1 #include <linux/bitops.h>
2 #include <linux/slab.h>
5 #include <linux/pagemap.h>
6 #include <linux/page-flags.h>
7 #include <linux/module.h>
8 #include <linux/spinlock.h>
9 #include <linux/blkdev.h>
10 #include <linux/swap.h>
11 #include <linux/writeback.h>
12 #include <linux/pagevec.h>
13 #include <linux/prefetch.h>
14 #include <linux/cleancache.h>
15 #include "extent_io.h"
16 #include "extent_map.h"
19 #include "btrfs_inode.h"
21 #include "check-integrity.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;
48 /* tells writepage not to lock the state bits for this range
49 * it still does the unlocking
51 unsigned int extent_locked:1;
53 /* tells the submit_bio code to use a WRITE_SYNC */
54 unsigned int sync_io:1;
57 static noinline void flush_write_bio(void *data);
58 static inline struct btrfs_fs_info *
59 tree_fs_info(struct extent_io_tree *tree)
61 return btrfs_sb(tree->mapping->host->i_sb);
64 int __init extent_io_init(void)
66 extent_state_cache = kmem_cache_create("extent_state",
67 sizeof(struct extent_state), 0,
68 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
69 if (!extent_state_cache)
72 extent_buffer_cache = kmem_cache_create("extent_buffers",
73 sizeof(struct extent_buffer), 0,
74 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
75 if (!extent_buffer_cache)
76 goto free_state_cache;
80 kmem_cache_destroy(extent_state_cache);
84 void extent_io_exit(void)
86 struct extent_state *state;
87 struct extent_buffer *eb;
89 while (!list_empty(&states)) {
90 state = list_entry(states.next, struct extent_state, leak_list);
91 printk(KERN_ERR "btrfs state leak: start %llu end %llu "
92 "state %lu in tree %p refs %d\n",
93 (unsigned long long)state->start,
94 (unsigned long long)state->end,
95 state->state, state->tree, atomic_read(&state->refs));
96 list_del(&state->leak_list);
97 kmem_cache_free(extent_state_cache, state);
101 while (!list_empty(&buffers)) {
102 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
103 printk(KERN_ERR "btrfs buffer leak start %llu len %lu "
104 "refs %d\n", (unsigned long long)eb->start,
105 eb->len, atomic_read(&eb->refs));
106 list_del(&eb->leak_list);
107 kmem_cache_free(extent_buffer_cache, eb);
109 if (extent_state_cache)
110 kmem_cache_destroy(extent_state_cache);
111 if (extent_buffer_cache)
112 kmem_cache_destroy(extent_buffer_cache);
115 void extent_io_tree_init(struct extent_io_tree *tree,
116 struct address_space *mapping)
118 tree->state = RB_ROOT;
119 INIT_RADIX_TREE(&tree->buffer, GFP_ATOMIC);
121 tree->dirty_bytes = 0;
122 spin_lock_init(&tree->lock);
123 spin_lock_init(&tree->buffer_lock);
124 tree->mapping = mapping;
127 static struct extent_state *alloc_extent_state(gfp_t mask)
129 struct extent_state *state;
134 state = kmem_cache_alloc(extent_state_cache, mask);
141 spin_lock_irqsave(&leak_lock, flags);
142 list_add(&state->leak_list, &states);
143 spin_unlock_irqrestore(&leak_lock, flags);
145 atomic_set(&state->refs, 1);
146 init_waitqueue_head(&state->wq);
147 trace_alloc_extent_state(state, mask, _RET_IP_);
151 void free_extent_state(struct extent_state *state)
155 if (atomic_dec_and_test(&state->refs)) {
159 WARN_ON(state->tree);
161 spin_lock_irqsave(&leak_lock, flags);
162 list_del(&state->leak_list);
163 spin_unlock_irqrestore(&leak_lock, flags);
165 trace_free_extent_state(state, _RET_IP_);
166 kmem_cache_free(extent_state_cache, state);
170 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
171 struct rb_node *node)
173 struct rb_node **p = &root->rb_node;
174 struct rb_node *parent = NULL;
175 struct tree_entry *entry;
179 entry = rb_entry(parent, struct tree_entry, rb_node);
181 if (offset < entry->start)
183 else if (offset > entry->end)
189 entry = rb_entry(node, struct tree_entry, rb_node);
190 rb_link_node(node, parent, p);
191 rb_insert_color(node, root);
195 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
196 struct rb_node **prev_ret,
197 struct rb_node **next_ret)
199 struct rb_root *root = &tree->state;
200 struct rb_node *n = root->rb_node;
201 struct rb_node *prev = NULL;
202 struct rb_node *orig_prev = NULL;
203 struct tree_entry *entry;
204 struct tree_entry *prev_entry = NULL;
207 entry = rb_entry(n, struct tree_entry, rb_node);
211 if (offset < entry->start)
213 else if (offset > entry->end)
221 while (prev && offset > prev_entry->end) {
222 prev = rb_next(prev);
223 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
230 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
231 while (prev && offset < prev_entry->start) {
232 prev = rb_prev(prev);
233 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
240 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
243 struct rb_node *prev = NULL;
246 ret = __etree_search(tree, offset, &prev, NULL);
252 static void merge_cb(struct extent_io_tree *tree, struct extent_state *new,
253 struct extent_state *other)
255 if (tree->ops && tree->ops->merge_extent_hook)
256 tree->ops->merge_extent_hook(tree->mapping->host, new,
261 * utility function to look for merge candidates inside a given range.
262 * Any extents with matching state are merged together into a single
263 * extent in the tree. Extents with EXTENT_IO in their state field
264 * are not merged because the end_io handlers need to be able to do
265 * operations on them without sleeping (or doing allocations/splits).
267 * This should be called with the tree lock held.
269 static void merge_state(struct extent_io_tree *tree,
270 struct extent_state *state)
272 struct extent_state *other;
273 struct rb_node *other_node;
275 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
278 other_node = rb_prev(&state->rb_node);
280 other = rb_entry(other_node, struct extent_state, rb_node);
281 if (other->end == state->start - 1 &&
282 other->state == state->state) {
283 merge_cb(tree, state, other);
284 state->start = other->start;
286 rb_erase(&other->rb_node, &tree->state);
287 free_extent_state(other);
290 other_node = rb_next(&state->rb_node);
292 other = rb_entry(other_node, struct extent_state, rb_node);
293 if (other->start == state->end + 1 &&
294 other->state == state->state) {
295 merge_cb(tree, state, other);
296 state->end = other->end;
298 rb_erase(&other->rb_node, &tree->state);
299 free_extent_state(other);
304 static void set_state_cb(struct extent_io_tree *tree,
305 struct extent_state *state, int *bits)
307 if (tree->ops && tree->ops->set_bit_hook)
308 tree->ops->set_bit_hook(tree->mapping->host, state, bits);
311 static void clear_state_cb(struct extent_io_tree *tree,
312 struct extent_state *state, int *bits)
314 if (tree->ops && tree->ops->clear_bit_hook)
315 tree->ops->clear_bit_hook(tree->mapping->host, state, bits);
318 static void set_state_bits(struct extent_io_tree *tree,
319 struct extent_state *state, int *bits);
322 * insert an extent_state struct into the tree. 'bits' are set on the
323 * struct before it is inserted.
325 * This may return -EEXIST if the extent is already there, in which case the
326 * state struct is freed.
328 * The tree lock is not taken internally. This is a utility function and
329 * probably isn't what you want to call (see set/clear_extent_bit).
331 static int insert_state(struct extent_io_tree *tree,
332 struct extent_state *state, u64 start, u64 end,
335 struct rb_node *node;
338 printk(KERN_ERR "btrfs end < start %llu %llu\n",
339 (unsigned long long)end,
340 (unsigned long long)start);
343 state->start = start;
346 set_state_bits(tree, state, bits);
348 node = tree_insert(&tree->state, end, &state->rb_node);
350 struct extent_state *found;
351 found = rb_entry(node, struct extent_state, rb_node);
352 printk(KERN_ERR "btrfs found node %llu %llu on insert of "
353 "%llu %llu\n", (unsigned long long)found->start,
354 (unsigned long long)found->end,
355 (unsigned long long)start, (unsigned long long)end);
359 merge_state(tree, state);
363 static void split_cb(struct extent_io_tree *tree, struct extent_state *orig,
366 if (tree->ops && tree->ops->split_extent_hook)
367 tree->ops->split_extent_hook(tree->mapping->host, orig, split);
371 * split a given extent state struct in two, inserting the preallocated
372 * struct 'prealloc' as the newly created second half. 'split' indicates an
373 * offset inside 'orig' where it should be split.
376 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
377 * are two extent state structs in the tree:
378 * prealloc: [orig->start, split - 1]
379 * orig: [ split, orig->end ]
381 * The tree locks are not taken by this function. They need to be held
384 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
385 struct extent_state *prealloc, u64 split)
387 struct rb_node *node;
389 split_cb(tree, orig, split);
391 prealloc->start = orig->start;
392 prealloc->end = split - 1;
393 prealloc->state = orig->state;
396 node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
398 free_extent_state(prealloc);
401 prealloc->tree = tree;
405 static struct extent_state *next_state(struct extent_state *state)
407 struct rb_node *next = rb_next(&state->rb_node);
409 return rb_entry(next, struct extent_state, rb_node);
415 * utility function to clear some bits in an extent state struct.
416 * it will optionally wake up any one waiting on this state (wake == 1)
418 * If no bits are set on the state struct after clearing things, the
419 * struct is freed and removed from the tree
421 static struct extent_state *clear_state_bit(struct extent_io_tree *tree,
422 struct extent_state *state,
425 struct extent_state *next;
426 int bits_to_clear = *bits & ~EXTENT_CTLBITS;
428 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
429 u64 range = state->end - state->start + 1;
430 WARN_ON(range > tree->dirty_bytes);
431 tree->dirty_bytes -= range;
433 clear_state_cb(tree, state, bits);
434 state->state &= ~bits_to_clear;
437 if (state->state == 0) {
438 next = next_state(state);
440 rb_erase(&state->rb_node, &tree->state);
442 free_extent_state(state);
447 merge_state(tree, state);
448 next = next_state(state);
453 static struct extent_state *
454 alloc_extent_state_atomic(struct extent_state *prealloc)
457 prealloc = alloc_extent_state(GFP_ATOMIC);
462 void extent_io_tree_panic(struct extent_io_tree *tree, int err)
464 btrfs_panic(tree_fs_info(tree), err, "Locking error: "
465 "Extent tree was modified by another "
466 "thread while locked.");
470 * clear some bits on a range in the tree. This may require splitting
471 * or inserting elements in the tree, so the gfp mask is used to
472 * indicate which allocations or sleeping are allowed.
474 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
475 * the given range from the tree regardless of state (ie for truncate).
477 * the range [start, end] is inclusive.
479 * This takes the tree lock, and returns 0 on success and < 0 on error.
481 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
482 int bits, int wake, int delete,
483 struct extent_state **cached_state,
486 struct extent_state *state;
487 struct extent_state *cached;
488 struct extent_state *prealloc = NULL;
489 struct rb_node *node;
495 bits |= ~EXTENT_CTLBITS;
496 bits |= EXTENT_FIRST_DELALLOC;
498 if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY))
501 if (!prealloc && (mask & __GFP_WAIT)) {
502 prealloc = alloc_extent_state(mask);
507 spin_lock(&tree->lock);
509 cached = *cached_state;
512 *cached_state = NULL;
516 if (cached && cached->tree && cached->start <= start &&
517 cached->end > start) {
519 atomic_dec(&cached->refs);
524 free_extent_state(cached);
527 * this search will find the extents that end after
530 node = tree_search(tree, start);
533 state = rb_entry(node, struct extent_state, rb_node);
535 if (state->start > end)
537 WARN_ON(state->end < start);
538 last_end = state->end;
540 /* the state doesn't have the wanted bits, go ahead */
541 if (!(state->state & bits)) {
542 state = next_state(state);
547 * | ---- desired range ---- |
549 * | ------------- state -------------- |
551 * We need to split the extent we found, and may flip
552 * bits on second half.
554 * If the extent we found extends past our range, we
555 * just split and search again. It'll get split again
556 * the next time though.
558 * If the extent we found is inside our range, we clear
559 * the desired bit on it.
562 if (state->start < start) {
563 prealloc = alloc_extent_state_atomic(prealloc);
565 err = split_state(tree, state, prealloc, start);
567 extent_io_tree_panic(tree, err);
572 if (state->end <= end) {
573 clear_state_bit(tree, state, &bits, wake);
574 if (last_end == (u64)-1)
576 start = last_end + 1;
581 * | ---- desired range ---- |
583 * We need to split the extent, and clear the bit
586 if (state->start <= end && state->end > end) {
587 prealloc = alloc_extent_state_atomic(prealloc);
589 err = split_state(tree, state, prealloc, end + 1);
591 extent_io_tree_panic(tree, err);
596 clear_state_bit(tree, prealloc, &bits, wake);
602 state = clear_state_bit(tree, state, &bits, wake);
604 if (last_end == (u64)-1)
606 start = last_end + 1;
607 if (start <= end && state && !need_resched())
612 spin_unlock(&tree->lock);
614 free_extent_state(prealloc);
621 spin_unlock(&tree->lock);
622 if (mask & __GFP_WAIT)
627 static void wait_on_state(struct extent_io_tree *tree,
628 struct extent_state *state)
629 __releases(tree->lock)
630 __acquires(tree->lock)
633 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
634 spin_unlock(&tree->lock);
636 spin_lock(&tree->lock);
637 finish_wait(&state->wq, &wait);
641 * waits for one or more bits to clear on a range in the state tree.
