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/writeback.h>
13 #include <linux/pagevec.h>
14 #include "extent_io.h"
15 #include "extent_map.h"
18 #include "btrfs_inode.h"
20 static struct kmem_cache *extent_state_cache;
21 static struct kmem_cache *extent_buffer_cache;
23 static LIST_HEAD(buffers);
24 static LIST_HEAD(states);
28 static DEFINE_SPINLOCK(leak_lock);
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;
44 /* tells writepage not to lock the state bits for this range
45 * it still does the unlocking
47 unsigned int extent_locked:1;
49 /* tells the submit_bio code to use a WRITE_SYNC */
50 unsigned int sync_io:1;
53 int __init extent_io_init(void)
55 extent_state_cache = kmem_cache_create("extent_state",
56 sizeof(struct extent_state), 0,
57 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
58 if (!extent_state_cache)
61 extent_buffer_cache = kmem_cache_create("extent_buffers",
62 sizeof(struct extent_buffer), 0,
63 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
64 if (!extent_buffer_cache)
65 goto free_state_cache;
69 kmem_cache_destroy(extent_state_cache);
73 void extent_io_exit(void)
75 struct extent_state *state;
76 struct extent_buffer *eb;
78 while (!list_empty(&states)) {
79 state = list_entry(states.next, struct extent_state, leak_list);
80 printk(KERN_ERR "btrfs state leak: start %llu end %llu "
81 "state %lu in tree %p refs %d\n",
82 (unsigned long long)state->start,
83 (unsigned long long)state->end,
84 state->state, state->tree, atomic_read(&state->refs));
85 list_del(&state->leak_list);
86 kmem_cache_free(extent_state_cache, state);
90 while (!list_empty(&buffers)) {
91 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
92 printk(KERN_ERR "btrfs buffer leak start %llu len %lu "
93 "refs %d\n", (unsigned long long)eb->start,
94 eb->len, atomic_read(&eb->refs));
95 list_del(&eb->leak_list);
96 kmem_cache_free(extent_buffer_cache, eb);
98 if (extent_state_cache)
99 kmem_cache_destroy(extent_state_cache);
100 if (extent_buffer_cache)
101 kmem_cache_destroy(extent_buffer_cache);
104 void extent_io_tree_init(struct extent_io_tree *tree,
105 struct address_space *mapping, gfp_t mask)
107 tree->state.rb_node = NULL;
108 tree->buffer.rb_node = NULL;
110 tree->dirty_bytes = 0;
111 spin_lock_init(&tree->lock);
112 spin_lock_init(&tree->buffer_lock);
113 tree->mapping = mapping;
116 static struct extent_state *alloc_extent_state(gfp_t mask)
118 struct extent_state *state;
123 state = kmem_cache_alloc(extent_state_cache, mask);
130 spin_lock_irqsave(&leak_lock, flags);
131 list_add(&state->leak_list, &states);
132 spin_unlock_irqrestore(&leak_lock, flags);
134 atomic_set(&state->refs, 1);
135 init_waitqueue_head(&state->wq);
139 static void free_extent_state(struct extent_state *state)
143 if (atomic_dec_and_test(&state->refs)) {
147 WARN_ON(state->tree);
149 spin_lock_irqsave(&leak_lock, flags);
150 list_del(&state->leak_list);
151 spin_unlock_irqrestore(&leak_lock, flags);
153 kmem_cache_free(extent_state_cache, state);
157 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
158 struct rb_node *node)
160 struct rb_node **p = &root->rb_node;
161 struct rb_node *parent = NULL;
162 struct tree_entry *entry;
166 entry = rb_entry(parent, struct tree_entry, rb_node);
168 if (offset < entry->start)
170 else if (offset > entry->end)
176 entry = rb_entry(node, struct tree_entry, rb_node);
177 rb_link_node(node, parent, p);
178 rb_insert_color(node, root);
182 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
183 struct rb_node **prev_ret,
184 struct rb_node **next_ret)
186 struct rb_root *root = &tree->state;
187 struct rb_node *n = root->rb_node;
188 struct rb_node *prev = NULL;
189 struct rb_node *orig_prev = NULL;
190 struct tree_entry *entry;
191 struct tree_entry *prev_entry = NULL;
194 entry = rb_entry(n, struct tree_entry, rb_node);
198 if (offset < entry->start)
200 else if (offset > entry->end)
208 while (prev && offset > prev_entry->end) {
209 prev = rb_next(prev);
210 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
217 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
218 while (prev && offset < prev_entry->start) {
219 prev = rb_prev(prev);
220 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
227 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
230 struct rb_node *prev = NULL;
233 ret = __etree_search(tree, offset, &prev, NULL);
239 static struct extent_buffer *buffer_tree_insert(struct extent_io_tree *tree,
240 u64 offset, struct rb_node *node)
242 struct rb_root *root = &tree->buffer;
243 struct rb_node **p = &root->rb_node;
244 struct rb_node *parent = NULL;
245 struct extent_buffer *eb;
249 eb = rb_entry(parent, struct extent_buffer, rb_node);
251 if (offset < eb->start)
253 else if (offset > eb->start)
259 rb_link_node(node, parent, p);
260 rb_insert_color(node, root);
264 static struct extent_buffer *buffer_search(struct extent_io_tree *tree,
267 struct rb_root *root = &tree->buffer;
268 struct rb_node *n = root->rb_node;
269 struct extent_buffer *eb;
272 eb = rb_entry(n, struct extent_buffer, rb_node);
273 if (offset < eb->start)
275 else if (offset > eb->start)
284 * utility function to look for merge candidates inside a given range.
285 * Any extents with matching state are merged together into a single
286 * extent in the tree. Extents with EXTENT_IO in their state field
287 * are not merged because the end_io handlers need to be able to do
288 * operations on them without sleeping (or doing allocations/splits).
290 * This should be called with the tree lock held.
292 static int merge_state(struct extent_io_tree *tree,
293 struct extent_state *state)
295 struct extent_state *other;
296 struct rb_node *other_node;
298 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
301 other_node = rb_prev(&state->rb_node);
303 other = rb_entry(other_node, struct extent_state, rb_node);
304 if (other->end == state->start - 1 &&
305 other->state == state->state) {
306 state->start = other->start;
308 rb_erase(&other->rb_node, &tree->state);
309 free_extent_state(other);
312 other_node = rb_next(&state->rb_node);
314 other = rb_entry(other_node, struct extent_state, rb_node);
315 if (other->start == state->end + 1 &&
316 other->state == state->state) {
317 other->start = state->start;
319 rb_erase(&state->rb_node, &tree->state);
320 free_extent_state(state);
326 static void set_state_cb(struct extent_io_tree *tree,
327 struct extent_state *state,
330 if (tree->ops && tree->ops->set_bit_hook) {
331 tree->ops->set_bit_hook(tree->mapping->host, state->start,
332 state->end, state->state, bits);
336 static void clear_state_cb(struct extent_io_tree *tree,
337 struct extent_state *state,
340 if (tree->ops && tree->ops->clear_bit_hook) {
341 tree->ops->clear_bit_hook(tree->mapping->host, state->start,
342 state->end, state->state, bits);
347 * insert an extent_state struct into the tree. 'bits' are set on the
348 * struct before it is inserted.
350 * This may return -EEXIST if the extent is already there, in which case the
351 * state struct is freed.
353 * The tree lock is not taken internally. This is a utility function and
354 * probably isn't what you want to call (see set/clear_extent_bit).
356 static int insert_state(struct extent_io_tree *tree,
357 struct extent_state *state, u64 start, u64 end,
360 struct rb_node *node;
363 printk(KERN_ERR "btrfs end < start %llu %llu\n",
364 (unsigned long long)end,
365 (unsigned long long)start);
368 if (bits & EXTENT_DIRTY)
369 tree->dirty_bytes += end - start + 1;
370 state->start = start;
372 set_state_cb(tree, state, bits);
373 state->state |= bits;
374 node = tree_insert(&tree->state, end, &state->rb_node);
376 struct extent_state *found;
377 found = rb_entry(node, struct extent_state, rb_node);
378 printk(KERN_ERR "btrfs found node %llu %llu on insert of "
379 "%llu %llu\n", (unsigned long long)found->start,
380 (unsigned long long)found->end,
381 (unsigned long long)start, (unsigned long long)end);
382 free_extent_state(state);
386 merge_state(tree, state);
391 * split a given extent state struct in two, inserting the preallocated
392 * struct 'prealloc' as the newly created second half. 'split' indicates an
393 * offset inside 'orig' where it should be split.
396 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
397 * are two extent state structs in the tree:
398 * prealloc: [orig->start, split - 1]
399 * orig: [ split, orig->end ]
401 * The tree locks are not taken by this function. They need to be held
404 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
405 struct extent_state *prealloc, u64 split)
407 struct rb_node *node;
408 prealloc->start = orig->start;
409 prealloc->end = split - 1;
410 prealloc->state = orig->state;
413 node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
415 free_extent_state(prealloc);
418 prealloc->tree = tree;
423 * utility function to clear some bits in an extent state struct.
424 * it will optionally wake up any one waiting on this state (wake == 1), or
425 * forcibly remove the state from the tree (delete == 1).
427 * If no bits are set on the state struct after clearing things, the
428 * struct is freed and removed from the tree
430 static int clear_state_bit(struct extent_io_tree *tree,
431 struct extent_state *state, int bits, int wake,
434 int ret = state->state & bits;
436 if ((bits & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
437 u64 range = state->end - state->start + 1;
438 WARN_ON(range > tree->dirty_bytes);
439 tree->dirty_bytes -= range;
441 clear_state_cb(tree, state, bits);
442 state->state &= ~bits;
445 if (delete || state->state == 0) {
447 clear_state_cb(tree, state, state->state);
448 rb_erase(&state->rb_node, &tree->state);
450 free_extent_state(state);
455 merge_state(tree, state);
461 * clear some bits on a range in the tree. This may require splitting
462 * or inserting elements in the tree, so the gfp mask is used to
463 * indicate which allocations or sleeping are allowed.
