1 #include <linux/bitops.h>
2 #include <linux/slab.h>
5 #include <linux/pagemap.h>
6 #include <linux/page-flags.h>
7 #include <linux/spinlock.h>
8 #include <linux/blkdev.h>
9 #include <linux/swap.h>
10 #include <linux/writeback.h>
11 #include <linux/pagevec.h>
12 #include <linux/prefetch.h>
13 #include <linux/cleancache.h>
14 #include "extent_io.h"
15 #include "extent_map.h"
18 #include "btrfs_inode.h"
20 #include "check-integrity.h"
22 #include "rcu-string.h"
24 static struct kmem_cache *extent_state_cache;
25 static struct kmem_cache *extent_buffer_cache;
27 static LIST_HEAD(buffers);
28 static LIST_HEAD(states);
32 static DEFINE_SPINLOCK(leak_lock);
35 #define BUFFER_LRU_MAX 64
40 struct rb_node rb_node;
43 struct extent_page_data {
45 struct extent_io_tree *tree;
46 get_extent_t *get_extent;
47 unsigned long bio_flags;
49 /* tells writepage not to lock the state bits for this range
50 * it still does the unlocking
52 unsigned int extent_locked:1;
54 /* tells the submit_bio code to use a WRITE_SYNC */
55 unsigned int sync_io:1;
58 static noinline void flush_write_bio(void *data);
59 static inline struct btrfs_fs_info *
60 tree_fs_info(struct extent_io_tree *tree)
62 return btrfs_sb(tree->mapping->host->i_sb);
65 int __init extent_io_init(void)
67 extent_state_cache = kmem_cache_create("btrfs_extent_state",
68 sizeof(struct extent_state), 0,
69 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
70 if (!extent_state_cache)
73 extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer",
74 sizeof(struct extent_buffer), 0,
75 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
76 if (!extent_buffer_cache)
77 goto free_state_cache;
81 kmem_cache_destroy(extent_state_cache);
85 void extent_io_exit(void)
87 struct extent_state *state;
88 struct extent_buffer *eb;
90 while (!list_empty(&states)) {
91 state = list_entry(states.next, struct extent_state, leak_list);
92 printk(KERN_ERR "btrfs state leak: start %llu end %llu "
93 "state %lu in tree %p refs %d\n",
94 (unsigned long long)state->start,
95 (unsigned long long)state->end,
96 state->state, state->tree, atomic_read(&state->refs));
97 list_del(&state->leak_list);
98 kmem_cache_free(extent_state_cache, state);
102 while (!list_empty(&buffers)) {
103 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
104 printk(KERN_ERR "btrfs buffer leak start %llu len %lu "
105 "refs %d\n", (unsigned long long)eb->start,
106 eb->len, atomic_read(&eb->refs));
107 list_del(&eb->leak_list);
108 kmem_cache_free(extent_buffer_cache, eb);
112 * Make sure all delayed rcu free are flushed before we
116 if (extent_state_cache)
117 kmem_cache_destroy(extent_state_cache);
118 if (extent_buffer_cache)
119 kmem_cache_destroy(extent_buffer_cache);
122 void extent_io_tree_init(struct extent_io_tree *tree,
123 struct address_space *mapping)
125 tree->state = RB_ROOT;
126 INIT_RADIX_TREE(&tree->buffer, GFP_ATOMIC);
128 tree->dirty_bytes = 0;
129 spin_lock_init(&tree->lock);
130 spin_lock_init(&tree->buffer_lock);
131 tree->mapping = mapping;
134 static struct extent_state *alloc_extent_state(gfp_t mask)
136 struct extent_state *state;
141 state = kmem_cache_alloc(extent_state_cache, mask);
148 spin_lock_irqsave(&leak_lock, flags);
149 list_add(&state->leak_list, &states);
150 spin_unlock_irqrestore(&leak_lock, flags);
152 atomic_set(&state->refs, 1);
153 init_waitqueue_head(&state->wq);
154 trace_alloc_extent_state(state, mask, _RET_IP_);
158 void free_extent_state(struct extent_state *state)
162 if (atomic_dec_and_test(&state->refs)) {
166 WARN_ON(state->tree);
168 spin_lock_irqsave(&leak_lock, flags);
169 list_del(&state->leak_list);
170 spin_unlock_irqrestore(&leak_lock, flags);
172 trace_free_extent_state(state, _RET_IP_);
173 kmem_cache_free(extent_state_cache, state);
177 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
178 struct rb_node *node)
180 struct rb_node **p = &root->rb_node;
181 struct rb_node *parent = NULL;
182 struct tree_entry *entry;
186 entry = rb_entry(parent, struct tree_entry, rb_node);
188 if (offset < entry->start)
190 else if (offset > entry->end)
196 rb_link_node(node, parent, p);
197 rb_insert_color(node, root);
201 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
202 struct rb_node **prev_ret,
203 struct rb_node **next_ret)
205 struct rb_root *root = &tree->state;
206 struct rb_node *n = root->rb_node;
207 struct rb_node *prev = NULL;
208 struct rb_node *orig_prev = NULL;
209 struct tree_entry *entry;
210 struct tree_entry *prev_entry = NULL;
213 entry = rb_entry(n, struct tree_entry, rb_node);
217 if (offset < entry->start)
219 else if (offset > entry->end)
227 while (prev && offset > prev_entry->end) {
228 prev = rb_next(prev);
229 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
236 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
237 while (prev && offset < prev_entry->start) {
238 prev = rb_prev(prev);
239 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
246 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
249 struct rb_node *prev = NULL;
252 ret = __etree_search(tree, offset, &prev, NULL);
258 static void merge_cb(struct extent_io_tree *tree, struct extent_state *new,
259 struct extent_state *other)
261 if (tree->ops && tree->ops->merge_extent_hook)
262 tree->ops->merge_extent_hook(tree->mapping->host, new,
267 * utility function to look for merge candidates inside a given range.
268 * Any extents with matching state are merged together into a single
269 * extent in the tree. Extents with EXTENT_IO in their state field
270 * are not merged because the end_io handlers need to be able to do
271 * operations on them without sleeping (or doing allocations/splits).
273 * This should be called with the tree lock held.
275 static void merge_state(struct extent_io_tree *tree,
276 struct extent_state *state)
278 struct extent_state *other;
279 struct rb_node *other_node;
281 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
284 other_node = rb_prev(&state->rb_node);
286 other = rb_entry(other_node, struct extent_state, rb_node);
287 if (other->end == state->start - 1 &&
288 other->state == state->state) {
289 merge_cb(tree, state, other);
290 state->start = other->start;
292 rb_erase(&other->rb_node, &tree->state);
293 free_extent_state(other);
296 other_node = rb_next(&state->rb_node);
298 other = rb_entry(other_node, struct extent_state, rb_node);
299 if (other->start == state->end + 1 &&
300 other->state == state->state) {
301 merge_cb(tree, state, other);
302 state->end = other->end;
304 rb_erase(&other->rb_node, &tree->state);
305 free_extent_state(other);
310 static void set_state_cb(struct extent_io_tree *tree,
311 struct extent_state *state, int *bits)
313 if (tree->ops && tree->ops->set_bit_hook)
314 tree->ops->set_bit_hook(tree->mapping->host, state, bits);
317 static void clear_state_cb(struct extent_io_tree *tree,
318 struct extent_state *state, int *bits)
320 if (tree->ops && tree->ops->clear_bit_hook)
321 tree->ops->clear_bit_hook(tree->mapping->host, state, bits);
324 static void set_state_bits(struct extent_io_tree *tree,
325 struct extent_state *state, int *bits);
328 * insert an extent_state struct into the tree. 'bits' are set on the
329 * struct before it is inserted.
331 * This may return -EEXIST if the extent is already there, in which case the
332 * state struct is freed.
334 * The tree lock is not taken internally. This is a utility function and
335 * probably isn't what you want to call (see set/clear_extent_bit).
337 static int insert_state(struct extent_io_tree *tree,
338 struct extent_state *state, u64 start, u64 end,
341 struct rb_node *node;
344 WARN(1, KERN_ERR "btrfs end < start %llu %llu\n",
345 (unsigned long long)end,
346 (unsigned long long)start);
347 state->start = start;
350 set_state_bits(tree, state, bits);
352 node = tree_insert(&tree->state, end, &state->rb_node);
354 struct extent_state *found;
355 found = rb_entry(node, struct extent_state, rb_node);
356 printk(KERN_ERR "btrfs found node %llu %llu on insert of "
357 "%llu %llu\n", (unsigned long long)found->start,
358 (unsigned long long)found->end,
359 (unsigned long long)start, (unsigned long long)end);
363 merge_state(tree, state);
367 static void split_cb(struct extent_io_tree *tree, struct extent_state *orig,
370 if (tree->ops && tree->ops->split_extent_hook)
371 tree->ops->split_extent_hook(tree->mapping->host, orig, split);
375 * split a given extent state struct in two, inserting the preallocated
376 * struct 'prealloc' as the newly created second half. 'split' indicates an
377 * offset inside 'orig' where it should be split.
380 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
381 * are two extent state structs in the tree:
382 * prealloc: [orig->start, split - 1]
383 * orig: [ split, orig->end ]
385 * The tree locks are not taken by this function. They need to be held
388 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
389 struct extent_state *prealloc, u64 split)
391 struct rb_node *node;
393 split_cb(tree, orig, split);
395 prealloc->start = orig->start;
396 prealloc->end = split - 1;
397 prealloc->state = orig->state;
400 node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
402 free_extent_state(prealloc);
405 prealloc->tree = tree;
409 static struct extent_state *next_state(struct extent_state *state)
411 struct rb_node *next = rb_next(&state->rb_node);
413 return rb_entry(next, struct extent_state, rb_node);
419 * utility function to clear some bits in an extent state struct.
420 * it will optionally wake up any one waiting on this state (wake == 1).
422 * If no bits are set on the state struct after clearing things, the
423 * struct is freed and removed from the tree
425 static struct extent_state *clear_state_bit(struct extent_io_tree *tree,
426 struct extent_state *state,
429 struct extent_state *next;
430 int bits_to_clear = *bits & ~EXTENT_CTLBITS;
432 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
433 u64 range = state->end - state->start + 1;
434 WARN_ON(range > tree->dirty_bytes);
435 tree->dirty_bytes -= range;
437 clear_state_cb(tree, state, bits);
438 state->state &= ~bits_to_clear;
441 if (state->state == 0) {
442 next = next_state(state);
444 rb_erase(&state->rb_node, &tree->state);
446 free_extent_state(state);
451 merge_state(tree, state);
452 next = next_state(state);
457 static struct extent_state *
458 alloc_extent_state_atomic(struct extent_state *prealloc)
461 prealloc = alloc_extent_state(GFP_ATOMIC);
466 void extent_io_tree_panic(struct extent_io_tree *tree, int err)
468 btrfs_panic(tree_fs_info(tree), err, "Locking error: "
469 "Extent tree was modified by another "
470 "thread while locked.");
474 * clear some bits on a range in the tree. This may require splitting
475 * or inserting elements in the tree, so the gfp mask is used to
476 * indicate which allocations or sleeping are allowed.
478 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
479 * the given range from the tree regardless of state (ie for truncate).
481 * the range [start, end] is inclusive.
483 * This takes the tree lock, and returns 0 on success and < 0 on error.
485 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
486 int bits, int wake, int delete,
487 struct extent_state **cached_state,
490 struct extent_state *state;
491 struct extent_state *cached;
492 struct extent_state *prealloc = NULL;
493 struct rb_node *node;
499 bits |= ~EXTENT_CTLBITS;
500 bits |= EXTENT_FIRST_DELALLOC;
502 if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY))
505 if (!prealloc && (mask & __GFP_WAIT)) {
506 prealloc = alloc_extent_state(mask);
511 spin_lock(&tree->lock);
513 cached = *cached_state;
516 *cached_state = NULL;
520 if (cached && cached->tree && cached->start <= start &&
521 cached->end > start) {
523 atomic_dec(&cached->refs);
528 free_extent_state(cached);
531 * this search will find the extents that end after
534 node = tree_search(tree, start);
537 state = rb_entry(node, struct extent_state, rb_node);
539 if (state->start > end)
541 WARN_ON(state->end < start);
542 last_end = state->end;
544 /* the state doesn't have the wanted bits, go ahead */
545 if (!(state->state & bits)) {
546 state = next_state(state);
551 * | ---- desired range ---- |
553 * | ------------- state -------------- |
555 * We need to split the extent we found, and may flip
556 * bits on second half.
558 * If the extent we found extends past our range, we
559 * just split and search again. It'll get split again
560 * the next time though.
562 * If the extent we found is inside our range, we clear
563 * the desired bit on it.
566 if (state->start < start) {
567 prealloc = alloc_extent_state_atomic(prealloc);
569 err = split_state(tree, state, prealloc, start);
571 extent_io_tree_panic(tree, err);
576 if (state->end <= end) {
577 state = clear_state_bit(tree, state, &bits, wake);
583 * | ---- desired range ---- |
585 * We need to split the extent, and clear the bit
588 if (state->start <= end && state->end > end) {
589 prealloc = alloc_extent_state_atomic(prealloc);
591 err = split_state(tree, state, prealloc, end + 1);
593 extent_io_tree_panic(tree, err);
598 clear_state_bit(tree, prealloc, &bits, wake);
604 state = clear_state_bit(tree, state, &bits, wake);
606 if (last_end == (u64)-1)
608 start = last_end + 1;
609 if (start <= end && state && !need_resched())
614 spin_unlock(&tree->lock);
616 free_extent_state(prealloc);
623 spin_unlock(&tree->lock);
624 if (mask & __GFP_WAIT)
629 static void wait_on_state(struct extent_io_tree *tree,
630 struct extent_state *state)
631 __releases(tree->lock)
632 __acquires(tree->lock)
635 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
636 spin_unlock(&tree->lock);
638 spin_lock(&tree->lock);
639 finish_wait(&state->wq, &wait);
643 * waits for one or more bits to clear on a range in the state tree.