642 * The range [start, end] is inclusive.
643 * The tree lock is taken by this function
645 void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits)
647 struct extent_state *state;
648 struct rb_node *node;
650 spin_lock(&tree->lock);
654 * this search will find all the extents that end after
657 node = tree_search(tree, start);
661 state = rb_entry(node, struct extent_state, rb_node);
663 if (state->start > end)
666 if (state->state & bits) {
667 start = state->start;
668 atomic_inc(&state->refs);
669 wait_on_state(tree, state);
670 free_extent_state(state);
673 start = state->end + 1;
678 cond_resched_lock(&tree->lock);
681 spin_unlock(&tree->lock);
684 static void set_state_bits(struct extent_io_tree *tree,
685 struct extent_state *state,
688 int bits_to_set = *bits & ~EXTENT_CTLBITS;
690 set_state_cb(tree, state, bits);
691 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
692 u64 range = state->end - state->start + 1;
693 tree->dirty_bytes += range;
695 state->state |= bits_to_set;
698 static void cache_state(struct extent_state *state,
699 struct extent_state **cached_ptr)
701 if (cached_ptr && !(*cached_ptr)) {
702 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) {
704 atomic_inc(&state->refs);
709 static void uncache_state(struct extent_state **cached_ptr)
711 if (cached_ptr && (*cached_ptr)) {
712 struct extent_state *state = *cached_ptr;
714 free_extent_state(state);
719 * set some bits on a range in the tree. This may require allocations or
720 * sleeping, so the gfp mask is used to indicate what is allowed.
722 * If any of the exclusive bits are set, this will fail with -EEXIST if some
723 * part of the range already has the desired bits set. The start of the
724 * existing range is returned in failed_start in this case.
726 * [start, end] is inclusive This takes the tree lock.
729 static int __must_check
730 __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
731 int bits, int exclusive_bits, u64 *failed_start,
732 struct extent_state **cached_state, gfp_t mask)
734 struct extent_state *state;
735 struct extent_state *prealloc = NULL;
736 struct rb_node *node;
741 bits |= EXTENT_FIRST_DELALLOC;
743 if (!prealloc && (mask & __GFP_WAIT)) {
744 prealloc = alloc_extent_state(mask);
748 spin_lock(&tree->lock);
749 if (cached_state && *cached_state) {
750 state = *cached_state;
751 if (state->start <= start && state->end > start &&
753 node = &state->rb_node;
758 * this search will find all the extents that end after
761 node = tree_search(tree, start);
763 prealloc = alloc_extent_state_atomic(prealloc);
765 err = insert_state(tree, prealloc, start, end, &bits);
767 extent_io_tree_panic(tree, err);
772 state = rb_entry(node, struct extent_state, rb_node);
774 last_start = state->start;
775 last_end = state->end;
778 * | ---- desired range ---- |
781 * Just lock what we found and keep going
783 if (state->start == start && state->end <= end) {
784 struct rb_node *next_node;
785 if (state->state & exclusive_bits) {
786 *failed_start = state->start;
791 set_state_bits(tree, state, &bits);
793 cache_state(state, cached_state);
794 merge_state(tree, state);
795 if (last_end == (u64)-1)
798 start = last_end + 1;
799 next_node = rb_next(&state->rb_node);
800 if (next_node && start < end && prealloc && !need_resched()) {
801 state = rb_entry(next_node, struct extent_state,
803 if (state->start == start)
810 * | ---- desired range ---- |
813 * | ------------- state -------------- |
815 * We need to split the extent we found, and may flip bits on
818 * If the extent we found extends past our
819 * range, we just split and search again. It'll get split
820 * again the next time though.
822 * If the extent we found is inside our range, we set the
825 if (state->start < start) {
826 if (state->state & exclusive_bits) {
827 *failed_start = start;
832 prealloc = alloc_extent_state_atomic(prealloc);
834 err = split_state(tree, state, prealloc, start);
836 extent_io_tree_panic(tree, err);
841 if (state->end <= end) {
842 set_state_bits(tree, state, &bits);
843 cache_state(state, cached_state);
844 merge_state(tree, state);
845 if (last_end == (u64)-1)
847 start = last_end + 1;
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 - convert all bits in a given range from one bit to another
937 * @tree: the io tree to search
938 * @start: the start offset in bytes
939 * @end: the end offset in bytes (inclusive)
940 * @bits: the bits to set in this range
941 * @clear_bits: the bits to clear in this range
942 * @mask: the allocation mask
944 * This will go through and set bits for the given range. If any states exist
945 * already in this range they are set with the given bit and cleared of the
946 * clear_bits. This is only meant to be used by things that are mergeable, ie
947 * converting from say DELALLOC to DIRTY. This is not meant to be used with
948 * boundary bits like LOCK.
950 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
951 int bits, int clear_bits, gfp_t mask)
953 struct extent_state *state;
954 struct extent_state *prealloc = NULL;
955 struct rb_node *node;
961 if (!prealloc && (mask & __GFP_WAIT)) {
962 prealloc = alloc_extent_state(mask);
967 spin_lock(&tree->lock);
969 * this search will find all the extents that end after
972 node = tree_search(tree, start);
974 prealloc = alloc_extent_state_atomic(prealloc);
979 err = insert_state(tree, prealloc, start, end, &bits);
982 extent_io_tree_panic(tree, err);
985 state = rb_entry(node, struct extent_state, rb_node);
987 last_start = state->start;
988 last_end = state->end;
991 * | ---- desired range ---- |
994 * Just lock what we found and keep going
996 if (state->start == start && state->end <= end) {
997 struct rb_node *next_node;
999 set_state_bits(tree, state, &bits);
1000 clear_state_bit(tree, state, &clear_bits, 0);
1001 if (last_end == (u64)-1)
1004 start = last_end + 1;
1005 next_node = rb_next(&state->rb_node);
1006 if (next_node && start < end && prealloc && !need_resched()) {
1007 state = rb_entry(next_node, struct extent_state,
1009 if (state->start == start)
1016 * | ---- desired range ---- |
1019 * | ------------- state -------------- |
1021 * We need to split the extent we found, and may flip bits on
1024 * If the extent we found extends past our
1025 * range, we just split and search again. It'll get split
1026 * again the next time though.
1028 * If the extent we found is inside our range, we set the
1029 * desired bit on it.
1031 if (state->start < start) {
1032 prealloc = alloc_extent_state_atomic(prealloc);
1037 err = split_state(tree, state, prealloc, start);
1039 extent_io_tree_panic(tree, err);
1043 if (state->end <= end) {
1044 set_state_bits(tree, state, &bits);
1045 clear_state_bit(tree, state, &clear_bits, 0);
1046 if (last_end == (u64)-1)
1048 start = last_end + 1;
1053 * | ---- desired range ---- |
1054 * | state | or | state |
1056 * There's a hole, we need to insert something in it and
1057 * ignore the extent we found.
1059 if (state->start > start) {
1061 if (end < last_start)
1064 this_end = last_start - 1;
1066 prealloc = alloc_extent_state_atomic(prealloc);
1073 * Avoid to free 'prealloc' if it can be merged with
1076 err = insert_state(tree, prealloc, start, this_end,
1079 extent_io_tree_panic(tree, err);
1081 start = this_end + 1;
1085 * | ---- desired range ---- |
1087 * We need to split the extent, and set the bit
1090 if (state->start <= end && state->end > end) {
1091 prealloc = alloc_extent_state_atomic(prealloc);
1097 err = split_state(tree, state, prealloc, end + 1);
1099 extent_io_tree_panic(tree, err);
1101 set_state_bits(tree, prealloc, &bits);
1102 clear_state_bit(tree, prealloc, &clear_bits, 0);
1110 spin_unlock(&tree->lock);
1112 free_extent_state(prealloc);
1119 spin_unlock(&tree->lock);
1120 if (mask & __GFP_WAIT)
1125 /* wrappers around set/clear extent bit */
1126 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1129 return set_extent_bit(tree, start, end, EXTENT_DIRTY, NULL,
1133 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1134 int bits, gfp_t mask)
1136 return set_extent_bit(tree, start, end, bits, NULL,
1140 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1141 int bits, gfp_t mask)
1143 return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask);
1146 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
1147 struct extent_state **cached_state, gfp_t mask)
1149 return set_extent_bit(tree, start, end,
1150 EXTENT_DELALLOC | EXTENT_UPTODATE,
1151 NULL, cached_state, mask);
1154 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1157 return clear_extent_bit(tree, start, end,
1158 EXTENT_DIRTY | EXTENT_DELALLOC |
1159 EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask);
1162 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
1165 return set_extent_bit(tree, start, end, EXTENT_NEW, NULL,
1169 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
1170 struct extent_state **cached_state, gfp_t mask)
1172 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0,
1173 cached_state, mask);
1176 static int clear_extent_uptodate(struct extent_io_tree *tree, u64 start,
1177 u64 end, struct extent_state **cached_state,
1180 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0,
1181 cached_state, mask);
1185 * either insert or lock state struct between start and end use mask to tell
1186 * us if waiting is desired.
1188 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1189 int bits, struct extent_state **cached_state)
1194 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED | bits,
1195 EXTENT_LOCKED, &failed_start,
1196 cached_state, GFP_NOFS);
1197 if (err == -EEXIST) {
1198 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1199 start = failed_start;
1202 WARN_ON(start > end);
1207 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1209 return lock_extent_bits(tree, start, end, 0, NULL);
1212 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1217 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1218 &failed_start, NULL, GFP_NOFS);
1219 if (err == -EEXIST) {
1220 if (failed_start > start)
1221 clear_extent_bit(tree, start, failed_start - 1,
1222 EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS);
1228 int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end,
1229 struct extent_state **cached, gfp_t mask)
1231 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached,
1235 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1237 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL,
1242 * helper function to set both pages and extents in the tree writeback
1244 static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
1246 unsigned long index = start >> PAGE_CACHE_SHIFT;
1247 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1250 while (index <= end_index) {
1251 page = find_get_page(tree->mapping, index);
1252 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1253 set_page_writeback(page);
1254 page_cache_release(page);
1260 /* find the first state struct with 'bits' set after 'start', and
1261 * return it. tree->lock must be held. NULL will returned if
1262 * nothing was found after 'start'
1264 struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree,
1265 u64 start, int bits)
1267 struct rb_node *node;
1268 struct extent_state *state;
1271 * this search will find all the extents that end after
1274 node = tree_search(tree, start);
1279 state = rb_entry(node, struct extent_state, rb_node);
1280 if (state->end >= start && (state->state & bits))
1283 node = rb_next(node);
1292 * find the first offset in the io tree with 'bits' set. zero is
1293 * returned if we find something, and *start_ret and *end_ret are
1294 * set to reflect the state struct that was found.
