1192 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1196 while (index <= end_index) {
1197 page = grab_cache_page(tree->mapping, index);
1203 err = PTR_ERR(page);
1208 lock_extent(tree, start, end, GFP_NOFS);
1213 * we failed above in getting the page at 'index', so we undo here
1214 * up to but not including the page at 'index'
1217 index = start >> PAGE_CACHE_SHIFT;
1218 while (index < end_index) {
1219 page = find_get_page(tree->mapping, index);
1221 page_cache_release(page);
1226 EXPORT_SYMBOL(lock_range);
1229 * helper function to unlock both pages and extents in the tree.
1231 int unlock_range(struct extent_map_tree *tree, u64 start, u64 end)
1233 unsigned long index = start >> PAGE_CACHE_SHIFT;
1234 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1237 while (index <= end_index) {
1238 page = find_get_page(tree->mapping, index);
1240 page_cache_release(page);
1243 unlock_extent(tree, start, end, GFP_NOFS);
1246 EXPORT_SYMBOL(unlock_range);
1248 int set_state_private(struct extent_map_tree *tree, u64 start, u64 private)
1250 struct rb_node *node;
1251 struct extent_state *state;
1254 write_lock_irq(&tree->lock);
1256 * this search will find all the extents that end after
1259 node = tree_search(&tree->state, start);
1260 if (!node || IS_ERR(node)) {
1264 state = rb_entry(node, struct extent_state, rb_node);
1265 if (state->start != start) {
1269 state->private = private;
1271 write_unlock_irq(&tree->lock);
1275 int get_state_private(struct extent_map_tree *tree, u64 start, u64 *private)
1277 struct rb_node *node;
1278 struct extent_state *state;
1281 read_lock_irq(&tree->lock);
1283 * this search will find all the extents that end after
1286 node = tree_search(&tree->state, start);
1287 if (!node || IS_ERR(node)) {
1291 state = rb_entry(node, struct extent_state, rb_node);
1292 if (state->start != start) {
1296 *private = state->private;
1298 read_unlock_irq(&tree->lock);
1303 * searches a range in the state tree for a given mask.
1304 * If 'filled' == 1, this returns 1 only if ever extent in the tree
1305 * has the bits set. Otherwise, 1 is returned if any bit in the
1306 * range is found set.
1308 int test_range_bit(struct extent_map_tree *tree, u64 start, u64 end,
1309 int bits, int filled)
1311 struct extent_state *state = NULL;
1312 struct rb_node *node;
1315 read_lock_irq(&tree->lock);
1316 node = tree_search(&tree->state, start);
1317 while (node && start <= end) {
1318 state = rb_entry(node, struct extent_state, rb_node);
1320 if (filled && state->start > start) {
1325 if (state->start > end)
1328 if (state->state & bits) {
1332 } else if (filled) {
1336 start = state->end + 1;
1339 node = rb_next(node);
1341 read_unlock_irq(&tree->lock);
1344 EXPORT_SYMBOL(test_range_bit);
1347 * helper function to set a given page up to date if all the
1348 * extents in the tree for that page are up to date
1350 static int check_page_uptodate(struct extent_map_tree *tree,
1353 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1354 u64 end = start + PAGE_CACHE_SIZE - 1;
1355 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1))
1356 SetPageUptodate(page);
1361 * helper function to unlock a page if all the extents in the tree
1362 * for that page are unlocked
1364 static int check_page_locked(struct extent_map_tree *tree,
1367 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1368 u64 end = start + PAGE_CACHE_SIZE - 1;
1369 if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0))
1375 * helper function to end page writeback if all the extents
1376 * in the tree for that page are done with writeback
1378 static int check_page_writeback(struct extent_map_tree *tree,
1381 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1382 u64 end = start + PAGE_CACHE_SIZE - 1;
1383 if (!test_range_bit(tree, start, end, EXTENT_WRITEBACK, 0))
1384 end_page_writeback(page);
1388 /* lots and lots of room for performance fixes in the end_bio funcs */
1391 * after a writepage IO is done, we need to:
1392 * clear the uptodate bits on error
1393 * clear the writeback bits in the extent tree for this IO
1394 * end_page_writeback if the page has no more pending IO
1396 * Scheduling is not allowed, so the extent state tree is expected
1397 * to have one and only one object corresponding to this IO.
1399 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1400 static void end_bio_extent_writepage(struct bio *bio, int err)
1402 static int end_bio_extent_writepage(struct bio *bio,
1403 unsigned int bytes_done, int err)
1406 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1407 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1408 struct extent_map_tree *tree = bio->bi_private;
1413 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1419 struct page *page = bvec->bv_page;
1420 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1422 end = start + bvec->bv_len - 1;
1424 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1429 if (--bvec >= bio->bi_io_vec)
1430 prefetchw(&bvec->bv_page->flags);
1433 clear_extent_uptodate(tree, start, end, GFP_ATOMIC);
1434 ClearPageUptodate(page);
1437 clear_extent_writeback(tree, start, end, GFP_ATOMIC);
1440 end_page_writeback(page);
1442 check_page_writeback(tree, page);
1443 if (tree->ops && tree->ops->writepage_end_io_hook)
1444 tree->ops->writepage_end_io_hook(page, start, end);
1445 } while (bvec >= bio->bi_io_vec);
1448 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1454 * after a readpage IO is done, we need to:
1455 * clear the uptodate bits on error
1456 * set the uptodate bits if things worked
1457 * set the page up to date if all extents in the tree are uptodate
1458 * clear the lock bit in the extent tree
1459 * unlock the page if there are no other extents locked for it
1461 * Scheduling is not allowed, so the extent state tree is expected
1462 * to have one and only one object corresponding to this IO.
