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;
48 unsigned long bio_flags;
50 /* tells writepage not to lock the state bits for this range
51 * it still does the unlocking
53 unsigned int extent_locked:1;
55 /* tells the submit_bio code to use a WRITE_SYNC */
56 unsigned int sync_io:1;
59 static noinline void flush_write_bio(void *data);
60 static inline struct btrfs_fs_info *
61 tree_fs_info(struct extent_io_tree *tree)
63 return btrfs_sb(tree->mapping->host->i_sb);
66 int __init extent_io_init(void)
68 extent_state_cache = kmem_cache_create("btrfs_extent_state",
69 sizeof(struct extent_state), 0,
70 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
71 if (!extent_state_cache)
74 extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer",
75 sizeof(struct extent_buffer), 0,
76 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
77 if (!extent_buffer_cache)
78 goto free_state_cache;
82 kmem_cache_destroy(extent_state_cache);
86 void extent_io_exit(void)
88 struct extent_state *state;
89 struct extent_buffer *eb;
91 while (!list_empty(&states)) {
92 state = list_entry(states.next, struct extent_state, leak_list);
93 printk(KERN_ERR "btrfs state leak: start %llu end %llu "
94 "state %lu in tree %p refs %d\n",
95 (unsigned long long)state->start,
96 (unsigned long long)state->end,
97 state->state, state->tree, atomic_read(&state->refs));
98 list_del(&state->leak_list);
99 kmem_cache_free(extent_state_cache, state);
103 while (!list_empty(&buffers)) {
104 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
105 printk(KERN_ERR "btrfs buffer leak start %llu len %lu "
106 "refs %d\n", (unsigned long long)eb->start,
107 eb->len, atomic_read(&eb->refs));
108 list_del(&eb->leak_list);
109 kmem_cache_free(extent_buffer_cache, eb);
113 * Make sure all delayed rcu free are flushed before we
117 if (extent_state_cache)
118 kmem_cache_destroy(extent_state_cache);
119 if (extent_buffer_cache)
120 kmem_cache_destroy(extent_buffer_cache);
123 void extent_io_tree_init(struct extent_io_tree *tree,
124 struct address_space *mapping)
126 tree->state = RB_ROOT;
127 INIT_RADIX_TREE(&tree->buffer, GFP_ATOMIC);
129 tree->dirty_bytes = 0;
130 spin_lock_init(&tree->lock);
131 spin_lock_init(&tree->buffer_lock);
132 tree->mapping = mapping;
135 static struct extent_state *alloc_extent_state(gfp_t mask)
137 struct extent_state *state;
142 state = kmem_cache_alloc(extent_state_cache, mask);
149 spin_lock_irqsave(&leak_lock, flags);
150 list_add(&state->leak_list, &states);
151 spin_unlock_irqrestore(&leak_lock, flags);
153 atomic_set(&state->refs, 1);
154 init_waitqueue_head(&state->wq);
155 trace_alloc_extent_state(state, mask, _RET_IP_);
159 void free_extent_state(struct extent_state *state)
163 if (atomic_dec_and_test(&state->refs)) {
167 WARN_ON(state->tree);
169 spin_lock_irqsave(&leak_lock, flags);
170 list_del(&state->leak_list);
171 spin_unlock_irqrestore(&leak_lock, flags);
173 trace_free_extent_state(state, _RET_IP_);
174 kmem_cache_free(extent_state_cache, state);
178 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
179 struct rb_node *node)
181 struct rb_node **p = &root->rb_node;
182 struct rb_node *parent = NULL;
183 struct tree_entry *entry;
187 entry = rb_entry(parent, struct tree_entry, rb_node);
189 if (offset < entry->start)
191 else if (offset > entry->end)
197 rb_link_node(node, parent, p);
198 rb_insert_color(node, root);
202 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
203 struct rb_node **prev_ret,
204 struct rb_node **next_ret)
206 struct rb_root *root = &tree->state;
207 struct rb_node *n = root->rb_node;
208 struct rb_node *prev = NULL;
209 struct rb_node *orig_prev = NULL;
210 struct tree_entry *entry;
211 struct tree_entry *prev_entry = NULL;
214 entry = rb_entry(n, struct tree_entry, rb_node);
218 if (offset < entry->start)
220 else if (offset > entry->end)
228 while (prev && offset > prev_entry->end) {
229 prev = rb_next(prev);
230 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
237 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
238 while (prev && offset < prev_entry->start) {
239 prev = rb_prev(prev);
240 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
247 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
250 struct rb_node *prev = NULL;
253 ret = __etree_search(tree, offset, &prev, NULL);
259 static void merge_cb(struct extent_io_tree *tree, struct extent_state *new,
260 struct extent_state *other)
262 if (tree->ops && tree->ops->merge_extent_hook)
263 tree->ops->merge_extent_hook(tree->mapping->host, new,
268 * utility function to look for merge candidates inside a given range.
269 * Any extents with matching state are merged together into a single
270 * extent in the tree. Extents with EXTENT_IO in their state field
271 * are not merged because the end_io handlers need to be able to do
272 * operations on them without sleeping (or doing allocations/splits).
274 * This should be called with the tree lock held.