1296 * If nothing was found, 1 is returned, < 0 on error
1298 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1299 u64 *start_ret, u64 *end_ret, int bits)
1301 struct extent_state *state;
1304 spin_lock(&tree->lock);
1305 state = find_first_extent_bit_state(tree, start, bits);
1307 *start_ret = state->start;
1308 *end_ret = state->end;
1311 spin_unlock(&tree->lock);
1316 * find a contiguous range of bytes in the file marked as delalloc, not
1317 * more than 'max_bytes'. start and end are used to return the range,
1319 * 1 is returned if we find something, 0 if nothing was in the tree
1321 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1322 u64 *start, u64 *end, u64 max_bytes,
1323 struct extent_state **cached_state)
1325 struct rb_node *node;
1326 struct extent_state *state;
1327 u64 cur_start = *start;
1329 u64 total_bytes = 0;
1331 spin_lock(&tree->lock);
1334 * this search will find all the extents that end after
1337 node = tree_search(tree, cur_start);
1345 state = rb_entry(node, struct extent_state, rb_node);
1346 if (found && (state->start != cur_start ||
1347 (state->state & EXTENT_BOUNDARY))) {
1350 if (!(state->state & EXTENT_DELALLOC)) {
1356 *start = state->start;
1357 *cached_state = state;
1358 atomic_inc(&state->refs);
1362 cur_start = state->end + 1;
1363 node = rb_next(node);
1366 total_bytes += state->end - state->start + 1;
1367 if (total_bytes >= max_bytes)
1371 spin_unlock(&tree->lock);
1375 static noinline void __unlock_for_delalloc(struct inode *inode,
1376 struct page *locked_page,
1380 struct page *pages[16];
1381 unsigned long index = start >> PAGE_CACHE_SHIFT;
1382 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1383 unsigned long nr_pages = end_index - index + 1;
1386 if (index == locked_page->index && end_index == index)
1389 while (nr_pages > 0) {
1390 ret = find_get_pages_contig(inode->i_mapping, index,
1391 min_t(unsigned long, nr_pages,
1392 ARRAY_SIZE(pages)), pages);
1393 for (i = 0; i < ret; i++) {
1394 if (pages[i] != locked_page)
1395 unlock_page(pages[i]);
1396 page_cache_release(pages[i]);
1404 static noinline int lock_delalloc_pages(struct inode *inode,
1405 struct page *locked_page,
1409 unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT;
1410 unsigned long start_index = index;
1411 unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT;
1412 unsigned long pages_locked = 0;
1413 struct page *pages[16];
1414 unsigned long nrpages;
1418 /* the caller is responsible for locking the start index */
1419 if (index == locked_page->index && index == end_index)
1422 /* skip the page at the start index */
1423 nrpages = end_index - index + 1;
1424 while (nrpages > 0) {
1425 ret = find_get_pages_contig(inode->i_mapping, index,
1426 min_t(unsigned long,
1427 nrpages, ARRAY_SIZE(pages)), pages);
1432 /* now we have an array of pages, lock them all */
1433 for (i = 0; i < ret; i++) {
1435 * the caller is taking responsibility for
1438 if (pages[i] != locked_page) {
1439 lock_page(pages[i]);
1440 if (!PageDirty(pages[i]) ||
1441 pages[i]->mapping != inode->i_mapping) {
1443 unlock_page(pages[i]);
1444 page_cache_release(pages[i]);
1448 page_cache_release(pages[i]);
1457 if (ret && pages_locked) {
1458 __unlock_for_delalloc(inode, locked_page,
1460 ((u64)(start_index + pages_locked - 1)) <<
1467 * find a contiguous range of bytes in the file marked as delalloc, not
1468 * more than 'max_bytes'. start and end are used to return the range,
1470 * 1 is returned if we find something, 0 if nothing was in the tree
1472 static noinline u64 find_lock_delalloc_range(struct inode *inode,
1473 struct extent_io_tree *tree,
1474 struct page *locked_page,
1475 u64 *start, u64 *end,
1481 struct extent_state *cached_state = NULL;
1486 /* step one, find a bunch of delalloc bytes starting at start */
1487 delalloc_start = *start;
1489 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1490 max_bytes, &cached_state);
1491 if (!found || delalloc_end <= *start) {
1492 *start = delalloc_start;
1493 *end = delalloc_end;
1494 free_extent_state(cached_state);
1499 * start comes from the offset of locked_page. We have to lock
1500 * pages in order, so we can't process delalloc bytes before
1503 if (delalloc_start < *start)
1504 delalloc_start = *start;
1507 * make sure to limit the number of pages we try to lock down
1510 if (delalloc_end + 1 - delalloc_start > max_bytes && loops)
1511 delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1;
1513 /* step two, lock all the pages after the page that has start */
1514 ret = lock_delalloc_pages(inode, locked_page,
1515 delalloc_start, delalloc_end);
1516 if (ret == -EAGAIN) {
1517 /* some of the pages are gone, lets avoid looping by
1518 * shortening the size of the delalloc range we're searching
1520 free_extent_state(cached_state);
1522 unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1);
1523 max_bytes = PAGE_CACHE_SIZE - offset;
1531 BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */
1533 /* step three, lock the state bits for the whole range */
1534 lock_extent_bits(tree, delalloc_start, delalloc_end, 0, &cached_state);
1536 /* then test to make sure it is all still delalloc */
1537 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1538 EXTENT_DELALLOC, 1, cached_state);
1540 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1541 &cached_state, GFP_NOFS);
1542 __unlock_for_delalloc(inode, locked_page,
1543 delalloc_start, delalloc_end);
1547 free_extent_state(cached_state);
1548 *start = delalloc_start;
1549 *end = delalloc_end;
1554 int extent_clear_unlock_delalloc(struct inode *inode,
1555 struct extent_io_tree *tree,
1556 u64 start, u64 end, struct page *locked_page,
1560 struct page *pages[16];
1561 unsigned long index = start >> PAGE_CACHE_SHIFT;
1562 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1563 unsigned long nr_pages = end_index - index + 1;
1567 if (op & EXTENT_CLEAR_UNLOCK)
1568 clear_bits |= EXTENT_LOCKED;
1569 if (op & EXTENT_CLEAR_DIRTY)
1570 clear_bits |= EXTENT_DIRTY;
1572 if (op & EXTENT_CLEAR_DELALLOC)
1573 clear_bits |= EXTENT_DELALLOC;
1575 clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS);
1576 if (!(op & (EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
1577 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK |
1578 EXTENT_SET_PRIVATE2)))
1581 while (nr_pages > 0) {
1582 ret = find_get_pages_contig(inode->i_mapping, index,
1583 min_t(unsigned long,
1584 nr_pages, ARRAY_SIZE(pages)), pages);
1585 for (i = 0; i < ret; i++) {
1587 if (op & EXTENT_SET_PRIVATE2)
1588 SetPagePrivate2(pages[i]);
1590 if (pages[i] == locked_page) {
1591 page_cache_release(pages[i]);
1594 if (op & EXTENT_CLEAR_DIRTY)
1595 clear_page_dirty_for_io(pages[i]);
1596 if (op & EXTENT_SET_WRITEBACK)
1597 set_page_writeback(pages[i]);
1598 if (op & EXTENT_END_WRITEBACK)
1599 end_page_writeback(pages[i]);
1600 if (op & EXTENT_CLEAR_UNLOCK_PAGE)
1601 unlock_page(pages[i]);
1602 page_cache_release(pages[i]);
1612 * count the number of bytes in the tree that have a given bit(s)
1613 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1614 * cached. The total number found is returned.
1616 u64 count_range_bits(struct extent_io_tree *tree,
1617 u64 *start, u64 search_end, u64 max_bytes,
1618 unsigned long bits, int contig)
1620 struct rb_node *node;
1621 struct extent_state *state;
1622 u64 cur_start = *start;
1623 u64 total_bytes = 0;
1627 if (search_end <= cur_start) {
1632 spin_lock(&tree->lock);
1633 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1634 total_bytes = tree->dirty_bytes;
1638 * this search will find all the extents that end after
1641 node = tree_search(tree, cur_start);
1646 state = rb_entry(node, struct extent_state, rb_node);
1647 if (state->start > search_end)
1649 if (contig && found && state->start > last + 1)
1651 if (state->end >= cur_start && (state->state & bits) == bits) {
1652 total_bytes += min(search_end, state->end) + 1 -
1653 max(cur_start, state->start);
1654 if (total_bytes >= max_bytes)
1657 *start = max(cur_start, state->start);
1661 } else if (contig && found) {
1664 node = rb_next(node);
1669 spin_unlock(&tree->lock);
1674 * set the private field for a given byte offset in the tree. If there isn't
1675 * an extent_state there already, this does nothing.
1677 int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1679 struct rb_node *node;
1680 struct extent_state *state;
1683 spin_lock(&tree->lock);
1685 * this search will find all the extents that end after
1688 node = tree_search(tree, start);
1693 state = rb_entry(node, struct extent_state, rb_node);
1694 if (state->start != start) {
1698 state->private = private;
1700 spin_unlock(&tree->lock);
1704 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1706 struct rb_node *node;
1707 struct extent_state *state;
1710 spin_lock(&tree->lock);
1712 * this search will find all the extents that end after
1715 node = tree_search(tree, start);
1720 state = rb_entry(node, struct extent_state, rb_node);
1721 if (state->start != start) {
1725 *private = state->private;
1727 spin_unlock(&tree->lock);
1732 * searches a range in the state tree for a given mask.
1733 * If 'filled' == 1, this returns 1 only if every extent in the tree
1734 * has the bits set. Otherwise, 1 is returned if any bit in the
1735 * range is found set.
1737 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1738 int bits, int filled, struct extent_state *cached)
1740 struct extent_state *state = NULL;
1741 struct rb_node *node;
1744 spin_lock(&tree->lock);
1745 if (cached && cached->tree && cached->start <= start &&
1746 cached->end > start)
1747 node = &cached->rb_node;
1749 node = tree_search(tree, start);
1750 while (node && start <= end) {
1751 state = rb_entry(node, struct extent_state, rb_node);
1753 if (filled && state->start > start) {
1758 if (state->start > end)
1761 if (state->state & bits) {
1765 } else if (filled) {
1770 if (state->end == (u64)-1)
1773 start = state->end + 1;
1776 node = rb_next(node);
1783 spin_unlock(&tree->lock);
1788 * helper function to set a given page up to date if all the
1789 * extents in the tree for that page are up to date
1791 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
1793 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1794 u64 end = start + PAGE_CACHE_SIZE - 1;
1795 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1796 SetPageUptodate(page);
1800 * helper function to unlock a page if all the extents in the tree
1801 * for that page are unlocked
1803 static void check_page_locked(struct extent_io_tree *tree, struct page *page)
1805 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1806 u64 end = start + PAGE_CACHE_SIZE - 1;
1807 if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL))
1812 * helper function to end page writeback if all the extents
1813 * in the tree for that page are done with writeback
1815 static void check_page_writeback(struct extent_io_tree *tree,
1818 end_page_writeback(page);
1822 * When IO fails, either with EIO or csum verification fails, we
1823 * try other mirrors that might have a good copy of the data. This
1824 * io_failure_record is used to record state as we go through all the
1825 * mirrors. If another mirror has good data, the page is set up to date
1826 * and things continue. If a good mirror can't be found, the original
1827 * bio end_io callback is called to indicate things have failed.
1829 struct io_failure_record {
1834 unsigned long bio_flags;
1840 static int free_io_failure(struct inode *inode, struct io_failure_record *rec,
1845 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1847 set_state_private(failure_tree, rec->start, 0);
1848 ret = clear_extent_bits(failure_tree, rec->start,
1849 rec->start + rec->len - 1,
1850 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1855 ret = clear_extent_bits(&BTRFS_I(inode)->io_tree, rec->start,
1856 rec->start + rec->len - 1,
1857 EXTENT_DAMAGED, GFP_NOFS);
1866 static void repair_io_failure_callback(struct bio *bio, int err)
1868 complete(bio->bi_private);
1872 * this bypasses the standard btrfs submit functions deliberately, as
1873 * the standard behavior is to write all copies in a raid setup. here we only
1874 * want to write the one bad copy. so we do the mapping for ourselves and issue
1875 * submit_bio directly.
1876 * to avoid any synchonization issues, wait for the data after writing, which
1877 * actually prevents the read that triggered the error from finishing.
1878 * currently, there can be no more than two copies of every data bit. thus,
1879 * exactly one rewrite is required.