644 * The range [start, end] is inclusive.
645 * The tree lock is taken by this function
647 void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits)
649 struct extent_state *state;
650 struct rb_node *node;
652 spin_lock(&tree->lock);
656 * this search will find all the extents that end after
659 node = tree_search(tree, start);
663 state = rb_entry(node, struct extent_state, rb_node);
665 if (state->start > end)
668 if (state->state & bits) {
669 start = state->start;
670 atomic_inc(&state->refs);
671 wait_on_state(tree, state);
672 free_extent_state(state);
675 start = state->end + 1;
680 cond_resched_lock(&tree->lock);
683 spin_unlock(&tree->lock);
686 static void set_state_bits(struct extent_io_tree *tree,
687 struct extent_state *state,
690 int bits_to_set = *bits & ~EXTENT_CTLBITS;
692 set_state_cb(tree, state, bits);
693 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
694 u64 range = state->end - state->start + 1;
695 tree->dirty_bytes += range;
697 state->state |= bits_to_set;
700 static void cache_state(struct extent_state *state,
701 struct extent_state **cached_ptr)
703 if (cached_ptr && !(*cached_ptr)) {
704 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) {
706 atomic_inc(&state->refs);
711 static void uncache_state(struct extent_state **cached_ptr)
713 if (cached_ptr && (*cached_ptr)) {
714 struct extent_state *state = *cached_ptr;
716 free_extent_state(state);
721 * set some bits on a range in the tree. This may require allocations or
722 * sleeping, so the gfp mask is used to indicate what is allowed.
724 * If any of the exclusive bits are set, this will fail with -EEXIST if some
725 * part of the range already has the desired bits set. The start of the
726 * existing range is returned in failed_start in this case.
728 * [start, end] is inclusive This takes the tree lock.
731 static int __must_check
732 __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
733 int bits, int exclusive_bits, u64 *failed_start,
734 struct extent_state **cached_state, gfp_t mask)
736 struct extent_state *state;
737 struct extent_state *prealloc = NULL;
738 struct rb_node *node;
743 bits |= EXTENT_FIRST_DELALLOC;
745 if (!prealloc && (mask & __GFP_WAIT)) {
746 prealloc = alloc_extent_state(mask);
750 spin_lock(&tree->lock);
751 if (cached_state && *cached_state) {
752 state = *cached_state;
753 if (state->start <= start && state->end > start &&
755 node = &state->rb_node;
760 * this search will find all the extents that end after
763 node = tree_search(tree, start);
765 prealloc = alloc_extent_state_atomic(prealloc);
767 err = insert_state(tree, prealloc, start, end, &bits);
769 extent_io_tree_panic(tree, err);
774 state = rb_entry(node, struct extent_state, rb_node);
776 last_start = state->start;
777 last_end = state->end;
780 * | ---- desired range ---- |
783 * Just lock what we found and keep going
785 if (state->start == start && state->end <= end) {
786 if (state->state & exclusive_bits) {
787 *failed_start = state->start;
792 set_state_bits(tree, state, &bits);
793 cache_state(state, cached_state);
794 merge_state(tree, state);
795 if (last_end == (u64)-1)
797 start = last_end + 1;
798 state = next_state(state);
799 if (start < end && state && state->start == start &&
806 * | ---- desired range ---- |
809 * | ------------- state -------------- |
811 * We need to split the extent we found, and may flip bits on
814 * If the extent we found extends past our
815 * range, we just split and search again. It'll get split
816 * again the next time though.
818 * If the extent we found is inside our range, we set the
821 if (state->start < start) {
822 if (state->state & exclusive_bits) {
823 *failed_start = start;
828 prealloc = alloc_extent_state_atomic(prealloc);
830 err = split_state(tree, state, prealloc, start);
832 extent_io_tree_panic(tree, err);
837 if (state->end <= end) {
838 set_state_bits(tree, state, &bits);
839 cache_state(state, cached_state);
840 merge_state(tree, state);
841 if (last_end == (u64)-1)
843 start = last_end + 1;
844 state = next_state(state);
845 if (start < end && state && state->start == start &&
852 * | ---- desired range ---- |
853 * | state | or | state |
855 * There's a hole, we need to insert something in it and
856 * ignore the extent we found.
858 if (state->start > start) {
860 if (end < last_start)
863 this_end = last_start - 1;
865 prealloc = alloc_extent_state_atomic(prealloc);
869 * Avoid to free 'prealloc' if it can be merged with
872 err = insert_state(tree, prealloc, start, this_end,
875 extent_io_tree_panic(tree, err);
877 cache_state(prealloc, cached_state);
879 start = this_end + 1;
883 * | ---- desired range ---- |
885 * We need to split the extent, and set the bit
888 if (state->start <= end && state->end > end) {
889 if (state->state & exclusive_bits) {
890 *failed_start = start;
895 prealloc = alloc_extent_state_atomic(prealloc);
897 err = split_state(tree, state, prealloc, end + 1);
899 extent_io_tree_panic(tree, err);
901 set_state_bits(tree, prealloc, &bits);
902 cache_state(prealloc, cached_state);
903 merge_state(tree, prealloc);
911 spin_unlock(&tree->lock);
913 free_extent_state(prealloc);
920 spin_unlock(&tree->lock);
921 if (mask & __GFP_WAIT)
926 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits,
927 u64 *failed_start, struct extent_state **cached_state,
930 return __set_extent_bit(tree, start, end, bits, 0, failed_start,
936 * convert_extent_bit - convert all bits in a given range from one bit to
938 * @tree: the io tree to search
939 * @start: the start offset in bytes
940 * @end: the end offset in bytes (inclusive)
941 * @bits: the bits to set in this range
942 * @clear_bits: the bits to clear in this range
943 * @cached_state: state that we're going to cache
944 * @mask: the allocation mask
946 * This will go through and set bits for the given range. If any states exist
947 * already in this range they are set with the given bit and cleared of the
948 * clear_bits. This is only meant to be used by things that are mergeable, ie
949 * converting from say DELALLOC to DIRTY. This is not meant to be used with
950 * boundary bits like LOCK.
952 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
953 int bits, int clear_bits,
954 struct extent_state **cached_state, gfp_t mask)
956 struct extent_state *state;
957 struct extent_state *prealloc = NULL;
958 struct rb_node *node;
964 if (!prealloc && (mask & __GFP_WAIT)) {
965 prealloc = alloc_extent_state(mask);
970 spin_lock(&tree->lock);
971 if (cached_state && *cached_state) {
972 state = *cached_state;
973 if (state->start <= start && state->end > start &&
975 node = &state->rb_node;
981 * this search will find all the extents that end after
984 node = tree_search(tree, start);
986 prealloc = alloc_extent_state_atomic(prealloc);
991 err = insert_state(tree, prealloc, start, end, &bits);
994 extent_io_tree_panic(tree, err);
997 state = rb_entry(node, struct extent_state, rb_node);
999 last_start = state->start;
1000 last_end = state->end;
1003 * | ---- desired range ---- |
1006 * Just lock what we found and keep going
1008 if (state->start == start && state->end <= end) {
1009 set_state_bits(tree, state, &bits);
1010 cache_state(state, cached_state);
1011 state = clear_state_bit(tree, state, &clear_bits, 0);
1012 if (last_end == (u64)-1)
1014 start = last_end + 1;
1015 if (start < end && state && state->start == start &&
1022 * | ---- desired range ---- |
1025 * | ------------- state -------------- |
1027 * We need to split the extent we found, and may flip bits on
1030 * If the extent we found extends past our
1031 * range, we just split and search again. It'll get split
1032 * again the next time though.
1034 * If the extent we found is inside our range, we set the
1035 * desired bit on it.
1037 if (state->start < start) {
1038 prealloc = alloc_extent_state_atomic(prealloc);
1043 err = split_state(tree, state, prealloc, start);
1045 extent_io_tree_panic(tree, err);
1049 if (state->end <= end) {
1050 set_state_bits(tree, state, &bits);
1051 cache_state(state, cached_state);
1052 state = clear_state_bit(tree, state, &clear_bits, 0);
1053 if (last_end == (u64)-1)
1055 start = last_end + 1;
1056 if (start < end && state && state->start == start &&
1063 * | ---- desired range ---- |
1064 * | state | or | state |
1066 * There's a hole, we need to insert something in it and
1067 * ignore the extent we found.
1069 if (state->start > start) {
1071 if (end < last_start)
1074 this_end = last_start - 1;
1076 prealloc = alloc_extent_state_atomic(prealloc);
1083 * Avoid to free 'prealloc' if it can be merged with
1086 err = insert_state(tree, prealloc, start, this_end,
1089 extent_io_tree_panic(tree, err);
1090 cache_state(prealloc, cached_state);
1092 start = this_end + 1;
1096 * | ---- desired range ---- |
1098 * We need to split the extent, and set the bit
1101 if (state->start <= end && state->end > end) {
1102 prealloc = alloc_extent_state_atomic(prealloc);
1108 err = split_state(tree, state, prealloc, end + 1);
1110 extent_io_tree_panic(tree, err);
1112 set_state_bits(tree, prealloc, &bits);
1113 cache_state(prealloc, cached_state);
1114 clear_state_bit(tree, prealloc, &clear_bits, 0);
1122 spin_unlock(&tree->lock);
1124 free_extent_state(prealloc);
1131 spin_unlock(&tree->lock);
1132 if (mask & __GFP_WAIT)
1137 /* wrappers around set/clear extent bit */
1138 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1141 return set_extent_bit(tree, start, end, EXTENT_DIRTY, NULL,
1145 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1146 int bits, gfp_t mask)
1148 return set_extent_bit(tree, start, end, bits, NULL,
1152 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1153 int bits, gfp_t mask)
1155 return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask);
1158 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
1159 struct extent_state **cached_state, gfp_t mask)
1161 return set_extent_bit(tree, start, end,
1162 EXTENT_DELALLOC | EXTENT_UPTODATE,
1163 NULL, cached_state, mask);
1166 int set_extent_defrag(struct extent_io_tree *tree, u64 start, u64 end,
1167 struct extent_state **cached_state, gfp_t mask)
1169 return set_extent_bit(tree, start, end,
1170 EXTENT_DELALLOC | EXTENT_UPTODATE | EXTENT_DEFRAG,
1171 NULL, cached_state, mask);
1174 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1177 return clear_extent_bit(tree, start, end,
1178 EXTENT_DIRTY | EXTENT_DELALLOC |
1179 EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask);
1182 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
1185 return set_extent_bit(tree, start, end, EXTENT_NEW, NULL,
1189 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
1190 struct extent_state **cached_state, gfp_t mask)
1192 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, NULL,
1193 cached_state, mask);
1196 int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
1197 struct extent_state **cached_state, gfp_t mask)
1199 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0,
1200 cached_state, mask);
1204 * either insert or lock state struct between start and end use mask to tell
1205 * us if waiting is desired.
1207 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1208 int bits, struct extent_state **cached_state)
1213 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED | bits,
1214 EXTENT_LOCKED, &failed_start,
1215 cached_state, GFP_NOFS);
1216 if (err == -EEXIST) {
1217 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1218 start = failed_start;
1221 WARN_ON(start > end);
1226 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1228 return lock_extent_bits(tree, start, end, 0, NULL);
1231 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1236 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1237 &failed_start, NULL, GFP_NOFS);
1238 if (err == -EEXIST) {
1239 if (failed_start > start)
1240 clear_extent_bit(tree, start, failed_start - 1,
1241 EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS);
1247 int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end,
1248 struct extent_state **cached, gfp_t mask)
1250 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached,
1254 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1256 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL,
1260 int extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
1262 unsigned long index = start >> PAGE_CACHE_SHIFT;
1263 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1266 while (index <= end_index) {
1267 page = find_get_page(inode->i_mapping, index);
1268 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1269 clear_page_dirty_for_io(page);
1270 page_cache_release(page);
1276 int extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
1278 unsigned long index = start >> PAGE_CACHE_SHIFT;
1279 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1282 while (index <= end_index) {
1283 page = find_get_page(inode->i_mapping, index);
1284 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1285 account_page_redirty(page);
1286 __set_page_dirty_nobuffers(page);
1287 page_cache_release(page);
1294 * helper function to set both pages and extents in the tree writeback
1296 static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
1298 unsigned long index = start >> PAGE_CACHE_SHIFT;
1299 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1302 while (index <= end_index) {
1303 page = find_get_page(tree->mapping, index);
1304 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1305 set_page_writeback(page);
1306 page_cache_release(page);
1312 /* find the first state struct with 'bits' set after 'start', and
1313 * return it. tree->lock must be held. NULL will returned if
1314 * nothing was found after 'start'
1316 struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree,
1317 u64 start, int bits)
1319 struct rb_node *node;
1320 struct extent_state *state;
1323 * this search will find all the extents that end after
1326 node = tree_search(tree, start);
1331 state = rb_entry(node, struct extent_state, rb_node);
1332 if (state->end >= start && (state->state & bits))
1335 node = rb_next(node);
1344 * find the first offset in the io tree with 'bits' set. zero is
1345 * returned if we find something, and *start_ret and *end_ret are
1346 * set to reflect the state struct that was found.
1348 * If nothing was found, 1 is returned. If found something, return 0.
