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;
95 tree->dirty_bytes = 0;
96 spin_lock_init(&tree->lock);
97 spin_lock_init(&tree->lru_lock);
98 tree->mapping = mapping;
99 INIT_LIST_HEAD(&tree->buffer_lru);
103 EXPORT_SYMBOL(extent_io_tree_init);
105 void extent_io_tree_empty_lru(struct extent_io_tree *tree)
107 struct extent_buffer *eb;
108 while(!list_empty(&tree->buffer_lru)) {
109 eb = list_entry(tree->buffer_lru.next, struct extent_buffer,
111 list_del_init(&eb->lru);
112 free_extent_buffer(eb);
115 EXPORT_SYMBOL(extent_io_tree_empty_lru);
117 struct extent_state *alloc_extent_state(gfp_t mask)
119 struct extent_state *state;
122 state = kmem_cache_alloc(extent_state_cache, mask);
128 spin_lock_irqsave(&leak_lock, flags);
129 list_add(&state->leak_list, &states);
130 spin_unlock_irqrestore(&leak_lock, flags);
132 atomic_set(&state->refs, 1);
133 init_waitqueue_head(&state->wq);
136 EXPORT_SYMBOL(alloc_extent_state);
138 void free_extent_state(struct extent_state *state)
142 if (atomic_dec_and_test(&state->refs)) {
144 WARN_ON(state->tree);
145 spin_lock_irqsave(&leak_lock, flags);
146 list_del(&state->leak_list);
147 spin_unlock_irqrestore(&leak_lock, flags);
148 kmem_cache_free(extent_state_cache, state);
151 EXPORT_SYMBOL(free_extent_state);
153 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
154 struct rb_node *node)
156 struct rb_node ** p = &root->rb_node;
157 struct rb_node * parent = NULL;
158 struct tree_entry *entry;
162 entry = rb_entry(parent, struct tree_entry, rb_node);
164 if (offset < entry->start)
166 else if (offset > entry->end)
172 entry = rb_entry(node, struct tree_entry, rb_node);
173 rb_link_node(node, parent, p);
174 rb_insert_color(node, root);
178 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
179 struct rb_node **prev_ret,
180 struct rb_node **next_ret)
182 struct rb_root *root = &tree->state;
183 struct rb_node * n = root->rb_node;
184 struct rb_node *prev = NULL;
185 struct rb_node *orig_prev = NULL;
186 struct tree_entry *entry;
187 struct tree_entry *prev_entry = NULL;
190 struct extent_state *state;
192 if (state->start <= offset && offset <= state->end)
193 return &tree->last->rb_node;
196 entry = rb_entry(n, struct tree_entry, rb_node);
200 if (offset < entry->start)
202 else if (offset > entry->end)
205 tree->last = rb_entry(n, struct extent_state, rb_node);
212 while(prev && offset > prev_entry->end) {
213 prev = rb_next(prev);
214 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
221 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
222 while(prev && offset < prev_entry->start) {
223 prev = rb_prev(prev);
224 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
231 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
234 struct rb_node *prev = NULL;
237 ret = __etree_search(tree, offset, &prev, NULL);
240 tree->last = rb_entry(prev, struct extent_state,
249 * utility function to look for merge candidates inside a given range.
250 * Any extents with matching state are merged together into a single
251 * extent in the tree. Extents with EXTENT_IO in their state field
252 * are not merged because the end_io handlers need to be able to do
253 * operations on them without sleeping (or doing allocations/splits).
255 * This should be called with the tree lock held.
257 static int merge_state(struct extent_io_tree *tree,
258 struct extent_state *state)
260 struct extent_state *other;
261 struct rb_node *other_node;
263 if (state->state & EXTENT_IOBITS)
266 other_node = rb_prev(&state->rb_node);
268 other = rb_entry(other_node, struct extent_state, rb_node);
269 if (other->end == state->start - 1 &&
270 other->state == state->state) {
271 state->start = other->start;
273 if (tree->last == other)
275 rb_erase(&other->rb_node, &tree->state);
276 free_extent_state(other);
279 other_node = rb_next(&state->rb_node);
281 other = rb_entry(other_node, struct extent_state, rb_node);
282 if (other->start == state->end + 1 &&
283 other->state == state->state) {
284 other->start = state->start;
286 if (tree->last == state)
288 rb_erase(&state->rb_node, &tree->state);
289 free_extent_state(state);
295 static void set_state_cb(struct extent_io_tree *tree,
296 struct extent_state *state,
299 if (tree->ops && tree->ops->set_bit_hook) {
300 tree->ops->set_bit_hook(tree->mapping->host, state->start,
301 state->end, state->state, bits);
305 static void clear_state_cb(struct extent_io_tree *tree,
306 struct extent_state *state,
309 if (tree->ops && tree->ops->set_bit_hook) {
310 tree->ops->clear_bit_hook(tree->mapping->host, state->start,
311 state->end, state->state, bits);
316 * insert an extent_state struct into the tree. 'bits' are set on the
317 * struct before it is inserted.
319 * This may return -EEXIST if the extent is already there, in which case the
320 * state struct is freed.
322 * The tree lock is not taken internally. This is a utility function and
323 * probably isn't what you want to call (see set/clear_extent_bit).
325 static int insert_state(struct extent_io_tree *tree,
326 struct extent_state *state, u64 start, u64 end,
329 struct rb_node *node;
332 printk("end < start %Lu %Lu\n", end, start);
335 if (bits & EXTENT_DIRTY)
336 tree->dirty_bytes += end - start + 1;
337 set_state_cb(tree, state, bits);
338 state->state |= bits;
339 state->start = start;
341 node = tree_insert(&tree->state, end, &state->rb_node);
343 struct extent_state *found;
344 found = rb_entry(node, struct extent_state, rb_node);
345 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, start, end);
346 free_extent_state(state);
351 merge_state(tree, state);
356 * split a given extent state struct in two, inserting the preallocated
357 * struct 'prealloc' as the newly created second half. 'split' indicates an
358 * offset inside 'orig' where it should be split.
361 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
362 * are two extent state structs in the tree:
363 * prealloc: [orig->start, split - 1]
364 * orig: [ split, orig->end ]
366 * The tree locks are not taken by this function. They need to be held
369 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
370 struct extent_state *prealloc, u64 split)
372 struct rb_node *node;
373 prealloc->start = orig->start;
374 prealloc->end = split - 1;
375 prealloc->state = orig->state;
378 node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
380 struct extent_state *found;
381 found = rb_entry(node, struct extent_state, rb_node);
382 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, prealloc->start, prealloc->end);
383 free_extent_state(prealloc);
386 prealloc->tree = tree;
391 * utility function to clear some bits in an extent state struct.
392 * it will optionally wake up any one waiting on this state (wake == 1), or
393 * forcibly remove the state from the tree (delete == 1).
395 * If no bits are set on the state struct after clearing things, the
396 * struct is freed and removed from the tree
398 static int clear_state_bit(struct extent_io_tree *tree,
399 struct extent_state *state, int bits, int wake,
402 int ret = state->state & bits;
404 if ((bits & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
405 u64 range = state->end - state->start + 1;
406 WARN_ON(range > tree->dirty_bytes);
407 tree->dirty_bytes -= range;
409 clear_state_cb(tree, state, bits);
410 state->state &= ~bits;
413 if (delete || state->state == 0) {
415 clear_state_cb(tree, state, state->state);
416 if (tree->last == state) {
417 tree->last = extent_state_next(state);
419 rb_erase(&state->rb_node, &tree->state);
421 free_extent_state(state);
426 merge_state(tree, state);
432 * clear some bits on a range in the tree. This may require splitting
433 * or inserting elements in the tree, so the gfp mask is used to
434 * indicate which allocations or sleeping are allowed.
436 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
437 * the given range from the tree regardless of state (ie for truncate).
439 * the range [start, end] is inclusive.
441 * This takes the tree lock, and returns < 0 on error, > 0 if any of the
442 * bits were already set, or zero if none of the bits were already set.
444 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
445 int bits, int wake, int delete, gfp_t mask)
447 struct extent_state *state;
448 struct extent_state *prealloc = NULL;
449 struct rb_node *node;
455 if (!prealloc && (mask & __GFP_WAIT)) {
456 prealloc = alloc_extent_state(mask);
461 spin_lock_irqsave(&tree->lock, flags);
463 * this search will find the extents that end after
466 node = tree_search(tree, start);
469 state = rb_entry(node, struct extent_state, rb_node);
470 if (state->start > end)
472 WARN_ON(state->end < start);
475 * | ---- desired range ---- |
477 * | ------------- state -------------- |
479 * We need to split the extent we found, and may flip
480 * bits on second half.
482 * If the extent we found extends past our range, we
483 * just split and search again. It'll get split again
484 * the next time though.
486 * If the extent we found is inside our range, we clear
487 * the desired bit on it.
490 if (state->start < start) {
492 prealloc = alloc_extent_state(GFP_ATOMIC);
493 err = split_state(tree, state, prealloc, start);
494 BUG_ON(err == -EEXIST);
498 if (state->end <= end) {
499 start = state->end + 1;
500 set |= clear_state_bit(tree, state, bits,
503 start = state->start;
508 * | ---- desired range ---- |
510 * We need to split the extent, and clear the bit
513 if (state->start <= end && state->end > end) {
515 prealloc = alloc_extent_state(GFP_ATOMIC);
516 err = split_state(tree, state, prealloc, end + 1);
517 BUG_ON(err == -EEXIST);
521 set |= clear_state_bit(tree, prealloc, bits,
527 start = state->end + 1;
528 set |= clear_state_bit(tree, state, bits, wake, delete);
532 spin_unlock_irqrestore(&tree->lock, flags);
534 free_extent_state(prealloc);
541 spin_unlock_irqrestore(&tree->lock, flags);
542 if (mask & __GFP_WAIT)
546 EXPORT_SYMBOL(clear_extent_bit);
548 static int wait_on_state(struct extent_io_tree *tree,
549 struct extent_state *state)
552 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
553 spin_unlock_irq(&tree->lock);
555 spin_lock_irq(&tree->lock);
556 finish_wait(&state->wq, &wait);
561 * waits for one or more bits to clear on a range in the state tree.