1881 int repair_io_failure(struct btrfs_mapping_tree *map_tree, u64 start,
1882 u64 length, u64 logical, struct page *page,
1886 struct btrfs_device *dev;
1887 DECLARE_COMPLETION_ONSTACK(compl);
1890 struct btrfs_bio *bbio = NULL;
1893 BUG_ON(!mirror_num);
1895 bio = bio_alloc(GFP_NOFS, 1);
1898 bio->bi_private = &compl;
1899 bio->bi_end_io = repair_io_failure_callback;
1901 map_length = length;
1903 ret = btrfs_map_block(map_tree, WRITE, logical,
1904 &map_length, &bbio, mirror_num);
1909 BUG_ON(mirror_num != bbio->mirror_num);
1910 sector = bbio->stripes[mirror_num-1].physical >> 9;
1911 bio->bi_sector = sector;
1912 dev = bbio->stripes[mirror_num-1].dev;
1914 if (!dev || !dev->bdev || !dev->writeable) {
1918 bio->bi_bdev = dev->bdev;
1919 bio_add_page(bio, page, length, start-page_offset(page));
1920 btrfsic_submit_bio(WRITE_SYNC, bio);
1921 wait_for_completion(&compl);
1923 if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) {
1924 /* try to remap that extent elsewhere? */
1929 printk(KERN_INFO "btrfs read error corrected: ino %lu off %llu (dev %s "
1930 "sector %llu)\n", page->mapping->host->i_ino, start,
1937 int repair_eb_io_failure(struct btrfs_root *root, struct extent_buffer *eb,
1940 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1941 u64 start = eb->start;
1942 unsigned long i, num_pages = num_extent_pages(eb->start, eb->len);
1945 for (i = 0; i < num_pages; i++) {
1946 struct page *p = extent_buffer_page(eb, i);
1947 ret = repair_io_failure(map_tree, start, PAGE_CACHE_SIZE,
1948 start, p, mirror_num);
1951 start += PAGE_CACHE_SIZE;
1958 * each time an IO finishes, we do a fast check in the IO failure tree
1959 * to see if we need to process or clean up an io_failure_record
1961 static int clean_io_failure(u64 start, struct page *page)
1964 u64 private_failure;
1965 struct io_failure_record *failrec;
1966 struct btrfs_mapping_tree *map_tree;
1967 struct extent_state *state;
1971 struct inode *inode = page->mapping->host;
1974 ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1975 (u64)-1, 1, EXTENT_DIRTY, 0);
1979 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, start,
1984 failrec = (struct io_failure_record *)(unsigned long) private_failure;
1985 BUG_ON(!failrec->this_mirror);
1987 if (failrec->in_validation) {
1988 /* there was no real error, just free the record */
1989 pr_debug("clean_io_failure: freeing dummy error at %llu\n",
1995 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1996 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1999 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
2001 if (state && state->start == failrec->start) {
2002 map_tree = &BTRFS_I(inode)->root->fs_info->mapping_tree;
2003 num_copies = btrfs_num_copies(map_tree, failrec->logical,
2005 if (num_copies > 1) {
2006 ret = repair_io_failure(map_tree, start, failrec->len,
2007 failrec->logical, page,
2008 failrec->failed_mirror);
2015 ret = free_io_failure(inode, failrec, did_repair);
2021 * this is a generic handler for readpage errors (default
2022 * readpage_io_failed_hook). if other copies exist, read those and write back
2023 * good data to the failed position. does not investigate in remapping the
2024 * failed extent elsewhere, hoping the device will be smart enough to do this as
2028 static int bio_readpage_error(struct bio *failed_bio, struct page *page,
2029 u64 start, u64 end, int failed_mirror,
2030 struct extent_state *state)
2032 struct io_failure_record *failrec = NULL;
2034 struct extent_map *em;
2035 struct inode *inode = page->mapping->host;
2036 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2037 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2038 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2045 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
2047 ret = get_state_private(failure_tree, start, &private);
2049 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2052 failrec->start = start;
2053 failrec->len = end - start + 1;
2054 failrec->this_mirror = 0;
2055 failrec->bio_flags = 0;
2056 failrec->in_validation = 0;
2058 read_lock(&em_tree->lock);
2059 em = lookup_extent_mapping(em_tree, start, failrec->len);
2061 read_unlock(&em_tree->lock);
2066 if (em->start > start || em->start + em->len < start) {
2067 free_extent_map(em);
2070 read_unlock(&em_tree->lock);
2072 if (!em || IS_ERR(em)) {
2076 logical = start - em->start;
2077 logical = em->block_start + logical;
2078 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2079 logical = em->block_start;
2080 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2081 extent_set_compress_type(&failrec->bio_flags,
2084 pr_debug("bio_readpage_error: (new) logical=%llu, start=%llu, "
2085 "len=%llu\n", logical, start, failrec->len);
2086 failrec->logical = logical;
2087 free_extent_map(em);
2089 /* set the bits in the private failure tree */
2090 ret = set_extent_bits(failure_tree, start, end,
2091 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
2093 ret = set_state_private(failure_tree, start,
2094 (u64)(unsigned long)failrec);
2095 /* set the bits in the inode's tree */
2097 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED,
2104 failrec = (struct io_failure_record *)(unsigned long)private;
2105 pr_debug("bio_readpage_error: (found) logical=%llu, "
2106 "start=%llu, len=%llu, validation=%d\n",
2107 failrec->logical, failrec->start, failrec->len,
2108 failrec->in_validation);
2110 * when data can be on disk more than twice, add to failrec here
2111 * (e.g. with a list for failed_mirror) to make
2112 * clean_io_failure() clean all those errors at once.
2115 num_copies = btrfs_num_copies(
2116 &BTRFS_I(inode)->root->fs_info->mapping_tree,
2117 failrec->logical, failrec->len);
2118 if (num_copies == 1) {
2120 * we only have a single copy of the data, so don't bother with
2121 * all the retry and error correction code that follows. no
2122 * matter what the error is, it is very likely to persist.
2124 pr_debug("bio_readpage_error: cannot repair, num_copies == 1. "
2125 "state=%p, num_copies=%d, next_mirror %d, "
2126 "failed_mirror %d\n", state, num_copies,
2127 failrec->this_mirror, failed_mirror);
2128 free_io_failure(inode, failrec, 0);
2133 spin_lock(&tree->lock);
2134 state = find_first_extent_bit_state(tree, failrec->start,
2136 if (state && state->start != failrec->start)
2138 spin_unlock(&tree->lock);
2142 * there are two premises:
2143 * a) deliver good data to the caller
2144 * b) correct the bad sectors on disk
2146 if (failed_bio->bi_vcnt > 1) {
2148 * to fulfill b), we need to know the exact failing sectors, as
2149 * we don't want to rewrite any more than the failed ones. thus,
2150 * we need separate read requests for the failed bio
2152 * if the following BUG_ON triggers, our validation request got
2153 * merged. we need separate requests for our algorithm to work.
2155 BUG_ON(failrec->in_validation);
2156 failrec->in_validation = 1;
2157 failrec->this_mirror = failed_mirror;
2158 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
2161 * we're ready to fulfill a) and b) alongside. get a good copy
2162 * of the failed sector and if we succeed, we have setup
2163 * everything for repair_io_failure to do the rest for us.
2165 if (failrec->in_validation) {
2166 BUG_ON(failrec->this_mirror != failed_mirror);
2167 failrec->in_validation = 0;
2168 failrec->this_mirror = 0;
2170 failrec->failed_mirror = failed_mirror;
2171 failrec->this_mirror++;
2172 if (failrec->this_mirror == failed_mirror)
2173 failrec->this_mirror++;
2174 read_mode = READ_SYNC;
2177 if (!state || failrec->this_mirror > num_copies) {
2178 pr_debug("bio_readpage_error: (fail) state=%p, num_copies=%d, "
2179 "next_mirror %d, failed_mirror %d\n", state,
2180 num_copies, failrec->this_mirror, failed_mirror);
2181 free_io_failure(inode, failrec, 0);
2185 bio = bio_alloc(GFP_NOFS, 1);
2187 free_io_failure(inode, failrec, 0);
2190 bio->bi_private = state;
2191 bio->bi_end_io = failed_bio->bi_end_io;
2192 bio->bi_sector = failrec->logical >> 9;
2193 bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
2196 bio_add_page(bio, page, failrec->len, start - page_offset(page));
2198 pr_debug("bio_readpage_error: submitting new read[%#x] to "
2199 "this_mirror=%d, num_copies=%d, in_validation=%d\n", read_mode,
2200 failrec->this_mirror, num_copies, failrec->in_validation);
2202 ret = tree->ops->submit_bio_hook(inode, read_mode, bio,
2203 failrec->this_mirror,
2204 failrec->bio_flags, 0);
2208 /* lots and lots of room for performance fixes in the end_bio funcs */
2210 int end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2212 int uptodate = (err == 0);
2213 struct extent_io_tree *tree;
2216 tree = &BTRFS_I(page->mapping->host)->io_tree;
2218 if (tree->ops && tree->ops->writepage_end_io_hook) {
2219 ret = tree->ops->writepage_end_io_hook(page, start,
2220 end, NULL, uptodate);
2225 if (!uptodate && tree->ops &&
2226 tree->ops->writepage_io_failed_hook) {
2227 ret = tree->ops->writepage_io_failed_hook(NULL, page,
2229 /* Writeback already completed */
2235 clear_extent_uptodate(tree, start, end, NULL, GFP_NOFS);
2236 ClearPageUptodate(page);
2243 * after a writepage IO is done, we need to:
2244 * clear the uptodate bits on error
2245 * clear the writeback bits in the extent tree for this IO
2246 * end_page_writeback if the page has no more pending IO
2248 * Scheduling is not allowed, so the extent state tree is expected
2249 * to have one and only one object corresponding to this IO.
2251 static void end_bio_extent_writepage(struct bio *bio, int err)
2253 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2254 struct extent_io_tree *tree;
2260 struct page *page = bvec->bv_page;
2261 tree = &BTRFS_I(page->mapping->host)->io_tree;
2263 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
2265 end = start + bvec->bv_len - 1;
2267 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
2272 if (--bvec >= bio->bi_io_vec)
2273 prefetchw(&bvec->bv_page->flags);
2275 if (end_extent_writepage(page, err, start, end))
2279 end_page_writeback(page);
2281 check_page_writeback(tree, page);
2282 } while (bvec >= bio->bi_io_vec);
2288 * after a readpage IO is done, we need to:
2289 * clear the uptodate bits on error
2290 * set the uptodate bits if things worked
2291 * set the page up to date if all extents in the tree are uptodate
2292 * clear the lock bit in the extent tree
2293 * unlock the page if there are no other extents locked for it
2295 * Scheduling is not allowed, so the extent state tree is expected
2296 * to have one and only one object corresponding to this IO.
2298 static void end_bio_extent_readpage(struct bio *bio, int err)
2300 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
2301 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
2302 struct bio_vec *bvec = bio->bi_io_vec;
2303 struct extent_io_tree *tree;
2314 struct page *page = bvec->bv_page;
2315 struct extent_state *cached = NULL;
2316 struct extent_state *state;
2318 pr_debug("end_bio_extent_readpage: bi_vcnt=%d, idx=%d, err=%d, "
2319 "mirror=%ld\n", bio->bi_vcnt, bio->bi_idx, err,
2320 (long int)bio->bi_bdev);
2321 tree = &BTRFS_I(page->mapping->host)->io_tree;
2323 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
2325 end = start + bvec->bv_len - 1;
2327 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
2332 if (++bvec <= bvec_end)
2333 prefetchw(&bvec->bv_page->flags);
2335 spin_lock(&tree->lock);
2336 state = find_first_extent_bit_state(tree, start, EXTENT_LOCKED);
2337 if (state && state->start == start) {
2339 * take a reference on the state, unlock will drop
2342 cache_state(state, &cached);
2344 spin_unlock(&tree->lock);
2346 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
2347 ret = tree->ops->readpage_end_io_hook(page, start, end,
2352 clean_io_failure(start, page);
2356 failed_mirror = (int)(unsigned long)bio->bi_bdev;
2358 if (!uptodate && tree->ops && tree->ops->readpage_io_failed_hook) {
2359 ret = tree->ops->readpage_io_failed_hook(page, failed_mirror);
2361 test_bit(BIO_UPTODATE, &bio->bi_flags))
2363 } else if (!uptodate) {
2365 * The generic bio_readpage_error handles errors the
2366 * following way: If possible, new read requests are
2367 * created and submitted and will end up in
2368 * end_bio_extent_readpage as well (if we're lucky, not
2369 * in the !uptodate case). In that case it returns 0 and
2370 * we just go on with the next page in our bio. If it
2371 * can't handle the error it will return -EIO and we
2372 * remain responsible for that page.
2374 ret = bio_readpage_error(bio, page, start, end,
2375 failed_mirror, NULL);
2378 test_bit(BIO_UPTODATE, &bio->bi_flags);
2381 uncache_state(&cached);
2386 if (uptodate && tree->track_uptodate) {
2387 set_extent_uptodate(tree, start, end, &cached,
2390 unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC);
2394 SetPageUptodate(page);
2396 ClearPageUptodate(page);
2402 check_page_uptodate(tree, page);
2404 ClearPageUptodate(page);
2407 check_page_locked(tree, page);
2409 } while (bvec <= bvec_end);
2415 btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
2420 bio = bio_alloc(gfp_flags, nr_vecs);
2422 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
2423 while (!bio && (nr_vecs /= 2))
2424 bio = bio_alloc(gfp_flags, nr_vecs);
2429 bio->bi_bdev = bdev;
2430 bio->bi_sector = first_sector;
2436 * Since writes are async, they will only return -ENOMEM.
2437 * Reads can return the full range of I/O error conditions.