1350 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1351 u64 *start_ret, u64 *end_ret, int bits,
1352 struct extent_state **cached_state)
1354 struct extent_state *state;
1358 spin_lock(&tree->lock);
1359 if (cached_state && *cached_state) {
1360 state = *cached_state;
1361 if (state->end == start - 1 && state->tree) {
1362 n = rb_next(&state->rb_node);
1364 state = rb_entry(n, struct extent_state,
1366 if (state->state & bits)
1370 free_extent_state(*cached_state);
1371 *cached_state = NULL;
1374 free_extent_state(*cached_state);
1375 *cached_state = NULL;
1378 state = find_first_extent_bit_state(tree, start, bits);
1381 cache_state(state, cached_state);
1382 *start_ret = state->start;
1383 *end_ret = state->end;
1387 spin_unlock(&tree->lock);
1392 * find a contiguous range of bytes in the file marked as delalloc, not
1393 * more than 'max_bytes'. start and end are used to return the range,
1395 * 1 is returned if we find something, 0 if nothing was in the tree
1397 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1398 u64 *start, u64 *end, u64 max_bytes,
1399 struct extent_state **cached_state)
1401 struct rb_node *node;
1402 struct extent_state *state;
1403 u64 cur_start = *start;
1405 u64 total_bytes = 0;
1407 spin_lock(&tree->lock);
1410 * this search will find all the extents that end after
1413 node = tree_search(tree, cur_start);
1421 state = rb_entry(node, struct extent_state, rb_node);
1422 if (found && (state->start != cur_start ||
1423 (state->state & EXTENT_BOUNDARY))) {
1426 if (!(state->state & EXTENT_DELALLOC)) {
1432 *start = state->start;
1433 *cached_state = state;
1434 atomic_inc(&state->refs);
1438 cur_start = state->end + 1;
1439 node = rb_next(node);
1442 total_bytes += state->end - state->start + 1;
1443 if (total_bytes >= max_bytes)
1447 spin_unlock(&tree->lock);
1451 static noinline void __unlock_for_delalloc(struct inode *inode,
1452 struct page *locked_page,
1456 struct page *pages[16];
1457 unsigned long index = start >> PAGE_CACHE_SHIFT;
1458 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1459 unsigned long nr_pages = end_index - index + 1;
1462 if (index == locked_page->index && end_index == index)
1465 while (nr_pages > 0) {
1466 ret = find_get_pages_contig(inode->i_mapping, index,
1467 min_t(unsigned long, nr_pages,
1468 ARRAY_SIZE(pages)), pages);
1469 for (i = 0; i < ret; i++) {
1470 if (pages[i] != locked_page)
1471 unlock_page(pages[i]);
1472 page_cache_release(pages[i]);
1480 static noinline int lock_delalloc_pages(struct inode *inode,
1481 struct page *locked_page,
1485 unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT;
1486 unsigned long start_index = index;
1487 unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT;
1488 unsigned long pages_locked = 0;
1489 struct page *pages[16];
1490 unsigned long nrpages;
1494 /* the caller is responsible for locking the start index */
1495 if (index == locked_page->index && index == end_index)
1498 /* skip the page at the start index */
1499 nrpages = end_index - index + 1;
1500 while (nrpages > 0) {
1501 ret = find_get_pages_contig(inode->i_mapping, index,
1502 min_t(unsigned long,
1503 nrpages, ARRAY_SIZE(pages)), pages);
1508 /* now we have an array of pages, lock them all */
1509 for (i = 0; i < ret; i++) {
1511 * the caller is taking responsibility for
1514 if (pages[i] != locked_page) {
1515 lock_page(pages[i]);
1516 if (!PageDirty(pages[i]) ||
1517 pages[i]->mapping != inode->i_mapping) {
1519 unlock_page(pages[i]);
1520 page_cache_release(pages[i]);
1524 page_cache_release(pages[i]);
1533 if (ret && pages_locked) {
1534 __unlock_for_delalloc(inode, locked_page,
1536 ((u64)(start_index + pages_locked - 1)) <<
1543 * find a contiguous range of bytes in the file marked as delalloc, not
1544 * more than 'max_bytes'. start and end are used to return the range,
1546 * 1 is returned if we find something, 0 if nothing was in the tree
1548 static noinline u64 find_lock_delalloc_range(struct inode *inode,
1549 struct extent_io_tree *tree,
1550 struct page *locked_page,
1551 u64 *start, u64 *end,
1557 struct extent_state *cached_state = NULL;
1562 /* step one, find a bunch of delalloc bytes starting at start */
1563 delalloc_start = *start;
1565 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1566 max_bytes, &cached_state);
1567 if (!found || delalloc_end <= *start) {
1568 *start = delalloc_start;
1569 *end = delalloc_end;
1570 free_extent_state(cached_state);
1575 * start comes from the offset of locked_page. We have to lock
1576 * pages in order, so we can't process delalloc bytes before
1579 if (delalloc_start < *start)
1580 delalloc_start = *start;
1583 * make sure to limit the number of pages we try to lock down
1586 if (delalloc_end + 1 - delalloc_start > max_bytes && loops)
1587 delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1;
1589 /* step two, lock all the pages after the page that has start */
1590 ret = lock_delalloc_pages(inode, locked_page,
1591 delalloc_start, delalloc_end);
1592 if (ret == -EAGAIN) {
1593 /* some of the pages are gone, lets avoid looping by
1594 * shortening the size of the delalloc range we're searching
1596 free_extent_state(cached_state);
1598 unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1);
1599 max_bytes = PAGE_CACHE_SIZE - offset;
1607 BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */
1609 /* step three, lock the state bits for the whole range */
1610 lock_extent_bits(tree, delalloc_start, delalloc_end, 0, &cached_state);
1612 /* then test to make sure it is all still delalloc */
1613 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1614 EXTENT_DELALLOC, 1, cached_state);
1616 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1617 &cached_state, GFP_NOFS);
1618 __unlock_for_delalloc(inode, locked_page,
1619 delalloc_start, delalloc_end);
1623 free_extent_state(cached_state);
1624 *start = delalloc_start;
1625 *end = delalloc_end;
1630 int extent_clear_unlock_delalloc(struct inode *inode,
1631 struct extent_io_tree *tree,
1632 u64 start, u64 end, struct page *locked_page,
1636 struct page *pages[16];
1637 unsigned long index = start >> PAGE_CACHE_SHIFT;
1638 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1639 unsigned long nr_pages = end_index - index + 1;
1643 if (op & EXTENT_CLEAR_UNLOCK)
1644 clear_bits |= EXTENT_LOCKED;
1645 if (op & EXTENT_CLEAR_DIRTY)
1646 clear_bits |= EXTENT_DIRTY;
1648 if (op & EXTENT_CLEAR_DELALLOC)
1649 clear_bits |= EXTENT_DELALLOC;
1651 clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS);
1652 if (!(op & (EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
1653 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK |
1654 EXTENT_SET_PRIVATE2)))
1657 while (nr_pages > 0) {
1658 ret = find_get_pages_contig(inode->i_mapping, index,
1659 min_t(unsigned long,
1660 nr_pages, ARRAY_SIZE(pages)), pages);
1661 for (i = 0; i < ret; i++) {
1663 if (op & EXTENT_SET_PRIVATE2)
1664 SetPagePrivate2(pages[i]);
1666 if (pages[i] == locked_page) {
1667 page_cache_release(pages[i]);
1670 if (op & EXTENT_CLEAR_DIRTY)
1671 clear_page_dirty_for_io(pages[i]);
1672 if (op & EXTENT_SET_WRITEBACK)
1673 set_page_writeback(pages[i]);
1674 if (op & EXTENT_END_WRITEBACK)
1675 end_page_writeback(pages[i]);
1676 if (op & EXTENT_CLEAR_UNLOCK_PAGE)
1677 unlock_page(pages[i]);
1678 page_cache_release(pages[i]);
1688 * count the number of bytes in the tree that have a given bit(s)
1689 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1690 * cached. The total number found is returned.
1692 u64 count_range_bits(struct extent_io_tree *tree,
1693 u64 *start, u64 search_end, u64 max_bytes,
1694 unsigned long bits, int contig)
1696 struct rb_node *node;
1697 struct extent_state *state;
1698 u64 cur_start = *start;
1699 u64 total_bytes = 0;
1703 if (search_end <= cur_start) {
1708 spin_lock(&tree->lock);
1709 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1710 total_bytes = tree->dirty_bytes;
1714 * this search will find all the extents that end after
1717 node = tree_search(tree, cur_start);
1722 state = rb_entry(node, struct extent_state, rb_node);
1723 if (state->start > search_end)
1725 if (contig && found && state->start > last + 1)
1727 if (state->end >= cur_start && (state->state & bits) == bits) {
1728 total_bytes += min(search_end, state->end) + 1 -
1729 max(cur_start, state->start);
1730 if (total_bytes >= max_bytes)
1733 *start = max(cur_start, state->start);
1737 } else if (contig && found) {
1740 node = rb_next(node);
1745 spin_unlock(&tree->lock);
1750 * set the private field for a given byte offset in the tree. If there isn't
1751 * an extent_state there already, this does nothing.
1753 int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1755 struct rb_node *node;
1756 struct extent_state *state;
1759 spin_lock(&tree->lock);
1761 * this search will find all the extents that end after
1764 node = tree_search(tree, start);
1769 state = rb_entry(node, struct extent_state, rb_node);
1770 if (state->start != start) {
1774 state->private = private;
1776 spin_unlock(&tree->lock);
1780 void extent_cache_csums_dio(struct extent_io_tree *tree, u64 start, u32 csums[],
1783 struct rb_node *node;
1784 struct extent_state *state;
1786 spin_lock(&tree->lock);
1788 * this search will find all the extents that end after
1791 node = tree_search(tree, start);
1794 state = rb_entry(node, struct extent_state, rb_node);
1795 BUG_ON(state->start != start);
1798 state->private = *csums++;
1800 state = next_state(state);
1802 spin_unlock(&tree->lock);
1805 static inline u64 __btrfs_get_bio_offset(struct bio *bio, int bio_index)
1807 struct bio_vec *bvec = bio->bi_io_vec + bio_index;
1809 return page_offset(bvec->bv_page) + bvec->bv_offset;
1812 void extent_cache_csums(struct extent_io_tree *tree, struct bio *bio, int bio_index,
1813 u32 csums[], int count)
1815 struct rb_node *node;
1816 struct extent_state *state = NULL;
1819 spin_lock(&tree->lock);
1821 start = __btrfs_get_bio_offset(bio, bio_index);
1822 if (state == NULL || state->start != start) {
1823 node = tree_search(tree, start);
1826 state = rb_entry(node, struct extent_state, rb_node);
1827 BUG_ON(state->start != start);
1829 state->private = *csums++;
1833 state = next_state(state);
1835 spin_unlock(&tree->lock);
1838 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1840 struct rb_node *node;
1841 struct extent_state *state;
1844 spin_lock(&tree->lock);
1846 * this search will find all the extents that end after
1849 node = tree_search(tree, start);
1854 state = rb_entry(node, struct extent_state, rb_node);
1855 if (state->start != start) {
1859 *private = state->private;
1861 spin_unlock(&tree->lock);
1866 * searches a range in the state tree for a given mask.
1867 * If 'filled' == 1, this returns 1 only if every extent in the tree
1868 * has the bits set. Otherwise, 1 is returned if any bit in the
1869 * range is found set.
1871 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1872 int bits, int filled, struct extent_state *cached)
1874 struct extent_state *state = NULL;
1875 struct rb_node *node;
1878 spin_lock(&tree->lock);
1879 if (cached && cached->tree && cached->start <= start &&
1880 cached->end > start)
1881 node = &cached->rb_node;
1883 node = tree_search(tree, start);
1884 while (node && start <= end) {
1885 state = rb_entry(node, struct extent_state, rb_node);
1887 if (filled && state->start > start) {
1892 if (state->start > end)
1895 if (state->state & bits) {
1899 } else if (filled) {
1904 if (state->end == (u64)-1)
1907 start = state->end + 1;
1910 node = rb_next(node);
1917 spin_unlock(&tree->lock);
1922 * helper function to set a given page up to date if all the
1923 * extents in the tree for that page are up to date
1925 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
1927 u64 start = page_offset(page);
1928 u64 end = start + PAGE_CACHE_SIZE - 1;
1929 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1930 SetPageUptodate(page);
1934 * helper function to unlock a page if all the extents in the tree
1935 * for that page are unlocked
1937 static void check_page_locked(struct extent_io_tree *tree, struct page *page)
1939 u64 start = page_offset(page);
1940 u64 end = start + PAGE_CACHE_SIZE - 1;
1941 if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL))
1946 * helper function to end page writeback if all the extents
1947 * in the tree for that page are done with writeback
1949 static void check_page_writeback(struct extent_io_tree *tree,
1952 end_page_writeback(page);
1956 * When IO fails, either with EIO or csum verification fails, we
1957 * try other mirrors that might have a good copy of the data. This
1958 * io_failure_record is used to record state as we go through all the
1959 * mirrors. If another mirror has good data, the page is set up to date
1960 * and things continue. If a good mirror can't be found, the original
1961 * bio end_io callback is called to indicate things have failed.
1963 struct io_failure_record {
1968 unsigned long bio_flags;
1974 static int free_io_failure(struct inode *inode, struct io_failure_record *rec,
1979 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1981 set_state_private(failure_tree, rec->start, 0);
1982 ret = clear_extent_bits(failure_tree, rec->start,
1983 rec->start + rec->len - 1,
1984 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1988 ret = clear_extent_bits(&BTRFS_I(inode)->io_tree, rec->start,
1989 rec->start + rec->len - 1,
1990 EXTENT_DAMAGED, GFP_NOFS);
1998 static void repair_io_failure_callback(struct bio *bio, int err)
2000 complete(bio->bi_private);
2004 * this bypasses the standard btrfs submit functions deliberately, as
2005 * the standard behavior is to write all copies in a raid setup. here we only
2006 * want to write the one bad copy. so we do the mapping for ourselves and issue
2007 * submit_bio directly.
2008 * to avoid any synchronization issues, wait for the data after writing, which
2009 * actually prevents the read that triggered the error from finishing.
2010 * currently, there can be no more than two copies of every data bit. thus,
2011 * exactly one rewrite is required.