1464 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1465 static void end_bio_extent_readpage(struct bio *bio, int err)
1467 static int end_bio_extent_readpage(struct bio *bio,
1468 unsigned int bytes_done, int err)
1471 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1472 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1473 struct extent_map_tree *tree = bio->bi_private;
1479 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1485 struct page *page = bvec->bv_page;
1486 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1488 end = start + bvec->bv_len - 1;
1490 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1495 if (--bvec >= bio->bi_io_vec)
1496 prefetchw(&bvec->bv_page->flags);
1498 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
1499 ret = tree->ops->readpage_end_io_hook(page, start, end);
1504 set_extent_uptodate(tree, start, end, GFP_ATOMIC);
1506 SetPageUptodate(page);
1508 check_page_uptodate(tree, page);
1510 ClearPageUptodate(page);
1514 unlock_extent(tree, start, end, GFP_ATOMIC);
1519 check_page_locked(tree, page);
1520 } while (bvec >= bio->bi_io_vec);
1523 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1529 * IO done from prepare_write is pretty simple, we just unlock
1530 * the structs in the extent tree when done, and set the uptodate bits
1533 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1534 static void end_bio_extent_preparewrite(struct bio *bio, int err)
1536 static int end_bio_extent_preparewrite(struct bio *bio,
1537 unsigned int bytes_done, int err)
1540 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1541 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1542 struct extent_map_tree *tree = bio->bi_private;
1546 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1552 struct page *page = bvec->bv_page;
1553 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1555 end = start + bvec->bv_len - 1;
1557 if (--bvec >= bio->bi_io_vec)
1558 prefetchw(&bvec->bv_page->flags);
1561 set_extent_uptodate(tree, start, end, GFP_ATOMIC);
1563 ClearPageUptodate(page);
1567 unlock_extent(tree, start, end, GFP_ATOMIC);
1569 } while (bvec >= bio->bi_io_vec);
1572 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1578 extent_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
1583 bio = bio_alloc(gfp_flags, nr_vecs);
1585 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
1586 while (!bio && (nr_vecs /= 2))
1587 bio = bio_alloc(gfp_flags, nr_vecs);
1591 bio->bi_bdev = bdev;
1592 bio->bi_sector = first_sector;
1597 static int submit_one_bio(int rw, struct bio *bio)
1604 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
1605 if (maxsector < bio->bi_sector) {
1606 printk("sector too large max %Lu got %llu\n", maxsector,
1607 (unsigned long long)bio->bi_sector);
1611 submit_bio(rw, bio);
1612 if (bio_flagged(bio, BIO_EOPNOTSUPP))
1618 static int submit_extent_page(int rw, struct extent_map_tree *tree,
1619 struct page *page, sector_t sector,
1620 size_t size, unsigned long offset,
1621 struct block_device *bdev,
1622 struct bio **bio_ret,
1623 unsigned long max_pages,
1624 bio_end_io_t end_io_func)
1630 if (bio_ret && *bio_ret) {
1632 if (bio->bi_sector + (bio->bi_size >> 9) != sector ||
1633 bio_add_page(bio, page, size, offset) < size) {
1634 ret = submit_one_bio(rw, bio);
1640 nr = min_t(int, max_pages, bio_get_nr_vecs(bdev));
1641 bio = extent_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
1643 printk("failed to allocate bio nr %d\n", nr);
1645 bio_add_page(bio, page, size, offset);
1646 bio->bi_end_io = end_io_func;
1647 bio->bi_private = tree;
1651 ret = submit_one_bio(rw, bio);
1657 void set_page_extent_mapped(struct page *page)
1659 if (!PagePrivate(page)) {
1660 SetPagePrivate(page);
1661 WARN_ON(!page->mapping->a_ops->invalidatepage);
1662 set_page_private(page, EXTENT_PAGE_PRIVATE);
1663 page_cache_get(page);
1668 * basic readpage implementation. Locked extent state structs are inserted
1669 * into the tree that are removed when the IO is done (by the end_io
1672 static int __extent_read_full_page(struct extent_map_tree *tree,
1674 get_extent_t *get_extent,
1677 struct inode *inode = page->mapping->host;
1678 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1679 u64 page_end = start + PAGE_CACHE_SIZE - 1;
1683 u64 last_byte = i_size_read(inode);
1687 struct extent_map *em;
1688 struct block_device *bdev;
1691 size_t page_offset = 0;
1693 size_t blocksize = inode->i_sb->s_blocksize;
1695 set_page_extent_mapped(page);
1698 lock_extent(tree, start, end, GFP_NOFS);
1700 while (cur <= end) {
1701 if (cur >= last_byte) {
1703 iosize = PAGE_CACHE_SIZE - page_offset;
1704 userpage = kmap_atomic(page, KM_USER0);
1705 memset(userpage + page_offset, 0, iosize);
1706 flush_dcache_page(page);
1707 kunmap_atomic(userpage, KM_USER0);
1708 set_extent_uptodate(tree, cur, cur + iosize - 1,
1710 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1713 em = get_extent(inode, page, page_offset, cur, end, 0);
1714 if (IS_ERR(em) || !em) {
1716 unlock_extent(tree, cur, end, GFP_NOFS);
1720 extent_offset = cur - em->start;
1721 BUG_ON(em->end < cur);
1724 iosize = min(em->end - cur, end - cur) + 1;
1725 cur_end = min(em->end, end);
1726 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
1727 sector = (em->block_start + extent_offset) >> 9;
1729 block_start = em->block_start;
1730 free_extent_map(em);
1733 /* we've found a hole, just zero and go on */
1734 if (block_start == EXTENT_MAP_HOLE) {
1736 userpage = kmap_atomic(page, KM_USER0);
1737 memset(userpage + page_offset, 0, iosize);
1738 flush_dcache_page(page);