276 static void merge_state(struct extent_io_tree *tree,
277 struct extent_state *state)
279 struct extent_state *other;
280 struct rb_node *other_node;
282 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
285 other_node = rb_prev(&state->rb_node);
287 other = rb_entry(other_node, struct extent_state, rb_node);
288 if (other->end == state->start - 1 &&
289 other->state == state->state) {
290 merge_cb(tree, state, other);
291 state->start = other->start;
293 rb_erase(&other->rb_node, &tree->state);
294 free_extent_state(other);
297 other_node = rb_next(&state->rb_node);
299 other = rb_entry(other_node, struct extent_state, rb_node);
300 if (other->start == state->end + 1 &&
301 other->state == state->state) {
302 merge_cb(tree, state, other);
303 state->end = other->end;
305 rb_erase(&other->rb_node, &tree->state);
306 free_extent_state(other);
311 static void set_state_cb(struct extent_io_tree *tree,
312 struct extent_state *state, int *bits)
314 if (tree->ops && tree->ops->set_bit_hook)
315 tree->ops->set_bit_hook(tree->mapping->host, state, bits);
318 static void clear_state_cb(struct extent_io_tree *tree,
319 struct extent_state *state, int *bits)
321 if (tree->ops && tree->ops->clear_bit_hook)
322 tree->ops->clear_bit_hook(tree->mapping->host, state, bits);
325 static void set_state_bits(struct extent_io_tree *tree,
326 struct extent_state *state, int *bits);
329 * insert an extent_state struct into the tree. 'bits' are set on the
330 * struct before it is inserted.
332 * This may return -EEXIST if the extent is already there, in which case the
333 * state struct is freed.
335 * The tree lock is not taken internally. This is a utility function and
336 * probably isn't what you want to call (see set/clear_extent_bit).
338 static int insert_state(struct extent_io_tree *tree,
339 struct extent_state *state, u64 start, u64 end,
342 struct rb_node *node;
345 WARN(1, KERN_ERR "btrfs end < start %llu %llu\n",
346 (unsigned long long)end,
347 (unsigned long long)start);
348 state->start = start;
351 set_state_bits(tree, state, bits);
353 node = tree_insert(&tree->state, end, &state->rb_node);
355 struct extent_state *found;
356 found = rb_entry(node, struct extent_state, rb_node);
357 printk(KERN_ERR "btrfs found node %llu %llu on insert of "
358 "%llu %llu\n", (unsigned long long)found->start,
359 (unsigned long long)found->end,
360 (unsigned long long)start, (unsigned long long)end);
364 merge_state(tree, state);
368 static void split_cb(struct extent_io_tree *tree, struct extent_state *orig,
371 if (tree->ops && tree->ops->split_extent_hook)
372 tree->ops->split_extent_hook(tree->mapping->host, orig, split);
376 * split a given extent state struct in two, inserting the preallocated
377 * struct 'prealloc' as the newly created second half. 'split' indicates an
378 * offset inside 'orig' where it should be split.
381 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
382 * are two extent state structs in the tree:
383 * prealloc: [orig->start, split - 1]
384 * orig: [ split, orig->end ]
386 * The tree locks are not taken by this function. They need to be held
389 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
390 struct extent_state *prealloc, u64 split)
392 struct rb_node *node;
394 split_cb(tree, orig, split);
396 prealloc->start = orig->start;
397 prealloc->end = split - 1;
398 prealloc->state = orig->state;
401 node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
403 free_extent_state(prealloc);
406 prealloc->tree = tree;
410 static struct extent_state *next_state(struct extent_state *state)
412 struct rb_node *next = rb_next(&state->rb_node);
414 return rb_entry(next, struct extent_state, rb_node);
420 * utility function to clear some bits in an extent state struct.
421 * it will optionally wake up any one waiting on this state (wake == 1).
423 * If no bits are set on the state struct after clearing things, the
424 * struct is freed and removed from the tree
426 static struct extent_state *clear_state_bit(struct extent_io_tree *tree,
427 struct extent_state *state,
430 struct extent_state *next;
431 int bits_to_clear = *bits & ~EXTENT_CTLBITS;
433 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
434 u64 range = state->end - state->start + 1;
435 WARN_ON(range > tree->dirty_bytes);
436 tree->dirty_bytes -= range;
438 clear_state_cb(tree, state, bits);
439 state->state &= ~bits_to_clear;
442 if (state->state == 0) {
443 next = next_state(state);
445 rb_erase(&state->rb_node, &tree->state);
447 free_extent_state(state);
452 merge_state(tree, state);
453 next = next_state(state);
458 static struct extent_state *
459 alloc_extent_state_atomic(struct extent_state *prealloc)
462 prealloc = alloc_extent_state(GFP_ATOMIC);
467 void extent_io_tree_panic(struct extent_io_tree *tree, int err)
469 btrfs_panic(tree_fs_info(tree), err, "Locking error: "
470 "Extent tree was modified by another "
471 "thread while locked.");
475 * clear some bits on a range in the tree. This may require splitting
476 * or inserting elements in the tree, so the gfp mask is used to
477 * indicate which allocations or sleeping are allowed.
479 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
480 * the given range from the tree regardless of state (ie for truncate).
482 * the range [start, end] is inclusive.
484 * This takes the tree lock, and returns 0 on success and < 0 on error.