465 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
466 * the given range from the tree regardless of state (ie for truncate).
468 * the range [start, end] is inclusive.
470 * This takes the tree lock, and returns < 0 on error, > 0 if any of the
471 * bits were already set, or zero if none of the bits were already set.
473 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
474 int bits, int wake, int delete,
475 struct extent_state **cached_state,
478 struct extent_state *state;
479 struct extent_state *cached;
480 struct extent_state *prealloc = NULL;
481 struct rb_node *next_node;
482 struct rb_node *node;
488 if (!prealloc && (mask & __GFP_WAIT)) {
489 prealloc = alloc_extent_state(mask);
494 spin_lock(&tree->lock);
496 cached = *cached_state;
497 *cached_state = NULL;
498 if (cached->tree && cached->start == start) {
499 atomic_dec(&cached->refs);
501 last_end = state->end;
504 free_extent_state(cached);
507 * this search will find the extents that end after
510 node = tree_search(tree, start);
513 state = rb_entry(node, struct extent_state, rb_node);
515 if (state->start > end)
517 WARN_ON(state->end < start);
518 last_end = state->end;
521 * | ---- desired range ---- |
523 * | ------------- state -------------- |
525 * We need to split the extent we found, and may flip
526 * bits on second half.
528 * If the extent we found extends past our range, we
529 * just split and search again. It'll get split again
530 * the next time though.
532 * If the extent we found is inside our range, we clear
533 * the desired bit on it.
536 if (state->start < start) {
538 prealloc = alloc_extent_state(GFP_ATOMIC);
539 err = split_state(tree, state, prealloc, start);
540 BUG_ON(err == -EEXIST);
544 if (state->end <= end) {
545 set |= clear_state_bit(tree, state, bits,
547 if (last_end == (u64)-1)
549 start = last_end + 1;
551 start = state->start;
556 * | ---- desired range ---- |
558 * We need to split the extent, and clear the bit
561 if (state->start <= end && state->end > end) {
563 prealloc = alloc_extent_state(GFP_ATOMIC);
564 err = split_state(tree, state, prealloc, end + 1);
565 BUG_ON(err == -EEXIST);
569 set |= clear_state_bit(tree, prealloc, bits,
575 if (state->end < end && prealloc && !need_resched())
576 next_node = rb_next(&state->rb_node);
579 set |= clear_state_bit(tree, state, bits, wake, delete);
580 if (last_end == (u64)-1)
582 start = last_end + 1;
583 if (start <= end && next_node) {
584 state = rb_entry(next_node, struct extent_state,
586 if (state->start == start)
592 spin_unlock(&tree->lock);
594 free_extent_state(prealloc);
601 spin_unlock(&tree->lock);
602 if (mask & __GFP_WAIT)
607 static int wait_on_state(struct extent_io_tree *tree,
608 struct extent_state *state)
609 __releases(tree->lock)
610 __acquires(tree->lock)
613 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
614 spin_unlock(&tree->lock);
616 spin_lock(&tree->lock);
617 finish_wait(&state->wq, &wait);
622 * waits for one or more bits to clear on a range in the state tree.
623 * The range [start, end] is inclusive.
624 * The tree lock is taken by this function
626 int wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits)
628 struct extent_state *state;
629 struct rb_node *node;
631 spin_lock(&tree->lock);
635 * this search will find all the extents that end after
638 node = tree_search(tree, start);
642 state = rb_entry(node, struct extent_state, rb_node);
644 if (state->start > end)
647 if (state->state & bits) {
648 start = state->start;
649 atomic_inc(&state->refs);
650 wait_on_state(tree, state);
651 free_extent_state(state);
654 start = state->end + 1;
659 if (need_resched()) {
660 spin_unlock(&tree->lock);
662 spin_lock(&tree->lock);
666 spin_unlock(&tree->lock);
670 static void set_state_bits(struct extent_io_tree *tree,
671 struct extent_state *state,
674 if ((bits & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
675 u64 range = state->end - state->start + 1;
676 tree->dirty_bytes += range;
678 set_state_cb(tree, state, bits);
679 state->state |= bits;
682 static void cache_state(struct extent_state *state,
683 struct extent_state **cached_ptr)
685 if (cached_ptr && !(*cached_ptr)) {
686 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) {
688 atomic_inc(&state->refs);
694 * set some bits on a range in the tree. This may require allocations or
695 * sleeping, so the gfp mask is used to indicate what is allowed.
697 * If any of the exclusive bits are set, this will fail with -EEXIST if some
698 * part of the range already has the desired bits set. The start of the
699 * existing range is returned in failed_start in this case.
701 * [start, end] is inclusive This takes the tree lock.
704 static int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
705 int bits, int exclusive_bits, u64 *failed_start,
706 struct extent_state **cached_state,
709 struct extent_state *state;
710 struct extent_state *prealloc = NULL;
711 struct rb_node *node;
716 if (!prealloc && (mask & __GFP_WAIT)) {
717 prealloc = alloc_extent_state(mask);
722 spin_lock(&tree->lock);
724 * this search will find all the extents that end after
727 node = tree_search(tree, start);
729 err = insert_state(tree, prealloc, start, end, bits);
731 BUG_ON(err == -EEXIST);
734 state = rb_entry(node, struct extent_state, rb_node);
736 last_start = state->start;
737 last_end = state->end;
740 * | ---- desired range ---- |
743 * Just lock what we found and keep going
745 if (state->start == start && state->end <= end) {
746 struct rb_node *next_node;
747 if (state->state & exclusive_bits) {
748 *failed_start = state->start;
752 set_state_bits(tree, state, bits);
753 cache_state(state, cached_state);
754 merge_state(tree, state);
755 if (last_end == (u64)-1)
758 start = last_end + 1;
759 if (start < end && prealloc && !need_resched()) {
760 next_node = rb_next(node);
762 state = rb_entry(next_node, struct extent_state,
764 if (state->start == start)
772 * | ---- desired range ---- |
775 * | ------------- state -------------- |
777 * We need to split the extent we found, and may flip bits on
780 * If the extent we found extends past our
781 * range, we just split and search again. It'll get split
782 * again the next time though.
784 * If the extent we found is inside our range, we set the
787 if (state->start < start) {
788 if (state->state & exclusive_bits) {
789 *failed_start = start;
793 err = split_state(tree, state, prealloc, start);
794 BUG_ON(err == -EEXIST);
798 if (state->end <= end) {
799 set_state_bits(tree, state, bits);
800 cache_state(state, cached_state);
801 merge_state(tree, state);
802 if (last_end == (u64)-1)
804 start = last_end + 1;
806 start = state->start;
811 * | ---- desired range ---- |
812 * | state | or | state |
814 * There's a hole, we need to insert something in it and
815 * ignore the extent we found.
817 if (state->start > start) {
819 if (end < last_start)
822 this_end = last_start - 1;
823 err = insert_state(tree, prealloc, start, this_end,
825 cache_state(prealloc, cached_state);
827 BUG_ON(err == -EEXIST);
830 start = this_end + 1;
834 * | ---- desired range ---- |
836 * We need to split the extent, and set the bit
839 if (state->start <= end && state->end > end) {
840 if (state->state & exclusive_bits) {
841 *failed_start = start;
845 err = split_state(tree, state, prealloc, end + 1);
846 BUG_ON(err == -EEXIST);
848 set_state_bits(tree, prealloc, bits);
849 cache_state(prealloc, cached_state);
850 merge_state(tree, prealloc);
858 spin_unlock(&tree->lock);
860 free_extent_state(prealloc);
867 spin_unlock(&tree->lock);
868 if (mask & __GFP_WAIT)
873 /* wrappers around set/clear extent bit */
874 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
877 return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL,
881 int set_extent_ordered(struct extent_io_tree *tree, u64 start, u64 end,
884 return set_extent_bit(tree, start, end, EXTENT_ORDERED, 0, NULL, NULL,
888 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
889 int bits, gfp_t mask)
891 return set_extent_bit(tree, start, end, bits, 0, NULL,
895 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
896 int bits, gfp_t mask)
898 return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask);
901 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
904 return set_extent_bit(tree, start, end,
905 EXTENT_DELALLOC | EXTENT_DIRTY | EXTENT_UPTODATE,
906 0, NULL, NULL, mask);
909 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
912 return clear_extent_bit(tree, start, end,
913 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0,
917 int clear_extent_ordered(struct extent_io_tree *tree, u64 start, u64 end,
920 return clear_extent_bit(tree, start, end, EXTENT_ORDERED, 1, 0,
924 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
927 return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL,
931 static int clear_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
934 return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0,
938 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
941 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, NULL,
945 static int clear_extent_uptodate(struct extent_io_tree *tree, u64 start,
948 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0,
952 int wait_on_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end)
954 return wait_extent_bit(tree, start, end, EXTENT_WRITEBACK);
958 * either insert or lock state struct between start and end use mask to tell
959 * us if waiting is desired.
961 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
962 int bits, struct extent_state **cached_state, gfp_t mask)
967 err = set_extent_bit(tree, start, end, EXTENT_LOCKED | bits,
968 EXTENT_LOCKED, &failed_start,
970 if (err == -EEXIST && (mask & __GFP_WAIT)) {
971 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
972 start = failed_start;
976 WARN_ON(start > end);
981 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
983 return lock_extent_bits(tree, start, end, 0, NULL, mask);
986 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end,
992 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
993 &failed_start, NULL, mask);
994 if (err == -EEXIST) {
995 if (failed_start > start)
996 clear_extent_bit(tree, start, failed_start - 1,
997 EXTENT_LOCKED, 1, 0, NULL, mask);
1003 int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end,
1004 struct extent_state **cached, gfp_t mask)
1006 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached,
1010 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end,
1013 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL,
1018 * helper function to set pages and extents in the tree dirty
1020 int set_range_dirty(struct extent_io_tree *tree, u64 start, u64 end)
1022 unsigned long index = start >> PAGE_CACHE_SHIFT;
1023 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1026 while (index <= end_index) {
1027 page = find_get_page(tree->mapping, index);
1029 __set_page_dirty_nobuffers(page);
1030 page_cache_release(page);
1037 * helper function to set both pages and extents in the tree writeback
1039 static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
1041 unsigned long index = start >> PAGE_CACHE_SHIFT;
1042 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1045 while (index <= end_index) {
1046 page = find_get_page(tree->mapping, index);
1048 set_page_writeback(page);
1049 page_cache_release(page);
1056 * find the first offset in the io tree with 'bits' set. zero is
1057 * returned if we find something, and *start_ret and *end_ret are
1058 * set to reflect the state struct that was found.
