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"
23 #include "rcu-string.h"
25 static struct kmem_cache *extent_state_cache;
26 static struct kmem_cache *extent_buffer_cache;
28 static LIST_HEAD(buffers);
29 static LIST_HEAD(states);
33 static DEFINE_SPINLOCK(leak_lock);
36 #define BUFFER_LRU_MAX 64
41 struct rb_node rb_node;
44 struct extent_page_data {
46 struct extent_io_tree *tree;
47 get_extent_t *get_extent;
49 /* tells writepage not to lock the state bits for this range
50 * it still does the unlocking
52 unsigned int extent_locked:1;
54 /* tells the submit_bio code to use a WRITE_SYNC */
55 unsigned int sync_io:1;
58 static noinline void flush_write_bio(void *data);
59 static inline struct btrfs_fs_info *
60 tree_fs_info(struct extent_io_tree *tree)
62 return btrfs_sb(tree->mapping->host->i_sb);
65 int __init extent_io_init(void)
67 extent_state_cache = kmem_cache_create("extent_state",
68 sizeof(struct extent_state), 0,
69 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
70 if (!extent_state_cache)
73 extent_buffer_cache = kmem_cache_create("extent_buffers",
74 sizeof(struct extent_buffer), 0,
75 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
76 if (!extent_buffer_cache)
77 goto free_state_cache;
81 kmem_cache_destroy(extent_state_cache);
85 void extent_io_exit(void)
87 struct extent_state *state;
88 struct extent_buffer *eb;
90 while (!list_empty(&states)) {
91 state = list_entry(states.next, struct extent_state, leak_list);
92 printk(KERN_ERR "btrfs state leak: start %llu end %llu "
93 "state %lu in tree %p refs %d\n",
94 (unsigned long long)state->start,
95 (unsigned long long)state->end,
96 state->state, state->tree, atomic_read(&state->refs));
97 list_del(&state->leak_list);
98 kmem_cache_free(extent_state_cache, state);
102 while (!list_empty(&buffers)) {
103 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
104 printk(KERN_ERR "btrfs buffer leak start %llu len %lu "
105 "refs %d\n", (unsigned long long)eb->start,
106 eb->len, atomic_read(&eb->refs));
107 list_del(&eb->leak_list);
108 kmem_cache_free(extent_buffer_cache, eb);
110 if (extent_state_cache)
111 kmem_cache_destroy(extent_state_cache);
112 if (extent_buffer_cache)
113 kmem_cache_destroy(extent_buffer_cache);
116 void extent_io_tree_init(struct extent_io_tree *tree,
117 struct address_space *mapping)
119 tree->state = RB_ROOT;
120 INIT_RADIX_TREE(&tree->buffer, GFP_ATOMIC);
122 tree->dirty_bytes = 0;
123 spin_lock_init(&tree->lock);
124 spin_lock_init(&tree->buffer_lock);
125 tree->mapping = mapping;
128 static struct extent_state *alloc_extent_state(gfp_t mask)
130 struct extent_state *state;
135 state = kmem_cache_alloc(extent_state_cache, mask);
142 spin_lock_irqsave(&leak_lock, flags);
143 list_add(&state->leak_list, &states);
144 spin_unlock_irqrestore(&leak_lock, flags);
146 atomic_set(&state->refs, 1);
147 init_waitqueue_head(&state->wq);
148 trace_alloc_extent_state(state, mask, _RET_IP_);
152 void free_extent_state(struct extent_state *state)
156 if (atomic_dec_and_test(&state->refs)) {
160 WARN_ON(state->tree);
162 spin_lock_irqsave(&leak_lock, flags);
163 list_del(&state->leak_list);
164 spin_unlock_irqrestore(&leak_lock, flags);
166 trace_free_extent_state(state, _RET_IP_);
167 kmem_cache_free(extent_state_cache, state);
171 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
172 struct rb_node *node)
174 struct rb_node **p = &root->rb_node;
175 struct rb_node *parent = NULL;
176 struct tree_entry *entry;
180 entry = rb_entry(parent, struct tree_entry, rb_node);
182 if (offset < entry->start)
184 else if (offset > entry->end)
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 state = clear_state_bit(tree, state, &bits, wake);
579 * | ---- desired range ---- |
581 * We need to split the extent, and clear the bit
584 if (state->start <= end && state->end > end) {
585 prealloc = alloc_extent_state_atomic(prealloc);
587 err = split_state(tree, state, prealloc, end + 1);
589 extent_io_tree_panic(tree, err);
594 clear_state_bit(tree, prealloc, &bits, wake);
600 state = clear_state_bit(tree, state, &bits, wake);
602 if (last_end == (u64)-1)
604 start = last_end + 1;
605 if (start <= end && state && !need_resched())
610 spin_unlock(&tree->lock);
612 free_extent_state(prealloc);
619 spin_unlock(&tree->lock);
620 if (mask & __GFP_WAIT)
625 static void wait_on_state(struct extent_io_tree *tree,
626 struct extent_state *state)
627 __releases(tree->lock)
628 __acquires(tree->lock)
631 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
632 spin_unlock(&tree->lock);
634 spin_lock(&tree->lock);
635 finish_wait(&state->wq, &wait);
639 * waits for one or more bits to clear on a range in the state tree.
640 * The range [start, end] is inclusive.
641 * The tree lock is taken by this function
643 void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits)
645 struct extent_state *state;
646 struct rb_node *node;
648 spin_lock(&tree->lock);
652 * this search will find all the extents that end after
655 node = tree_search(tree, start);
659 state = rb_entry(node, struct extent_state, rb_node);
661 if (state->start > end)
664 if (state->state & bits) {
665 start = state->start;
666 atomic_inc(&state->refs);
667 wait_on_state(tree, state);
668 free_extent_state(state);
671 start = state->end + 1;
676 cond_resched_lock(&tree->lock);
679 spin_unlock(&tree->lock);
682 static void set_state_bits(struct extent_io_tree *tree,
683 struct extent_state *state,
686 int bits_to_set = *bits & ~EXTENT_CTLBITS;
688 set_state_cb(tree, state, bits);
689 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
690 u64 range = state->end - state->start + 1;
691 tree->dirty_bytes += range;
693 state->state |= bits_to_set;
696 static void cache_state(struct extent_state *state,
697 struct extent_state **cached_ptr)
699 if (cached_ptr && !(*cached_ptr)) {
700 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) {
702 atomic_inc(&state->refs);
707 static void uncache_state(struct extent_state **cached_ptr)
709 if (cached_ptr && (*cached_ptr)) {
710 struct extent_state *state = *cached_ptr;
712 free_extent_state(state);
717 * set some bits on a range in the tree. This may require allocations or
718 * sleeping, so the gfp mask is used to indicate what is allowed.
720 * If any of the exclusive bits are set, this will fail with -EEXIST if some
721 * part of the range already has the desired bits set. The start of the
722 * existing range is returned in failed_start in this case.
724 * [start, end] is inclusive This takes the tree lock.
727 static int __must_check
728 __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
729 int bits, int exclusive_bits, u64 *failed_start,
730 struct extent_state **cached_state, gfp_t mask)
732 struct extent_state *state;
733 struct extent_state *prealloc = NULL;
734 struct rb_node *node;
739 bits |= EXTENT_FIRST_DELALLOC;
741 if (!prealloc && (mask & __GFP_WAIT)) {
742 prealloc = alloc_extent_state(mask);
746 spin_lock(&tree->lock);
747 if (cached_state && *cached_state) {
748 state = *cached_state;
749 if (state->start <= start && state->end > start &&
751 node = &state->rb_node;
756 * this search will find all the extents that end after
759 node = tree_search(tree, start);
761 prealloc = alloc_extent_state_atomic(prealloc);
763 err = insert_state(tree, prealloc, start, end, &bits);
765 extent_io_tree_panic(tree, err);
770 state = rb_entry(node, struct extent_state, rb_node);
772 last_start = state->start;
773 last_end = state->end;
776 * | ---- desired range ---- |
779 * Just lock what we found and keep going
781 if (state->start == start && state->end <= end) {
782 if (state->state & exclusive_bits) {
783 *failed_start = state->start;
788 set_state_bits(tree, state, &bits);
789 cache_state(state, cached_state);
790 merge_state(tree, state);
791 if (last_end == (u64)-1)
793 start = last_end + 1;
794 state = next_state(state);
795 if (start < end && state && state->start == start &&
802 * | ---- desired range ---- |
805 * | ------------- state -------------- |
807 * We need to split the extent we found, and may flip bits on
810 * If the extent we found extends past our
811 * range, we just split and search again. It'll get split
812 * again the next time though.
814 * If the extent we found is inside our range, we set the
817 if (state->start < start) {
818 if (state->state & exclusive_bits) {
819 *failed_start = start;
824 prealloc = alloc_extent_state_atomic(prealloc);
826 err = split_state(tree, state, prealloc, start);
828 extent_io_tree_panic(tree, err);
833 if (state->end <= end) {
834 set_state_bits(tree, state, &bits);
835 cache_state(state, cached_state);
836 merge_state(tree, state);
837 if (last_end == (u64)-1)
839 start = last_end + 1;
840 state = next_state(state);
841 if (start < end && state && state->start == start &&
848 * | ---- desired range ---- |
849 * | state | or | state |
851 * There's a hole, we need to insert something in it and
852 * ignore the extent we found.
854 if (state->start > start) {
856 if (end < last_start)
859 this_end = last_start - 1;
861 prealloc = alloc_extent_state_atomic(prealloc);
865 * Avoid to free 'prealloc' if it can be merged with
868 err = insert_state(tree, prealloc, start, this_end,
871 extent_io_tree_panic(tree, err);
873 cache_state(prealloc, cached_state);
875 start = this_end + 1;
879 * | ---- desired range ---- |
881 * We need to split the extent, and set the bit
884 if (state->start <= end && state->end > end) {
885 if (state->state & exclusive_bits) {
886 *failed_start = start;
891 prealloc = alloc_extent_state_atomic(prealloc);
893 err = split_state(tree, state, prealloc, end + 1);
895 extent_io_tree_panic(tree, err);
897 set_state_bits(tree, prealloc, &bits);
898 cache_state(prealloc, cached_state);
899 merge_state(tree, prealloc);
907 spin_unlock(&tree->lock);
909 free_extent_state(prealloc);
916 spin_unlock(&tree->lock);
917 if (mask & __GFP_WAIT)
922 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits,
923 u64 *failed_start, struct extent_state **cached_state,
926 return __set_extent_bit(tree, start, end, bits, 0, failed_start,
932 * convert_extent_bit - convert all bits in a given range from one bit to
934 * @tree: the io tree to search
935 * @start: the start offset in bytes
936 * @end: the end offset in bytes (inclusive)
937 * @bits: the bits to set in this range
938 * @clear_bits: the bits to clear in this range
939 * @mask: the allocation mask
941 * This will go through and set bits for the given range. If any states exist
942 * already in this range they are set with the given bit and cleared of the
943 * clear_bits. This is only meant to be used by things that are mergeable, ie
944 * converting from say DELALLOC to DIRTY. This is not meant to be used with
945 * boundary bits like LOCK.
947 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
948 int bits, int clear_bits, gfp_t mask)
950 struct extent_state *state;
951 struct extent_state *prealloc = NULL;
952 struct rb_node *node;
958 if (!prealloc && (mask & __GFP_WAIT)) {
959 prealloc = alloc_extent_state(mask);
964 spin_lock(&tree->lock);
966 * this search will find all the extents that end after
969 node = tree_search(tree, start);
971 prealloc = alloc_extent_state_atomic(prealloc);
976 err = insert_state(tree, prealloc, start, end, &bits);
979 extent_io_tree_panic(tree, err);
982 state = rb_entry(node, struct extent_state, rb_node);
984 last_start = state->start;
985 last_end = state->end;
988 * | ---- desired range ---- |
991 * Just lock what we found and keep going
993 if (state->start == start && state->end <= end) {
994 set_state_bits(tree, state, &bits);
995 state = clear_state_bit(tree, state, &clear_bits, 0);
996 if (last_end == (u64)-1)
998 start = last_end + 1;
999 if (start < end && state && state->start == start &&
1006 * | ---- desired range ---- |
1009 * | ------------- state -------------- |
1011 * We need to split the extent we found, and may flip bits on
1014 * If the extent we found extends past our
1015 * range, we just split and search again. It'll get split
1016 * again the next time though.