562 * The range [start, end] is inclusive.
563 * The tree lock is taken by this function
565 int wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits)
567 struct extent_state *state;
568 struct rb_node *node;
570 spin_lock_irq(&tree->lock);
574 * this search will find all the extents that end after
577 node = tree_search(tree, start);
581 state = rb_entry(node, struct extent_state, rb_node);
583 if (state->start > end)
586 if (state->state & bits) {
587 start = state->start;
588 atomic_inc(&state->refs);
589 wait_on_state(tree, state);
590 free_extent_state(state);
593 start = state->end + 1;
598 if (need_resched()) {
599 spin_unlock_irq(&tree->lock);
601 spin_lock_irq(&tree->lock);
605 spin_unlock_irq(&tree->lock);
608 EXPORT_SYMBOL(wait_extent_bit);
610 static void set_state_bits(struct extent_io_tree *tree,
611 struct extent_state *state,
614 if ((bits & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
615 u64 range = state->end - state->start + 1;
616 tree->dirty_bytes += range;
618 set_state_cb(tree, state, bits);
619 state->state |= bits;
623 * set some bits on a range in the tree. This may require allocations
624 * or sleeping, so the gfp mask is used to indicate what is allowed.
626 * If 'exclusive' == 1, this will fail with -EEXIST if some part of the
627 * range already has the desired bits set. The start of the existing
628 * range is returned in failed_start in this case.
630 * [start, end] is inclusive
631 * This takes the tree lock.
633 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits,
634 int exclusive, u64 *failed_start, gfp_t mask)
636 struct extent_state *state;
637 struct extent_state *prealloc = NULL;
638 struct rb_node *node;
645 if (!prealloc && (mask & __GFP_WAIT)) {
646 prealloc = alloc_extent_state(mask);
651 spin_lock_irqsave(&tree->lock, flags);
653 * this search will find all the extents that end after
656 node = tree_search(tree, start);
658 err = insert_state(tree, prealloc, start, end, bits);
660 BUG_ON(err == -EEXIST);
664 state = rb_entry(node, struct extent_state, rb_node);
665 last_start = state->start;
666 last_end = state->end;
669 * | ---- desired range ---- |
672 * Just lock what we found and keep going
674 if (state->start == start && state->end <= end) {
675 set = state->state & bits;
676 if (set && exclusive) {
677 *failed_start = state->start;
681 set_state_bits(tree, state, bits);
682 start = state->end + 1;
683 merge_state(tree, state);
688 * | ---- desired range ---- |
691 * | ------------- state -------------- |
693 * We need to split the extent we found, and may flip bits on
696 * If the extent we found extends past our
697 * range, we just split and search again. It'll get split
698 * again the next time though.
700 * If the extent we found is inside our range, we set the
703 if (state->start < start) {
704 set = state->state & bits;
705 if (exclusive && set) {
706 *failed_start = start;
710 err = split_state(tree, state, prealloc, start);
711 BUG_ON(err == -EEXIST);
715 if (state->end <= end) {
716 set_state_bits(tree, state, bits);
717 start = state->end + 1;
718 merge_state(tree, state);
720 start = state->start;
725 * | ---- desired range ---- |
726 * | state | or | state |
728 * There's a hole, we need to insert something in it and
729 * ignore the extent we found.
731 if (state->start > start) {
733 if (end < last_start)
736 this_end = last_start -1;
737 err = insert_state(tree, prealloc, start, this_end,
740 BUG_ON(err == -EEXIST);
743 start = this_end + 1;
747 * | ---- desired range ---- |
749 * We need to split the extent, and set the bit
752 if (state->start <= end && state->end > end) {
753 set = state->state & bits;
754 if (exclusive && set) {
755 *failed_start = start;
759 err = split_state(tree, state, prealloc, end + 1);
760 BUG_ON(err == -EEXIST);
762 set_state_bits(tree, prealloc, bits);
763 merge_state(tree, prealloc);
771 spin_unlock_irqrestore(&tree->lock, flags);
773 free_extent_state(prealloc);
780 spin_unlock_irqrestore(&tree->lock, flags);
781 if (mask & __GFP_WAIT)
785 EXPORT_SYMBOL(set_extent_bit);
787 /* wrappers around set/clear extent bit */
788 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
791 return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL,
794 EXPORT_SYMBOL(set_extent_dirty);
796 int set_extent_ordered(struct extent_io_tree *tree, u64 start, u64 end,
799 return set_extent_bit(tree, start, end, EXTENT_ORDERED, 0, NULL, mask);
801 EXPORT_SYMBOL(set_extent_ordered);
803 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
804 int bits, gfp_t mask)
806 return set_extent_bit(tree, start, end, bits, 0, NULL,
809 EXPORT_SYMBOL(set_extent_bits);
811 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
812 int bits, gfp_t mask)
814 return clear_extent_bit(tree, start, end, bits, 0, 0, mask);
816 EXPORT_SYMBOL(clear_extent_bits);
818 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
821 return set_extent_bit(tree, start, end,
822 EXTENT_DELALLOC | EXTENT_DIRTY,
825 EXPORT_SYMBOL(set_extent_delalloc);
827 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
830 return clear_extent_bit(tree, start, end,
831 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, mask);
833 EXPORT_SYMBOL(clear_extent_dirty);
835 int clear_extent_ordered(struct extent_io_tree *tree, u64 start, u64 end,
838 return clear_extent_bit(tree, start, end, EXTENT_ORDERED, 1, 0, mask);
840 EXPORT_SYMBOL(clear_extent_ordered);
842 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
845 return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL,
848 EXPORT_SYMBOL(set_extent_new);
850 int clear_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
853 return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0, mask);
855 EXPORT_SYMBOL(clear_extent_new);
857 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
860 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, NULL,
863 EXPORT_SYMBOL(set_extent_uptodate);
865 int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
868 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, mask);
870 EXPORT_SYMBOL(clear_extent_uptodate);
872 int set_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end,
875 return set_extent_bit(tree, start, end, EXTENT_WRITEBACK,
878 EXPORT_SYMBOL(set_extent_writeback);
880 int clear_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end,
883 return clear_extent_bit(tree, start, end, EXTENT_WRITEBACK, 1, 0, mask);
885 EXPORT_SYMBOL(clear_extent_writeback);
887 int wait_on_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end)
889 return wait_extent_bit(tree, start, end, EXTENT_WRITEBACK);
891 EXPORT_SYMBOL(wait_on_extent_writeback);
893 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
898 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1,
899 &failed_start, mask);
900 if (err == -EEXIST && (mask & __GFP_WAIT)) {
901 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
902 start = failed_start;
906 WARN_ON(start > end);
910 EXPORT_SYMBOL(lock_extent);
912 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end,
915 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, mask);
917 EXPORT_SYMBOL(unlock_extent);
920 * helper function to set pages and extents in the tree dirty
922 int set_range_dirty(struct extent_io_tree *tree, u64 start, u64 end)
924 unsigned long index = start >> PAGE_CACHE_SHIFT;
925 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
928 while (index <= end_index) {
929 page = find_get_page(tree->mapping, index);
931 __set_page_dirty_nobuffers(page);
932 page_cache_release(page);
935 set_extent_dirty(tree, start, end, GFP_NOFS);
938 EXPORT_SYMBOL(set_range_dirty);
941 * helper function to set both pages and extents in the tree writeback
943 int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
945 unsigned long index = start >> PAGE_CACHE_SHIFT;
946 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
949 while (index <= end_index) {
950 page = find_get_page(tree->mapping, index);
952 set_page_writeback(page);
953 page_cache_release(page);
956 set_extent_writeback(tree, start, end, GFP_NOFS);
959 EXPORT_SYMBOL(set_range_writeback);
961 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
962 u64 *start_ret, u64 *end_ret, int bits)
964 struct rb_node *node;
965 struct extent_state *state;
968 spin_lock_irq(&tree->lock);
970 * this search will find all the extents that end after
973 node = tree_search(tree, start);
979 state = rb_entry(node, struct extent_state, rb_node);
980 if (state->end >= start && (state->state & bits)) {
981 *start_ret = state->start;
982 *end_ret = state->end;
986 node = rb_next(node);
991 spin_unlock_irq(&tree->lock);
994 EXPORT_SYMBOL(find_first_extent_bit);
996 struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree,
999 struct rb_node *node;
1000 struct extent_state *state;
1003 * this search will find all the extents that end after
1006 node = tree_search(tree, start);
1012 state = rb_entry(node, struct extent_state, rb_node);
1013 if (state->end >= start && (state->state & bits)) {
1016 node = rb_next(node);
1023 EXPORT_SYMBOL(find_first_extent_bit_state);
1025 u64 find_lock_delalloc_range(struct extent_io_tree *tree,
1026 u64 *start, u64 *end, u64 max_bytes)
1028 struct rb_node *node;
1029 struct extent_state *state;
1030 u64 cur_start = *start;
1032 u64 total_bytes = 0;
1034 spin_lock_irq(&tree->lock);
1036 * this search will find all the extents that end after
1040 node = tree_search(tree, cur_start);
1048 state = rb_entry(node, struct extent_state, rb_node);
1049 if (found && state->start != cur_start) {
1052 if (!(state->state & EXTENT_DELALLOC)) {
1058 struct extent_state *prev_state;
1059 struct rb_node *prev_node = node;
1061 prev_node = rb_prev(prev_node);
1064 prev_state = rb_entry(prev_node,
1065 struct extent_state,
1067 if (!(prev_state->state & EXTENT_DELALLOC))
1073 if (state->state & EXTENT_LOCKED) {
1075 atomic_inc(&state->refs);
1076 prepare_to_wait(&state->wq, &wait,
1077 TASK_UNINTERRUPTIBLE);
1078 spin_unlock_irq(&tree->lock);
1080 spin_lock_irq(&tree->lock);
1081 finish_wait(&state->wq, &wait);
1082 free_extent_state(state);
1085 set_state_cb(tree, state, EXTENT_LOCKED);
1086 state->state |= EXTENT_LOCKED;
1088 *start = state->start;
1091 cur_start = state->end + 1;
1092 node = rb_next(node);
1095 total_bytes += state->end - state->start + 1;
1096 if (total_bytes >= max_bytes)
1100 spin_unlock_irq(&tree->lock);
1104 u64 count_range_bits(struct extent_io_tree *tree,
1105 u64 *start, u64 search_end, u64 max_bytes,
1108 struct rb_node *node;
1109 struct extent_state *state;
1110 u64 cur_start = *start;
1111 u64 total_bytes = 0;
1114 if (search_end <= cur_start) {
1115 printk("search_end %Lu start %Lu\n", search_end, cur_start);
1120 spin_lock_irq(&tree->lock);
1121 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1122 total_bytes = tree->dirty_bytes;
1126 * this search will find all the extents that end after
1129 node = tree_search(tree, cur_start);
1135 state = rb_entry(node, struct extent_state, rb_node);
1136 if (state->start > search_end)
1138 if (state->end >= cur_start && (state->state & bits)) {
1139 total_bytes += min(search_end, state->end) + 1 -
1140 max(cur_start, state->start);
1141 if (total_bytes >= max_bytes)
1144 *start = state->start;
1148 node = rb_next(node);
1153 spin_unlock_irq(&tree->lock);
1157 * helper function to lock both pages and extents in the tree.