2439 static int __must_check submit_one_bio(int rw, struct bio *bio,
2440 int mirror_num, unsigned long bio_flags)
2443 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2444 struct page *page = bvec->bv_page;
2445 struct extent_io_tree *tree = bio->bi_private;
2448 start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
2450 bio->bi_private = NULL;
2454 if (tree->ops && tree->ops->submit_bio_hook)
2455 ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
2456 mirror_num, bio_flags, start);
2458 btrfsic_submit_bio(rw, bio);
2460 if (bio_flagged(bio, BIO_EOPNOTSUPP))
2466 static int merge_bio(struct extent_io_tree *tree, struct page *page,
2467 unsigned long offset, size_t size, struct bio *bio,
2468 unsigned long bio_flags)
2471 if (tree->ops && tree->ops->merge_bio_hook)
2472 ret = tree->ops->merge_bio_hook(page, offset, size, bio,
2479 static int submit_extent_page(int rw, struct extent_io_tree *tree,
2480 struct page *page, sector_t sector,
2481 size_t size, unsigned long offset,
2482 struct block_device *bdev,
2483 struct bio **bio_ret,
2484 unsigned long max_pages,
2485 bio_end_io_t end_io_func,
2487 unsigned long prev_bio_flags,
2488 unsigned long bio_flags)
2494 int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED;
2495 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
2496 size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE);
2498 if (bio_ret && *bio_ret) {
2501 contig = bio->bi_sector == sector;
2503 contig = bio->bi_sector + (bio->bi_size >> 9) ==
2506 if (prev_bio_flags != bio_flags || !contig ||
2507 merge_bio(tree, page, offset, page_size, bio, bio_flags) ||
2508 bio_add_page(bio, page, page_size, offset) < page_size) {
2509 ret = submit_one_bio(rw, bio, mirror_num,
2518 if (this_compressed)
2521 nr = bio_get_nr_vecs(bdev);
2523 bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
2527 bio_add_page(bio, page, page_size, offset);
2528 bio->bi_end_io = end_io_func;
2529 bio->bi_private = tree;
2534 ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
2539 void attach_extent_buffer_page(struct extent_buffer *eb, struct page *page)
2541 if (!PagePrivate(page)) {
2542 SetPagePrivate(page);
2543 page_cache_get(page);
2544 set_page_private(page, (unsigned long)eb);
2546 WARN_ON(page->private != (unsigned long)eb);
2550 void set_page_extent_mapped(struct page *page)
2552 if (!PagePrivate(page)) {
2553 SetPagePrivate(page);
2554 page_cache_get(page);
2555 set_page_private(page, EXTENT_PAGE_PRIVATE);
2560 * basic readpage implementation. Locked extent state structs are inserted
2561 * into the tree that are removed when the IO is done (by the end_io
2563 * XXX JDM: This needs looking at to ensure proper page locking
2565 static int __extent_read_full_page(struct extent_io_tree *tree,
2567 get_extent_t *get_extent,
2568 struct bio **bio, int mirror_num,
2569 unsigned long *bio_flags)
2571 struct inode *inode = page->mapping->host;
2572 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2573 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2577 u64 last_byte = i_size_read(inode);
2581 struct extent_map *em;
2582 struct block_device *bdev;
2583 struct btrfs_ordered_extent *ordered;
2586 size_t pg_offset = 0;
2588 size_t disk_io_size;
2589 size_t blocksize = inode->i_sb->s_blocksize;
2590 unsigned long this_bio_flag = 0;
2592 set_page_extent_mapped(page);
2594 if (!PageUptodate(page)) {
2595 if (cleancache_get_page(page) == 0) {
2596 BUG_ON(blocksize != PAGE_SIZE);
2603 lock_extent(tree, start, end);
2604 ordered = btrfs_lookup_ordered_extent(inode, start);
2607 unlock_extent(tree, start, end);
2608 btrfs_start_ordered_extent(inode, ordered, 1);
2609 btrfs_put_ordered_extent(ordered);
2612 if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
2614 size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
2617 iosize = PAGE_CACHE_SIZE - zero_offset;
2618 userpage = kmap_atomic(page, KM_USER0);
2619 memset(userpage + zero_offset, 0, iosize);
2620 flush_dcache_page(page);
2621 kunmap_atomic(userpage, KM_USER0);
2624 while (cur <= end) {
2625 if (cur >= last_byte) {
2627 struct extent_state *cached = NULL;
2629 iosize = PAGE_CACHE_SIZE - pg_offset;
2630 userpage = kmap_atomic(page, KM_USER0);
2631 memset(userpage + pg_offset, 0, iosize);
2632 flush_dcache_page(page);
2633 kunmap_atomic(userpage, KM_USER0);
2634 set_extent_uptodate(tree, cur, cur + iosize - 1,
2636 unlock_extent_cached(tree, cur, cur + iosize - 1,
2640 em = get_extent(inode, page, pg_offset, cur,
2642 if (IS_ERR_OR_NULL(em)) {
2644 unlock_extent(tree, cur, end);
2647 extent_offset = cur - em->start;
2648 BUG_ON(extent_map_end(em) <= cur);
2651 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2652 this_bio_flag = EXTENT_BIO_COMPRESSED;
2653 extent_set_compress_type(&this_bio_flag,
2657 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2658 cur_end = min(extent_map_end(em) - 1, end);
2659 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2660 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2661 disk_io_size = em->block_len;
2662 sector = em->block_start >> 9;
2664 sector = (em->block_start + extent_offset) >> 9;
2665 disk_io_size = iosize;
2668 block_start = em->block_start;
2669 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2670 block_start = EXTENT_MAP_HOLE;
2671 free_extent_map(em);
2674 /* we've found a hole, just zero and go on */
2675 if (block_start == EXTENT_MAP_HOLE) {
2677 struct extent_state *cached = NULL;
2679 userpage = kmap_atomic(page, KM_USER0);
2680 memset(userpage + pg_offset, 0, iosize);
2681 flush_dcache_page(page);
2682 kunmap_atomic(userpage, KM_USER0);
2684 set_extent_uptodate(tree, cur, cur + iosize - 1,
2686 unlock_extent_cached(tree, cur, cur + iosize - 1,
2689 pg_offset += iosize;
2692 /* the get_extent function already copied into the page */
2693 if (test_range_bit(tree, cur, cur_end,
2694 EXTENT_UPTODATE, 1, NULL)) {
2695 check_page_uptodate(tree, page);
2696 unlock_extent(tree, cur, cur + iosize - 1);
2698 pg_offset += iosize;
2701 /* we have an inline extent but it didn't get marked up
2702 * to date. Error out
2704 if (block_start == EXTENT_MAP_INLINE) {
2706 unlock_extent(tree, cur, cur + iosize - 1);
2708 pg_offset += iosize;
2713 if (tree->ops && tree->ops->readpage_io_hook) {
2714 ret = tree->ops->readpage_io_hook(page, cur,
2718 unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
2720 ret = submit_extent_page(READ, tree, page,
2721 sector, disk_io_size, pg_offset,
2723 end_bio_extent_readpage, mirror_num,
2726 BUG_ON(ret == -ENOMEM);
2728 *bio_flags = this_bio_flag;
2733 pg_offset += iosize;
2737 if (!PageError(page))
2738 SetPageUptodate(page);
2744 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
2745 get_extent_t *get_extent, int mirror_num)
2747 struct bio *bio = NULL;
2748 unsigned long bio_flags = 0;
2751 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
2754 ret = submit_one_bio(READ, bio, mirror_num, bio_flags);
2758 static noinline void update_nr_written(struct page *page,
2759 struct writeback_control *wbc,
2760 unsigned long nr_written)
2762 wbc->nr_to_write -= nr_written;
2763 if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
2764 wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
2765 page->mapping->writeback_index = page->index + nr_written;
2769 * the writepage semantics are similar to regular writepage. extent
2770 * records are inserted to lock ranges in the tree, and as dirty areas
2771 * are found, they are marked writeback. Then the lock bits are removed
2772 * and the end_io handler clears the writeback ranges
2774 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2777 struct inode *inode = page->mapping->host;
2778 struct extent_page_data *epd = data;
2779 struct extent_io_tree *tree = epd->tree;
2780 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2782 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2786 u64 last_byte = i_size_read(inode);
2790 struct extent_state *cached_state = NULL;
2791 struct extent_map *em;
2792 struct block_device *bdev;
2795 size_t pg_offset = 0;
2797 loff_t i_size = i_size_read(inode);
2798 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
2804 unsigned long nr_written = 0;
2805 bool fill_delalloc = true;
2807 if (wbc->sync_mode == WB_SYNC_ALL)
2808 write_flags = WRITE_SYNC;
2810 write_flags = WRITE;
2812 trace___extent_writepage(page, inode, wbc);
2814 WARN_ON(!PageLocked(page));
2816 ClearPageError(page);
2818 pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
2819 if (page->index > end_index ||
2820 (page->index == end_index && !pg_offset)) {
2821 page->mapping->a_ops->invalidatepage(page, 0);
2826 if (page->index == end_index) {
2829 userpage = kmap_atomic(page, KM_USER0);
2830 memset(userpage + pg_offset, 0,
2831 PAGE_CACHE_SIZE - pg_offset);
2832 kunmap_atomic(userpage, KM_USER0);
2833 flush_dcache_page(page);
2837 set_page_extent_mapped(page);
2839 if (!tree->ops || !tree->ops->fill_delalloc)
2840 fill_delalloc = false;
2842 delalloc_start = start;
2845 if (!epd->extent_locked && fill_delalloc) {
2846 u64 delalloc_to_write = 0;
2848 * make sure the wbc mapping index is at least updated
2851 update_nr_written(page, wbc, 0);
2853 while (delalloc_end < page_end) {
2854 nr_delalloc = find_lock_delalloc_range(inode, tree,
2859 if (nr_delalloc == 0) {
2860 delalloc_start = delalloc_end + 1;
2863 ret = tree->ops->fill_delalloc(inode, page,
2868 /* File system has been set read-only */
2874 * delalloc_end is already one less than the total
2875 * length, so we don't subtract one from
2878 delalloc_to_write += (delalloc_end - delalloc_start +
2881 delalloc_start = delalloc_end + 1;
2883 if (wbc->nr_to_write < delalloc_to_write) {
2886 if (delalloc_to_write < thresh * 2)
2887 thresh = delalloc_to_write;
2888 wbc->nr_to_write = min_t(u64, delalloc_to_write,
2892 /* did the fill delalloc function already unlock and start
2898 * we've unlocked the page, so we can't update
2899 * the mapping's writeback index, just update
2902 wbc->nr_to_write -= nr_written;
2906 if (tree->ops && tree->ops->writepage_start_hook) {
2907 ret = tree->ops->writepage_start_hook(page, start,
2910 /* Fixup worker will requeue */
2912 wbc->pages_skipped++;
2914 redirty_page_for_writepage(wbc, page);
2915 update_nr_written(page, wbc, nr_written);
2923 * we don't want to touch the inode after unlocking the page,
2924 * so we update the mapping writeback index now
2926 update_nr_written(page, wbc, nr_written + 1);
2929 if (last_byte <= start) {
2930 if (tree->ops && tree->ops->writepage_end_io_hook)
2931 tree->ops->writepage_end_io_hook(page, start,
2936 blocksize = inode->i_sb->s_blocksize;
2938 while (cur <= end) {
2939 if (cur >= last_byte) {
2940 if (tree->ops && tree->ops->writepage_end_io_hook)
2941 tree->ops->writepage_end_io_hook(page, cur,
2945 em = epd->get_extent(inode, page, pg_offset, cur,
2947 if (IS_ERR_OR_NULL(em)) {
2952 extent_offset = cur - em->start;
2953 BUG_ON(extent_map_end(em) <= cur);
2955 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2956 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2957 sector = (em->block_start + extent_offset) >> 9;
2959 block_start = em->block_start;
2960 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
2961 free_extent_map(em);
2965 * compressed and inline extents are written through other
2968 if (compressed || block_start == EXTENT_MAP_HOLE ||
2969 block_start == EXTENT_MAP_INLINE) {
2971 * end_io notification does not happen here for
2972 * compressed extents
2974 if (!compressed && tree->ops &&
2975 tree->ops->writepage_end_io_hook)
2976 tree->ops->writepage_end_io_hook(page, cur,
2979 else if (compressed) {
2980 /* we don't want to end_page_writeback on
2981 * a compressed extent. this happens
2988 pg_offset += iosize;
2991 /* leave this out until we have a page_mkwrite call */
2992 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
2993 EXTENT_DIRTY, 0, NULL)) {
2995 pg_offset += iosize;
2999 if (tree->ops && tree->ops->writepage_io_hook) {
3000 ret = tree->ops->writepage_io_hook(page, cur,
3008 unsigned long max_nr = end_index + 1;
3010 set_range_writeback(tree, cur, cur + iosize - 1);
3011 if (!PageWriteback(page)) {
3012 printk(KERN_ERR "btrfs warning page %lu not "
3013 "writeback, cur %llu end %llu\n",
3014 page->index, (unsigned long long)cur,
3015 (unsigned long long)end);
3018 ret = submit_extent_page(write_flags, tree, page,
3019 sector, iosize, pg_offset,
3020 bdev, &epd->bio, max_nr,
3021 end_bio_extent_writepage,
3027 pg_offset += iosize;
3032 /* make sure the mapping tag for page dirty gets cleared */
3033 set_page_writeback(page);
3034 end_page_writeback(page);
3040 /* drop our reference on any cached states */
3041 free_extent_state(cached_state);
3045 static int eb_wait(void *word)
3051 static void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3053 wait_on_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK, eb_wait,
3054 TASK_UNINTERRUPTIBLE);
3057 static int lock_extent_buffer_for_io(struct extent_buffer *eb,
3058 struct btrfs_fs_info *fs_info,
3059 struct extent_page_data *epd)
3061 unsigned long i, num_pages;
3065 if (!btrfs_try_tree_write_lock(eb)) {
3067 flush_write_bio(epd);
3068 btrfs_tree_lock(eb);
3071 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3072 btrfs_tree_unlock(eb);
3076 flush_write_bio(epd);
3080 wait_on_extent_buffer_writeback(eb);
3081 btrfs_tree_lock(eb);
3082 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3084 btrfs_tree_unlock(eb);
3088 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3089 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3090 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3091 spin_lock(&fs_info->delalloc_lock);
3092 if (fs_info->dirty_metadata_bytes >= eb->len)
3093 fs_info->dirty_metadata_bytes -= eb->len;
3096 spin_unlock(&fs_info->delalloc_lock);
3100 btrfs_tree_unlock(eb);
3105 num_pages = num_extent_pages(eb->start, eb->len);
3106 for (i = 0; i < num_pages; i++) {
3107 struct page *p = extent_buffer_page(eb, i);
3109 if (!trylock_page(p)) {
3111 flush_write_bio(epd);
3121 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3123 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3124 smp_mb__after_clear_bit();
3125 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3128 static void end_bio_extent_buffer_writepage(struct bio *bio, int err)
3130 int uptodate = err == 0;
3131 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
3132 struct extent_buffer *eb;
3136 struct page *page = bvec->bv_page;
3139 eb = (struct extent_buffer *)page->private;
3141 done = atomic_dec_and_test(&eb->io_pages);
3143 if (!uptodate || test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
3144 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3145 ClearPageUptodate(page);
3149 end_page_writeback(page);
3154 end_extent_buffer_writeback(eb);
3155 } while (bvec >= bio->bi_io_vec);
3161 static int write_one_eb(struct extent_buffer *eb,
3162 struct btrfs_fs_info *fs_info,
3163 struct writeback_control *wbc,
3164 struct extent_page_data *epd)
3166 struct block_device *bdev = fs_info->fs_devices->latest_bdev;
3167 u64 offset = eb->start;
3168 unsigned long i, num_pages;
3169 int rw = (epd->sync_io ? WRITE_SYNC : WRITE);
3172 clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3173 num_pages = num_extent_pages(eb->start, eb->len);
3174 atomic_set(&eb->io_pages, num_pages);
3175 for (i = 0; i < num_pages; i++) {
3176 struct page *p = extent_buffer_page(eb, i);
3178 clear_page_dirty_for_io(p);
3179 set_page_writeback(p);
3180 ret = submit_extent_page(rw, eb->tree, p, offset >> 9,
3181 PAGE_CACHE_SIZE, 0, bdev, &epd->bio,
3182 -1, end_bio_extent_buffer_writepage,
3185 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3187 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3188 end_extent_buffer_writeback(eb);
3192 offset += PAGE_CACHE_SIZE;
3193 update_nr_written(p, wbc, 1);
3197 if (unlikely(ret)) {
3198 for (; i < num_pages; i++) {
3199 struct page *p = extent_buffer_page(eb, i);
3207 int btree_write_cache_pages(struct address_space *mapping,
3208 struct writeback_control *wbc)
3210 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
3211 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
3212 struct extent_buffer *eb, *prev_eb = NULL;
3213 struct extent_page_data epd = {
3217 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3221 int nr_to_write_done = 0;
3222 struct pagevec pvec;
3225 pgoff_t end; /* Inclusive */
3229 pagevec_init(&pvec, 0);
3230 if (wbc->range_cyclic) {
3231 index = mapping->writeback_index; /* Start from prev offset */
3234 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3235 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3238 if (wbc->sync_mode == WB_SYNC_ALL)
3239 tag = PAGECACHE_TAG_TOWRITE;
3241 tag = PAGECACHE_TAG_DIRTY;
3243 if (wbc->sync_mode == WB_SYNC_ALL)
3244 tag_pages_for_writeback(mapping, index, end);
3245 while (!done && !nr_to_write_done && (index <= end) &&
3246 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3247 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3251 for (i = 0; i < nr_pages; i++) {
3252 struct page *page = pvec.pages[i];
3254 if (!PagePrivate(page))
3257 if (!wbc->range_cyclic && page->index > end) {
3262 eb = (struct extent_buffer *)page->private;
3271 if (!atomic_inc_not_zero(&eb->refs)) {
3277 ret = lock_extent_buffer_for_io(eb, fs_info, &epd);
3279 free_extent_buffer(eb);
3283 ret = write_one_eb(eb, fs_info, wbc, &epd);
3286 free_extent_buffer(eb);
3289 free_extent_buffer(eb);
3292 * the filesystem may choose to bump up nr_to_write.