1060 * If nothing was found, 1 is returned, < 0 on error
1062 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1063 u64 *start_ret, u64 *end_ret, int bits)
1065 struct rb_node *node;
1066 struct extent_state *state;
1069 spin_lock(&tree->lock);
1071 * this search will find all the extents that end after
1074 node = tree_search(tree, start);
1079 state = rb_entry(node, struct extent_state, rb_node);
1080 if (state->end >= start && (state->state & bits)) {
1081 *start_ret = state->start;
1082 *end_ret = state->end;
1086 node = rb_next(node);
1091 spin_unlock(&tree->lock);
1095 /* find the first state struct with 'bits' set after 'start', and
1096 * return it. tree->lock must be held. NULL will returned if
1097 * nothing was found after 'start'
1099 struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree,
1100 u64 start, int bits)
1102 struct rb_node *node;
1103 struct extent_state *state;
1106 * this search will find all the extents that end after
1109 node = tree_search(tree, start);
1114 state = rb_entry(node, struct extent_state, rb_node);
1115 if (state->end >= start && (state->state & bits))
1118 node = rb_next(node);
1127 * find a contiguous range of bytes in the file marked as delalloc, not
1128 * more than 'max_bytes'. start and end are used to return the range,
1130 * 1 is returned if we find something, 0 if nothing was in the tree
1132 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1133 u64 *start, u64 *end, u64 max_bytes)
1135 struct rb_node *node;
1136 struct extent_state *state;
1137 u64 cur_start = *start;
1139 u64 total_bytes = 0;
1141 spin_lock(&tree->lock);
1144 * this search will find all the extents that end after
1147 node = tree_search(tree, cur_start);
1155 state = rb_entry(node, struct extent_state, rb_node);
1156 if (found && (state->start != cur_start ||
1157 (state->state & EXTENT_BOUNDARY))) {
1160 if (!(state->state & EXTENT_DELALLOC)) {
1166 *start = state->start;
1169 cur_start = state->end + 1;
1170 node = rb_next(node);
1173 total_bytes += state->end - state->start + 1;
1174 if (total_bytes >= max_bytes)
1178 spin_unlock(&tree->lock);
1182 static noinline int __unlock_for_delalloc(struct inode *inode,
1183 struct page *locked_page,
1187 struct page *pages[16];
1188 unsigned long index = start >> PAGE_CACHE_SHIFT;
1189 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1190 unsigned long nr_pages = end_index - index + 1;
1193 if (index == locked_page->index && end_index == index)
1196 while (nr_pages > 0) {
1197 ret = find_get_pages_contig(inode->i_mapping, index,
1198 min_t(unsigned long, nr_pages,
1199 ARRAY_SIZE(pages)), pages);
1200 for (i = 0; i < ret; i++) {
1201 if (pages[i] != locked_page)
1202 unlock_page(pages[i]);
1203 page_cache_release(pages[i]);
1212 static noinline int lock_delalloc_pages(struct inode *inode,
1213 struct page *locked_page,
1217 unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT;
1218 unsigned long start_index = index;
1219 unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT;
1220 unsigned long pages_locked = 0;
1221 struct page *pages[16];
1222 unsigned long nrpages;
1226 /* the caller is responsible for locking the start index */
1227 if (index == locked_page->index && index == end_index)
1230 /* skip the page at the start index */
1231 nrpages = end_index - index + 1;
1232 while (nrpages > 0) {
1233 ret = find_get_pages_contig(inode->i_mapping, index,
1234 min_t(unsigned long,
1235 nrpages, ARRAY_SIZE(pages)), pages);
1240 /* now we have an array of pages, lock them all */
1241 for (i = 0; i < ret; i++) {
1243 * the caller is taking responsibility for
1246 if (pages[i] != locked_page) {
1247 lock_page(pages[i]);
1248 if (!PageDirty(pages[i]) ||
1249 pages[i]->mapping != inode->i_mapping) {
1251 unlock_page(pages[i]);
1252 page_cache_release(pages[i]);
1256 page_cache_release(pages[i]);
1265 if (ret && pages_locked) {
1266 __unlock_for_delalloc(inode, locked_page,
1268 ((u64)(start_index + pages_locked - 1)) <<
1275 * find a contiguous range of bytes in the file marked as delalloc, not
1276 * more than 'max_bytes'. start and end are used to return the range,
1278 * 1 is returned if we find something, 0 if nothing was in the tree
1280 static noinline u64 find_lock_delalloc_range(struct inode *inode,
1281 struct extent_io_tree *tree,
1282 struct page *locked_page,
1283 u64 *start, u64 *end,
1293 /* step one, find a bunch of delalloc bytes starting at start */
1294 delalloc_start = *start;
1296 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1298 if (!found || delalloc_end <= *start) {
1299 *start = delalloc_start;
1300 *end = delalloc_end;
1305 * start comes from the offset of locked_page. We have to lock
1306 * pages in order, so we can't process delalloc bytes before
1309 if (delalloc_start < *start)
1310 delalloc_start = *start;
1313 * make sure to limit the number of pages we try to lock down
1316 if (delalloc_end + 1 - delalloc_start > max_bytes && loops)
1317 delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1;
1319 /* step two, lock all the pages after the page that has start */
1320 ret = lock_delalloc_pages(inode, locked_page,
1321 delalloc_start, delalloc_end);
1322 if (ret == -EAGAIN) {
1323 /* some of the pages are gone, lets avoid looping by
1324 * shortening the size of the delalloc range we're searching
1327 unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1);
1328 max_bytes = PAGE_CACHE_SIZE - offset;
1338 /* step three, lock the state bits for the whole range */
1339 lock_extent(tree, delalloc_start, delalloc_end, GFP_NOFS);
1341 /* then test to make sure it is all still delalloc */
1342 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1343 EXTENT_DELALLOC, 1);
1345 unlock_extent(tree, delalloc_start, delalloc_end, GFP_NOFS);
1346 __unlock_for_delalloc(inode, locked_page,
1347 delalloc_start, delalloc_end);
1351 *start = delalloc_start;
1352 *end = delalloc_end;
1357 int extent_clear_unlock_delalloc(struct inode *inode,
1358 struct extent_io_tree *tree,
1359 u64 start, u64 end, struct page *locked_page,
1362 int clear_delalloc, int clear_dirty,
1367 struct page *pages[16];
1368 unsigned long index = start >> PAGE_CACHE_SHIFT;
1369 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1370 unsigned long nr_pages = end_index - index + 1;
1375 clear_bits |= EXTENT_LOCKED;
1377 clear_bits |= EXTENT_DIRTY;
1380 clear_bits |= EXTENT_DELALLOC;
1382 clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS);
1383 if (!(unlock_pages || clear_dirty || set_writeback || end_writeback))
1386 while (nr_pages > 0) {
1387 ret = find_get_pages_contig(inode->i_mapping, index,
1388 min_t(unsigned long,
1389 nr_pages, ARRAY_SIZE(pages)), pages);
1390 for (i = 0; i < ret; i++) {
1391 if (pages[i] == locked_page) {
1392 page_cache_release(pages[i]);
1396 clear_page_dirty_for_io(pages[i]);
1398 set_page_writeback(pages[i]);
1400 end_page_writeback(pages[i]);
1402 unlock_page(pages[i]);
1403 page_cache_release(pages[i]);
1413 * count the number of bytes in the tree that have a given bit(s)
1414 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1415 * cached. The total number found is returned.
1417 u64 count_range_bits(struct extent_io_tree *tree,
1418 u64 *start, u64 search_end, u64 max_bytes,
1421 struct rb_node *node;
1422 struct extent_state *state;
1423 u64 cur_start = *start;
1424 u64 total_bytes = 0;
1427 if (search_end <= cur_start) {
1432 spin_lock(&tree->lock);
1433 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1434 total_bytes = tree->dirty_bytes;
1438 * this search will find all the extents that end after
1441 node = tree_search(tree, cur_start);
1446 state = rb_entry(node, struct extent_state, rb_node);
1447 if (state->start > search_end)
1449 if (state->end >= cur_start && (state->state & bits)) {
1450 total_bytes += min(search_end, state->end) + 1 -
1451 max(cur_start, state->start);
1452 if (total_bytes >= max_bytes)
1455 *start = state->start;
1459 node = rb_next(node);
1464 spin_unlock(&tree->lock);
1469 * set the private field for a given byte offset in the tree. If there isn't
1470 * an extent_state there already, this does nothing.
1472 int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1474 struct rb_node *node;
1475 struct extent_state *state;
1478 spin_lock(&tree->lock);
1480 * this search will find all the extents that end after
1483 node = tree_search(tree, start);
1488 state = rb_entry(node, struct extent_state, rb_node);
1489 if (state->start != start) {
1493 state->private = private;
1495 spin_unlock(&tree->lock);
1499 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1501 struct rb_node *node;
1502 struct extent_state *state;
1505 spin_lock(&tree->lock);
1507 * this search will find all the extents that end after
1510 node = tree_search(tree, start);
1515 state = rb_entry(node, struct extent_state, rb_node);
1516 if (state->start != start) {
1520 *private = state->private;
1522 spin_unlock(&tree->lock);
1527 * searches a range in the state tree for a given mask.