1158 * pages must be locked first.
1160 int lock_range(struct extent_io_tree *tree, u64 start, u64 end)
1162 unsigned long index = start >> PAGE_CACHE_SHIFT;
1163 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1167 while (index <= end_index) {
1168 page = grab_cache_page(tree->mapping, index);
1174 err = PTR_ERR(page);
1179 lock_extent(tree, start, end, GFP_NOFS);
1184 * we failed above in getting the page at 'index', so we undo here
1185 * up to but not including the page at 'index'
1188 index = start >> PAGE_CACHE_SHIFT;
1189 while (index < end_index) {
1190 page = find_get_page(tree->mapping, index);
1192 page_cache_release(page);
1197 EXPORT_SYMBOL(lock_range);
1200 * helper function to unlock both pages and extents in the tree.
1202 int unlock_range(struct extent_io_tree *tree, u64 start, u64 end)
1204 unsigned long index = start >> PAGE_CACHE_SHIFT;
1205 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1208 while (index <= end_index) {
1209 page = find_get_page(tree->mapping, index);
1211 page_cache_release(page);
1214 unlock_extent(tree, start, end, GFP_NOFS);
1217 EXPORT_SYMBOL(unlock_range);
1219 int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1221 struct rb_node *node;
1222 struct extent_state *state;
1225 spin_lock_irq(&tree->lock);
1227 * this search will find all the extents that end after
1230 node = tree_search(tree, start);
1235 state = rb_entry(node, struct extent_state, rb_node);
1236 if (state->start != start) {
1240 state->private = private;
1242 spin_unlock_irq(&tree->lock);
1246 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1248 struct rb_node *node;
1249 struct extent_state *state;
1252 spin_lock_irq(&tree->lock);
1254 * this search will find all the extents that end after
1257 node = tree_search(tree, start);
1262 state = rb_entry(node, struct extent_state, rb_node);
1263 if (state->start != start) {
1267 *private = state->private;
1269 spin_unlock_irq(&tree->lock);
1274 * searches a range in the state tree for a given mask.
1275 * If 'filled' == 1, this returns 1 only if every extent in the tree
1276 * has the bits set. Otherwise, 1 is returned if any bit in the
1277 * range is found set.
1279 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1280 int bits, int filled)
1282 struct extent_state *state = NULL;
1283 struct rb_node *node;
1285 unsigned long flags;
1287 spin_lock_irqsave(&tree->lock, flags);
1288 node = tree_search(tree, start);
1289 while (node && start <= end) {
1290 state = rb_entry(node, struct extent_state, rb_node);
1292 if (filled && state->start > start) {
1297 if (state->start > end)
1300 if (state->state & bits) {
1304 } else if (filled) {
1308 start = state->end + 1;
1311 node = rb_next(node);
1318 spin_unlock_irqrestore(&tree->lock, flags);
1321 EXPORT_SYMBOL(test_range_bit);
1324 * helper function to set a given page up to date if all the
1325 * extents in the tree for that page are up to date
1327 static int check_page_uptodate(struct extent_io_tree *tree,
1330 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1331 u64 end = start + PAGE_CACHE_SIZE - 1;
1332 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1))
1333 SetPageUptodate(page);
1338 * helper function to unlock a page if all the extents in the tree
1339 * for that page are unlocked
1341 static int check_page_locked(struct extent_io_tree *tree,
1344 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1345 u64 end = start + PAGE_CACHE_SIZE - 1;
1346 if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0))
1352 * helper function to end page writeback if all the extents
1353 * in the tree for that page are done with writeback
1355 static int check_page_writeback(struct extent_io_tree *tree,
1358 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1359 u64 end = start + PAGE_CACHE_SIZE - 1;
1360 if (!test_range_bit(tree, start, end, EXTENT_WRITEBACK, 0))
1361 end_page_writeback(page);
1365 /* lots and lots of room for performance fixes in the end_bio funcs */
1368 * after a writepage IO is done, we need to:
1369 * clear the uptodate bits on error
1370 * clear the writeback bits in the extent tree for this IO
1371 * end_page_writeback if the page has no more pending IO
1373 * Scheduling is not allowed, so the extent state tree is expected
1374 * to have one and only one object corresponding to this IO.
1376 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1377 static void end_bio_extent_writepage(struct bio *bio, int err)
1379 static int end_bio_extent_writepage(struct bio *bio,
1380 unsigned int bytes_done, int err)
1383 int uptodate = err == 0;
1384 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1385 struct extent_state *state = bio->bi_private;
1386 struct extent_io_tree *tree = state->tree;
1387 struct rb_node *node;
1393 unsigned long flags;
1395 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1400 struct page *page = bvec->bv_page;
1401 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1403 end = start + bvec->bv_len - 1;
1405 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1410 if (--bvec >= bio->bi_io_vec)
1411 prefetchw(&bvec->bv_page->flags);
1412 if (tree->ops && tree->ops->writepage_end_io_hook) {
1413 ret = tree->ops->writepage_end_io_hook(page, start,
1414 end, state, uptodate);
1419 if (!uptodate && tree->ops &&
1420 tree->ops->writepage_io_failed_hook) {
1421 ret = tree->ops->writepage_io_failed_hook(bio, page,
1425 uptodate = (err == 0);
1431 clear_extent_uptodate(tree, start, end, GFP_ATOMIC);
1432 ClearPageUptodate(page);
1437 * bios can get merged in funny ways, and so we need to
1438 * be careful with the state variable. We know the
1439 * state won't be merged with others because it has
1440 * WRITEBACK set, but we can't be sure each biovec is
1441 * sequential in the file. So, if our cached state
1442 * doesn't match the expected end, search the tree
1443 * for the correct one.
1446 spin_lock_irqsave(&tree->lock, flags);
1447 if (!state || state->end != end) {
1449 node = __etree_search(tree, start, NULL, NULL);
1451 state = rb_entry(node, struct extent_state,
1453 if (state->end != end ||
1454 !(state->state & EXTENT_WRITEBACK))
1458 spin_unlock_irqrestore(&tree->lock, flags);
1459 clear_extent_writeback(tree, start,
1466 struct extent_state *clear = state;
1468 node = rb_prev(&state->rb_node);
1470 state = rb_entry(node,
1471 struct extent_state,
1477 clear_state_bit(tree, clear, EXTENT_WRITEBACK,
1488 /* before releasing the lock, make sure the next state
1489 * variable has the expected bits set and corresponds
1490 * to the correct offsets in the file
1492 if (state && (state->end + 1 != start ||
1493 !(state->state & EXTENT_WRITEBACK))) {
1496 spin_unlock_irqrestore(&tree->lock, flags);
1500 end_page_writeback(page);
1502 check_page_writeback(tree, page);
1503 } while (bvec >= bio->bi_io_vec);
1505 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1511 * after a readpage IO is done, we need to:
1512 * clear the uptodate bits on error
1513 * set the uptodate bits if things worked
1514 * set the page up to date if all extents in the tree are uptodate
1515 * clear the lock bit in the extent tree
1516 * unlock the page if there are no other extents locked for it
1518 * Scheduling is not allowed, so the extent state tree is expected
1519 * to have one and only one object corresponding to this IO.
