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_io.h"
16 #include "extent_map.h"
18 /* temporary define until extent_map moves out of btrfs */
19 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
20 unsigned long extra_flags,
21 void (*ctor)(void *, struct kmem_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);
29 static spinlock_t leak_lock = SPIN_LOCK_UNLOCKED;
31 #define BUFFER_LRU_MAX 64
36 struct rb_node rb_node;
39 struct extent_page_data {
41 struct extent_io_tree *tree;
42 get_extent_t *get_extent;
45 int __init extent_io_init(void)
47 extent_state_cache = btrfs_cache_create("extent_state",
48 sizeof(struct extent_state), 0,
50 if (!extent_state_cache)
53 extent_buffer_cache = btrfs_cache_create("extent_buffers",
54 sizeof(struct extent_buffer), 0,
56 if (!extent_buffer_cache)
57 goto free_state_cache;
61 kmem_cache_destroy(extent_state_cache);
65 void extent_io_exit(void)
67 struct extent_state *state;
68 struct extent_buffer *eb;
70 while (!list_empty(&states)) {
71 state = list_entry(states.next, struct extent_state, leak_list);
72 printk("state leak: start %Lu end %Lu state %lu in tree %p refs %d\n", state->start, state->end, state->state, state->tree, atomic_read(&state->refs));
73 list_del(&state->leak_list);
74 kmem_cache_free(extent_state_cache, state);
78 while (!list_empty(&buffers)) {
79 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
80 printk("buffer leak start %Lu len %lu refs %d\n", eb->start, eb->len, atomic_read(&eb->refs));
81 list_del(&eb->leak_list);
82 kmem_cache_free(extent_buffer_cache, eb);
84 if (extent_state_cache)
85 kmem_cache_destroy(extent_state_cache);
86 if (extent_buffer_cache)
87 kmem_cache_destroy(extent_buffer_cache);
90 void extent_io_tree_init(struct extent_io_tree *tree,
91 struct address_space *mapping, gfp_t mask)
93 tree->state.rb_node = NULL;
94 tree->buffer.rb_node = NULL;
96 tree->dirty_bytes = 0;
97 spin_lock_init(&tree->lock);
98 spin_lock_init(&tree->buffer_lock);
99 tree->mapping = mapping;
101 EXPORT_SYMBOL(extent_io_tree_init);
103 struct extent_state *alloc_extent_state(gfp_t mask)
105 struct extent_state *state;
108 state = kmem_cache_alloc(extent_state_cache, mask);
114 spin_lock_irqsave(&leak_lock, flags);
115 list_add(&state->leak_list, &states);
116 spin_unlock_irqrestore(&leak_lock, flags);
118 atomic_set(&state->refs, 1);
119 init_waitqueue_head(&state->wq);
122 EXPORT_SYMBOL(alloc_extent_state);
124 void free_extent_state(struct extent_state *state)
128 if (atomic_dec_and_test(&state->refs)) {
130 WARN_ON(state->tree);
131 spin_lock_irqsave(&leak_lock, flags);
132 list_del(&state->leak_list);
133 spin_unlock_irqrestore(&leak_lock, flags);
134 kmem_cache_free(extent_state_cache, state);
137 EXPORT_SYMBOL(free_extent_state);
139 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
140 struct rb_node *node)
142 struct rb_node ** p = &root->rb_node;
143 struct rb_node * parent = NULL;
144 struct tree_entry *entry;
148 entry = rb_entry(parent, struct tree_entry, rb_node);
150 if (offset < entry->start)
152 else if (offset > entry->end)
158 entry = rb_entry(node, struct tree_entry, rb_node);
159 rb_link_node(node, parent, p);
160 rb_insert_color(node, root);
164 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
165 struct rb_node **prev_ret,
166 struct rb_node **next_ret)
168 struct rb_root *root = &tree->state;
169 struct rb_node * n = root->rb_node;
170 struct rb_node *prev = NULL;
171 struct rb_node *orig_prev = NULL;
172 struct tree_entry *entry;
173 struct tree_entry *prev_entry = NULL;
176 entry = rb_entry(n, struct tree_entry, rb_node);
180 if (offset < entry->start)
182 else if (offset > entry->end)
191 while(prev && offset > prev_entry->end) {
192 prev = rb_next(prev);
193 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
200 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
201 while(prev && offset < prev_entry->start) {
202 prev = rb_prev(prev);
203 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
210 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
213 struct rb_node *prev = NULL;
216 ret = __etree_search(tree, offset, &prev, NULL);
223 static struct extent_buffer *buffer_tree_insert(struct extent_io_tree *tree,
224 u64 offset, struct rb_node *node)
226 struct rb_root *root = &tree->buffer;
227 struct rb_node ** p = &root->rb_node;
228 struct rb_node * parent = NULL;
229 struct extent_buffer *eb;
233 eb = rb_entry(parent, struct extent_buffer, rb_node);
235 if (offset < eb->start)
237 else if (offset > eb->start)
243 rb_link_node(node, parent, p);
244 rb_insert_color(node, root);
248 static struct extent_buffer *buffer_search(struct extent_io_tree *tree,
251 struct rb_root *root = &tree->buffer;
252 struct rb_node * n = root->rb_node;
253 struct extent_buffer *eb;
256 eb = rb_entry(n, struct extent_buffer, rb_node);
257 if (offset < eb->start)
259 else if (offset > eb->start)
268 * utility function to look for merge candidates inside a given range.
269 * Any extents with matching state are merged together into a single
270 * extent in the tree. Extents with EXTENT_IO in their state field
271 * are not merged because the end_io handlers need to be able to do
272 * operations on them without sleeping (or doing allocations/splits).
274 * This should be called with the tree lock held.
276 static int merge_state(struct extent_io_tree *tree,
277 struct extent_state *state)
279 struct extent_state *other;
280 struct rb_node *other_node;
282 if (state->state & EXTENT_IOBITS)
285 other_node = rb_prev(&state->rb_node);
287 other = rb_entry(other_node, struct extent_state, rb_node);
288 if (other->end == state->start - 1 &&
289 other->state == state->state) {
290 state->start = other->start;
292 rb_erase(&other->rb_node, &tree->state);
293 free_extent_state(other);
296 other_node = rb_next(&state->rb_node);
298 other = rb_entry(other_node, struct extent_state, rb_node);
299 if (other->start == state->end + 1 &&
300 other->state == state->state) {
301 other->start = state->start;
303 rb_erase(&state->rb_node, &tree->state);
304 free_extent_state(state);
310 static void set_state_cb(struct extent_io_tree *tree,
311 struct extent_state *state,
314 if (tree->ops && tree->ops->set_bit_hook) {
315 tree->ops->set_bit_hook(tree->mapping->host, state->start,
316 state->end, state->state, bits);
320 static void clear_state_cb(struct extent_io_tree *tree,
321 struct extent_state *state,
324 if (tree->ops && tree->ops->set_bit_hook) {
325 tree->ops->clear_bit_hook(tree->mapping->host, state->start,
326 state->end, state->state, bits);
331 * insert an extent_state struct into the tree. 'bits' are set on the
332 * struct before it is inserted.
334 * This may return -EEXIST if the extent is already there, in which case the
335 * state struct is freed.
337 * The tree lock is not taken internally. This is a utility function and
338 * probably isn't what you want to call (see set/clear_extent_bit).
340 static int insert_state(struct extent_io_tree *tree,
341 struct extent_state *state, u64 start, u64 end,
344 struct rb_node *node;
347 printk("end < start %Lu %Lu\n", end, start);
350 if (bits & EXTENT_DIRTY)
351 tree->dirty_bytes += end - start + 1;
352 set_state_cb(tree, state, bits);
353 state->state |= bits;
354 state->start = start;
356 node = tree_insert(&tree->state, end, &state->rb_node);
358 struct extent_state *found;
359 found = rb_entry(node, struct extent_state, rb_node);
360 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, start, end);
361 free_extent_state(state);
365 merge_state(tree, state);
370 * split a given extent state struct in two, inserting the preallocated
371 * struct 'prealloc' as the newly created second half. 'split' indicates an
372 * offset inside 'orig' where it should be split.
375 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
376 * are two extent state structs in the tree:
377 * prealloc: [orig->start, split - 1]
378 * orig: [ split, orig->end ]
380 * The tree locks are not taken by this function. They need to be held
383 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
384 struct extent_state *prealloc, u64 split)
386 struct rb_node *node;
387 prealloc->start = orig->start;
388 prealloc->end = split - 1;
389 prealloc->state = orig->state;
392 node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
394 struct extent_state *found;
395 found = rb_entry(node, struct extent_state, rb_node);
396 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, prealloc->start, prealloc->end);
397 free_extent_state(prealloc);
400 prealloc->tree = tree;
405 * utility function to clear some bits in an extent state struct.
406 * it will optionally wake up any one waiting on this state (wake == 1), or
407 * forcibly remove the state from the tree (delete == 1).
409 * If no bits are set on the state struct after clearing things, the
410 * struct is freed and removed from the tree
412 static int clear_state_bit(struct extent_io_tree *tree,
413 struct extent_state *state, int bits, int wake,
416 int ret = state->state & bits;
418 if ((bits & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
419 u64 range = state->end - state->start + 1;
420 WARN_ON(range > tree->dirty_bytes);
421 tree->dirty_bytes -= range;
423 clear_state_cb(tree, state, bits);
424 state->state &= ~bits;
427 if (delete || state->state == 0) {
429 clear_state_cb(tree, state, state->state);
430 rb_erase(&state->rb_node, &tree->state);
432 free_extent_state(state);
437 merge_state(tree, state);
443 * clear some bits on a range in the tree. This may require splitting
444 * or inserting elements in the tree, so the gfp mask is used to
445 * indicate which allocations or sleeping are allowed.