2013 int repair_io_failure(struct btrfs_fs_info *fs_info, u64 start,
2014 u64 length, u64 logical, struct page *page,
2018 struct btrfs_device *dev;
2019 DECLARE_COMPLETION_ONSTACK(compl);
2022 struct btrfs_bio *bbio = NULL;
2023 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
2026 BUG_ON(!mirror_num);
2028 /* we can't repair anything in raid56 yet */
2029 if (btrfs_is_parity_mirror(map_tree, logical, length, mirror_num))
2032 bio = bio_alloc(GFP_NOFS, 1);
2035 bio->bi_private = &compl;
2036 bio->bi_end_io = repair_io_failure_callback;
2038 map_length = length;
2040 ret = btrfs_map_block(fs_info, WRITE, logical,
2041 &map_length, &bbio, mirror_num);
2046 BUG_ON(mirror_num != bbio->mirror_num);
2047 sector = bbio->stripes[mirror_num-1].physical >> 9;
2048 bio->bi_sector = sector;
2049 dev = bbio->stripes[mirror_num-1].dev;
2051 if (!dev || !dev->bdev || !dev->writeable) {
2055 bio->bi_bdev = dev->bdev;
2056 bio_add_page(bio, page, length, start - page_offset(page));
2057 btrfsic_submit_bio(WRITE_SYNC, bio);
2058 wait_for_completion(&compl);
2060 if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) {
2061 /* try to remap that extent elsewhere? */
2063 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
2067 printk_ratelimited_in_rcu(KERN_INFO "btrfs read error corrected: ino %lu off %llu "
2068 "(dev %s sector %llu)\n", page->mapping->host->i_ino,
2069 start, rcu_str_deref(dev->name), sector);
2075 int repair_eb_io_failure(struct btrfs_root *root, struct extent_buffer *eb,
2078 u64 start = eb->start;
2079 unsigned long i, num_pages = num_extent_pages(eb->start, eb->len);
2082 for (i = 0; i < num_pages; i++) {
2083 struct page *p = extent_buffer_page(eb, i);
2084 ret = repair_io_failure(root->fs_info, start, PAGE_CACHE_SIZE,
2085 start, p, mirror_num);
2088 start += PAGE_CACHE_SIZE;
2095 * each time an IO finishes, we do a fast check in the IO failure tree
2096 * to see if we need to process or clean up an io_failure_record
2098 static int clean_io_failure(u64 start, struct page *page)
2101 u64 private_failure;
2102 struct io_failure_record *failrec;
2103 struct btrfs_fs_info *fs_info;
2104 struct extent_state *state;
2108 struct inode *inode = page->mapping->host;
2111 ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
2112 (u64)-1, 1, EXTENT_DIRTY, 0);
2116 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, start,
2121 failrec = (struct io_failure_record *)(unsigned long) private_failure;
2122 BUG_ON(!failrec->this_mirror);
2124 if (failrec->in_validation) {
2125 /* there was no real error, just free the record */
2126 pr_debug("clean_io_failure: freeing dummy error at %llu\n",
2132 spin_lock(&BTRFS_I(inode)->io_tree.lock);
2133 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
2136 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
2138 if (state && state->start == failrec->start) {
2139 fs_info = BTRFS_I(inode)->root->fs_info;
2140 num_copies = btrfs_num_copies(fs_info, failrec->logical,
2142 if (num_copies > 1) {
2143 ret = repair_io_failure(fs_info, start, failrec->len,
2144 failrec->logical, page,
2145 failrec->failed_mirror);
2153 ret = free_io_failure(inode, failrec, did_repair);
2159 * this is a generic handler for readpage errors (default
2160 * readpage_io_failed_hook). if other copies exist, read those and write back
2161 * good data to the failed position. does not investigate in remapping the
2162 * failed extent elsewhere, hoping the device will be smart enough to do this as
2166 static int bio_readpage_error(struct bio *failed_bio, struct page *page,
2167 u64 start, u64 end, int failed_mirror,
2168 struct extent_state *state)
2170 struct io_failure_record *failrec = NULL;
2172 struct extent_map *em;
2173 struct inode *inode = page->mapping->host;
2174 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2175 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2176 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2183 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
2185 ret = get_state_private(failure_tree, start, &private);
2187 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2190 failrec->start = start;
2191 failrec->len = end - start + 1;
2192 failrec->this_mirror = 0;
2193 failrec->bio_flags = 0;
2194 failrec->in_validation = 0;
2196 read_lock(&em_tree->lock);
2197 em = lookup_extent_mapping(em_tree, start, failrec->len);
2199 read_unlock(&em_tree->lock);
2204 if (em->start > start || em->start + em->len < start) {
2205 free_extent_map(em);
2208 read_unlock(&em_tree->lock);
2214 logical = start - em->start;
2215 logical = em->block_start + logical;
2216 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2217 logical = em->block_start;
2218 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2219 extent_set_compress_type(&failrec->bio_flags,
2222 pr_debug("bio_readpage_error: (new) logical=%llu, start=%llu, "
2223 "len=%llu\n", logical, start, failrec->len);
2224 failrec->logical = logical;
2225 free_extent_map(em);
2227 /* set the bits in the private failure tree */
2228 ret = set_extent_bits(failure_tree, start, end,
2229 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
2231 ret = set_state_private(failure_tree, start,
2232 (u64)(unsigned long)failrec);
2233 /* set the bits in the inode's tree */
2235 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED,
2242 failrec = (struct io_failure_record *)(unsigned long)private;
2243 pr_debug("bio_readpage_error: (found) logical=%llu, "
2244 "start=%llu, len=%llu, validation=%d\n",
2245 failrec->logical, failrec->start, failrec->len,
2246 failrec->in_validation);
2248 * when data can be on disk more than twice, add to failrec here
2249 * (e.g. with a list for failed_mirror) to make
2250 * clean_io_failure() clean all those errors at once.
2253 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
2254 failrec->logical, failrec->len);
2255 if (num_copies == 1) {
2257 * we only have a single copy of the data, so don't bother with
2258 * all the retry and error correction code that follows. no
2259 * matter what the error is, it is very likely to persist.
2261 pr_debug("bio_readpage_error: cannot repair, num_copies == 1. "
2262 "state=%p, num_copies=%d, next_mirror %d, "
2263 "failed_mirror %d\n", state, num_copies,
2264 failrec->this_mirror, failed_mirror);
2265 free_io_failure(inode, failrec, 0);
2270 spin_lock(&tree->lock);
2271 state = find_first_extent_bit_state(tree, failrec->start,
2273 if (state && state->start != failrec->start)
2275 spin_unlock(&tree->lock);
2279 * there are two premises:
2280 * a) deliver good data to the caller
2281 * b) correct the bad sectors on disk
2283 if (failed_bio->bi_vcnt > 1) {
2285 * to fulfill b), we need to know the exact failing sectors, as
2286 * we don't want to rewrite any more than the failed ones. thus,
2287 * we need separate read requests for the failed bio
2289 * if the following BUG_ON triggers, our validation request got
2290 * merged. we need separate requests for our algorithm to work.
2292 BUG_ON(failrec->in_validation);
2293 failrec->in_validation = 1;
2294 failrec->this_mirror = failed_mirror;
2295 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
2298 * we're ready to fulfill a) and b) alongside. get a good copy
2299 * of the failed sector and if we succeed, we have setup
2300 * everything for repair_io_failure to do the rest for us.
2302 if (failrec->in_validation) {
2303 BUG_ON(failrec->this_mirror != failed_mirror);
2304 failrec->in_validation = 0;
2305 failrec->this_mirror = 0;
2307 failrec->failed_mirror = failed_mirror;
2308 failrec->this_mirror++;
2309 if (failrec->this_mirror == failed_mirror)
2310 failrec->this_mirror++;
2311 read_mode = READ_SYNC;
2314 if (!state || failrec->this_mirror > num_copies) {
2315 pr_debug("bio_readpage_error: (fail) state=%p, num_copies=%d, "
2316 "next_mirror %d, failed_mirror %d\n", state,
2317 num_copies, failrec->this_mirror, failed_mirror);
2318 free_io_failure(inode, failrec, 0);
2322 bio = bio_alloc(GFP_NOFS, 1);
2324 free_io_failure(inode, failrec, 0);
2327 bio->bi_private = state;
2328 bio->bi_end_io = failed_bio->bi_end_io;
2329 bio->bi_sector = failrec->logical >> 9;
2330 bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
2333 bio_add_page(bio, page, failrec->len, start - page_offset(page));
2335 pr_debug("bio_readpage_error: submitting new read[%#x] to "
2336 "this_mirror=%d, num_copies=%d, in_validation=%d\n", read_mode,
2337 failrec->this_mirror, num_copies, failrec->in_validation);
2339 ret = tree->ops->submit_bio_hook(inode, read_mode, bio,
2340 failrec->this_mirror,
2341 failrec->bio_flags, 0);
2345 /* lots and lots of room for performance fixes in the end_bio funcs */
2347 int end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2349 int uptodate = (err == 0);
2350 struct extent_io_tree *tree;
2353 tree = &BTRFS_I(page->mapping->host)->io_tree;
2355 if (tree->ops && tree->ops->writepage_end_io_hook) {
2356 ret = tree->ops->writepage_end_io_hook(page, start,
2357 end, NULL, uptodate);
2363 ClearPageUptodate(page);
2370 * after a writepage IO is done, we need to:
2371 * clear the uptodate bits on error
2372 * clear the writeback bits in the extent tree for this IO
2373 * end_page_writeback if the page has no more pending IO
2375 * Scheduling is not allowed, so the extent state tree is expected
2376 * to have one and only one object corresponding to this IO.
2378 static void end_bio_extent_writepage(struct bio *bio, int err)
2380 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2381 struct extent_io_tree *tree;
2387 struct page *page = bvec->bv_page;
2388 tree = &BTRFS_I(page->mapping->host)->io_tree;
2390 start = page_offset(page) + bvec->bv_offset;
2391 end = start + bvec->bv_len - 1;
2393 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
2398 if (--bvec >= bio->bi_io_vec)
2399 prefetchw(&bvec->bv_page->flags);
2401 if (end_extent_writepage(page, err, start, end))
2405 end_page_writeback(page);
2407 check_page_writeback(tree, page);
2408 } while (bvec >= bio->bi_io_vec);
2414 * after a readpage IO is done, we need to:
2415 * clear the uptodate bits on error
2416 * set the uptodate bits if things worked
2417 * set the page up to date if all extents in the tree are uptodate
2418 * clear the lock bit in the extent tree
2419 * unlock the page if there are no other extents locked for it
2421 * Scheduling is not allowed, so the extent state tree is expected
2422 * to have one and only one object corresponding to this IO.
2424 static void end_bio_extent_readpage(struct bio *bio, int err)
2426 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
2427 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
2428 struct bio_vec *bvec = bio->bi_io_vec;
2429 struct extent_io_tree *tree;
2440 struct page *page = bvec->bv_page;
2441 struct extent_state *cached = NULL;
2442 struct extent_state *state;
2444 pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, "
2445 "mirror=%ld\n", (u64)bio->bi_sector, err,
2446 (long int)bio->bi_bdev);
2447 tree = &BTRFS_I(page->mapping->host)->io_tree;
2449 start = page_offset(page) + bvec->bv_offset;
2450 end = start + bvec->bv_len - 1;
2452 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
2457 if (++bvec <= bvec_end)
2458 prefetchw(&bvec->bv_page->flags);
2460 spin_lock(&tree->lock);
2461 state = find_first_extent_bit_state(tree, start, EXTENT_LOCKED);
2462 if (state && state->start == start) {
2464 * take a reference on the state, unlock will drop
2467 cache_state(state, &cached);
2469 spin_unlock(&tree->lock);
2471 mirror = (int)(unsigned long)bio->bi_bdev;
2472 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
2473 ret = tree->ops->readpage_end_io_hook(page, start, end,
2478 clean_io_failure(start, page);
2481 if (!uptodate && tree->ops && tree->ops->readpage_io_failed_hook) {
2482 ret = tree->ops->readpage_io_failed_hook(page, mirror);
2484 test_bit(BIO_UPTODATE, &bio->bi_flags))
2486 } else if (!uptodate) {
2488 * The generic bio_readpage_error handles errors the
2489 * following way: If possible, new read requests are
2490 * created and submitted and will end up in
2491 * end_bio_extent_readpage as well (if we're lucky, not
2492 * in the !uptodate case). In that case it returns 0 and
2493 * we just go on with the next page in our bio. If it
2494 * can't handle the error it will return -EIO and we
2495 * remain responsible for that page.