1521 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1522 static void end_bio_extent_readpage(struct bio *bio, int err)
1524 static int end_bio_extent_readpage(struct bio *bio,
1525 unsigned int bytes_done, int err)
1528 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1529 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1530 struct extent_state *state = bio->bi_private;
1531 struct extent_io_tree *tree = state->tree;
1532 struct rb_node *node;
1536 unsigned long flags;
1540 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1546 struct page *page = bvec->bv_page;
1547 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1549 end = start + bvec->bv_len - 1;
1551 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1556 if (--bvec >= bio->bi_io_vec)
1557 prefetchw(&bvec->bv_page->flags);
1559 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
1560 ret = tree->ops->readpage_end_io_hook(page, start, end,
1565 if (!uptodate && tree->ops &&
1566 tree->ops->readpage_io_failed_hook) {
1567 ret = tree->ops->readpage_io_failed_hook(bio, page,
1572 test_bit(BIO_UPTODATE, &bio->bi_flags);
1577 spin_lock_irqsave(&tree->lock, flags);
1578 if (!state || state->end != end) {
1580 node = __etree_search(tree, start, NULL, NULL);
1582 state = rb_entry(node, struct extent_state,
1584 if (state->end != end ||
1585 !(state->state & EXTENT_LOCKED))
1589 spin_unlock_irqrestore(&tree->lock, flags);
1591 set_extent_uptodate(tree, start, end,
1593 unlock_extent(tree, start, end, GFP_ATOMIC);
1600 struct extent_state *clear = state;
1602 node = rb_prev(&state->rb_node);
1604 state = rb_entry(node,
1605 struct extent_state,
1611 set_state_cb(tree, clear, EXTENT_UPTODATE);
1612 clear->state |= EXTENT_UPTODATE;
1614 clear_state_bit(tree, clear, EXTENT_LOCKED,
1625 /* before releasing the lock, make sure the next state
1626 * variable has the expected bits set and corresponds
1627 * to the correct offsets in the file
1629 if (state && (state->end + 1 != start ||
1630 !(state->state & EXTENT_LOCKED))) {
1633 spin_unlock_irqrestore(&tree->lock, flags);
1637 SetPageUptodate(page);
1639 ClearPageUptodate(page);
1645 check_page_uptodate(tree, page);
1647 ClearPageUptodate(page);
1650 check_page_locked(tree, page);
1652 } while (bvec >= bio->bi_io_vec);
1655 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1661 * IO done from prepare_write is pretty simple, we just unlock
1662 * the structs in the extent tree when done, and set the uptodate bits
1665 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1666 static void end_bio_extent_preparewrite(struct bio *bio, int err)
1668 static int end_bio_extent_preparewrite(struct bio *bio,
1669 unsigned int bytes_done, int err)
1672 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1673 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1674 struct extent_state *state = bio->bi_private;
1675 struct extent_io_tree *tree = state->tree;
1679 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1685 struct page *page = bvec->bv_page;
1686 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1688 end = start + bvec->bv_len - 1;
1690 if (--bvec >= bio->bi_io_vec)
1691 prefetchw(&bvec->bv_page->flags);
1694 set_extent_uptodate(tree, start, end, GFP_ATOMIC);
1696 ClearPageUptodate(page);
1700 unlock_extent(tree, start, end, GFP_ATOMIC);
1702 } while (bvec >= bio->bi_io_vec);
1705 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1711 extent_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
1716 bio = bio_alloc(gfp_flags, nr_vecs);
1718 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
1719 while (!bio && (nr_vecs /= 2))
1720 bio = bio_alloc(gfp_flags, nr_vecs);
1725 bio->bi_bdev = bdev;
1726 bio->bi_sector = first_sector;
1731 static int submit_one_bio(int rw, struct bio *bio, int mirror_num)
1734 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1735 struct page *page = bvec->bv_page;
1736 struct extent_io_tree *tree = bio->bi_private;
1737 struct rb_node *node;
1738 struct extent_state *state;
1742 start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
1743 end = start + bvec->bv_len - 1;
1745 spin_lock_irq(&tree->lock);
1746 node = __etree_search(tree, start, NULL, NULL);
1748 state = rb_entry(node, struct extent_state, rb_node);
1749 while(state->end < end) {
1750 node = rb_next(node);
1751 state = rb_entry(node, struct extent_state, rb_node);
1753 BUG_ON(state->end != end);
1754 spin_unlock_irq(&tree->lock);
1756 bio->bi_private = state;
1760 if (tree->ops && tree->ops->submit_bio_hook)
1761 tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
1764 submit_bio(rw, bio);
1765 if (bio_flagged(bio, BIO_EOPNOTSUPP))
1771 static int submit_extent_page(int rw, struct extent_io_tree *tree,
1772 struct page *page, sector_t sector,
1773 size_t size, unsigned long offset,
1774 struct block_device *bdev,
1775 struct bio **bio_ret,
1776 unsigned long max_pages,
1777 bio_end_io_t end_io_func,
1784 if (bio_ret && *bio_ret) {
1786 if (bio->bi_sector + (bio->bi_size >> 9) != sector ||
1787 (tree->ops && tree->ops->merge_bio_hook &&
1788 tree->ops->merge_bio_hook(page, offset, size, bio)) ||
1789 bio_add_page(bio, page, size, offset) < size) {
1790 ret = submit_one_bio(rw, bio, mirror_num);
1796 nr = bio_get_nr_vecs(bdev);
1797 bio = extent_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
1799 printk("failed to allocate bio nr %d\n", nr);
1803 bio_add_page(bio, page, size, offset);
1804 bio->bi_end_io = end_io_func;
1805 bio->bi_private = tree;
1810 ret = submit_one_bio(rw, bio, mirror_num);
1816 void set_page_extent_mapped(struct page *page)
1818 if (!PagePrivate(page)) {
1819 SetPagePrivate(page);
1820 WARN_ON(!page->mapping->a_ops->invalidatepage);
1821 set_page_private(page, EXTENT_PAGE_PRIVATE);
1822 page_cache_get(page);
1826 void set_page_extent_head(struct page *page, unsigned long len)
1828 set_page_private(page, EXTENT_PAGE_PRIVATE_FIRST_PAGE | len << 2);
1832 * basic readpage implementation. Locked extent state structs are inserted
1833 * into the tree that are removed when the IO is done (by the end_io
1836 static int __extent_read_full_page(struct extent_io_tree *tree,
1838 get_extent_t *get_extent,
1839 struct bio **bio, int mirror_num)
1841 struct inode *inode = page->mapping->host;
1842 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1843 u64 page_end = start + PAGE_CACHE_SIZE - 1;
1847 u64 last_byte = i_size_read(inode);
1851 struct extent_map *em;
1852 struct block_device *bdev;
1855 size_t page_offset = 0;
1857 size_t blocksize = inode->i_sb->s_blocksize;
1859 set_page_extent_mapped(page);
1862 lock_extent(tree, start, end, GFP_NOFS);
1864 while (cur <= end) {
1865 if (cur >= last_byte) {
1867 iosize = PAGE_CACHE_SIZE - page_offset;
1868 userpage = kmap_atomic(page, KM_USER0);
1869 memset(userpage + page_offset, 0, iosize);
1870 flush_dcache_page(page);
1871 kunmap_atomic(userpage, KM_USER0);
1872 set_extent_uptodate(tree, cur, cur + iosize - 1,
1874 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1877 em = get_extent(inode, page, page_offset, cur,
1879 if (IS_ERR(em) || !em) {
1881 unlock_extent(tree, cur, end, GFP_NOFS);
1884 extent_offset = cur - em->start;
1885 if (extent_map_end(em) <= cur) {
1886 printk("bad mapping em [%Lu %Lu] cur %Lu\n", em->start, extent_map_end(em), cur);
1888 BUG_ON(extent_map_end(em) <= cur);
1890 printk("2bad mapping end %Lu cur %Lu\n", end, cur);
1894 iosize = min(extent_map_end(em) - cur, end - cur + 1);
1895 cur_end = min(extent_map_end(em) - 1, end);
1896 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
1897 sector = (em->block_start + extent_offset) >> 9;
1899 block_start = em->block_start;
1900 free_extent_map(em);
1903 /* we've found a hole, just zero and go on */
1904 if (block_start == EXTENT_MAP_HOLE) {
1906 userpage = kmap_atomic(page, KM_USER0);
1907 memset(userpage + page_offset, 0, iosize);
1908 flush_dcache_page(page);
1909 kunmap_atomic(userpage, KM_USER0);
1911 set_extent_uptodate(tree, cur, cur + iosize - 1,
1913 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1915 page_offset += iosize;
1918 /* the get_extent function already copied into the page */
1919 if (test_range_bit(tree, cur, cur_end, EXTENT_UPTODATE, 1)) {