447 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
448 * the given range from the tree regardless of state (ie for truncate).
450 * the range [start, end] is inclusive.
452 * This takes the tree lock, and returns < 0 on error, > 0 if any of the
453 * bits were already set, or zero if none of the bits were already set.
455 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
456 int bits, int wake, int delete, gfp_t mask)
458 struct extent_state *state;
459 struct extent_state *prealloc = NULL;
460 struct rb_node *node;
466 if (!prealloc && (mask & __GFP_WAIT)) {
467 prealloc = alloc_extent_state(mask);
472 spin_lock_irqsave(&tree->lock, flags);
474 * this search will find the extents that end after
477 node = tree_search(tree, start);
480 state = rb_entry(node, struct extent_state, rb_node);
481 if (state->start > end)
483 WARN_ON(state->end < start);
486 * | ---- desired range ---- |
488 * | ------------- state -------------- |
490 * We need to split the extent we found, and may flip
491 * bits on second half.
493 * If the extent we found extends past our range, we
494 * just split and search again. It'll get split again
495 * the next time though.
497 * If the extent we found is inside our range, we clear
498 * the desired bit on it.
501 if (state->start < start) {
503 prealloc = alloc_extent_state(GFP_ATOMIC);
504 err = split_state(tree, state, prealloc, start);
505 BUG_ON(err == -EEXIST);
509 if (state->end <= end) {
510 start = state->end + 1;
511 set |= clear_state_bit(tree, state, bits,
514 start = state->start;
519 * | ---- desired range ---- |
521 * We need to split the extent, and clear the bit
524 if (state->start <= end && state->end > end) {
526 prealloc = alloc_extent_state(GFP_ATOMIC);
527 err = split_state(tree, state, prealloc, end + 1);
528 BUG_ON(err == -EEXIST);
532 set |= clear_state_bit(tree, prealloc, bits,
538 start = state->end + 1;
539 set |= clear_state_bit(tree, state, bits, wake, delete);
543 spin_unlock_irqrestore(&tree->lock, flags);
545 free_extent_state(prealloc);
552 spin_unlock_irqrestore(&tree->lock, flags);
553 if (mask & __GFP_WAIT)
557 EXPORT_SYMBOL(clear_extent_bit);
559 static int wait_on_state(struct extent_io_tree *tree,
560 struct extent_state *state)
563 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
564 spin_unlock_irq(&tree->lock);
566 spin_lock_irq(&tree->lock);
567 finish_wait(&state->wq, &wait);
572 * waits for one or more bits to clear on a range in the state tree.
573 * The range [start, end] is inclusive.
574 * The tree lock is taken by this function
576 int wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits)
578 struct extent_state *state;
579 struct rb_node *node;
581 spin_lock_irq(&tree->lock);
585 * this search will find all the extents that end after
588 node = tree_search(tree, start);
592 state = rb_entry(node, struct extent_state, rb_node);
594 if (state->start > end)
597 if (state->state & bits) {
598 start = state->start;
599 atomic_inc(&state->refs);
600 wait_on_state(tree, state);
601 free_extent_state(state);
604 start = state->end + 1;
609 if (need_resched()) {
610 spin_unlock_irq(&tree->lock);
612 spin_lock_irq(&tree->lock);
616 spin_unlock_irq(&tree->lock);
619 EXPORT_SYMBOL(wait_extent_bit);
621 static void set_state_bits(struct extent_io_tree *tree,
622 struct extent_state *state,
625 if ((bits & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
626 u64 range = state->end - state->start + 1;
627 tree->dirty_bytes += range;
629 set_state_cb(tree, state, bits);
630 state->state |= bits;
634 * set some bits on a range in the tree. This may require allocations
635 * or sleeping, so the gfp mask is used to indicate what is allowed.
637 * If 'exclusive' == 1, this will fail with -EEXIST if some part of the
638 * range already has the desired bits set. The start of the existing
639 * range is returned in failed_start in this case.
641 * [start, end] is inclusive
642 * This takes the tree lock.
644 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits,
645 int exclusive, u64 *failed_start, gfp_t mask)
647 struct extent_state *state;
648 struct extent_state *prealloc = NULL;
649 struct rb_node *node;
656 if (!prealloc && (mask & __GFP_WAIT)) {
657 prealloc = alloc_extent_state(mask);
662 spin_lock_irqsave(&tree->lock, flags);
664 * this search will find all the extents that end after
667 node = tree_search(tree, start);
669 err = insert_state(tree, prealloc, start, end, bits);
671 BUG_ON(err == -EEXIST);
675 state = rb_entry(node, struct extent_state, rb_node);
676 last_start = state->start;
677 last_end = state->end;
680 * | ---- desired range ---- |
683 * Just lock what we found and keep going
685 if (state->start == start && state->end <= end) {
686 set = state->state & bits;
687 if (set && exclusive) {
688 *failed_start = state->start;
692 set_state_bits(tree, state, bits);
693 start = state->end + 1;
694 merge_state(tree, state);
699 * | ---- desired range ---- |
702 * | ------------- state -------------- |
704 * We need to split the extent we found, and may flip bits on
707 * If the extent we found extends past our
708 * range, we just split and search again. It'll get split
709 * again the next time though.
711 * If the extent we found is inside our range, we set the
714 if (state->start < start) {
715 set = state->state & bits;
716 if (exclusive && set) {
717 *failed_start = start;
721 err = split_state(tree, state, prealloc, start);
722 BUG_ON(err == -EEXIST);
726 if (state->end <= end) {
727 set_state_bits(tree, state, bits);
728 start = state->end + 1;
729 merge_state(tree, state);
731 start = state->start;
736 * | ---- desired range ---- |
737 * | state | or | state |
739 * There's a hole, we need to insert something in it and
740 * ignore the extent we found.
742 if (state->start > start) {
744 if (end < last_start)
747 this_end = last_start -1;
748 err = insert_state(tree, prealloc, start, this_end,
751 BUG_ON(err == -EEXIST);
754 start = this_end + 1;
758 * | ---- desired range ---- |
760 * We need to split the extent, and set the bit
763 if (state->start <= end && state->end > end) {
764 set = state->state & bits;
765 if (exclusive && set) {
766 *failed_start = start;
770 err = split_state(tree, state, prealloc, end + 1);
771 BUG_ON(err == -EEXIST);
773 set_state_bits(tree, prealloc, bits);
774 merge_state(tree, prealloc);
782 spin_unlock_irqrestore(&tree->lock, flags);
784 free_extent_state(prealloc);
791 spin_unlock_irqrestore(&tree->lock, flags);
792 if (mask & __GFP_WAIT)
796 EXPORT_SYMBOL(set_extent_bit);
798 /* wrappers around set/clear extent bit */
799 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
802 return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL,
805 EXPORT_SYMBOL(set_extent_dirty);
807 int set_extent_ordered(struct extent_io_tree *tree, u64 start, u64 end,
810 return set_extent_bit(tree, start, end, EXTENT_ORDERED, 0, NULL, mask);
812 EXPORT_SYMBOL(set_extent_ordered);
814 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
815 int bits, gfp_t mask)
817 return set_extent_bit(tree, start, end, bits, 0, NULL,
820 EXPORT_SYMBOL(set_extent_bits);
822 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
823 int bits, gfp_t mask)
825 return clear_extent_bit(tree, start, end, bits, 0, 0, mask);
827 EXPORT_SYMBOL(clear_extent_bits);
829 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
832 return set_extent_bit(tree, start, end,
833 EXTENT_DELALLOC | EXTENT_DIRTY,
836 EXPORT_SYMBOL(set_extent_delalloc);
838 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
841 return clear_extent_bit(tree, start, end,
842 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, mask);
844 EXPORT_SYMBOL(clear_extent_dirty);
846 int clear_extent_ordered(struct extent_io_tree *tree, u64 start, u64 end,
849 return clear_extent_bit(tree, start, end, EXTENT_ORDERED, 1, 0, mask);
851 EXPORT_SYMBOL(clear_extent_ordered);
853 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
856 return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL,
859 EXPORT_SYMBOL(set_extent_new);
861 int clear_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
864 return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0, mask);
866 EXPORT_SYMBOL(clear_extent_new);
868 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
871 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, NULL,
874 EXPORT_SYMBOL(set_extent_uptodate);
876 int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
879 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, mask);
881 EXPORT_SYMBOL(clear_extent_uptodate);
883 int set_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end,
886 return set_extent_bit(tree, start, end, EXTENT_WRITEBACK,
889 EXPORT_SYMBOL(set_extent_writeback);
891 int clear_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end,
894 return clear_extent_bit(tree, start, end, EXTENT_WRITEBACK, 1, 0, mask);
896 EXPORT_SYMBOL(clear_extent_writeback);
898 int wait_on_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end)
900 return wait_extent_bit(tree, start, end, EXTENT_WRITEBACK);
902 EXPORT_SYMBOL(wait_on_extent_writeback);
904 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
909 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1,
910 &failed_start, mask);
911 if (err == -EEXIST && (mask & __GFP_WAIT)) {
912 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
913 start = failed_start;
917 WARN_ON(start > end);
921 EXPORT_SYMBOL(lock_extent);
923 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end,
926 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, mask);
928 EXPORT_SYMBOL(unlock_extent);
931 * helper function to set pages and extents in the tree dirty
933 int set_range_dirty(struct extent_io_tree *tree, u64 start, u64 end)
935 unsigned long index = start >> PAGE_CACHE_SHIFT;
936 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
939 while (index <= end_index) {