2497 ret = bio_readpage_error(bio, page, start, end, mirror, NULL);
2500 test_bit(BIO_UPTODATE, &bio->bi_flags);
2503 uncache_state(&cached);
2508 if (uptodate && tree->track_uptodate) {
2509 set_extent_uptodate(tree, start, end, &cached,
2512 unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC);
2516 SetPageUptodate(page);
2518 ClearPageUptodate(page);
2524 check_page_uptodate(tree, page);
2526 ClearPageUptodate(page);
2529 check_page_locked(tree, page);
2531 } while (bvec <= bvec_end);
2537 btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
2542 bio = bio_alloc(gfp_flags, nr_vecs);
2544 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
2545 while (!bio && (nr_vecs /= 2))
2546 bio = bio_alloc(gfp_flags, nr_vecs);
2551 bio->bi_bdev = bdev;
2552 bio->bi_sector = first_sector;
2557 static int __must_check submit_one_bio(int rw, struct bio *bio,
2558 int mirror_num, unsigned long bio_flags)
2561 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2562 struct page *page = bvec->bv_page;
2563 struct extent_io_tree *tree = bio->bi_private;
2566 start = page_offset(page) + bvec->bv_offset;
2568 bio->bi_private = NULL;
2572 if (tree->ops && tree->ops->submit_bio_hook)
2573 ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
2574 mirror_num, bio_flags, start);
2576 btrfsic_submit_bio(rw, bio);
2578 if (bio_flagged(bio, BIO_EOPNOTSUPP))
2584 static int merge_bio(int rw, struct extent_io_tree *tree, struct page *page,
2585 unsigned long offset, size_t size, struct bio *bio,
2586 unsigned long bio_flags)
2589 if (tree->ops && tree->ops->merge_bio_hook)
2590 ret = tree->ops->merge_bio_hook(rw, page, offset, size, bio,
2597 static int submit_extent_page(int rw, struct extent_io_tree *tree,
2598 struct page *page, sector_t sector,
2599 size_t size, unsigned long offset,
2600 struct block_device *bdev,
2601 struct bio **bio_ret,
2602 unsigned long max_pages,
2603 bio_end_io_t end_io_func,
2605 unsigned long prev_bio_flags,
2606 unsigned long bio_flags)
2612 int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED;
2613 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
2614 size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE);
2616 if (bio_ret && *bio_ret) {
2619 contig = bio->bi_sector == sector;
2621 contig = bio->bi_sector + (bio->bi_size >> 9) ==
2624 if (prev_bio_flags != bio_flags || !contig ||
2625 merge_bio(rw, tree, page, offset, page_size, bio, bio_flags) ||
2626 bio_add_page(bio, page, page_size, offset) < page_size) {
2627 ret = submit_one_bio(rw, bio, mirror_num,
2636 if (this_compressed)
2639 nr = bio_get_nr_vecs(bdev);
2641 bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
2645 bio_add_page(bio, page, page_size, offset);
2646 bio->bi_end_io = end_io_func;
2647 bio->bi_private = tree;
2652 ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
2657 void attach_extent_buffer_page(struct extent_buffer *eb, struct page *page)
2659 if (!PagePrivate(page)) {
2660 SetPagePrivate(page);
2661 page_cache_get(page);
2662 set_page_private(page, (unsigned long)eb);
2664 WARN_ON(page->private != (unsigned long)eb);
2668 void set_page_extent_mapped(struct page *page)
2670 if (!PagePrivate(page)) {
2671 SetPagePrivate(page);
2672 page_cache_get(page);
2673 set_page_private(page, EXTENT_PAGE_PRIVATE);
2678 * basic readpage implementation. Locked extent state structs are inserted
2679 * into the tree that are removed when the IO is done (by the end_io
2681 * XXX JDM: This needs looking at to ensure proper page locking
2683 static int __extent_read_full_page(struct extent_io_tree *tree,
2685 get_extent_t *get_extent,
2686 struct bio **bio, int mirror_num,
2687 unsigned long *bio_flags)
2689 struct inode *inode = page->mapping->host;
2690 u64 start = page_offset(page);
2691 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2695 u64 last_byte = i_size_read(inode);
2699 struct extent_map *em;
2700 struct block_device *bdev;
2701 struct btrfs_ordered_extent *ordered;
2704 size_t pg_offset = 0;
2706 size_t disk_io_size;
2707 size_t blocksize = inode->i_sb->s_blocksize;
2708 unsigned long this_bio_flag = 0;
2710 set_page_extent_mapped(page);
2712 if (!PageUptodate(page)) {
2713 if (cleancache_get_page(page) == 0) {
2714 BUG_ON(blocksize != PAGE_SIZE);
2721 lock_extent(tree, start, end);
2722 ordered = btrfs_lookup_ordered_extent(inode, start);
2725 unlock_extent(tree, start, end);
2726 btrfs_start_ordered_extent(inode, ordered, 1);
2727 btrfs_put_ordered_extent(ordered);
2730 if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
2732 size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
2735 iosize = PAGE_CACHE_SIZE - zero_offset;
2736 userpage = kmap_atomic(page);
2737 memset(userpage + zero_offset, 0, iosize);
2738 flush_dcache_page(page);
2739 kunmap_atomic(userpage);
2742 while (cur <= end) {
2743 unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
2745 if (cur >= last_byte) {
2747 struct extent_state *cached = NULL;
2749 iosize = PAGE_CACHE_SIZE - pg_offset;
2750 userpage = kmap_atomic(page);
2751 memset(userpage + pg_offset, 0, iosize);
2752 flush_dcache_page(page);
2753 kunmap_atomic(userpage);
2754 set_extent_uptodate(tree, cur, cur + iosize - 1,
2756 unlock_extent_cached(tree, cur, cur + iosize - 1,
2760 em = get_extent(inode, page, pg_offset, cur,
2762 if (IS_ERR_OR_NULL(em)) {
2764 unlock_extent(tree, cur, end);
2767 extent_offset = cur - em->start;
2768 BUG_ON(extent_map_end(em) <= cur);
2771 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2772 this_bio_flag = EXTENT_BIO_COMPRESSED;
2773 extent_set_compress_type(&this_bio_flag,
2777 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2778 cur_end = min(extent_map_end(em) - 1, end);
2779 iosize = ALIGN(iosize, blocksize);
2780 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2781 disk_io_size = em->block_len;
2782 sector = em->block_start >> 9;
2784 sector = (em->block_start + extent_offset) >> 9;
2785 disk_io_size = iosize;
2788 block_start = em->block_start;
2789 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2790 block_start = EXTENT_MAP_HOLE;
2791 free_extent_map(em);
2794 /* we've found a hole, just zero and go on */
2795 if (block_start == EXTENT_MAP_HOLE) {
2797 struct extent_state *cached = NULL;
2799 userpage = kmap_atomic(page);
2800 memset(userpage + pg_offset, 0, iosize);
2801 flush_dcache_page(page);
2802 kunmap_atomic(userpage);
2804 set_extent_uptodate(tree, cur, cur + iosize - 1,
2806 unlock_extent_cached(tree, cur, cur + iosize - 1,
2809 pg_offset += iosize;
2812 /* the get_extent function already copied into the page */
2813 if (test_range_bit(tree, cur, cur_end,
2814 EXTENT_UPTODATE, 1, NULL)) {
2815 check_page_uptodate(tree, page);
2816 unlock_extent(tree, cur, cur + iosize - 1);
2818 pg_offset += iosize;
2821 /* we have an inline extent but it didn't get marked up
2822 * to date. Error out
2824 if (block_start == EXTENT_MAP_INLINE) {
2826 unlock_extent(tree, cur, cur + iosize - 1);
2828 pg_offset += iosize;
2833 ret = submit_extent_page(READ, tree, page,
2834 sector, disk_io_size, pg_offset,
2836 end_bio_extent_readpage, mirror_num,
2841 *bio_flags = this_bio_flag;
2844 unlock_extent(tree, cur, cur + iosize - 1);
2847 pg_offset += iosize;
2851 if (!PageError(page))
2852 SetPageUptodate(page);
2858 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
2859 get_extent_t *get_extent, int mirror_num)
2861 struct bio *bio = NULL;
2862 unsigned long bio_flags = 0;
2865 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
2868 ret = submit_one_bio(READ, bio, mirror_num, bio_flags);
2872 static noinline void update_nr_written(struct page *page,
2873 struct writeback_control *wbc,
2874 unsigned long nr_written)
2876 wbc->nr_to_write -= nr_written;
2877 if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
2878 wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
2879 page->mapping->writeback_index = page->index + nr_written;
2883 * the writepage semantics are similar to regular writepage. extent
2884 * records are inserted to lock ranges in the tree, and as dirty areas
2885 * are found, they are marked writeback. Then the lock bits are removed
2886 * and the end_io handler clears the writeback ranges
2888 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2891 struct inode *inode = page->mapping->host;
2892 struct extent_page_data *epd = data;
2893 struct extent_io_tree *tree = epd->tree;
2894 u64 start = page_offset(page);
2896 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2900 u64 last_byte = i_size_read(inode);
2904 struct extent_state *cached_state = NULL;
2905 struct extent_map *em;
2906 struct block_device *bdev;
2909 size_t pg_offset = 0;
2911 loff_t i_size = i_size_read(inode);
2912 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
2918 unsigned long nr_written = 0;
2919 bool fill_delalloc = true;
2921 if (wbc->sync_mode == WB_SYNC_ALL)
2922 write_flags = WRITE_SYNC;
2924 write_flags = WRITE;
2926 trace___extent_writepage(page, inode, wbc);
2928 WARN_ON(!PageLocked(page));
2930 ClearPageError(page);
2932 pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
2933 if (page->index > end_index ||
2934 (page->index == end_index && !pg_offset)) {
2935 page->mapping->a_ops->invalidatepage(page, 0);
2940 if (page->index == end_index) {
2943 userpage = kmap_atomic(page);
2944 memset(userpage + pg_offset, 0,
2945 PAGE_CACHE_SIZE - pg_offset);
2946 kunmap_atomic(userpage);
2947 flush_dcache_page(page);
2951 set_page_extent_mapped(page);
2953 if (!tree->ops || !tree->ops->fill_delalloc)
2954 fill_delalloc = false;
2956 delalloc_start = start;
2959 if (!epd->extent_locked && fill_delalloc) {
2960 u64 delalloc_to_write = 0;
2962 * make sure the wbc mapping index is at least updated
2965 update_nr_written(page, wbc, 0);
2967 while (delalloc_end < page_end) {
2968 nr_delalloc = find_lock_delalloc_range(inode, tree,
2973 if (nr_delalloc == 0) {
2974 delalloc_start = delalloc_end + 1;
2977 ret = tree->ops->fill_delalloc(inode, page,
2982 /* File system has been set read-only */
2988 * delalloc_end is already one less than the total
2989 * length, so we don't subtract one from
2992 delalloc_to_write += (delalloc_end - delalloc_start +
2995 delalloc_start = delalloc_end + 1;
2997 if (wbc->nr_to_write < delalloc_to_write) {
3000 if (delalloc_to_write < thresh * 2)
3001 thresh = delalloc_to_write;
3002 wbc->nr_to_write = min_t(u64, delalloc_to_write,
3006 /* did the fill delalloc function already unlock and start
3012 * we've unlocked the page, so we can't update
3013 * the mapping's writeback index, just update
3016 wbc->nr_to_write -= nr_written;
3020 if (tree->ops && tree->ops->writepage_start_hook) {
3021 ret = tree->ops->writepage_start_hook(page, start,
3024 /* Fixup worker will requeue */
3026 wbc->pages_skipped++;
3028 redirty_page_for_writepage(wbc, page);
3029 update_nr_written(page, wbc, nr_written);
3037 * we don't want to touch the inode after unlocking the page,
3038 * so we update the mapping writeback index now
3040 update_nr_written(page, wbc, nr_written + 1);
3043 if (last_byte <= start) {
3044 if (tree->ops && tree->ops->writepage_end_io_hook)
3045 tree->ops->writepage_end_io_hook(page, start,
3050 blocksize = inode->i_sb->s_blocksize;
3052 while (cur <= end) {
3053 if (cur >= last_byte) {
3054 if (tree->ops && tree->ops->writepage_end_io_hook)
3055 tree->ops->writepage_end_io_hook(page, cur,
3059 em = epd->get_extent(inode, page, pg_offset, cur,
3061 if (IS_ERR_OR_NULL(em)) {
3066 extent_offset = cur - em->start;
3067 BUG_ON(extent_map_end(em) <= cur);
3069 iosize = min(extent_map_end(em) - cur, end - cur + 1);
3070 iosize = ALIGN(iosize, blocksize);
3071 sector = (em->block_start + extent_offset) >> 9;
3073 block_start = em->block_start;
3074 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3075 free_extent_map(em);
3079 * compressed and inline extents are written through other
3082 if (compressed || block_start == EXTENT_MAP_HOLE ||
3083 block_start == EXTENT_MAP_INLINE) {
3085 * end_io notification does not happen here for
3086 * compressed extents
3088 if (!compressed && tree->ops &&
3089 tree->ops->writepage_end_io_hook)
3090 tree->ops->writepage_end_io_hook(page, cur,
3093 else if (compressed) {
3094 /* we don't want to end_page_writeback on
3095 * a compressed extent. this happens
3102 pg_offset += iosize;
3105 /* leave this out until we have a page_mkwrite call */
3106 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
3107 EXTENT_DIRTY, 0, NULL)) {
3109 pg_offset += iosize;
3113 if (tree->ops && tree->ops->writepage_io_hook) {
3114 ret = tree->ops->writepage_io_hook(page, cur,
3122 unsigned long max_nr = end_index + 1;
3124 set_range_writeback(tree, cur, cur + iosize - 1);
3125 if (!PageWriteback(page)) {
3126 printk(KERN_ERR "btrfs warning page %lu not "
3127 "writeback, cur %llu end %llu\n",
3128 page->index, (unsigned long long)cur,
3129 (unsigned long long)end);
3132 ret = submit_extent_page(write_flags, tree, page,
3133 sector, iosize, pg_offset,
3134 bdev, &epd->bio, max_nr,
3135 end_bio_extent_writepage,
3141 pg_offset += iosize;
3146 /* make sure the mapping tag for page dirty gets cleared */
3147 set_page_writeback(page);
3148 end_page_writeback(page);
3154 /* drop our reference on any cached states */
3155 free_extent_state(cached_state);
3159 static int eb_wait(void *word)
3165 static void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3167 wait_on_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK, eb_wait,
3168 TASK_UNINTERRUPTIBLE);
3171 static int lock_extent_buffer_for_io(struct extent_buffer *eb,
3172 struct btrfs_fs_info *fs_info,
3173 struct extent_page_data *epd)
3175 unsigned long i, num_pages;
3179 if (!btrfs_try_tree_write_lock(eb)) {
3181 flush_write_bio(epd);
3182 btrfs_tree_lock(eb);
3185 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3186 btrfs_tree_unlock(eb);
3190 flush_write_bio(epd);
3194 wait_on_extent_buffer_writeback(eb);
3195 btrfs_tree_lock(eb);
3196 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3198 btrfs_tree_unlock(eb);
3203 * We need to do this to prevent races in people who check if the eb is
3204 * under IO since we can end up having no IO bits set for a short period
3207 spin_lock(&eb->refs_lock);
3208 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3209 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3210 spin_unlock(&eb->refs_lock);
3211 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3212 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
3214 fs_info->dirty_metadata_batch);
3217 spin_unlock(&eb->refs_lock);
3220 btrfs_tree_unlock(eb);
3225 num_pages = num_extent_pages(eb->start, eb->len);
3226 for (i = 0; i < num_pages; i++) {
3227 struct page *p = extent_buffer_page(eb, i);
3229 if (!trylock_page(p)) {
3231 flush_write_bio(epd);
3241 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3243 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3244 smp_mb__after_clear_bit();
3245 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3248 static void end_bio_extent_buffer_writepage(struct bio *bio, int err)
3250 int uptodate = err == 0;
3251 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
3252 struct extent_buffer *eb;
3256 struct page *page = bvec->bv_page;
3259 eb = (struct extent_buffer *)page->private;
3261 done = atomic_dec_and_test(&eb->io_pages);
3263 if (!uptodate || test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
3264 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3265 ClearPageUptodate(page);
3269 end_page_writeback(page);
3274 end_extent_buffer_writeback(eb);
3275 } while (bvec >= bio->bi_io_vec);
3281 static int write_one_eb(struct extent_buffer *eb,
3282 struct btrfs_fs_info *fs_info,
3283 struct writeback_control *wbc,
3284 struct extent_page_data *epd)
3286 struct block_device *bdev = fs_info->fs_devices->latest_bdev;
3287 u64 offset = eb->start;
3288 unsigned long i, num_pages;
3289 unsigned long bio_flags = 0;
3290 int rw = (epd->sync_io ? WRITE_SYNC : WRITE);
3293 clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3294 num_pages = num_extent_pages(eb->start, eb->len);
3295 atomic_set(&eb->io_pages, num_pages);
3296 if (btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID)
3297 bio_flags = EXTENT_BIO_TREE_LOG;
3299 for (i = 0; i < num_pages; i++) {
3300 struct page *p = extent_buffer_page(eb, i);
3302 clear_page_dirty_for_io(p);
3303 set_page_writeback(p);
3304 ret = submit_extent_page(rw, eb->tree, p, offset >> 9,
3305 PAGE_CACHE_SIZE, 0, bdev, &epd->bio,
3306 -1, end_bio_extent_buffer_writepage,
3307 0, epd->bio_flags, bio_flags);
3308 epd->bio_flags = bio_flags;
3310 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3312 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3313 end_extent_buffer_writeback(eb);
3317 offset += PAGE_CACHE_SIZE;
3318 update_nr_written(p, wbc, 1);
3322 if (unlikely(ret)) {
3323 for (; i < num_pages; i++) {
3324 struct page *p = extent_buffer_page(eb, i);
3332 int btree_write_cache_pages(struct address_space *mapping,
3333 struct writeback_control *wbc)
3335 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
3336 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
3337 struct extent_buffer *eb, *prev_eb = NULL;
3338 struct extent_page_data epd = {
3342 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3347 int nr_to_write_done = 0;
3348 struct pagevec pvec;
3351 pgoff_t end; /* Inclusive */
3355 pagevec_init(&pvec, 0);
3356 if (wbc->range_cyclic) {
3357 index = mapping->writeback_index; /* Start from prev offset */
3360 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3361 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3364 if (wbc->sync_mode == WB_SYNC_ALL)
3365 tag = PAGECACHE_TAG_TOWRITE;
3367 tag = PAGECACHE_TAG_DIRTY;
3369 if (wbc->sync_mode == WB_SYNC_ALL)
3370 tag_pages_for_writeback(mapping, index, end);
3371 while (!done && !nr_to_write_done && (index <= end) &&
3372 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3373 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3377 for (i = 0; i < nr_pages; i++) {
3378 struct page *page = pvec.pages[i];
3380 if (!PagePrivate(page))
3383 if (!wbc->range_cyclic && page->index > end) {
3388 spin_lock(&mapping->private_lock);
3389 if (!PagePrivate(page)) {
3390 spin_unlock(&mapping->private_lock);
3394 eb = (struct extent_buffer *)page->private;
3397 * Shouldn't happen and normally this would be a BUG_ON
3398 * but no sense in crashing the users box for something
3399 * we can survive anyway.