1920 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1922 page_offset += iosize;
1925 /* we have an inline extent but it didn't get marked up
1926 * to date. Error out
1928 if (block_start == EXTENT_MAP_INLINE) {
1930 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1932 page_offset += iosize;
1937 if (tree->ops && tree->ops->readpage_io_hook) {
1938 ret = tree->ops->readpage_io_hook(page, cur,
1942 unsigned long nr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
1944 ret = submit_extent_page(READ, tree, page,
1945 sector, iosize, page_offset,
1947 end_bio_extent_readpage, mirror_num);
1952 page_offset += iosize;
1956 if (!PageError(page))
1957 SetPageUptodate(page);
1963 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
1964 get_extent_t *get_extent)
1966 struct bio *bio = NULL;
1969 ret = __extent_read_full_page(tree, page, get_extent, &bio, 0);
1971 submit_one_bio(READ, bio, 0);
1974 EXPORT_SYMBOL(extent_read_full_page);
1977 * the writepage semantics are similar to regular writepage. extent
1978 * records are inserted to lock ranges in the tree, and as dirty areas
1979 * are found, they are marked writeback. Then the lock bits are removed
1980 * and the end_io handler clears the writeback ranges
1982 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
1985 struct inode *inode = page->mapping->host;
1986 struct extent_page_data *epd = data;
1987 struct extent_io_tree *tree = epd->tree;
1988 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1990 u64 page_end = start + PAGE_CACHE_SIZE - 1;
1994 u64 last_byte = i_size_read(inode);
1999 struct extent_map *em;
2000 struct block_device *bdev;
2003 size_t pg_offset = 0;
2005 loff_t i_size = i_size_read(inode);
2006 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
2010 WARN_ON(!PageLocked(page));
2011 pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
2012 if (page->index > end_index ||
2013 (page->index == end_index && !pg_offset)) {
2014 page->mapping->a_ops->invalidatepage(page, 0);
2019 if (page->index == end_index) {
2022 userpage = kmap_atomic(page, KM_USER0);
2023 memset(userpage + pg_offset, 0,
2024 PAGE_CACHE_SIZE - pg_offset);
2025 kunmap_atomic(userpage, KM_USER0);
2026 flush_dcache_page(page);
2030 set_page_extent_mapped(page);
2032 delalloc_start = start;
2034 while(delalloc_end < page_end) {
2035 nr_delalloc = find_lock_delalloc_range(tree, &delalloc_start,
2038 if (nr_delalloc == 0) {
2039 delalloc_start = delalloc_end + 1;
2042 tree->ops->fill_delalloc(inode, delalloc_start,
2044 clear_extent_bit(tree, delalloc_start,
2046 EXTENT_LOCKED | EXTENT_DELALLOC,
2048 delalloc_start = delalloc_end + 1;
2050 lock_extent(tree, start, page_end, GFP_NOFS);
2051 unlock_start = start;
2053 if (tree->ops && tree->ops->writepage_start_hook) {
2054 ret = tree->ops->writepage_start_hook(page, start, page_end);
2055 if (ret == -EAGAIN) {
2056 unlock_extent(tree, start, page_end, GFP_NOFS);
2057 redirty_page_for_writepage(wbc, page);
2064 if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0)) {
2065 printk("found delalloc bits after lock_extent\n");
2068 if (last_byte <= start) {
2069 clear_extent_dirty(tree, start, page_end, GFP_NOFS);
2070 unlock_extent(tree, start, page_end, GFP_NOFS);
2071 if (tree->ops && tree->ops->writepage_end_io_hook)
2072 tree->ops->writepage_end_io_hook(page, start,
2074 unlock_start = page_end + 1;
2078 set_extent_uptodate(tree, start, page_end, GFP_NOFS);
2079 blocksize = inode->i_sb->s_blocksize;
2081 while (cur <= end) {
2082 if (cur >= last_byte) {
2083 clear_extent_dirty(tree, cur, page_end, GFP_NOFS);
2084 unlock_extent(tree, unlock_start, page_end, GFP_NOFS);
2085 if (tree->ops && tree->ops->writepage_end_io_hook)
2086 tree->ops->writepage_end_io_hook(page, cur,
2088 unlock_start = page_end + 1;
2091 em = epd->get_extent(inode, page, pg_offset, cur,
2093 if (IS_ERR(em) || !em) {
2098 extent_offset = cur - em->start;
2099 BUG_ON(extent_map_end(em) <= cur);
2101 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2102 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2103 sector = (em->block_start + extent_offset) >> 9;
2105 block_start = em->block_start;
2106 free_extent_map(em);
2109 if (block_start == EXTENT_MAP_HOLE ||
2110 block_start == EXTENT_MAP_INLINE) {
2111 clear_extent_dirty(tree, cur,
2112 cur + iosize - 1, GFP_NOFS);
2114 unlock_extent(tree, unlock_start, cur + iosize -1,
2117 if (tree->ops && tree->ops->writepage_end_io_hook)
2118 tree->ops->writepage_end_io_hook(page, cur,
2122 pg_offset += iosize;
2127 /* leave this out until we have a page_mkwrite call */
2128 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
2131 pg_offset += iosize;
2134 clear_extent_dirty(tree, cur, cur + iosize - 1, GFP_NOFS);
2135 if (tree->ops && tree->ops->writepage_io_hook) {
2136 ret = tree->ops->writepage_io_hook(page, cur,
2144 unsigned long max_nr = end_index + 1;
2146 set_range_writeback(tree, cur, cur + iosize - 1);
2147 if (!PageWriteback(page)) {
2148 printk("warning page %lu not writeback, "
2149 "cur %llu end %llu\n", page->index,
2150 (unsigned long long)cur,
2151 (unsigned long long)end);
2154 ret = submit_extent_page(WRITE, tree, page, sector,
2155 iosize, pg_offset, bdev,
2157 end_bio_extent_writepage, 0);
2162 pg_offset += iosize;
2167 /* make sure the mapping tag for page dirty gets cleared */
2168 set_page_writeback(page);
2169 end_page_writeback(page);
2171 if (unlock_start <= page_end)
2172 unlock_extent(tree, unlock_start, page_end, GFP_NOFS);
2177 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,22)
2178 /* Taken directly from 2.6.23 for 2.6.18 back port */
2179 typedef int (*writepage_t)(struct page *page, struct writeback_control *wbc,
2183 * write_cache_pages - walk the list of dirty pages of the given address space
2184 * and write all of them.
2185 * @mapping: address space structure to write
2186 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2187 * @writepage: function called for each page
2188 * @data: data passed to writepage function
2190 * If a page is already under I/O, write_cache_pages() skips it, even
2191 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
2192 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
2193 * and msync() need to guarantee that all the data which was dirty at the time
2194 * the call was made get new I/O started against them. If wbc->sync_mode is
2195 * WB_SYNC_ALL then we were called for data integrity and we must wait for
2196 * existing IO to complete.
2198 static int write_cache_pages(struct address_space *mapping,
2199 struct writeback_control *wbc, writepage_t writepage,
2202 struct backing_dev_info *bdi = mapping->backing_dev_info;
2205 struct pagevec pvec;
2208 pgoff_t end; /* Inclusive */
2210 int range_whole = 0;
2212 if (wbc->nonblocking && bdi_write_congested(bdi)) {
2213 wbc->encountered_congestion = 1;
2217 pagevec_init(&pvec, 0);
2218 if (wbc->range_cyclic) {
2219 index = mapping->writeback_index; /* Start from prev offset */
2222 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2223 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2224 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2229 while (!done && (index <= end) &&
2230 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
2231 PAGECACHE_TAG_DIRTY,
2232 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
2236 for (i = 0; i < nr_pages; i++) {
2237 struct page *page = pvec.pages[i];
2240 * At this point we hold neither mapping->tree_lock nor
2241 * lock on the page itself: the page may be truncated or
2242 * invalidated (changing page->mapping to NULL), or even
2243 * swizzled back from swapper_space to tmpfs file
2248 if (unlikely(page->mapping != mapping)) {
2253 if (!wbc->range_cyclic && page->index > end) {
2259 if (wbc->sync_mode != WB_SYNC_NONE)
2260 wait_on_page_writeback(page);
2262 if (PageWriteback(page) ||
2263 !clear_page_dirty_for_io(page)) {
2268 ret = (*writepage)(page, wbc, data);
2270 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
2274 if (ret || (--(wbc->nr_to_write) <= 0))
2276 if (wbc->nonblocking && bdi_write_congested(bdi)) {
2277 wbc->encountered_congestion = 1;
2281 pagevec_release(&pvec);
2284 if (!scanned && !done) {
2286 * We hit the last page and there is more work to be done: wrap
