1 #include <linux/bitops.h>
2 #include <linux/slab.h>
5 #include <linux/pagemap.h>
6 #include <linux/page-flags.h>
7 #include <linux/module.h>
8 #include <linux/spinlock.h>
9 #include <linux/blkdev.h>
10 #include <linux/swap.h>
11 #include <linux/writeback.h>
12 #include <linux/pagevec.h>
13 #include "extent_io.h"
14 #include "extent_map.h"
17 #include "btrfs_inode.h"
19 static struct kmem_cache *extent_state_cache;
20 static struct kmem_cache *extent_buffer_cache;
22 static LIST_HEAD(buffers);
23 static LIST_HEAD(states);
27 static DEFINE_SPINLOCK(leak_lock);
30 #define BUFFER_LRU_MAX 64
35 struct rb_node rb_node;
38 struct extent_page_data {
40 struct extent_io_tree *tree;
41 get_extent_t *get_extent;
43 /* tells writepage not to lock the state bits for this range
44 * it still does the unlocking
46 unsigned int extent_locked:1;
48 /* tells the submit_bio code to use a WRITE_SYNC */
49 unsigned int sync_io:1;
52 int __init extent_io_init(void)
54 extent_state_cache = kmem_cache_create("extent_state",
55 sizeof(struct extent_state), 0,
56 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
57 if (!extent_state_cache)
60 extent_buffer_cache = kmem_cache_create("extent_buffers",
61 sizeof(struct extent_buffer), 0,
62 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
63 if (!extent_buffer_cache)
64 goto free_state_cache;
68 kmem_cache_destroy(extent_state_cache);
72 void extent_io_exit(void)
74 struct extent_state *state;
75 struct extent_buffer *eb;
77 while (!list_empty(&states)) {
78 state = list_entry(states.next, struct extent_state, leak_list);
79 printk(KERN_ERR "btrfs state leak: start %llu end %llu "
80 "state %lu in tree %p refs %d\n",
81 (unsigned long long)state->start,
82 (unsigned long long)state->end,
83 state->state, state->tree, atomic_read(&state->refs));
84 list_del(&state->leak_list);
85 kmem_cache_free(extent_state_cache, state);
89 while (!list_empty(&buffers)) {
90 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
91 printk(KERN_ERR "btrfs buffer leak start %llu len %lu "
92 "refs %d\n", (unsigned long long)eb->start,
93 eb->len, atomic_read(&eb->refs));
94 list_del(&eb->leak_list);
95 kmem_cache_free(extent_buffer_cache, eb);
97 if (extent_state_cache)
98 kmem_cache_destroy(extent_state_cache);
99 if (extent_buffer_cache)
100 kmem_cache_destroy(extent_buffer_cache);
103 void extent_io_tree_init(struct extent_io_tree *tree,
104 struct address_space *mapping)
106 tree->state = RB_ROOT;
107 INIT_RADIX_TREE(&tree->buffer, GFP_ATOMIC);
109 tree->dirty_bytes = 0;
110 spin_lock_init(&tree->lock);
111 spin_lock_init(&tree->buffer_lock);
112 tree->mapping = mapping;
115 static struct extent_state *alloc_extent_state(gfp_t mask)
117 struct extent_state *state;
122 state = kmem_cache_alloc(extent_state_cache, mask);
129 spin_lock_irqsave(&leak_lock, flags);
130 list_add(&state->leak_list, &states);
131 spin_unlock_irqrestore(&leak_lock, flags);
133 atomic_set(&state->refs, 1);
134 init_waitqueue_head(&state->wq);
138 void free_extent_state(struct extent_state *state)
142 if (atomic_dec_and_test(&state->refs)) {
146 WARN_ON(state->tree);
148 spin_lock_irqsave(&leak_lock, flags);
149 list_del(&state->leak_list);
150 spin_unlock_irqrestore(&leak_lock, flags);
152 kmem_cache_free(extent_state_cache, state);
156 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
157 struct rb_node *node)
159 struct rb_node **p = &root->rb_node;
160 struct rb_node *parent = NULL;
161 struct tree_entry *entry;
165 entry = rb_entry(parent, struct tree_entry, rb_node);
167 if (offset < entry->start)
169 else if (offset > entry->end)
175 entry = rb_entry(node, struct tree_entry, rb_node);
176 rb_link_node(node, parent, p);
177 rb_insert_color(node, root);
181 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
182 struct rb_node **prev_ret,
183 struct rb_node **next_ret)
185 struct rb_root *root = &tree->state;
186 struct rb_node *n = root->rb_node;
187 struct rb_node *prev = NULL;
188 struct rb_node *orig_prev = NULL;
189 struct tree_entry *entry;
190 struct tree_entry *prev_entry = NULL;
193 entry = rb_entry(n, struct tree_entry, rb_node);
197 if (offset < entry->start)
199 else if (offset > entry->end)
207 while (prev && offset > prev_entry->end) {
208 prev = rb_next(prev);
209 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
216 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
217 while (prev && offset < prev_entry->start) {
218 prev = rb_prev(prev);
219 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
226 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
229 struct rb_node *prev = NULL;
232 ret = __etree_search(tree, offset, &prev, NULL);
238 static void merge_cb(struct extent_io_tree *tree, struct extent_state *new,
239 struct extent_state *other)
241 if (tree->ops && tree->ops->merge_extent_hook)
242 tree->ops->merge_extent_hook(tree->mapping->host, new,
247 * utility function to look for merge candidates inside a given range.
248 * Any extents with matching state are merged together into a single
249 * extent in the tree. Extents with EXTENT_IO in their state field
250 * are not merged because the end_io handlers need to be able to do
251 * operations on them without sleeping (or doing allocations/splits).
253 * This should be called with the tree lock held.
255 static int merge_state(struct extent_io_tree *tree,
256 struct extent_state *state)
258 struct extent_state *other;
259 struct rb_node *other_node;
261 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
264 other_node = rb_prev(&state->rb_node);
266 other = rb_entry(other_node, struct extent_state, rb_node);
267 if (other->end == state->start - 1 &&
268 other->state == state->state) {
269 merge_cb(tree, state, other);
270 state->start = other->start;
272 rb_erase(&other->rb_node, &tree->state);
273 free_extent_state(other);
276 other_node = rb_next(&state->rb_node);
278 other = rb_entry(other_node, struct extent_state, rb_node);
279 if (other->start == state->end + 1 &&
280 other->state == state->state) {
281 merge_cb(tree, state, other);
282 state->end = other->end;
284 rb_erase(&other->rb_node, &tree->state);
285 free_extent_state(other);
292 static int set_state_cb(struct extent_io_tree *tree,
293 struct extent_state *state, int *bits)
295 if (tree->ops && tree->ops->set_bit_hook) {
296 return tree->ops->set_bit_hook(tree->mapping->host,
303 static void clear_state_cb(struct extent_io_tree *tree,
304 struct extent_state *state, int *bits)
306 if (tree->ops && tree->ops->clear_bit_hook)
307 tree->ops->clear_bit_hook(tree->mapping->host, state, bits);
311 * insert an extent_state struct into the tree. 'bits' are set on the
312 * struct before it is inserted.
314 * This may return -EEXIST if the extent is already there, in which case the
315 * state struct is freed.
317 * The tree lock is not taken internally. This is a utility function and
318 * probably isn't what you want to call (see set/clear_extent_bit).
320 static int insert_state(struct extent_io_tree *tree,
321 struct extent_state *state, u64 start, u64 end,
324 struct rb_node *node;
325 int bits_to_set = *bits & ~EXTENT_CTLBITS;
329 printk(KERN_ERR "btrfs end < start %llu %llu\n",
330 (unsigned long long)end,
331 (unsigned long long)start);
334 state->start = start;
336 ret = set_state_cb(tree, state, bits);
340 if (bits_to_set & EXTENT_DIRTY)
341 tree->dirty_bytes += end - start + 1;
342 state->state |= bits_to_set;
343 node = tree_insert(&tree->state, end, &state->rb_node);
345 struct extent_state *found;
346 found = rb_entry(node, struct extent_state, rb_node);
347 printk(KERN_ERR "btrfs found node %llu %llu on insert of "
348 "%llu %llu\n", (unsigned long long)found->start,
349 (unsigned long long)found->end,
350 (unsigned long long)start, (unsigned long long)end);
354 merge_state(tree, state);
358 static int split_cb(struct extent_io_tree *tree, struct extent_state *orig,
361 if (tree->ops && tree->ops->split_extent_hook)
362 return tree->ops->split_extent_hook(tree->mapping->host,
368 * split a given extent state struct in two, inserting the preallocated
369 * struct 'prealloc' as the newly created second half. 'split' indicates an
370 * offset inside 'orig' where it should be split.
373 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
374 * are two extent state structs in the tree:
375 * prealloc: [orig->start, split - 1]
376 * orig: [ split, orig->end ]
378 * The tree locks are not taken by this function. They need to be held
381 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
382 struct extent_state *prealloc, u64 split)
384 struct rb_node *node;
386 split_cb(tree, orig, split);
388 prealloc->start = orig->start;
389 prealloc->end = split - 1;
390 prealloc->state = orig->state;
393 node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
395 free_extent_state(prealloc);
398 prealloc->tree = tree;
403 * utility function to clear some bits in an extent state struct.
404 * it will optionally wake up any one waiting on this state (wake == 1), or
405 * forcibly remove the state from the tree (delete == 1).
407 * If no bits are set on the state struct after clearing things, the
408 * struct is freed and removed from the tree
410 static int clear_state_bit(struct extent_io_tree *tree,
411 struct extent_state *state,
414 int bits_to_clear = *bits & ~EXTENT_CTLBITS;
415 int ret = state->state & bits_to_clear;
417 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
418 u64 range = state->end - state->start + 1;
419 WARN_ON(range > tree->dirty_bytes);
420 tree->dirty_bytes -= range;
422 clear_state_cb(tree, state, bits);
423 state->state &= ~bits_to_clear;
426 if (state->state == 0) {
428 rb_erase(&state->rb_node, &tree->state);
430 free_extent_state(state);
435 merge_state(tree, state);
440 static struct extent_state *
441 alloc_extent_state_atomic(struct extent_state *prealloc)
444 prealloc = alloc_extent_state(GFP_ATOMIC);
450 * clear some bits on a range in the tree. This may require splitting
451 * or inserting elements in the tree, so the gfp mask is used to
452 * indicate which allocations or sleeping are allowed.