1018 * If the extent we found is inside our range, we set the
1019 * desired bit on it.
1021 if (state->start < start) {
1022 prealloc = alloc_extent_state_atomic(prealloc);
1027 err = split_state(tree, state, prealloc, start);
1029 extent_io_tree_panic(tree, err);
1033 if (state->end <= end) {
1034 set_state_bits(tree, state, &bits);
1035 state = clear_state_bit(tree, state, &clear_bits, 0);
1036 if (last_end == (u64)-1)
1038 start = last_end + 1;
1039 if (start < end && state && state->start == start &&
1046 * | ---- desired range ---- |
1047 * | state | or | state |
1049 * There's a hole, we need to insert something in it and
1050 * ignore the extent we found.
1052 if (state->start > start) {
1054 if (end < last_start)
1057 this_end = last_start - 1;
1059 prealloc = alloc_extent_state_atomic(prealloc);
1066 * Avoid to free 'prealloc' if it can be merged with
1069 err = insert_state(tree, prealloc, start, this_end,
1072 extent_io_tree_panic(tree, err);
1074 start = this_end + 1;
1078 * | ---- desired range ---- |
1080 * We need to split the extent, and set the bit
1083 if (state->start <= end && state->end > end) {
1084 prealloc = alloc_extent_state_atomic(prealloc);
1090 err = split_state(tree, state, prealloc, end + 1);
1092 extent_io_tree_panic(tree, err);
1094 set_state_bits(tree, prealloc, &bits);
1095 clear_state_bit(tree, prealloc, &clear_bits, 0);
1103 spin_unlock(&tree->lock);
1105 free_extent_state(prealloc);
1112 spin_unlock(&tree->lock);
1113 if (mask & __GFP_WAIT)
1118 /* wrappers around set/clear extent bit */
1119 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1122 return set_extent_bit(tree, start, end, EXTENT_DIRTY, NULL,
1126 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1127 int bits, gfp_t mask)
1129 return set_extent_bit(tree, start, end, bits, NULL,
1133 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1134 int bits, gfp_t mask)
1136 return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask);
1139 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
1140 struct extent_state **cached_state, gfp_t mask)
1142 return set_extent_bit(tree, start, end,
1143 EXTENT_DELALLOC | EXTENT_UPTODATE,
1144 NULL, cached_state, mask);
1147 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1150 return clear_extent_bit(tree, start, end,
1151 EXTENT_DIRTY | EXTENT_DELALLOC |
1152 EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask);
1155 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
1158 return set_extent_bit(tree, start, end, EXTENT_NEW, NULL,
1162 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
1163 struct extent_state **cached_state, gfp_t mask)
1165 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0,
1166 cached_state, mask);
1169 int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
1170 struct extent_state **cached_state, gfp_t mask)
1172 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0,
1173 cached_state, mask);
1177 * either insert or lock state struct between start and end use mask to tell
1178 * us if waiting is desired.
1180 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1181 int bits, struct extent_state **cached_state)
1186 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED | bits,
1187 EXTENT_LOCKED, &failed_start,
1188 cached_state, GFP_NOFS);
1189 if (err == -EEXIST) {
1190 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1191 start = failed_start;
1194 WARN_ON(start > end);
1199 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1201 return lock_extent_bits(tree, start, end, 0, NULL);
1204 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1209 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1210 &failed_start, NULL, GFP_NOFS);
1211 if (err == -EEXIST) {
1212 if (failed_start > start)
1213 clear_extent_bit(tree, start, failed_start - 1,
1214 EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS);
1220 int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end,
1221 struct extent_state **cached, gfp_t mask)
1223 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached,
1227 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1229 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL,
1234 * helper function to set both pages and extents in the tree writeback
1236 static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
1238 unsigned long index = start >> PAGE_CACHE_SHIFT;
1239 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1242 while (index <= end_index) {
1243 page = find_get_page(tree->mapping, index);
1244 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1245 set_page_writeback(page);
1246 page_cache_release(page);
1252 /* find the first state struct with 'bits' set after 'start', and
1253 * return it. tree->lock must be held. NULL will returned if
1254 * nothing was found after 'start'
1256 struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree,
1257 u64 start, int bits)
1259 struct rb_node *node;
1260 struct extent_state *state;
1263 * this search will find all the extents that end after
1266 node = tree_search(tree, start);
1271 state = rb_entry(node, struct extent_state, rb_node);
1272 if (state->end >= start && (state->state & bits))
1275 node = rb_next(node);
1284 * find the first offset in the io tree with 'bits' set. zero is
1285 * returned if we find something, and *start_ret and *end_ret are
1286 * set to reflect the state struct that was found.
1288 * If nothing was found, 1 is returned. If found something, return 0.
1290 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1291 u64 *start_ret, u64 *end_ret, int bits)
1293 struct extent_state *state;
1296 spin_lock(&tree->lock);
1297 state = find_first_extent_bit_state(tree, start, bits);
1299 *start_ret = state->start;
1300 *end_ret = state->end;
1303 spin_unlock(&tree->lock);
1308 * find a contiguous range of bytes in the file marked as delalloc, not
1309 * more than 'max_bytes'. start and end are used to return the range,
1311 * 1 is returned if we find something, 0 if nothing was in the tree
1313 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1314 u64 *start, u64 *end, u64 max_bytes,
1315 struct extent_state **cached_state)
1317 struct rb_node *node;
1318 struct extent_state *state;
1319 u64 cur_start = *start;
1321 u64 total_bytes = 0;
1323 spin_lock(&tree->lock);
1326 * this search will find all the extents that end after
1329 node = tree_search(tree, cur_start);
1337 state = rb_entry(node, struct extent_state, rb_node);
1338 if (found && (state->start != cur_start ||
1339 (state->state & EXTENT_BOUNDARY))) {
1342 if (!(state->state & EXTENT_DELALLOC)) {
1348 *start = state->start;
1349 *cached_state = state;
1350 atomic_inc(&state->refs);
1354 cur_start = state->end + 1;
1355 node = rb_next(node);
1358 total_bytes += state->end - state->start + 1;
1359 if (total_bytes >= max_bytes)
1363 spin_unlock(&tree->lock);
1367 static noinline void __unlock_for_delalloc(struct inode *inode,
1368 struct page *locked_page,
1372 struct page *pages[16];
1373 unsigned long index = start >> PAGE_CACHE_SHIFT;
1374 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1375 unsigned long nr_pages = end_index - index + 1;
1378 if (index == locked_page->index && end_index == index)
1381 while (nr_pages > 0) {
1382 ret = find_get_pages_contig(inode->i_mapping, index,
1383 min_t(unsigned long, nr_pages,
1384 ARRAY_SIZE(pages)), pages);
1385 for (i = 0; i < ret; i++) {
1386 if (pages[i] != locked_page)
1387 unlock_page(pages[i]);
1388 page_cache_release(pages[i]);
1396 static noinline int lock_delalloc_pages(struct inode *inode,
1397 struct page *locked_page,
1401 unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT;
1402 unsigned long start_index = index;
1403 unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT;
1404 unsigned long pages_locked = 0;
1405 struct page *pages[16];
1406 unsigned long nrpages;
1410 /* the caller is responsible for locking the start index */
1411 if (index == locked_page->index && index == end_index)
1414 /* skip the page at the start index */
1415 nrpages = end_index - index + 1;
1416 while (nrpages > 0) {
1417 ret = find_get_pages_contig(inode->i_mapping, index,
1418 min_t(unsigned long,
1419 nrpages, ARRAY_SIZE(pages)), pages);
1424 /* now we have an array of pages, lock them all */
1425 for (i = 0; i < ret; i++) {
1427 * the caller is taking responsibility for
1430 if (pages[i] != locked_page) {
1431 lock_page(pages[i]);
1432 if (!PageDirty(pages[i]) ||
1433 pages[i]->mapping != inode->i_mapping) {
1435 unlock_page(pages[i]);
1436 page_cache_release(pages[i]);
1440 page_cache_release(pages[i]);
1449 if (ret && pages_locked) {
1450 __unlock_for_delalloc(inode, locked_page,
1452 ((u64)(start_index + pages_locked - 1)) <<
1459 * find a contiguous range of bytes in the file marked as delalloc, not
1460 * more than 'max_bytes'. start and end are used to return the range,
1462 * 1 is returned if we find something, 0 if nothing was in the tree
1464 static noinline u64 find_lock_delalloc_range(struct inode *inode,
1465 struct extent_io_tree *tree,
1466 struct page *locked_page,
1467 u64 *start, u64 *end,
1473 struct extent_state *cached_state = NULL;
1478 /* step one, find a bunch of delalloc bytes starting at start */
1479 delalloc_start = *start;
1481 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1482 max_bytes, &cached_state);
1483 if (!found || delalloc_end <= *start) {
1484 *start = delalloc_start;
1485 *end = delalloc_end;
1486 free_extent_state(cached_state);
1491 * start comes from the offset of locked_page. We have to lock
1492 * pages in order, so we can't process delalloc bytes before
1495 if (delalloc_start < *start)
1496 delalloc_start = *start;
1499 * make sure to limit the number of pages we try to lock down
1502 if (delalloc_end + 1 - delalloc_start > max_bytes && loops)
1503 delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1;
1505 /* step two, lock all the pages after the page that has start */
1506 ret = lock_delalloc_pages(inode, locked_page,
1507 delalloc_start, delalloc_end);
1508 if (ret == -EAGAIN) {
1509 /* some of the pages are gone, lets avoid looping by
1510 * shortening the size of the delalloc range we're searching
1512 free_extent_state(cached_state);
1514 unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1);
1515 max_bytes = PAGE_CACHE_SIZE - offset;
1523 BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */
1525 /* step three, lock the state bits for the whole range */
1526 lock_extent_bits(tree, delalloc_start, delalloc_end, 0, &cached_state);
1528 /* then test to make sure it is all still delalloc */
1529 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1530 EXTENT_DELALLOC, 1, cached_state);
1532 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1533 &cached_state, GFP_NOFS);
1534 __unlock_for_delalloc(inode, locked_page,
1535 delalloc_start, delalloc_end);
1539 free_extent_state(cached_state);
1540 *start = delalloc_start;
1541 *end = delalloc_end;
1546 int extent_clear_unlock_delalloc(struct inode *inode,
1547 struct extent_io_tree *tree,
1548 u64 start, u64 end, struct page *locked_page,
1552 struct page *pages[16];
1553 unsigned long index = start >> PAGE_CACHE_SHIFT;
1554 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1555 unsigned long nr_pages = end_index - index + 1;
1559 if (op & EXTENT_CLEAR_UNLOCK)
1560 clear_bits |= EXTENT_LOCKED;
1561 if (op & EXTENT_CLEAR_DIRTY)
1562 clear_bits |= EXTENT_DIRTY;
1564 if (op & EXTENT_CLEAR_DELALLOC)
1565 clear_bits |= EXTENT_DELALLOC;
1567 clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS);
1568 if (!(op & (EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
1569 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK |
1570 EXTENT_SET_PRIVATE2)))
1573 while (nr_pages > 0) {
1574 ret = find_get_pages_contig(inode->i_mapping, index,
1575 min_t(unsigned long,
1576 nr_pages, ARRAY_SIZE(pages)), pages);
1577 for (i = 0; i < ret; i++) {
1579 if (op & EXTENT_SET_PRIVATE2)
1580 SetPagePrivate2(pages[i]);
1582 if (pages[i] == locked_page) {
1583 page_cache_release(pages[i]);
1586 if (op & EXTENT_CLEAR_DIRTY)
1587 clear_page_dirty_for_io(pages[i]);
1588 if (op & EXTENT_SET_WRITEBACK)
1589 set_page_writeback(pages[i]);
1590 if (op & EXTENT_END_WRITEBACK)
1591 end_page_writeback(pages[i]);
1592 if (op & EXTENT_CLEAR_UNLOCK_PAGE)
1593 unlock_page(pages[i]);
1594 page_cache_release(pages[i]);
1604 * count the number of bytes in the tree that have a given bit(s)
1605 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1606 * cached. The total number found is returned.