940 page = find_get_page(tree->mapping, index);
942 __set_page_dirty_nobuffers(page);
943 page_cache_release(page);
946 set_extent_dirty(tree, start, end, GFP_NOFS);
949 EXPORT_SYMBOL(set_range_dirty);
952 * helper function to set both pages and extents in the tree writeback
954 int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
956 unsigned long index = start >> PAGE_CACHE_SHIFT;
957 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
960 while (index <= end_index) {
961 page = find_get_page(tree->mapping, index);
963 set_page_writeback(page);
964 page_cache_release(page);
967 set_extent_writeback(tree, start, end, GFP_NOFS);
970 EXPORT_SYMBOL(set_range_writeback);
972 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
973 u64 *start_ret, u64 *end_ret, int bits)
975 struct rb_node *node;
976 struct extent_state *state;
979 spin_lock_irq(&tree->lock);
981 * this search will find all the extents that end after
984 node = tree_search(tree, start);
990 state = rb_entry(node, struct extent_state, rb_node);
991 if (state->end >= start && (state->state & bits)) {
992 *start_ret = state->start;
993 *end_ret = state->end;
997 node = rb_next(node);
1002 spin_unlock_irq(&tree->lock);
1005 EXPORT_SYMBOL(find_first_extent_bit);
1007 struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree,
1008 u64 start, int bits)
1010 struct rb_node *node;
1011 struct extent_state *state;
1014 * this search will find all the extents that end after
1017 node = tree_search(tree, start);
1023 state = rb_entry(node, struct extent_state, rb_node);
1024 if (state->end >= start && (state->state & bits)) {
1027 node = rb_next(node);
1034 EXPORT_SYMBOL(find_first_extent_bit_state);
1036 u64 find_lock_delalloc_range(struct extent_io_tree *tree,
1037 u64 *start, u64 *end, u64 max_bytes)
1039 struct rb_node *node;
1040 struct extent_state *state;
1041 u64 cur_start = *start;
1043 u64 total_bytes = 0;
1045 spin_lock_irq(&tree->lock);
1047 * this search will find all the extents that end after
1051 node = tree_search(tree, cur_start);
1059 state = rb_entry(node, struct extent_state, rb_node);
1060 if (found && state->start != cur_start) {
1063 if (!(state->state & EXTENT_DELALLOC)) {
1069 struct extent_state *prev_state;
1070 struct rb_node *prev_node = node;
1072 prev_node = rb_prev(prev_node);
1075 prev_state = rb_entry(prev_node,
1076 struct extent_state,
1078 if (!(prev_state->state & EXTENT_DELALLOC))
1084 if (state->state & EXTENT_LOCKED) {
1086 atomic_inc(&state->refs);
1087 prepare_to_wait(&state->wq, &wait,
1088 TASK_UNINTERRUPTIBLE);
1089 spin_unlock_irq(&tree->lock);
1091 spin_lock_irq(&tree->lock);
1092 finish_wait(&state->wq, &wait);
1093 free_extent_state(state);
1096 set_state_cb(tree, state, EXTENT_LOCKED);
1097 state->state |= EXTENT_LOCKED;
1099 *start = state->start;
1102 cur_start = state->end + 1;
1103 node = rb_next(node);
1106 total_bytes += state->end - state->start + 1;
1107 if (total_bytes >= max_bytes)
1111 spin_unlock_irq(&tree->lock);
1115 u64 count_range_bits(struct extent_io_tree *tree,
1116 u64 *start, u64 search_end, u64 max_bytes,
1119 struct rb_node *node;
1120 struct extent_state *state;
1121 u64 cur_start = *start;
1122 u64 total_bytes = 0;
1125 if (search_end <= cur_start) {
1126 printk("search_end %Lu start %Lu\n", search_end, cur_start);
1131 spin_lock_irq(&tree->lock);
1132 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1133 total_bytes = tree->dirty_bytes;
1137 * this search will find all the extents that end after
1140 node = tree_search(tree, cur_start);
1146 state = rb_entry(node, struct extent_state, rb_node);
1147 if (state->start > search_end)
1149 if (state->end >= cur_start && (state->state & bits)) {
1150 total_bytes += min(search_end, state->end) + 1 -
1151 max(cur_start, state->start);
1152 if (total_bytes >= max_bytes)
1155 *start = state->start;
1159 node = rb_next(node);
1164 spin_unlock_irq(&tree->lock);
1168 * helper function to lock both pages and extents in the tree.
1169 * pages must be locked first.
1171 int lock_range(struct extent_io_tree *tree, u64 start, u64 end)
1173 unsigned long index = start >> PAGE_CACHE_SHIFT;
1174 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1178 while (index <= end_index) {
1179 page = grab_cache_page(tree->mapping, index);
1185 err = PTR_ERR(page);
1190 lock_extent(tree, start, end, GFP_NOFS);
1195 * we failed above in getting the page at 'index', so we undo here
1196 * up to but not including the page at 'index'
1199 index = start >> PAGE_CACHE_SHIFT;
1200 while (index < end_index) {
1201 page = find_get_page(tree->mapping, index);
1203 page_cache_release(page);
1208 EXPORT_SYMBOL(lock_range);
1211 * helper function to unlock both pages and extents in the tree.
1213 int unlock_range(struct extent_io_tree *tree, u64 start, u64 end)
1215 unsigned long index = start >> PAGE_CACHE_SHIFT;
1216 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1219 while (index <= end_index) {
1220 page = find_get_page(tree->mapping, index);
1222 page_cache_release(page);
1225 unlock_extent(tree, start, end, GFP_NOFS);
1228 EXPORT_SYMBOL(unlock_range);
1230 int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1232 struct rb_node *node;
1233 struct extent_state *state;
1236 spin_lock_irq(&tree->lock);
1238 * this search will find all the extents that end after
1241 node = tree_search(tree, start);
1246 state = rb_entry(node, struct extent_state, rb_node);
1247 if (state->start != start) {
1251 state->private = private;
1253 spin_unlock_irq(&tree->lock);
1257 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1259 struct rb_node *node;
1260 struct extent_state *state;
1263 spin_lock_irq(&tree->lock);
1265 * this search will find all the extents that end after
1268 node = tree_search(tree, start);
1273 state = rb_entry(node, struct extent_state, rb_node);
1274 if (state->start != start) {
1278 *private = state->private;
1280 spin_unlock_irq(&tree->lock);
1285 * searches a range in the state tree for a given mask.
1286 * If 'filled' == 1, this returns 1 only if every extent in the tree
1287 * has the bits set. Otherwise, 1 is returned if any bit in the
1288 * range is found set.
1290 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1291 int bits, int filled)
1293 struct extent_state *state = NULL;
1294 struct rb_node *node;
1296 unsigned long flags;
1298 spin_lock_irqsave(&tree->lock, flags);
1299 node = tree_search(tree, start);
1300 while (node && start <= end) {
1301 state = rb_entry(node, struct extent_state, rb_node);
1303 if (filled && state->start > start) {
1308 if (state->start > end)
1311 if (state->state & bits) {
1315 } else if (filled) {
1319 start = state->end + 1;
1322 node = rb_next(node);
1329 spin_unlock_irqrestore(&tree->lock, flags);
1332 EXPORT_SYMBOL(test_range_bit);
1335 * helper function to set a given page up to date if all the
1336 * extents in the tree for that page are up to date
1338 static int check_page_uptodate(struct extent_io_tree *tree,
1341 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1342 u64 end = start + PAGE_CACHE_SIZE - 1;
1343 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1))
1344 SetPageUptodate(page);
1349 * helper function to unlock a page if all the extents in the tree
1350 * for that page are unlocked
1352 static int check_page_locked(struct extent_io_tree *tree,
1355 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1356 u64 end = start + PAGE_CACHE_SIZE - 1;
1357 if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0))
1363 * helper function to end page writeback if all the extents
1364 * in the tree for that page are done with writeback
1366 static int check_page_writeback(struct extent_io_tree *tree,
1369 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1370 u64 end = start + PAGE_CACHE_SIZE - 1;
1371 if (!test_range_bit(tree, start, end, EXTENT_WRITEBACK, 0))
1372 end_page_writeback(page);
1376 /* lots and lots of room for performance fixes in the end_bio funcs */
1379 * after a writepage IO is done, we need to:
1380 * clear the uptodate bits on error
1381 * clear the writeback bits in the extent tree for this IO
1382 * end_page_writeback if the page has no more pending IO
1384 * Scheduling is not allowed, so the extent state tree is expected
1385 * to have one and only one object corresponding to this IO.
1387 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1388 static void end_bio_extent_writepage(struct bio *bio, int err)
1390 static int end_bio_extent_writepage(struct bio *bio,
1391 unsigned int bytes_done, int err)
1394 int uptodate = err == 0;
1395 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1396 struct extent_state *state = bio->bi_private;
1397 struct extent_io_tree *tree = state->tree;
1398 struct rb_node *node;
1404 unsigned long flags;
1406 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1411 struct page *page = bvec->bv_page;
1412 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1414 end = start + bvec->bv_len - 1;
1416 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1421 if (--bvec >= bio->bi_io_vec)
1422 prefetchw(&bvec->bv_page->flags);
1423 if (tree->ops && tree->ops->writepage_end_io_hook) {
1424 ret = tree->ops->writepage_end_io_hook(page, start,
1425 end, state, uptodate);
1430 if (!uptodate && tree->ops &&
1431 tree->ops->writepage_io_failed_hook) {
1432 ret = tree->ops->writepage_io_failed_hook(bio, page,
1436 uptodate = (err == 0);
1442 clear_extent_uptodate(tree, start, end, GFP_ATOMIC);
1443 ClearPageUptodate(page);
1448 * bios can get merged in funny ways, and so we need to
1449 * be careful with the state variable. We know the
1450 * state won't be merged with others because it has
1451 * WRITEBACK set, but we can't be sure each biovec is
1452 * sequential in the file. So, if our cached state
1453 * doesn't match the expected end, search the tree
1454 * for the correct one.