1528 * If 'filled' == 1, this returns 1 only if every extent in the tree
1529 * has the bits set. Otherwise, 1 is returned if any bit in the
1530 * range is found set.
1532 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1533 int bits, int filled)
1535 struct extent_state *state = NULL;
1536 struct rb_node *node;
1539 spin_lock(&tree->lock);
1540 node = tree_search(tree, start);
1541 while (node && start <= end) {
1542 state = rb_entry(node, struct extent_state, rb_node);
1544 if (filled && state->start > start) {
1549 if (state->start > end)
1552 if (state->state & bits) {
1556 } else if (filled) {
1560 start = state->end + 1;
1563 node = rb_next(node);
1570 spin_unlock(&tree->lock);
1575 * helper function to set a given page up to date if all the
1576 * extents in the tree for that page are up to date
1578 static int check_page_uptodate(struct extent_io_tree *tree,
1581 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1582 u64 end = start + PAGE_CACHE_SIZE - 1;
1583 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1))
1584 SetPageUptodate(page);
1589 * helper function to unlock a page if all the extents in the tree
1590 * for that page are unlocked
1592 static int check_page_locked(struct extent_io_tree *tree,
1595 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1596 u64 end = start + PAGE_CACHE_SIZE - 1;
1597 if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0))
1603 * helper function to end page writeback if all the extents
1604 * in the tree for that page are done with writeback
1606 static int check_page_writeback(struct extent_io_tree *tree,
1609 end_page_writeback(page);
1613 /* lots and lots of room for performance fixes in the end_bio funcs */
1616 * after a writepage IO is done, we need to:
1617 * clear the uptodate bits on error
1618 * clear the writeback bits in the extent tree for this IO
1619 * end_page_writeback if the page has no more pending IO
1621 * Scheduling is not allowed, so the extent state tree is expected
1622 * to have one and only one object corresponding to this IO.
1624 static void end_bio_extent_writepage(struct bio *bio, int err)
1626 int uptodate = err == 0;
1627 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1628 struct extent_io_tree *tree;
1635 struct page *page = bvec->bv_page;
1636 tree = &BTRFS_I(page->mapping->host)->io_tree;
1638 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1640 end = start + bvec->bv_len - 1;
1642 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1647 if (--bvec >= bio->bi_io_vec)
1648 prefetchw(&bvec->bv_page->flags);
1649 if (tree->ops && tree->ops->writepage_end_io_hook) {
1650 ret = tree->ops->writepage_end_io_hook(page, start,
1651 end, NULL, uptodate);
1656 if (!uptodate && tree->ops &&
1657 tree->ops->writepage_io_failed_hook) {
1658 ret = tree->ops->writepage_io_failed_hook(bio, page,
1661 uptodate = (err == 0);
1667 clear_extent_uptodate(tree, start, end, GFP_NOFS);
1668 ClearPageUptodate(page);
1673 end_page_writeback(page);
1675 check_page_writeback(tree, page);
1676 } while (bvec >= bio->bi_io_vec);
1682 * after a readpage IO is done, we need to:
1683 * clear the uptodate bits on error
1684 * set the uptodate bits if things worked
1685 * set the page up to date if all extents in the tree are uptodate
1686 * clear the lock bit in the extent tree
1687 * unlock the page if there are no other extents locked for it
1689 * Scheduling is not allowed, so the extent state tree is expected
1690 * to have one and only one object corresponding to this IO.
1692 static void end_bio_extent_readpage(struct bio *bio, int err)
1694 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1695 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1696 struct extent_io_tree *tree;
1706 struct page *page = bvec->bv_page;
1707 tree = &BTRFS_I(page->mapping->host)->io_tree;
1709 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1711 end = start + bvec->bv_len - 1;
1713 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1718 if (--bvec >= bio->bi_io_vec)
1719 prefetchw(&bvec->bv_page->flags);
1721 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
1722 ret = tree->ops->readpage_end_io_hook(page, start, end,
1727 if (!uptodate && tree->ops &&
1728 tree->ops->readpage_io_failed_hook) {
1729 ret = tree->ops->readpage_io_failed_hook(bio, page,
1733 test_bit(BIO_UPTODATE, &bio->bi_flags);
1741 set_extent_uptodate(tree, start, end,
1744 unlock_extent(tree, start, end, GFP_ATOMIC);
1748 SetPageUptodate(page);
1750 ClearPageUptodate(page);
1756 check_page_uptodate(tree, page);
1758 ClearPageUptodate(page);
1761 check_page_locked(tree, page);
1763 } while (bvec >= bio->bi_io_vec);
1769 * IO done from prepare_write is pretty simple, we just unlock
1770 * the structs in the extent tree when done, and set the uptodate bits
1773 static void end_bio_extent_preparewrite(struct bio *bio, int err)
1775 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1776 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1777 struct extent_io_tree *tree;
1782 struct page *page = bvec->bv_page;
1783 tree = &BTRFS_I(page->mapping->host)->io_tree;
1785 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1787 end = start + bvec->bv_len - 1;
1789 if (--bvec >= bio->bi_io_vec)
1790 prefetchw(&bvec->bv_page->flags);
1793 set_extent_uptodate(tree, start, end, GFP_ATOMIC);
1795 ClearPageUptodate(page);
1799 unlock_extent(tree, start, end, GFP_ATOMIC);
1801 } while (bvec >= bio->bi_io_vec);
1807 extent_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
1812 bio = bio_alloc(gfp_flags, nr_vecs);
1814 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
1815 while (!bio && (nr_vecs /= 2))
1816 bio = bio_alloc(gfp_flags, nr_vecs);
1821 bio->bi_bdev = bdev;
1822 bio->bi_sector = first_sector;
1827 static int submit_one_bio(int rw, struct bio *bio, int mirror_num,
1828 unsigned long bio_flags)
1831 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1832 struct page *page = bvec->bv_page;
1833 struct extent_io_tree *tree = bio->bi_private;
1837 start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
1838 end = start + bvec->bv_len - 1;
1840 bio->bi_private = NULL;
1844 if (tree->ops && tree->ops->submit_bio_hook)
1845 tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
1846 mirror_num, bio_flags);
1848 submit_bio(rw, bio);
1849 if (bio_flagged(bio, BIO_EOPNOTSUPP))
1855 static int submit_extent_page(int rw, struct extent_io_tree *tree,
1856 struct page *page, sector_t sector,
1857 size_t size, unsigned long offset,
1858 struct block_device *bdev,
1859 struct bio **bio_ret,
1860 unsigned long max_pages,
1861 bio_end_io_t end_io_func,
1863 unsigned long prev_bio_flags,
1864 unsigned long bio_flags)
1870 int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED;
1871 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
1872 size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE);
1874 if (bio_ret && *bio_ret) {
1877 contig = bio->bi_sector == sector;
1879 contig = bio->bi_sector + (bio->bi_size >> 9) ==
1882 if (prev_bio_flags != bio_flags || !contig ||
1883 (tree->ops && tree->ops->merge_bio_hook &&
1884 tree->ops->merge_bio_hook(page, offset, page_size, bio,
1886 bio_add_page(bio, page, page_size, offset) < page_size) {
1887 ret = submit_one_bio(rw, bio, mirror_num,
1894 if (this_compressed)
1897 nr = bio_get_nr_vecs(bdev);
1899 bio = extent_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
1901 bio_add_page(bio, page, page_size, offset);
1902 bio->bi_end_io = end_io_func;
1903 bio->bi_private = tree;
1908 ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
1913 void set_page_extent_mapped(struct page *page)
1915 if (!PagePrivate(page)) {
1916 SetPagePrivate(page);
1917 page_cache_get(page);
1918 set_page_private(page, EXTENT_PAGE_PRIVATE);
1922 static void set_page_extent_head(struct page *page, unsigned long len)
1924 set_page_private(page, EXTENT_PAGE_PRIVATE_FIRST_PAGE | len << 2);
1928 * basic readpage implementation. Locked extent state structs are inserted
1929 * into the tree that are removed when the IO is done (by the end_io
1932 static int __extent_read_full_page(struct extent_io_tree *tree,
1934 get_extent_t *get_extent,
1935 struct bio **bio, int mirror_num,
1936 unsigned long *bio_flags)
1938 struct inode *inode = page->mapping->host;
1939 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1940 u64 page_end = start + PAGE_CACHE_SIZE - 1;
1944 u64 last_byte = i_size_read(inode);
1948 struct extent_map *em;
1949 struct block_device *bdev;
1952 size_t page_offset = 0;
1954 size_t disk_io_size;
1955 size_t blocksize = inode->i_sb->s_blocksize;
1956 unsigned long this_bio_flag = 0;
1958 set_page_extent_mapped(page);
1961 lock_extent(tree, start, end, GFP_NOFS);
1963 if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
1965 size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
1968 iosize = PAGE_CACHE_SIZE - zero_offset;
1969 userpage = kmap_atomic(page, KM_USER0);
1970 memset(userpage + zero_offset, 0, iosize);