1739 kunmap_atomic(userpage, KM_USER0);
1741 set_extent_uptodate(tree, cur, cur + iosize - 1,
1743 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1745 page_offset += iosize;
1748 /* the get_extent function already copied into the page */
1749 if (test_range_bit(tree, cur, cur_end, EXTENT_UPTODATE, 1)) {
1750 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1752 page_offset += iosize;
1757 if (tree->ops && tree->ops->readpage_io_hook) {
1758 ret = tree->ops->readpage_io_hook(page, cur,
1762 unsigned long nr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
1764 ret = submit_extent_page(READ, tree, page,
1765 sector, iosize, page_offset,
1767 end_bio_extent_readpage);
1772 page_offset += iosize;
1776 if (!PageError(page))
1777 SetPageUptodate(page);
1783 int extent_read_full_page(struct extent_map_tree *tree, struct page *page,
1784 get_extent_t *get_extent)
1786 struct bio *bio = NULL;
1789 ret = __extent_read_full_page(tree, page, get_extent, &bio);
1791 submit_one_bio(READ, bio);
1794 EXPORT_SYMBOL(extent_read_full_page);
1797 * the writepage semantics are similar to regular writepage. extent
1798 * records are inserted to lock ranges in the tree, and as dirty areas
1799 * are found, they are marked writeback. Then the lock bits are removed
1800 * and the end_io handler clears the writeback ranges
1802 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
1805 struct inode *inode = page->mapping->host;
1806 struct extent_page_data *epd = data;
1807 struct extent_map_tree *tree = epd->tree;
1808 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1810 u64 page_end = start + PAGE_CACHE_SIZE - 1;
1814 u64 last_byte = i_size_read(inode);
1818 struct extent_map *em;
1819 struct block_device *bdev;
1822 size_t page_offset = 0;
1824 loff_t i_size = i_size_read(inode);
1825 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
1829 WARN_ON(!PageLocked(page));
1830 if (page->index > end_index) {
1831 clear_extent_dirty(tree, start, page_end, GFP_NOFS);
1836 if (page->index == end_index) {
1839 size_t offset = i_size & (PAGE_CACHE_SIZE - 1);
1841 userpage = kmap_atomic(page, KM_USER0);
1842 memset(userpage + offset, 0, PAGE_CACHE_SIZE - offset);
1843 flush_dcache_page(page);
1844 kunmap_atomic(userpage, KM_USER0);
1847 set_page_extent_mapped(page);
1849 delalloc_start = start;
1851 while(delalloc_end < page_end) {
1852 nr_delalloc = find_lock_delalloc_range(tree, &delalloc_start,
1855 if (nr_delalloc == 0) {
1856 delalloc_start = delalloc_end + 1;
1859 tree->ops->fill_delalloc(inode, delalloc_start,
1861 clear_extent_bit(tree, delalloc_start,
1863 EXTENT_LOCKED | EXTENT_DELALLOC,
1865 delalloc_start = delalloc_end + 1;
1867 lock_extent(tree, start, page_end, GFP_NOFS);
1870 if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0)) {
1871 printk("found delalloc bits after lock_extent\n");
1874 if (last_byte <= start) {
1875 clear_extent_dirty(tree, start, page_end, GFP_NOFS);
1879 set_extent_uptodate(tree, start, page_end, GFP_NOFS);
1880 blocksize = inode->i_sb->s_blocksize;
1882 while (cur <= end) {
1883 if (cur >= last_byte) {
1884 clear_extent_dirty(tree, cur, page_end, GFP_NOFS);
1887 em = epd->get_extent(inode, page, page_offset, cur, end, 1);
1888 if (IS_ERR(em) || !em) {
1893 extent_offset = cur - em->start;
1894 BUG_ON(em->end < cur);
1896 iosize = min(em->end - cur, end - cur) + 1;
1897 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
1898 sector = (em->block_start + extent_offset) >> 9;
1900 block_start = em->block_start;
1901 free_extent_map(em);
1904 if (block_start == EXTENT_MAP_HOLE ||
1905 block_start == EXTENT_MAP_INLINE) {
1906 clear_extent_dirty(tree, cur,
1907 cur + iosize - 1, GFP_NOFS);
1909 page_offset += iosize;
1913 /* leave this out until we have a page_mkwrite call */
1914 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
1917 page_offset += iosize;
1920 clear_extent_dirty(tree, cur, cur + iosize - 1, GFP_NOFS);
1921 if (tree->ops && tree->ops->writepage_io_hook) {
1922 ret = tree->ops->writepage_io_hook(page, cur,
1930 unsigned long max_nr = end_index + 1;
1931 set_range_writeback(tree, cur, cur + iosize - 1);
1932 if (!PageWriteback(page)) {
1933 printk("warning page %lu not writeback, "
1934 "cur %llu end %llu\n", page->index,
1935 (unsigned long long)cur,
1936 (unsigned long long)end);
1939 ret = submit_extent_page(WRITE, tree, page, sector,
1940 iosize, page_offset, bdev,
1942 end_bio_extent_writepage);
1947 page_offset += iosize;
1952 /* make sure the mapping tag for page dirty gets cleared */
1953 set_page_writeback(page);
1954 end_page_writeback(page);
1956 unlock_extent(tree, start, page_end, GFP_NOFS);
1961 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1963 /* Taken directly from 2.6.23 for 2.6.18 back port */
1964 typedef int (*writepage_t)(struct page *page, struct writeback_control *wbc,
1968 * write_cache_pages - walk the list of dirty pages of the given address space
1969 * and write all of them.
1970 * @mapping: address space structure to write
1971 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
1972 * @writepage: function called for each page
1973 * @data: data passed to writepage function
1975 * If a page is already under I/O, write_cache_pages() skips it, even
1976 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
1977 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
1978 * and msync() need to guarantee that all the data which was dirty at the time
1979 * the call was made get new I/O started against them. If wbc->sync_mode is
1980 * WB_SYNC_ALL then we were called for data integrity and we must wait for
1981 * existing IO to complete.