3402 spin_unlock(&mapping->private_lock);
3407 if (eb == prev_eb) {
3408 spin_unlock(&mapping->private_lock);
3412 ret = atomic_inc_not_zero(&eb->refs);
3413 spin_unlock(&mapping->private_lock);
3418 ret = lock_extent_buffer_for_io(eb, fs_info, &epd);
3420 free_extent_buffer(eb);
3424 ret = write_one_eb(eb, fs_info, wbc, &epd);
3427 free_extent_buffer(eb);
3430 free_extent_buffer(eb);
3433 * the filesystem may choose to bump up nr_to_write.
3434 * We have to make sure to honor the new nr_to_write
3437 nr_to_write_done = wbc->nr_to_write <= 0;
3439 pagevec_release(&pvec);
3442 if (!scanned && !done) {
3444 * We hit the last page and there is more work to be done: wrap
3445 * back to the start of the file
3451 flush_write_bio(&epd);
3456 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3457 * @mapping: address space structure to write
3458 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3459 * @writepage: function called for each page
3460 * @data: data passed to writepage function
3462 * If a page is already under I/O, write_cache_pages() skips it, even
3463 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3464 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3465 * and msync() need to guarantee that all the data which was dirty at the time
3466 * the call was made get new I/O started against them. If wbc->sync_mode is
3467 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3468 * existing IO to complete.
3470 static int extent_write_cache_pages(struct extent_io_tree *tree,
3471 struct address_space *mapping,
3472 struct writeback_control *wbc,
3473 writepage_t writepage, void *data,
3474 void (*flush_fn)(void *))
3476 struct inode *inode = mapping->host;
3479 int nr_to_write_done = 0;
3480 struct pagevec pvec;
3483 pgoff_t end; /* Inclusive */
3488 * We have to hold onto the inode so that ordered extents can do their
3489 * work when the IO finishes. The alternative to this is failing to add
3490 * an ordered extent if the igrab() fails there and that is a huge pain
3491 * to deal with, so instead just hold onto the inode throughout the
3492 * writepages operation. If it fails here we are freeing up the inode
3493 * anyway and we'd rather not waste our time writing out stuff that is
3494 * going to be truncated anyway.
3499 pagevec_init(&pvec, 0);
3500 if (wbc->range_cyclic) {
3501 index = mapping->writeback_index; /* Start from prev offset */
3504 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3505 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3508 if (wbc->sync_mode == WB_SYNC_ALL)
3509 tag = PAGECACHE_TAG_TOWRITE;
3511 tag = PAGECACHE_TAG_DIRTY;
3513 if (wbc->sync_mode == WB_SYNC_ALL)
3514 tag_pages_for_writeback(mapping, index, end);
3515 while (!done && !nr_to_write_done && (index <= end) &&
3516 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3517 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3521 for (i = 0; i < nr_pages; i++) {
3522 struct page *page = pvec.pages[i];
3525 * At this point we hold neither mapping->tree_lock nor
3526 * lock on the page itself: the page may be truncated or
3527 * invalidated (changing page->mapping to NULL), or even
3528 * swizzled back from swapper_space to tmpfs file
3531 if (!trylock_page(page)) {
3536 if (unlikely(page->mapping != mapping)) {
3541 if (!wbc->range_cyclic && page->index > end) {
3547 if (wbc->sync_mode != WB_SYNC_NONE) {
3548 if (PageWriteback(page))
3550 wait_on_page_writeback(page);
3553 if (PageWriteback(page) ||
3554 !clear_page_dirty_for_io(page)) {
3559 ret = (*writepage)(page, wbc, data);
3561 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
3569 * the filesystem may choose to bump up nr_to_write.
3570 * We have to make sure to honor the new nr_to_write
3573 nr_to_write_done = wbc->nr_to_write <= 0;
3575 pagevec_release(&pvec);
3578 if (!scanned && !done) {
3580 * We hit the last page and there is more work to be done: wrap
3581 * back to the start of the file
3587 btrfs_add_delayed_iput(inode);
3591 static void flush_epd_write_bio(struct extent_page_data *epd)
3600 ret = submit_one_bio(rw, epd->bio, 0, epd->bio_flags);
3601 BUG_ON(ret < 0); /* -ENOMEM */
3606 static noinline void flush_write_bio(void *data)
3608 struct extent_page_data *epd = data;
3609 flush_epd_write_bio(epd);
3612 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
3613 get_extent_t *get_extent,
3614 struct writeback_control *wbc)
3617 struct extent_page_data epd = {
3620 .get_extent = get_extent,
3622 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3626 ret = __extent_writepage(page, wbc, &epd);
3628 flush_epd_write_bio(&epd);
3632 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
3633 u64 start, u64 end, get_extent_t *get_extent,
3637 struct address_space *mapping = inode->i_mapping;
3639 unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
3642 struct extent_page_data epd = {
3645 .get_extent = get_extent,
3647 .sync_io = mode == WB_SYNC_ALL,
3650 struct writeback_control wbc_writepages = {
3652 .nr_to_write = nr_pages * 2,
3653 .range_start = start,
3654 .range_end = end + 1,
3657 while (start <= end) {
3658 page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
3659 if (clear_page_dirty_for_io(page))
3660 ret = __extent_writepage(page, &wbc_writepages, &epd);
3662 if (tree->ops && tree->ops->writepage_end_io_hook)
3663 tree->ops->writepage_end_io_hook(page, start,
3664 start + PAGE_CACHE_SIZE - 1,
3668 page_cache_release(page);
3669 start += PAGE_CACHE_SIZE;
3672 flush_epd_write_bio(&epd);
3676 int extent_writepages(struct extent_io_tree *tree,
3677 struct address_space *mapping,
3678 get_extent_t *get_extent,
3679 struct writeback_control *wbc)
3682 struct extent_page_data epd = {
3685 .get_extent = get_extent,
3687 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3691 ret = extent_write_cache_pages(tree, mapping, wbc,
3692 __extent_writepage, &epd,
3694 flush_epd_write_bio(&epd);
3698 int extent_readpages(struct extent_io_tree *tree,
3699 struct address_space *mapping,
3700 struct list_head *pages, unsigned nr_pages,
3701 get_extent_t get_extent)
3703 struct bio *bio = NULL;
3705 unsigned long bio_flags = 0;
3706 struct page *pagepool[16];
3711 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
3712 page = list_entry(pages->prev, struct page, lru);
3714 prefetchw(&page->flags);
3715 list_del(&page->lru);
3716 if (add_to_page_cache_lru(page, mapping,
3717 page->index, GFP_NOFS)) {
3718 page_cache_release(page);
3722 pagepool[nr++] = page;
3723 if (nr < ARRAY_SIZE(pagepool))
3725 for (i = 0; i < nr; i++) {
3726 __extent_read_full_page(tree, pagepool[i], get_extent,
3727 &bio, 0, &bio_flags);
3728 page_cache_release(pagepool[i]);
3732 for (i = 0; i < nr; i++) {
3733 __extent_read_full_page(tree, pagepool[i], get_extent,
3734 &bio, 0, &bio_flags);
3735 page_cache_release(pagepool[i]);
3738 BUG_ON(!list_empty(pages));
3740 return submit_one_bio(READ, bio, 0, bio_flags);
3745 * basic invalidatepage code, this waits on any locked or writeback
3746 * ranges corresponding to the page, and then deletes any extent state
3747 * records from the tree
3749 int extent_invalidatepage(struct extent_io_tree *tree,
3750 struct page *page, unsigned long offset)
3752 struct extent_state *cached_state = NULL;
3753 u64 start = page_offset(page);
3754 u64 end = start + PAGE_CACHE_SIZE - 1;
3755 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
3757 start += ALIGN(offset, blocksize);
3761 lock_extent_bits(tree, start, end, 0, &cached_state);
3762 wait_on_page_writeback(page);
3763 clear_extent_bit(tree, start, end,
3764 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
3765 EXTENT_DO_ACCOUNTING,
3766 1, 1, &cached_state, GFP_NOFS);
3771 * a helper for releasepage, this tests for areas of the page that
3772 * are locked or under IO and drops the related state bits if it is safe
3775 int try_release_extent_state(struct extent_map_tree *map,
3776 struct extent_io_tree *tree, struct page *page,
3779 u64 start = page_offset(page);
3780 u64 end = start + PAGE_CACHE_SIZE - 1;
3783 if (test_range_bit(tree, start, end,
3784 EXTENT_IOBITS, 0, NULL))
3787 if ((mask & GFP_NOFS) == GFP_NOFS)
3790 * at this point we can safely clear everything except the
3791 * locked bit and the nodatasum bit
3793 ret = clear_extent_bit(tree, start, end,
3794 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
3797 /* if clear_extent_bit failed for enomem reasons,
3798 * we can't allow the release to continue.
3809 * a helper for releasepage. As long as there are no locked extents
3810 * in the range corresponding to the page, both state records and extent
3811 * map records are removed
3813 int try_release_extent_mapping(struct extent_map_tree *map,
3814 struct extent_io_tree *tree, struct page *page,
3817 struct extent_map *em;
3818 u64 start = page_offset(page);
3819 u64 end = start + PAGE_CACHE_SIZE - 1;
3821 if ((mask & __GFP_WAIT) &&
3822 page->mapping->host->i_size > 16 * 1024 * 1024) {
3824 while (start <= end) {
3825 len = end - start + 1;
3826 write_lock(&map->lock);
3827 em = lookup_extent_mapping(map, start, len);
3829 write_unlock(&map->lock);
3832 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
3833 em->start != start) {
3834 write_unlock(&map->lock);
3835 free_extent_map(em);
3838 if (!test_range_bit(tree, em->start,
3839 extent_map_end(em) - 1,
3840 EXTENT_LOCKED | EXTENT_WRITEBACK,
3842 remove_extent_mapping(map, em);
3843 /* once for the rb tree */
3844 free_extent_map(em);
3846 start = extent_map_end(em);
3847 write_unlock(&map->lock);
3850 free_extent_map(em);
3853 return try_release_extent_state(map, tree, page, mask);
3857 * helper function for fiemap, which doesn't want to see any holes.