1608 u64 count_range_bits(struct extent_io_tree *tree,
1609 u64 *start, u64 search_end, u64 max_bytes,
1610 unsigned long bits, int contig)
1612 struct rb_node *node;
1613 struct extent_state *state;
1614 u64 cur_start = *start;
1615 u64 total_bytes = 0;
1619 if (search_end <= cur_start) {
1624 spin_lock(&tree->lock);
1625 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1626 total_bytes = tree->dirty_bytes;
1630 * this search will find all the extents that end after
1633 node = tree_search(tree, cur_start);
1638 state = rb_entry(node, struct extent_state, rb_node);
1639 if (state->start > search_end)
1641 if (contig && found && state->start > last + 1)
1643 if (state->end >= cur_start && (state->state & bits) == bits) {
1644 total_bytes += min(search_end, state->end) + 1 -
1645 max(cur_start, state->start);
1646 if (total_bytes >= max_bytes)
1649 *start = max(cur_start, state->start);
1653 } else if (contig && found) {
1656 node = rb_next(node);
1661 spin_unlock(&tree->lock);
1666 * set the private field for a given byte offset in the tree. If there isn't
1667 * an extent_state there already, this does nothing.
1669 int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1671 struct rb_node *node;
1672 struct extent_state *state;
1675 spin_lock(&tree->lock);
1677 * this search will find all the extents that end after
1680 node = tree_search(tree, start);
1685 state = rb_entry(node, struct extent_state, rb_node);
1686 if (state->start != start) {
1690 state->private = private;
1692 spin_unlock(&tree->lock);
1696 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1698 struct rb_node *node;
1699 struct extent_state *state;
1702 spin_lock(&tree->lock);
1704 * this search will find all the extents that end after
1707 node = tree_search(tree, start);
1712 state = rb_entry(node, struct extent_state, rb_node);
1713 if (state->start != start) {
1717 *private = state->private;
1719 spin_unlock(&tree->lock);
1724 * searches a range in the state tree for a given mask.
1725 * If 'filled' == 1, this returns 1 only if every extent in the tree
1726 * has the bits set. Otherwise, 1 is returned if any bit in the
1727 * range is found set.
1729 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1730 int bits, int filled, struct extent_state *cached)
1732 struct extent_state *state = NULL;
1733 struct rb_node *node;
1736 spin_lock(&tree->lock);
1737 if (cached && cached->tree && cached->start <= start &&
1738 cached->end > start)
1739 node = &cached->rb_node;
1741 node = tree_search(tree, start);
1742 while (node && start <= end) {
1743 state = rb_entry(node, struct extent_state, rb_node);
1745 if (filled && state->start > start) {
1750 if (state->start > end)
1753 if (state->state & bits) {
1757 } else if (filled) {
1762 if (state->end == (u64)-1)
1765 start = state->end + 1;
1768 node = rb_next(node);
1775 spin_unlock(&tree->lock);
1780 * helper function to set a given page up to date if all the
1781 * extents in the tree for that page are up to date
1783 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
1785 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1786 u64 end = start + PAGE_CACHE_SIZE - 1;
1787 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1788 SetPageUptodate(page);
1792 * helper function to unlock a page if all the extents in the tree
1793 * for that page are unlocked
1795 static void check_page_locked(struct extent_io_tree *tree, struct page *page)
1797 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1798 u64 end = start + PAGE_CACHE_SIZE - 1;
1799 if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL))
1804 * helper function to end page writeback if all the extents
1805 * in the tree for that page are done with writeback
1807 static void check_page_writeback(struct extent_io_tree *tree,
1810 end_page_writeback(page);
1814 * When IO fails, either with EIO or csum verification fails, we
1815 * try other mirrors that might have a good copy of the data. This
1816 * io_failure_record is used to record state as we go through all the
1817 * mirrors. If another mirror has good data, the page is set up to date
1818 * and things continue. If a good mirror can't be found, the original
1819 * bio end_io callback is called to indicate things have failed.
1821 struct io_failure_record {
1826 unsigned long bio_flags;
1832 static int free_io_failure(struct inode *inode, struct io_failure_record *rec,
1837 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1839 set_state_private(failure_tree, rec->start, 0);
1840 ret = clear_extent_bits(failure_tree, rec->start,
1841 rec->start + rec->len - 1,
1842 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1847 ret = clear_extent_bits(&BTRFS_I(inode)->io_tree, rec->start,
1848 rec->start + rec->len - 1,
1849 EXTENT_DAMAGED, GFP_NOFS);
1858 static void repair_io_failure_callback(struct bio *bio, int err)
1860 complete(bio->bi_private);
1864 * this bypasses the standard btrfs submit functions deliberately, as
1865 * the standard behavior is to write all copies in a raid setup. here we only
1866 * want to write the one bad copy. so we do the mapping for ourselves and issue
1867 * submit_bio directly.
1868 * to avoid any synchonization issues, wait for the data after writing, which
1869 * actually prevents the read that triggered the error from finishing.
1870 * currently, there can be no more than two copies of every data bit. thus,
1871 * exactly one rewrite is required.
1873 int repair_io_failure(struct btrfs_mapping_tree *map_tree, u64 start,
1874 u64 length, u64 logical, struct page *page,
1878 struct btrfs_device *dev;
1879 DECLARE_COMPLETION_ONSTACK(compl);
1882 struct btrfs_bio *bbio = NULL;
1885 BUG_ON(!mirror_num);
1887 bio = bio_alloc(GFP_NOFS, 1);
1890 bio->bi_private = &compl;
1891 bio->bi_end_io = repair_io_failure_callback;
1893 map_length = length;
1895 ret = btrfs_map_block(map_tree, WRITE, logical,
1896 &map_length, &bbio, mirror_num);
1901 BUG_ON(mirror_num != bbio->mirror_num);
1902 sector = bbio->stripes[mirror_num-1].physical >> 9;
1903 bio->bi_sector = sector;
1904 dev = bbio->stripes[mirror_num-1].dev;
1906 if (!dev || !dev->bdev || !dev->writeable) {
1910 bio->bi_bdev = dev->bdev;
1911 bio_add_page(bio, page, length, start-page_offset(page));
1912 btrfsic_submit_bio(WRITE_SYNC, bio);
1913 wait_for_completion(&compl);
1915 if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) {
1916 /* try to remap that extent elsewhere? */
1918 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
1922 printk_ratelimited_in_rcu(KERN_INFO "btrfs read error corrected: ino %lu off %llu "
1923 "(dev %s sector %llu)\n", page->mapping->host->i_ino,
1924 start, rcu_str_deref(dev->name), sector);
1930 int repair_eb_io_failure(struct btrfs_root *root, struct extent_buffer *eb,
1933 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1934 u64 start = eb->start;
1935 unsigned long i, num_pages = num_extent_pages(eb->start, eb->len);
1938 for (i = 0; i < num_pages; i++) {
1939 struct page *p = extent_buffer_page(eb, i);
1940 ret = repair_io_failure(map_tree, start, PAGE_CACHE_SIZE,
1941 start, p, mirror_num);
1944 start += PAGE_CACHE_SIZE;
1951 * each time an IO finishes, we do a fast check in the IO failure tree
1952 * to see if we need to process or clean up an io_failure_record
1954 static int clean_io_failure(u64 start, struct page *page)
1957 u64 private_failure;
1958 struct io_failure_record *failrec;
1959 struct btrfs_mapping_tree *map_tree;
1960 struct extent_state *state;
1964 struct inode *inode = page->mapping->host;
1967 ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1968 (u64)-1, 1, EXTENT_DIRTY, 0);
1972 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, start,
1977 failrec = (struct io_failure_record *)(unsigned long) private_failure;
1978 BUG_ON(!failrec->this_mirror);
1980 if (failrec->in_validation) {
1981 /* there was no real error, just free the record */
1982 pr_debug("clean_io_failure: freeing dummy error at %llu\n",
1988 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1989 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1992 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1994 if (state && state->start == failrec->start) {
1995 map_tree = &BTRFS_I(inode)->root->fs_info->mapping_tree;
1996 num_copies = btrfs_num_copies(map_tree, failrec->logical,
1998 if (num_copies > 1) {
1999 ret = repair_io_failure(map_tree, start, failrec->len,
2000 failrec->logical, page,
2001 failrec->failed_mirror);
2008 ret = free_io_failure(inode, failrec, did_repair);
2014 * this is a generic handler for readpage errors (default
2015 * readpage_io_failed_hook). if other copies exist, read those and write back
2016 * good data to the failed position. does not investigate in remapping the
2017 * failed extent elsewhere, hoping the device will be smart enough to do this as
2021 static int bio_readpage_error(struct bio *failed_bio, struct page *page,
2022 u64 start, u64 end, int failed_mirror,
2023 struct extent_state *state)
2025 struct io_failure_record *failrec = NULL;
2027 struct extent_map *em;
2028 struct inode *inode = page->mapping->host;
2029 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2030 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2031 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2038 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
2040 ret = get_state_private(failure_tree, start, &private);
2042 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2045 failrec->start = start;
2046 failrec->len = end - start + 1;
2047 failrec->this_mirror = 0;
2048 failrec->bio_flags = 0;
2049 failrec->in_validation = 0;
2051 read_lock(&em_tree->lock);
2052 em = lookup_extent_mapping(em_tree, start, failrec->len);
2054 read_unlock(&em_tree->lock);
2059 if (em->start > start || em->start + em->len < start) {
2060 free_extent_map(em);
2063 read_unlock(&em_tree->lock);
2065 if (!em || IS_ERR(em)) {
2069 logical = start - em->start;
2070 logical = em->block_start + logical;
2071 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2072 logical = em->block_start;
2073 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2074 extent_set_compress_type(&failrec->bio_flags,
2077 pr_debug("bio_readpage_error: (new) logical=%llu, start=%llu, "
2078 "len=%llu\n", logical, start, failrec->len);
2079 failrec->logical = logical;
2080 free_extent_map(em);
2082 /* set the bits in the private failure tree */
2083 ret = set_extent_bits(failure_tree, start, end,
2084 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
2086 ret = set_state_private(failure_tree, start,
2087 (u64)(unsigned long)failrec);
2088 /* set the bits in the inode's tree */
2090 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED,
2097 failrec = (struct io_failure_record *)(unsigned long)private;
2098 pr_debug("bio_readpage_error: (found) logical=%llu, "
2099 "start=%llu, len=%llu, validation=%d\n",
2100 failrec->logical, failrec->start, failrec->len,
2101 failrec->in_validation);
2103 * when data can be on disk more than twice, add to failrec here
2104 * (e.g. with a list for failed_mirror) to make
2105 * clean_io_failure() clean all those errors at once.
2108 num_copies = btrfs_num_copies(
2109 &BTRFS_I(inode)->root->fs_info->mapping_tree,
2110 failrec->logical, failrec->len);
2111 if (num_copies == 1) {
2113 * we only have a single copy of the data, so don't bother with
2114 * all the retry and error correction code that follows. no
2115 * matter what the error is, it is very likely to persist.
2117 pr_debug("bio_readpage_error: cannot repair, num_copies == 1. "
2118 "state=%p, num_copies=%d, next_mirror %d, "
2119 "failed_mirror %d\n", state, num_copies,
2120 failrec->this_mirror, failed_mirror);
2121 free_io_failure(inode, failrec, 0);
2126 spin_lock(&tree->lock);
2127 state = find_first_extent_bit_state(tree, failrec->start,
2129 if (state && state->start != failrec->start)
2131 spin_unlock(&tree->lock);
2135 * there are two premises:
2136 * a) deliver good data to the caller
2137 * b) correct the bad sectors on disk
2139 if (failed_bio->bi_vcnt > 1) {
2141 * to fulfill b), we need to know the exact failing sectors, as
2142 * we don't want to rewrite any more than the failed ones. thus,
2143 * we need separate read requests for the failed bio
2145 * if the following BUG_ON triggers, our validation request got
2146 * merged. we need separate requests for our algorithm to work.
2148 BUG_ON(failrec->in_validation);
2149 failrec->in_validation = 1;
2150 failrec->this_mirror = failed_mirror;
2151 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
2154 * we're ready to fulfill a) and b) alongside. get a good copy
2155 * of the failed sector and if we succeed, we have setup
2156 * everything for repair_io_failure to do the rest for us.