1983 static int write_cache_pages(struct address_space *mapping,
1984 struct writeback_control *wbc, writepage_t writepage,
1987 struct backing_dev_info *bdi = mapping->backing_dev_info;
1990 struct pagevec pvec;
1993 pgoff_t end; /* Inclusive */
1995 int range_whole = 0;
1997 if (wbc->nonblocking && bdi_write_congested(bdi)) {
1998 wbc->encountered_congestion = 1;
2002 pagevec_init(&pvec, 0);
2003 if (wbc->range_cyclic) {
2004 index = mapping->writeback_index; /* Start from prev offset */
2007 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2008 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2009 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2014 while (!done && (index <= end) &&
2015 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
2016 PAGECACHE_TAG_DIRTY,
2017 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
2021 for (i = 0; i < nr_pages; i++) {
2022 struct page *page = pvec.pages[i];
2025 * At this point we hold neither mapping->tree_lock nor
2026 * lock on the page itself: the page may be truncated or
2027 * invalidated (changing page->mapping to NULL), or even
2028 * swizzled back from swapper_space to tmpfs file
2033 if (unlikely(page->mapping != mapping)) {
2038 if (!wbc->range_cyclic && page->index > end) {
2044 if (wbc->sync_mode != WB_SYNC_NONE)
2045 wait_on_page_writeback(page);
2047 if (PageWriteback(page) ||
2048 !clear_page_dirty_for_io(page)) {
2053 ret = (*writepage)(page, wbc, data);
2055 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
2059 if (ret || (--(wbc->nr_to_write) <= 0))
2061 if (wbc->nonblocking && bdi_write_congested(bdi)) {
2062 wbc->encountered_congestion = 1;
2066 pagevec_release(&pvec);
2069 if (!scanned && !done) {
2071 * We hit the last page and there is more work to be done: wrap
2072 * back to the start of the file
2078 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2079 mapping->writeback_index = index;
2084 int extent_write_full_page(struct extent_map_tree *tree, struct page *page,
2085 get_extent_t *get_extent,
2086 struct writeback_control *wbc)
2089 struct address_space *mapping = page->mapping;
2090 struct extent_page_data epd = {
2093 .get_extent = get_extent,
2095 struct writeback_control wbc_writepages = {
2097 .sync_mode = WB_SYNC_NONE,
2098 .older_than_this = NULL,
2100 .range_start = page_offset(page) + PAGE_CACHE_SIZE,
2101 .range_end = (loff_t)-1,
2105 ret = __extent_writepage(page, wbc, &epd);
2107 write_cache_pages(mapping, &wbc_writepages, __extent_writepage, &epd);
2109 submit_one_bio(WRITE, epd.bio);
2113 EXPORT_SYMBOL(extent_write_full_page);
2116 int extent_writepages(struct extent_map_tree *tree,
2117 struct address_space *mapping,
2118 get_extent_t *get_extent,
2119 struct writeback_control *wbc)
2122 struct extent_page_data epd = {
2125 .get_extent = get_extent,
2128 ret = write_cache_pages(mapping, wbc, __extent_writepage, &epd);
2130 submit_one_bio(WRITE, epd.bio);
2134 EXPORT_SYMBOL(extent_writepages);
2136 int extent_readpages(struct extent_map_tree *tree,
2137 struct address_space *mapping,
2138 struct list_head *pages, unsigned nr_pages,
2139 get_extent_t get_extent)
2141 struct bio *bio = NULL;
2143 struct pagevec pvec;
2145 pagevec_init(&pvec, 0);
2146 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
2147 struct page *page = list_entry(pages->prev, struct page, lru);
2149 prefetchw(&page->flags);
2150 list_del(&page->lru);
2152 * what we want to do here is call add_to_page_cache_lru,
2153 * but that isn't exported, so we reproduce it here
2155 if (!add_to_page_cache(page, mapping,
2156 page->index, GFP_KERNEL)) {
2158 /* open coding of lru_cache_add, also not exported */
2159 page_cache_get(page);
2160 if (!pagevec_add(&pvec, page))
2161 __pagevec_lru_add(&pvec);
2162 __extent_read_full_page(tree, page, get_extent, &bio);
2164 page_cache_release(page);
2166 if (pagevec_count(&pvec))
2167 __pagevec_lru_add(&pvec);
2168 BUG_ON(!list_empty(pages));
2170 submit_one_bio(READ, bio);
2173 EXPORT_SYMBOL(extent_readpages);
2176 * basic invalidatepage code, this waits on any locked or writeback
2177 * ranges corresponding to the page, and then deletes any extent state
2178 * records from the tree
2180 int extent_invalidatepage(struct extent_map_tree *tree,
2181 struct page *page, unsigned long offset)
2183 u64 start = ((u64)page->index << PAGE_CACHE_SHIFT);
2184 u64 end = start + PAGE_CACHE_SIZE - 1;
2185 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
2187 start += (offset + blocksize -1) & ~(blocksize - 1);
2191 lock_extent(tree, start, end, GFP_NOFS);
2192 wait_on_extent_writeback(tree, start, end);
2193 clear_extent_bit(tree, start, end,
2194 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC,
2198 EXPORT_SYMBOL(extent_invalidatepage);
2201 * simple commit_write call, set_range_dirty is used to mark both
2202 * the pages and the extent records as dirty
2204 int extent_commit_write(struct extent_map_tree *tree,
2205 struct inode *inode, struct page *page,
2206 unsigned from, unsigned to)
2208 loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
2210 set_page_extent_mapped(page);
2211 set_page_dirty(page);
2213 if (pos > inode->i_size) {
2214 i_size_write(inode, pos);
2215 mark_inode_dirty(inode);
2219 EXPORT_SYMBOL(extent_commit_write);
2221 int extent_prepare_write(struct extent_map_tree *tree,
2222 struct inode *inode, struct page *page,
2223 unsigned from, unsigned to, get_extent_t *get_extent)
2225 u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2226 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
2228 u64 orig_block_start;
2231 struct extent_map *em;
2232 unsigned blocksize = 1 << inode->i_blkbits;
2233 size_t page_offset = 0;
2234 size_t block_off_start;
2235 size_t block_off_end;
2241 set_page_extent_mapped(page);
2243 block_start = (page_start + from) & ~((u64)blocksize - 1);
2244 block_end = (page_start + to - 1) | (blocksize - 1);
2245 orig_block_start = block_start;
2247 lock_extent(tree, page_start, page_end, GFP_NOFS);
2248 while(block_start <= block_end) {
2249 em = get_extent(inode, page, page_offset, block_start,
2251 if (IS_ERR(em) || !em) {
2254 cur_end = min(block_end, em->end);
2255 block_off_start = block_start & (PAGE_CACHE_SIZE - 1);
2256 block_off_end = block_off_start + blocksize;
2257 isnew = clear_extent_new(tree, block_start, cur_end, GFP_NOFS);
2259 if (!PageUptodate(page) && isnew &&
2260 (block_off_end > to || block_off_start < from)) {
2263 kaddr = kmap_atomic(page, KM_USER0);
2264 if (block_off_end > to)
2265 memset(kaddr + to, 0, block_off_end - to);
2266 if (block_off_start < from)
2267 memset(kaddr + block_off_start, 0,
2268 from - block_off_start);
2269 flush_dcache_page(page);
2270 kunmap_atomic(kaddr, KM_USER0);
2272 if ((em->block_start != EXTENT_MAP_HOLE &&
2273 em->block_start != EXTENT_MAP_INLINE) &&
2274 !isnew && !PageUptodate(page) &&
2275 (block_off_end > to || block_off_start < from) &&
2276 !test_range_bit(tree, block_start, cur_end,
2277 EXTENT_UPTODATE, 1)) {
2279 u64 extent_offset = block_start - em->start;
2281 sector = (em->block_start + extent_offset) >> 9;
2282 iosize = (cur_end - block_start + blocksize - 1) &
2283 ~((u64)blocksize - 1);
2285 * we've already got the extent locked, but we
2286 * need to split the state such that our end_bio
2287 * handler can clear the lock.