2287 * back to the start of the file
2293 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2294 mapping->writeback_index = index;
2299 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
2300 get_extent_t *get_extent,
2301 struct writeback_control *wbc)
2304 struct address_space *mapping = page->mapping;
2305 struct extent_page_data epd = {
2308 .get_extent = get_extent,
2310 struct writeback_control wbc_writepages = {
2312 .sync_mode = WB_SYNC_NONE,
2313 .older_than_this = NULL,
2315 .range_start = page_offset(page) + PAGE_CACHE_SIZE,
2316 .range_end = (loff_t)-1,
2320 ret = __extent_writepage(page, wbc, &epd);
2322 write_cache_pages(mapping, &wbc_writepages, __extent_writepage, &epd);
2324 submit_one_bio(WRITE, epd.bio, 0);
2328 EXPORT_SYMBOL(extent_write_full_page);
2331 int extent_writepages(struct extent_io_tree *tree,
2332 struct address_space *mapping,
2333 get_extent_t *get_extent,
2334 struct writeback_control *wbc)
2337 struct extent_page_data epd = {
2340 .get_extent = get_extent,
2343 ret = write_cache_pages(mapping, wbc, __extent_writepage, &epd);
2345 submit_one_bio(WRITE, epd.bio, 0);
2349 EXPORT_SYMBOL(extent_writepages);
2351 int extent_readpages(struct extent_io_tree *tree,
2352 struct address_space *mapping,
2353 struct list_head *pages, unsigned nr_pages,
2354 get_extent_t get_extent)
2356 struct bio *bio = NULL;
2358 struct pagevec pvec;
2360 pagevec_init(&pvec, 0);
2361 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
2362 struct page *page = list_entry(pages->prev, struct page, lru);
2364 prefetchw(&page->flags);
2365 list_del(&page->lru);
2367 * what we want to do here is call add_to_page_cache_lru,
2368 * but that isn't exported, so we reproduce it here
2370 if (!add_to_page_cache(page, mapping,
2371 page->index, GFP_KERNEL)) {
2373 /* open coding of lru_cache_add, also not exported */
2374 page_cache_get(page);
2375 if (!pagevec_add(&pvec, page))
2376 __pagevec_lru_add(&pvec);
2377 __extent_read_full_page(tree, page, get_extent,
2380 page_cache_release(page);
2382 if (pagevec_count(&pvec))
2383 __pagevec_lru_add(&pvec);
2384 BUG_ON(!list_empty(pages));
2386 submit_one_bio(READ, bio, 0);
2389 EXPORT_SYMBOL(extent_readpages);
2392 * basic invalidatepage code, this waits on any locked or writeback
2393 * ranges corresponding to the page, and then deletes any extent state
2394 * records from the tree
2396 int extent_invalidatepage(struct extent_io_tree *tree,
2397 struct page *page, unsigned long offset)
2399 u64 start = ((u64)page->index << PAGE_CACHE_SHIFT);
2400 u64 end = start + PAGE_CACHE_SIZE - 1;
2401 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
2403 start += (offset + blocksize -1) & ~(blocksize - 1);
2407 lock_extent(tree, start, end, GFP_NOFS);
2408 wait_on_extent_writeback(tree, start, end);
2409 clear_extent_bit(tree, start, end,
2410 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC,
2414 EXPORT_SYMBOL(extent_invalidatepage);
2417 * simple commit_write call, set_range_dirty is used to mark both
2418 * the pages and the extent records as dirty
2420 int extent_commit_write(struct extent_io_tree *tree,
2421 struct inode *inode, struct page *page,
2422 unsigned from, unsigned to)
2424 loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
2426 set_page_extent_mapped(page);
2427 set_page_dirty(page);
2429 if (pos > inode->i_size) {
2430 i_size_write(inode, pos);
2431 mark_inode_dirty(inode);
2435 EXPORT_SYMBOL(extent_commit_write);
2437 int extent_prepare_write(struct extent_io_tree *tree,
2438 struct inode *inode, struct page *page,
2439 unsigned from, unsigned to, get_extent_t *get_extent)
2441 u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2442 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
2444 u64 orig_block_start;
2447 struct extent_map *em;
2448 unsigned blocksize = 1 << inode->i_blkbits;
2449 size_t page_offset = 0;
2450 size_t block_off_start;
2451 size_t block_off_end;
2457 set_page_extent_mapped(page);
2459 block_start = (page_start + from) & ~((u64)blocksize - 1);
2460 block_end = (page_start + to - 1) | (blocksize - 1);
2461 orig_block_start = block_start;
2463 lock_extent(tree, page_start, page_end, GFP_NOFS);
2464 while(block_start <= block_end) {
2465 em = get_extent(inode, page, page_offset, block_start,
2466 block_end - block_start + 1, 1);
2467 if (IS_ERR(em) || !em) {
2470 cur_end = min(block_end, extent_map_end(em) - 1);
2471 block_off_start = block_start & (PAGE_CACHE_SIZE - 1);
2472 block_off_end = block_off_start + blocksize;
2473 isnew = clear_extent_new(tree, block_start, cur_end, GFP_NOFS);
2475 if (!PageUptodate(page) && isnew &&
2476 (block_off_end > to || block_off_start < from)) {
2479 kaddr = kmap_atomic(page, KM_USER0);
2480 if (block_off_end > to)
2481 memset(kaddr + to, 0, block_off_end - to);
2482 if (block_off_start < from)
2483 memset(kaddr + block_off_start, 0,
2484 from - block_off_start);
2485 flush_dcache_page(page);
2486 kunmap_atomic(kaddr, KM_USER0);
2488 if ((em->block_start != EXTENT_MAP_HOLE &&
2489 em->block_start != EXTENT_MAP_INLINE) &&
2490 !isnew && !PageUptodate(page) &&
2491 (block_off_end > to || block_off_start < from) &&
2492 !test_range_bit(tree, block_start, cur_end,
2493 EXTENT_UPTODATE, 1)) {
2495 u64 extent_offset = block_start - em->start;
2497 sector = (em->block_start + extent_offset) >> 9;
2498 iosize = (cur_end - block_start + blocksize) &
2499 ~((u64)blocksize - 1);
2501 * we've already got the extent locked, but we
2502 * need to split the state such that our end_bio
2503 * handler can clear the lock.
2505 set_extent_bit(tree, block_start,
2506 block_start + iosize - 1,
2507 EXTENT_LOCKED, 0, NULL, GFP_NOFS);
2508 ret = submit_extent_page(READ, tree, page,
2509 sector, iosize, page_offset, em->bdev,
2511 end_bio_extent_preparewrite, 0);
2513 block_start = block_start + iosize;
2515 set_extent_uptodate(tree, block_start, cur_end,
2517 unlock_extent(tree, block_start, cur_end, GFP_NOFS);
2518 block_start = cur_end + 1;
2520 page_offset = block_start & (PAGE_CACHE_SIZE - 1);
2521 free_extent_map(em);
2524 wait_extent_bit(tree, orig_block_start,
2525 block_end, EXTENT_LOCKED);
2527 check_page_uptodate(tree, page);
2529 /* FIXME, zero out newly allocated blocks on error */
2532 EXPORT_SYMBOL(extent_prepare_write);
2535 * a helper for releasepage, this tests for areas of the page that
2536 * are locked or under IO and drops the related state bits if it is safe
2539 int try_release_extent_state(struct extent_map_tree *map,
2540 struct extent_io_tree *tree, struct page *page,
2543 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2544 u64 end = start + PAGE_CACHE_SIZE - 1;
2547 if (test_range_bit(tree, start, end,
2548 EXTENT_IOBITS | EXTENT_ORDERED, 0))
2551 if ((mask & GFP_NOFS) == GFP_NOFS)
2553 clear_extent_bit(tree, start, end, EXTENT_UPTODATE,
2558 EXPORT_SYMBOL(try_release_extent_state);
2561 * a helper for releasepage. As long as there are no locked extents
2562 * in the range corresponding to the page, both state records and extent
2563 * map records are removed
2565 int try_release_extent_mapping(struct extent_map_tree *map,
2566 struct extent_io_tree *tree, struct page *page,
2569 struct extent_map *em;
2570 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2571 u64 end = start + PAGE_CACHE_SIZE - 1;
2573 if ((mask & __GFP_WAIT) &&
2574 page->mapping->host->i_size > 16 * 1024 * 1024) {
2576 while (start <= end) {
2577 len = end - start + 1;
2578 spin_lock(&map->lock);
2579 em = lookup_extent_mapping(map, start, len);
2580 if (!em || IS_ERR(em)) {
2581 spin_unlock(&map->lock);
2584 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
2585 em->start != start) {
2586 spin_unlock(&map->lock);
2587 free_extent_map(em);
2590 if (!test_range_bit(tree, em->start,
2591 extent_map_end(em) - 1,
2592 EXTENT_LOCKED, 0)) {
2593 remove_extent_mapping(map, em);
2594 /* once for the rb tree */
2595 free_extent_map(em);
2597 start = extent_map_end(em);
2598 spin_unlock(&map->lock);
2601 free_extent_map(em);
2604 return try_release_extent_state(map, tree, page, mask);
2606 EXPORT_SYMBOL(try_release_extent_mapping);
2608 sector_t extent_bmap(struct address_space *mapping, sector_t iblock,
2609 get_extent_t *get_extent)