2158 if (failrec->in_validation) {
2159 BUG_ON(failrec->this_mirror != failed_mirror);
2160 failrec->in_validation = 0;
2161 failrec->this_mirror = 0;
2163 failrec->failed_mirror = failed_mirror;
2164 failrec->this_mirror++;
2165 if (failrec->this_mirror == failed_mirror)
2166 failrec->this_mirror++;
2167 read_mode = READ_SYNC;
2170 if (!state || failrec->this_mirror > num_copies) {
2171 pr_debug("bio_readpage_error: (fail) state=%p, num_copies=%d, "
2172 "next_mirror %d, failed_mirror %d\n", state,
2173 num_copies, failrec->this_mirror, failed_mirror);
2174 free_io_failure(inode, failrec, 0);
2178 bio = bio_alloc(GFP_NOFS, 1);
2180 free_io_failure(inode, failrec, 0);
2183 bio->bi_private = state;
2184 bio->bi_end_io = failed_bio->bi_end_io;
2185 bio->bi_sector = failrec->logical >> 9;
2186 bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
2189 bio_add_page(bio, page, failrec->len, start - page_offset(page));
2191 pr_debug("bio_readpage_error: submitting new read[%#x] to "
2192 "this_mirror=%d, num_copies=%d, in_validation=%d\n", read_mode,
2193 failrec->this_mirror, num_copies, failrec->in_validation);
2195 ret = tree->ops->submit_bio_hook(inode, read_mode, bio,
2196 failrec->this_mirror,
2197 failrec->bio_flags, 0);
2201 /* lots and lots of room for performance fixes in the end_bio funcs */
2203 int end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2205 int uptodate = (err == 0);
2206 struct extent_io_tree *tree;
2209 tree = &BTRFS_I(page->mapping->host)->io_tree;
2211 if (tree->ops && tree->ops->writepage_end_io_hook) {
2212 ret = tree->ops->writepage_end_io_hook(page, start,
2213 end, NULL, uptodate);
2219 ClearPageUptodate(page);
2226 * after a writepage IO is done, we need to:
2227 * clear the uptodate bits on error
2228 * clear the writeback bits in the extent tree for this IO
2229 * end_page_writeback if the page has no more pending IO
2231 * Scheduling is not allowed, so the extent state tree is expected
2232 * to have one and only one object corresponding to this IO.
2234 static void end_bio_extent_writepage(struct bio *bio, int err)
2236 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2237 struct extent_io_tree *tree;
2243 struct page *page = bvec->bv_page;
2244 tree = &BTRFS_I(page->mapping->host)->io_tree;
2246 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
2248 end = start + bvec->bv_len - 1;
2250 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
2255 if (--bvec >= bio->bi_io_vec)
2256 prefetchw(&bvec->bv_page->flags);
2258 if (end_extent_writepage(page, err, start, end))
2262 end_page_writeback(page);
2264 check_page_writeback(tree, page);
2265 } while (bvec >= bio->bi_io_vec);
2271 * after a readpage IO is done, we need to:
2272 * clear the uptodate bits on error
2273 * set the uptodate bits if things worked
2274 * set the page up to date if all extents in the tree are uptodate
2275 * clear the lock bit in the extent tree
2276 * unlock the page if there are no other extents locked for it
2278 * Scheduling is not allowed, so the extent state tree is expected
2279 * to have one and only one object corresponding to this IO.
2281 static void end_bio_extent_readpage(struct bio *bio, int err)
2283 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
2284 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
2285 struct bio_vec *bvec = bio->bi_io_vec;
2286 struct extent_io_tree *tree;
2297 struct page *page = bvec->bv_page;
2298 struct extent_state *cached = NULL;
2299 struct extent_state *state;
2301 pr_debug("end_bio_extent_readpage: bi_vcnt=%d, idx=%d, err=%d, "
2302 "mirror=%ld\n", bio->bi_vcnt, bio->bi_idx, err,
2303 (long int)bio->bi_bdev);
2304 tree = &BTRFS_I(page->mapping->host)->io_tree;
2306 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
2308 end = start + bvec->bv_len - 1;
2310 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
2315 if (++bvec <= bvec_end)
2316 prefetchw(&bvec->bv_page->flags);
2318 spin_lock(&tree->lock);
2319 state = find_first_extent_bit_state(tree, start, EXTENT_LOCKED);
2320 if (state && state->start == start) {
2322 * take a reference on the state, unlock will drop
2325 cache_state(state, &cached);
2327 spin_unlock(&tree->lock);
2329 mirror = (int)(unsigned long)bio->bi_bdev;
2330 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
2331 ret = tree->ops->readpage_end_io_hook(page, start, end,
2336 clean_io_failure(start, page);
2339 if (!uptodate && tree->ops && tree->ops->readpage_io_failed_hook) {
2340 ret = tree->ops->readpage_io_failed_hook(page, mirror);
2342 test_bit(BIO_UPTODATE, &bio->bi_flags))
2344 } else if (!uptodate) {
2346 * The generic bio_readpage_error handles errors the
2347 * following way: If possible, new read requests are
2348 * created and submitted and will end up in
2349 * end_bio_extent_readpage as well (if we're lucky, not
2350 * in the !uptodate case). In that case it returns 0 and
2351 * we just go on with the next page in our bio. If it
2352 * can't handle the error it will return -EIO and we
2353 * remain responsible for that page.
2355 ret = bio_readpage_error(bio, page, start, end, mirror, NULL);
2358 test_bit(BIO_UPTODATE, &bio->bi_flags);
2361 uncache_state(&cached);
2366 if (uptodate && tree->track_uptodate) {
2367 set_extent_uptodate(tree, start, end, &cached,
2370 unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC);
2374 SetPageUptodate(page);
2376 ClearPageUptodate(page);
2382 check_page_uptodate(tree, page);
2384 ClearPageUptodate(page);
2387 check_page_locked(tree, page);
2389 } while (bvec <= bvec_end);
2395 btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
2400 bio = bio_alloc(gfp_flags, nr_vecs);
2402 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
2403 while (!bio && (nr_vecs /= 2))
2404 bio = bio_alloc(gfp_flags, nr_vecs);
2409 bio->bi_bdev = bdev;
2410 bio->bi_sector = first_sector;
2416 * Since writes are async, they will only return -ENOMEM.
2417 * Reads can return the full range of I/O error conditions.
2419 static int __must_check submit_one_bio(int rw, struct bio *bio,
2420 int mirror_num, unsigned long bio_flags)
2423 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2424 struct page *page = bvec->bv_page;
2425 struct extent_io_tree *tree = bio->bi_private;
2428 start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
2430 bio->bi_private = NULL;
2434 if (tree->ops && tree->ops->submit_bio_hook)
2435 ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
2436 mirror_num, bio_flags, start);
2438 btrfsic_submit_bio(rw, bio);
2440 if (bio_flagged(bio, BIO_EOPNOTSUPP))
2446 static int merge_bio(struct extent_io_tree *tree, struct page *page,
2447 unsigned long offset, size_t size, struct bio *bio,
2448 unsigned long bio_flags)
2451 if (tree->ops && tree->ops->merge_bio_hook)
2452 ret = tree->ops->merge_bio_hook(page, offset, size, bio,
2459 static int submit_extent_page(int rw, struct extent_io_tree *tree,
2460 struct page *page, sector_t sector,
2461 size_t size, unsigned long offset,
2462 struct block_device *bdev,
2463 struct bio **bio_ret,
2464 unsigned long max_pages,
2465 bio_end_io_t end_io_func,
2467 unsigned long prev_bio_flags,
2468 unsigned long bio_flags)
2474 int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED;
2475 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
2476 size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE);
2478 if (bio_ret && *bio_ret) {
2481 contig = bio->bi_sector == sector;
2483 contig = bio->bi_sector + (bio->bi_size >> 9) ==
2486 if (prev_bio_flags != bio_flags || !contig ||
2487 merge_bio(tree, page, offset, page_size, bio, bio_flags) ||
2488 bio_add_page(bio, page, page_size, offset) < page_size) {
2489 ret = submit_one_bio(rw, bio, mirror_num,
2498 if (this_compressed)
2501 nr = bio_get_nr_vecs(bdev);
2503 bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
2507 bio_add_page(bio, page, page_size, offset);
2508 bio->bi_end_io = end_io_func;
2509 bio->bi_private = tree;
2514 ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
2519 void attach_extent_buffer_page(struct extent_buffer *eb, struct page *page)
2521 if (!PagePrivate(page)) {
2522 SetPagePrivate(page);
2523 page_cache_get(page);
2524 set_page_private(page, (unsigned long)eb);
2526 WARN_ON(page->private != (unsigned long)eb);
2530 void set_page_extent_mapped(struct page *page)
2532 if (!PagePrivate(page)) {
2533 SetPagePrivate(page);
2534 page_cache_get(page);
2535 set_page_private(page, EXTENT_PAGE_PRIVATE);
2540 * basic readpage implementation. Locked extent state structs are inserted
2541 * into the tree that are removed when the IO is done (by the end_io
2543 * XXX JDM: This needs looking at to ensure proper page locking
2545 static int __extent_read_full_page(struct extent_io_tree *tree,
2547 get_extent_t *get_extent,
2548 struct bio **bio, int mirror_num,
2549 unsigned long *bio_flags)
2551 struct inode *inode = page->mapping->host;
2552 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2553 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2557 u64 last_byte = i_size_read(inode);
2561 struct extent_map *em;
2562 struct block_device *bdev;
2563 struct btrfs_ordered_extent *ordered;
2566 size_t pg_offset = 0;
2568 size_t disk_io_size;
2569 size_t blocksize = inode->i_sb->s_blocksize;
2570 unsigned long this_bio_flag = 0;
2572 set_page_extent_mapped(page);
2574 if (!PageUptodate(page)) {
2575 if (cleancache_get_page(page) == 0) {
2576 BUG_ON(blocksize != PAGE_SIZE);
2583 lock_extent(tree, start, end);
2584 ordered = btrfs_lookup_ordered_extent(inode, start);
2587 unlock_extent(tree, start, end);
2588 btrfs_start_ordered_extent(inode, ordered, 1);
2589 btrfs_put_ordered_extent(ordered);
2592 if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
2594 size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
2597 iosize = PAGE_CACHE_SIZE - zero_offset;
2598 userpage = kmap_atomic(page);
2599 memset(userpage + zero_offset, 0, iosize);
2600 flush_dcache_page(page);
2601 kunmap_atomic(userpage);
2604 while (cur <= end) {
2605 if (cur >= last_byte) {
2607 struct extent_state *cached = NULL;
2609 iosize = PAGE_CACHE_SIZE - pg_offset;
2610 userpage = kmap_atomic(page);
2611 memset(userpage + pg_offset, 0, iosize);
2612 flush_dcache_page(page);
2613 kunmap_atomic(userpage);
2614 set_extent_uptodate(tree, cur, cur + iosize - 1,
2616 unlock_extent_cached(tree, cur, cur + iosize - 1,
2620 em = get_extent(inode, page, pg_offset, cur,
2622 if (IS_ERR_OR_NULL(em)) {
2624 unlock_extent(tree, cur, end);
2627 extent_offset = cur - em->start;
2628 BUG_ON(extent_map_end(em) <= cur);
2631 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2632 this_bio_flag = EXTENT_BIO_COMPRESSED;
2633 extent_set_compress_type(&this_bio_flag,
2637 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2638 cur_end = min(extent_map_end(em) - 1, end);
2639 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2640 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2641 disk_io_size = em->block_len;
2642 sector = em->block_start >> 9;
2644 sector = (em->block_start + extent_offset) >> 9;
2645 disk_io_size = iosize;
2648 block_start = em->block_start;
2649 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2650 block_start = EXTENT_MAP_HOLE;
2651 free_extent_map(em);
2654 /* we've found a hole, just zero and go on */
2655 if (block_start == EXTENT_MAP_HOLE) {
2657 struct extent_state *cached = NULL;
2659 userpage = kmap_atomic(page);
2660 memset(userpage + pg_offset, 0, iosize);
2661 flush_dcache_page(page);
2662 kunmap_atomic(userpage);
2664 set_extent_uptodate(tree, cur, cur + iosize - 1,
2666 unlock_extent_cached(tree, cur, cur + iosize - 1,
2669 pg_offset += iosize;
2672 /* the get_extent function already copied into the page */
2673 if (test_range_bit(tree, cur, cur_end,
2674 EXTENT_UPTODATE, 1, NULL)) {
2675 check_page_uptodate(tree, page);
2676 unlock_extent(tree, cur, cur + iosize - 1);
2678 pg_offset += iosize;
2681 /* we have an inline extent but it didn't get marked up
2682 * to date. Error out
2684 if (block_start == EXTENT_MAP_INLINE) {
2686 unlock_extent(tree, cur, cur + iosize - 1);
2688 pg_offset += iosize;
2693 if (tree->ops && tree->ops->readpage_io_hook) {
2694 ret = tree->ops->readpage_io_hook(page, cur,
2698 unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
2700 ret = submit_extent_page(READ, tree, page,
2701 sector, disk_io_size, pg_offset,