454 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
455 * the given range from the tree regardless of state (ie for truncate).
457 * the range [start, end] is inclusive.
459 * This takes the tree lock, and returns < 0 on error, > 0 if any of the
460 * bits were already set, or zero if none of the bits were already set.
462 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
463 int bits, int wake, int delete,
464 struct extent_state **cached_state,
467 struct extent_state *state;
468 struct extent_state *cached;
469 struct extent_state *prealloc = NULL;
470 struct rb_node *next_node;
471 struct rb_node *node;
478 bits |= ~EXTENT_CTLBITS;
479 bits |= EXTENT_FIRST_DELALLOC;
481 if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY))
484 if (!prealloc && (mask & __GFP_WAIT)) {
485 prealloc = alloc_extent_state(mask);
490 spin_lock(&tree->lock);
492 cached = *cached_state;
495 *cached_state = NULL;
499 if (cached && cached->tree && cached->start <= start &&
500 cached->end > start) {
502 atomic_dec(&cached->refs);
507 free_extent_state(cached);
510 * this search will find the extents that end after
513 node = tree_search(tree, start);
516 state = rb_entry(node, struct extent_state, rb_node);
518 if (state->start > end)
520 WARN_ON(state->end < start);
521 last_end = state->end;
524 * | ---- desired range ---- |
526 * | ------------- state -------------- |
528 * We need to split the extent we found, and may flip
529 * bits on second half.
531 * If the extent we found extends past our range, we
532 * just split and search again. It'll get split again
533 * the next time though.
535 * If the extent we found is inside our range, we clear
536 * the desired bit on it.
539 if (state->start < start) {
540 prealloc = alloc_extent_state_atomic(prealloc);
542 err = split_state(tree, state, prealloc, start);
543 BUG_ON(err == -EEXIST);
547 if (state->end <= end) {
548 set |= clear_state_bit(tree, state, &bits, wake);
549 if (last_end == (u64)-1)
551 start = last_end + 1;
556 * | ---- desired range ---- |
558 * We need to split the extent, and clear the bit
561 if (state->start <= end && state->end > end) {
562 prealloc = alloc_extent_state_atomic(prealloc);
564 err = split_state(tree, state, prealloc, end + 1);
565 BUG_ON(err == -EEXIST);
569 set |= clear_state_bit(tree, prealloc, &bits, wake);
575 if (state->end < end && prealloc && !need_resched())
576 next_node = rb_next(&state->rb_node);
580 set |= clear_state_bit(tree, state, &bits, wake);
581 if (last_end == (u64)-1)
583 start = last_end + 1;
584 if (start <= end && next_node) {
585 state = rb_entry(next_node, struct extent_state,
587 if (state->start == start)
593 spin_unlock(&tree->lock);
595 free_extent_state(prealloc);
602 spin_unlock(&tree->lock);
603 if (mask & __GFP_WAIT)
608 static int wait_on_state(struct extent_io_tree *tree,
609 struct extent_state *state)
610 __releases(tree->lock)
611 __acquires(tree->lock)
614 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
615 spin_unlock(&tree->lock);
617 spin_lock(&tree->lock);
618 finish_wait(&state->wq, &wait);
623 * waits for one or more bits to clear on a range in the state tree.
624 * The range [start, end] is inclusive.
625 * The tree lock is taken by this function
627 int wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits)
629 struct extent_state *state;
630 struct rb_node *node;
632 spin_lock(&tree->lock);
636 * this search will find all the extents that end after
639 node = tree_search(tree, start);
643 state = rb_entry(node, struct extent_state, rb_node);
645 if (state->start > end)
648 if (state->state & bits) {
649 start = state->start;
650 atomic_inc(&state->refs);
651 wait_on_state(tree, state);
652 free_extent_state(state);
655 start = state->end + 1;
660 if (need_resched()) {
661 spin_unlock(&tree->lock);
663 spin_lock(&tree->lock);
667 spin_unlock(&tree->lock);
671 static int set_state_bits(struct extent_io_tree *tree,
672 struct extent_state *state,
676 int bits_to_set = *bits & ~EXTENT_CTLBITS;
678 ret = set_state_cb(tree, state, bits);
681 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
682 u64 range = state->end - state->start + 1;
683 tree->dirty_bytes += range;
685 state->state |= bits_to_set;
690 static void cache_state(struct extent_state *state,
691 struct extent_state **cached_ptr)
693 if (cached_ptr && !(*cached_ptr)) {
694 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) {
696 atomic_inc(&state->refs);
701 static void uncache_state(struct extent_state **cached_ptr)
703 if (cached_ptr && (*cached_ptr)) {
704 struct extent_state *state = *cached_ptr;
706 free_extent_state(state);
711 * set some bits on a range in the tree. This may require allocations or
712 * sleeping, so the gfp mask is used to indicate what is allowed.
714 * If any of the exclusive bits are set, this will fail with -EEXIST if some
715 * part of the range already has the desired bits set. The start of the
716 * existing range is returned in failed_start in this case.
718 * [start, end] is inclusive This takes the tree lock.
721 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
722 int bits, int exclusive_bits, u64 *failed_start,
723 struct extent_state **cached_state, gfp_t mask)
725 struct extent_state *state;
726 struct extent_state *prealloc = NULL;
727 struct rb_node *node;
732 bits |= EXTENT_FIRST_DELALLOC;
734 if (!prealloc && (mask & __GFP_WAIT)) {
735 prealloc = alloc_extent_state(mask);
739 spin_lock(&tree->lock);
740 if (cached_state && *cached_state) {
741 state = *cached_state;
742 if (state->start <= start && state->end > start &&
744 node = &state->rb_node;
749 * this search will find all the extents that end after
752 node = tree_search(tree, start);
754 prealloc = alloc_extent_state_atomic(prealloc);
756 err = insert_state(tree, prealloc, start, end, &bits);
758 BUG_ON(err == -EEXIST);
761 state = rb_entry(node, struct extent_state, rb_node);
763 last_start = state->start;
764 last_end = state->end;
767 * | ---- desired range ---- |
770 * Just lock what we found and keep going
772 if (state->start == start && state->end <= end) {
773 struct rb_node *next_node;
774 if (state->state & exclusive_bits) {
775 *failed_start = state->start;
780 err = set_state_bits(tree, state, &bits);
784 cache_state(state, cached_state);
785 merge_state(tree, state);
786 if (last_end == (u64)-1)
789 start = last_end + 1;
790 next_node = rb_next(&state->rb_node);
791 if (next_node && start < end && prealloc && !need_resched()) {
792 state = rb_entry(next_node, struct extent_state,
794 if (state->start == start)
801 * | ---- desired range ---- |
804 * | ------------- state -------------- |
806 * We need to split the extent we found, and may flip bits on
809 * If the extent we found extends past our
810 * range, we just split and search again. It'll get split
811 * again the next time though.
813 * If the extent we found is inside our range, we set the
816 if (state->start < start) {
817 if (state->state & exclusive_bits) {
818 *failed_start = start;
823 prealloc = alloc_extent_state_atomic(prealloc);
825 err = split_state(tree, state, prealloc, start);
826 BUG_ON(err == -EEXIST);
830 if (state->end <= end) {
831 err = set_state_bits(tree, state, &bits);
834 cache_state(state, cached_state);
835 merge_state(tree, state);
836 if (last_end == (u64)-1)
838 start = last_end + 1;
843 * | ---- desired range ---- |
844 * | state | or | state |
846 * There's a hole, we need to insert something in it and
847 * ignore the extent we found.
849 if (state->start > start) {
851 if (end < last_start)
854 this_end = last_start - 1;
856 prealloc = alloc_extent_state_atomic(prealloc);
860 * Avoid to free 'prealloc' if it can be merged with
863 err = insert_state(tree, prealloc, start, this_end,
865 BUG_ON(err == -EEXIST);
867 free_extent_state(prealloc);
871 cache_state(prealloc, cached_state);
873 start = this_end + 1;
877 * | ---- desired range ---- |
879 * We need to split the extent, and set the bit
882 if (state->start <= end && state->end > end) {
883 if (state->state & exclusive_bits) {
884 *failed_start = start;
889 prealloc = alloc_extent_state_atomic(prealloc);
891 err = split_state(tree, state, prealloc, end + 1);
892 BUG_ON(err == -EEXIST);
894 err = set_state_bits(tree, prealloc, &bits);
899 cache_state(prealloc, cached_state);
900 merge_state(tree, prealloc);
908 spin_unlock(&tree->lock);
910 free_extent_state(prealloc);
917 spin_unlock(&tree->lock);
918 if (mask & __GFP_WAIT)
923 /* wrappers around set/clear extent bit */
924 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
927 return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL,
931 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
932 int bits, gfp_t mask)
934 return set_extent_bit(tree, start, end, bits, 0, NULL,
938 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
939 int bits, gfp_t mask)
941 return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask);
944 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
945 struct extent_state **cached_state, gfp_t mask)
947 return set_extent_bit(tree, start, end,
948 EXTENT_DELALLOC | EXTENT_DIRTY | EXTENT_UPTODATE,
949 0, NULL, cached_state, mask);
952 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
955 return clear_extent_bit(tree, start, end,
956 EXTENT_DIRTY | EXTENT_DELALLOC |
957 EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask);
960 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
963 return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL,
967 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
968 struct extent_state **cached_state, gfp_t mask)
970 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0,
971 NULL, cached_state, mask);
974 static int clear_extent_uptodate(struct extent_io_tree *tree, u64 start,
975 u64 end, struct extent_state **cached_state,
978 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0,
983 * either insert or lock state struct between start and end use mask to tell
984 * us if waiting is desired.