2289 set_extent_bit(tree, block_start,
2290 block_start + iosize - 1,
2291 EXTENT_LOCKED, 0, NULL, GFP_NOFS);
2292 ret = submit_extent_page(READ, tree, page,
2293 sector, iosize, page_offset, em->bdev,
2295 end_bio_extent_preparewrite);
2297 block_start = block_start + iosize;
2299 set_extent_uptodate(tree, block_start, cur_end,
2301 unlock_extent(tree, block_start, cur_end, GFP_NOFS);
2302 block_start = cur_end + 1;
2304 page_offset = block_start & (PAGE_CACHE_SIZE - 1);
2305 free_extent_map(em);
2308 wait_extent_bit(tree, orig_block_start,
2309 block_end, EXTENT_LOCKED);
2311 check_page_uptodate(tree, page);
2313 /* FIXME, zero out newly allocated blocks on error */
2316 EXPORT_SYMBOL(extent_prepare_write);
2319 * a helper for releasepage. As long as there are no locked extents
2320 * in the range corresponding to the page, both state records and extent
2321 * map records are removed
2323 int try_release_extent_mapping(struct extent_map_tree *tree, struct page *page)
2325 struct extent_map *em;
2326 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2327 u64 end = start + PAGE_CACHE_SIZE - 1;
2328 u64 orig_start = start;
2331 while (start <= end) {
2332 em = lookup_extent_mapping(tree, start, end);
2333 if (!em || IS_ERR(em))
2335 if (!test_range_bit(tree, em->start, em->end,
2336 EXTENT_LOCKED, 0)) {
2337 remove_extent_mapping(tree, em);
2338 /* once for the rb tree */
2339 free_extent_map(em);
2341 start = em->end + 1;
2343 free_extent_map(em);
2345 if (test_range_bit(tree, orig_start, end, EXTENT_LOCKED, 0))
2348 clear_extent_bit(tree, orig_start, end, EXTENT_UPTODATE,
2352 EXPORT_SYMBOL(try_release_extent_mapping);
2354 sector_t extent_bmap(struct address_space *mapping, sector_t iblock,
2355 get_extent_t *get_extent)
2357 struct inode *inode = mapping->host;
2358 u64 start = iblock << inode->i_blkbits;
2359 u64 end = start + (1 << inode->i_blkbits) - 1;
2360 sector_t sector = 0;
2361 struct extent_map *em;
2363 em = get_extent(inode, NULL, 0, start, end, 0);
2364 if (!em || IS_ERR(em))
2367 if (em->block_start == EXTENT_MAP_INLINE ||
2368 em->block_start == EXTENT_MAP_HOLE)
2371 sector = (em->block_start + start - em->start) >> inode->i_blkbits;
2373 free_extent_map(em);
2377 static int add_lru(struct extent_map_tree *tree, struct extent_buffer *eb)
2379 if (list_empty(&eb->lru)) {
2380 extent_buffer_get(eb);
2381 list_add(&eb->lru, &tree->buffer_lru);
2383 if (tree->lru_size >= BUFFER_LRU_MAX) {
2384 struct extent_buffer *rm;
2385 rm = list_entry(tree->buffer_lru.prev,
2386 struct extent_buffer, lru);
2388 list_del_init(&rm->lru);
2389 free_extent_buffer(rm);
2392 list_move(&eb->lru, &tree->buffer_lru);
2395 static struct extent_buffer *find_lru(struct extent_map_tree *tree,
2396 u64 start, unsigned long len)
2398 struct list_head *lru = &tree->buffer_lru;
2399 struct list_head *cur = lru->next;
2400 struct extent_buffer *eb;
2402 if (list_empty(lru))
2406 eb = list_entry(cur, struct extent_buffer, lru);
2407 if (eb->start == start && eb->len == len) {
2408 extent_buffer_get(eb);
2412 } while (cur != lru);
2416 static inline unsigned long num_extent_pages(u64 start, u64 len)
2418 return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
2419 (start >> PAGE_CACHE_SHIFT);
2422 static inline struct page *extent_buffer_page(struct extent_buffer *eb,
2426 struct address_space *mapping;
2429 return eb->first_page;
2430 i += eb->start >> PAGE_CACHE_SHIFT;
2431 mapping = eb->first_page->mapping;
2432 read_lock_irq(&mapping->tree_lock);
2433 p = radix_tree_lookup(&mapping->page_tree, i);
2434 read_unlock_irq(&mapping->tree_lock);
2438 static struct extent_buffer *__alloc_extent_buffer(struct extent_map_tree *tree,
2443 struct extent_buffer *eb = NULL;
2445 spin_lock(&tree->lru_lock);
2446 eb = find_lru(tree, start, len);
2447 spin_unlock(&tree->lru_lock);
2452 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
2453 INIT_LIST_HEAD(&eb->lru);
2456 atomic_set(&eb->refs, 1);
2461 static void __free_extent_buffer(struct extent_buffer *eb)
2463 kmem_cache_free(extent_buffer_cache, eb);
2466 struct extent_buffer *alloc_extent_buffer(struct extent_map_tree *tree,
2467 u64 start, unsigned long len,
2471 unsigned long num_pages = num_extent_pages(start, len);
2473 unsigned long index = start >> PAGE_CACHE_SHIFT;
2474 struct extent_buffer *eb;
2476 struct address_space *mapping = tree->mapping;
2479 eb = __alloc_extent_buffer(tree, start, len, mask);
2480 if (!eb || IS_ERR(eb))
2483 if (eb->flags & EXTENT_BUFFER_FILLED)
2487 eb->first_page = page0;