2703 end_bio_extent_readpage, mirror_num,
2706 BUG_ON(ret == -ENOMEM);
2708 *bio_flags = this_bio_flag;
2713 pg_offset += iosize;
2717 if (!PageError(page))
2718 SetPageUptodate(page);
2724 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
2725 get_extent_t *get_extent, int mirror_num)
2727 struct bio *bio = NULL;
2728 unsigned long bio_flags = 0;
2731 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
2734 ret = submit_one_bio(READ, bio, mirror_num, bio_flags);
2738 static noinline void update_nr_written(struct page *page,
2739 struct writeback_control *wbc,
2740 unsigned long nr_written)
2742 wbc->nr_to_write -= nr_written;
2743 if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
2744 wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
2745 page->mapping->writeback_index = page->index + nr_written;
2749 * the writepage semantics are similar to regular writepage. extent
2750 * records are inserted to lock ranges in the tree, and as dirty areas
2751 * are found, they are marked writeback. Then the lock bits are removed
2752 * and the end_io handler clears the writeback ranges
2754 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2757 struct inode *inode = page->mapping->host;
2758 struct extent_page_data *epd = data;
2759 struct extent_io_tree *tree = epd->tree;
2760 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2762 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2766 u64 last_byte = i_size_read(inode);
2770 struct extent_state *cached_state = NULL;
2771 struct extent_map *em;
2772 struct block_device *bdev;
2775 size_t pg_offset = 0;
2777 loff_t i_size = i_size_read(inode);
2778 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
2784 unsigned long nr_written = 0;
2785 bool fill_delalloc = true;
2787 if (wbc->sync_mode == WB_SYNC_ALL)
2788 write_flags = WRITE_SYNC;
2790 write_flags = WRITE;
2792 trace___extent_writepage(page, inode, wbc);
2794 WARN_ON(!PageLocked(page));
2796 ClearPageError(page);
2798 pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
2799 if (page->index > end_index ||
2800 (page->index == end_index && !pg_offset)) {
2801 page->mapping->a_ops->invalidatepage(page, 0);
2806 if (page->index == end_index) {
2809 userpage = kmap_atomic(page);
2810 memset(userpage + pg_offset, 0,
2811 PAGE_CACHE_SIZE - pg_offset);
2812 kunmap_atomic(userpage);
2813 flush_dcache_page(page);
2817 set_page_extent_mapped(page);
2819 if (!tree->ops || !tree->ops->fill_delalloc)
2820 fill_delalloc = false;
2822 delalloc_start = start;
2825 if (!epd->extent_locked && fill_delalloc) {
2826 u64 delalloc_to_write = 0;
2828 * make sure the wbc mapping index is at least updated
2831 update_nr_written(page, wbc, 0);
2833 while (delalloc_end < page_end) {
2834 nr_delalloc = find_lock_delalloc_range(inode, tree,
2839 if (nr_delalloc == 0) {
2840 delalloc_start = delalloc_end + 1;
2843 ret = tree->ops->fill_delalloc(inode, page,
2848 /* File system has been set read-only */
2854 * delalloc_end is already one less than the total
2855 * length, so we don't subtract one from
2858 delalloc_to_write += (delalloc_end - delalloc_start +
2861 delalloc_start = delalloc_end + 1;
2863 if (wbc->nr_to_write < delalloc_to_write) {
2866 if (delalloc_to_write < thresh * 2)
2867 thresh = delalloc_to_write;
2868 wbc->nr_to_write = min_t(u64, delalloc_to_write,
2872 /* did the fill delalloc function already unlock and start
2878 * we've unlocked the page, so we can't update
2879 * the mapping's writeback index, just update
2882 wbc->nr_to_write -= nr_written;
2886 if (tree->ops && tree->ops->writepage_start_hook) {
2887 ret = tree->ops->writepage_start_hook(page, start,
2890 /* Fixup worker will requeue */
2892 wbc->pages_skipped++;
2894 redirty_page_for_writepage(wbc, page);
2895 update_nr_written(page, wbc, nr_written);
2903 * we don't want to touch the inode after unlocking the page,
2904 * so we update the mapping writeback index now
2906 update_nr_written(page, wbc, nr_written + 1);
2909 if (last_byte <= start) {
2910 if (tree->ops && tree->ops->writepage_end_io_hook)
2911 tree->ops->writepage_end_io_hook(page, start,
2916 blocksize = inode->i_sb->s_blocksize;
2918 while (cur <= end) {
2919 if (cur >= last_byte) {
2920 if (tree->ops && tree->ops->writepage_end_io_hook)
2921 tree->ops->writepage_end_io_hook(page, cur,
2925 em = epd->get_extent(inode, page, pg_offset, cur,
2927 if (IS_ERR_OR_NULL(em)) {
2932 extent_offset = cur - em->start;
2933 BUG_ON(extent_map_end(em) <= cur);
2935 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2936 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2937 sector = (em->block_start + extent_offset) >> 9;
2939 block_start = em->block_start;
2940 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
2941 free_extent_map(em);
2945 * compressed and inline extents are written through other
2948 if (compressed || block_start == EXTENT_MAP_HOLE ||
2949 block_start == EXTENT_MAP_INLINE) {
2951 * end_io notification does not happen here for
2952 * compressed extents
2954 if (!compressed && tree->ops &&
2955 tree->ops->writepage_end_io_hook)
2956 tree->ops->writepage_end_io_hook(page, cur,
2959 else if (compressed) {
2960 /* we don't want to end_page_writeback on
2961 * a compressed extent. this happens
2968 pg_offset += iosize;
2971 /* leave this out until we have a page_mkwrite call */
2972 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
2973 EXTENT_DIRTY, 0, NULL)) {
2975 pg_offset += iosize;
2979 if (tree->ops && tree->ops->writepage_io_hook) {
2980 ret = tree->ops->writepage_io_hook(page, cur,
2988 unsigned long max_nr = end_index + 1;
2990 set_range_writeback(tree, cur, cur + iosize - 1);
2991 if (!PageWriteback(page)) {
2992 printk(KERN_ERR "btrfs warning page %lu not "
2993 "writeback, cur %llu end %llu\n",
2994 page->index, (unsigned long long)cur,
2995 (unsigned long long)end);
2998 ret = submit_extent_page(write_flags, tree, page,
2999 sector, iosize, pg_offset,
3000 bdev, &epd->bio, max_nr,
3001 end_bio_extent_writepage,
3007 pg_offset += iosize;
3012 /* make sure the mapping tag for page dirty gets cleared */
3013 set_page_writeback(page);
3014 end_page_writeback(page);
3020 /* drop our reference on any cached states */
3021 free_extent_state(cached_state);
3025 static int eb_wait(void *word)
3031 static void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3033 wait_on_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK, eb_wait,
3034 TASK_UNINTERRUPTIBLE);
3037 static int lock_extent_buffer_for_io(struct extent_buffer *eb,
3038 struct btrfs_fs_info *fs_info,
3039 struct extent_page_data *epd)
3041 unsigned long i, num_pages;
3045 if (!btrfs_try_tree_write_lock(eb)) {
3047 flush_write_bio(epd);
3048 btrfs_tree_lock(eb);
3051 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3052 btrfs_tree_unlock(eb);
3056 flush_write_bio(epd);
3060 wait_on_extent_buffer_writeback(eb);
3061 btrfs_tree_lock(eb);
3062 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3064 btrfs_tree_unlock(eb);
3069 * We need to do this to prevent races in people who check if the eb is
3070 * under IO since we can end up having no IO bits set for a short period
3073 spin_lock(&eb->refs_lock);
3074 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3075 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3076 spin_unlock(&eb->refs_lock);
3077 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3078 spin_lock(&fs_info->delalloc_lock);
3079 if (fs_info->dirty_metadata_bytes >= eb->len)
3080 fs_info->dirty_metadata_bytes -= eb->len;
3083 spin_unlock(&fs_info->delalloc_lock);
3086 spin_unlock(&eb->refs_lock);
3089 btrfs_tree_unlock(eb);
3094 num_pages = num_extent_pages(eb->start, eb->len);
3095 for (i = 0; i < num_pages; i++) {
3096 struct page *p = extent_buffer_page(eb, i);
3098 if (!trylock_page(p)) {
3100 flush_write_bio(epd);
3110 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3112 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3113 smp_mb__after_clear_bit();
3114 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3117 static void end_bio_extent_buffer_writepage(struct bio *bio, int err)
3119 int uptodate = err == 0;
3120 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
3121 struct extent_buffer *eb;
3125 struct page *page = bvec->bv_page;
3128 eb = (struct extent_buffer *)page->private;
3130 done = atomic_dec_and_test(&eb->io_pages);
3132 if (!uptodate || test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
3133 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3134 ClearPageUptodate(page);
3138 end_page_writeback(page);
3143 end_extent_buffer_writeback(eb);
3144 } while (bvec >= bio->bi_io_vec);
3150 static int write_one_eb(struct extent_buffer *eb,
3151 struct btrfs_fs_info *fs_info,
3152 struct writeback_control *wbc,
3153 struct extent_page_data *epd)
3155 struct block_device *bdev = fs_info->fs_devices->latest_bdev;
3156 u64 offset = eb->start;
3157 unsigned long i, num_pages;
3158 int rw = (epd->sync_io ? WRITE_SYNC : WRITE);
3161 clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3162 num_pages = num_extent_pages(eb->start, eb->len);
3163 atomic_set(&eb->io_pages, num_pages);
3164 for (i = 0; i < num_pages; i++) {
3165 struct page *p = extent_buffer_page(eb, i);
3167 clear_page_dirty_for_io(p);
3168 set_page_writeback(p);
3169 ret = submit_extent_page(rw, eb->tree, p, offset >> 9,
3170 PAGE_CACHE_SIZE, 0, bdev, &epd->bio,
3171 -1, end_bio_extent_buffer_writepage,
3174 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3176 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3177 end_extent_buffer_writeback(eb);
3181 offset += PAGE_CACHE_SIZE;
3182 update_nr_written(p, wbc, 1);
3186 if (unlikely(ret)) {
3187 for (; i < num_pages; i++) {
3188 struct page *p = extent_buffer_page(eb, i);
3196 int btree_write_cache_pages(struct address_space *mapping,
3197 struct writeback_control *wbc)
3199 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
3200 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
3201 struct extent_buffer *eb, *prev_eb = NULL;
3202 struct extent_page_data epd = {
3206 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3210 int nr_to_write_done = 0;
3211 struct pagevec pvec;
3214 pgoff_t end; /* Inclusive */
3218 pagevec_init(&pvec, 0);
3219 if (wbc->range_cyclic) {
3220 index = mapping->writeback_index; /* Start from prev offset */
3223 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3224 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3227 if (wbc->sync_mode == WB_SYNC_ALL)
3228 tag = PAGECACHE_TAG_TOWRITE;
3230 tag = PAGECACHE_TAG_DIRTY;
3232 if (wbc->sync_mode == WB_SYNC_ALL)
3233 tag_pages_for_writeback(mapping, index, end);
3234 while (!done && !nr_to_write_done && (index <= end) &&
3235 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3236 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3240 for (i = 0; i < nr_pages; i++) {
3241 struct page *page = pvec.pages[i];
3243 if (!PagePrivate(page))
3246 if (!wbc->range_cyclic && page->index > end) {
3251 eb = (struct extent_buffer *)page->private;
3260 if (!atomic_inc_not_zero(&eb->refs)) {
3266 ret = lock_extent_buffer_for_io(eb, fs_info, &epd);
3268 free_extent_buffer(eb);
3272 ret = write_one_eb(eb, fs_info, wbc, &epd);
3275 free_extent_buffer(eb);
3278 free_extent_buffer(eb);
3281 * the filesystem may choose to bump up nr_to_write.
3282 * We have to make sure to honor the new nr_to_write
3285 nr_to_write_done = wbc->nr_to_write <= 0;
3287 pagevec_release(&pvec);
3290 if (!scanned && !done) {
3292 * We hit the last page and there is more work to be done: wrap
3293 * back to the start of the file
3299 flush_write_bio(&epd);
3304 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3305 * @mapping: address space structure to write
3306 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3307 * @writepage: function called for each page
3308 * @data: data passed to writepage function
3310 * If a page is already under I/O, write_cache_pages() skips it, even
3311 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3312 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3313 * and msync() need to guarantee that all the data which was dirty at the time
3314 * the call was made get new I/O started against them. If wbc->sync_mode is
3315 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3316 * existing IO to complete.