1457 spin_lock_irqsave(&tree->lock, flags);
1458 if (!state || state->end != end) {
1460 node = __etree_search(tree, start, NULL, NULL);
1462 state = rb_entry(node, struct extent_state,
1464 if (state->end != end ||
1465 !(state->state & EXTENT_WRITEBACK))
1469 spin_unlock_irqrestore(&tree->lock, flags);
1470 clear_extent_writeback(tree, start,
1477 struct extent_state *clear = state;
1479 node = rb_prev(&state->rb_node);
1481 state = rb_entry(node,
1482 struct extent_state,
1488 clear_state_bit(tree, clear, EXTENT_WRITEBACK,
1499 /* before releasing the lock, make sure the next state
1500 * variable has the expected bits set and corresponds
1501 * to the correct offsets in the file
1503 if (state && (state->end + 1 != start ||
1504 !(state->state & EXTENT_WRITEBACK))) {
1507 spin_unlock_irqrestore(&tree->lock, flags);
1511 end_page_writeback(page);
1513 check_page_writeback(tree, page);
1514 } while (bvec >= bio->bi_io_vec);
1516 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1522 * after a readpage IO is done, we need to:
1523 * clear the uptodate bits on error
1524 * set the uptodate bits if things worked
1525 * set the page up to date if all extents in the tree are uptodate
1526 * clear the lock bit in the extent tree
1527 * unlock the page if there are no other extents locked for it
1529 * Scheduling is not allowed, so the extent state tree is expected
1530 * to have one and only one object corresponding to this IO.
1532 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1533 static void end_bio_extent_readpage(struct bio *bio, int err)
1535 static int end_bio_extent_readpage(struct bio *bio,
1536 unsigned int bytes_done, int err)
1539 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1540 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1541 struct extent_state *state = bio->bi_private;
1542 struct extent_io_tree *tree = state->tree;
1543 struct rb_node *node;
1547 unsigned long flags;
1551 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1557 struct page *page = bvec->bv_page;
1558 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1560 end = start + bvec->bv_len - 1;
1562 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1567 if (--bvec >= bio->bi_io_vec)
1568 prefetchw(&bvec->bv_page->flags);
1570 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
1571 ret = tree->ops->readpage_end_io_hook(page, start, end,
1576 if (!uptodate && tree->ops &&
1577 tree->ops->readpage_io_failed_hook) {
1578 ret = tree->ops->readpage_io_failed_hook(bio, page,
1583 test_bit(BIO_UPTODATE, &bio->bi_flags);
1588 spin_lock_irqsave(&tree->lock, flags);
1589 if (!state || state->end != end) {
1591 node = __etree_search(tree, start, NULL, NULL);
1593 state = rb_entry(node, struct extent_state,
1595 if (state->end != end ||
1596 !(state->state & EXTENT_LOCKED))
1600 spin_unlock_irqrestore(&tree->lock, flags);
1602 set_extent_uptodate(tree, start, end,
1604 unlock_extent(tree, start, end, GFP_ATOMIC);
1611 struct extent_state *clear = state;
1613 node = rb_prev(&state->rb_node);
1615 state = rb_entry(node,
1616 struct extent_state,
1622 set_state_cb(tree, clear, EXTENT_UPTODATE);
1623 clear->state |= EXTENT_UPTODATE;
1625 clear_state_bit(tree, clear, EXTENT_LOCKED,
1636 /* before releasing the lock, make sure the next state
1637 * variable has the expected bits set and corresponds
1638 * to the correct offsets in the file
1640 if (state && (state->end + 1 != start ||
1641 !(state->state & EXTENT_LOCKED))) {
1644 spin_unlock_irqrestore(&tree->lock, flags);
1648 SetPageUptodate(page);
1650 ClearPageUptodate(page);
1656 check_page_uptodate(tree, page);
1658 ClearPageUptodate(page);
1661 check_page_locked(tree, page);
1663 } while (bvec >= bio->bi_io_vec);
1666 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1672 * IO done from prepare_write is pretty simple, we just unlock
1673 * the structs in the extent tree when done, and set the uptodate bits
1676 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1677 static void end_bio_extent_preparewrite(struct bio *bio, int err)
1679 static int end_bio_extent_preparewrite(struct bio *bio,
1680 unsigned int bytes_done, int err)
1683 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1684 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1685 struct extent_state *state = bio->bi_private;
1686 struct extent_io_tree *tree = state->tree;
1690 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1696 struct page *page = bvec->bv_page;
1697 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1699 end = start + bvec->bv_len - 1;
1701 if (--bvec >= bio->bi_io_vec)
1702 prefetchw(&bvec->bv_page->flags);
1705 set_extent_uptodate(tree, start, end, GFP_ATOMIC);
1707 ClearPageUptodate(page);
1711 unlock_extent(tree, start, end, GFP_ATOMIC);
1713 } while (bvec >= bio->bi_io_vec);
1716 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1722 extent_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
1727 bio = bio_alloc(gfp_flags, nr_vecs);
1729 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
1730 while (!bio && (nr_vecs /= 2))
1731 bio = bio_alloc(gfp_flags, nr_vecs);
1736 bio->bi_bdev = bdev;
1737 bio->bi_sector = first_sector;
1742 static int submit_one_bio(int rw, struct bio *bio, int mirror_num)
1745 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1746 struct page *page = bvec->bv_page;
1747 struct extent_io_tree *tree = bio->bi_private;
1748 struct rb_node *node;
1749 struct extent_state *state;
1753 start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
1754 end = start + bvec->bv_len - 1;
1756 spin_lock_irq(&tree->lock);
1757 node = __etree_search(tree, start, NULL, NULL);
1759 state = rb_entry(node, struct extent_state, rb_node);
1760 while(state->end < end) {
1761 node = rb_next(node);
1762 state = rb_entry(node, struct extent_state, rb_node);
1764 BUG_ON(state->end != end);
1765 spin_unlock_irq(&tree->lock);
1767 bio->bi_private = state;
1771 if (tree->ops && tree->ops->submit_bio_hook)
1772 tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
1775 submit_bio(rw, bio);
1776 if (bio_flagged(bio, BIO_EOPNOTSUPP))
1782 static int submit_extent_page(int rw, struct extent_io_tree *tree,
1783 struct page *page, sector_t sector,
1784 size_t size, unsigned long offset,
1785 struct block_device *bdev,
1786 struct bio **bio_ret,
1787 unsigned long max_pages,
1788 bio_end_io_t end_io_func,
1795 if (bio_ret && *bio_ret) {
1797 if (bio->bi_sector + (bio->bi_size >> 9) != sector ||
1798 (tree->ops && tree->ops->merge_bio_hook &&
1799 tree->ops->merge_bio_hook(page, offset, size, bio)) ||
1800 bio_add_page(bio, page, size, offset) < size) {
1801 ret = submit_one_bio(rw, bio, mirror_num);
1807 nr = bio_get_nr_vecs(bdev);
1808 bio = extent_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
1810 printk("failed to allocate bio nr %d\n", nr);
1814 bio_add_page(bio, page, size, offset);
1815 bio->bi_end_io = end_io_func;
1816 bio->bi_private = tree;
1821 ret = submit_one_bio(rw, bio, mirror_num);
1827 void set_page_extent_mapped(struct page *page)
1829 if (!PagePrivate(page)) {
1830 SetPagePrivate(page);
1831 page_cache_get(page);
1832 set_page_private(page, EXTENT_PAGE_PRIVATE);
1836 void set_page_extent_head(struct page *page, unsigned long len)
1838 set_page_private(page, EXTENT_PAGE_PRIVATE_FIRST_PAGE | len << 2);
1842 * basic readpage implementation. Locked extent state structs are inserted
1843 * into the tree that are removed when the IO is done (by the end_io
1846 static int __extent_read_full_page(struct extent_io_tree *tree,
1848 get_extent_t *get_extent,
1849 struct bio **bio, int mirror_num)
1851 struct inode *inode = page->mapping->host;
1852 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1853 u64 page_end = start + PAGE_CACHE_SIZE - 1;
1857 u64 last_byte = i_size_read(inode);
1861 struct extent_map *em;
1862 struct block_device *bdev;
1865 size_t page_offset = 0;
1867 size_t blocksize = inode->i_sb->s_blocksize;
1869 set_page_extent_mapped(page);
1872 lock_extent(tree, start, end, GFP_NOFS);
1874 while (cur <= end) {