486 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
487 int bits, int wake, int delete,
488 struct extent_state **cached_state,
491 struct extent_state *state;
492 struct extent_state *cached;
493 struct extent_state *prealloc = NULL;
494 struct rb_node *node;
500 bits |= ~EXTENT_CTLBITS;
501 bits |= EXTENT_FIRST_DELALLOC;
503 if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY))
506 if (!prealloc && (mask & __GFP_WAIT)) {
507 prealloc = alloc_extent_state(mask);
512 spin_lock(&tree->lock);
514 cached = *cached_state;
517 *cached_state = NULL;
521 if (cached && cached->tree && cached->start <= start &&
522 cached->end > start) {
524 atomic_dec(&cached->refs);
529 free_extent_state(cached);
532 * this search will find the extents that end after
535 node = tree_search(tree, start);
538 state = rb_entry(node, struct extent_state, rb_node);
540 if (state->start > end)
542 WARN_ON(state->end < start);
543 last_end = state->end;
545 /* the state doesn't have the wanted bits, go ahead */
546 if (!(state->state & bits)) {
547 state = next_state(state);
552 * | ---- desired range ---- |
554 * | ------------- state -------------- |
556 * We need to split the extent we found, and may flip
557 * bits on second half.
559 * If the extent we found extends past our range, we
560 * just split and search again. It'll get split again
561 * the next time though.
563 * If the extent we found is inside our range, we clear
564 * the desired bit on it.
567 if (state->start < start) {
568 prealloc = alloc_extent_state_atomic(prealloc);
570 err = split_state(tree, state, prealloc, start);
572 extent_io_tree_panic(tree, err);
577 if (state->end <= end) {
578 state = clear_state_bit(tree, state, &bits, wake);
584 * | ---- desired range ---- |
586 * We need to split the extent, and clear the bit
589 if (state->start <= end && state->end > end) {
590 prealloc = alloc_extent_state_atomic(prealloc);
592 err = split_state(tree, state, prealloc, end + 1);
594 extent_io_tree_panic(tree, err);
599 clear_state_bit(tree, prealloc, &bits, wake);
605 state = clear_state_bit(tree, state, &bits, wake);
607 if (last_end == (u64)-1)
609 start = last_end + 1;
610 if (start <= end && state && !need_resched())
615 spin_unlock(&tree->lock);
617 free_extent_state(prealloc);
624 spin_unlock(&tree->lock);
625 if (mask & __GFP_WAIT)
630 static void wait_on_state(struct extent_io_tree *tree,
631 struct extent_state *state)
632 __releases(tree->lock)
633 __acquires(tree->lock)
636 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
637 spin_unlock(&tree->lock);
639 spin_lock(&tree->lock);
640 finish_wait(&state->wq, &wait);
644 * waits for one or more bits to clear on a range in the state tree.
645 * The range [start, end] is inclusive.
646 * The tree lock is taken by this function
648 void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits)
650 struct extent_state *state;
651 struct rb_node *node;
653 spin_lock(&tree->lock);
657 * this search will find all the extents that end after
660 node = tree_search(tree, start);
664 state = rb_entry(node, struct extent_state, rb_node);
666 if (state->start > end)
669 if (state->state & bits) {
670 start = state->start;
671 atomic_inc(&state->refs);
672 wait_on_state(tree, state);
673 free_extent_state(state);
676 start = state->end + 1;
681 cond_resched_lock(&tree->lock);
684 spin_unlock(&tree->lock);
687 static void set_state_bits(struct extent_io_tree *tree,
688 struct extent_state *state,
691 int bits_to_set = *bits & ~EXTENT_CTLBITS;
693 set_state_cb(tree, state, bits);
694 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
695 u64 range = state->end - state->start + 1;
696 tree->dirty_bytes += range;
698 state->state |= bits_to_set;
701 static void cache_state(struct extent_state *state,
702 struct extent_state **cached_ptr)
704 if (cached_ptr && !(*cached_ptr)) {
705 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) {
707 atomic_inc(&state->refs);
712 static void uncache_state(struct extent_state **cached_ptr)
714 if (cached_ptr && (*cached_ptr)) {
715 struct extent_state *state = *cached_ptr;
717 free_extent_state(state);
722 * set some bits on a range in the tree. This may require allocations or
723 * sleeping, so the gfp mask is used to indicate what is allowed.
725 * If any of the exclusive bits are set, this will fail with -EEXIST if some
726 * part of the range already has the desired bits set. The start of the
727 * existing range is returned in failed_start in this case.
729 * [start, end] is inclusive This takes the tree lock.
732 static int __must_check
733 __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
734 int bits, int exclusive_bits, u64 *failed_start,
735 struct extent_state **cached_state, gfp_t mask)
737 struct extent_state *state;
738 struct extent_state *prealloc = NULL;
739 struct rb_node *node;
744 bits |= EXTENT_FIRST_DELALLOC;
746 if (!prealloc && (mask & __GFP_WAIT)) {
747 prealloc = alloc_extent_state(mask);
751 spin_lock(&tree->lock);
752 if (cached_state && *cached_state) {
753 state = *cached_state;
754 if (state->start <= start && state->end > start &&
756 node = &state->rb_node;
761 * this search will find all the extents that end after
764 node = tree_search(tree, start);
766 prealloc = alloc_extent_state_atomic(prealloc);
768 err = insert_state(tree, prealloc, start, end, &bits);
770 extent_io_tree_panic(tree, err);
775 state = rb_entry(node, struct extent_state, rb_node);
777 last_start = state->start;
778 last_end = state->end;
781 * | ---- desired range ---- |
784 * Just lock what we found and keep going
786 if (state->start == start && state->end <= end) {
787 if (state->state & exclusive_bits) {
788 *failed_start = state->start;
793 set_state_bits(tree, state, &bits);
794 cache_state(state, cached_state);
795 merge_state(tree, state);
796 if (last_end == (u64)-1)
798 start = last_end + 1;
799 state = next_state(state);
800 if (start < end && state && state->start == start &&
807 * | ---- desired range ---- |
810 * | ------------- state -------------- |
812 * We need to split the extent we found, and may flip bits on
815 * If the extent we found extends past our
816 * range, we just split and search again. It'll get split
817 * again the next time though.