986 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
987 int bits, struct extent_state **cached_state, gfp_t mask)
992 err = set_extent_bit(tree, start, end, EXTENT_LOCKED | bits,
993 EXTENT_LOCKED, &failed_start,
995 if (err == -EEXIST && (mask & __GFP_WAIT)) {
996 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
997 start = failed_start;
1001 WARN_ON(start > end);
1006 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
1008 return lock_extent_bits(tree, start, end, 0, NULL, mask);
1011 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end,
1017 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1018 &failed_start, NULL, mask);
1019 if (err == -EEXIST) {
1020 if (failed_start > start)
1021 clear_extent_bit(tree, start, failed_start - 1,
1022 EXTENT_LOCKED, 1, 0, NULL, mask);
1028 int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end,
1029 struct extent_state **cached, gfp_t mask)
1031 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached,
1035 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
1037 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL,
1042 * helper function to set both pages and extents in the tree writeback
1044 static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
1046 unsigned long index = start >> PAGE_CACHE_SHIFT;
1047 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1050 while (index <= end_index) {
1051 page = find_get_page(tree->mapping, index);
1053 set_page_writeback(page);
1054 page_cache_release(page);
1061 * find the first offset in the io tree with 'bits' set. zero is
1062 * returned if we find something, and *start_ret and *end_ret are
1063 * set to reflect the state struct that was found.
1065 * If nothing was found, 1 is returned, < 0 on error
1067 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1068 u64 *start_ret, u64 *end_ret, int bits)
1070 struct rb_node *node;
1071 struct extent_state *state;
1074 spin_lock(&tree->lock);
1076 * this search will find all the extents that end after
1079 node = tree_search(tree, start);
1084 state = rb_entry(node, struct extent_state, rb_node);
1085 if (state->end >= start && (state->state & bits)) {
1086 *start_ret = state->start;
1087 *end_ret = state->end;
1091 node = rb_next(node);
1096 spin_unlock(&tree->lock);
1100 /* find the first state struct with 'bits' set after 'start', and
1101 * return it. tree->lock must be held. NULL will returned if
1102 * nothing was found after 'start'
1104 struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree,
1105 u64 start, int bits)
1107 struct rb_node *node;
1108 struct extent_state *state;
1111 * this search will find all the extents that end after
1114 node = tree_search(tree, start);
1119 state = rb_entry(node, struct extent_state, rb_node);
1120 if (state->end >= start && (state->state & bits))
1123 node = rb_next(node);
1132 * find a contiguous range of bytes in the file marked as delalloc, not
1133 * more than 'max_bytes'. start and end are used to return the range,
1135 * 1 is returned if we find something, 0 if nothing was in the tree
1137 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1138 u64 *start, u64 *end, u64 max_bytes,
1139 struct extent_state **cached_state)
1141 struct rb_node *node;
1142 struct extent_state *state;
1143 u64 cur_start = *start;
1145 u64 total_bytes = 0;
1147 spin_lock(&tree->lock);
1150 * this search will find all the extents that end after
1153 node = tree_search(tree, cur_start);
1161 state = rb_entry(node, struct extent_state, rb_node);
1162 if (found && (state->start != cur_start ||
1163 (state->state & EXTENT_BOUNDARY))) {
1166 if (!(state->state & EXTENT_DELALLOC)) {
1172 *start = state->start;
1173 *cached_state = state;
1174 atomic_inc(&state->refs);
1178 cur_start = state->end + 1;
1179 node = rb_next(node);
1182 total_bytes += state->end - state->start + 1;
1183 if (total_bytes >= max_bytes)
1187 spin_unlock(&tree->lock);
1191 static noinline int __unlock_for_delalloc(struct inode *inode,
1192 struct page *locked_page,
1196 struct page *pages[16];
1197 unsigned long index = start >> PAGE_CACHE_SHIFT;
1198 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1199 unsigned long nr_pages = end_index - index + 1;
1202 if (index == locked_page->index && end_index == index)
1205 while (nr_pages > 0) {
1206 ret = find_get_pages_contig(inode->i_mapping, index,
1207 min_t(unsigned long, nr_pages,
1208 ARRAY_SIZE(pages)), pages);
1209 for (i = 0; i < ret; i++) {
1210 if (pages[i] != locked_page)
1211 unlock_page(pages[i]);
1212 page_cache_release(pages[i]);
1221 static noinline int lock_delalloc_pages(struct inode *inode,
1222 struct page *locked_page,
1226 unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT;
1227 unsigned long start_index = index;
1228 unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT;
1229 unsigned long pages_locked = 0;
1230 struct page *pages[16];
1231 unsigned long nrpages;
1235 /* the caller is responsible for locking the start index */
1236 if (index == locked_page->index && index == end_index)
1239 /* skip the page at the start index */
1240 nrpages = end_index - index + 1;
1241 while (nrpages > 0) {
1242 ret = find_get_pages_contig(inode->i_mapping, index,
1243 min_t(unsigned long,
1244 nrpages, ARRAY_SIZE(pages)), pages);
1249 /* now we have an array of pages, lock them all */
1250 for (i = 0; i < ret; i++) {
1252 * the caller is taking responsibility for
1255 if (pages[i] != locked_page) {
1256 lock_page(pages[i]);
1257 if (!PageDirty(pages[i]) ||
1258 pages[i]->mapping != inode->i_mapping) {
1260 unlock_page(pages[i]);
1261 page_cache_release(pages[i]);
1265 page_cache_release(pages[i]);
1274 if (ret && pages_locked) {
1275 __unlock_for_delalloc(inode, locked_page,
1277 ((u64)(start_index + pages_locked - 1)) <<
1284 * find a contiguous range of bytes in the file marked as delalloc, not
1285 * more than 'max_bytes'. start and end are used to return the range,
1287 * 1 is returned if we find something, 0 if nothing was in the tree
1289 static noinline u64 find_lock_delalloc_range(struct inode *inode,
1290 struct extent_io_tree *tree,
1291 struct page *locked_page,
1292 u64 *start, u64 *end,
1298 struct extent_state *cached_state = NULL;
1303 /* step one, find a bunch of delalloc bytes starting at start */
1304 delalloc_start = *start;
1306 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1307 max_bytes, &cached_state);
1308 if (!found || delalloc_end <= *start) {
1309 *start = delalloc_start;
1310 *end = delalloc_end;
1311 free_extent_state(cached_state);
1316 * start comes from the offset of locked_page. We have to lock
1317 * pages in order, so we can't process delalloc bytes before
1320 if (delalloc_start < *start)
1321 delalloc_start = *start;
1324 * make sure to limit the number of pages we try to lock down
1327 if (delalloc_end + 1 - delalloc_start > max_bytes && loops)
1328 delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1;
1330 /* step two, lock all the pages after the page that has start */
1331 ret = lock_delalloc_pages(inode, locked_page,
1332 delalloc_start, delalloc_end);
1333 if (ret == -EAGAIN) {
1334 /* some of the pages are gone, lets avoid looping by
1335 * shortening the size of the delalloc range we're searching
1337 free_extent_state(cached_state);
1339 unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1);
1340 max_bytes = PAGE_CACHE_SIZE - offset;
1350 /* step three, lock the state bits for the whole range */
1351 lock_extent_bits(tree, delalloc_start, delalloc_end,
1352 0, &cached_state, GFP_NOFS);
1354 /* then test to make sure it is all still delalloc */
1355 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1356 EXTENT_DELALLOC, 1, cached_state);
1358 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1359 &cached_state, GFP_NOFS);
1360 __unlock_for_delalloc(inode, locked_page,
1361 delalloc_start, delalloc_end);
1365 free_extent_state(cached_state);
1366 *start = delalloc_start;
1367 *end = delalloc_end;
1372 int extent_clear_unlock_delalloc(struct inode *inode,
1373 struct extent_io_tree *tree,
1374 u64 start, u64 end, struct page *locked_page,
1378 struct page *pages[16];
1379 unsigned long index = start >> PAGE_CACHE_SHIFT;
1380 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1381 unsigned long nr_pages = end_index - index + 1;
1385 if (op & EXTENT_CLEAR_UNLOCK)
1386 clear_bits |= EXTENT_LOCKED;
1387 if (op & EXTENT_CLEAR_DIRTY)
1388 clear_bits |= EXTENT_DIRTY;
1390 if (op & EXTENT_CLEAR_DELALLOC)
1391 clear_bits |= EXTENT_DELALLOC;
1393 clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS);
1394 if (!(op & (EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
1395 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK |
1396 EXTENT_SET_PRIVATE2)))
1399 while (nr_pages > 0) {
1400 ret = find_get_pages_contig(inode->i_mapping, index,
1401 min_t(unsigned long,
1402 nr_pages, ARRAY_SIZE(pages)), pages);
1403 for (i = 0; i < ret; i++) {
1405 if (op & EXTENT_SET_PRIVATE2)
1406 SetPagePrivate2(pages[i]);
1408 if (pages[i] == locked_page) {
1409 page_cache_release(pages[i]);
1412 if (op & EXTENT_CLEAR_DIRTY)
1413 clear_page_dirty_for_io(pages[i]);
1414 if (op & EXTENT_SET_WRITEBACK)
1415 set_page_writeback(pages[i]);
1416 if (op & EXTENT_END_WRITEBACK)
1417 end_page_writeback(pages[i]);
1418 if (op & EXTENT_CLEAR_UNLOCK_PAGE)
1419 unlock_page(pages[i]);
1420 page_cache_release(pages[i]);
1430 * count the number of bytes in the tree that have a given bit(s)
1431 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1432 * cached. The total number found is returned.
1434 u64 count_range_bits(struct extent_io_tree *tree,
1435 u64 *start, u64 search_end, u64 max_bytes,
1436 unsigned long bits, int contig)
1438 struct rb_node *node;
1439 struct extent_state *state;
1440 u64 cur_start = *start;
1441 u64 total_bytes = 0;
1445 if (search_end <= cur_start) {
1450 spin_lock(&tree->lock);
1451 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1452 total_bytes = tree->dirty_bytes;
1456 * this search will find all the extents that end after
1459 node = tree_search(tree, cur_start);
1464 state = rb_entry(node, struct extent_state, rb_node);
1465 if (state->start > search_end)
1467 if (contig && found && state->start > last + 1)
1469 if (state->end >= cur_start && (state->state & bits) == bits) {
1470 total_bytes += min(search_end, state->end) + 1 -
1471 max(cur_start, state->start);
1472 if (total_bytes >= max_bytes)
1475 *start = max(cur_start, state->start);
1479 } else if (contig && found) {
1482 node = rb_next(node);
1487 spin_unlock(&tree->lock);
1492 * set the private field for a given byte offset in the tree. If there isn't
1493 * an extent_state there already, this does nothing.