1875 if (cur >= last_byte) {
1877 iosize = PAGE_CACHE_SIZE - page_offset;
1878 userpage = kmap_atomic(page, KM_USER0);
1879 memset(userpage + page_offset, 0, iosize);
1880 flush_dcache_page(page);
1881 kunmap_atomic(userpage, KM_USER0);
1882 set_extent_uptodate(tree, cur, cur + iosize - 1,
1884 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1887 em = get_extent(inode, page, page_offset, cur,
1889 if (IS_ERR(em) || !em) {
1891 unlock_extent(tree, cur, end, GFP_NOFS);
1894 extent_offset = cur - em->start;
1895 if (extent_map_end(em) <= cur) {
1896 printk("bad mapping em [%Lu %Lu] cur %Lu\n", em->start, extent_map_end(em), cur);
1898 BUG_ON(extent_map_end(em) <= cur);
1900 printk("2bad mapping end %Lu cur %Lu\n", end, cur);
1904 iosize = min(extent_map_end(em) - cur, end - cur + 1);
1905 cur_end = min(extent_map_end(em) - 1, end);
1906 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
1907 sector = (em->block_start + extent_offset) >> 9;
1909 block_start = em->block_start;
1910 free_extent_map(em);
1913 /* we've found a hole, just zero and go on */
1914 if (block_start == EXTENT_MAP_HOLE) {
1916 userpage = kmap_atomic(page, KM_USER0);
1917 memset(userpage + page_offset, 0, iosize);
1918 flush_dcache_page(page);
1919 kunmap_atomic(userpage, KM_USER0);
1921 set_extent_uptodate(tree, cur, cur + iosize - 1,
1923 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1925 page_offset += iosize;
1928 /* the get_extent function already copied into the page */
1929 if (test_range_bit(tree, cur, cur_end, EXTENT_UPTODATE, 1)) {
1930 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1932 page_offset += iosize;
1935 /* we have an inline extent but it didn't get marked up
1936 * to date. Error out
1938 if (block_start == EXTENT_MAP_INLINE) {
1940 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1942 page_offset += iosize;
1947 if (tree->ops && tree->ops->readpage_io_hook) {
1948 ret = tree->ops->readpage_io_hook(page, cur,
1952 unsigned long nr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
1954 ret = submit_extent_page(READ, tree, page,
1955 sector, iosize, page_offset,
1957 end_bio_extent_readpage, mirror_num);
1962 page_offset += iosize;
1966 if (!PageError(page))
1967 SetPageUptodate(page);
1973 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
1974 get_extent_t *get_extent)
1976 struct bio *bio = NULL;
1979 ret = __extent_read_full_page(tree, page, get_extent, &bio, 0);
1981 submit_one_bio(READ, bio, 0);
1984 EXPORT_SYMBOL(extent_read_full_page);
1987 * the writepage semantics are similar to regular writepage. extent
1988 * records are inserted to lock ranges in the tree, and as dirty areas
1989 * are found, they are marked writeback. Then the lock bits are removed
1990 * and the end_io handler clears the writeback ranges
1992 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
1995 struct inode *inode = page->mapping->host;
1996 struct extent_page_data *epd = data;
1997 struct extent_io_tree *tree = epd->tree;
1998 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2000 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2004 u64 last_byte = i_size_read(inode);
2009 struct extent_map *em;
2010 struct block_device *bdev;
2013 size_t pg_offset = 0;
2015 loff_t i_size = i_size_read(inode);
2016 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
2020 WARN_ON(!PageLocked(page));
2021 pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
2022 if (page->index > end_index ||
2023 (page->index == end_index && !pg_offset)) {
2024 page->mapping->a_ops->invalidatepage(page, 0);
2029 if (page->index == end_index) {
2032 userpage = kmap_atomic(page, KM_USER0);
2033 memset(userpage + pg_offset, 0,
2034 PAGE_CACHE_SIZE - pg_offset);
2035 kunmap_atomic(userpage, KM_USER0);
2036 flush_dcache_page(page);
2040 set_page_extent_mapped(page);
2042 delalloc_start = start;
2044 while(delalloc_end < page_end) {
2045 nr_delalloc = find_lock_delalloc_range(tree, &delalloc_start,
2048 if (nr_delalloc == 0) {
2049 delalloc_start = delalloc_end + 1;
2052 tree->ops->fill_delalloc(inode, delalloc_start,
2054 clear_extent_bit(tree, delalloc_start,
2056 EXTENT_LOCKED | EXTENT_DELALLOC,
2058 delalloc_start = delalloc_end + 1;
2060 lock_extent(tree, start, page_end, GFP_NOFS);
2061 unlock_start = start;
2063 if (tree->ops && tree->ops->writepage_start_hook) {
2064 ret = tree->ops->writepage_start_hook(page, start, page_end);
2065 if (ret == -EAGAIN) {
2066 unlock_extent(tree, start, page_end, GFP_NOFS);
2067 redirty_page_for_writepage(wbc, page);
2074 if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0)) {
2075 printk("found delalloc bits after lock_extent\n");
2078 if (last_byte <= start) {
2079 clear_extent_dirty(tree, start, page_end, GFP_NOFS);
2080 unlock_extent(tree, start, page_end, GFP_NOFS);
2081 if (tree->ops && tree->ops->writepage_end_io_hook)
2082 tree->ops->writepage_end_io_hook(page, start,
2084 unlock_start = page_end + 1;
2088 set_extent_uptodate(tree, start, page_end, GFP_NOFS);
2089 blocksize = inode->i_sb->s_blocksize;
2091 while (cur <= end) {
2092 if (cur >= last_byte) {
2093 clear_extent_dirty(tree, cur, page_end, GFP_NOFS);
2094 unlock_extent(tree, unlock_start, page_end, GFP_NOFS);
2095 if (tree->ops && tree->ops->writepage_end_io_hook)
2096 tree->ops->writepage_end_io_hook(page, cur,
2098 unlock_start = page_end + 1;
2101 em = epd->get_extent(inode, page, pg_offset, cur,
2103 if (IS_ERR(em) || !em) {
2108 extent_offset = cur - em->start;
2109 BUG_ON(extent_map_end(em) <= cur);
2111 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2112 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2113 sector = (em->block_start + extent_offset) >> 9;
2115 block_start = em->block_start;
2116 free_extent_map(em);
2119 if (block_start == EXTENT_MAP_HOLE ||
2120 block_start == EXTENT_MAP_INLINE) {
2121 clear_extent_dirty(tree, cur,
2122 cur + iosize - 1, GFP_NOFS);
2124 unlock_extent(tree, unlock_start, cur + iosize -1,
2127 if (tree->ops && tree->ops->writepage_end_io_hook)
2128 tree->ops->writepage_end_io_hook(page, cur,
2132 pg_offset += iosize;
2137 /* leave this out until we have a page_mkwrite call */
2138 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
2141 pg_offset += iosize;
2144 clear_extent_dirty(tree, cur, cur + iosize - 1, GFP_NOFS);
2145 if (tree->ops && tree->ops->writepage_io_hook) {
2146 ret = tree->ops->writepage_io_hook(page, cur,
2154 unsigned long max_nr = end_index + 1;
2156 set_range_writeback(tree, cur, cur + iosize - 1);
2157 if (!PageWriteback(page)) {
2158 printk("warning page %lu not writeback, "
2159 "cur %llu end %llu\n", page->index,
2160 (unsigned long long)cur,
2161 (unsigned long long)end);
2164 ret = submit_extent_page(WRITE, tree, page, sector,
2165 iosize, pg_offset, bdev,
2167 end_bio_extent_writepage, 0);
2172 pg_offset += iosize;
2177 /* make sure the mapping tag for page dirty gets cleared */
2178 set_page_writeback(page);
2179 end_page_writeback(page);
2181 if (unlock_start <= page_end)
2182 unlock_extent(tree, unlock_start, page_end, GFP_NOFS);
2187 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,22)
2188 /* Taken directly from 2.6.23 for 2.6.18 back port */
2189 typedef int (*writepage_t)(struct page *page, struct writeback_control *wbc,
2193 * write_cache_pages - walk the list of dirty pages of the given address space
2194 * and write all of them.
2195 * @mapping: address space structure to write
2196 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2197 * @writepage: function called for each page
2198 * @data: data passed to writepage function
2200 * If a page is already under I/O, write_cache_pages() skips it, even
2201 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
2202 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
2203 * and msync() need to guarantee that all the data which was dirty at the time
2204 * the call was made get new I/O started against them. If wbc->sync_mode is
2205 * WB_SYNC_ALL then we were called for data integrity and we must wait for
2206 * existing IO to complete.