819 * If the extent we found is inside our range, we set the
822 if (state->start < start) {
823 if (state->state & exclusive_bits) {
824 *failed_start = start;
829 prealloc = alloc_extent_state_atomic(prealloc);
831 err = split_state(tree, state, prealloc, start);
833 extent_io_tree_panic(tree, err);
838 if (state->end <= end) {
839 set_state_bits(tree, state, &bits);
840 cache_state(state, cached_state);
841 merge_state(tree, state);
842 if (last_end == (u64)-1)
844 start = last_end + 1;
845 state = next_state(state);
846 if (start < end && state && state->start == start &&
853 * | ---- desired range ---- |
854 * | state | or | state |
856 * There's a hole, we need to insert something in it and
857 * ignore the extent we found.
859 if (state->start > start) {
861 if (end < last_start)
864 this_end = last_start - 1;
866 prealloc = alloc_extent_state_atomic(prealloc);
870 * Avoid to free 'prealloc' if it can be merged with
873 err = insert_state(tree, prealloc, start, this_end,
876 extent_io_tree_panic(tree, err);
878 cache_state(prealloc, cached_state);
880 start = this_end + 1;
884 * | ---- desired range ---- |
886 * We need to split the extent, and set the bit
889 if (state->start <= end && state->end > end) {
890 if (state->state & exclusive_bits) {
891 *failed_start = start;
896 prealloc = alloc_extent_state_atomic(prealloc);
898 err = split_state(tree, state, prealloc, end + 1);
900 extent_io_tree_panic(tree, err);
902 set_state_bits(tree, prealloc, &bits);
903 cache_state(prealloc, cached_state);
904 merge_state(tree, prealloc);
912 spin_unlock(&tree->lock);
914 free_extent_state(prealloc);
921 spin_unlock(&tree->lock);
922 if (mask & __GFP_WAIT)
927 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits,
928 u64 *failed_start, struct extent_state **cached_state,
931 return __set_extent_bit(tree, start, end, bits, 0, failed_start,
937 * convert_extent_bit - convert all bits in a given range from one bit to
939 * @tree: the io tree to search
940 * @start: the start offset in bytes
941 * @end: the end offset in bytes (inclusive)
942 * @bits: the bits to set in this range
943 * @clear_bits: the bits to clear in this range
944 * @cached_state: state that we're going to cache
945 * @mask: the allocation mask
947 * This will go through and set bits for the given range. If any states exist
948 * already in this range they are set with the given bit and cleared of the
949 * clear_bits. This is only meant to be used by things that are mergeable, ie
950 * converting from say DELALLOC to DIRTY. This is not meant to be used with
951 * boundary bits like LOCK.
953 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
954 int bits, int clear_bits,
955 struct extent_state **cached_state, gfp_t mask)
957 struct extent_state *state;
958 struct extent_state *prealloc = NULL;
959 struct rb_node *node;
965 if (!prealloc && (mask & __GFP_WAIT)) {
966 prealloc = alloc_extent_state(mask);
971 spin_lock(&tree->lock);
972 if (cached_state && *cached_state) {
973 state = *cached_state;
974 if (state->start <= start && state->end > start &&
976 node = &state->rb_node;
982 * this search will find all the extents that end after
985 node = tree_search(tree, start);
987 prealloc = alloc_extent_state_atomic(prealloc);
992 err = insert_state(tree, prealloc, start, end, &bits);
995 extent_io_tree_panic(tree, err);
998 state = rb_entry(node, struct extent_state, rb_node);
1000 last_start = state->start;
1001 last_end = state->end;
1004 * | ---- desired range ---- |
1007 * Just lock what we found and keep going
1009 if (state->start == start && state->end <= end) {
1010 set_state_bits(tree, state, &bits);
1011 cache_state(state, cached_state);
1012 state = clear_state_bit(tree, state, &clear_bits, 0);
1013 if (last_end == (u64)-1)
1015 start = last_end + 1;
1016 if (start < end && state && state->start == start &&
1023 * | ---- desired range ---- |
1026 * | ------------- state -------------- |
1028 * We need to split the extent we found, and may flip bits on
1031 * If the extent we found extends past our
1032 * range, we just split and search again. It'll get split
1033 * again the next time though.
1035 * If the extent we found is inside our range, we set the
1036 * desired bit on it.