1971 flush_dcache_page(page);
1972 kunmap_atomic(userpage, KM_USER0);
1975 while (cur <= end) {
1976 if (cur >= last_byte) {
1978 iosize = PAGE_CACHE_SIZE - page_offset;
1979 userpage = kmap_atomic(page, KM_USER0);
1980 memset(userpage + page_offset, 0, iosize);
1981 flush_dcache_page(page);
1982 kunmap_atomic(userpage, KM_USER0);
1983 set_extent_uptodate(tree, cur, cur + iosize - 1,
1985 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1988 em = get_extent(inode, page, page_offset, cur,
1990 if (IS_ERR(em) || !em) {
1992 unlock_extent(tree, cur, end, GFP_NOFS);
1995 extent_offset = cur - em->start;
1996 BUG_ON(extent_map_end(em) <= cur);
1999 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
2000 this_bio_flag = EXTENT_BIO_COMPRESSED;
2002 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2003 cur_end = min(extent_map_end(em) - 1, end);
2004 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2005 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2006 disk_io_size = em->block_len;
2007 sector = em->block_start >> 9;
2009 sector = (em->block_start + extent_offset) >> 9;
2010 disk_io_size = iosize;
2013 block_start = em->block_start;
2014 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2015 block_start = EXTENT_MAP_HOLE;
2016 free_extent_map(em);
2019 /* we've found a hole, just zero and go on */
2020 if (block_start == EXTENT_MAP_HOLE) {
2022 userpage = kmap_atomic(page, KM_USER0);
2023 memset(userpage + page_offset, 0, iosize);
2024 flush_dcache_page(page);
2025 kunmap_atomic(userpage, KM_USER0);
2027 set_extent_uptodate(tree, cur, cur + iosize - 1,
2029 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
2031 page_offset += iosize;
2034 /* the get_extent function already copied into the page */
2035 if (test_range_bit(tree, cur, cur_end, EXTENT_UPTODATE, 1)) {
2036 check_page_uptodate(tree, page);
2037 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
2039 page_offset += iosize;
2042 /* we have an inline extent but it didn't get marked up
2043 * to date. Error out
2045 if (block_start == EXTENT_MAP_INLINE) {
2047 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
2049 page_offset += iosize;
2054 if (tree->ops && tree->ops->readpage_io_hook) {
2055 ret = tree->ops->readpage_io_hook(page, cur,
2059 unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
2061 ret = submit_extent_page(READ, tree, page,
2062 sector, disk_io_size, page_offset,
2064 end_bio_extent_readpage, mirror_num,
2068 *bio_flags = this_bio_flag;
2073 page_offset += iosize;
2076 if (!PageError(page))
2077 SetPageUptodate(page);
2083 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
2084 get_extent_t *get_extent)
2086 struct bio *bio = NULL;
2087 unsigned long bio_flags = 0;
2090 ret = __extent_read_full_page(tree, page, get_extent, &bio, 0,
2093 submit_one_bio(READ, bio, 0, bio_flags);
2097 static noinline void update_nr_written(struct page *page,
2098 struct writeback_control *wbc,
2099 unsigned long nr_written)
2101 wbc->nr_to_write -= nr_written;
2102 if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
2103 wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
2104 page->mapping->writeback_index = page->index + nr_written;
2108 * the writepage semantics are similar to regular writepage. extent
2109 * records are inserted to lock ranges in the tree, and as dirty areas
2110 * are found, they are marked writeback. Then the lock bits are removed
2111 * and the end_io handler clears the writeback ranges
2113 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2116 struct inode *inode = page->mapping->host;
2117 struct extent_page_data *epd = data;
2118 struct extent_io_tree *tree = epd->tree;
2119 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2121 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2125 u64 last_byte = i_size_read(inode);
2130 struct extent_state *cached_state = NULL;
2131 struct extent_map *em;
2132 struct block_device *bdev;
2135 size_t pg_offset = 0;
2137 loff_t i_size = i_size_read(inode);
2138 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
2144 unsigned long nr_written = 0;
2146 if (wbc->sync_mode == WB_SYNC_ALL)
2147 write_flags = WRITE_SYNC_PLUG;
2149 write_flags = WRITE;
2151 WARN_ON(!PageLocked(page));
2152 pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
2153 if (page->index > end_index ||
2154 (page->index == end_index && !pg_offset)) {
2155 page->mapping->a_ops->invalidatepage(page, 0);
2160 if (page->index == end_index) {
2163 userpage = kmap_atomic(page, KM_USER0);
2164 memset(userpage + pg_offset, 0,
2165 PAGE_CACHE_SIZE - pg_offset);
2166 kunmap_atomic(userpage, KM_USER0);
2167 flush_dcache_page(page);
2171 set_page_extent_mapped(page);
2173 delalloc_start = start;
2176 if (!epd->extent_locked) {
2177 u64 delalloc_to_write;
2179 * make sure the wbc mapping index is at least updated
2182 update_nr_written(page, wbc, 0);
2184 while (delalloc_end < page_end) {
2185 nr_delalloc = find_lock_delalloc_range(inode, tree,
2190 if (nr_delalloc == 0) {
2191 delalloc_start = delalloc_end + 1;
2194 tree->ops->fill_delalloc(inode, page, delalloc_start,
2195 delalloc_end, &page_started,
2197 delalloc_to_write = (delalloc_end -
2198 max_t(u64, page_offset(page),
2199 delalloc_start) + 1) >>
2201 if (wbc->nr_to_write < delalloc_to_write) {
2202 wbc->nr_to_write = min_t(long, 8192,
2205 delalloc_start = delalloc_end + 1;
2208 /* did the fill delalloc function already unlock and start
2214 * we've unlocked the page, so we can't update
2215 * the mapping's writeback index, just update
2218 wbc->nr_to_write -= nr_written;
2222 if (tree->ops && tree->ops->writepage_start_hook) {
2223 ret = tree->ops->writepage_start_hook(page, start,
2225 if (ret == -EAGAIN) {
2226 redirty_page_for_writepage(wbc, page);
2227 update_nr_written(page, wbc, nr_written);
2235 * we don't want to touch the inode after unlocking the page,
2236 * so we update the mapping writeback index now
2238 update_nr_written(page, wbc, nr_written + 1);
2241 if (last_byte <= start) {
2242 if (tree->ops && tree->ops->writepage_end_io_hook)
2243 tree->ops->writepage_end_io_hook(page, start,
2245 unlock_start = page_end + 1;
2249 blocksize = inode->i_sb->s_blocksize;
2251 while (cur <= end) {
2252 if (cur >= last_byte) {
2253 if (tree->ops && tree->ops->writepage_end_io_hook)
2254 tree->ops->writepage_end_io_hook(page, cur,
2256 unlock_start = page_end + 1;
2259 em = epd->get_extent(inode, page, pg_offset, cur,
2261 if (IS_ERR(em) || !em) {
2266 extent_offset = cur - em->start;
2267 BUG_ON(extent_map_end(em) <= cur);
2269 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2270 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2271 sector = (em->block_start + extent_offset) >> 9;
2273 block_start = em->block_start;
2274 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
2275 free_extent_map(em);
2279 * compressed and inline extents are written through other
2282 if (compressed || block_start == EXTENT_MAP_HOLE ||
2283 block_start == EXTENT_MAP_INLINE) {
2285 * end_io notification does not happen here for
2286 * compressed extents
2288 if (!compressed && tree->ops &&
2289 tree->ops->writepage_end_io_hook)
2290 tree->ops->writepage_end_io_hook(page, cur,
2293 else if (compressed) {
2294 /* we don't want to end_page_writeback on
2295 * a compressed extent. this happens
2302 pg_offset += iosize;
2306 /* leave this out until we have a page_mkwrite call */
2307 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
2310 pg_offset += iosize;
2314 if (tree->ops && tree->ops->writepage_io_hook) {
2315 ret = tree->ops->writepage_io_hook(page, cur,
2323 unsigned long max_nr = end_index + 1;
2325 set_range_writeback(tree, cur, cur + iosize - 1);
2326 if (!PageWriteback(page)) {
2327 printk(KERN_ERR "btrfs warning page %lu not "
2328 "writeback, cur %llu end %llu\n",
2329 page->index, (unsigned long long)cur,
2330 (unsigned long long)end);
2333 ret = submit_extent_page(write_flags, tree, page,
2334 sector, iosize, pg_offset,
2335 bdev, &epd->bio, max_nr,
2336 end_bio_extent_writepage,
2342 pg_offset += iosize;
2347 /* make sure the mapping tag for page dirty gets cleared */
2348 set_page_writeback(page);
2349 end_page_writeback(page);
2355 /* drop our reference on any cached states */
2356 free_extent_state(cached_state);
2361 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
2362 * @mapping: address space structure to write
2363 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2364 * @writepage: function called for each page
2365 * @data: data passed to writepage function
2367 * If a page is already under I/O, write_cache_pages() skips it, even
2368 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
2369 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
2370 * and msync() need to guarantee that all the data which was dirty at the time
2371 * the call was made get new I/O started against them. If wbc->sync_mode is
2372 * WB_SYNC_ALL then we were called for data integrity and we must wait for
2373 * existing IO to complete.