3858 * This maps until we find something past 'last'
3860 static struct extent_map *get_extent_skip_holes(struct inode *inode,
3863 get_extent_t *get_extent)
3865 u64 sectorsize = BTRFS_I(inode)->root->sectorsize;
3866 struct extent_map *em;
3873 len = last - offset;
3876 len = ALIGN(len, sectorsize);
3877 em = get_extent(inode, NULL, 0, offset, len, 0);
3878 if (IS_ERR_OR_NULL(em))
3881 /* if this isn't a hole return it */
3882 if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) &&
3883 em->block_start != EXTENT_MAP_HOLE) {
3887 /* this is a hole, advance to the next extent */
3888 offset = extent_map_end(em);
3889 free_extent_map(em);
3896 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
3897 __u64 start, __u64 len, get_extent_t *get_extent)
3901 u64 max = start + len;
3905 u64 last_for_get_extent = 0;
3907 u64 isize = i_size_read(inode);
3908 struct btrfs_key found_key;
3909 struct extent_map *em = NULL;
3910 struct extent_state *cached_state = NULL;
3911 struct btrfs_path *path;
3912 struct btrfs_file_extent_item *item;
3917 unsigned long emflags;
3922 path = btrfs_alloc_path();
3925 path->leave_spinning = 1;
3927 start = ALIGN(start, BTRFS_I(inode)->root->sectorsize);
3928 len = ALIGN(len, BTRFS_I(inode)->root->sectorsize);
3931 * lookup the last file extent. We're not using i_size here
3932 * because there might be preallocation past i_size
3934 ret = btrfs_lookup_file_extent(NULL, BTRFS_I(inode)->root,
3935 path, btrfs_ino(inode), -1, 0);
3937 btrfs_free_path(path);
3942 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3943 struct btrfs_file_extent_item);
3944 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
3945 found_type = btrfs_key_type(&found_key);
3947 /* No extents, but there might be delalloc bits */
3948 if (found_key.objectid != btrfs_ino(inode) ||
3949 found_type != BTRFS_EXTENT_DATA_KEY) {
3950 /* have to trust i_size as the end */
3952 last_for_get_extent = isize;
3955 * remember the start of the last extent. There are a
3956 * bunch of different factors that go into the length of the
3957 * extent, so its much less complex to remember where it started
3959 last = found_key.offset;
3960 last_for_get_extent = last + 1;
3962 btrfs_free_path(path);
3965 * we might have some extents allocated but more delalloc past those
3966 * extents. so, we trust isize unless the start of the last extent is
3971 last_for_get_extent = isize;
3974 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len, 0,
3977 em = get_extent_skip_holes(inode, start, last_for_get_extent,
3987 u64 offset_in_extent;
3989 /* break if the extent we found is outside the range */
3990 if (em->start >= max || extent_map_end(em) < off)
3994 * get_extent may return an extent that starts before our
3995 * requested range. We have to make sure the ranges
3996 * we return to fiemap always move forward and don't
3997 * overlap, so adjust the offsets here
3999 em_start = max(em->start, off);
4002 * record the offset from the start of the extent
4003 * for adjusting the disk offset below
4005 offset_in_extent = em_start - em->start;
4006 em_end = extent_map_end(em);
4007 em_len = em_end - em_start;
4008 emflags = em->flags;
4013 * bump off for our next call to get_extent
4015 off = extent_map_end(em);
4019 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
4021 flags |= FIEMAP_EXTENT_LAST;
4022 } else if (em->block_start == EXTENT_MAP_INLINE) {
4023 flags |= (FIEMAP_EXTENT_DATA_INLINE |
4024 FIEMAP_EXTENT_NOT_ALIGNED);
4025 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
4026 flags |= (FIEMAP_EXTENT_DELALLOC |
4027 FIEMAP_EXTENT_UNKNOWN);
4029 disko = em->block_start + offset_in_extent;
4031 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4032 flags |= FIEMAP_EXTENT_ENCODED;
4034 free_extent_map(em);
4036 if ((em_start >= last) || em_len == (u64)-1 ||
4037 (last == (u64)-1 && isize <= em_end)) {
4038 flags |= FIEMAP_EXTENT_LAST;
4042 /* now scan forward to see if this is really the last extent. */
4043 em = get_extent_skip_holes(inode, off, last_for_get_extent,
4050 flags |= FIEMAP_EXTENT_LAST;
4053 ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
4059 free_extent_map(em);
4061 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len,
4062 &cached_state, GFP_NOFS);
4066 static void __free_extent_buffer(struct extent_buffer *eb)
4069 unsigned long flags;
4070 spin_lock_irqsave(&leak_lock, flags);
4071 list_del(&eb->leak_list);
4072 spin_unlock_irqrestore(&leak_lock, flags);
4074 kmem_cache_free(extent_buffer_cache, eb);
4077 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
4082 struct extent_buffer *eb = NULL;
4084 unsigned long flags;
4087 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
4094 rwlock_init(&eb->lock);
4095 atomic_set(&eb->write_locks, 0);
4096 atomic_set(&eb->read_locks, 0);
4097 atomic_set(&eb->blocking_readers, 0);
4098 atomic_set(&eb->blocking_writers, 0);
4099 atomic_set(&eb->spinning_readers, 0);
4100 atomic_set(&eb->spinning_writers, 0);
4101 eb->lock_nested = 0;
4102 init_waitqueue_head(&eb->write_lock_wq);
4103 init_waitqueue_head(&eb->read_lock_wq);
4106 spin_lock_irqsave(&leak_lock, flags);
4107 list_add(&eb->leak_list, &buffers);
4108 spin_unlock_irqrestore(&leak_lock, flags);
4110 spin_lock_init(&eb->refs_lock);
4111 atomic_set(&eb->refs, 1);
4112 atomic_set(&eb->io_pages, 0);
4115 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4117 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4118 > MAX_INLINE_EXTENT_BUFFER_SIZE);
4119 BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE);
4124 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src)
4128 struct extent_buffer *new;
4129 unsigned long num_pages = num_extent_pages(src->start, src->len);
4131 new = __alloc_extent_buffer(NULL, src->start, src->len, GFP_ATOMIC);
4135 for (i = 0; i < num_pages; i++) {
4136 p = alloc_page(GFP_ATOMIC);
4138 attach_extent_buffer_page(new, p);
4139 WARN_ON(PageDirty(p));
4144 copy_extent_buffer(new, src, 0, 0, src->len);
4145 set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
4146 set_bit(EXTENT_BUFFER_DUMMY, &new->bflags);
4151 struct extent_buffer *alloc_dummy_extent_buffer(u64 start, unsigned long len)
4153 struct extent_buffer *eb;
4154 unsigned long num_pages = num_extent_pages(0, len);
4157 eb = __alloc_extent_buffer(NULL, start, len, GFP_ATOMIC);
4161 for (i = 0; i < num_pages; i++) {
4162 eb->pages[i] = alloc_page(GFP_ATOMIC);
4166 set_extent_buffer_uptodate(eb);
4167 btrfs_set_header_nritems(eb, 0);
4168 set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4173 __free_page(eb->pages[i - 1]);
4174 __free_extent_buffer(eb);
4178 static int extent_buffer_under_io(struct extent_buffer *eb)
4180 return (atomic_read(&eb->io_pages) ||
4181 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4182 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4186 * Helper for releasing extent buffer page.
4188 static void btrfs_release_extent_buffer_page(struct extent_buffer *eb,
4189 unsigned long start_idx)
4191 unsigned long index;
4192 unsigned long num_pages;
4194 int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4196 BUG_ON(extent_buffer_under_io(eb));
4198 num_pages = num_extent_pages(eb->start, eb->len);
4199 index = start_idx + num_pages;
4200 if (start_idx >= index)
4205 page = extent_buffer_page(eb, index);
4206 if (page && mapped) {
4207 spin_lock(&page->mapping->private_lock);
4209 * We do this since we'll remove the pages after we've
4210 * removed the eb from the radix tree, so we could race
4211 * and have this page now attached to the new eb. So
4212 * only clear page_private if it's still connected to
4215 if (PagePrivate(page) &&
4216 page->private == (unsigned long)eb) {
4217 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4218 BUG_ON(PageDirty(page));
4219 BUG_ON(PageWriteback(page));
4221 * We need to make sure we haven't be attached
4224 ClearPagePrivate(page);
4225 set_page_private(page, 0);
4226 /* One for the page private */
4227 page_cache_release(page);
4229 spin_unlock(&page->mapping->private_lock);
4233 /* One for when we alloced the page */
4234 page_cache_release(page);
4236 } while (index != start_idx);
4240 * Helper for releasing the extent buffer.
4242 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4244 btrfs_release_extent_buffer_page(eb, 0);
4245 __free_extent_buffer(eb);
4248 static void check_buffer_tree_ref(struct extent_buffer *eb)
4251 /* the ref bit is tricky. We have to make sure it is set
4252 * if we have the buffer dirty. Otherwise the
4253 * code to free a buffer can end up dropping a dirty
4256 * Once the ref bit is set, it won't go away while the
4257 * buffer is dirty or in writeback, and it also won't
4258 * go away while we have the reference count on the
4261 * We can't just set the ref bit without bumping the
4262 * ref on the eb because free_extent_buffer might
4263 * see the ref bit and try to clear it. If this happens
4264 * free_extent_buffer might end up dropping our original
4265 * ref by mistake and freeing the page before we are able
4266 * to add one more ref.
4268 * So bump the ref count first, then set the bit. If someone
4269 * beat us to it, drop the ref we added.
4271 refs = atomic_read(&eb->refs);
4272 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4275 spin_lock(&eb->refs_lock);
4276 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4277 atomic_inc(&eb->refs);
4278 spin_unlock(&eb->refs_lock);
4281 static void mark_extent_buffer_accessed(struct extent_buffer *eb)
4283 unsigned long num_pages, i;
4285 check_buffer_tree_ref(eb);
4287 num_pages = num_extent_pages(eb->start, eb->len);
4288 for (i = 0; i < num_pages; i++) {
4289 struct page *p = extent_buffer_page(eb, i);
4290 mark_page_accessed(p);
4294 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
4295 u64 start, unsigned long len)
4297 unsigned long num_pages = num_extent_pages(start, len);
4299 unsigned long index = start >> PAGE_CACHE_SHIFT;
4300 struct extent_buffer *eb;
4301 struct extent_buffer *exists = NULL;
4303 struct address_space *mapping = tree->mapping;
4308 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
4309 if (eb && atomic_inc_not_zero(&eb->refs)) {
4311 mark_extent_buffer_accessed(eb);
4316 eb = __alloc_extent_buffer(tree, start, len, GFP_NOFS);
4320 for (i = 0; i < num_pages; i++, index++) {
4321 p = find_or_create_page(mapping, index, GFP_NOFS);
4325 spin_lock(&mapping->private_lock);
4326 if (PagePrivate(p)) {
4328 * We could have already allocated an eb for this page
4329 * and attached one so lets see if we can get a ref on
4330 * the existing eb, and if we can we know it's good and
4331 * we can just return that one, else we know we can just
4332 * overwrite page->private.
4334 exists = (struct extent_buffer *)p->private;
4335 if (atomic_inc_not_zero(&exists->refs)) {
4336 spin_unlock(&mapping->private_lock);
4338 page_cache_release(p);
4339 mark_extent_buffer_accessed(exists);
4344 * Do this so attach doesn't complain and we need to
4345 * drop the ref the old guy had.
4347 ClearPagePrivate(p);
4348 WARN_ON(PageDirty(p));
4349 page_cache_release(p);
4351 attach_extent_buffer_page(eb, p);
4352 spin_unlock(&mapping->private_lock);
4353 WARN_ON(PageDirty(p));
4354 mark_page_accessed(p);
4356 if (!PageUptodate(p))
4360 * see below about how we avoid a nasty race with release page
4361 * and why we unlock later
4365 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4367 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
4371 spin_lock(&tree->buffer_lock);
4372 ret = radix_tree_insert(&tree->buffer, start >> PAGE_CACHE_SHIFT, eb);
4373 if (ret == -EEXIST) {
4374 exists = radix_tree_lookup(&tree->buffer,
4375 start >> PAGE_CACHE_SHIFT);
4376 if (!atomic_inc_not_zero(&exists->refs)) {
4377 spin_unlock(&tree->buffer_lock);
4378 radix_tree_preload_end();
4382 spin_unlock(&tree->buffer_lock);
4383 radix_tree_preload_end();
4384 mark_extent_buffer_accessed(exists);
4387 /* add one reference for the tree */
4388 check_buffer_tree_ref(eb);
4389 spin_unlock(&tree->buffer_lock);
4390 radix_tree_preload_end();
4393 * there is a race where release page may have
4394 * tried to find this extent buffer in the radix
4395 * but failed. It will tell the VM it is safe to
4396 * reclaim the, and it will clear the page private bit.