2490 page_cache_get(page0);
2491 mark_page_accessed(page0);
2492 set_page_extent_mapped(page0);
2493 WARN_ON(!PageUptodate(page0));
2494 set_page_private(page0, EXTENT_PAGE_PRIVATE_FIRST_PAGE |
2499 for (; i < num_pages; i++, index++) {
2500 p = find_or_create_page(mapping, index, mask | __GFP_HIGHMEM);
2505 set_page_extent_mapped(p);
2506 mark_page_accessed(p);
2509 set_page_private(p, EXTENT_PAGE_PRIVATE_FIRST_PAGE |
2512 set_page_private(p, EXTENT_PAGE_PRIVATE);
2514 if (!PageUptodate(p))
2519 eb->flags |= EXTENT_UPTODATE;
2520 eb->flags |= EXTENT_BUFFER_FILLED;
2523 spin_lock(&tree->lru_lock);
2525 spin_unlock(&tree->lru_lock);
2529 spin_lock(&tree->lru_lock);
2530 list_del_init(&eb->lru);
2531 spin_unlock(&tree->lru_lock);
2532 if (!atomic_dec_and_test(&eb->refs))
2534 for (index = 1; index < i; index++) {
2535 page_cache_release(extent_buffer_page(eb, index));
2538 page_cache_release(extent_buffer_page(eb, 0));
2539 __free_extent_buffer(eb);
2542 EXPORT_SYMBOL(alloc_extent_buffer);
2544 struct extent_buffer *find_extent_buffer(struct extent_map_tree *tree,
2545 u64 start, unsigned long len,
2548 unsigned long num_pages = num_extent_pages(start, len);
2550 unsigned long index = start >> PAGE_CACHE_SHIFT;
2551 struct extent_buffer *eb;
2553 struct address_space *mapping = tree->mapping;
2556 eb = __alloc_extent_buffer(tree, start, len, mask);
2557 if (!eb || IS_ERR(eb))
2560 if (eb->flags & EXTENT_BUFFER_FILLED)
2563 for (i = 0; i < num_pages; i++, index++) {
2564 p = find_lock_page(mapping, index);
2568 set_page_extent_mapped(p);
2569 mark_page_accessed(p);
2573 set_page_private(p, EXTENT_PAGE_PRIVATE_FIRST_PAGE |
2576 set_page_private(p, EXTENT_PAGE_PRIVATE);
2579 if (!PageUptodate(p))
2584 eb->flags |= EXTENT_UPTODATE;
2585 eb->flags |= EXTENT_BUFFER_FILLED;
2588 spin_lock(&tree->lru_lock);
2590 spin_unlock(&tree->lru_lock);
2593 spin_lock(&tree->lru_lock);
2594 list_del_init(&eb->lru);
2595 spin_unlock(&tree->lru_lock);
2596 if (!atomic_dec_and_test(&eb->refs))
2598 for (index = 1; index < i; index++) {
2599 page_cache_release(extent_buffer_page(eb, index));
2602 page_cache_release(extent_buffer_page(eb, 0));
2603 __free_extent_buffer(eb);
2606 EXPORT_SYMBOL(find_extent_buffer);
2608 void free_extent_buffer(struct extent_buffer *eb)
2611 unsigned long num_pages;
2616 if (!atomic_dec_and_test(&eb->refs))
2619 WARN_ON(!list_empty(&eb->lru));
2620 num_pages = num_extent_pages(eb->start, eb->len);
2622 for (i = 1; i < num_pages; i++) {
2623 page_cache_release(extent_buffer_page(eb, i));
2625 page_cache_release(extent_buffer_page(eb, 0));
2626 __free_extent_buffer(eb);
2628 EXPORT_SYMBOL(free_extent_buffer);
2630 int clear_extent_buffer_dirty(struct extent_map_tree *tree,
2631 struct extent_buffer *eb)
2635 unsigned long num_pages;
2638 u64 start = eb->start;
2639 u64 end = start + eb->len - 1;
2641 set = clear_extent_dirty(tree, start, end, GFP_NOFS);
2642 num_pages = num_extent_pages(eb->start, eb->len);
2644 for (i = 0; i < num_pages; i++) {
2645 page = extent_buffer_page(eb, i);
2648 * if we're on the last page or the first page and the
2649 * block isn't aligned on a page boundary, do extra checks
2650 * to make sure we don't clean page that is partially dirty
2652 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
2653 ((i == num_pages - 1) &&
2654 ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
2655 start = (u64)page->index << PAGE_CACHE_SHIFT;
2656 end = start + PAGE_CACHE_SIZE - 1;
2657 if (test_range_bit(tree, start, end,
2663 clear_page_dirty_for_io(page);
2664 write_lock_irq(&page->mapping->tree_lock);
2665 if (!PageDirty(page)) {
2666 radix_tree_tag_clear(&page->mapping->page_tree,
2668 PAGECACHE_TAG_DIRTY);
2670 write_unlock_irq(&page->mapping->tree_lock);
2675 EXPORT_SYMBOL(clear_extent_buffer_dirty);
2677 int wait_on_extent_buffer_writeback(struct extent_map_tree *tree,
2678 struct extent_buffer *eb)
2680 return wait_on_extent_writeback(tree, eb->start,
2681 eb->start + eb->len - 1);
2683 EXPORT_SYMBOL(wait_on_extent_buffer_writeback);
2685 int set_extent_buffer_dirty(struct extent_map_tree *tree,
2686 struct extent_buffer *eb)
2689 unsigned long num_pages;
2691 num_pages = num_extent_pages(eb->start, eb->len);
2692 for (i = 0; i < num_pages; i++) {
2693 struct page *page = extent_buffer_page(eb, i);
2694 /* writepage may need to do something special for the
2695 * first page, we have to make sure page->private is
2696 * properly set. releasepage may drop page->private
2697 * on us if the page isn't already dirty.