2375 static int extent_write_cache_pages(struct extent_io_tree *tree,
2376 struct address_space *mapping,
2377 struct writeback_control *wbc,
2378 writepage_t writepage, void *data,
2379 void (*flush_fn)(void *))
2383 struct pagevec pvec;
2386 pgoff_t end; /* Inclusive */
2388 int range_whole = 0;
2390 pagevec_init(&pvec, 0);
2391 if (wbc->range_cyclic) {
2392 index = mapping->writeback_index; /* Start from prev offset */
2395 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2396 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2397 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2402 while (!done && (index <= end) &&
2403 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
2404 PAGECACHE_TAG_DIRTY, min(end - index,
2405 (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
2409 for (i = 0; i < nr_pages; i++) {
2410 struct page *page = pvec.pages[i];
2413 * At this point we hold neither mapping->tree_lock nor
2414 * lock on the page itself: the page may be truncated or
2415 * invalidated (changing page->mapping to NULL), or even
2416 * swizzled back from swapper_space to tmpfs file
2419 if (tree->ops && tree->ops->write_cache_pages_lock_hook)
2420 tree->ops->write_cache_pages_lock_hook(page);
2424 if (unlikely(page->mapping != mapping)) {
2429 if (!wbc->range_cyclic && page->index > end) {
2435 if (wbc->sync_mode != WB_SYNC_NONE) {
2436 if (PageWriteback(page))
2438 wait_on_page_writeback(page);
2441 if (PageWriteback(page) ||
2442 !clear_page_dirty_for_io(page)) {
2447 ret = (*writepage)(page, wbc, data);
2449 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
2453 if (ret || wbc->nr_to_write <= 0)
2456 pagevec_release(&pvec);
2459 if (!scanned && !done) {
2461 * We hit the last page and there is more work to be done: wrap
2462 * back to the start of the file
2471 static void flush_epd_write_bio(struct extent_page_data *epd)
2475 submit_one_bio(WRITE_SYNC, epd->bio, 0, 0);
2477 submit_one_bio(WRITE, epd->bio, 0, 0);
2482 static noinline void flush_write_bio(void *data)
2484 struct extent_page_data *epd = data;
2485 flush_epd_write_bio(epd);
2488 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
2489 get_extent_t *get_extent,
2490 struct writeback_control *wbc)
2493 struct address_space *mapping = page->mapping;
2494 struct extent_page_data epd = {
2497 .get_extent = get_extent,
2499 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
2501 struct writeback_control wbc_writepages = {
2503 .sync_mode = wbc->sync_mode,
2504 .older_than_this = NULL,
2506 .range_start = page_offset(page) + PAGE_CACHE_SIZE,
2507 .range_end = (loff_t)-1,
2510 ret = __extent_writepage(page, wbc, &epd);
2512 extent_write_cache_pages(tree, mapping, &wbc_writepages,
2513 __extent_writepage, &epd, flush_write_bio);
2514 flush_epd_write_bio(&epd);
2518 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
2519 u64 start, u64 end, get_extent_t *get_extent,
2523 struct address_space *mapping = inode->i_mapping;
2525 unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
2528 struct extent_page_data epd = {
2531 .get_extent = get_extent,
2533 .sync_io = mode == WB_SYNC_ALL,
2535 struct writeback_control wbc_writepages = {
2536 .bdi = inode->i_mapping->backing_dev_info,
2538 .older_than_this = NULL,
2539 .nr_to_write = nr_pages * 2,
2540 .range_start = start,
2541 .range_end = end + 1,
2544 while (start <= end) {
2545 page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
2546 if (clear_page_dirty_for_io(page))
2547 ret = __extent_writepage(page, &wbc_writepages, &epd);
2549 if (tree->ops && tree->ops->writepage_end_io_hook)
2550 tree->ops->writepage_end_io_hook(page, start,
2551 start + PAGE_CACHE_SIZE - 1,
2555 page_cache_release(page);
2556 start += PAGE_CACHE_SIZE;
2559 flush_epd_write_bio(&epd);
2563 int extent_writepages(struct extent_io_tree *tree,
2564 struct address_space *mapping,
2565 get_extent_t *get_extent,
2566 struct writeback_control *wbc)
2569 struct extent_page_data epd = {
2572 .get_extent = get_extent,
2574 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
2577 ret = extent_write_cache_pages(tree, mapping, wbc,
2578 __extent_writepage, &epd,
2580 flush_epd_write_bio(&epd);
2584 int extent_readpages(struct extent_io_tree *tree,
2585 struct address_space *mapping,
2586 struct list_head *pages, unsigned nr_pages,
2587 get_extent_t get_extent)
2589 struct bio *bio = NULL;
2591 struct pagevec pvec;
2592 unsigned long bio_flags = 0;
2594 pagevec_init(&pvec, 0);
2595 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
2596 struct page *page = list_entry(pages->prev, struct page, lru);
2598 prefetchw(&page->flags);
2599 list_del(&page->lru);
2601 * what we want to do here is call add_to_page_cache_lru,
2602 * but that isn't exported, so we reproduce it here
2604 if (!add_to_page_cache(page, mapping,
2605 page->index, GFP_KERNEL)) {
2607 /* open coding of lru_cache_add, also not exported */
2608 page_cache_get(page);
2609 if (!pagevec_add(&pvec, page))
2610 __pagevec_lru_add_file(&pvec);
2611 __extent_read_full_page(tree, page, get_extent,
2612 &bio, 0, &bio_flags);
2614 page_cache_release(page);
2616 if (pagevec_count(&pvec))
2617 __pagevec_lru_add_file(&pvec);
2618 BUG_ON(!list_empty(pages));
2620 submit_one_bio(READ, bio, 0, bio_flags);
2625 * basic invalidatepage code, this waits on any locked or writeback
2626 * ranges corresponding to the page, and then deletes any extent state
2627 * records from the tree
2629 int extent_invalidatepage(struct extent_io_tree *tree,
2630 struct page *page, unsigned long offset)
2632 u64 start = ((u64)page->index << PAGE_CACHE_SHIFT);
2633 u64 end = start + PAGE_CACHE_SIZE - 1;
2634 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
2636 start += (offset + blocksize - 1) & ~(blocksize - 1);
2640 lock_extent(tree, start, end, GFP_NOFS);
2641 wait_on_page_writeback(page);
2642 clear_extent_bit(tree, start, end,
2643 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC,
2644 1, 1, NULL, GFP_NOFS);
2649 * simple commit_write call, set_range_dirty is used to mark both
2650 * the pages and the extent records as dirty
2652 int extent_commit_write(struct extent_io_tree *tree,
2653 struct inode *inode, struct page *page,
2654 unsigned from, unsigned to)
2656 loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
2658 set_page_extent_mapped(page);
2659 set_page_dirty(page);
2661 if (pos > inode->i_size) {
2662 i_size_write(inode, pos);
2663 mark_inode_dirty(inode);
2668 int extent_prepare_write(struct extent_io_tree *tree,
2669 struct inode *inode, struct page *page,
2670 unsigned from, unsigned to, get_extent_t *get_extent)
2672 u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2673 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
2675 u64 orig_block_start;
2678 struct extent_map *em;
2679 unsigned blocksize = 1 << inode->i_blkbits;
2680 size_t page_offset = 0;
2681 size_t block_off_start;
2682 size_t block_off_end;
2688 set_page_extent_mapped(page);
2690 block_start = (page_start + from) & ~((u64)blocksize - 1);
2691 block_end = (page_start + to - 1) | (blocksize - 1);
2692 orig_block_start = block_start;
2694 lock_extent(tree, page_start, page_end, GFP_NOFS);
2695 while (block_start <= block_end) {
2696 em = get_extent(inode, page, page_offset, block_start,
2697 block_end - block_start + 1, 1);
2698 if (IS_ERR(em) || !em)
2701 cur_end = min(block_end, extent_map_end(em) - 1);
2702 block_off_start = block_start & (PAGE_CACHE_SIZE - 1);
2703 block_off_end = block_off_start + blocksize;
2704 isnew = clear_extent_new(tree, block_start, cur_end, GFP_NOFS);
2706 if (!PageUptodate(page) && isnew &&
2707 (block_off_end > to || block_off_start < from)) {
2710 kaddr = kmap_atomic(page, KM_USER0);
2711 if (block_off_end > to)
2712 memset(kaddr + to, 0, block_off_end - to);
2713 if (block_off_start < from)
2714 memset(kaddr + block_off_start, 0,
2715 from - block_off_start);
2716 flush_dcache_page(page);
2717 kunmap_atomic(kaddr, KM_USER0);
2719 if ((em->block_start != EXTENT_MAP_HOLE &&
2720 em->block_start != EXTENT_MAP_INLINE) &&
2721 !isnew && !PageUptodate(page) &&
2722 (block_off_end > to || block_off_start < from) &&
2723 !test_range_bit(tree, block_start, cur_end,
2724 EXTENT_UPTODATE, 1)) {
2726 u64 extent_offset = block_start - em->start;
2728 sector = (em->block_start + extent_offset) >> 9;
2729 iosize = (cur_end - block_start + blocksize) &
2730 ~((u64)blocksize - 1);
2732 * we've already got the extent locked, but we
2733 * need to split the state such that our end_bio
2734 * handler can clear the lock.