2611 struct inode *inode = mapping->host;
2612 u64 start = iblock << inode->i_blkbits;
2613 sector_t sector = 0;
2614 struct extent_map *em;
2616 em = get_extent(inode, NULL, 0, start, (1 << inode->i_blkbits), 0);
2617 if (!em || IS_ERR(em))
2620 if (em->block_start == EXTENT_MAP_INLINE ||
2621 em->block_start == EXTENT_MAP_HOLE)
2624 sector = (em->block_start + start - em->start) >> inode->i_blkbits;
2626 free_extent_map(em);
2630 static int add_lru(struct extent_io_tree *tree, struct extent_buffer *eb)
2632 if (list_empty(&eb->lru)) {
2633 extent_buffer_get(eb);
2634 list_add(&eb->lru, &tree->buffer_lru);
2636 if (tree->lru_size >= BUFFER_LRU_MAX) {
2637 struct extent_buffer *rm;
2638 rm = list_entry(tree->buffer_lru.prev,
2639 struct extent_buffer, lru);
2641 list_del_init(&rm->lru);
2642 free_extent_buffer(rm);
2645 list_move(&eb->lru, &tree->buffer_lru);
2648 static struct extent_buffer *find_lru(struct extent_io_tree *tree,
2649 u64 start, unsigned long len)
2651 struct list_head *lru = &tree->buffer_lru;
2652 struct list_head *cur = lru->next;
2653 struct extent_buffer *eb;
2655 if (list_empty(lru))
2659 eb = list_entry(cur, struct extent_buffer, lru);
2660 if (eb->start == start && eb->len == len) {
2661 extent_buffer_get(eb);
2665 } while (cur != lru);
2669 static inline unsigned long num_extent_pages(u64 start, u64 len)
2671 return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
2672 (start >> PAGE_CACHE_SHIFT);
2675 static inline struct page *extent_buffer_page(struct extent_buffer *eb,
2679 struct address_space *mapping;
2682 return eb->first_page;
2683 i += eb->start >> PAGE_CACHE_SHIFT;
2684 mapping = eb->first_page->mapping;
2685 read_lock_irq(&mapping->tree_lock);
2686 p = radix_tree_lookup(&mapping->page_tree, i);
2687 read_unlock_irq(&mapping->tree_lock);
2691 int release_extent_buffer_tail_pages(struct extent_buffer *eb)
2693 unsigned long num_pages = num_extent_pages(eb->start, eb->len);
2699 for (i = 1; i < num_pages; i++) {
2700 page = extent_buffer_page(eb, i);
2701 page_cache_release(page);
2707 int invalidate_extent_lru(struct extent_io_tree *tree, u64 start,
2710 struct list_head *lru = &tree->buffer_lru;
2711 struct list_head *cur = lru->next;
2712 struct extent_buffer *eb;
2715 spin_lock(&tree->lru_lock);
2716 if (list_empty(lru))
2720 eb = list_entry(cur, struct extent_buffer, lru);
2721 if (eb->start <= start && eb->start + eb->len > start) {
2722 eb->flags &= ~EXTENT_UPTODATE;
2725 } while (cur != lru);
2727 spin_unlock(&tree->lru_lock);
2731 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
2736 struct extent_buffer *eb = NULL;
2737 unsigned long flags;
2739 spin_lock(&tree->lru_lock);
2740 eb = find_lru(tree, start, len);
2741 spin_unlock(&tree->lru_lock);
2746 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
2747 INIT_LIST_HEAD(&eb->lru);
2750 spin_lock_irqsave(&leak_lock, flags);
2751 list_add(&eb->leak_list, &buffers);
2752 spin_unlock_irqrestore(&leak_lock, flags);
2753 atomic_set(&eb->refs, 1);
2758 static void __free_extent_buffer(struct extent_buffer *eb)
2760 unsigned long flags;
2761 spin_lock_irqsave(&leak_lock, flags);
2762 list_del(&eb->leak_list);
2763 spin_unlock_irqrestore(&leak_lock, flags);
2764 kmem_cache_free(extent_buffer_cache, eb);
2767 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
2768 u64 start, unsigned long len,
2772 unsigned long num_pages = num_extent_pages(start, len);
2774 unsigned long index = start >> PAGE_CACHE_SHIFT;
2775 struct extent_buffer *eb;
2777 struct address_space *mapping = tree->mapping;
2780 eb = __alloc_extent_buffer(tree, start, len, mask);
2784 if (eb->flags & EXTENT_BUFFER_FILLED)
2788 eb->first_page = page0;
2791 page_cache_get(page0);
2792 mark_page_accessed(page0);
2793 set_page_extent_mapped(page0);
2794 set_page_extent_head(page0, len);
2795 uptodate = PageUptodate(page0);
2799 for (; i < num_pages; i++, index++) {
2800 p = find_or_create_page(mapping, index, mask | __GFP_HIGHMEM);
2805 set_page_extent_mapped(p);
2806 mark_page_accessed(p);
2809 set_page_extent_head(p, len);
2811 set_page_private(p, EXTENT_PAGE_PRIVATE);
2813 if (!PageUptodate(p))
2818 eb->flags |= EXTENT_UPTODATE;
2819 eb->flags |= EXTENT_BUFFER_FILLED;
2822 spin_lock(&tree->lru_lock);
2824 spin_unlock(&tree->lru_lock);
2828 spin_lock(&tree->lru_lock);
2829 list_del_init(&eb->lru);
2830 spin_unlock(&tree->lru_lock);
2831 if (!atomic_dec_and_test(&eb->refs))
2833 for (index = 1; index < i; index++) {
2834 page_cache_release(extent_buffer_page(eb, index));
2837 page_cache_release(extent_buffer_page(eb, 0));
2838 __free_extent_buffer(eb);
2841 EXPORT_SYMBOL(alloc_extent_buffer);
2843 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
2844 u64 start, unsigned long len,
2847 unsigned long num_pages = num_extent_pages(start, len);
2849 unsigned long index = start >> PAGE_CACHE_SHIFT;
2850 struct extent_buffer *eb;
2852 struct address_space *mapping = tree->mapping;
2855 eb = __alloc_extent_buffer(tree, start, len, mask);
2859 if (eb->flags & EXTENT_BUFFER_FILLED)
2862 for (i = 0; i < num_pages; i++, index++) {
2863 p = find_get_page(mapping, index);
2867 if (TestSetPageLocked(p)) {
2868 page_cache_release(p);
2872 set_page_extent_mapped(p);
2873 mark_page_accessed(p);
2877 set_page_extent_head(p, len);
2879 set_page_private(p, EXTENT_PAGE_PRIVATE);
2882 if (!PageUptodate(p))
2887 eb->flags |= EXTENT_UPTODATE;
2888 eb->flags |= EXTENT_BUFFER_FILLED;
2891 spin_lock(&tree->lru_lock);
2893 spin_unlock(&tree->lru_lock);
2896 spin_lock(&tree->lru_lock);
2897 list_del_init(&eb->lru);
2898 spin_unlock(&tree->lru_lock);
2899 if (!atomic_dec_and_test(&eb->refs))
2901 for (index = 1; index < i; index++) {
2902 page_cache_release(extent_buffer_page(eb, index));
2905 page_cache_release(extent_buffer_page(eb, 0));
2906 __free_extent_buffer(eb);
2909 EXPORT_SYMBOL(find_extent_buffer);
2911 void free_extent_buffer(struct extent_buffer *eb)
2914 unsigned long num_pages;
2919 if (!atomic_dec_and_test(&eb->refs))
2922 WARN_ON(!list_empty(&eb->lru));
2923 num_pages = num_extent_pages(eb->start, eb->len);
2925 for (i = 1; i < num_pages; i++) {
2926 page_cache_release(extent_buffer_page(eb, i));
2928 page_cache_release(extent_buffer_page(eb, 0));
2929 __free_extent_buffer(eb);
2931 EXPORT_SYMBOL(free_extent_buffer);
2933 int clear_extent_buffer_dirty(struct extent_io_tree *tree,
2934 struct extent_buffer *eb)
2938 unsigned long num_pages;
2941 u64 start = eb->start;
2942 u64 end = start + eb->len - 1;
2944 set = clear_extent_dirty(tree, start, end, GFP_NOFS);
2945 num_pages = num_extent_pages(eb->start, eb->len);
2947 for (i = 0; i < num_pages; i++) {
2948 page = extent_buffer_page(eb, i);
2950 set_page_extent_head(page, eb->len);
2952 set_page_private(page, EXTENT_PAGE_PRIVATE);
2955 * if we're on the last page or the first page and the
2956 * block isn't aligned on a page boundary, do extra checks
2957 * to make sure we don't clean page that is partially dirty
2959 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
2960 ((i == num_pages - 1) &&
2961 ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
2962 start = (u64)page->index << PAGE_CACHE_SHIFT;
2963 end = start + PAGE_CACHE_SIZE - 1;
2964 if (test_range_bit(tree, start, end,
2969 clear_page_dirty_for_io(page);
2970 read_lock_irq(&page->mapping->tree_lock);
2971 if (!PageDirty(page)) {
2972 radix_tree_tag_clear(&page->mapping->page_tree,
2974 PAGECACHE_TAG_DIRTY);
2976 read_unlock_irq(&page->mapping->tree_lock);
2980 EXPORT_SYMBOL(clear_extent_buffer_dirty);
2982 int wait_on_extent_buffer_writeback(struct extent_io_tree *tree,
2983 struct extent_buffer *eb)
2985 return wait_on_extent_writeback(tree, eb->start,
2986 eb->start + eb->len - 1);
2988 EXPORT_SYMBOL(wait_on_extent_buffer_writeback);
2990 int set_extent_buffer_dirty(struct extent_io_tree *tree,
2991 struct extent_buffer *eb)
2994 unsigned long num_pages;
2996 num_pages = num_extent_pages(eb->start, eb->len);
2997 for (i = 0; i < num_pages; i++) {
2998 struct page *page = extent_buffer_page(eb, i);
2999 /* writepage may need to do something special for the
3000 * first page, we have to make sure page->private is
3001 * properly set. releasepage may drop page->private
3002 * on us if the page isn't already dirty.