1038 if (state->start < start) {
1039 prealloc = alloc_extent_state_atomic(prealloc);
1044 err = split_state(tree, state, prealloc, start);
1046 extent_io_tree_panic(tree, err);
1050 if (state->end <= end) {
1051 set_state_bits(tree, state, &bits);
1052 cache_state(state, cached_state);
1053 state = clear_state_bit(tree, state, &clear_bits, 0);
1054 if (last_end == (u64)-1)
1056 start = last_end + 1;
1057 if (start < end && state && state->start == start &&
1064 * | ---- desired range ---- |
1065 * | state | or | state |
1067 * There's a hole, we need to insert something in it and
1068 * ignore the extent we found.
1070 if (state->start > start) {
1072 if (end < last_start)
1075 this_end = last_start - 1;
1077 prealloc = alloc_extent_state_atomic(prealloc);
1084 * Avoid to free 'prealloc' if it can be merged with
1087 err = insert_state(tree, prealloc, start, this_end,
1090 extent_io_tree_panic(tree, err);
1091 cache_state(prealloc, cached_state);
1093 start = this_end + 1;
1097 * | ---- desired range ---- |
1099 * We need to split the extent, and set the bit
1102 if (state->start <= end && state->end > end) {
1103 prealloc = alloc_extent_state_atomic(prealloc);
1109 err = split_state(tree, state, prealloc, end + 1);
1111 extent_io_tree_panic(tree, err);
1113 set_state_bits(tree, prealloc, &bits);
1114 cache_state(prealloc, cached_state);
1115 clear_state_bit(tree, prealloc, &clear_bits, 0);
1123 spin_unlock(&tree->lock);
1125 free_extent_state(prealloc);
1132 spin_unlock(&tree->lock);
1133 if (mask & __GFP_WAIT)
1138 /* wrappers around set/clear extent bit */
1139 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1142 return set_extent_bit(tree, start, end, EXTENT_DIRTY, NULL,
1146 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1147 int bits, gfp_t mask)
1149 return set_extent_bit(tree, start, end, bits, NULL,
1153 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1154 int bits, gfp_t mask)
1156 return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask);
1159 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
1160 struct extent_state **cached_state, gfp_t mask)
1162 return set_extent_bit(tree, start, end,
1163 EXTENT_DELALLOC | EXTENT_UPTODATE,
1164 NULL, cached_state, mask);
1167 int set_extent_defrag(struct extent_io_tree *tree, u64 start, u64 end,
1168 struct extent_state **cached_state, gfp_t mask)
1170 return set_extent_bit(tree, start, end,
1171 EXTENT_DELALLOC | EXTENT_UPTODATE | EXTENT_DEFRAG,
1172 NULL, cached_state, mask);
1175 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1178 return clear_extent_bit(tree, start, end,
1179 EXTENT_DIRTY | EXTENT_DELALLOC |
1180 EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask);
1183 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
1186 return set_extent_bit(tree, start, end, EXTENT_NEW, NULL,
1190 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
1191 struct extent_state **cached_state, gfp_t mask)
1193 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0,
1194 cached_state, mask);
1197 int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
1198 struct extent_state **cached_state, gfp_t mask)
1200 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0,
1201 cached_state, mask);
1205 * either insert or lock state struct between start and end use mask to tell
1206 * us if waiting is desired.
1208 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1209 int bits, struct extent_state **cached_state)
1214 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED | bits,
1215 EXTENT_LOCKED, &failed_start,
1216 cached_state, GFP_NOFS);
1217 if (err == -EEXIST) {
1218 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1219 start = failed_start;
1222 WARN_ON(start > end);
1227 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1229 return lock_extent_bits(tree, start, end, 0, NULL);
1232 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1237 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1238 &failed_start, NULL, GFP_NOFS);
1239 if (err == -EEXIST) {
1240 if (failed_start > start)
1241 clear_extent_bit(tree, start, failed_start - 1,
1242 EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS);
1248 int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end,
1249 struct extent_state **cached, gfp_t mask)
1251 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached,
1255 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1257 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL,
1262 * helper function to set both pages and extents in the tree writeback
1264 static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
1266 unsigned long index = start >> PAGE_CACHE_SHIFT;
1267 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1270 while (index <= end_index) {
1271 page = find_get_page(tree->mapping, index);
1272 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1273 set_page_writeback(page);
1274 page_cache_release(page);
1280 /* find the first state struct with 'bits' set after 'start', and
1281 * return it. tree->lock must be held. NULL will returned if
1282 * nothing was found after 'start'
1284 struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree,
1285 u64 start, int bits)
1287 struct rb_node *node;
1288 struct extent_state *state;
1291 * this search will find all the extents that end after
1294 node = tree_search(tree, start);
1299 state = rb_entry(node, struct extent_state, rb_node);
1300 if (state->end >= start && (state->state & bits))
1303 node = rb_next(node);
1312 * find the first offset in the io tree with 'bits' set. zero is
1313 * returned if we find something, and *start_ret and *end_ret are
1314 * set to reflect the state struct that was found.
1316 * If nothing was found, 1 is returned. If found something, return 0.