2736 set_extent_bit(tree, block_start,
2737 block_start + iosize - 1,
2738 EXTENT_LOCKED, 0, NULL, NULL, GFP_NOFS);
2739 ret = submit_extent_page(READ, tree, page,
2740 sector, iosize, page_offset, em->bdev,
2742 end_bio_extent_preparewrite, 0,
2745 block_start = block_start + iosize;
2747 set_extent_uptodate(tree, block_start, cur_end,
2749 unlock_extent(tree, block_start, cur_end, GFP_NOFS);
2750 block_start = cur_end + 1;
2752 page_offset = block_start & (PAGE_CACHE_SIZE - 1);
2753 free_extent_map(em);
2756 wait_extent_bit(tree, orig_block_start,
2757 block_end, EXTENT_LOCKED);
2759 check_page_uptodate(tree, page);
2761 /* FIXME, zero out newly allocated blocks on error */
2766 * a helper for releasepage, this tests for areas of the page that
2767 * are locked or under IO and drops the related state bits if it is safe
2770 int try_release_extent_state(struct extent_map_tree *map,
2771 struct extent_io_tree *tree, struct page *page,
2774 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2775 u64 end = start + PAGE_CACHE_SIZE - 1;
2778 if (test_range_bit(tree, start, end,
2779 EXTENT_IOBITS | EXTENT_ORDERED, 0))
2782 if ((mask & GFP_NOFS) == GFP_NOFS)
2784 clear_extent_bit(tree, start, end, EXTENT_UPTODATE,
2791 * a helper for releasepage. As long as there are no locked extents
2792 * in the range corresponding to the page, both state records and extent
2793 * map records are removed
2795 int try_release_extent_mapping(struct extent_map_tree *map,
2796 struct extent_io_tree *tree, struct page *page,
2799 struct extent_map *em;
2800 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2801 u64 end = start + PAGE_CACHE_SIZE - 1;
2803 if ((mask & __GFP_WAIT) &&
2804 page->mapping->host->i_size > 16 * 1024 * 1024) {
2806 while (start <= end) {
2807 len = end - start + 1;
2808 write_lock(&map->lock);
2809 em = lookup_extent_mapping(map, start, len);
2810 if (!em || IS_ERR(em)) {
2811 write_unlock(&map->lock);
2814 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
2815 em->start != start) {
2816 write_unlock(&map->lock);
2817 free_extent_map(em);
2820 if (!test_range_bit(tree, em->start,
2821 extent_map_end(em) - 1,
2822 EXTENT_LOCKED | EXTENT_WRITEBACK |
2825 remove_extent_mapping(map, em);
2826 /* once for the rb tree */
2827 free_extent_map(em);
2829 start = extent_map_end(em);
2830 write_unlock(&map->lock);
2833 free_extent_map(em);
2836 return try_release_extent_state(map, tree, page, mask);
2839 sector_t extent_bmap(struct address_space *mapping, sector_t iblock,
2840 get_extent_t *get_extent)
2842 struct inode *inode = mapping->host;
2843 u64 start = iblock << inode->i_blkbits;
2844 sector_t sector = 0;
2845 size_t blksize = (1 << inode->i_blkbits);
2846 struct extent_map *em;
2848 lock_extent(&BTRFS_I(inode)->io_tree, start, start + blksize - 1,
2850 em = get_extent(inode, NULL, 0, start, blksize, 0);
2851 unlock_extent(&BTRFS_I(inode)->io_tree, start, start + blksize - 1,
2853 if (!em || IS_ERR(em))
2856 if (em->block_start > EXTENT_MAP_LAST_BYTE)
2859 sector = (em->block_start + start - em->start) >> inode->i_blkbits;
2861 free_extent_map(em);
2865 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
2866 __u64 start, __u64 len, get_extent_t *get_extent)
2870 u64 max = start + len;
2873 struct extent_map *em = NULL;
2875 u64 em_start = 0, em_len = 0;
2876 unsigned long emflags;
2882 lock_extent(&BTRFS_I(inode)->io_tree, start, start + len,
2884 em = get_extent(inode, NULL, 0, off, max - off, 0);
2892 off = em->start + em->len;
2896 em_start = em->start;
2902 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
2904 flags |= FIEMAP_EXTENT_LAST;
2905 } else if (em->block_start == EXTENT_MAP_HOLE) {
2906 flags |= FIEMAP_EXTENT_UNWRITTEN;
2907 } else if (em->block_start == EXTENT_MAP_INLINE) {
2908 flags |= (FIEMAP_EXTENT_DATA_INLINE |
2909 FIEMAP_EXTENT_NOT_ALIGNED);
2910 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
2911 flags |= (FIEMAP_EXTENT_DELALLOC |
2912 FIEMAP_EXTENT_UNKNOWN);
2914 disko = em->block_start;
2916 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
2917 flags |= FIEMAP_EXTENT_ENCODED;
2919 emflags = em->flags;
2920 free_extent_map(em);
2924 em = get_extent(inode, NULL, 0, off, max - off, 0);
2931 emflags = em->flags;
2933 if (test_bit(EXTENT_FLAG_VACANCY, &emflags)) {
2934 flags |= FIEMAP_EXTENT_LAST;
2938 ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
2944 free_extent_map(em);
2946 unlock_extent(&BTRFS_I(inode)->io_tree, start, start + len,
2951 static inline struct page *extent_buffer_page(struct extent_buffer *eb,
2955 struct address_space *mapping;
2958 return eb->first_page;
2959 i += eb->start >> PAGE_CACHE_SHIFT;
2960 mapping = eb->first_page->mapping;
2965 * extent_buffer_page is only called after pinning the page
2966 * by increasing the reference count. So we know the page must
2967 * be in the radix tree.
2970 p = radix_tree_lookup(&mapping->page_tree, i);
2976 static inline unsigned long num_extent_pages(u64 start, u64 len)
2978 return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
2979 (start >> PAGE_CACHE_SHIFT);
2982 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
2987 struct extent_buffer *eb = NULL;
2989 unsigned long flags;
2992 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
2995 spin_lock_init(&eb->lock);
2996 init_waitqueue_head(&eb->lock_wq);
2999 spin_lock_irqsave(&leak_lock, flags);
3000 list_add(&eb->leak_list, &buffers);
3001 spin_unlock_irqrestore(&leak_lock, flags);
3003 atomic_set(&eb->refs, 1);
3008 static void __free_extent_buffer(struct extent_buffer *eb)
3011 unsigned long flags;
3012 spin_lock_irqsave(&leak_lock, flags);
3013 list_del(&eb->leak_list);
3014 spin_unlock_irqrestore(&leak_lock, flags);
3016 kmem_cache_free(extent_buffer_cache, eb);
3019 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
3020 u64 start, unsigned long len,
3024 unsigned long num_pages = num_extent_pages(start, len);
3026 unsigned long index = start >> PAGE_CACHE_SHIFT;
3027 struct extent_buffer *eb;
3028 struct extent_buffer *exists = NULL;
3030 struct address_space *mapping = tree->mapping;
3033 spin_lock(&tree->buffer_lock);
3034 eb = buffer_search(tree, start);
3036 atomic_inc(&eb->refs);
3037 spin_unlock(&tree->buffer_lock);
3038 mark_page_accessed(eb->first_page);
3041 spin_unlock(&tree->buffer_lock);
3043 eb = __alloc_extent_buffer(tree, start, len, mask);
3048 eb->first_page = page0;
3051 page_cache_get(page0);
3052 mark_page_accessed(page0);
3053 set_page_extent_mapped(page0);
3054 set_page_extent_head(page0, len);
3055 uptodate = PageUptodate(page0);
3059 for (; i < num_pages; i++, index++) {
3060 p = find_or_create_page(mapping, index, mask | __GFP_HIGHMEM);
3065 set_page_extent_mapped(p);
3066 mark_page_accessed(p);
3069 set_page_extent_head(p, len);
3071 set_page_private(p, EXTENT_PAGE_PRIVATE);
3073 if (!PageUptodate(p))
3078 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3080 spin_lock(&tree->buffer_lock);
3081 exists = buffer_tree_insert(tree, start, &eb->rb_node);
3083 /* add one reference for the caller */
3084 atomic_inc(&exists->refs);
3085 spin_unlock(&tree->buffer_lock);
3088 spin_unlock(&tree->buffer_lock);
3090 /* add one reference for the tree */
3091 atomic_inc(&eb->refs);
3095 if (!atomic_dec_and_test(&eb->refs))
3097 for (index = 1; index < i; index++)
3098 page_cache_release(extent_buffer_page(eb, index));
3099 page_cache_release(extent_buffer_page(eb, 0));
3100 __free_extent_buffer(eb);
3104 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
3105 u64 start, unsigned long len,
3108 struct extent_buffer *eb;
3110 spin_lock(&tree->buffer_lock);
3111 eb = buffer_search(tree, start);
3113 atomic_inc(&eb->refs);
3114 spin_unlock(&tree->buffer_lock);
3117 mark_page_accessed(eb->first_page);
3122 void free_extent_buffer(struct extent_buffer *eb)
3127 if (!atomic_dec_and_test(&eb->refs))
3133 int clear_extent_buffer_dirty(struct extent_io_tree *tree,
3134 struct extent_buffer *eb)
3137 unsigned long num_pages;
3140 num_pages = num_extent_pages(eb->start, eb->len);
3142 for (i = 0; i < num_pages; i++) {
3143 page = extent_buffer_page(eb, i);
3144 if (!PageDirty(page))
3149 set_page_extent_head(page, eb->len);
3151 set_page_private(page, EXTENT_PAGE_PRIVATE);
3153 clear_page_dirty_for_io(page);
3154 spin_lock_irq(&page->mapping->tree_lock);
3155 if (!PageDirty(page)) {
3156 radix_tree_tag_clear(&page->mapping->page_tree,
3158 PAGECACHE_TAG_DIRTY);
3160 spin_unlock_irq(&page->mapping->tree_lock);
3166 int wait_on_extent_buffer_writeback(struct extent_io_tree *tree,
3167 struct extent_buffer *eb)
3169 return wait_on_extent_writeback(tree, eb->start,
3170 eb->start + eb->len - 1);
3173 int set_extent_buffer_dirty(struct extent_io_tree *tree,
3174 struct extent_buffer *eb)
3177 unsigned long num_pages;
3180 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
3181 num_pages = num_extent_pages(eb->start, eb->len);
3182 for (i = 0; i < num_pages; i++)
3183 __set_page_dirty_nobuffers(extent_buffer_page(eb, i));
3187 int clear_extent_buffer_uptodate(struct extent_io_tree *tree,
3188 struct extent_buffer *eb)
3192 unsigned long num_pages;
3194 num_pages = num_extent_pages(eb->start, eb->len);
3195 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3197 clear_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
3199 for (i = 0; i < num_pages; i++) {
3200 page = extent_buffer_page(eb, i);
3202 ClearPageUptodate(page);
3207 int set_extent_buffer_uptodate(struct extent_io_tree *tree,
3208 struct extent_buffer *eb)
3212 unsigned long num_pages;
3214 num_pages = num_extent_pages(eb->start, eb->len);
3216 set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
3218 for (i = 0; i < num_pages; i++) {
3219 page = extent_buffer_page(eb, i);
3220 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
3221 ((i == num_pages - 1) &&
3222 ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
3223 check_page_uptodate(tree, page);
3226 SetPageUptodate(page);
3231 int extent_range_uptodate(struct extent_io_tree *tree,
3236 int pg_uptodate = 1;
3238 unsigned long index;
3240 ret = test_range_bit(tree, start, end, EXTENT_UPTODATE, 1);
3243 while (start <= end) {
3244 index = start >> PAGE_CACHE_SHIFT;
3245 page = find_get_page(tree->mapping, index);
3246 uptodate = PageUptodate(page);
3247 page_cache_release(page);
3252 start += PAGE_CACHE_SIZE;
3257 int extent_buffer_uptodate(struct extent_io_tree *tree,
3258 struct extent_buffer *eb)
3261 unsigned long num_pages;