1495 int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1497 struct rb_node *node;
1498 struct extent_state *state;
1501 spin_lock(&tree->lock);
1503 * this search will find all the extents that end after
1506 node = tree_search(tree, start);
1511 state = rb_entry(node, struct extent_state, rb_node);
1512 if (state->start != start) {
1516 state->private = private;
1518 spin_unlock(&tree->lock);
1522 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1524 struct rb_node *node;
1525 struct extent_state *state;
1528 spin_lock(&tree->lock);
1530 * this search will find all the extents that end after
1533 node = tree_search(tree, start);
1538 state = rb_entry(node, struct extent_state, rb_node);
1539 if (state->start != start) {
1543 *private = state->private;
1545 spin_unlock(&tree->lock);
1550 * searches a range in the state tree for a given mask.
1551 * If 'filled' == 1, this returns 1 only if every extent in the tree
1552 * has the bits set. Otherwise, 1 is returned if any bit in the
1553 * range is found set.
1555 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1556 int bits, int filled, struct extent_state *cached)
1558 struct extent_state *state = NULL;
1559 struct rb_node *node;
1562 spin_lock(&tree->lock);
1563 if (cached && cached->tree && cached->start <= start &&
1564 cached->end > start)
1565 node = &cached->rb_node;
1567 node = tree_search(tree, start);
1568 while (node && start <= end) {
1569 state = rb_entry(node, struct extent_state, rb_node);
1571 if (filled && state->start > start) {
1576 if (state->start > end)
1579 if (state->state & bits) {
1583 } else if (filled) {
1588 if (state->end == (u64)-1)
1591 start = state->end + 1;
1594 node = rb_next(node);
1601 spin_unlock(&tree->lock);
1606 * helper function to set a given page up to date if all the
1607 * extents in the tree for that page are up to date
1609 static int check_page_uptodate(struct extent_io_tree *tree,
1612 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1613 u64 end = start + PAGE_CACHE_SIZE - 1;
1614 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1615 SetPageUptodate(page);
1620 * helper function to unlock a page if all the extents in the tree
1621 * for that page are unlocked
1623 static int check_page_locked(struct extent_io_tree *tree,
1626 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1627 u64 end = start + PAGE_CACHE_SIZE - 1;
1628 if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL))
1634 * helper function to end page writeback if all the extents
1635 * in the tree for that page are done with writeback
1637 static int check_page_writeback(struct extent_io_tree *tree,
1640 end_page_writeback(page);
1644 /* lots and lots of room for performance fixes in the end_bio funcs */
1647 * after a writepage IO is done, we need to:
1648 * clear the uptodate bits on error
1649 * clear the writeback bits in the extent tree for this IO
1650 * end_page_writeback if the page has no more pending IO
1652 * Scheduling is not allowed, so the extent state tree is expected
1653 * to have one and only one object corresponding to this IO.
1655 static void end_bio_extent_writepage(struct bio *bio, int err)
1657 int uptodate = err == 0;
1658 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1659 struct extent_io_tree *tree;
1666 struct page *page = bvec->bv_page;
1667 tree = &BTRFS_I(page->mapping->host)->io_tree;
1669 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1671 end = start + bvec->bv_len - 1;
1673 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1678 if (--bvec >= bio->bi_io_vec)
1679 prefetchw(&bvec->bv_page->flags);
1680 if (tree->ops && tree->ops->writepage_end_io_hook) {
1681 ret = tree->ops->writepage_end_io_hook(page, start,
1682 end, NULL, uptodate);
1687 if (!uptodate && tree->ops &&
1688 tree->ops->writepage_io_failed_hook) {
1689 ret = tree->ops->writepage_io_failed_hook(bio, page,
1692 uptodate = (err == 0);
1698 clear_extent_uptodate(tree, start, end, NULL, GFP_NOFS);
1699 ClearPageUptodate(page);
1704 end_page_writeback(page);
1706 check_page_writeback(tree, page);
1707 } while (bvec >= bio->bi_io_vec);
1713 * after a readpage IO is done, we need to:
1714 * clear the uptodate bits on error
1715 * set the uptodate bits if things worked
1716 * set the page up to date if all extents in the tree are uptodate
1717 * clear the lock bit in the extent tree
1718 * unlock the page if there are no other extents locked for it
1720 * Scheduling is not allowed, so the extent state tree is expected
1721 * to have one and only one object corresponding to this IO.
1723 static void end_bio_extent_readpage(struct bio *bio, int err)
1725 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1726 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
1727 struct bio_vec *bvec = bio->bi_io_vec;
1728 struct extent_io_tree *tree;
1738 struct page *page = bvec->bv_page;
1739 struct extent_state *cached = NULL;
1740 struct extent_state *state;
1742 tree = &BTRFS_I(page->mapping->host)->io_tree;
1744 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1746 end = start + bvec->bv_len - 1;
1748 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1753 if (++bvec <= bvec_end)
1754 prefetchw(&bvec->bv_page->flags);
1756 spin_lock(&tree->lock);
1757 state = find_first_extent_bit_state(tree, start, EXTENT_LOCKED);
1758 if (state && state->start == start) {
1760 * take a reference on the state, unlock will drop
1763 cache_state(state, &cached);
1765 spin_unlock(&tree->lock);
1767 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
1768 ret = tree->ops->readpage_end_io_hook(page, start, end,
1773 if (!uptodate && tree->ops &&
1774 tree->ops->readpage_io_failed_hook) {
1775 ret = tree->ops->readpage_io_failed_hook(bio, page,
1779 test_bit(BIO_UPTODATE, &bio->bi_flags);
1782 uncache_state(&cached);
1788 set_extent_uptodate(tree, start, end, &cached,
1791 unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC);
1795 SetPageUptodate(page);
1797 ClearPageUptodate(page);
1803 check_page_uptodate(tree, page);
1805 ClearPageUptodate(page);
1808 check_page_locked(tree, page);
1810 } while (bvec <= bvec_end);
1816 btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
1821 bio = bio_alloc(gfp_flags, nr_vecs);
1823 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
1824 while (!bio && (nr_vecs /= 2))
1825 bio = bio_alloc(gfp_flags, nr_vecs);
1830 bio->bi_bdev = bdev;
1831 bio->bi_sector = first_sector;
1836 static int submit_one_bio(int rw, struct bio *bio, int mirror_num,
1837 unsigned long bio_flags)
1840 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1841 struct page *page = bvec->bv_page;
1842 struct extent_io_tree *tree = bio->bi_private;
1845 start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
1847 bio->bi_private = NULL;
1851 if (tree->ops && tree->ops->submit_bio_hook)
1852 ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
1853 mirror_num, bio_flags, start);
1855 submit_bio(rw, bio);
1856 if (bio_flagged(bio, BIO_EOPNOTSUPP))
1862 static int submit_extent_page(int rw, struct extent_io_tree *tree,
1863 struct page *page, sector_t sector,
1864 size_t size, unsigned long offset,
1865 struct block_device *bdev,
1866 struct bio **bio_ret,
1867 unsigned long max_pages,
1868 bio_end_io_t end_io_func,
1870 unsigned long prev_bio_flags,
1871 unsigned long bio_flags)
1877 int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED;
1878 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
1879 size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE);
1881 if (bio_ret && *bio_ret) {
1884 contig = bio->bi_sector == sector;
1886 contig = bio->bi_sector + (bio->bi_size >> 9) ==
1889 if (prev_bio_flags != bio_flags || !contig ||
1890 (tree->ops && tree->ops->merge_bio_hook &&
1891 tree->ops->merge_bio_hook(page, offset, page_size, bio,
1893 bio_add_page(bio, page, page_size, offset) < page_size) {
1894 ret = submit_one_bio(rw, bio, mirror_num,
1901 if (this_compressed)
1904 nr = bio_get_nr_vecs(bdev);
1906 bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
1910 bio_add_page(bio, page, page_size, offset);
1911 bio->bi_end_io = end_io_func;
1912 bio->bi_private = tree;
1917 ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
1922 void set_page_extent_mapped(struct page *page)
1924 if (!PagePrivate(page)) {
1925 SetPagePrivate(page);
1926 page_cache_get(page);
1927 set_page_private(page, EXTENT_PAGE_PRIVATE);
1931 static void set_page_extent_head(struct page *page, unsigned long len)
1933 WARN_ON(!PagePrivate(page));
1934 set_page_private(page, EXTENT_PAGE_PRIVATE_FIRST_PAGE | len << 2);
1938 * basic readpage implementation. Locked extent state structs are inserted
1939 * into the tree that are removed when the IO is done (by the end_io
1942 static int __extent_read_full_page(struct extent_io_tree *tree,
1944 get_extent_t *get_extent,
1945 struct bio **bio, int mirror_num,
1946 unsigned long *bio_flags)
1948 struct inode *inode = page->mapping->host;
1949 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1950 u64 page_end = start + PAGE_CACHE_SIZE - 1;
1954 u64 last_byte = i_size_read(inode);
1958 struct extent_map *em;
1959 struct block_device *bdev;
1960 struct btrfs_ordered_extent *ordered;
1963 size_t pg_offset = 0;
1965 size_t disk_io_size;
1966 size_t blocksize = inode->i_sb->s_blocksize;
1967 unsigned long this_bio_flag = 0;
1969 set_page_extent_mapped(page);
1973 lock_extent(tree, start, end, GFP_NOFS);
1974 ordered = btrfs_lookup_ordered_extent(inode, start);
1977 unlock_extent(tree, start, end, GFP_NOFS);
1978 btrfs_start_ordered_extent(inode, ordered, 1);