2208 static int write_cache_pages(struct address_space *mapping,
2209 struct writeback_control *wbc, writepage_t writepage,
2212 struct backing_dev_info *bdi = mapping->backing_dev_info;
2215 struct pagevec pvec;
2218 pgoff_t end; /* Inclusive */
2220 int range_whole = 0;
2222 if (wbc->nonblocking && bdi_write_congested(bdi)) {
2223 wbc->encountered_congestion = 1;
2227 pagevec_init(&pvec, 0);
2228 if (wbc->range_cyclic) {
2229 index = mapping->writeback_index; /* Start from prev offset */
2232 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2233 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2234 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2239 while (!done && (index <= end) &&
2240 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
2241 PAGECACHE_TAG_DIRTY,
2242 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
2246 for (i = 0; i < nr_pages; i++) {
2247 struct page *page = pvec.pages[i];
2250 * At this point we hold neither mapping->tree_lock nor
2251 * lock on the page itself: the page may be truncated or
2252 * invalidated (changing page->mapping to NULL), or even
2253 * swizzled back from swapper_space to tmpfs file
2258 if (unlikely(page->mapping != mapping)) {
2263 if (!wbc->range_cyclic && page->index > end) {
2269 if (wbc->sync_mode != WB_SYNC_NONE)
2270 wait_on_page_writeback(page);
2272 if (PageWriteback(page) ||
2273 !clear_page_dirty_for_io(page)) {
2278 ret = (*writepage)(page, wbc, data);
2280 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
2284 if (ret || (--(wbc->nr_to_write) <= 0))
2286 if (wbc->nonblocking && bdi_write_congested(bdi)) {
2287 wbc->encountered_congestion = 1;
2291 pagevec_release(&pvec);
2294 if (!scanned && !done) {
2296 * We hit the last page and there is more work to be done: wrap
2297 * back to the start of the file
2303 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2304 mapping->writeback_index = index;
2309 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
2310 get_extent_t *get_extent,
2311 struct writeback_control *wbc)
2314 struct address_space *mapping = page->mapping;
2315 struct extent_page_data epd = {
2318 .get_extent = get_extent,
2320 struct writeback_control wbc_writepages = {
2322 .sync_mode = WB_SYNC_NONE,
2323 .older_than_this = NULL,
2325 .range_start = page_offset(page) + PAGE_CACHE_SIZE,
2326 .range_end = (loff_t)-1,
2330 ret = __extent_writepage(page, wbc, &epd);
2332 write_cache_pages(mapping, &wbc_writepages, __extent_writepage, &epd);
2334 submit_one_bio(WRITE, epd.bio, 0);
2338 EXPORT_SYMBOL(extent_write_full_page);
2341 int extent_writepages(struct extent_io_tree *tree,
2342 struct address_space *mapping,
2343 get_extent_t *get_extent,
2344 struct writeback_control *wbc)
2347 struct extent_page_data epd = {
2350 .get_extent = get_extent,
2353 ret = write_cache_pages(mapping, wbc, __extent_writepage, &epd);
2355 submit_one_bio(WRITE, epd.bio, 0);
2359 EXPORT_SYMBOL(extent_writepages);
2361 int extent_readpages(struct extent_io_tree *tree,
2362 struct address_space *mapping,
2363 struct list_head *pages, unsigned nr_pages,
2364 get_extent_t get_extent)
2366 struct bio *bio = NULL;
2368 struct pagevec pvec;
2370 pagevec_init(&pvec, 0);
2371 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
2372 struct page *page = list_entry(pages->prev, struct page, lru);
2374 prefetchw(&page->flags);
2375 list_del(&page->lru);
2377 * what we want to do here is call add_to_page_cache_lru,
2378 * but that isn't exported, so we reproduce it here
2380 if (!add_to_page_cache(page, mapping,
2381 page->index, GFP_KERNEL)) {
2383 /* open coding of lru_cache_add, also not exported */
2384 page_cache_get(page);
2385 if (!pagevec_add(&pvec, page))
2386 __pagevec_lru_add(&pvec);
2387 __extent_read_full_page(tree, page, get_extent,
2390 page_cache_release(page);
2392 if (pagevec_count(&pvec))
2393 __pagevec_lru_add(&pvec);
2394 BUG_ON(!list_empty(pages));
2396 submit_one_bio(READ, bio, 0);
2399 EXPORT_SYMBOL(extent_readpages);
2402 * basic invalidatepage code, this waits on any locked or writeback
2403 * ranges corresponding to the page, and then deletes any extent state
2404 * records from the tree
2406 int extent_invalidatepage(struct extent_io_tree *tree,
2407 struct page *page, unsigned long offset)
2409 u64 start = ((u64)page->index << PAGE_CACHE_SHIFT);
2410 u64 end = start + PAGE_CACHE_SIZE - 1;
2411 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
2413 start += (offset + blocksize -1) & ~(blocksize - 1);
2417 lock_extent(tree, start, end, GFP_NOFS);
2418 wait_on_extent_writeback(tree, start, end);
2419 clear_extent_bit(tree, start, end,
2420 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC,
2424 EXPORT_SYMBOL(extent_invalidatepage);
2427 * simple commit_write call, set_range_dirty is used to mark both
2428 * the pages and the extent records as dirty
2430 int extent_commit_write(struct extent_io_tree *tree,
2431 struct inode *inode, struct page *page,
2432 unsigned from, unsigned to)
2434 loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
2436 set_page_extent_mapped(page);
2437 set_page_dirty(page);
2439 if (pos > inode->i_size) {
2440 i_size_write(inode, pos);
2441 mark_inode_dirty(inode);
2445 EXPORT_SYMBOL(extent_commit_write);
2447 int extent_prepare_write(struct extent_io_tree *tree,
2448 struct inode *inode, struct page *page,
2449 unsigned from, unsigned to, get_extent_t *get_extent)
2451 u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2452 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
2454 u64 orig_block_start;
2457 struct extent_map *em;
2458 unsigned blocksize = 1 << inode->i_blkbits;
2459 size_t page_offset = 0;
2460 size_t block_off_start;
2461 size_t block_off_end;
2467 set_page_extent_mapped(page);
2469 block_start = (page_start + from) & ~((u64)blocksize - 1);
2470 block_end = (page_start + to - 1) | (blocksize - 1);
2471 orig_block_start = block_start;
2473 lock_extent(tree, page_start, page_end, GFP_NOFS);
2474 while(block_start <= block_end) {
2475 em = get_extent(inode, page, page_offset, block_start,
2476 block_end - block_start + 1, 1);
2477 if (IS_ERR(em) || !em) {
2480 cur_end = min(block_end, extent_map_end(em) - 1);
2481 block_off_start = block_start & (PAGE_CACHE_SIZE - 1);
2482 block_off_end = block_off_start + blocksize;
2483 isnew = clear_extent_new(tree, block_start, cur_end, GFP_NOFS);
2485 if (!PageUptodate(page) && isnew &&
2486 (block_off_end > to || block_off_start < from)) {
2489 kaddr = kmap_atomic(page, KM_USER0);
2490 if (block_off_end > to)
2491 memset(kaddr + to, 0, block_off_end - to);
2492 if (block_off_start < from)
2493 memset(kaddr + block_off_start, 0,
2494 from - block_off_start);
2495 flush_dcache_page(page);
2496 kunmap_atomic(kaddr, KM_USER0);
2498 if ((em->block_start != EXTENT_MAP_HOLE &&
2499 em->block_start != EXTENT_MAP_INLINE) &&
2500 !isnew && !PageUptodate(page) &&
2501 (block_off_end > to || block_off_start < from) &&
2502 !test_range_bit(tree, block_start, cur_end,
2503 EXTENT_UPTODATE, 1)) {
2505 u64 extent_offset = block_start - em->start;
2507 sector = (em->block_start + extent_offset) >> 9;
2508 iosize = (cur_end - block_start + blocksize) &
2509 ~((u64)blocksize - 1);
2511 * we've already got the extent locked, but we
2512 * need to split the state such that our end_bio
2513 * handler can clear the lock.
2515 set_extent_bit(tree, block_start,
2516 block_start + iosize - 1,
2517 EXTENT_LOCKED, 0, NULL, GFP_NOFS);
2518 ret = submit_extent_page(READ, tree, page,
2519 sector, iosize, page_offset, em->bdev,
2521 end_bio_extent_preparewrite, 0);
2523 block_start = block_start + iosize;
2525 set_extent_uptodate(tree, block_start, cur_end,
2527 unlock_extent(tree, block_start, cur_end, GFP_NOFS);
2528 block_start = cur_end + 1;
2530 page_offset = block_start & (PAGE_CACHE_SIZE - 1);
2531 free_extent_map(em);
2534 wait_extent_bit(tree, orig_block_start,
2535 block_end, EXTENT_LOCKED);
2537 check_page_uptodate(tree, page);
2539 /* FIXME, zero out newly allocated blocks on error */
2542 EXPORT_SYMBOL(extent_prepare_write);
2545 * a helper for releasepage, this tests for areas of the page that
2546 * are locked or under IO and drops the related state bits if it is safe
2549 int try_release_extent_state(struct extent_map_tree *map,
2550 struct extent_io_tree *tree, struct page *page,
2553 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2554 u64 end = start + PAGE_CACHE_SIZE - 1;
2557 if (test_range_bit(tree, start, end,
2558 EXTENT_IOBITS | EXTENT_ORDERED, 0))
2561 if ((mask & GFP_NOFS) == GFP_NOFS)
2563 clear_extent_bit(tree, start, end, EXTENT_UPTODATE,
2568 EXPORT_SYMBOL(try_release_extent_state);
2571 * a helper for releasepage. As long as there are no locked extents
2572 * in the range corresponding to the page, both state records and extent
2573 * map records are removed
2575 int try_release_extent_mapping(struct extent_map_tree *map,
2576 struct extent_io_tree *tree, struct page *page,
2579 struct extent_map *em;
2580 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2581 u64 end = start + PAGE_CACHE_SIZE - 1;
2583 if ((mask & __GFP_WAIT) &&
2584 page->mapping->host->i_size > 16 * 1024 * 1024) {
2586 while (start <= end) {
2587 len = end - start + 1;
2588 spin_lock(&map->lock);
2589 em = lookup_extent_mapping(map, start, len);
2590 if (!em || IS_ERR(em)) {
2591 spin_unlock(&map->lock);
2594 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