3264 int pg_uptodate = 1;
3266 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
3269 ret = test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3270 EXTENT_UPTODATE, 1);
3274 num_pages = num_extent_pages(eb->start, eb->len);
3275 for (i = 0; i < num_pages; i++) {
3276 page = extent_buffer_page(eb, i);
3277 if (!PageUptodate(page)) {
3285 int read_extent_buffer_pages(struct extent_io_tree *tree,
3286 struct extent_buffer *eb,
3287 u64 start, int wait,
3288 get_extent_t *get_extent, int mirror_num)
3291 unsigned long start_i;
3295 int locked_pages = 0;
3296 int all_uptodate = 1;
3297 int inc_all_pages = 0;
3298 unsigned long num_pages;
3299 struct bio *bio = NULL;
3300 unsigned long bio_flags = 0;
3302 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
3305 if (test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3306 EXTENT_UPTODATE, 1)) {
3311 WARN_ON(start < eb->start);
3312 start_i = (start >> PAGE_CACHE_SHIFT) -
3313 (eb->start >> PAGE_CACHE_SHIFT);
3318 num_pages = num_extent_pages(eb->start, eb->len);
3319 for (i = start_i; i < num_pages; i++) {
3320 page = extent_buffer_page(eb, i);
3322 if (!trylock_page(page))
3328 if (!PageUptodate(page))
3333 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3337 for (i = start_i; i < num_pages; i++) {
3338 page = extent_buffer_page(eb, i);
3340 page_cache_get(page);
3341 if (!PageUptodate(page)) {
3344 ClearPageError(page);
3345 err = __extent_read_full_page(tree, page,
3347 mirror_num, &bio_flags);
3356 submit_one_bio(READ, bio, mirror_num, bio_flags);
3361 for (i = start_i; i < num_pages; i++) {
3362 page = extent_buffer_page(eb, i);
3363 wait_on_page_locked(page);
3364 if (!PageUptodate(page))
3369 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3374 while (locked_pages > 0) {
3375 page = extent_buffer_page(eb, i);
3383 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
3384 unsigned long start,
3391 char *dst = (char *)dstv;
3392 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3393 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3395 WARN_ON(start > eb->len);
3396 WARN_ON(start + len > eb->start + eb->len);
3398 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3401 page = extent_buffer_page(eb, i);
3403 cur = min(len, (PAGE_CACHE_SIZE - offset));
3404 kaddr = kmap_atomic(page, KM_USER1);
3405 memcpy(dst, kaddr + offset, cur);
3406 kunmap_atomic(kaddr, KM_USER1);
3415 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
3416 unsigned long min_len, char **token, char **map,
3417 unsigned long *map_start,
3418 unsigned long *map_len, int km)
3420 size_t offset = start & (PAGE_CACHE_SIZE - 1);
3423 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3424 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3425 unsigned long end_i = (start_offset + start + min_len - 1) >>
3432 offset = start_offset;
3436 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
3439 if (start + min_len > eb->len) {
3440 printk(KERN_ERR "btrfs bad mapping eb start %llu len %lu, "
3441 "wanted %lu %lu\n", (unsigned long long)eb->start,
3442 eb->len, start, min_len);
3446 p = extent_buffer_page(eb, i);
3447 kaddr = kmap_atomic(p, km);
3449 *map = kaddr + offset;
3450 *map_len = PAGE_CACHE_SIZE - offset;
3454 int map_extent_buffer(struct extent_buffer *eb, unsigned long start,
3455 unsigned long min_len,
3456 char **token, char **map,
3457 unsigned long *map_start,
3458 unsigned long *map_len, int km)
3462 if (eb->map_token) {
3463 unmap_extent_buffer(eb, eb->map_token, km);
3464 eb->map_token = NULL;
3467 err = map_private_extent_buffer(eb, start, min_len, token, map,
3468 map_start, map_len, km);
3470 eb->map_token = *token;
3472 eb->map_start = *map_start;
3473 eb->map_len = *map_len;
3478 void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km)
3480 kunmap_atomic(token, km);
3483 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
3484 unsigned long start,
3491 char *ptr = (char *)ptrv;
3492 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3493 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3496 WARN_ON(start > eb->len);
3497 WARN_ON(start + len > eb->start + eb->len);
3499 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3502 page = extent_buffer_page(eb, i);
3504 cur = min(len, (PAGE_CACHE_SIZE - offset));
3506 kaddr = kmap_atomic(page, KM_USER0);
3507 ret = memcmp(ptr, kaddr + offset, cur);
3508 kunmap_atomic(kaddr, KM_USER0);
3520 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
3521 unsigned long start, unsigned long len)
3527 char *src = (char *)srcv;
3528 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3529 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3531 WARN_ON(start > eb->len);
3532 WARN_ON(start + len > eb->start + eb->len);
3534 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3537 page = extent_buffer_page(eb, i);
3538 WARN_ON(!PageUptodate(page));
3540 cur = min(len, PAGE_CACHE_SIZE - offset);
3541 kaddr = kmap_atomic(page, KM_USER1);
3542 memcpy(kaddr + offset, src, cur);
3543 kunmap_atomic(kaddr, KM_USER1);
3552 void memset_extent_buffer(struct extent_buffer *eb, char c,
3553 unsigned long start, unsigned long len)
3559 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3560 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3562 WARN_ON(start > eb->len);
3563 WARN_ON(start + len > eb->start + eb->len);
3565 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3568 page = extent_buffer_page(eb, i);
3569 WARN_ON(!PageUptodate(page));
3571 cur = min(len, PAGE_CACHE_SIZE - offset);
3572 kaddr = kmap_atomic(page, KM_USER0);
3573 memset(kaddr + offset, c, cur);
3574 kunmap_atomic(kaddr, KM_USER0);
3582 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
3583 unsigned long dst_offset, unsigned long src_offset,
3586 u64 dst_len = dst->len;
3591 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3592 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3594 WARN_ON(src->len != dst_len);
3596 offset = (start_offset + dst_offset) &
3597 ((unsigned long)PAGE_CACHE_SIZE - 1);
3600 page = extent_buffer_page(dst, i);
3601 WARN_ON(!PageUptodate(page));
3603 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
3605 kaddr = kmap_atomic(page, KM_USER0);
3606 read_extent_buffer(src, kaddr + offset, src_offset, cur);
3607 kunmap_atomic(kaddr, KM_USER0);
3616 static void move_pages(struct page *dst_page, struct page *src_page,
3617 unsigned long dst_off, unsigned long src_off,
3620 char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3621 if (dst_page == src_page) {
3622 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
3624 char *src_kaddr = kmap_atomic(src_page, KM_USER1);
3625 char *p = dst_kaddr + dst_off + len;
3626 char *s = src_kaddr + src_off + len;
3631 kunmap_atomic(src_kaddr, KM_USER1);
3633 kunmap_atomic(dst_kaddr, KM_USER0);
3636 static void copy_pages(struct page *dst_page, struct page *src_page,
3637 unsigned long dst_off, unsigned long src_off,
3640 char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3643 if (dst_page != src_page)
3644 src_kaddr = kmap_atomic(src_page, KM_USER1);
3646 src_kaddr = dst_kaddr;
3648 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
3649 kunmap_atomic(dst_kaddr, KM_USER0);
3650 if (dst_page != src_page)
3651 kunmap_atomic(src_kaddr, KM_USER1);
3654 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3655 unsigned long src_offset, unsigned long len)
3658 size_t dst_off_in_page;
3659 size_t src_off_in_page;
3660 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3661 unsigned long dst_i;
3662 unsigned long src_i;
3664 if (src_offset + len > dst->len) {
3665 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
3666 "len %lu dst len %lu\n", src_offset, len, dst->len);
3669 if (dst_offset + len > dst->len) {
3670 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
3671 "len %lu dst len %lu\n", dst_offset, len, dst->len);
3676 dst_off_in_page = (start_offset + dst_offset) &
3677 ((unsigned long)PAGE_CACHE_SIZE - 1);
3678 src_off_in_page = (start_offset + src_offset) &
3679 ((unsigned long)PAGE_CACHE_SIZE - 1);
3681 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3682 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
3684 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
3686 cur = min_t(unsigned long, cur,
3687 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
3689 copy_pages(extent_buffer_page(dst, dst_i),
3690 extent_buffer_page(dst, src_i),
3691 dst_off_in_page, src_off_in_page, cur);
3699 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3700 unsigned long src_offset, unsigned long len)
3703 size_t dst_off_in_page;
3704 size_t src_off_in_page;
3705 unsigned long dst_end = dst_offset + len - 1;
3706 unsigned long src_end = src_offset + len - 1;
3707 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3708 unsigned long dst_i;
3709 unsigned long src_i;
3711 if (src_offset + len > dst->len) {
3712 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
3713 "len %lu len %lu\n", src_offset, len, dst->len);
3716 if (dst_offset + len > dst->len) {
3717 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
3718 "len %lu len %lu\n", dst_offset, len, dst->len);
3721 if (dst_offset < src_offset) {
3722 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
3726 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
3727 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
3729 dst_off_in_page = (start_offset + dst_end) &
3730 ((unsigned long)PAGE_CACHE_SIZE - 1);
3731 src_off_in_page = (start_offset + src_end) &
3732 ((unsigned long)PAGE_CACHE_SIZE - 1);
3734 cur = min_t(unsigned long, len, src_off_in_page + 1);
3735 cur = min(cur, dst_off_in_page + 1);
3736 move_pages(extent_buffer_page(dst, dst_i),
3737 extent_buffer_page(dst, src_i),
3738 dst_off_in_page - cur + 1,
3739 src_off_in_page - cur + 1, cur);
3747 int try_release_extent_buffer(struct extent_io_tree *tree, struct page *page)
3749 u64 start = page_offset(page);
3750 struct extent_buffer *eb;
3753 unsigned long num_pages;
3755 spin_lock(&tree->buffer_lock);
3756 eb = buffer_search(tree, start);
3760 if (atomic_read(&eb->refs) > 1) {
3764 if (test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3768 /* at this point we can safely release the extent buffer */
3769 num_pages = num_extent_pages(eb->start, eb->len);
3770 for (i = 0; i < num_pages; i++)
3771 page_cache_release(extent_buffer_page(eb, i));
3772 rb_erase(&eb->rb_node, &tree->buffer);
3773 __free_extent_buffer(eb);
3775 spin_unlock(&tree->buffer_lock);