3293 * We have to make sure to honor the new nr_to_write
3296 nr_to_write_done = wbc->nr_to_write <= 0;
3298 pagevec_release(&pvec);
3301 if (!scanned && !done) {
3303 * We hit the last page and there is more work to be done: wrap
3304 * back to the start of the file
3310 flush_write_bio(&epd);
3315 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3316 * @mapping: address space structure to write
3317 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3318 * @writepage: function called for each page
3319 * @data: data passed to writepage function
3321 * If a page is already under I/O, write_cache_pages() skips it, even
3322 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3323 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3324 * and msync() need to guarantee that all the data which was dirty at the time
3325 * the call was made get new I/O started against them. If wbc->sync_mode is
3326 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3327 * existing IO to complete.
3329 static int extent_write_cache_pages(struct extent_io_tree *tree,
3330 struct address_space *mapping,
3331 struct writeback_control *wbc,
3332 writepage_t writepage, void *data,
3333 void (*flush_fn)(void *))
3337 int nr_to_write_done = 0;
3338 struct pagevec pvec;
3341 pgoff_t end; /* Inclusive */
3345 pagevec_init(&pvec, 0);
3346 if (wbc->range_cyclic) {
3347 index = mapping->writeback_index; /* Start from prev offset */
3350 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3351 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3354 if (wbc->sync_mode == WB_SYNC_ALL)
3355 tag = PAGECACHE_TAG_TOWRITE;
3357 tag = PAGECACHE_TAG_DIRTY;
3359 if (wbc->sync_mode == WB_SYNC_ALL)
3360 tag_pages_for_writeback(mapping, index, end);
3361 while (!done && !nr_to_write_done && (index <= end) &&
3362 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3363 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3367 for (i = 0; i < nr_pages; i++) {
3368 struct page *page = pvec.pages[i];
3371 * At this point we hold neither mapping->tree_lock nor
3372 * lock on the page itself: the page may be truncated or
3373 * invalidated (changing page->mapping to NULL), or even
3374 * swizzled back from swapper_space to tmpfs file
3378 tree->ops->write_cache_pages_lock_hook) {
3379 tree->ops->write_cache_pages_lock_hook(page,
3382 if (!trylock_page(page)) {
3388 if (unlikely(page->mapping != mapping)) {
3393 if (!wbc->range_cyclic && page->index > end) {
3399 if (wbc->sync_mode != WB_SYNC_NONE) {
3400 if (PageWriteback(page))
3402 wait_on_page_writeback(page);
3405 if (PageWriteback(page) ||
3406 !clear_page_dirty_for_io(page)) {
3411 ret = (*writepage)(page, wbc, data);
3413 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
3421 * the filesystem may choose to bump up nr_to_write.
3422 * We have to make sure to honor the new nr_to_write
3425 nr_to_write_done = wbc->nr_to_write <= 0;
3427 pagevec_release(&pvec);
3430 if (!scanned && !done) {
3432 * We hit the last page and there is more work to be done: wrap
3433 * back to the start of the file
3442 static void flush_epd_write_bio(struct extent_page_data *epd)
3451 ret = submit_one_bio(rw, epd->bio, 0, 0);
3452 BUG_ON(ret < 0); /* -ENOMEM */
3457 static noinline void flush_write_bio(void *data)
3459 struct extent_page_data *epd = data;
3460 flush_epd_write_bio(epd);
3463 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
3464 get_extent_t *get_extent,
3465 struct writeback_control *wbc)
3468 struct extent_page_data epd = {
3471 .get_extent = get_extent,
3473 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3476 ret = __extent_writepage(page, wbc, &epd);
3478 flush_epd_write_bio(&epd);
3482 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
3483 u64 start, u64 end, get_extent_t *get_extent,
3487 struct address_space *mapping = inode->i_mapping;
3489 unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
3492 struct extent_page_data epd = {
3495 .get_extent = get_extent,
3497 .sync_io = mode == WB_SYNC_ALL,
3499 struct writeback_control wbc_writepages = {
3501 .nr_to_write = nr_pages * 2,
3502 .range_start = start,
3503 .range_end = end + 1,
3506 while (start <= end) {
3507 page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
3508 if (clear_page_dirty_for_io(page))
3509 ret = __extent_writepage(page, &wbc_writepages, &epd);
3511 if (tree->ops && tree->ops->writepage_end_io_hook)
3512 tree->ops->writepage_end_io_hook(page, start,
3513 start + PAGE_CACHE_SIZE - 1,
3517 page_cache_release(page);
3518 start += PAGE_CACHE_SIZE;
3521 flush_epd_write_bio(&epd);
3525 int extent_writepages(struct extent_io_tree *tree,
3526 struct address_space *mapping,
3527 get_extent_t *get_extent,
3528 struct writeback_control *wbc)
3531 struct extent_page_data epd = {
3534 .get_extent = get_extent,
3536 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3539 ret = extent_write_cache_pages(tree, mapping, wbc,
3540 __extent_writepage, &epd,
3542 flush_epd_write_bio(&epd);
3546 int extent_readpages(struct extent_io_tree *tree,
3547 struct address_space *mapping,
3548 struct list_head *pages, unsigned nr_pages,
3549 get_extent_t get_extent)
3551 struct bio *bio = NULL;
3553 unsigned long bio_flags = 0;
3555 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
3556 struct page *page = list_entry(pages->prev, struct page, lru);
3558 prefetchw(&page->flags);
3559 list_del(&page->lru);
3560 if (!add_to_page_cache_lru(page, mapping,
3561 page->index, GFP_NOFS)) {
3562 __extent_read_full_page(tree, page, get_extent,
3563 &bio, 0, &bio_flags);
3565 page_cache_release(page);
3567 BUG_ON(!list_empty(pages));
3569 return submit_one_bio(READ, bio, 0, bio_flags);
3574 * basic invalidatepage code, this waits on any locked or writeback
3575 * ranges corresponding to the page, and then deletes any extent state
3576 * records from the tree
3578 int extent_invalidatepage(struct extent_io_tree *tree,
3579 struct page *page, unsigned long offset)
3581 struct extent_state *cached_state = NULL;
3582 u64 start = ((u64)page->index << PAGE_CACHE_SHIFT);
3583 u64 end = start + PAGE_CACHE_SIZE - 1;
3584 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
3586 start += (offset + blocksize - 1) & ~(blocksize - 1);
3590 lock_extent_bits(tree, start, end, 0, &cached_state);
3591 wait_on_page_writeback(page);
3592 clear_extent_bit(tree, start, end,
3593 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
3594 EXTENT_DO_ACCOUNTING,
3595 1, 1, &cached_state, GFP_NOFS);
3600 * a helper for releasepage, this tests for areas of the page that
3601 * are locked or under IO and drops the related state bits if it is safe
3604 int try_release_extent_state(struct extent_map_tree *map,
3605 struct extent_io_tree *tree, struct page *page,
3608 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
3609 u64 end = start + PAGE_CACHE_SIZE - 1;
3612 if (test_range_bit(tree, start, end,
3613 EXTENT_IOBITS, 0, NULL))
3616 if ((mask & GFP_NOFS) == GFP_NOFS)
3619 * at this point we can safely clear everything except the
3620 * locked bit and the nodatasum bit
3622 ret = clear_extent_bit(tree, start, end,
3623 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
3626 /* if clear_extent_bit failed for enomem reasons,
3627 * we can't allow the release to continue.
3638 * a helper for releasepage. As long as there are no locked extents
3639 * in the range corresponding to the page, both state records and extent
3640 * map records are removed
3642 int try_release_extent_mapping(struct extent_map_tree *map,
3643 struct extent_io_tree *tree, struct page *page,
3646 struct extent_map *em;
3647 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
3648 u64 end = start + PAGE_CACHE_SIZE - 1;
3650 if ((mask & __GFP_WAIT) &&
3651 page->mapping->host->i_size > 16 * 1024 * 1024) {
3653 while (start <= end) {
3654 len = end - start + 1;
3655 write_lock(&map->lock);
3656 em = lookup_extent_mapping(map, start, len);
3658 write_unlock(&map->lock);
3661 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
3662 em->start != start) {
3663 write_unlock(&map->lock);
3664 free_extent_map(em);
3667 if (!test_range_bit(tree, em->start,
3668 extent_map_end(em) - 1,
3669 EXTENT_LOCKED | EXTENT_WRITEBACK,
3671 remove_extent_mapping(map, em);
3672 /* once for the rb tree */
3673 free_extent_map(em);
3675 start = extent_map_end(em);
3676 write_unlock(&map->lock);
3679 free_extent_map(em);
3682 return try_release_extent_state(map, tree, page, mask);
3686 * helper function for fiemap, which doesn't want to see any holes.