1318 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1319 u64 *start_ret, u64 *end_ret, int bits,
1320 struct extent_state **cached_state)
1322 struct extent_state *state;
1326 spin_lock(&tree->lock);
1327 if (cached_state && *cached_state) {
1328 state = *cached_state;
1329 if (state->end == start - 1 && state->tree) {
1330 n = rb_next(&state->rb_node);
1332 state = rb_entry(n, struct extent_state,
1334 if (state->state & bits)
1338 free_extent_state(*cached_state);
1339 *cached_state = NULL;
1342 free_extent_state(*cached_state);
1343 *cached_state = NULL;
1346 state = find_first_extent_bit_state(tree, start, bits);
1349 cache_state(state, cached_state);
1350 *start_ret = state->start;
1351 *end_ret = state->end;
1355 spin_unlock(&tree->lock);
1360 * find a contiguous range of bytes in the file marked as delalloc, not
1361 * more than 'max_bytes'. start and end are used to return the range,
1363 * 1 is returned if we find something, 0 if nothing was in the tree
1365 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1366 u64 *start, u64 *end, u64 max_bytes,
1367 struct extent_state **cached_state)
1369 struct rb_node *node;
1370 struct extent_state *state;
1371 u64 cur_start = *start;
1373 u64 total_bytes = 0;
1375 spin_lock(&tree->lock);
1378 * this search will find all the extents that end after
1381 node = tree_search(tree, cur_start);
1389 state = rb_entry(node, struct extent_state, rb_node);
1390 if (found && (state->start != cur_start ||
1391 (state->state & EXTENT_BOUNDARY))) {
1394 if (!(state->state & EXTENT_DELALLOC)) {
1400 *start = state->start;
1401 *cached_state = state;
1402 atomic_inc(&state->refs);
1406 cur_start = state->end + 1;
1407 node = rb_next(node);
1410 total_bytes += state->end - state->start + 1;
1411 if (total_bytes >= max_bytes)
1415 spin_unlock(&tree->lock);
1419 static noinline void __unlock_for_delalloc(struct inode *inode,
1420 struct page *locked_page,
1424 struct page *pages[16];
1425 unsigned long index = start >> PAGE_CACHE_SHIFT;
1426 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1427 unsigned long nr_pages = end_index - index + 1;
1430 if (index == locked_page->index && end_index == index)
1433 while (nr_pages > 0) {
1434 ret = find_get_pages_contig(inode->i_mapping, index,
1435 min_t(unsigned long, nr_pages,
1436 ARRAY_SIZE(pages)), pages);
1437 for (i = 0; i < ret; i++) {
1438 if (pages[i] != locked_page)
1439 unlock_page(pages[i]);
1440 page_cache_release(pages[i]);
1448 static noinline int lock_delalloc_pages(struct inode *inode,
1449 struct page *locked_page,
1453 unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT;
1454 unsigned long start_index = index;
1455 unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT;
1456 unsigned long pages_locked = 0;
1457 struct page *pages[16];
1458 unsigned long nrpages;
1462 /* the caller is responsible for locking the start index */
1463 if (index == locked_page->index && index == end_index)
1466 /* skip the page at the start index */
1467 nrpages = end_index - index + 1;
1468 while (nrpages > 0) {
1469 ret = find_get_pages_contig(inode->i_mapping, index,
1470 min_t(unsigned long,
1471 nrpages, ARRAY_SIZE(pages)), pages);
1476 /* now we have an array of pages, lock them all */
1477 for (i = 0; i < ret; i++) {
1479 * the caller is taking responsibility for
1482 if (pages[i] != locked_page) {
1483 lock_page(pages[i]);
1484 if (!PageDirty(pages[i]) ||
1485 pages[i]->mapping != inode->i_mapping) {
1487 unlock_page(pages[i]);
1488 page_cache_release(pages[i]);
1492 page_cache_release(pages[i]);
1501 if (ret && pages_locked) {
1502 __unlock_for_delalloc(inode, locked_page,
1504 ((u64)(start_index + pages_locked - 1)) <<
1511 * find a contiguous range of bytes in the file marked as delalloc, not
1512 * more than 'max_bytes'. start and end are used to return the range,
1514 * 1 is returned if we find something, 0 if nothing was in the tree
1516 static noinline u64 find_lock_delalloc_range(struct inode *inode,
1517 struct extent_io_tree *tree,
1518 struct page *locked_page,
1519 u64 *start, u64 *end,
1525 struct extent_state *cached_state = NULL;
1530 /* step one, find a bunch of delalloc bytes starting at start */
1531 delalloc_start = *start;
1533 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1534 max_bytes, &cached_state);
1535 if (!found || delalloc_end <= *start) {
1536 *start = delalloc_start;
1537 *end = delalloc_end;
1538 free_extent_state(cached_state);
1543 * start comes from the offset of locked_page. We have to lock
1544 * pages in order, so we can't process delalloc bytes before
1547 if (delalloc_start < *start)
1548 delalloc_start = *start;
1551 * make sure to limit the number of pages we try to lock down
1554 if (delalloc_end + 1 - delalloc_start > max_bytes && loops)
1555 delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1;