3318 static int extent_write_cache_pages(struct extent_io_tree *tree,
3319 struct address_space *mapping,
3320 struct writeback_control *wbc,
3321 writepage_t writepage, void *data,
3322 void (*flush_fn)(void *))
3324 struct inode *inode = mapping->host;
3327 int nr_to_write_done = 0;
3328 struct pagevec pvec;
3331 pgoff_t end; /* Inclusive */
3336 * We have to hold onto the inode so that ordered extents can do their
3337 * work when the IO finishes. The alternative to this is failing to add
3338 * an ordered extent if the igrab() fails there and that is a huge pain
3339 * to deal with, so instead just hold onto the inode throughout the
3340 * writepages operation. If it fails here we are freeing up the inode
3341 * anyway and we'd rather not waste our time writing out stuff that is
3342 * going to be truncated anyway.
3347 pagevec_init(&pvec, 0);
3348 if (wbc->range_cyclic) {
3349 index = mapping->writeback_index; /* Start from prev offset */
3352 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3353 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3356 if (wbc->sync_mode == WB_SYNC_ALL)
3357 tag = PAGECACHE_TAG_TOWRITE;
3359 tag = PAGECACHE_TAG_DIRTY;
3361 if (wbc->sync_mode == WB_SYNC_ALL)
3362 tag_pages_for_writeback(mapping, index, end);
3363 while (!done && !nr_to_write_done && (index <= end) &&
3364 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3365 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3369 for (i = 0; i < nr_pages; i++) {
3370 struct page *page = pvec.pages[i];
3373 * At this point we hold neither mapping->tree_lock nor
3374 * lock on the page itself: the page may be truncated or
3375 * invalidated (changing page->mapping to NULL), or even
3376 * swizzled back from swapper_space to tmpfs file
3380 tree->ops->write_cache_pages_lock_hook) {
3381 tree->ops->write_cache_pages_lock_hook(page,
3384 if (!trylock_page(page)) {
3390 if (unlikely(page->mapping != mapping)) {
3395 if (!wbc->range_cyclic && page->index > end) {
3401 if (wbc->sync_mode != WB_SYNC_NONE) {
3402 if (PageWriteback(page))
3404 wait_on_page_writeback(page);
3407 if (PageWriteback(page) ||
3408 !clear_page_dirty_for_io(page)) {
3413 ret = (*writepage)(page, wbc, data);
3415 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
3423 * the filesystem may choose to bump up nr_to_write.
3424 * We have to make sure to honor the new nr_to_write
3427 nr_to_write_done = wbc->nr_to_write <= 0;
3429 pagevec_release(&pvec);
3432 if (!scanned && !done) {
3434 * We hit the last page and there is more work to be done: wrap
3435 * back to the start of the file
3441 btrfs_add_delayed_iput(inode);
3445 static void flush_epd_write_bio(struct extent_page_data *epd)
3454 ret = submit_one_bio(rw, epd->bio, 0, 0);
3455 BUG_ON(ret < 0); /* -ENOMEM */
3460 static noinline void flush_write_bio(void *data)
3462 struct extent_page_data *epd = data;
3463 flush_epd_write_bio(epd);
3466 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
3467 get_extent_t *get_extent,
3468 struct writeback_control *wbc)
3471 struct extent_page_data epd = {
3474 .get_extent = get_extent,
3476 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3479 ret = __extent_writepage(page, wbc, &epd);
3481 flush_epd_write_bio(&epd);
3485 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
3486 u64 start, u64 end, get_extent_t *get_extent,
3490 struct address_space *mapping = inode->i_mapping;
3492 unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
3495 struct extent_page_data epd = {
3498 .get_extent = get_extent,
3500 .sync_io = mode == WB_SYNC_ALL,
3502 struct writeback_control wbc_writepages = {
3504 .nr_to_write = nr_pages * 2,
3505 .range_start = start,
3506 .range_end = end + 1,
3509 while (start <= end) {
3510 page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
3511 if (clear_page_dirty_for_io(page))
3512 ret = __extent_writepage(page, &wbc_writepages, &epd);
3514 if (tree->ops && tree->ops->writepage_end_io_hook)
3515 tree->ops->writepage_end_io_hook(page, start,
3516 start + PAGE_CACHE_SIZE - 1,
3520 page_cache_release(page);
3521 start += PAGE_CACHE_SIZE;
3524 flush_epd_write_bio(&epd);
3528 int extent_writepages(struct extent_io_tree *tree,
3529 struct address_space *mapping,
3530 get_extent_t *get_extent,
3531 struct writeback_control *wbc)
3534 struct extent_page_data epd = {
3537 .get_extent = get_extent,
3539 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3542 ret = extent_write_cache_pages(tree, mapping, wbc,
3543 __extent_writepage, &epd,
3545 flush_epd_write_bio(&epd);
3549 int extent_readpages(struct extent_io_tree *tree,
3550 struct address_space *mapping,
3551 struct list_head *pages, unsigned nr_pages,
3552 get_extent_t get_extent)
3554 struct bio *bio = NULL;
3556 unsigned long bio_flags = 0;
3557 struct page *pagepool[16];
3562 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
3563 page = list_entry(pages->prev, struct page, lru);
3565 prefetchw(&page->flags);
3566 list_del(&page->lru);
3567 if (add_to_page_cache_lru(page, mapping,
3568 page->index, GFP_NOFS)) {
3569 page_cache_release(page);
3573 pagepool[nr++] = page;
3574 if (nr < ARRAY_SIZE(pagepool))
3576 for (i = 0; i < nr; i++) {
3577 __extent_read_full_page(tree, pagepool[i], get_extent,
3578 &bio, 0, &bio_flags);
3579 page_cache_release(pagepool[i]);
3583 for (i = 0; i < nr; i++) {
3584 __extent_read_full_page(tree, pagepool[i], get_extent,
3585 &bio, 0, &bio_flags);
3586 page_cache_release(pagepool[i]);
3589 BUG_ON(!list_empty(pages));
3591 return submit_one_bio(READ, bio, 0, bio_flags);
3596 * basic invalidatepage code, this waits on any locked or writeback
3597 * ranges corresponding to the page, and then deletes any extent state
3598 * records from the tree
3600 int extent_invalidatepage(struct extent_io_tree *tree,
3601 struct page *page, unsigned long offset)
3603 struct extent_state *cached_state = NULL;
3604 u64 start = ((u64)page->index << PAGE_CACHE_SHIFT);
3605 u64 end = start + PAGE_CACHE_SIZE - 1;
3606 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
3608 start += (offset + blocksize - 1) & ~(blocksize - 1);
3612 lock_extent_bits(tree, start, end, 0, &cached_state);
3613 wait_on_page_writeback(page);
3614 clear_extent_bit(tree, start, end,
3615 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
3616 EXTENT_DO_ACCOUNTING,
3617 1, 1, &cached_state, GFP_NOFS);
3622 * a helper for releasepage, this tests for areas of the page that
3623 * are locked or under IO and drops the related state bits if it is safe
3626 int try_release_extent_state(struct extent_map_tree *map,
3627 struct extent_io_tree *tree, struct page *page,
3630 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
3631 u64 end = start + PAGE_CACHE_SIZE - 1;
3634 if (test_range_bit(tree, start, end,
3635 EXTENT_IOBITS, 0, NULL))
3638 if ((mask & GFP_NOFS) == GFP_NOFS)
3641 * at this point we can safely clear everything except the
3642 * locked bit and the nodatasum bit
3644 ret = clear_extent_bit(tree, start, end,
3645 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
3648 /* if clear_extent_bit failed for enomem reasons,
3649 * we can't allow the release to continue.
3660 * a helper for releasepage. As long as there are no locked extents
3661 * in the range corresponding to the page, both state records and extent
3662 * map records are removed
3664 int try_release_extent_mapping(struct extent_map_tree *map,
3665 struct extent_io_tree *tree, struct page *page,
3668 struct extent_map *em;
3669 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
3670 u64 end = start + PAGE_CACHE_SIZE - 1;
3672 if ((mask & __GFP_WAIT) &&
3673 page->mapping->host->i_size > 16 * 1024 * 1024) {
3675 while (start <= end) {
3676 len = end - start + 1;
3677 write_lock(&map->lock);
3678 em = lookup_extent_mapping(map, start, len);
3680 write_unlock(&map->lock);
3683 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
3684 em->start != start) {
3685 write_unlock(&map->lock);
3686 free_extent_map(em);
3689 if (!test_range_bit(tree, em->start,
3690 extent_map_end(em) - 1,
3691 EXTENT_LOCKED | EXTENT_WRITEBACK,
3693 remove_extent_mapping(map, em);
3694 /* once for the rb tree */
3695 free_extent_map(em);
3697 start = extent_map_end(em);
3698 write_unlock(&map->lock);
3701 free_extent_map(em);
3704 return try_release_extent_state(map, tree, page, mask);
3708 * helper function for fiemap, which doesn't want to see any holes.