3687 * This maps until we find something past 'last'
3689 static struct extent_map *get_extent_skip_holes(struct inode *inode,
3692 get_extent_t *get_extent)
3694 u64 sectorsize = BTRFS_I(inode)->root->sectorsize;
3695 struct extent_map *em;
3702 len = last - offset;
3705 len = (len + sectorsize - 1) & ~(sectorsize - 1);
3706 em = get_extent(inode, NULL, 0, offset, len, 0);
3707 if (IS_ERR_OR_NULL(em))
3710 /* if this isn't a hole return it */
3711 if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) &&
3712 em->block_start != EXTENT_MAP_HOLE) {
3716 /* this is a hole, advance to the next extent */
3717 offset = extent_map_end(em);
3718 free_extent_map(em);
3725 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
3726 __u64 start, __u64 len, get_extent_t *get_extent)
3730 u64 max = start + len;
3734 u64 last_for_get_extent = 0;
3736 u64 isize = i_size_read(inode);
3737 struct btrfs_key found_key;
3738 struct extent_map *em = NULL;
3739 struct extent_state *cached_state = NULL;
3740 struct btrfs_path *path;
3741 struct btrfs_file_extent_item *item;
3746 unsigned long emflags;
3751 path = btrfs_alloc_path();
3754 path->leave_spinning = 1;
3756 start = ALIGN(start, BTRFS_I(inode)->root->sectorsize);
3757 len = ALIGN(len, BTRFS_I(inode)->root->sectorsize);
3760 * lookup the last file extent. We're not using i_size here
3761 * because there might be preallocation past i_size
3763 ret = btrfs_lookup_file_extent(NULL, BTRFS_I(inode)->root,
3764 path, btrfs_ino(inode), -1, 0);
3766 btrfs_free_path(path);
3771 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3772 struct btrfs_file_extent_item);
3773 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
3774 found_type = btrfs_key_type(&found_key);
3776 /* No extents, but there might be delalloc bits */
3777 if (found_key.objectid != btrfs_ino(inode) ||
3778 found_type != BTRFS_EXTENT_DATA_KEY) {
3779 /* have to trust i_size as the end */
3781 last_for_get_extent = isize;
3784 * remember the start of the last extent. There are a
3785 * bunch of different factors that go into the length of the
3786 * extent, so its much less complex to remember where it started
3788 last = found_key.offset;
3789 last_for_get_extent = last + 1;
3791 btrfs_free_path(path);
3794 * we might have some extents allocated but more delalloc past those
3795 * extents. so, we trust isize unless the start of the last extent is
3800 last_for_get_extent = isize;
3803 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len, 0,
3806 em = get_extent_skip_holes(inode, start, last_for_get_extent,
3816 u64 offset_in_extent;
3818 /* break if the extent we found is outside the range */
3819 if (em->start >= max || extent_map_end(em) < off)
3823 * get_extent may return an extent that starts before our
3824 * requested range. We have to make sure the ranges
3825 * we return to fiemap always move forward and don't
3826 * overlap, so adjust the offsets here
3828 em_start = max(em->start, off);
3831 * record the offset from the start of the extent
3832 * for adjusting the disk offset below
3834 offset_in_extent = em_start - em->start;
3835 em_end = extent_map_end(em);
3836 em_len = em_end - em_start;
3837 emflags = em->flags;
3842 * bump off for our next call to get_extent
3844 off = extent_map_end(em);
3848 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
3850 flags |= FIEMAP_EXTENT_LAST;
3851 } else if (em->block_start == EXTENT_MAP_INLINE) {
3852 flags |= (FIEMAP_EXTENT_DATA_INLINE |
3853 FIEMAP_EXTENT_NOT_ALIGNED);
3854 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
3855 flags |= (FIEMAP_EXTENT_DELALLOC |
3856 FIEMAP_EXTENT_UNKNOWN);
3858 disko = em->block_start + offset_in_extent;
3860 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
3861 flags |= FIEMAP_EXTENT_ENCODED;
3863 free_extent_map(em);
3865 if ((em_start >= last) || em_len == (u64)-1 ||
3866 (last == (u64)-1 && isize <= em_end)) {
3867 flags |= FIEMAP_EXTENT_LAST;
3871 /* now scan forward to see if this is really the last extent. */
3872 em = get_extent_skip_holes(inode, off, last_for_get_extent,
3879 flags |= FIEMAP_EXTENT_LAST;
3882 ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
3888 free_extent_map(em);
3890 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len,
3891 &cached_state, GFP_NOFS);
3895 inline struct page *extent_buffer_page(struct extent_buffer *eb,
3898 return eb->pages[i];
3901 inline unsigned long num_extent_pages(u64 start, u64 len)
3903 return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
3904 (start >> PAGE_CACHE_SHIFT);
3907 static void __free_extent_buffer(struct extent_buffer *eb)
3910 unsigned long flags;
3911 spin_lock_irqsave(&leak_lock, flags);
3912 list_del(&eb->leak_list);
3913 spin_unlock_irqrestore(&leak_lock, flags);
3915 if (eb->pages && eb->pages != eb->inline_pages)
3917 kmem_cache_free(extent_buffer_cache, eb);
3920 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
3925 struct extent_buffer *eb = NULL;
3927 unsigned long flags;
3930 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
3936 rwlock_init(&eb->lock);
3937 atomic_set(&eb->write_locks, 0);
3938 atomic_set(&eb->read_locks, 0);
3939 atomic_set(&eb->blocking_readers, 0);
3940 atomic_set(&eb->blocking_writers, 0);
3941 atomic_set(&eb->spinning_readers, 0);
3942 atomic_set(&eb->spinning_writers, 0);
3943 eb->lock_nested = 0;
3944 init_waitqueue_head(&eb->write_lock_wq);
3945 init_waitqueue_head(&eb->read_lock_wq);
3948 spin_lock_irqsave(&leak_lock, flags);
3949 list_add(&eb->leak_list, &buffers);
3950 spin_unlock_irqrestore(&leak_lock, flags);
3952 spin_lock_init(&eb->refs_lock);
3953 atomic_set(&eb->refs, 1);
3954 atomic_set(&eb->io_pages, 0);
3956 if (len > MAX_INLINE_EXTENT_BUFFER_SIZE) {
3957 struct page **pages;
3958 int num_pages = (len + PAGE_CACHE_SIZE - 1) >>
3960 pages = kzalloc(num_pages, mask);
3962 __free_extent_buffer(eb);
3967 eb->pages = eb->inline_pages;
3973 static int extent_buffer_under_io(struct extent_buffer *eb)
3975 return (atomic_read(&eb->io_pages) ||
3976 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
3977 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
3981 * Helper for releasing extent buffer page.
3983 static void btrfs_release_extent_buffer_page(struct extent_buffer *eb,
3984 unsigned long start_idx)
3986 unsigned long index;
3989 BUG_ON(extent_buffer_under_io(eb));
3991 index = num_extent_pages(eb->start, eb->len);
3992 if (start_idx >= index)
3997 page = extent_buffer_page(eb, index);
3999 spin_lock(&page->mapping->private_lock);
4001 * We do this since we'll remove the pages after we've
4002 * removed the eb from the radix tree, so we could race
4003 * and have this page now attached to the new eb. So
4004 * only clear page_private if it's still connected to
4007 if (PagePrivate(page) &&
4008 page->private == (unsigned long)eb) {
4009 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4010 BUG_ON(PageDirty(page));
4011 BUG_ON(PageWriteback(page));
4013 * We need to make sure we haven't be attached
4016 ClearPagePrivate(page);
4017 set_page_private(page, 0);
4018 /* One for the page private */
4019 page_cache_release(page);
4021 spin_unlock(&page->mapping->private_lock);
4023 /* One for when we alloced the page */
4024 page_cache_release(page);
4026 } while (index != start_idx);
4030 * Helper for releasing the extent buffer.
4032 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4034 btrfs_release_extent_buffer_page(eb, 0);
4035 __free_extent_buffer(eb);
4038 static void check_buffer_tree_ref(struct extent_buffer *eb)
4040 /* the ref bit is tricky. We have to make sure it is set
4041 * if we have the buffer dirty. Otherwise the
4042 * code to free a buffer can end up dropping a dirty
4045 * Once the ref bit is set, it won't go away while the
4046 * buffer is dirty or in writeback, and it also won't
4047 * go away while we have the reference count on the
4050 * We can't just set the ref bit without bumping the
4051 * ref on the eb because free_extent_buffer might
4052 * see the ref bit and try to clear it. If this happens
4053 * free_extent_buffer might end up dropping our original
4054 * ref by mistake and freeing the page before we are able
4055 * to add one more ref.
4057 * So bump the ref count first, then set the bit. If someone
4058 * beat us to it, drop the ref we added.
4060 if (!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
4061 atomic_inc(&eb->refs);
4062 if (test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4063 atomic_dec(&eb->refs);
4067 static void mark_extent_buffer_accessed(struct extent_buffer *eb)
4069 unsigned long num_pages, i;
4071 check_buffer_tree_ref(eb);
4073 num_pages = num_extent_pages(eb->start, eb->len);
4074 for (i = 0; i < num_pages; i++) {
4075 struct page *p = extent_buffer_page(eb, i);
4076 mark_page_accessed(p);
4080 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
4081 u64 start, unsigned long len)
4083 unsigned long num_pages = num_extent_pages(start, len);
4085 unsigned long index = start >> PAGE_CACHE_SHIFT;
4086 struct extent_buffer *eb;
4087 struct extent_buffer *exists = NULL;
4089 struct address_space *mapping = tree->mapping;
4094 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
4095 if (eb && atomic_inc_not_zero(&eb->refs)) {
4097 mark_extent_buffer_accessed(eb);
4102 eb = __alloc_extent_buffer(tree, start, len, GFP_NOFS);
4106 for (i = 0; i < num_pages; i++, index++) {
4107 p = find_or_create_page(mapping, index, GFP_NOFS);
4113 spin_lock(&mapping->private_lock);
4114 if (PagePrivate(p)) {
4116 * We could have already allocated an eb for this page
4117 * and attached one so lets see if we can get a ref on
4118 * the existing eb, and if we can we know it's good and
4119 * we can just return that one, else we know we can just
4120 * overwrite page->private.
4122 exists = (struct extent_buffer *)p->private;
4123 if (atomic_inc_not_zero(&exists->refs)) {
4124 spin_unlock(&mapping->private_lock);
4126 mark_extent_buffer_accessed(exists);
4131 * Do this so attach doesn't complain and we need to
4132 * drop the ref the old guy had.
4134 ClearPagePrivate(p);
4135 WARN_ON(PageDirty(p));
4136 page_cache_release(p);
4138 attach_extent_buffer_page(eb, p);
4139 spin_unlock(&mapping->private_lock);
4140 WARN_ON(PageDirty(p));
4141 mark_page_accessed(p);
4143 if (!PageUptodate(p))
4147 * see below about how we avoid a nasty race with release page
4148 * and why we unlock later
4152 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4154 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
4158 spin_lock(&tree->buffer_lock);
4159 ret = radix_tree_insert(&tree->buffer, start >> PAGE_CACHE_SHIFT, eb);
4160 if (ret == -EEXIST) {
4161 exists = radix_tree_lookup(&tree->buffer,
4162 start >> PAGE_CACHE_SHIFT);
4163 if (!atomic_inc_not_zero(&exists->refs)) {
4164 spin_unlock(&tree->buffer_lock);
4165 radix_tree_preload_end();
4169 spin_unlock(&tree->buffer_lock);
4170 radix_tree_preload_end();
4171 mark_extent_buffer_accessed(exists);
4174 /* add one reference for the tree */
4175 spin_lock(&eb->refs_lock);
4176 check_buffer_tree_ref(eb);
4177 spin_unlock(&eb->refs_lock);
4178 spin_unlock(&tree->buffer_lock);
4179 radix_tree_preload_end();
4182 * there is a race where release page may have
4183 * tried to find this extent buffer in the radix
4184 * but failed. It will tell the VM it is safe to
4185 * reclaim the, and it will clear the page private bit.
4186 * We must make sure to set the page private bit properly
4187 * after the extent buffer is in the radix tree so
4188 * it doesn't get lost
4190 SetPageChecked(eb->pages[0]);
4191 for (i = 1; i < num_pages; i++) {
4192 p = extent_buffer_page(eb, i);
4193 ClearPageChecked(p);
4196 unlock_page(eb->pages[0]);
4200 for (i = 0; i < num_pages; i++) {
4202 unlock_page(eb->pages[i]);
4205 if (!atomic_dec_and_test(&eb->refs))
4207 btrfs_release_extent_buffer(eb);
4211 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
4212 u64 start, unsigned long len)
4214 struct extent_buffer *eb;
4217 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
4218 if (eb && atomic_inc_not_zero(&eb->refs)) {
4220 mark_extent_buffer_accessed(eb);
4228 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
4230 struct extent_buffer *eb =
4231 container_of(head, struct extent_buffer, rcu_head);
4233 __free_extent_buffer(eb);
4236 /* Expects to have eb->eb_lock already held */
4237 static void release_extent_buffer(struct extent_buffer *eb, gfp_t mask)
4239 WARN_ON(atomic_read(&eb->refs) == 0);
4240 if (atomic_dec_and_test(&eb->refs)) {
4241 struct extent_io_tree *tree = eb->tree;
4243 spin_unlock(&eb->refs_lock);
4245 spin_lock(&tree->buffer_lock);
4246 radix_tree_delete(&tree->buffer,
4247 eb->start >> PAGE_CACHE_SHIFT);
4248 spin_unlock(&tree->buffer_lock);
4250 /* Should be safe to release our pages at this point */
4251 btrfs_release_extent_buffer_page(eb, 0);
4253 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
4256 spin_unlock(&eb->refs_lock);
4259 void free_extent_buffer(struct extent_buffer *eb)
4264 spin_lock(&eb->refs_lock);
4265 if (atomic_read(&eb->refs) == 2 &&
4266 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
4267 !extent_buffer_under_io(eb) &&
4268 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4269 atomic_dec(&eb->refs);
4272 * I know this is terrible, but it's temporary until we stop tracking
4273 * the uptodate bits and such for the extent buffers.