1557 /* step two, lock all the pages after the page that has start */
1558 ret = lock_delalloc_pages(inode, locked_page,
1559 delalloc_start, delalloc_end);
1560 if (ret == -EAGAIN) {
1561 /* some of the pages are gone, lets avoid looping by
1562 * shortening the size of the delalloc range we're searching
1564 free_extent_state(cached_state);
1566 unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1);
1567 max_bytes = PAGE_CACHE_SIZE - offset;
1575 BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */
1577 /* step three, lock the state bits for the whole range */
1578 lock_extent_bits(tree, delalloc_start, delalloc_end, 0, &cached_state);
1580 /* then test to make sure it is all still delalloc */
1581 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1582 EXTENT_DELALLOC, 1, cached_state);
1584 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1585 &cached_state, GFP_NOFS);
1586 __unlock_for_delalloc(inode, locked_page,
1587 delalloc_start, delalloc_end);
1591 free_extent_state(cached_state);
1592 *start = delalloc_start;
1593 *end = delalloc_end;
1598 int extent_clear_unlock_delalloc(struct inode *inode,
1599 struct extent_io_tree *tree,
1600 u64 start, u64 end, struct page *locked_page,
1604 struct page *pages[16];
1605 unsigned long index = start >> PAGE_CACHE_SHIFT;
1606 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1607 unsigned long nr_pages = end_index - index + 1;
1611 if (op & EXTENT_CLEAR_UNLOCK)
1612 clear_bits |= EXTENT_LOCKED;
1613 if (op & EXTENT_CLEAR_DIRTY)
1614 clear_bits |= EXTENT_DIRTY;
1616 if (op & EXTENT_CLEAR_DELALLOC)
1617 clear_bits |= EXTENT_DELALLOC;
1619 clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS);
1620 if (!(op & (EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
1621 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK |
1622 EXTENT_SET_PRIVATE2)))
1625 while (nr_pages > 0) {
1626 ret = find_get_pages_contig(inode->i_mapping, index,
1627 min_t(unsigned long,
1628 nr_pages, ARRAY_SIZE(pages)), pages);
1629 for (i = 0; i < ret; i++) {
1631 if (op & EXTENT_SET_PRIVATE2)
1632 SetPagePrivate2(pages[i]);
1634 if (pages[i] == locked_page) {
1635 page_cache_release(pages[i]);
1638 if (op & EXTENT_CLEAR_DIRTY)
1639 clear_page_dirty_for_io(pages[i]);
1640 if (op & EXTENT_SET_WRITEBACK)
1641 set_page_writeback(pages[i]);
1642 if (op & EXTENT_END_WRITEBACK)
1643 end_page_writeback(pages[i]);
1644 if (op & EXTENT_CLEAR_UNLOCK_PAGE)
1645 unlock_page(pages[i]);
1646 page_cache_release(pages[i]);
1656 * count the number of bytes in the tree that have a given bit(s)
1657 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1658 * cached. The total number found is returned.
1660 u64 count_range_bits(struct extent_io_tree *tree,
1661 u64 *start, u64 search_end, u64 max_bytes,
1662 unsigned long bits, int contig)
1664 struct rb_node *node;
1665 struct extent_state *state;
1666 u64 cur_start = *start;
1667 u64 total_bytes = 0;
1671 if (search_end <= cur_start) {
1676 spin_lock(&tree->lock);
1677 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1678 total_bytes = tree->dirty_bytes;
1682 * this search will find all the extents that end after
1685 node = tree_search(tree, cur_start);
1690 state = rb_entry(node, struct extent_state, rb_node);
1691 if (state->start > search_end)
1693 if (contig && found && state->start > last + 1)
1695 if (state->end >= cur_start && (state->state & bits) == bits) {
1696 total_bytes += min(search_end, state->end) + 1 -
1697 max(cur_start, state->start);
1698 if (total_bytes >= max_bytes)
1701 *start = max(cur_start, state->start);
1705 } else if (contig && found) {
1708 node = rb_next(node);
1713 spin_unlock(&tree->lock);
1718 * set the private field for a given byte offset in the tree. If there isn't
1719 * an extent_state there already, this does nothing.
1721 int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1723 struct rb_node *node;
1724 struct extent_state *state;
1727 spin_lock(&tree->lock);
1729 * this search will find all the extents that end after
1732 node = tree_search(tree, start);
1737 state = rb_entry(node, struct extent_state, rb_node);
1738 if (state->start != start) {
1742 state->private = private;
1744 spin_unlock(&tree->lock);
1748 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1750 struct rb_node *node;
1751 struct extent_state *state;
1754 spin_lock(&tree->lock);
1756 * this search will find all the extents that end after
1759 node = tree_search(tree, start);
1764 state = rb_entry(node, struct extent_state, rb_node);
1765 if (state->start != start) {
1769 *private = state->private;
1771 spin_unlock(&tree->lock);
1776 * searches a range in the state tree for a given mask.
1777 * If 'filled' == 1, this returns 1 only if every extent in the tree
1778 * has the bits set. Otherwise, 1 is returned if any bit in the
1779 * range is found set.