2595 em->start != start) {
2596 spin_unlock(&map->lock);
2597 free_extent_map(em);
2600 if (!test_range_bit(tree, em->start,
2601 extent_map_end(em) - 1,
2602 EXTENT_LOCKED, 0)) {
2603 remove_extent_mapping(map, em);
2604 /* once for the rb tree */
2605 free_extent_map(em);
2607 start = extent_map_end(em);
2608 spin_unlock(&map->lock);
2611 free_extent_map(em);
2614 return try_release_extent_state(map, tree, page, mask);
2616 EXPORT_SYMBOL(try_release_extent_mapping);
2618 sector_t extent_bmap(struct address_space *mapping, sector_t iblock,
2619 get_extent_t *get_extent)
2621 struct inode *inode = mapping->host;
2622 u64 start = iblock << inode->i_blkbits;
2623 sector_t sector = 0;
2624 struct extent_map *em;
2626 em = get_extent(inode, NULL, 0, start, (1 << inode->i_blkbits), 0);
2627 if (!em || IS_ERR(em))
2630 if (em->block_start == EXTENT_MAP_INLINE ||
2631 em->block_start == EXTENT_MAP_HOLE)
2634 sector = (em->block_start + start - em->start) >> inode->i_blkbits;
2636 free_extent_map(em);
2640 static inline struct page *extent_buffer_page(struct extent_buffer *eb,
2644 struct address_space *mapping;
2647 return eb->first_page;
2648 i += eb->start >> PAGE_CACHE_SHIFT;
2649 mapping = eb->first_page->mapping;
2650 read_lock_irq(&mapping->tree_lock);
2651 p = radix_tree_lookup(&mapping->page_tree, i);
2652 read_unlock_irq(&mapping->tree_lock);
2656 static inline unsigned long num_extent_pages(u64 start, u64 len)
2658 return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
2659 (start >> PAGE_CACHE_SHIFT);
2662 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
2667 struct extent_buffer *eb = NULL;
2668 unsigned long flags;
2670 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
2673 mutex_init(&eb->mutex);
2674 spin_lock_irqsave(&leak_lock, flags);
2675 list_add(&eb->leak_list, &buffers);
2676 spin_unlock_irqrestore(&leak_lock, flags);
2677 atomic_set(&eb->refs, 1);
2682 static void __free_extent_buffer(struct extent_buffer *eb)
2684 unsigned long flags;
2685 spin_lock_irqsave(&leak_lock, flags);
2686 list_del(&eb->leak_list);
2687 spin_unlock_irqrestore(&leak_lock, flags);
2688 kmem_cache_free(extent_buffer_cache, eb);
2691 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
2692 u64 start, unsigned long len,
2696 unsigned long num_pages = num_extent_pages(start, len);
2698 unsigned long index = start >> PAGE_CACHE_SHIFT;
2699 struct extent_buffer *eb;
2700 struct extent_buffer *exists = NULL;
2702 struct address_space *mapping = tree->mapping;
2705 spin_lock(&tree->buffer_lock);
2706 eb = buffer_search(tree, start);
2708 atomic_inc(&eb->refs);
2709 spin_unlock(&tree->buffer_lock);
2712 spin_unlock(&tree->buffer_lock);
2714 eb = __alloc_extent_buffer(tree, start, len, mask);
2719 eb->first_page = page0;
2722 page_cache_get(page0);
2723 mark_page_accessed(page0);
2724 set_page_extent_mapped(page0);
2725 set_page_extent_head(page0, len);
2726 uptodate = PageUptodate(page0);
2730 for (; i < num_pages; i++, index++) {
2731 p = find_or_create_page(mapping, index, mask | __GFP_HIGHMEM);
2736 set_page_extent_mapped(p);
2737 mark_page_accessed(p);
2740 set_page_extent_head(p, len);
2742 set_page_private(p, EXTENT_PAGE_PRIVATE);
2744 if (!PageUptodate(p))
2749 eb->flags |= EXTENT_UPTODATE;
2750 eb->flags |= EXTENT_BUFFER_FILLED;
2752 spin_lock(&tree->buffer_lock);
2753 exists = buffer_tree_insert(tree, start, &eb->rb_node);
2755 /* add one reference for the caller */
2756 atomic_inc(&exists->refs);
2757 spin_unlock(&tree->buffer_lock);
2760 spin_unlock(&tree->buffer_lock);
2762 /* add one reference for the tree */
2763 atomic_inc(&eb->refs);
2767 if (!atomic_dec_and_test(&eb->refs))
2769 for (index = 1; index < i; index++)
2770 page_cache_release(extent_buffer_page(eb, index));
2771 page_cache_release(extent_buffer_page(eb, 0));
2772 __free_extent_buffer(eb);
2775 EXPORT_SYMBOL(alloc_extent_buffer);
2777 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
2778 u64 start, unsigned long len,
2781 struct extent_buffer *eb;
2783 spin_lock(&tree->buffer_lock);
2784 eb = buffer_search(tree, start);
2786 atomic_inc(&eb->refs);
2787 spin_unlock(&tree->buffer_lock);
2791 EXPORT_SYMBOL(find_extent_buffer);
2793 void free_extent_buffer(struct extent_buffer *eb)
2798 if (!atomic_dec_and_test(&eb->refs))
2803 EXPORT_SYMBOL(free_extent_buffer);
2805 int clear_extent_buffer_dirty(struct extent_io_tree *tree,
2806 struct extent_buffer *eb)
2810 unsigned long num_pages;
2813 u64 start = eb->start;
2814 u64 end = start + eb->len - 1;
2816 set = clear_extent_dirty(tree, start, end, GFP_NOFS);
2817 num_pages = num_extent_pages(eb->start, eb->len);
2819 for (i = 0; i < num_pages; i++) {
2820 page = extent_buffer_page(eb, i);
2823 set_page_extent_head(page, eb->len);
2825 set_page_private(page, EXTENT_PAGE_PRIVATE);
2828 * if we're on the last page or the first page and the
2829 * block isn't aligned on a page boundary, do extra checks
2830 * to make sure we don't clean page that is partially dirty
2832 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
2833 ((i == num_pages - 1) &&
2834 ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
2835 start = (u64)page->index << PAGE_CACHE_SHIFT;
2836 end = start + PAGE_CACHE_SIZE - 1;
2837 if (test_range_bit(tree, start, end,
2843 clear_page_dirty_for_io(page);
2844 read_lock_irq(&page->mapping->tree_lock);
2845 if (!PageDirty(page)) {
2846 radix_tree_tag_clear(&page->mapping->page_tree,
2848 PAGECACHE_TAG_DIRTY);
2850 read_unlock_irq(&page->mapping->tree_lock);
2855 EXPORT_SYMBOL(clear_extent_buffer_dirty);
2857 int wait_on_extent_buffer_writeback(struct extent_io_tree *tree,
2858 struct extent_buffer *eb)
2860 return wait_on_extent_writeback(tree, eb->start,
2861 eb->start + eb->len - 1);
2863 EXPORT_SYMBOL(wait_on_extent_buffer_writeback);
2865 int set_extent_buffer_dirty(struct extent_io_tree *tree,
2866 struct extent_buffer *eb)
2869 unsigned long num_pages;
2871 num_pages = num_extent_pages(eb->start, eb->len);
2872 for (i = 0; i < num_pages; i++) {
2873 struct page *page = extent_buffer_page(eb, i);
2874 /* writepage may need to do something special for the
2875 * first page, we have to make sure page->private is
2876 * properly set. releasepage may drop page->private
2877 * on us if the page isn't already dirty.
2881 set_page_extent_head(page, eb->len);
2882 } else if (PagePrivate(page) &&
2883 page->private != EXTENT_PAGE_PRIVATE) {
2885 set_page_extent_mapped(page);
2888 __set_page_dirty_nobuffers(extent_buffer_page(eb, i));
2892 return set_extent_dirty(tree, eb->start,
2893 eb->start + eb->len - 1, GFP_NOFS);
2895 EXPORT_SYMBOL(set_extent_buffer_dirty);
2897 int clear_extent_buffer_uptodate(struct extent_io_tree *tree,
2898 struct extent_buffer *eb)
2902 unsigned long num_pages;
2904 num_pages = num_extent_pages(eb->start, eb->len);
2905 eb->flags &= ~EXTENT_UPTODATE;
2907 clear_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
2909 for (i = 0; i < num_pages; i++) {
2910 page = extent_buffer_page(eb, i);
2911 ClearPageUptodate(page);
2916 int set_extent_buffer_uptodate(struct extent_io_tree *tree,
2917 struct extent_buffer *eb)
2921 unsigned long num_pages;
2923 num_pages = num_extent_pages(eb->start, eb->len);
2925 set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
2927 for (i = 0; i < num_pages; i++) {
2928 page = extent_buffer_page(eb, i);
2929 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
2930 ((i == num_pages - 1) &&
2931 ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
2932 check_page_uptodate(tree, page);
2935 SetPageUptodate(page);
2939 EXPORT_SYMBOL(set_extent_buffer_uptodate);
2941 int extent_range_uptodate(struct extent_io_tree *tree,
2946 int pg_uptodate = 1;
2948 unsigned long index;
2950 ret = test_range_bit(tree, start, end, EXTENT_UPTODATE, 1);
2953 while(start <= end) {
2954 index = start >> PAGE_CACHE_SHIFT;
2955 page = find_get_page(tree->mapping, index);
2956 uptodate = PageUptodate(page);
2957 page_cache_release(page);
2962 start += PAGE_CACHE_SIZE;
2967 int extent_buffer_uptodate(struct extent_io_tree *tree,
2968 struct extent_buffer *eb)
2971 unsigned long num_pages;
2974 int pg_uptodate = 1;
2976 if (eb->flags & EXTENT_UPTODATE)
2979 ret = test_range_bit(tree, eb->start, eb->start + eb->len - 1,
2980 EXTENT_UPTODATE, 1);
2984 num_pages = num_extent_pages(eb->start, eb->len);
2985 for (i = 0; i < num_pages; i++) {
2986 page = extent_buffer_page(eb, i);
2987 if (!PageUptodate(page)) {
2994 EXPORT_SYMBOL(extent_buffer_uptodate);
2996 int read_extent_buffer_pages(struct extent_io_tree *tree,
2997 struct extent_buffer *eb,
2998 u64 start, int wait,
2999 get_extent_t *get_extent, int mirror_num)
3002 unsigned long start_i;
3006 int locked_pages = 0;
3007 int all_uptodate = 1;
3008 int inc_all_pages = 0;
3009 unsigned long num_pages;
3010 struct bio *bio = NULL;
3012 if (eb->flags & EXTENT_UPTODATE)
3015 if (test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3016 EXTENT_UPTODATE, 1)) {
3021 WARN_ON(start < eb->start);
3022 start_i = (start >> PAGE_CACHE_SHIFT) -
3023 (eb->start >> PAGE_CACHE_SHIFT);
3028 num_pages = num_extent_pages(eb->start, eb->len);
3029 for (i = start_i; i < num_pages; i++) {
3030 page = extent_buffer_page(eb, i);
3032 if (TestSetPageLocked(page))
3038 if (!PageUptodate(page)) {
3044 eb->flags |= EXTENT_UPTODATE;
3048 for (i = start_i; i < num_pages; i++) {
3049 page = extent_buffer_page(eb, i);