4275 release_extent_buffer(eb, GFP_ATOMIC);
4278 void free_extent_buffer_stale(struct extent_buffer *eb)
4283 spin_lock(&eb->refs_lock);
4284 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
4286 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
4287 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4288 atomic_dec(&eb->refs);
4289 release_extent_buffer(eb, GFP_NOFS);
4292 void clear_extent_buffer_dirty(struct extent_buffer *eb)
4295 unsigned long num_pages;
4298 num_pages = num_extent_pages(eb->start, eb->len);
4300 for (i = 0; i < num_pages; i++) {
4301 page = extent_buffer_page(eb, i);
4302 if (!PageDirty(page))
4306 WARN_ON(!PagePrivate(page));
4308 clear_page_dirty_for_io(page);
4309 spin_lock_irq(&page->mapping->tree_lock);
4310 if (!PageDirty(page)) {
4311 radix_tree_tag_clear(&page->mapping->page_tree,
4313 PAGECACHE_TAG_DIRTY);
4315 spin_unlock_irq(&page->mapping->tree_lock);
4316 ClearPageError(page);
4319 WARN_ON(atomic_read(&eb->refs) == 0);
4322 int set_extent_buffer_dirty(struct extent_buffer *eb)
4325 unsigned long num_pages;
4328 check_buffer_tree_ref(eb);
4330 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
4332 num_pages = num_extent_pages(eb->start, eb->len);
4333 WARN_ON(atomic_read(&eb->refs) == 0);
4334 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
4336 for (i = 0; i < num_pages; i++)
4337 set_page_dirty(extent_buffer_page(eb, i));
4341 static int range_straddles_pages(u64 start, u64 len)
4343 if (len < PAGE_CACHE_SIZE)
4345 if (start & (PAGE_CACHE_SIZE - 1))
4347 if ((start + len) & (PAGE_CACHE_SIZE - 1))
4352 int clear_extent_buffer_uptodate(struct extent_buffer *eb)
4356 unsigned long num_pages;
4358 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4359 num_pages = num_extent_pages(eb->start, eb->len);
4360 for (i = 0; i < num_pages; i++) {
4361 page = extent_buffer_page(eb, i);
4363 ClearPageUptodate(page);
4368 int set_extent_buffer_uptodate(struct extent_buffer *eb)
4372 unsigned long num_pages;
4374 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4375 num_pages = num_extent_pages(eb->start, eb->len);
4376 for (i = 0; i < num_pages; i++) {
4377 page = extent_buffer_page(eb, i);
4378 SetPageUptodate(page);
4383 int extent_range_uptodate(struct extent_io_tree *tree,
4388 int pg_uptodate = 1;
4390 unsigned long index;
4392 if (range_straddles_pages(start, end - start + 1)) {
4393 ret = test_range_bit(tree, start, end,
4394 EXTENT_UPTODATE, 1, NULL);
4398 while (start <= end) {
4399 index = start >> PAGE_CACHE_SHIFT;
4400 page = find_get_page(tree->mapping, index);
4403 uptodate = PageUptodate(page);
4404 page_cache_release(page);
4409 start += PAGE_CACHE_SIZE;
4414 int extent_buffer_uptodate(struct extent_buffer *eb)
4416 return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4419 int read_extent_buffer_pages(struct extent_io_tree *tree,
4420 struct extent_buffer *eb, u64 start, int wait,
4421 get_extent_t *get_extent, int mirror_num)
4424 unsigned long start_i;
4428 int locked_pages = 0;
4429 int all_uptodate = 1;
4430 unsigned long num_pages;
4431 unsigned long num_reads = 0;
4432 struct bio *bio = NULL;
4433 unsigned long bio_flags = 0;
4435 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
4439 WARN_ON(start < eb->start);
4440 start_i = (start >> PAGE_CACHE_SHIFT) -
4441 (eb->start >> PAGE_CACHE_SHIFT);
4446 num_pages = num_extent_pages(eb->start, eb->len);
4447 for (i = start_i; i < num_pages; i++) {
4448 page = extent_buffer_page(eb, i);
4449 if (wait == WAIT_NONE) {
4450 if (!trylock_page(page))
4456 if (!PageUptodate(page)) {
4463 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4467 clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
4468 eb->failed_mirror = 0;
4469 atomic_set(&eb->io_pages, num_reads);
4470 for (i = start_i; i < num_pages; i++) {
4471 page = extent_buffer_page(eb, i);
4472 if (!PageUptodate(page)) {
4473 ClearPageError(page);
4474 err = __extent_read_full_page(tree, page,
4476 mirror_num, &bio_flags);
4485 err = submit_one_bio(READ, bio, mirror_num, bio_flags);
4490 if (ret || wait != WAIT_COMPLETE)
4493 for (i = start_i; i < num_pages; i++) {
4494 page = extent_buffer_page(eb, i);
4495 wait_on_page_locked(page);
4496 if (!PageUptodate(page))
4504 while (locked_pages > 0) {
4505 page = extent_buffer_page(eb, i);
4513 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
4514 unsigned long start,
4521 char *dst = (char *)dstv;
4522 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4523 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4525 WARN_ON(start > eb->len);
4526 WARN_ON(start + len > eb->start + eb->len);
4528 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4531 page = extent_buffer_page(eb, i);
4533 cur = min(len, (PAGE_CACHE_SIZE - offset));
4534 kaddr = page_address(page);
4535 memcpy(dst, kaddr + offset, cur);
4544 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
4545 unsigned long min_len, char **map,
4546 unsigned long *map_start,
4547 unsigned long *map_len)
4549 size_t offset = start & (PAGE_CACHE_SIZE - 1);
4552 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4553 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4554 unsigned long end_i = (start_offset + start + min_len - 1) >>
4561 offset = start_offset;
4565 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
4568 if (start + min_len > eb->len) {
4569 printk(KERN_ERR "btrfs bad mapping eb start %llu len %lu, "
4570 "wanted %lu %lu\n", (unsigned long long)eb->start,
4571 eb->len, start, min_len);
4576 p = extent_buffer_page(eb, i);
4577 kaddr = page_address(p);
4578 *map = kaddr + offset;
4579 *map_len = PAGE_CACHE_SIZE - offset;
4583 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
4584 unsigned long start,
4591 char *ptr = (char *)ptrv;
4592 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4593 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4596 WARN_ON(start > eb->len);
4597 WARN_ON(start + len > eb->start + eb->len);
4599 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4602 page = extent_buffer_page(eb, i);
4604 cur = min(len, (PAGE_CACHE_SIZE - offset));
4606 kaddr = page_address(page);
4607 ret = memcmp(ptr, kaddr + offset, cur);
4619 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
4620 unsigned long start, unsigned long len)
4626 char *src = (char *)srcv;
4627 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4628 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4630 WARN_ON(start > eb->len);
4631 WARN_ON(start + len > eb->start + eb->len);
4633 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4636 page = extent_buffer_page(eb, i);
4637 WARN_ON(!PageUptodate(page));
4639 cur = min(len, PAGE_CACHE_SIZE - offset);
4640 kaddr = page_address(page);
4641 memcpy(kaddr + offset, src, cur);
4650 void memset_extent_buffer(struct extent_buffer *eb, char c,
4651 unsigned long start, unsigned long len)
4657 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4658 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4660 WARN_ON(start > eb->len);
4661 WARN_ON(start + len > eb->start + eb->len);
4663 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4666 page = extent_buffer_page(eb, i);
4667 WARN_ON(!PageUptodate(page));
4669 cur = min(len, PAGE_CACHE_SIZE - offset);
4670 kaddr = page_address(page);
4671 memset(kaddr + offset, c, cur);
4679 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
4680 unsigned long dst_offset, unsigned long src_offset,
4683 u64 dst_len = dst->len;
4688 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4689 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
4691 WARN_ON(src->len != dst_len);
4693 offset = (start_offset + dst_offset) &
4694 ((unsigned long)PAGE_CACHE_SIZE - 1);
4697 page = extent_buffer_page(dst, i);
4698 WARN_ON(!PageUptodate(page));
4700 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
4702 kaddr = page_address(page);
4703 read_extent_buffer(src, kaddr + offset, src_offset, cur);
4712 static void move_pages(struct page *dst_page, struct page *src_page,
4713 unsigned long dst_off, unsigned long src_off,
4716 char *dst_kaddr = page_address(dst_page);
4717 if (dst_page == src_page) {
4718 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
4720 char *src_kaddr = page_address(src_page);
4721 char *p = dst_kaddr + dst_off + len;
4722 char *s = src_kaddr + src_off + len;
4729 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
4731 unsigned long distance = (src > dst) ? src - dst : dst - src;
4732 return distance < len;
4735 static void copy_pages(struct page *dst_page, struct page *src_page,
4736 unsigned long dst_off, unsigned long src_off,
4739 char *dst_kaddr = page_address(dst_page);
4741 int must_memmove = 0;
4743 if (dst_page != src_page) {
4744 src_kaddr = page_address(src_page);
4746 src_kaddr = dst_kaddr;
4747 if (areas_overlap(src_off, dst_off, len))
4752 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
4754 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
4757 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
4758 unsigned long src_offset, unsigned long len)
4761 size_t dst_off_in_page;
4762 size_t src_off_in_page;
4763 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4764 unsigned long dst_i;
4765 unsigned long src_i;
4767 if (src_offset + len > dst->len) {
4768 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
4769 "len %lu dst len %lu\n", src_offset, len, dst->len);
4772 if (dst_offset + len > dst->len) {
4773 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
4774 "len %lu dst len %lu\n", dst_offset, len, dst->len);
4779 dst_off_in_page = (start_offset + dst_offset) &
4780 ((unsigned long)PAGE_CACHE_SIZE - 1);
4781 src_off_in_page = (start_offset + src_offset) &
4782 ((unsigned long)PAGE_CACHE_SIZE - 1);
4784 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
4785 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
4787 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
4789 cur = min_t(unsigned long, cur,
4790 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
4792 copy_pages(extent_buffer_page(dst, dst_i),
4793 extent_buffer_page(dst, src_i),
4794 dst_off_in_page, src_off_in_page, cur);
4802 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
4803 unsigned long src_offset, unsigned long len)
4806 size_t dst_off_in_page;
4807 size_t src_off_in_page;
4808 unsigned long dst_end = dst_offset + len - 1;
4809 unsigned long src_end = src_offset + len - 1;
4810 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4811 unsigned long dst_i;
4812 unsigned long src_i;
4814 if (src_offset + len > dst->len) {
4815 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
4816 "len %lu len %lu\n", src_offset, len, dst->len);
4819 if (dst_offset + len > dst->len) {
4820 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
4821 "len %lu len %lu\n", dst_offset, len, dst->len);
4824 if (dst_offset < src_offset) {
4825 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
4829 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
4830 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
4832 dst_off_in_page = (start_offset + dst_end) &
4833 ((unsigned long)PAGE_CACHE_SIZE - 1);
4834 src_off_in_page = (start_offset + src_end) &
4835 ((unsigned long)PAGE_CACHE_SIZE - 1);
4837 cur = min_t(unsigned long, len, src_off_in_page + 1);
4838 cur = min(cur, dst_off_in_page + 1);
4839 move_pages(extent_buffer_page(dst, dst_i),
4840 extent_buffer_page(dst, src_i),
4841 dst_off_in_page - cur + 1,
4842 src_off_in_page - cur + 1, cur);
4850 int try_release_extent_buffer(struct page *page, gfp_t mask)
4852 struct extent_buffer *eb;
4855 * We need to make sure noboody is attaching this page to an eb right
4858 spin_lock(&page->mapping->private_lock);
4859 if (!PagePrivate(page)) {
4860 spin_unlock(&page->mapping->private_lock);
4864 eb = (struct extent_buffer *)page->private;
4868 * This is a little awful but should be ok, we need to make sure that
4869 * the eb doesn't disappear out from under us while we're looking at
4872 spin_lock(&eb->refs_lock);
4873 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
4874 spin_unlock(&eb->refs_lock);
4875 spin_unlock(&page->mapping->private_lock);
4878 spin_unlock(&page->mapping->private_lock);
4880 if ((mask & GFP_NOFS) == GFP_NOFS)
4884 * If tree ref isn't set then we know the ref on this eb is a real ref,
4885 * so just return, this page will likely be freed soon anyway.
4887 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
4888 spin_unlock(&eb->refs_lock);
4891 release_extent_buffer(eb, mask);