1979 btrfs_put_ordered_extent(ordered);
1982 if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
1984 size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
1987 iosize = PAGE_CACHE_SIZE - zero_offset;
1988 userpage = kmap_atomic(page, KM_USER0);
1989 memset(userpage + zero_offset, 0, iosize);
1990 flush_dcache_page(page);
1991 kunmap_atomic(userpage, KM_USER0);
1994 while (cur <= end) {
1995 if (cur >= last_byte) {
1997 struct extent_state *cached = NULL;
1999 iosize = PAGE_CACHE_SIZE - pg_offset;
2000 userpage = kmap_atomic(page, KM_USER0);
2001 memset(userpage + pg_offset, 0, iosize);
2002 flush_dcache_page(page);
2003 kunmap_atomic(userpage, KM_USER0);
2004 set_extent_uptodate(tree, cur, cur + iosize - 1,
2006 unlock_extent_cached(tree, cur, cur + iosize - 1,
2010 em = get_extent(inode, page, pg_offset, cur,
2012 if (IS_ERR_OR_NULL(em)) {
2014 unlock_extent(tree, cur, end, GFP_NOFS);
2017 extent_offset = cur - em->start;
2018 BUG_ON(extent_map_end(em) <= cur);
2021 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2022 this_bio_flag = EXTENT_BIO_COMPRESSED;
2023 extent_set_compress_type(&this_bio_flag,
2027 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2028 cur_end = min(extent_map_end(em) - 1, end);
2029 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2030 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2031 disk_io_size = em->block_len;
2032 sector = em->block_start >> 9;
2034 sector = (em->block_start + extent_offset) >> 9;
2035 disk_io_size = iosize;
2038 block_start = em->block_start;
2039 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2040 block_start = EXTENT_MAP_HOLE;
2041 free_extent_map(em);
2044 /* we've found a hole, just zero and go on */
2045 if (block_start == EXTENT_MAP_HOLE) {
2047 struct extent_state *cached = NULL;
2049 userpage = kmap_atomic(page, KM_USER0);
2050 memset(userpage + pg_offset, 0, iosize);
2051 flush_dcache_page(page);
2052 kunmap_atomic(userpage, KM_USER0);
2054 set_extent_uptodate(tree, cur, cur + iosize - 1,
2056 unlock_extent_cached(tree, cur, cur + iosize - 1,
2059 pg_offset += iosize;
2062 /* the get_extent function already copied into the page */
2063 if (test_range_bit(tree, cur, cur_end,
2064 EXTENT_UPTODATE, 1, NULL)) {
2065 check_page_uptodate(tree, page);
2066 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
2068 pg_offset += iosize;
2071 /* we have an inline extent but it didn't get marked up
2072 * to date. Error out
2074 if (block_start == EXTENT_MAP_INLINE) {
2076 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
2078 pg_offset += iosize;
2083 if (tree->ops && tree->ops->readpage_io_hook) {
2084 ret = tree->ops->readpage_io_hook(page, cur,
2088 unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
2090 ret = submit_extent_page(READ, tree, page,
2091 sector, disk_io_size, pg_offset,
2093 end_bio_extent_readpage, mirror_num,
2097 *bio_flags = this_bio_flag;
2102 pg_offset += iosize;
2105 if (!PageError(page))
2106 SetPageUptodate(page);
2112 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
2113 get_extent_t *get_extent)
2115 struct bio *bio = NULL;
2116 unsigned long bio_flags = 0;
2119 ret = __extent_read_full_page(tree, page, get_extent, &bio, 0,
2122 ret = submit_one_bio(READ, bio, 0, bio_flags);
2126 static noinline void update_nr_written(struct page *page,
2127 struct writeback_control *wbc,
2128 unsigned long nr_written)
2130 wbc->nr_to_write -= nr_written;
2131 if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
2132 wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
2133 page->mapping->writeback_index = page->index + nr_written;
2137 * the writepage semantics are similar to regular writepage. extent
2138 * records are inserted to lock ranges in the tree, and as dirty areas
2139 * are found, they are marked writeback. Then the lock bits are removed
2140 * and the end_io handler clears the writeback ranges
2142 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2145 struct inode *inode = page->mapping->host;
2146 struct extent_page_data *epd = data;
2147 struct extent_io_tree *tree = epd->tree;
2148 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2150 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2154 u64 last_byte = i_size_read(inode);
2158 struct extent_state *cached_state = NULL;
2159 struct extent_map *em;
2160 struct block_device *bdev;
2163 size_t pg_offset = 0;
2165 loff_t i_size = i_size_read(inode);
2166 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
2172 unsigned long nr_written = 0;
2174 if (wbc->sync_mode == WB_SYNC_ALL)
2175 write_flags = WRITE_SYNC;
2177 write_flags = WRITE;
2179 trace___extent_writepage(page, inode, wbc);
2181 WARN_ON(!PageLocked(page));
2182 pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
2183 if (page->index > end_index ||
2184 (page->index == end_index && !pg_offset)) {
2185 page->mapping->a_ops->invalidatepage(page, 0);
2190 if (page->index == end_index) {
2193 userpage = kmap_atomic(page, KM_USER0);
2194 memset(userpage + pg_offset, 0,
2195 PAGE_CACHE_SIZE - pg_offset);
2196 kunmap_atomic(userpage, KM_USER0);
2197 flush_dcache_page(page);
2201 set_page_extent_mapped(page);
2203 delalloc_start = start;
2206 if (!epd->extent_locked) {
2207 u64 delalloc_to_write = 0;
2209 * make sure the wbc mapping index is at least updated
2212 update_nr_written(page, wbc, 0);
2214 while (delalloc_end < page_end) {
2215 nr_delalloc = find_lock_delalloc_range(inode, tree,
2220 if (nr_delalloc == 0) {
2221 delalloc_start = delalloc_end + 1;
2224 tree->ops->fill_delalloc(inode, page, delalloc_start,
2225 delalloc_end, &page_started,
2228 * delalloc_end is already one less than the total
2229 * length, so we don't subtract one from
2232 delalloc_to_write += (delalloc_end - delalloc_start +
2235 delalloc_start = delalloc_end + 1;
2237 if (wbc->nr_to_write < delalloc_to_write) {
2240 if (delalloc_to_write < thresh * 2)
2241 thresh = delalloc_to_write;
2242 wbc->nr_to_write = min_t(u64, delalloc_to_write,
2246 /* did the fill delalloc function already unlock and start
2252 * we've unlocked the page, so we can't update
2253 * the mapping's writeback index, just update
2256 wbc->nr_to_write -= nr_written;
2260 if (tree->ops && tree->ops->writepage_start_hook) {
2261 ret = tree->ops->writepage_start_hook(page, start,
2263 if (ret == -EAGAIN) {
2264 redirty_page_for_writepage(wbc, page);
2265 update_nr_written(page, wbc, nr_written);
2273 * we don't want to touch the inode after unlocking the page,
2274 * so we update the mapping writeback index now
2276 update_nr_written(page, wbc, nr_written + 1);
2279 if (last_byte <= start) {
2280 if (tree->ops && tree->ops->writepage_end_io_hook)
2281 tree->ops->writepage_end_io_hook(page, start,
2286 blocksize = inode->i_sb->s_blocksize;
2288 while (cur <= end) {
2289 if (cur >= last_byte) {
2290 if (tree->ops && tree->ops->writepage_end_io_hook)
2291 tree->ops->writepage_end_io_hook(page, cur,
2295 em = epd->get_extent(inode, page, pg_offset, cur,
2297 if (IS_ERR_OR_NULL(em)) {
2302 extent_offset = cur - em->start;
2303 BUG_ON(extent_map_end(em) <= cur);
2305 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2306 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2307 sector = (em->block_start + extent_offset) >> 9;
2309 block_start = em->block_start;
2310 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
2311 free_extent_map(em);
2315 * compressed and inline extents are written through other
2318 if (compressed || block_start == EXTENT_MAP_HOLE ||
2319 block_start == EXTENT_MAP_INLINE) {
2321 * end_io notification does not happen here for
2322 * compressed extents
2324 if (!compressed && tree->ops &&
2325 tree->ops->writepage_end_io_hook)
2326 tree->ops->writepage_end_io_hook(page, cur,
2329 else if (compressed) {
2330 /* we don't want to end_page_writeback on
2331 * a compressed extent. this happens
2338 pg_offset += iosize;
2341 /* leave this out until we have a page_mkwrite call */
2342 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
2343 EXTENT_DIRTY, 0, NULL)) {
2345 pg_offset += iosize;
2349 if (tree->ops && tree->ops->writepage_io_hook) {
2350 ret = tree->ops->writepage_io_hook(page, cur,
2358 unsigned long max_nr = end_index + 1;
2360 set_range_writeback(tree, cur, cur + iosize - 1);
2361 if (!PageWriteback(page)) {
2362 printk(KERN_ERR "btrfs warning page %lu not "
2363 "writeback, cur %llu end %llu\n",
2364 page->index, (unsigned long long)cur,
2365 (unsigned long long)end);
2368 ret = submit_extent_page(write_flags, tree, page,
2369 sector, iosize, pg_offset,
2370 bdev, &epd->bio, max_nr,
2371 end_bio_extent_writepage,
2377 pg_offset += iosize;
2382 /* make sure the mapping tag for page dirty gets cleared */
2383 set_page_writeback(page);
2384 end_page_writeback(page);
2390 /* drop our reference on any cached states */
2391 free_extent_state(cached_state);
2396 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
2397 * @mapping: address space structure to write
2398 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2399 * @writepage: function called for each page
2400 * @data: data passed to writepage function
2402 * If a page is already under I/O, write_cache_pages() skips it, even
2403 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
2404 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
2405 * and msync() need to guarantee that all the data which was dirty at the time
2406 * the call was made get new I/O started against them. If wbc->sync_mode is
2407 * WB_SYNC_ALL then we were called for data integrity and we must wait for
2408 * existing IO to complete.