2701 set_page_private(page,
2702 EXTENT_PAGE_PRIVATE_FIRST_PAGE |
2705 __set_page_dirty_nobuffers(extent_buffer_page(eb, i));
2709 return set_extent_dirty(tree, eb->start,
2710 eb->start + eb->len - 1, GFP_NOFS);
2712 EXPORT_SYMBOL(set_extent_buffer_dirty);
2714 int set_extent_buffer_uptodate(struct extent_map_tree *tree,
2715 struct extent_buffer *eb)
2719 unsigned long num_pages;
2721 num_pages = num_extent_pages(eb->start, eb->len);
2723 set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
2725 for (i = 0; i < num_pages; i++) {
2726 page = extent_buffer_page(eb, i);
2727 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
2728 ((i == num_pages - 1) &&
2729 ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
2730 check_page_uptodate(tree, page);
2733 SetPageUptodate(page);
2737 EXPORT_SYMBOL(set_extent_buffer_uptodate);
2739 int extent_buffer_uptodate(struct extent_map_tree *tree,
2740 struct extent_buffer *eb)
2742 if (eb->flags & EXTENT_UPTODATE)
2744 return test_range_bit(tree, eb->start, eb->start + eb->len - 1,
2745 EXTENT_UPTODATE, 1);
2747 EXPORT_SYMBOL(extent_buffer_uptodate);
2749 int read_extent_buffer_pages(struct extent_map_tree *tree,
2750 struct extent_buffer *eb,
2755 unsigned long start_i;
2759 unsigned long num_pages;
2761 if (eb->flags & EXTENT_UPTODATE)
2764 if (0 && test_range_bit(tree, eb->start, eb->start + eb->len - 1,
2765 EXTENT_UPTODATE, 1)) {
2770 WARN_ON(start < eb->start);
2771 start_i = (start >> PAGE_CACHE_SHIFT) -
2772 (eb->start >> PAGE_CACHE_SHIFT);
2777 num_pages = num_extent_pages(eb->start, eb->len);
2778 for (i = start_i; i < num_pages; i++) {
2779 page = extent_buffer_page(eb, i);
2780 if (PageUptodate(page)) {
2784 if (TestSetPageLocked(page)) {
2790 if (!PageUptodate(page)) {
2791 err = page->mapping->a_ops->readpage(NULL, page);
2804 for (i = start_i; i < num_pages; i++) {
2805 page = extent_buffer_page(eb, i);
2806 wait_on_page_locked(page);
2807 if (!PageUptodate(page)) {
2812 eb->flags |= EXTENT_UPTODATE;
2815 EXPORT_SYMBOL(read_extent_buffer_pages);
2817 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
2818 unsigned long start,
2825 char *dst = (char *)dstv;
2826 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
2827 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
2828 unsigned long num_pages = num_extent_pages(eb->start, eb->len);
2830 WARN_ON(start > eb->len);
2831 WARN_ON(start + len > eb->start + eb->len);
2833 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
2836 page = extent_buffer_page(eb, i);
2837 if (!PageUptodate(page)) {
2838 printk("page %lu not up to date i %lu, total %lu, len %lu\n", page->index, i, num_pages, eb->len);
2841 WARN_ON(!PageUptodate(page));
2843 cur = min(len, (PAGE_CACHE_SIZE - offset));
2844 kaddr = kmap_atomic(page, KM_USER1);
2845 memcpy(dst, kaddr + offset, cur);
2846 kunmap_atomic(kaddr, KM_USER1);
2854 EXPORT_SYMBOL(read_extent_buffer);
2856 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
2857 unsigned long min_len, char **token, char **map,
2858 unsigned long *map_start,
2859 unsigned long *map_len, int km)
2861 size_t offset = start & (PAGE_CACHE_SIZE - 1);
2864 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
2865 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
2866 unsigned long end_i = (start_offset + start + min_len - 1) >>
2873 offset = start_offset;
2877 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
2879 if (start + min_len > eb->len) {
2880 printk("bad mapping eb start %Lu len %lu, wanted %lu %lu\n", eb->start, eb->len, start, min_len);
2884 p = extent_buffer_page(eb, i);
2885 WARN_ON(!PageUptodate(p));
2886 kaddr = kmap_atomic(p, km);
2888 *map = kaddr + offset;
2889 *map_len = PAGE_CACHE_SIZE - offset;
2892 EXPORT_SYMBOL(map_private_extent_buffer);
2894 int map_extent_buffer(struct extent_buffer *eb, unsigned long start,
2895 unsigned long min_len,
2896 char **token, char **map,
2897 unsigned long *map_start,
2898 unsigned long *map_len, int km)
2902 if (eb->map_token) {
2903 unmap_extent_buffer(eb, eb->map_token, km);
2904 eb->map_token = NULL;
2907 err = map_private_extent_buffer(eb, start, min_len, token, map,
2908 map_start, map_len, km);
2910 eb->map_token = *token;
2912 eb->map_start = *map_start;
2913 eb->map_len = *map_len;
2917 EXPORT_SYMBOL(map_extent_buffer);
2919 void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km)
2921 kunmap_atomic(token, km);
2923 EXPORT_SYMBOL(unmap_extent_buffer);
2925 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
2926 unsigned long start,
2933 char *ptr = (char *)ptrv;
2934 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
2935 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
2938 WARN_ON(start > eb->len);
2939 WARN_ON(start + len > eb->start + eb->len);
2941 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
2944 page = extent_buffer_page(eb, i);
2945 WARN_ON(!PageUptodate(page));
2947 cur = min(len, (PAGE_CACHE_SIZE - offset));
2949 kaddr = kmap_atomic(page, KM_USER0);
2950 ret = memcmp(ptr, kaddr + offset, cur);
2951 kunmap_atomic(kaddr, KM_USER0);
2962 EXPORT_SYMBOL(memcmp_extent_buffer);
2964 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
2965 unsigned long start, unsigned long len)
2971 char *src = (char *)srcv;
2972 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