3709 * This maps until we find something past 'last'
3711 static struct extent_map *get_extent_skip_holes(struct inode *inode,
3714 get_extent_t *get_extent)
3716 u64 sectorsize = BTRFS_I(inode)->root->sectorsize;
3717 struct extent_map *em;
3724 len = last - offset;
3727 len = (len + sectorsize - 1) & ~(sectorsize - 1);
3728 em = get_extent(inode, NULL, 0, offset, len, 0);
3729 if (IS_ERR_OR_NULL(em))
3732 /* if this isn't a hole return it */
3733 if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) &&
3734 em->block_start != EXTENT_MAP_HOLE) {
3738 /* this is a hole, advance to the next extent */
3739 offset = extent_map_end(em);
3740 free_extent_map(em);
3747 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
3748 __u64 start, __u64 len, get_extent_t *get_extent)
3752 u64 max = start + len;
3756 u64 last_for_get_extent = 0;
3758 u64 isize = i_size_read(inode);
3759 struct btrfs_key found_key;
3760 struct extent_map *em = NULL;
3761 struct extent_state *cached_state = NULL;
3762 struct btrfs_path *path;
3763 struct btrfs_file_extent_item *item;
3768 unsigned long emflags;
3773 path = btrfs_alloc_path();
3776 path->leave_spinning = 1;
3778 start = ALIGN(start, BTRFS_I(inode)->root->sectorsize);
3779 len = ALIGN(len, BTRFS_I(inode)->root->sectorsize);
3782 * lookup the last file extent. We're not using i_size here
3783 * because there might be preallocation past i_size
3785 ret = btrfs_lookup_file_extent(NULL, BTRFS_I(inode)->root,
3786 path, btrfs_ino(inode), -1, 0);
3788 btrfs_free_path(path);
3793 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3794 struct btrfs_file_extent_item);
3795 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
3796 found_type = btrfs_key_type(&found_key);
3798 /* No extents, but there might be delalloc bits */
3799 if (found_key.objectid != btrfs_ino(inode) ||
3800 found_type != BTRFS_EXTENT_DATA_KEY) {
3801 /* have to trust i_size as the end */
3803 last_for_get_extent = isize;
3806 * remember the start of the last extent. There are a
3807 * bunch of different factors that go into the length of the
3808 * extent, so its much less complex to remember where it started
3810 last = found_key.offset;
3811 last_for_get_extent = last + 1;
3813 btrfs_free_path(path);
3816 * we might have some extents allocated but more delalloc past those
3817 * extents. so, we trust isize unless the start of the last extent is
3822 last_for_get_extent = isize;
3825 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len, 0,
3828 em = get_extent_skip_holes(inode, start, last_for_get_extent,
3838 u64 offset_in_extent;
3840 /* break if the extent we found is outside the range */
3841 if (em->start >= max || extent_map_end(em) < off)
3845 * get_extent may return an extent that starts before our
3846 * requested range. We have to make sure the ranges
3847 * we return to fiemap always move forward and don't
3848 * overlap, so adjust the offsets here
3850 em_start = max(em->start, off);
3853 * record the offset from the start of the extent
3854 * for adjusting the disk offset below
3856 offset_in_extent = em_start - em->start;
3857 em_end = extent_map_end(em);
3858 em_len = em_end - em_start;
3859 emflags = em->flags;
3864 * bump off for our next call to get_extent
3866 off = extent_map_end(em);
3870 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
3872 flags |= FIEMAP_EXTENT_LAST;
3873 } else if (em->block_start == EXTENT_MAP_INLINE) {
3874 flags |= (FIEMAP_EXTENT_DATA_INLINE |
3875 FIEMAP_EXTENT_NOT_ALIGNED);
3876 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
3877 flags |= (FIEMAP_EXTENT_DELALLOC |
3878 FIEMAP_EXTENT_UNKNOWN);
3880 disko = em->block_start + offset_in_extent;
3882 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
3883 flags |= FIEMAP_EXTENT_ENCODED;
3885 free_extent_map(em);
3887 if ((em_start >= last) || em_len == (u64)-1 ||
3888 (last == (u64)-1 && isize <= em_end)) {
3889 flags |= FIEMAP_EXTENT_LAST;
3893 /* now scan forward to see if this is really the last extent. */
3894 em = get_extent_skip_holes(inode, off, last_for_get_extent,
3901 flags |= FIEMAP_EXTENT_LAST;
3904 ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
3910 free_extent_map(em);
3912 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len,
3913 &cached_state, GFP_NOFS);
3917 inline struct page *extent_buffer_page(struct extent_buffer *eb,
3920 return eb->pages[i];
3923 inline unsigned long num_extent_pages(u64 start, u64 len)
3925 return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
3926 (start >> PAGE_CACHE_SHIFT);
3929 static void __free_extent_buffer(struct extent_buffer *eb)
3932 unsigned long flags;
3933 spin_lock_irqsave(&leak_lock, flags);
3934 list_del(&eb->leak_list);
3935 spin_unlock_irqrestore(&leak_lock, flags);
3937 if (eb->pages && eb->pages != eb->inline_pages)
3939 kmem_cache_free(extent_buffer_cache, eb);
3942 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
3947 struct extent_buffer *eb = NULL;
3949 unsigned long flags;
3952 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
3959 rwlock_init(&eb->lock);
3960 atomic_set(&eb->write_locks, 0);
3961 atomic_set(&eb->read_locks, 0);
3962 atomic_set(&eb->blocking_readers, 0);
3963 atomic_set(&eb->blocking_writers, 0);
3964 atomic_set(&eb->spinning_readers, 0);
3965 atomic_set(&eb->spinning_writers, 0);
3966 eb->lock_nested = 0;
3967 init_waitqueue_head(&eb->write_lock_wq);
3968 init_waitqueue_head(&eb->read_lock_wq);
3971 spin_lock_irqsave(&leak_lock, flags);
3972 list_add(&eb->leak_list, &buffers);
3973 spin_unlock_irqrestore(&leak_lock, flags);
3975 spin_lock_init(&eb->refs_lock);
3976 atomic_set(&eb->refs, 1);
3977 atomic_set(&eb->io_pages, 0);
3979 if (len > MAX_INLINE_EXTENT_BUFFER_SIZE) {
3980 struct page **pages;
3981 int num_pages = (len + PAGE_CACHE_SIZE - 1) >>
3983 pages = kzalloc(num_pages, mask);
3985 __free_extent_buffer(eb);
3990 eb->pages = eb->inline_pages;
3996 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src)
4000 struct extent_buffer *new;
4001 unsigned long num_pages = num_extent_pages(src->start, src->len);
4003 new = __alloc_extent_buffer(NULL, src->start, src->len, GFP_ATOMIC);
4007 for (i = 0; i < num_pages; i++) {
4008 p = alloc_page(GFP_ATOMIC);
4010 attach_extent_buffer_page(new, p);
4011 WARN_ON(PageDirty(p));
4016 copy_extent_buffer(new, src, 0, 0, src->len);
4017 set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
4018 set_bit(EXTENT_BUFFER_DUMMY, &new->bflags);
4023 struct extent_buffer *alloc_dummy_extent_buffer(u64 start, unsigned long len)
4025 struct extent_buffer *eb;
4026 unsigned long num_pages = num_extent_pages(0, len);
4029 eb = __alloc_extent_buffer(NULL, start, len, GFP_ATOMIC);
4033 for (i = 0; i < num_pages; i++) {
4034 eb->pages[i] = alloc_page(GFP_ATOMIC);
4038 set_extent_buffer_uptodate(eb);
4039 btrfs_set_header_nritems(eb, 0);
4040 set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4044 for (i--; i > 0; i--)
4045 __free_page(eb->pages[i]);
4046 __free_extent_buffer(eb);
4050 static int extent_buffer_under_io(struct extent_buffer *eb)
4052 return (atomic_read(&eb->io_pages) ||
4053 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4054 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4058 * Helper for releasing extent buffer page.
4060 static void btrfs_release_extent_buffer_page(struct extent_buffer *eb,
4061 unsigned long start_idx)
4063 unsigned long index;
4064 unsigned long num_pages;
4066 int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4068 BUG_ON(extent_buffer_under_io(eb));
4070 num_pages = num_extent_pages(eb->start, eb->len);
4071 index = start_idx + num_pages;
4072 if (start_idx >= index)
4077 page = extent_buffer_page(eb, index);
4078 if (page && mapped) {
4079 spin_lock(&page->mapping->private_lock);
4081 * We do this since we'll remove the pages after we've
4082 * removed the eb from the radix tree, so we could race
4083 * and have this page now attached to the new eb. So
4084 * only clear page_private if it's still connected to
4087 if (PagePrivate(page) &&
4088 page->private == (unsigned long)eb) {
4089 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4090 BUG_ON(PageDirty(page));
4091 BUG_ON(PageWriteback(page));
4093 * We need to make sure we haven't be attached
4096 ClearPagePrivate(page);
4097 set_page_private(page, 0);
4098 /* One for the page private */
4099 page_cache_release(page);
4101 spin_unlock(&page->mapping->private_lock);
4105 /* One for when we alloced the page */
4106 page_cache_release(page);
4108 } while (index != start_idx);
4112 * Helper for releasing the extent buffer.
4114 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4116 btrfs_release_extent_buffer_page(eb, 0);
4117 __free_extent_buffer(eb);
4120 static void check_buffer_tree_ref(struct extent_buffer *eb)
4122 /* the ref bit is tricky. We have to make sure it is set
4123 * if we have the buffer dirty. Otherwise the
4124 * code to free a buffer can end up dropping a dirty
4127 * Once the ref bit is set, it won't go away while the
4128 * buffer is dirty or in writeback, and it also won't
4129 * go away while we have the reference count on the
4132 * We can't just set the ref bit without bumping the
4133 * ref on the eb because free_extent_buffer might
4134 * see the ref bit and try to clear it. If this happens
4135 * free_extent_buffer might end up dropping our original
4136 * ref by mistake and freeing the page before we are able
4137 * to add one more ref.
4139 * So bump the ref count first, then set the bit. If someone
4140 * beat us to it, drop the ref we added.
4142 spin_lock(&eb->refs_lock);
4143 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4144 atomic_inc(&eb->refs);
4145 spin_unlock(&eb->refs_lock);
4148 static void mark_extent_buffer_accessed(struct extent_buffer *eb)
4150 unsigned long num_pages, i;
4152 check_buffer_tree_ref(eb);
4154 num_pages = num_extent_pages(eb->start, eb->len);
4155 for (i = 0; i < num_pages; i++) {
4156 struct page *p = extent_buffer_page(eb, i);
4157 mark_page_accessed(p);
4161 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
4162 u64 start, unsigned long len)
4164 unsigned long num_pages = num_extent_pages(start, len);
4166 unsigned long index = start >> PAGE_CACHE_SHIFT;
4167 struct extent_buffer *eb;
4168 struct extent_buffer *exists = NULL;
4170 struct address_space *mapping = tree->mapping;
4175 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
4176 if (eb && atomic_inc_not_zero(&eb->refs)) {
4178 mark_extent_buffer_accessed(eb);
4183 eb = __alloc_extent_buffer(tree, start, len, GFP_NOFS);
4187 for (i = 0; i < num_pages; i++, index++) {
4188 p = find_or_create_page(mapping, index, GFP_NOFS);
4194 spin_lock(&mapping->private_lock);
4195 if (PagePrivate(p)) {
4197 * We could have already allocated an eb for this page
4198 * and attached one so lets see if we can get a ref on
4199 * the existing eb, and if we can we know it's good and
4200 * we can just return that one, else we know we can just
4201 * overwrite page->private.
4203 exists = (struct extent_buffer *)p->private;
4204 if (atomic_inc_not_zero(&exists->refs)) {
4205 spin_unlock(&mapping->private_lock);
4207 page_cache_release(p);
4208 mark_extent_buffer_accessed(exists);
4213 * Do this so attach doesn't complain and we need to
4214 * drop the ref the old guy had.
4216 ClearPagePrivate(p);
4217 WARN_ON(PageDirty(p));
4218 page_cache_release(p);
4220 attach_extent_buffer_page(eb, p);
4221 spin_unlock(&mapping->private_lock);
4222 WARN_ON(PageDirty(p));
4223 mark_page_accessed(p);
4225 if (!PageUptodate(p))
4229 * see below about how we avoid a nasty race with release page
4230 * and why we unlock later
4234 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4236 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
4240 spin_lock(&tree->buffer_lock);
4241 ret = radix_tree_insert(&tree->buffer, start >> PAGE_CACHE_SHIFT, eb);
4242 if (ret == -EEXIST) {
4243 exists = radix_tree_lookup(&tree->buffer,
4244 start >> PAGE_CACHE_SHIFT);
4245 if (!atomic_inc_not_zero(&exists->refs)) {
4246 spin_unlock(&tree->buffer_lock);
4247 radix_tree_preload_end();
4251 spin_unlock(&tree->buffer_lock);
4252 radix_tree_preload_end();
4253 mark_extent_buffer_accessed(exists);
4256 /* add one reference for the tree */
4257 check_buffer_tree_ref(eb);
4258 spin_unlock(&tree->buffer_lock);
4259 radix_tree_preload_end();
4262 * there is a race where release page may have
4263 * tried to find this extent buffer in the radix
4264 * but failed. It will tell the VM it is safe to
4265 * reclaim the, and it will clear the page private bit.
4266 * We must make sure to set the page private bit properly
4267 * after the extent buffer is in the radix tree so
4268 * it doesn't get lost
4270 SetPageChecked(eb->pages[0]);
4271 for (i = 1; i < num_pages; i++) {
4272 p = extent_buffer_page(eb, i);
4273 ClearPageChecked(p);
4276 unlock_page(eb->pages[0]);
4280 for (i = 0; i < num_pages; i++) {
4282 unlock_page(eb->pages[i]);
4285 WARN_ON(!atomic_dec_and_test(&eb->refs));
4286 btrfs_release_extent_buffer(eb);
4290 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
4291 u64 start, unsigned long len)
4293 struct extent_buffer *eb;
4296 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
4297 if (eb && atomic_inc_not_zero(&eb->refs)) {
4299 mark_extent_buffer_accessed(eb);
4307 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
4309 struct extent_buffer *eb =
4310 container_of(head, struct extent_buffer, rcu_head);
4312 __free_extent_buffer(eb);
4315 /* Expects to have eb->eb_lock already held */
4316 static int release_extent_buffer(struct extent_buffer *eb, gfp_t mask)
4318 WARN_ON(atomic_read(&eb->refs) == 0);
4319 if (atomic_dec_and_test(&eb->refs)) {
4320 if (test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags)) {
4321 spin_unlock(&eb->refs_lock);
4323 struct extent_io_tree *tree = eb->tree;
4325 spin_unlock(&eb->refs_lock);
4327 spin_lock(&tree->buffer_lock);
4328 radix_tree_delete(&tree->buffer,
4329 eb->start >> PAGE_CACHE_SHIFT);
4330 spin_unlock(&tree->buffer_lock);
4333 /* Should be safe to release our pages at this point */
4334 btrfs_release_extent_buffer_page(eb, 0);
4336 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
4339 spin_unlock(&eb->refs_lock);
4344 void free_extent_buffer(struct extent_buffer *eb)
4349 spin_lock(&eb->refs_lock);
4350 if (atomic_read(&eb->refs) == 2 &&
4351 test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))
4352 atomic_dec(&eb->refs);
4354 if (atomic_read(&eb->refs) == 2 &&
4355 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
4356 !extent_buffer_under_io(eb) &&
4357 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4358 atomic_dec(&eb->refs);
4361 * I know this is terrible, but it's temporary until we stop tracking
4362 * the uptodate bits and such for the extent buffers.