2410 static int extent_write_cache_pages(struct extent_io_tree *tree,
2411 struct address_space *mapping,
2412 struct writeback_control *wbc,
2413 writepage_t writepage, void *data,
2414 void (*flush_fn)(void *))
2418 int nr_to_write_done = 0;
2419 struct pagevec pvec;
2422 pgoff_t end; /* Inclusive */
2426 pagevec_init(&pvec, 0);
2427 if (wbc->range_cyclic) {
2428 index = mapping->writeback_index; /* Start from prev offset */
2431 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2432 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2435 if (wbc->sync_mode == WB_SYNC_ALL)
2436 tag = PAGECACHE_TAG_TOWRITE;
2438 tag = PAGECACHE_TAG_DIRTY;
2440 if (wbc->sync_mode == WB_SYNC_ALL)
2441 tag_pages_for_writeback(mapping, index, end);
2442 while (!done && !nr_to_write_done && (index <= end) &&
2443 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2444 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
2448 for (i = 0; i < nr_pages; i++) {
2449 struct page *page = pvec.pages[i];
2452 * At this point we hold neither mapping->tree_lock nor
2453 * lock on the page itself: the page may be truncated or
2454 * invalidated (changing page->mapping to NULL), or even
2455 * swizzled back from swapper_space to tmpfs file
2458 if (tree->ops && tree->ops->write_cache_pages_lock_hook)
2459 tree->ops->write_cache_pages_lock_hook(page);
2463 if (unlikely(page->mapping != mapping)) {
2468 if (!wbc->range_cyclic && page->index > end) {
2474 if (wbc->sync_mode != WB_SYNC_NONE) {
2475 if (PageWriteback(page))
2477 wait_on_page_writeback(page);
2480 if (PageWriteback(page) ||
2481 !clear_page_dirty_for_io(page)) {
2486 ret = (*writepage)(page, wbc, data);
2488 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
2496 * the filesystem may choose to bump up nr_to_write.
2497 * We have to make sure to honor the new nr_to_write
2500 nr_to_write_done = wbc->nr_to_write <= 0;
2502 pagevec_release(&pvec);
2505 if (!scanned && !done) {
2507 * We hit the last page and there is more work to be done: wrap
2508 * back to the start of the file
2517 static void flush_epd_write_bio(struct extent_page_data *epd)
2521 submit_one_bio(WRITE_SYNC, epd->bio, 0, 0);
2523 submit_one_bio(WRITE, epd->bio, 0, 0);
2528 static noinline void flush_write_bio(void *data)
2530 struct extent_page_data *epd = data;
2531 flush_epd_write_bio(epd);
2534 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
2535 get_extent_t *get_extent,
2536 struct writeback_control *wbc)
2539 struct address_space *mapping = page->mapping;
2540 struct extent_page_data epd = {
2543 .get_extent = get_extent,
2545 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
2547 struct writeback_control wbc_writepages = {
2548 .sync_mode = wbc->sync_mode,
2549 .older_than_this = NULL,
2551 .range_start = page_offset(page) + PAGE_CACHE_SIZE,
2552 .range_end = (loff_t)-1,
2555 ret = __extent_writepage(page, wbc, &epd);
2557 extent_write_cache_pages(tree, mapping, &wbc_writepages,
2558 __extent_writepage, &epd, flush_write_bio);
2559 flush_epd_write_bio(&epd);
2563 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
2564 u64 start, u64 end, get_extent_t *get_extent,
2568 struct address_space *mapping = inode->i_mapping;
2570 unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
2573 struct extent_page_data epd = {
2576 .get_extent = get_extent,
2578 .sync_io = mode == WB_SYNC_ALL,
2580 struct writeback_control wbc_writepages = {
2582 .older_than_this = NULL,
2583 .nr_to_write = nr_pages * 2,
2584 .range_start = start,
2585 .range_end = end + 1,
2588 while (start <= end) {
2589 page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
2590 if (clear_page_dirty_for_io(page))
2591 ret = __extent_writepage(page, &wbc_writepages, &epd);
2593 if (tree->ops && tree->ops->writepage_end_io_hook)
2594 tree->ops->writepage_end_io_hook(page, start,
2595 start + PAGE_CACHE_SIZE - 1,
2599 page_cache_release(page);
2600 start += PAGE_CACHE_SIZE;
2603 flush_epd_write_bio(&epd);
2607 int extent_writepages(struct extent_io_tree *tree,
2608 struct address_space *mapping,
2609 get_extent_t *get_extent,
2610 struct writeback_control *wbc)
2613 struct extent_page_data epd = {
2616 .get_extent = get_extent,
2618 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
2621 ret = extent_write_cache_pages(tree, mapping, wbc,
2622 __extent_writepage, &epd,
2624 flush_epd_write_bio(&epd);
2628 int extent_readpages(struct extent_io_tree *tree,
2629 struct address_space *mapping,
2630 struct list_head *pages, unsigned nr_pages,
2631 get_extent_t get_extent)
2633 struct bio *bio = NULL;
2635 unsigned long bio_flags = 0;
2637 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
2638 struct page *page = list_entry(pages->prev, struct page, lru);
2640 prefetchw(&page->flags);
2641 list_del(&page->lru);
2642 if (!add_to_page_cache_lru(page, mapping,
2643 page->index, GFP_NOFS)) {
2644 __extent_read_full_page(tree, page, get_extent,
2645 &bio, 0, &bio_flags);
2647 page_cache_release(page);
2649 BUG_ON(!list_empty(pages));
2651 submit_one_bio(READ, bio, 0, bio_flags);
2656 * basic invalidatepage code, this waits on any locked or writeback
2657 * ranges corresponding to the page, and then deletes any extent state
2658 * records from the tree
2660 int extent_invalidatepage(struct extent_io_tree *tree,
2661 struct page *page, unsigned long offset)
2663 struct extent_state *cached_state = NULL;
2664 u64 start = ((u64)page->index << PAGE_CACHE_SHIFT);
2665 u64 end = start + PAGE_CACHE_SIZE - 1;
2666 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
2668 start += (offset + blocksize - 1) & ~(blocksize - 1);
2672 lock_extent_bits(tree, start, end, 0, &cached_state, GFP_NOFS);
2673 wait_on_page_writeback(page);
2674 clear_extent_bit(tree, start, end,
2675 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
2676 EXTENT_DO_ACCOUNTING,
2677 1, 1, &cached_state, GFP_NOFS);
2682 * a helper for releasepage, this tests for areas of the page that
2683 * are locked or under IO and drops the related state bits if it is safe
2686 int try_release_extent_state(struct extent_map_tree *map,
2687 struct extent_io_tree *tree, struct page *page,
2690 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2691 u64 end = start + PAGE_CACHE_SIZE - 1;
2694 if (test_range_bit(tree, start, end,
2695 EXTENT_IOBITS, 0, NULL))
2698 if ((mask & GFP_NOFS) == GFP_NOFS)
2701 * at this point we can safely clear everything except the
2702 * locked bit and the nodatasum bit
2704 ret = clear_extent_bit(tree, start, end,
2705 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
2708 /* if clear_extent_bit failed for enomem reasons,
2709 * we can't allow the release to continue.
2720 * a helper for releasepage. As long as there are no locked extents
2721 * in the range corresponding to the page, both state records and extent
2722 * map records are removed
2724 int try_release_extent_mapping(struct extent_map_tree *map,
2725 struct extent_io_tree *tree, struct page *page,
2728 struct extent_map *em;
2729 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2730 u64 end = start + PAGE_CACHE_SIZE - 1;
2732 if ((mask & __GFP_WAIT) &&
2733 page->mapping->host->i_size > 16 * 1024 * 1024) {
2735 while (start <= end) {
2736 len = end - start + 1;
2737 write_lock(&map->lock);
2738 em = lookup_extent_mapping(map, start, len);
2739 if (IS_ERR_OR_NULL(em)) {
2740 write_unlock(&map->lock);
2743 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
2744 em->start != start) {
2745 write_unlock(&map->lock);
2746 free_extent_map(em);
2749 if (!test_range_bit(tree, em->start,
2750 extent_map_end(em) - 1,
2751 EXTENT_LOCKED | EXTENT_WRITEBACK,
2753 remove_extent_mapping(map, em);
2754 /* once for the rb tree */
2755 free_extent_map(em);
2757 start = extent_map_end(em);
2758 write_unlock(&map->lock);
2761 free_extent_map(em);
2764 return try_release_extent_state(map, tree, page, mask);
2768 * helper function for fiemap, which doesn't want to see any holes.
2769 * This maps until we find something past 'last'
2771 static struct extent_map *get_extent_skip_holes(struct inode *inode,
2774 get_extent_t *get_extent)
2776 u64 sectorsize = BTRFS_I(inode)->root->sectorsize;
2777 struct extent_map *em;
2784 len = last - offset;
2787 len = (len + sectorsize - 1) & ~(sectorsize - 1);
2788 em = get_extent(inode, NULL, 0, offset, len, 0);
2789 if (IS_ERR_OR_NULL(em))
2792 /* if this isn't a hole return it */
2793 if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) &&
2794 em->block_start != EXTENT_MAP_HOLE) {
2798 /* this is a hole, advance to the next extent */
2799 offset = extent_map_end(em);
2800 free_extent_map(em);
2807 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
2808 __u64 start, __u64 len, get_extent_t *get_extent)
2812 u64 max = start + len;
2816 u64 last_for_get_extent = 0;
2818 u64 isize = i_size_read(inode);
2819 struct btrfs_key found_key;
2820 struct extent_map *em = NULL;
2821 struct extent_state *cached_state = NULL;
2822 struct btrfs_path *path;
2823 struct btrfs_file_extent_item *item;
2828 unsigned long emflags;
2833 path = btrfs_alloc_path();
2836 path->leave_spinning = 1;
2839 * lookup the last file extent. We're not using i_size here
2840 * because there might be preallocation past i_size
2842 ret = btrfs_lookup_file_extent(NULL, BTRFS_I(inode)->root,
2843 path, btrfs_ino(inode), -1, 0);
2845 btrfs_free_path(path);
2850 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2851 struct btrfs_file_extent_item);
2852 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
2853 found_type = btrfs_key_type(&found_key);
2855 /* No extents, but there might be delalloc bits */
2856 if (found_key.objectid != btrfs_ino(inode) ||
2857 found_type != BTRFS_EXTENT_DATA_KEY) {
2858 /* have to trust i_size as the end */
2860 last_for_get_extent = isize;
2863 * remember the start of the last extent. There are a
2864 * bunch of different factors that go into the length of the
2865 * extent, so its much less complex to remember where it started
2867 last = found_key.offset;
2868 last_for_get_extent = last + 1;
2870 btrfs_free_path(path);
2873 * we might have some extents allocated but more delalloc past those
2874 * extents. so, we trust isize unless the start of the last extent is
2879 last_for_get_extent = isize;
2882 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len, 0,
2883 &cached_state, GFP_NOFS);
2885 em = get_extent_skip_holes(inode, off, last_for_get_extent,
2895 u64 offset_in_extent;
2897 /* break if the extent we found is outside the range */
2898 if (em->start >= max || extent_map_end(em) < off)
2902 * get_extent may return an extent that starts before our
2903 * requested range. We have to make sure the ranges
2904 * we return to fiemap always move forward and don't
2905 * overlap, so adjust the offsets here
2907 em_start = max(em->start, off);
2910 * record the offset from the start of the extent
2911 * for adjusting the disk offset below
2913 offset_in_extent = em_start - em->start;
2914 em_end = extent_map_end(em);
2915 em_len = em_end - em_start;
2916 emflags = em->flags;
2921 * bump off for our next call to get_extent
2923 off = extent_map_end(em);
2927 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
2929 flags |= FIEMAP_EXTENT_LAST;
2930 } else if (em->block_start == EXTENT_MAP_INLINE) {
2931 flags |= (FIEMAP_EXTENT_DATA_INLINE |
2932 FIEMAP_EXTENT_NOT_ALIGNED);
2933 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
2934 flags |= (FIEMAP_EXTENT_DELALLOC |
2935 FIEMAP_EXTENT_UNKNOWN);
2937 disko = em->block_start + offset_in_extent;
2939 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
2940 flags |= FIEMAP_EXTENT_ENCODED;
2942 free_extent_map(em);
2944 if ((em_start >= last) || em_len == (u64)-1 ||
2945 (last == (u64)-1 && isize <= em_end)) {
2946 flags |= FIEMAP_EXTENT_LAST;
2950 /* now scan forward to see if this is really the last extent. */
2951 em = get_extent_skip_holes(inode, off, last_for_get_extent,
2958 flags |= FIEMAP_EXTENT_LAST;
2961 ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
2967 free_extent_map(em);
2969 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len,
2970 &cached_state, GFP_NOFS);
2974 static inline struct page *extent_buffer_page(struct extent_buffer *eb,
2978 struct address_space *mapping;
2981 return eb->first_page;
2982 i += eb->start >> PAGE_CACHE_SHIFT;
2983 mapping = eb->first_page->mapping;
2988 * extent_buffer_page is only called after pinning the page
2989 * by increasing the reference count. So we know the page must
2990 * be in the radix tree.