2973 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
2975 WARN_ON(start > eb->len);
2976 WARN_ON(start + len > eb->start + eb->len);
2978 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
2981 page = extent_buffer_page(eb, i);
2982 WARN_ON(!PageUptodate(page));
2984 cur = min(len, PAGE_CACHE_SIZE - offset);
2985 kaddr = kmap_atomic(page, KM_USER1);
2986 memcpy(kaddr + offset, src, cur);
2987 kunmap_atomic(kaddr, KM_USER1);
2995 EXPORT_SYMBOL(write_extent_buffer);
2997 void memset_extent_buffer(struct extent_buffer *eb, char c,
2998 unsigned long start, unsigned long len)
3004 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3005 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3007 WARN_ON(start > eb->len);
3008 WARN_ON(start + len > eb->start + eb->len);
3010 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3013 page = extent_buffer_page(eb, i);
3014 WARN_ON(!PageUptodate(page));
3016 cur = min(len, PAGE_CACHE_SIZE - offset);
3017 kaddr = kmap_atomic(page, KM_USER0);
3018 memset(kaddr + offset, c, cur);
3019 kunmap_atomic(kaddr, KM_USER0);
3026 EXPORT_SYMBOL(memset_extent_buffer);
3028 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
3029 unsigned long dst_offset, unsigned long src_offset,
3032 u64 dst_len = dst->len;
3037 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3038 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3040 WARN_ON(src->len != dst_len);
3042 offset = (start_offset + dst_offset) &
3043 ((unsigned long)PAGE_CACHE_SIZE - 1);
3046 page = extent_buffer_page(dst, i);
3047 WARN_ON(!PageUptodate(page));
3049 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
3051 kaddr = kmap_atomic(page, KM_USER0);
3052 read_extent_buffer(src, kaddr + offset, src_offset, cur);
3053 kunmap_atomic(kaddr, KM_USER0);
3061 EXPORT_SYMBOL(copy_extent_buffer);
3063 static void move_pages(struct page *dst_page, struct page *src_page,
3064 unsigned long dst_off, unsigned long src_off,
3067 char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3068 if (dst_page == src_page) {
3069 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
3071 char *src_kaddr = kmap_atomic(src_page, KM_USER1);
3072 char *p = dst_kaddr + dst_off + len;
3073 char *s = src_kaddr + src_off + len;
3078 kunmap_atomic(src_kaddr, KM_USER1);
3080 kunmap_atomic(dst_kaddr, KM_USER0);
3083 static void copy_pages(struct page *dst_page, struct page *src_page,
3084 unsigned long dst_off, unsigned long src_off,
3087 char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3090 if (dst_page != src_page)
3091 src_kaddr = kmap_atomic(src_page, KM_USER1);
3093 src_kaddr = dst_kaddr;
3095 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
3096 kunmap_atomic(dst_kaddr, KM_USER0);
3097 if (dst_page != src_page)
3098 kunmap_atomic(src_kaddr, KM_USER1);
3101 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3102 unsigned long src_offset, unsigned long len)
3105 size_t dst_off_in_page;
3106 size_t src_off_in_page;
3107 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3108 unsigned long dst_i;
3109 unsigned long src_i;
3111 if (src_offset + len > dst->len) {
3112 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
3113 src_offset, len, dst->len);
3116 if (dst_offset + len > dst->len) {
3117 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
3118 dst_offset, len, dst->len);
3123 dst_off_in_page = (start_offset + dst_offset) &
3124 ((unsigned long)PAGE_CACHE_SIZE - 1);
3125 src_off_in_page = (start_offset + src_offset) &
3126 ((unsigned long)PAGE_CACHE_SIZE - 1);
3128 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3129 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
3131 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
3133 cur = min_t(unsigned long, cur,
3134 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
3136 copy_pages(extent_buffer_page(dst, dst_i),
3137 extent_buffer_page(dst, src_i),
3138 dst_off_in_page, src_off_in_page, cur);
3145 EXPORT_SYMBOL(memcpy_extent_buffer);
3147 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3148 unsigned long src_offset, unsigned long len)
3151 size_t dst_off_in_page;
3152 size_t src_off_in_page;
3153 unsigned long dst_end = dst_offset + len - 1;
3154 unsigned long src_end = src_offset + len - 1;
3155 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3156 unsigned long dst_i;
3157 unsigned long src_i;
3159 if (src_offset + len > dst->len) {
3160 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
3161 src_offset, len, dst->len);
3164 if (dst_offset + len > dst->len) {
3165 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
3166 dst_offset, len, dst->len);
3169 if (dst_offset < src_offset) {
3170 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
3174 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
3175 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
3177 dst_off_in_page = (start_offset + dst_end) &
3178 ((unsigned long)PAGE_CACHE_SIZE - 1);
3179 src_off_in_page = (start_offset + src_end) &
3180 ((unsigned long)PAGE_CACHE_SIZE - 1);
3182 cur = min_t(unsigned long, len, src_off_in_page + 1);
3183 cur = min(cur, dst_off_in_page + 1);
3184 move_pages(extent_buffer_page(dst, dst_i),
3185 extent_buffer_page(dst, src_i),
3186 dst_off_in_page - cur + 1,
3187 src_off_in_page - cur + 1, cur);
3194 EXPORT_SYMBOL(memmove_extent_buffer);