3005 set_page_extent_head(page, eb->len);
3006 } else if (PagePrivate(page) &&
3007 page->private != EXTENT_PAGE_PRIVATE) {
3008 set_page_extent_mapped(page);
3010 __set_page_dirty_nobuffers(extent_buffer_page(eb, i));
3012 return set_extent_dirty(tree, eb->start,
3013 eb->start + eb->len - 1, GFP_NOFS);
3015 EXPORT_SYMBOL(set_extent_buffer_dirty);
3017 int clear_extent_buffer_uptodate(struct extent_io_tree *tree,
3018 struct extent_buffer *eb)
3022 unsigned long num_pages;
3024 num_pages = num_extent_pages(eb->start, eb->len);
3025 eb->flags &= ~EXTENT_UPTODATE;
3027 clear_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
3029 for (i = 0; i < num_pages; i++) {
3030 page = extent_buffer_page(eb, i);
3031 ClearPageUptodate(page);
3036 int set_extent_buffer_uptodate(struct extent_io_tree *tree,
3037 struct extent_buffer *eb)
3041 unsigned long num_pages;
3043 num_pages = num_extent_pages(eb->start, eb->len);
3045 set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
3047 for (i = 0; i < num_pages; i++) {
3048 page = extent_buffer_page(eb, i);
3049 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
3050 ((i == num_pages - 1) &&
3051 ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
3052 check_page_uptodate(tree, page);
3055 SetPageUptodate(page);
3059 EXPORT_SYMBOL(set_extent_buffer_uptodate);
3061 int extent_range_uptodate(struct extent_io_tree *tree,
3066 int pg_uptodate = 1;
3068 unsigned long index;
3070 ret = test_range_bit(tree, start, end, EXTENT_UPTODATE, 1);
3073 while(start <= end) {
3074 index = start >> PAGE_CACHE_SHIFT;
3075 page = find_get_page(tree->mapping, index);
3076 uptodate = PageUptodate(page);
3077 page_cache_release(page);
3082 start += PAGE_CACHE_SIZE;
3087 int extent_buffer_uptodate(struct extent_io_tree *tree,
3088 struct extent_buffer *eb)
3091 unsigned long num_pages;
3094 int pg_uptodate = 1;
3096 if (eb->flags & EXTENT_UPTODATE)
3099 ret = test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3100 EXTENT_UPTODATE, 1);
3104 num_pages = num_extent_pages(eb->start, eb->len);
3105 for (i = 0; i < num_pages; i++) {
3106 page = extent_buffer_page(eb, i);
3107 if (!PageUptodate(page)) {
3114 EXPORT_SYMBOL(extent_buffer_uptodate);
3116 int read_extent_buffer_pages(struct extent_io_tree *tree,
3117 struct extent_buffer *eb,
3118 u64 start, int wait,
3119 get_extent_t *get_extent, int mirror_num)
3122 unsigned long start_i;
3126 int locked_pages = 0;
3127 int all_uptodate = 1;
3128 int inc_all_pages = 0;
3129 unsigned long num_pages;
3130 struct bio *bio = NULL;
3132 if (eb->flags & EXTENT_UPTODATE)
3135 if (test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3136 EXTENT_UPTODATE, 1)) {
3141 WARN_ON(start < eb->start);
3142 start_i = (start >> PAGE_CACHE_SHIFT) -
3143 (eb->start >> PAGE_CACHE_SHIFT);
3148 num_pages = num_extent_pages(eb->start, eb->len);
3149 for (i = start_i; i < num_pages; i++) {
3150 page = extent_buffer_page(eb, i);
3152 if (TestSetPageLocked(page))
3158 if (!PageUptodate(page)) {
3164 eb->flags |= EXTENT_UPTODATE;
3168 for (i = start_i; i < num_pages; i++) {
3169 page = extent_buffer_page(eb, i);
3171 page_cache_get(page);
3172 if (!PageUptodate(page)) {
3175 ClearPageError(page);
3176 err = __extent_read_full_page(tree, page,
3188 submit_one_bio(READ, bio, mirror_num);
3193 for (i = start_i; i < num_pages; i++) {
3194 page = extent_buffer_page(eb, i);
3195 wait_on_page_locked(page);
3196 if (!PageUptodate(page)) {
3201 eb->flags |= EXTENT_UPTODATE;
3206 while(locked_pages > 0) {
3207 page = extent_buffer_page(eb, i);
3214 EXPORT_SYMBOL(read_extent_buffer_pages);
3216 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
3217 unsigned long start,
3224 char *dst = (char *)dstv;
3225 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3226 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3228 WARN_ON(start > eb->len);
3229 WARN_ON(start + len > eb->start + eb->len);
3231 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3234 page = extent_buffer_page(eb, i);
3236 cur = min(len, (PAGE_CACHE_SIZE - offset));
3237 kaddr = kmap_atomic(page, KM_USER1);
3238 memcpy(dst, kaddr + offset, cur);
3239 kunmap_atomic(kaddr, KM_USER1);
3247 EXPORT_SYMBOL(read_extent_buffer);
3249 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
3250 unsigned long min_len, char **token, char **map,
3251 unsigned long *map_start,
3252 unsigned long *map_len, int km)
3254 size_t offset = start & (PAGE_CACHE_SIZE - 1);
3257 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3258 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3259 unsigned long end_i = (start_offset + start + min_len - 1) >>
3266 offset = start_offset;
3270 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
3272 if (start + min_len > eb->len) {
3273 printk("bad mapping eb start %Lu len %lu, wanted %lu %lu\n", eb->start, eb->len, start, min_len);
3277 p = extent_buffer_page(eb, i);
3278 kaddr = kmap_atomic(p, km);
3280 *map = kaddr + offset;
3281 *map_len = PAGE_CACHE_SIZE - offset;
3284 EXPORT_SYMBOL(map_private_extent_buffer);
3286 int map_extent_buffer(struct extent_buffer *eb, unsigned long start,
3287 unsigned long min_len,
3288 char **token, char **map,
3289 unsigned long *map_start,
3290 unsigned long *map_len, int km)
3294 if (eb->map_token) {
3295 unmap_extent_buffer(eb, eb->map_token, km);
3296 eb->map_token = NULL;
3299 err = map_private_extent_buffer(eb, start, min_len, token, map,
3300 map_start, map_len, km);
3302 eb->map_token = *token;
3304 eb->map_start = *map_start;
3305 eb->map_len = *map_len;
3309 EXPORT_SYMBOL(map_extent_buffer);
3311 void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km)
3313 kunmap_atomic(token, km);
3315 EXPORT_SYMBOL(unmap_extent_buffer);
3317 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
3318 unsigned long start,
3325 char *ptr = (char *)ptrv;
3326 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3327 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3330 WARN_ON(start > eb->len);
3331 WARN_ON(start + len > eb->start + eb->len);
3333 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3336 page = extent_buffer_page(eb, i);
3338 cur = min(len, (PAGE_CACHE_SIZE - offset));
3340 kaddr = kmap_atomic(page, KM_USER0);
3341 ret = memcmp(ptr, kaddr + offset, cur);
3342 kunmap_atomic(kaddr, KM_USER0);
3353 EXPORT_SYMBOL(memcmp_extent_buffer);
3355 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
3356 unsigned long start, unsigned long len)
3362 char *src = (char *)srcv;
3363 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3364 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3366 WARN_ON(start > eb->len);
3367 WARN_ON(start + len > eb->start + eb->len);
3369 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3372 page = extent_buffer_page(eb, i);
3373 WARN_ON(!PageUptodate(page));
3375 cur = min(len, PAGE_CACHE_SIZE - offset);
3376 kaddr = kmap_atomic(page, KM_USER1);
3377 memcpy(kaddr + offset, src, cur);
3378 kunmap_atomic(kaddr, KM_USER1);
3386 EXPORT_SYMBOL(write_extent_buffer);
3388 void memset_extent_buffer(struct extent_buffer *eb, char c,
3389 unsigned long start, unsigned long len)
3395 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3396 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3398 WARN_ON(start > eb->len);
3399 WARN_ON(start + len > eb->start + eb->len);
3401 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3404 page = extent_buffer_page(eb, i);
3405 WARN_ON(!PageUptodate(page));
3407 cur = min(len, PAGE_CACHE_SIZE - offset);
3408 kaddr = kmap_atomic(page, KM_USER0);
3409 memset(kaddr + offset, c, cur);
3410 kunmap_atomic(kaddr, KM_USER0);
3417 EXPORT_SYMBOL(memset_extent_buffer);
3419 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
3420 unsigned long dst_offset, unsigned long src_offset,
3423 u64 dst_len = dst->len;
3428 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3429 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3431 WARN_ON(src->len != dst_len);
3433 offset = (start_offset + dst_offset) &
3434 ((unsigned long)PAGE_CACHE_SIZE - 1);
3437 page = extent_buffer_page(dst, i);
3438 WARN_ON(!PageUptodate(page));
3440 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
3442 kaddr = kmap_atomic(page, KM_USER0);
3443 read_extent_buffer(src, kaddr + offset, src_offset, cur);
3444 kunmap_atomic(kaddr, KM_USER0);
3452 EXPORT_SYMBOL(copy_extent_buffer);
3454 static void move_pages(struct page *dst_page, struct page *src_page,
3455 unsigned long dst_off, unsigned long src_off,
3458 char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3459 if (dst_page == src_page) {
3460 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
3462 char *src_kaddr = kmap_atomic(src_page, KM_USER1);
3463 char *p = dst_kaddr + dst_off + len;
3464 char *s = src_kaddr + src_off + len;
3469 kunmap_atomic(src_kaddr, KM_USER1);
3471 kunmap_atomic(dst_kaddr, KM_USER0);
3474 static void copy_pages(struct page *dst_page, struct page *src_page,
3475 unsigned long dst_off, unsigned long src_off,
3478 char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3481 if (dst_page != src_page)
3482 src_kaddr = kmap_atomic(src_page, KM_USER1);
3484 src_kaddr = dst_kaddr;
3486 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
3487 kunmap_atomic(dst_kaddr, KM_USER0);
3488 if (dst_page != src_page)
3489 kunmap_atomic(src_kaddr, KM_USER1);
3492 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3493 unsigned long src_offset, unsigned long len)
3496 size_t dst_off_in_page;
3497 size_t src_off_in_page;
3498 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3499 unsigned long dst_i;
3500 unsigned long src_i;
3502 if (src_offset + len > dst->len) {
3503 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
3504 src_offset, len, dst->len);
3507 if (dst_offset + len > dst->len) {
3508 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
3509 dst_offset, len, dst->len);
3514 dst_off_in_page = (start_offset + dst_offset) &
3515 ((unsigned long)PAGE_CACHE_SIZE - 1);
3516 src_off_in_page = (start_offset + src_offset) &
3517 ((unsigned long)PAGE_CACHE_SIZE - 1);
3519 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3520 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
3522 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
3524 cur = min_t(unsigned long, cur,
3525 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
3527 copy_pages(extent_buffer_page(dst, dst_i),
3528 extent_buffer_page(dst, src_i),
3529 dst_off_in_page, src_off_in_page, cur);
3536 EXPORT_SYMBOL(memcpy_extent_buffer);
3538 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3539 unsigned long src_offset, unsigned long len)
3542 size_t dst_off_in_page;
3543 size_t src_off_in_page;
3544 unsigned long dst_end = dst_offset + len - 1;
3545 unsigned long src_end = src_offset + len - 1;
3546 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3547 unsigned long dst_i;
3548 unsigned long src_i;
3550 if (src_offset + len > dst->len) {
3551 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
3552 src_offset, len, dst->len);
3555 if (dst_offset + len > dst->len) {
3556 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
3557 dst_offset, len, dst->len);
3560 if (dst_offset < src_offset) {
3561 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
3565 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
3566 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
3568 dst_off_in_page = (start_offset + dst_end) &
3569 ((unsigned long)PAGE_CACHE_SIZE - 1);
3570 src_off_in_page = (start_offset + src_end) &
3571 ((unsigned long)PAGE_CACHE_SIZE - 1);
3573 cur = min_t(unsigned long, len, src_off_in_page + 1);
3574 cur = min(cur, dst_off_in_page + 1);
3575 move_pages(extent_buffer_page(dst, dst_i),
3576 extent_buffer_page(dst, src_i),
3577 dst_off_in_page - cur + 1,
3578 src_off_in_page - cur + 1, cur);
3585 EXPORT_SYMBOL(memmove_extent_buffer);