2993 p = radix_tree_lookup(&mapping->page_tree, i);
2999 static inline unsigned long num_extent_pages(u64 start, u64 len)
3001 return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
3002 (start >> PAGE_CACHE_SHIFT);
3005 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
3010 struct extent_buffer *eb = NULL;
3012 unsigned long flags;
3015 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
3020 rwlock_init(&eb->lock);
3021 atomic_set(&eb->write_locks, 0);
3022 atomic_set(&eb->read_locks, 0);
3023 atomic_set(&eb->blocking_readers, 0);
3024 atomic_set(&eb->blocking_writers, 0);
3025 atomic_set(&eb->spinning_readers, 0);
3026 atomic_set(&eb->spinning_writers, 0);
3027 init_waitqueue_head(&eb->write_lock_wq);
3028 init_waitqueue_head(&eb->read_lock_wq);
3031 spin_lock_irqsave(&leak_lock, flags);
3032 list_add(&eb->leak_list, &buffers);
3033 spin_unlock_irqrestore(&leak_lock, flags);
3035 atomic_set(&eb->refs, 1);
3040 static void __free_extent_buffer(struct extent_buffer *eb)
3043 unsigned long flags;
3044 spin_lock_irqsave(&leak_lock, flags);
3045 list_del(&eb->leak_list);
3046 spin_unlock_irqrestore(&leak_lock, flags);
3048 kmem_cache_free(extent_buffer_cache, eb);
3052 * Helper for releasing extent buffer page.
3054 static void btrfs_release_extent_buffer_page(struct extent_buffer *eb,
3055 unsigned long start_idx)
3057 unsigned long index;
3060 if (!eb->first_page)
3063 index = num_extent_pages(eb->start, eb->len);
3064 if (start_idx >= index)
3069 page = extent_buffer_page(eb, index);
3071 page_cache_release(page);
3072 } while (index != start_idx);
3076 * Helper for releasing the extent buffer.
3078 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
3080 btrfs_release_extent_buffer_page(eb, 0);
3081 __free_extent_buffer(eb);
3084 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
3085 u64 start, unsigned long len,
3088 unsigned long num_pages = num_extent_pages(start, len);
3090 unsigned long index = start >> PAGE_CACHE_SHIFT;
3091 struct extent_buffer *eb;
3092 struct extent_buffer *exists = NULL;
3094 struct address_space *mapping = tree->mapping;
3099 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
3100 if (eb && atomic_inc_not_zero(&eb->refs)) {
3102 mark_page_accessed(eb->first_page);
3107 eb = __alloc_extent_buffer(tree, start, len, GFP_NOFS);
3112 eb->first_page = page0;
3115 page_cache_get(page0);
3116 mark_page_accessed(page0);
3117 set_page_extent_mapped(page0);
3118 set_page_extent_head(page0, len);
3119 uptodate = PageUptodate(page0);
3123 for (; i < num_pages; i++, index++) {
3124 p = find_or_create_page(mapping, index, GFP_NOFS);
3129 set_page_extent_mapped(p);
3130 mark_page_accessed(p);
3133 set_page_extent_head(p, len);
3135 set_page_private(p, EXTENT_PAGE_PRIVATE);
3137 if (!PageUptodate(p))
3141 * see below about how we avoid a nasty race with release page
3142 * and why we unlock later
3148 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3150 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
3154 spin_lock(&tree->buffer_lock);
3155 ret = radix_tree_insert(&tree->buffer, start >> PAGE_CACHE_SHIFT, eb);
3156 if (ret == -EEXIST) {
3157 exists = radix_tree_lookup(&tree->buffer,
3158 start >> PAGE_CACHE_SHIFT);
3159 /* add one reference for the caller */
3160 atomic_inc(&exists->refs);
3161 spin_unlock(&tree->buffer_lock);
3162 radix_tree_preload_end();
3165 /* add one reference for the tree */
3166 atomic_inc(&eb->refs);
3167 spin_unlock(&tree->buffer_lock);
3168 radix_tree_preload_end();
3171 * there is a race where release page may have
3172 * tried to find this extent buffer in the radix
3173 * but failed. It will tell the VM it is safe to
3174 * reclaim the, and it will clear the page private bit.
3175 * We must make sure to set the page private bit properly
3176 * after the extent buffer is in the radix tree so
3177 * it doesn't get lost
3179 set_page_extent_mapped(eb->first_page);
3180 set_page_extent_head(eb->first_page, eb->len);
3182 unlock_page(eb->first_page);
3186 if (eb->first_page && !page0)
3187 unlock_page(eb->first_page);
3189 if (!atomic_dec_and_test(&eb->refs))
3191 btrfs_release_extent_buffer(eb);
3195 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
3196 u64 start, unsigned long len)
3198 struct extent_buffer *eb;
3201 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
3202 if (eb && atomic_inc_not_zero(&eb->refs)) {
3204 mark_page_accessed(eb->first_page);
3212 void free_extent_buffer(struct extent_buffer *eb)
3217 if (!atomic_dec_and_test(&eb->refs))
3223 int clear_extent_buffer_dirty(struct extent_io_tree *tree,
3224 struct extent_buffer *eb)
3227 unsigned long num_pages;
3230 num_pages = num_extent_pages(eb->start, eb->len);
3232 for (i = 0; i < num_pages; i++) {
3233 page = extent_buffer_page(eb, i);
3234 if (!PageDirty(page))
3238 WARN_ON(!PagePrivate(page));
3240 set_page_extent_mapped(page);
3242 set_page_extent_head(page, eb->len);
3244 clear_page_dirty_for_io(page);
3245 spin_lock_irq(&page->mapping->tree_lock);
3246 if (!PageDirty(page)) {
3247 radix_tree_tag_clear(&page->mapping->page_tree,
3249 PAGECACHE_TAG_DIRTY);
3251 spin_unlock_irq(&page->mapping->tree_lock);
3257 int set_extent_buffer_dirty(struct extent_io_tree *tree,
3258 struct extent_buffer *eb)
3261 unsigned long num_pages;
3264 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
3265 num_pages = num_extent_pages(eb->start, eb->len);
3266 for (i = 0; i < num_pages; i++)
3267 __set_page_dirty_nobuffers(extent_buffer_page(eb, i));
3271 static int __eb_straddles_pages(u64 start, u64 len)
3273 if (len < PAGE_CACHE_SIZE)
3275 if (start & (PAGE_CACHE_SIZE - 1))
3277 if ((start + len) & (PAGE_CACHE_SIZE - 1))
3282 static int eb_straddles_pages(struct extent_buffer *eb)
3284 return __eb_straddles_pages(eb->start, eb->len);
3287 int clear_extent_buffer_uptodate(struct extent_io_tree *tree,
3288 struct extent_buffer *eb,
3289 struct extent_state **cached_state)
3293 unsigned long num_pages;
3295 num_pages = num_extent_pages(eb->start, eb->len);
3296 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3298 if (eb_straddles_pages(eb)) {
3299 clear_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
3300 cached_state, GFP_NOFS);
3302 for (i = 0; i < num_pages; i++) {
3303 page = extent_buffer_page(eb, i);
3305 ClearPageUptodate(page);
3310 int set_extent_buffer_uptodate(struct extent_io_tree *tree,
3311 struct extent_buffer *eb)
3315 unsigned long num_pages;
3317 num_pages = num_extent_pages(eb->start, eb->len);
3319 if (eb_straddles_pages(eb)) {
3320 set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
3323 for (i = 0; i < num_pages; i++) {
3324 page = extent_buffer_page(eb, i);
3325 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
3326 ((i == num_pages - 1) &&
3327 ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
3328 check_page_uptodate(tree, page);
3331 SetPageUptodate(page);
3336 int extent_range_uptodate(struct extent_io_tree *tree,
3341 int pg_uptodate = 1;
3343 unsigned long index;
3345 if (__eb_straddles_pages(start, end - start + 1)) {
3346 ret = test_range_bit(tree, start, end,
3347 EXTENT_UPTODATE, 1, NULL);
3351 while (start <= end) {
3352 index = start >> PAGE_CACHE_SHIFT;
3353 page = find_get_page(tree->mapping, index);
3354 uptodate = PageUptodate(page);
3355 page_cache_release(page);
3360 start += PAGE_CACHE_SIZE;
3365 int extent_buffer_uptodate(struct extent_io_tree *tree,
3366 struct extent_buffer *eb,
3367 struct extent_state *cached_state)
3370 unsigned long num_pages;
3373 int pg_uptodate = 1;
3375 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
3378 if (eb_straddles_pages(eb)) {
3379 ret = test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3380 EXTENT_UPTODATE, 1, cached_state);
3385 num_pages = num_extent_pages(eb->start, eb->len);
3386 for (i = 0; i < num_pages; i++) {
3387 page = extent_buffer_page(eb, i);
3388 if (!PageUptodate(page)) {
3396 int read_extent_buffer_pages(struct extent_io_tree *tree,
3397 struct extent_buffer *eb,
3398 u64 start, int wait,
3399 get_extent_t *get_extent, int mirror_num)
3402 unsigned long start_i;
3406 int locked_pages = 0;
3407 int all_uptodate = 1;
3408 int inc_all_pages = 0;
3409 unsigned long num_pages;
3410 struct bio *bio = NULL;
3411 unsigned long bio_flags = 0;
3413 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
3416 if (eb_straddles_pages(eb)) {
3417 if (test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3418 EXTENT_UPTODATE, 1, NULL)) {
3424 WARN_ON(start < eb->start);
3425 start_i = (start >> PAGE_CACHE_SHIFT) -
3426 (eb->start >> PAGE_CACHE_SHIFT);
3431 num_pages = num_extent_pages(eb->start, eb->len);
3432 for (i = start_i; i < num_pages; i++) {
3433 page = extent_buffer_page(eb, i);
3435 if (!trylock_page(page))
3441 if (!PageUptodate(page))