4397 * We must make sure to set the page private bit properly
4398 * after the extent buffer is in the radix tree so
4399 * it doesn't get lost
4401 SetPageChecked(eb->pages[0]);
4402 for (i = 1; i < num_pages; i++) {
4403 p = extent_buffer_page(eb, i);
4404 ClearPageChecked(p);
4407 unlock_page(eb->pages[0]);
4411 for (i = 0; i < num_pages; i++) {
4413 unlock_page(eb->pages[i]);
4416 WARN_ON(!atomic_dec_and_test(&eb->refs));
4417 btrfs_release_extent_buffer(eb);
4421 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
4422 u64 start, unsigned long len)
4424 struct extent_buffer *eb;
4427 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
4428 if (eb && atomic_inc_not_zero(&eb->refs)) {
4430 mark_extent_buffer_accessed(eb);
4438 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
4440 struct extent_buffer *eb =
4441 container_of(head, struct extent_buffer, rcu_head);
4443 __free_extent_buffer(eb);
4446 /* Expects to have eb->eb_lock already held */
4447 static int release_extent_buffer(struct extent_buffer *eb, gfp_t mask)
4449 WARN_ON(atomic_read(&eb->refs) == 0);
4450 if (atomic_dec_and_test(&eb->refs)) {
4451 if (test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags)) {
4452 spin_unlock(&eb->refs_lock);
4454 struct extent_io_tree *tree = eb->tree;
4456 spin_unlock(&eb->refs_lock);
4458 spin_lock(&tree->buffer_lock);
4459 radix_tree_delete(&tree->buffer,
4460 eb->start >> PAGE_CACHE_SHIFT);
4461 spin_unlock(&tree->buffer_lock);
4464 /* Should be safe to release our pages at this point */
4465 btrfs_release_extent_buffer_page(eb, 0);
4466 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
4469 spin_unlock(&eb->refs_lock);
4474 void free_extent_buffer(struct extent_buffer *eb)
4482 refs = atomic_read(&eb->refs);
4485 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
4490 spin_lock(&eb->refs_lock);
4491 if (atomic_read(&eb->refs) == 2 &&
4492 test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))
4493 atomic_dec(&eb->refs);
4495 if (atomic_read(&eb->refs) == 2 &&
4496 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
4497 !extent_buffer_under_io(eb) &&
4498 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4499 atomic_dec(&eb->refs);
4502 * I know this is terrible, but it's temporary until we stop tracking
4503 * the uptodate bits and such for the extent buffers.
4505 release_extent_buffer(eb, GFP_ATOMIC);
4508 void free_extent_buffer_stale(struct extent_buffer *eb)
4513 spin_lock(&eb->refs_lock);
4514 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
4516 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
4517 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4518 atomic_dec(&eb->refs);
4519 release_extent_buffer(eb, GFP_NOFS);
4522 void clear_extent_buffer_dirty(struct extent_buffer *eb)
4525 unsigned long num_pages;
4528 num_pages = num_extent_pages(eb->start, eb->len);
4530 for (i = 0; i < num_pages; i++) {
4531 page = extent_buffer_page(eb, i);
4532 if (!PageDirty(page))
4536 WARN_ON(!PagePrivate(page));
4538 clear_page_dirty_for_io(page);
4539 spin_lock_irq(&page->mapping->tree_lock);
4540 if (!PageDirty(page)) {
4541 radix_tree_tag_clear(&page->mapping->page_tree,
4543 PAGECACHE_TAG_DIRTY);
4545 spin_unlock_irq(&page->mapping->tree_lock);
4546 ClearPageError(page);
4549 WARN_ON(atomic_read(&eb->refs) == 0);
4552 int set_extent_buffer_dirty(struct extent_buffer *eb)
4555 unsigned long num_pages;
4558 check_buffer_tree_ref(eb);
4560 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
4562 num_pages = num_extent_pages(eb->start, eb->len);
4563 WARN_ON(atomic_read(&eb->refs) == 0);
4564 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
4566 for (i = 0; i < num_pages; i++)
4567 set_page_dirty(extent_buffer_page(eb, i));
4571 static int range_straddles_pages(u64 start, u64 len)
4573 if (len < PAGE_CACHE_SIZE)
4575 if (start & (PAGE_CACHE_SIZE - 1))
4577 if ((start + len) & (PAGE_CACHE_SIZE - 1))
4582 int clear_extent_buffer_uptodate(struct extent_buffer *eb)
4586 unsigned long num_pages;
4588 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4589 num_pages = num_extent_pages(eb->start, eb->len);
4590 for (i = 0; i < num_pages; i++) {
4591 page = extent_buffer_page(eb, i);
4593 ClearPageUptodate(page);
4598 int set_extent_buffer_uptodate(struct extent_buffer *eb)
4602 unsigned long num_pages;
4604 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4605 num_pages = num_extent_pages(eb->start, eb->len);
4606 for (i = 0; i < num_pages; i++) {
4607 page = extent_buffer_page(eb, i);
4608 SetPageUptodate(page);
4613 int extent_range_uptodate(struct extent_io_tree *tree,
4618 int pg_uptodate = 1;
4620 unsigned long index;
4622 if (range_straddles_pages(start, end - start + 1)) {
4623 ret = test_range_bit(tree, start, end,
4624 EXTENT_UPTODATE, 1, NULL);
4628 while (start <= end) {
4629 index = start >> PAGE_CACHE_SHIFT;
4630 page = find_get_page(tree->mapping, index);
4633 uptodate = PageUptodate(page);
4634 page_cache_release(page);
4639 start += PAGE_CACHE_SIZE;
4644 int extent_buffer_uptodate(struct extent_buffer *eb)
4646 return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4649 int read_extent_buffer_pages(struct extent_io_tree *tree,
4650 struct extent_buffer *eb, u64 start, int wait,
4651 get_extent_t *get_extent, int mirror_num)
4654 unsigned long start_i;
4658 int locked_pages = 0;
4659 int all_uptodate = 1;
4660 unsigned long num_pages;
4661 unsigned long num_reads = 0;
4662 struct bio *bio = NULL;
4663 unsigned long bio_flags = 0;
4665 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
4669 WARN_ON(start < eb->start);
4670 start_i = (start >> PAGE_CACHE_SHIFT) -
4671 (eb->start >> PAGE_CACHE_SHIFT);
4676 num_pages = num_extent_pages(eb->start, eb->len);
4677 for (i = start_i; i < num_pages; i++) {
4678 page = extent_buffer_page(eb, i);
4679 if (wait == WAIT_NONE) {
4680 if (!trylock_page(page))
4686 if (!PageUptodate(page)) {
4693 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4697 clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
4698 eb->read_mirror = 0;
4699 atomic_set(&eb->io_pages, num_reads);
4700 for (i = start_i; i < num_pages; i++) {
4701 page = extent_buffer_page(eb, i);
4702 if (!PageUptodate(page)) {
4703 ClearPageError(page);
4704 err = __extent_read_full_page(tree, page,
4706 mirror_num, &bio_flags);
4715 err = submit_one_bio(READ, bio, mirror_num, bio_flags);
4720 if (ret || wait != WAIT_COMPLETE)
4723 for (i = start_i; i < num_pages; i++) {
4724 page = extent_buffer_page(eb, i);
4725 wait_on_page_locked(page);
4726 if (!PageUptodate(page))
4734 while (locked_pages > 0) {
4735 page = extent_buffer_page(eb, i);
4743 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
4744 unsigned long start,
4751 char *dst = (char *)dstv;
4752 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4753 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4755 WARN_ON(start > eb->len);
4756 WARN_ON(start + len > eb->start + eb->len);
4758 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4761 page = extent_buffer_page(eb, i);
4763 cur = min(len, (PAGE_CACHE_SIZE - offset));
4764 kaddr = page_address(page);
4765 memcpy(dst, kaddr + offset, cur);
4774 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
4775 unsigned long min_len, char **map,
4776 unsigned long *map_start,
4777 unsigned long *map_len)
4779 size_t offset = start & (PAGE_CACHE_SIZE - 1);
4782 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4783 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4784 unsigned long end_i = (start_offset + start + min_len - 1) >>
4791 offset = start_offset;
4795 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
4798 if (start + min_len > eb->len) {
4799 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, "
4800 "wanted %lu %lu\n", (unsigned long long)eb->start,
4801 eb->len, start, min_len);
4805 p = extent_buffer_page(eb, i);
4806 kaddr = page_address(p);
4807 *map = kaddr + offset;
4808 *map_len = PAGE_CACHE_SIZE - offset;
4812 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
4813 unsigned long start,
4820 char *ptr = (char *)ptrv;
4821 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4822 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4825 WARN_ON(start > eb->len);
4826 WARN_ON(start + len > eb->start + eb->len);
4828 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4831 page = extent_buffer_page(eb, i);
4833 cur = min(len, (PAGE_CACHE_SIZE - offset));
4835 kaddr = page_address(page);
4836 ret = memcmp(ptr, kaddr + offset, cur);
4848 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
4849 unsigned long start, unsigned long len)
4855 char *src = (char *)srcv;
4856 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4857 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4859 WARN_ON(start > eb->len);
4860 WARN_ON(start + len > eb->start + eb->len);
4862 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4865 page = extent_buffer_page(eb, i);
4866 WARN_ON(!PageUptodate(page));
4868 cur = min(len, PAGE_CACHE_SIZE - offset);
4869 kaddr = page_address(page);
4870 memcpy(kaddr + offset, src, cur);
4879 void memset_extent_buffer(struct extent_buffer *eb, char c,
4880 unsigned long start, unsigned long len)
4886 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4887 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4889 WARN_ON(start > eb->len);
4890 WARN_ON(start + len > eb->start + eb->len);
4892 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4895 page = extent_buffer_page(eb, i);
4896 WARN_ON(!PageUptodate(page));
4898 cur = min(len, PAGE_CACHE_SIZE - offset);
4899 kaddr = page_address(page);
4900 memset(kaddr + offset, c, cur);
4908 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
4909 unsigned long dst_offset, unsigned long src_offset,
4912 u64 dst_len = dst->len;
4917 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4918 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
4920 WARN_ON(src->len != dst_len);
4922 offset = (start_offset + dst_offset) &
4923 ((unsigned long)PAGE_CACHE_SIZE - 1);
4926 page = extent_buffer_page(dst, i);
4927 WARN_ON(!PageUptodate(page));
4929 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
4931 kaddr = page_address(page);
4932 read_extent_buffer(src, kaddr + offset, src_offset, cur);
4941 static void move_pages(struct page *dst_page, struct page *src_page,
4942 unsigned long dst_off, unsigned long src_off,
4945 char *dst_kaddr = page_address(dst_page);
4946 if (dst_page == src_page) {
4947 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
4949 char *src_kaddr = page_address(src_page);
4950 char *p = dst_kaddr + dst_off + len;
4951 char *s = src_kaddr + src_off + len;
4958 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
4960 unsigned long distance = (src > dst) ? src - dst : dst - src;
4961 return distance < len;
4964 static void copy_pages(struct page *dst_page, struct page *src_page,
4965 unsigned long dst_off, unsigned long src_off,
4968 char *dst_kaddr = page_address(dst_page);
4970 int must_memmove = 0;
4972 if (dst_page != src_page) {
4973 src_kaddr = page_address(src_page);
4975 src_kaddr = dst_kaddr;
4976 if (areas_overlap(src_off, dst_off, len))
4981 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
4983 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
4986 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
4987 unsigned long src_offset, unsigned long len)
4990 size_t dst_off_in_page;
4991 size_t src_off_in_page;
4992 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4993 unsigned long dst_i;
4994 unsigned long src_i;
4996 if (src_offset + len > dst->len) {
4997 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
4998 "len %lu dst len %lu\n", src_offset, len, dst->len);
5001 if (dst_offset + len > dst->len) {
5002 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
5003 "len %lu dst len %lu\n", dst_offset, len, dst->len);
5008 dst_off_in_page = (start_offset + dst_offset) &
5009 ((unsigned long)PAGE_CACHE_SIZE - 1);
5010 src_off_in_page = (start_offset + src_offset) &
5011 ((unsigned long)PAGE_CACHE_SIZE - 1);
5013 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
5014 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
5016 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
5018 cur = min_t(unsigned long, cur,
5019 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
5021 copy_pages(extent_buffer_page(dst, dst_i),
5022 extent_buffer_page(dst, src_i),
5023 dst_off_in_page, src_off_in_page, cur);
5031 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5032 unsigned long src_offset, unsigned long len)
5035 size_t dst_off_in_page;
5036 size_t src_off_in_page;
5037 unsigned long dst_end = dst_offset + len - 1;
5038 unsigned long src_end = src_offset + len - 1;
5039 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5040 unsigned long dst_i;
5041 unsigned long src_i;
5043 if (src_offset + len > dst->len) {
5044 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
5045 "len %lu len %lu\n", src_offset, len, dst->len);
5048 if (dst_offset + len > dst->len) {
5049 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
5050 "len %lu len %lu\n", dst_offset, len, dst->len);
5053 if (dst_offset < src_offset) {
5054 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
5058 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
5059 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
5061 dst_off_in_page = (start_offset + dst_end) &
5062 ((unsigned long)PAGE_CACHE_SIZE - 1);
5063 src_off_in_page = (start_offset + src_end) &
5064 ((unsigned long)PAGE_CACHE_SIZE - 1);
5066 cur = min_t(unsigned long, len, src_off_in_page + 1);
5067 cur = min(cur, dst_off_in_page + 1);
5068 move_pages(extent_buffer_page(dst, dst_i),
5069 extent_buffer_page(dst, src_i),
5070 dst_off_in_page - cur + 1,
5071 src_off_in_page - cur + 1, cur);
5079 int try_release_extent_buffer(struct page *page, gfp_t mask)
5081 struct extent_buffer *eb;
5084 * We need to make sure noboody is attaching this page to an eb right
5087 spin_lock(&page->mapping->private_lock);
5088 if (!PagePrivate(page)) {
5089 spin_unlock(&page->mapping->private_lock);
5093 eb = (struct extent_buffer *)page->private;
5097 * This is a little awful but should be ok, we need to make sure that
5098 * the eb doesn't disappear out from under us while we're looking at
5101 spin_lock(&eb->refs_lock);
5102 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
5103 spin_unlock(&eb->refs_lock);
5104 spin_unlock(&page->mapping->private_lock);
5107 spin_unlock(&page->mapping->private_lock);
5109 if ((mask & GFP_NOFS) == GFP_NOFS)
5113 * If tree ref isn't set then we know the ref on this eb is a real ref,
5114 * so just return, this page will likely be freed soon anyway.
5116 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
5117 spin_unlock(&eb->refs_lock);
5121 return release_extent_buffer(eb, mask);