3051 page_cache_get(page);
3052 if (!PageUptodate(page)) {
3055 ClearPageError(page);
3056 err = __extent_read_full_page(tree, page,
3068 submit_one_bio(READ, bio, mirror_num);
3073 for (i = start_i; i < num_pages; i++) {
3074 page = extent_buffer_page(eb, i);
3075 wait_on_page_locked(page);
3076 if (!PageUptodate(page)) {
3081 eb->flags |= EXTENT_UPTODATE;
3086 while(locked_pages > 0) {
3087 page = extent_buffer_page(eb, i);
3094 EXPORT_SYMBOL(read_extent_buffer_pages);
3096 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
3097 unsigned long start,
3104 char *dst = (char *)dstv;
3105 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3106 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3108 WARN_ON(start > eb->len);
3109 WARN_ON(start + len > eb->start + eb->len);
3111 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3114 page = extent_buffer_page(eb, i);
3116 cur = min(len, (PAGE_CACHE_SIZE - offset));
3117 kaddr = kmap_atomic(page, KM_USER1);
3118 memcpy(dst, kaddr + offset, cur);
3119 kunmap_atomic(kaddr, KM_USER1);
3127 EXPORT_SYMBOL(read_extent_buffer);
3129 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
3130 unsigned long min_len, char **token, char **map,
3131 unsigned long *map_start,
3132 unsigned long *map_len, int km)
3134 size_t offset = start & (PAGE_CACHE_SIZE - 1);
3137 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3138 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3139 unsigned long end_i = (start_offset + start + min_len - 1) >>
3146 offset = start_offset;
3150 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
3152 if (start + min_len > eb->len) {
3153 printk("bad mapping eb start %Lu len %lu, wanted %lu %lu\n", eb->start, eb->len, start, min_len);
3157 p = extent_buffer_page(eb, i);
3158 kaddr = kmap_atomic(p, km);
3160 *map = kaddr + offset;
3161 *map_len = PAGE_CACHE_SIZE - offset;
3164 EXPORT_SYMBOL(map_private_extent_buffer);
3166 int map_extent_buffer(struct extent_buffer *eb, unsigned long start,
3167 unsigned long min_len,
3168 char **token, char **map,
3169 unsigned long *map_start,
3170 unsigned long *map_len, int km)
3174 if (eb->map_token) {
3175 unmap_extent_buffer(eb, eb->map_token, km);
3176 eb->map_token = NULL;
3179 err = map_private_extent_buffer(eb, start, min_len, token, map,
3180 map_start, map_len, km);
3182 eb->map_token = *token;
3184 eb->map_start = *map_start;
3185 eb->map_len = *map_len;
3189 EXPORT_SYMBOL(map_extent_buffer);
3191 void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km)
3193 kunmap_atomic(token, km);
3195 EXPORT_SYMBOL(unmap_extent_buffer);
3197 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
3198 unsigned long start,
3205 char *ptr = (char *)ptrv;
3206 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3207 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3210 WARN_ON(start > eb->len);
3211 WARN_ON(start + len > eb->start + eb->len);
3213 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3216 page = extent_buffer_page(eb, i);
3218 cur = min(len, (PAGE_CACHE_SIZE - offset));
3220 kaddr = kmap_atomic(page, KM_USER0);
3221 ret = memcmp(ptr, kaddr + offset, cur);
3222 kunmap_atomic(kaddr, KM_USER0);
3233 EXPORT_SYMBOL(memcmp_extent_buffer);
3235 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
3236 unsigned long start, unsigned long len)
3242 char *src = (char *)srcv;
3243 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3244 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3246 WARN_ON(start > eb->len);
3247 WARN_ON(start + len > eb->start + eb->len);
3249 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3252 page = extent_buffer_page(eb, i);
3253 WARN_ON(!PageUptodate(page));
3255 cur = min(len, PAGE_CACHE_SIZE - offset);
3256 kaddr = kmap_atomic(page, KM_USER1);
3257 memcpy(kaddr + offset, src, cur);
3258 kunmap_atomic(kaddr, KM_USER1);
3266 EXPORT_SYMBOL(write_extent_buffer);
3268 void memset_extent_buffer(struct extent_buffer *eb, char c,
3269 unsigned long start, unsigned long len)
3275 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3276 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3278 WARN_ON(start > eb->len);
3279 WARN_ON(start + len > eb->start + eb->len);
3281 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3284 page = extent_buffer_page(eb, i);
3285 WARN_ON(!PageUptodate(page));
3287 cur = min(len, PAGE_CACHE_SIZE - offset);
3288 kaddr = kmap_atomic(page, KM_USER0);
3289 memset(kaddr + offset, c, cur);
3290 kunmap_atomic(kaddr, KM_USER0);
3297 EXPORT_SYMBOL(memset_extent_buffer);
3299 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
3300 unsigned long dst_offset, unsigned long src_offset,
3303 u64 dst_len = dst->len;
3308 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3309 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3311 WARN_ON(src->len != dst_len);
3313 offset = (start_offset + dst_offset) &
3314 ((unsigned long)PAGE_CACHE_SIZE - 1);
3317 page = extent_buffer_page(dst, i);
3318 WARN_ON(!PageUptodate(page));
3320 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
3322 kaddr = kmap_atomic(page, KM_USER0);
3323 read_extent_buffer(src, kaddr + offset, src_offset, cur);
3324 kunmap_atomic(kaddr, KM_USER0);
3332 EXPORT_SYMBOL(copy_extent_buffer);
3334 static void move_pages(struct page *dst_page, struct page *src_page,
3335 unsigned long dst_off, unsigned long src_off,
3338 char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3339 if (dst_page == src_page) {
3340 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
3342 char *src_kaddr = kmap_atomic(src_page, KM_USER1);
3343 char *p = dst_kaddr + dst_off + len;
3344 char *s = src_kaddr + src_off + len;
3349 kunmap_atomic(src_kaddr, KM_USER1);
3351 kunmap_atomic(dst_kaddr, KM_USER0);
3354 static void copy_pages(struct page *dst_page, struct page *src_page,
3355 unsigned long dst_off, unsigned long src_off,
3358 char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3361 if (dst_page != src_page)
3362 src_kaddr = kmap_atomic(src_page, KM_USER1);
3364 src_kaddr = dst_kaddr;
3366 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
3367 kunmap_atomic(dst_kaddr, KM_USER0);
3368 if (dst_page != src_page)
3369 kunmap_atomic(src_kaddr, KM_USER1);
3372 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3373 unsigned long src_offset, unsigned long len)
3376 size_t dst_off_in_page;
3377 size_t src_off_in_page;
3378 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3379 unsigned long dst_i;
3380 unsigned long src_i;
3382 if (src_offset + len > dst->len) {
3383 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
3384 src_offset, len, dst->len);
3387 if (dst_offset + len > dst->len) {
3388 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
3389 dst_offset, len, dst->len);
3394 dst_off_in_page = (start_offset + dst_offset) &
3395 ((unsigned long)PAGE_CACHE_SIZE - 1);
3396 src_off_in_page = (start_offset + src_offset) &
3397 ((unsigned long)PAGE_CACHE_SIZE - 1);
3399 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3400 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
3402 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
3404 cur = min_t(unsigned long, cur,
3405 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
3407 copy_pages(extent_buffer_page(dst, dst_i),
3408 extent_buffer_page(dst, src_i),
3409 dst_off_in_page, src_off_in_page, cur);
3416 EXPORT_SYMBOL(memcpy_extent_buffer);
3418 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3419 unsigned long src_offset, unsigned long len)
3422 size_t dst_off_in_page;
3423 size_t src_off_in_page;
3424 unsigned long dst_end = dst_offset + len - 1;
3425 unsigned long src_end = src_offset + len - 1;
3426 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3427 unsigned long dst_i;
3428 unsigned long src_i;
3430 if (src_offset + len > dst->len) {
3431 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
3432 src_offset, len, dst->len);
3435 if (dst_offset + len > dst->len) {
3436 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
3437 dst_offset, len, dst->len);
3440 if (dst_offset < src_offset) {
3441 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
3445 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
3446 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
3448 dst_off_in_page = (start_offset + dst_end) &
3449 ((unsigned long)PAGE_CACHE_SIZE - 1);
3450 src_off_in_page = (start_offset + src_end) &
3451 ((unsigned long)PAGE_CACHE_SIZE - 1);
3453 cur = min_t(unsigned long, len, src_off_in_page + 1);
3454 cur = min(cur, dst_off_in_page + 1);
3455 move_pages(extent_buffer_page(dst, dst_i),
3456 extent_buffer_page(dst, src_i),
3457 dst_off_in_page - cur + 1,
3458 src_off_in_page - cur + 1, cur);
3465 EXPORT_SYMBOL(memmove_extent_buffer);
3467 int try_release_extent_buffer(struct extent_io_tree *tree, struct page *page)
3469 u64 start = page_offset(page);
3470 struct extent_buffer *eb;
3473 unsigned long num_pages;
3475 spin_lock(&tree->buffer_lock);
3476 eb = buffer_search(tree, start);
3480 if (atomic_read(&eb->refs) > 1) {
3484 /* at this point we can safely release the extent buffer */
3485 num_pages = num_extent_pages(eb->start, eb->len);
3486 for (i = 0; i < num_pages; i++) {
3487 struct page *page = extent_buffer_page(eb, i);
3488 page_cache_release(page);
3490 rb_erase(&eb->rb_node, &tree->buffer);
3491 __free_extent_buffer(eb);
3493 spin_unlock(&tree->buffer_lock);
3496 EXPORT_SYMBOL(try_release_extent_buffer);