3446 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3450 for (i = start_i; i < num_pages; i++) {
3451 page = extent_buffer_page(eb, i);
3453 WARN_ON(!PagePrivate(page));
3455 set_page_extent_mapped(page);
3457 set_page_extent_head(page, eb->len);
3460 page_cache_get(page);
3461 if (!PageUptodate(page)) {
3464 ClearPageError(page);
3465 err = __extent_read_full_page(tree, page,
3467 mirror_num, &bio_flags);
3476 submit_one_bio(READ, bio, mirror_num, bio_flags);
3481 for (i = start_i; i < num_pages; i++) {
3482 page = extent_buffer_page(eb, i);
3483 wait_on_page_locked(page);
3484 if (!PageUptodate(page))
3489 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3494 while (locked_pages > 0) {
3495 page = extent_buffer_page(eb, i);
3503 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
3504 unsigned long start,
3511 char *dst = (char *)dstv;
3512 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3513 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3515 WARN_ON(start > eb->len);
3516 WARN_ON(start + len > eb->start + eb->len);
3518 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3521 page = extent_buffer_page(eb, i);
3523 cur = min(len, (PAGE_CACHE_SIZE - offset));
3524 kaddr = page_address(page);
3525 memcpy(dst, kaddr + offset, cur);
3534 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
3535 unsigned long min_len, char **map,
3536 unsigned long *map_start,
3537 unsigned long *map_len)
3539 size_t offset = start & (PAGE_CACHE_SIZE - 1);
3542 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3543 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3544 unsigned long end_i = (start_offset + start + min_len - 1) >>
3551 offset = start_offset;
3555 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
3558 if (start + min_len > eb->len) {
3559 printk(KERN_ERR "btrfs bad mapping eb start %llu len %lu, "
3560 "wanted %lu %lu\n", (unsigned long long)eb->start,
3561 eb->len, start, min_len);
3566 p = extent_buffer_page(eb, i);
3567 kaddr = page_address(p);
3568 *map = kaddr + offset;
3569 *map_len = PAGE_CACHE_SIZE - offset;
3573 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
3574 unsigned long start,
3581 char *ptr = (char *)ptrv;
3582 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3583 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3586 WARN_ON(start > eb->len);
3587 WARN_ON(start + len > eb->start + eb->len);
3589 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3592 page = extent_buffer_page(eb, i);
3594 cur = min(len, (PAGE_CACHE_SIZE - offset));
3596 kaddr = page_address(page);
3597 ret = memcmp(ptr, kaddr + offset, cur);
3609 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
3610 unsigned long start, unsigned long len)
3616 char *src = (char *)srcv;
3617 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3618 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3620 WARN_ON(start > eb->len);
3621 WARN_ON(start + len > eb->start + eb->len);
3623 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3626 page = extent_buffer_page(eb, i);
3627 WARN_ON(!PageUptodate(page));
3629 cur = min(len, PAGE_CACHE_SIZE - offset);
3630 kaddr = page_address(page);
3631 memcpy(kaddr + offset, src, cur);
3640 void memset_extent_buffer(struct extent_buffer *eb, char c,
3641 unsigned long start, unsigned long len)
3647 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3648 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3650 WARN_ON(start > eb->len);
3651 WARN_ON(start + len > eb->start + eb->len);
3653 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3656 page = extent_buffer_page(eb, i);
3657 WARN_ON(!PageUptodate(page));
3659 cur = min(len, PAGE_CACHE_SIZE - offset);
3660 kaddr = page_address(page);
3661 memset(kaddr + offset, c, cur);
3669 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
3670 unsigned long dst_offset, unsigned long src_offset,
3673 u64 dst_len = dst->len;
3678 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3679 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3681 WARN_ON(src->len != dst_len);
3683 offset = (start_offset + dst_offset) &
3684 ((unsigned long)PAGE_CACHE_SIZE - 1);
3687 page = extent_buffer_page(dst, i);
3688 WARN_ON(!PageUptodate(page));
3690 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
3692 kaddr = page_address(page);
3693 read_extent_buffer(src, kaddr + offset, src_offset, cur);
3702 static void move_pages(struct page *dst_page, struct page *src_page,
3703 unsigned long dst_off, unsigned long src_off,
3706 char *dst_kaddr = page_address(dst_page);
3707 if (dst_page == src_page) {
3708 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
3710 char *src_kaddr = page_address(src_page);
3711 char *p = dst_kaddr + dst_off + len;
3712 char *s = src_kaddr + src_off + len;
3719 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
3721 unsigned long distance = (src > dst) ? src - dst : dst - src;
3722 return distance < len;
3725 static void copy_pages(struct page *dst_page, struct page *src_page,
3726 unsigned long dst_off, unsigned long src_off,
3729 char *dst_kaddr = page_address(dst_page);
3732 if (dst_page != src_page) {
3733 src_kaddr = page_address(src_page);
3735 src_kaddr = dst_kaddr;
3736 BUG_ON(areas_overlap(src_off, dst_off, len));
3739 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
3742 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3743 unsigned long src_offset, unsigned long len)
3746 size_t dst_off_in_page;
3747 size_t src_off_in_page;
3748 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3749 unsigned long dst_i;
3750 unsigned long src_i;
3752 if (src_offset + len > dst->len) {
3753 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
3754 "len %lu dst len %lu\n", src_offset, len, dst->len);
3757 if (dst_offset + len > dst->len) {
3758 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
3759 "len %lu dst len %lu\n", dst_offset, len, dst->len);
3764 dst_off_in_page = (start_offset + dst_offset) &
3765 ((unsigned long)PAGE_CACHE_SIZE - 1);
3766 src_off_in_page = (start_offset + src_offset) &
3767 ((unsigned long)PAGE_CACHE_SIZE - 1);
3769 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3770 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
3772 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
3774 cur = min_t(unsigned long, cur,
3775 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
3777 copy_pages(extent_buffer_page(dst, dst_i),
3778 extent_buffer_page(dst, src_i),
3779 dst_off_in_page, src_off_in_page, cur);
3787 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3788 unsigned long src_offset, unsigned long len)
3791 size_t dst_off_in_page;
3792 size_t src_off_in_page;
3793 unsigned long dst_end = dst_offset + len - 1;
3794 unsigned long src_end = src_offset + len - 1;
3795 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3796 unsigned long dst_i;
3797 unsigned long src_i;
3799 if (src_offset + len > dst->len) {
3800 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
3801 "len %lu len %lu\n", src_offset, len, dst->len);
3804 if (dst_offset + len > dst->len) {
3805 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
3806 "len %lu len %lu\n", dst_offset, len, dst->len);
3809 if (!areas_overlap(src_offset, dst_offset, len)) {
3810 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
3814 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
3815 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
3817 dst_off_in_page = (start_offset + dst_end) &
3818 ((unsigned long)PAGE_CACHE_SIZE - 1);
3819 src_off_in_page = (start_offset + src_end) &
3820 ((unsigned long)PAGE_CACHE_SIZE - 1);
3822 cur = min_t(unsigned long, len, src_off_in_page + 1);
3823 cur = min(cur, dst_off_in_page + 1);
3824 move_pages(extent_buffer_page(dst, dst_i),
3825 extent_buffer_page(dst, src_i),
3826 dst_off_in_page - cur + 1,
3827 src_off_in_page - cur + 1, cur);
3835 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
3837 struct extent_buffer *eb =
3838 container_of(head, struct extent_buffer, rcu_head);
3840 btrfs_release_extent_buffer(eb);
3843 int try_release_extent_buffer(struct extent_io_tree *tree, struct page *page)
3845 u64 start = page_offset(page);
3846 struct extent_buffer *eb;
3849 spin_lock(&tree->buffer_lock);
3850 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
3852 spin_unlock(&tree->buffer_lock);
3856 if (test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3862 * set @eb->refs to 0 if it is already 1, and then release the @eb.
3865 if (atomic_cmpxchg(&eb->refs, 1, 0) != 1) {
3870 radix_tree_delete(&tree->buffer, start >> PAGE_CACHE_SHIFT);
3872 spin_unlock(&tree->buffer_lock);
3874 /* at this point we can safely release the extent buffer */
3875 if (atomic_read(&eb->refs) == 0)
3876 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);