4364 release_extent_buffer(eb, GFP_ATOMIC);
4367 void free_extent_buffer_stale(struct extent_buffer *eb)
4372 spin_lock(&eb->refs_lock);
4373 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
4375 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
4376 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4377 atomic_dec(&eb->refs);
4378 release_extent_buffer(eb, GFP_NOFS);
4381 void clear_extent_buffer_dirty(struct extent_buffer *eb)
4384 unsigned long num_pages;
4387 num_pages = num_extent_pages(eb->start, eb->len);
4389 for (i = 0; i < num_pages; i++) {
4390 page = extent_buffer_page(eb, i);
4391 if (!PageDirty(page))
4395 WARN_ON(!PagePrivate(page));
4397 clear_page_dirty_for_io(page);
4398 spin_lock_irq(&page->mapping->tree_lock);
4399 if (!PageDirty(page)) {
4400 radix_tree_tag_clear(&page->mapping->page_tree,
4402 PAGECACHE_TAG_DIRTY);
4404 spin_unlock_irq(&page->mapping->tree_lock);
4405 ClearPageError(page);
4408 WARN_ON(atomic_read(&eb->refs) == 0);
4411 int set_extent_buffer_dirty(struct extent_buffer *eb)
4414 unsigned long num_pages;
4417 check_buffer_tree_ref(eb);
4419 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
4421 num_pages = num_extent_pages(eb->start, eb->len);
4422 WARN_ON(atomic_read(&eb->refs) == 0);
4423 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
4425 for (i = 0; i < num_pages; i++)
4426 set_page_dirty(extent_buffer_page(eb, i));
4430 static int range_straddles_pages(u64 start, u64 len)
4432 if (len < PAGE_CACHE_SIZE)
4434 if (start & (PAGE_CACHE_SIZE - 1))
4436 if ((start + len) & (PAGE_CACHE_SIZE - 1))
4441 int clear_extent_buffer_uptodate(struct extent_buffer *eb)
4445 unsigned long num_pages;
4447 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4448 num_pages = num_extent_pages(eb->start, eb->len);
4449 for (i = 0; i < num_pages; i++) {
4450 page = extent_buffer_page(eb, i);
4452 ClearPageUptodate(page);
4457 int set_extent_buffer_uptodate(struct extent_buffer *eb)
4461 unsigned long num_pages;
4463 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4464 num_pages = num_extent_pages(eb->start, eb->len);
4465 for (i = 0; i < num_pages; i++) {
4466 page = extent_buffer_page(eb, i);
4467 SetPageUptodate(page);
4472 int extent_range_uptodate(struct extent_io_tree *tree,
4477 int pg_uptodate = 1;
4479 unsigned long index;
4481 if (range_straddles_pages(start, end - start + 1)) {
4482 ret = test_range_bit(tree, start, end,
4483 EXTENT_UPTODATE, 1, NULL);
4487 while (start <= end) {
4488 index = start >> PAGE_CACHE_SHIFT;
4489 page = find_get_page(tree->mapping, index);
4492 uptodate = PageUptodate(page);
4493 page_cache_release(page);
4498 start += PAGE_CACHE_SIZE;
4503 int extent_buffer_uptodate(struct extent_buffer *eb)
4505 return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4508 int read_extent_buffer_pages(struct extent_io_tree *tree,
4509 struct extent_buffer *eb, u64 start, int wait,
4510 get_extent_t *get_extent, int mirror_num)
4513 unsigned long start_i;
4517 int locked_pages = 0;
4518 int all_uptodate = 1;
4519 unsigned long num_pages;
4520 unsigned long num_reads = 0;
4521 struct bio *bio = NULL;
4522 unsigned long bio_flags = 0;
4524 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
4528 WARN_ON(start < eb->start);
4529 start_i = (start >> PAGE_CACHE_SHIFT) -
4530 (eb->start >> PAGE_CACHE_SHIFT);
4535 num_pages = num_extent_pages(eb->start, eb->len);
4536 for (i = start_i; i < num_pages; i++) {
4537 page = extent_buffer_page(eb, i);
4538 if (wait == WAIT_NONE) {
4539 if (!trylock_page(page))
4545 if (!PageUptodate(page)) {
4552 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4556 clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
4557 eb->read_mirror = 0;
4558 atomic_set(&eb->io_pages, num_reads);
4559 for (i = start_i; i < num_pages; i++) {
4560 page = extent_buffer_page(eb, i);
4561 if (!PageUptodate(page)) {
4562 ClearPageError(page);
4563 err = __extent_read_full_page(tree, page,
4565 mirror_num, &bio_flags);
4574 err = submit_one_bio(READ, bio, mirror_num, bio_flags);
4579 if (ret || wait != WAIT_COMPLETE)
4582 for (i = start_i; i < num_pages; i++) {
4583 page = extent_buffer_page(eb, i);
4584 wait_on_page_locked(page);
4585 if (!PageUptodate(page))
4593 while (locked_pages > 0) {
4594 page = extent_buffer_page(eb, i);
4602 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
4603 unsigned long start,
4610 char *dst = (char *)dstv;
4611 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4612 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4614 WARN_ON(start > eb->len);
4615 WARN_ON(start + len > eb->start + eb->len);
4617 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4620 page = extent_buffer_page(eb, i);
4622 cur = min(len, (PAGE_CACHE_SIZE - offset));
4623 kaddr = page_address(page);
4624 memcpy(dst, kaddr + offset, cur);
4633 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
4634 unsigned long min_len, char **map,
4635 unsigned long *map_start,
4636 unsigned long *map_len)
4638 size_t offset = start & (PAGE_CACHE_SIZE - 1);
4641 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4642 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4643 unsigned long end_i = (start_offset + start + min_len - 1) >>
4650 offset = start_offset;
4654 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
4657 if (start + min_len > eb->len) {
4658 printk(KERN_ERR "btrfs bad mapping eb start %llu len %lu, "
4659 "wanted %lu %lu\n", (unsigned long long)eb->start,
4660 eb->len, start, min_len);
4665 p = extent_buffer_page(eb, i);
4666 kaddr = page_address(p);
4667 *map = kaddr + offset;
4668 *map_len = PAGE_CACHE_SIZE - offset;
4672 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
4673 unsigned long start,
4680 char *ptr = (char *)ptrv;
4681 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4682 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4685 WARN_ON(start > eb->len);
4686 WARN_ON(start + len > eb->start + eb->len);
4688 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4691 page = extent_buffer_page(eb, i);
4693 cur = min(len, (PAGE_CACHE_SIZE - offset));
4695 kaddr = page_address(page);
4696 ret = memcmp(ptr, kaddr + offset, cur);
4708 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
4709 unsigned long start, unsigned long len)
4715 char *src = (char *)srcv;
4716 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4717 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4719 WARN_ON(start > eb->len);
4720 WARN_ON(start + len > eb->start + eb->len);
4722 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4725 page = extent_buffer_page(eb, i);
4726 WARN_ON(!PageUptodate(page));
4728 cur = min(len, PAGE_CACHE_SIZE - offset);
4729 kaddr = page_address(page);
4730 memcpy(kaddr + offset, src, cur);
4739 void memset_extent_buffer(struct extent_buffer *eb, char c,
4740 unsigned long start, unsigned long len)
4746 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4747 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4749 WARN_ON(start > eb->len);
4750 WARN_ON(start + len > eb->start + eb->len);
4752 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4755 page = extent_buffer_page(eb, i);
4756 WARN_ON(!PageUptodate(page));
4758 cur = min(len, PAGE_CACHE_SIZE - offset);
4759 kaddr = page_address(page);
4760 memset(kaddr + offset, c, cur);
4768 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
4769 unsigned long dst_offset, unsigned long src_offset,
4772 u64 dst_len = dst->len;
4777 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4778 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
4780 WARN_ON(src->len != dst_len);
4782 offset = (start_offset + dst_offset) &
4783 ((unsigned long)PAGE_CACHE_SIZE - 1);
4786 page = extent_buffer_page(dst, i);
4787 WARN_ON(!PageUptodate(page));
4789 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
4791 kaddr = page_address(page);
4792 read_extent_buffer(src, kaddr + offset, src_offset, cur);
4801 static void move_pages(struct page *dst_page, struct page *src_page,
4802 unsigned long dst_off, unsigned long src_off,
4805 char *dst_kaddr = page_address(dst_page);
4806 if (dst_page == src_page) {
4807 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
4809 char *src_kaddr = page_address(src_page);
4810 char *p = dst_kaddr + dst_off + len;
4811 char *s = src_kaddr + src_off + len;
4818 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
4820 unsigned long distance = (src > dst) ? src - dst : dst - src;
4821 return distance < len;
4824 static void copy_pages(struct page *dst_page, struct page *src_page,
4825 unsigned long dst_off, unsigned long src_off,
4828 char *dst_kaddr = page_address(dst_page);
4830 int must_memmove = 0;
4832 if (dst_page != src_page) {
4833 src_kaddr = page_address(src_page);
4835 src_kaddr = dst_kaddr;
4836 if (areas_overlap(src_off, dst_off, len))
4841 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
4843 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
4846 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
4847 unsigned long src_offset, unsigned long len)
4850 size_t dst_off_in_page;
4851 size_t src_off_in_page;
4852 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4853 unsigned long dst_i;
4854 unsigned long src_i;
4856 if (src_offset + len > dst->len) {
4857 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
4858 "len %lu dst len %lu\n", src_offset, len, dst->len);
4861 if (dst_offset + len > dst->len) {
4862 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
4863 "len %lu dst len %lu\n", dst_offset, len, dst->len);
4868 dst_off_in_page = (start_offset + dst_offset) &
4869 ((unsigned long)PAGE_CACHE_SIZE - 1);
4870 src_off_in_page = (start_offset + src_offset) &
4871 ((unsigned long)PAGE_CACHE_SIZE - 1);
4873 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
4874 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
4876 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
4878 cur = min_t(unsigned long, cur,
4879 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
4881 copy_pages(extent_buffer_page(dst, dst_i),
4882 extent_buffer_page(dst, src_i),
4883 dst_off_in_page, src_off_in_page, cur);
4891 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
4892 unsigned long src_offset, unsigned long len)
4895 size_t dst_off_in_page;
4896 size_t src_off_in_page;
4897 unsigned long dst_end = dst_offset + len - 1;
4898 unsigned long src_end = src_offset + len - 1;
4899 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4900 unsigned long dst_i;
4901 unsigned long src_i;
4903 if (src_offset + len > dst->len) {
4904 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
4905 "len %lu len %lu\n", src_offset, len, dst->len);
4908 if (dst_offset + len > dst->len) {
4909 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
4910 "len %lu len %lu\n", dst_offset, len, dst->len);
4913 if (dst_offset < src_offset) {
4914 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
4918 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
4919 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
4921 dst_off_in_page = (start_offset + dst_end) &
4922 ((unsigned long)PAGE_CACHE_SIZE - 1);
4923 src_off_in_page = (start_offset + src_end) &
4924 ((unsigned long)PAGE_CACHE_SIZE - 1);
4926 cur = min_t(unsigned long, len, src_off_in_page + 1);
4927 cur = min(cur, dst_off_in_page + 1);
4928 move_pages(extent_buffer_page(dst, dst_i),
4929 extent_buffer_page(dst, src_i),
4930 dst_off_in_page - cur + 1,
4931 src_off_in_page - cur + 1, cur);
4939 int try_release_extent_buffer(struct page *page, gfp_t mask)
4941 struct extent_buffer *eb;
4944 * We need to make sure noboody is attaching this page to an eb right
4947 spin_lock(&page->mapping->private_lock);
4948 if (!PagePrivate(page)) {
4949 spin_unlock(&page->mapping->private_lock);
4953 eb = (struct extent_buffer *)page->private;
4957 * This is a little awful but should be ok, we need to make sure that
4958 * the eb doesn't disappear out from under us while we're looking at
4961 spin_lock(&eb->refs_lock);
4962 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
4963 spin_unlock(&eb->refs_lock);
4964 spin_unlock(&page->mapping->private_lock);
4967 spin_unlock(&page->mapping->private_lock);
4969 if ((mask & GFP_NOFS) == GFP_NOFS)
4973 * If tree ref isn't set then we know the ref on this eb is a real ref,
4974 * so just return, this page will likely be freed soon anyway.
4976 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
4977 spin_unlock(&eb->refs_lock);
4981 return release_extent_buffer(eb, mask);