1781 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1782 int bits, int filled, struct extent_state *cached)
1784 struct extent_state *state = NULL;
1785 struct rb_node *node;
1788 spin_lock(&tree->lock);
1789 if (cached && cached->tree && cached->start <= start &&
1790 cached->end > start)
1791 node = &cached->rb_node;
1793 node = tree_search(tree, start);
1794 while (node && start <= end) {
1795 state = rb_entry(node, struct extent_state, rb_node);
1797 if (filled && state->start > start) {
1802 if (state->start > end)
1805 if (state->state & bits) {
1809 } else if (filled) {
1814 if (state->end == (u64)-1)
1817 start = state->end + 1;
1820 node = rb_next(node);
1827 spin_unlock(&tree->lock);
1832 * helper function to set a given page up to date if all the
1833 * extents in the tree for that page are up to date
1835 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
1837 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1838 u64 end = start + PAGE_CACHE_SIZE - 1;
1839 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1840 SetPageUptodate(page);
1844 * helper function to unlock a page if all the extents in the tree
1845 * for that page are unlocked
1847 static void check_page_locked(struct extent_io_tree *tree, struct page *page)
1849 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1850 u64 end = start + PAGE_CACHE_SIZE - 1;
1851 if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL))
1856 * helper function to end page writeback if all the extents
1857 * in the tree for that page are done with writeback
1859 static void check_page_writeback(struct extent_io_tree *tree,
1862 end_page_writeback(page);
1866 * When IO fails, either with EIO or csum verification fails, we
1867 * try other mirrors that might have a good copy of the data. This
1868 * io_failure_record is used to record state as we go through all the
1869 * mirrors. If another mirror has good data, the page is set up to date
1870 * and things continue. If a good mirror can't be found, the original
1871 * bio end_io callback is called to indicate things have failed.
1873 struct io_failure_record {
1878 unsigned long bio_flags;
1884 static int free_io_failure(struct inode *inode, struct io_failure_record *rec,
1889 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1891 set_state_private(failure_tree, rec->start, 0);
1892 ret = clear_extent_bits(failure_tree, rec->start,
1893 rec->start + rec->len - 1,
1894 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1898 ret = clear_extent_bits(&BTRFS_I(inode)->io_tree, rec->start,
1899 rec->start + rec->len - 1,
1900 EXTENT_DAMAGED, GFP_NOFS);
1908 static void repair_io_failure_callback(struct bio *bio, int err)
1910 complete(bio->bi_private);
1914 * this bypasses the standard btrfs submit functions deliberately, as
1915 * the standard behavior is to write all copies in a raid setup. here we only
1916 * want to write the one bad copy. so we do the mapping for ourselves and issue
1917 * submit_bio directly.
1918 * to avoid any synchronization issues, wait for the data after writing, which
1919 * actually prevents the read that triggered the error from finishing.
1920 * currently, there can be no more than two copies of every data bit. thus,
1921 * exactly one rewrite is required.
1923 int repair_io_failure(struct btrfs_fs_info *fs_info, u64 start,
1924 u64 length, u64 logical, struct page *page,
1928 struct btrfs_device *dev;
1929 DECLARE_COMPLETION_ONSTACK(compl);
1932 struct btrfs_bio *bbio = NULL;
1933 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1936 BUG_ON(!mirror_num);
1938 /* we can't repair anything in raid56 yet */
1939 if (btrfs_is_parity_mirror(map_tree, logical, length, mirror_num))
1942 bio = bio_alloc(GFP_NOFS, 1);
1945 bio->bi_private = &compl;
1946 bio->bi_end_io = repair_io_failure_callback;
1948 map_length = length;
1950 ret = btrfs_map_block(fs_info, WRITE, logical,
1951 &map_length, &bbio, mirror_num);
1956 BUG_ON(mirror_num != bbio->mirror_num);
1957 sector = bbio->stripes[mirror_num-1].physical >> 9;
1958 bio->bi_sector = sector;
1959 dev = bbio->stripes[mirror_num-1].dev;
1961 if (!dev || !dev->bdev || !dev->writeable) {
1965 bio->bi_bdev = dev->bdev;
1966 bio_add_page(bio, page, length, start-page_offset(page));
1967 btrfsic_submit_bio(WRITE_SYNC, bio);
1968 wait_for_completion(&compl);
1970 if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) {
1971 /* try to remap that extent elsewhere? */
1973 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
1977 printk_ratelimited_in_rcu(KERN_INFO "btrfs read error corrected: ino %lu off %llu "
1978 "(dev %s sector %llu)\n", page->mapping->host->i_ino,
1979 start, rcu_str_deref(dev->name), sector);
1985 int repair_eb_io_failure(struct btrfs_root *root, struct extent_buffer *eb,
1988 u64 start = eb->start;
1989 unsigned long i, num_pages = num_extent_pages(eb->start, eb->len);
1992 for (i = 0; i < num_pages; i++) {
1993 struct page *p = extent_buffer_page(eb, i);
1994 ret = repair_io_failure(root->fs_info, start, PAGE_CACHE_SIZE,
1995 start, p, mirror_num);
1998 start += PAGE_CACHE_SIZE;
2005 * each time an IO finishes, we do a fast check in the IO failure tree
2006 * to see if we need to process or clean up an io_failure_record
2008 static int clean_io_failure(u64 start, struct page *page)
2011 u64 private_failure;
2012 struct io_failure_record *failrec;
2013 struct btrfs_fs_info *fs_info;
2014 struct extent_state *state;
2018 struct inode *inode = page->mapping->host;
2021 ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
2022 (u64)-1, 1, EXTENT_DIRTY, 0);
2026 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, start,
2031 failrec = (struct io_failure_record *)(unsigned long) private_failure;
2032 BUG_ON(!failrec->this_mirror);
2034 if (failrec->in_validation) {
2035 /* there was no real error, just free the record */
2036 pr_debug("clean_io_failure: freeing dummy error at %llu\n",
2042 spin_lock(&BTRFS_I(inode)->io_tree.lock);
2043 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
2046 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
2048 if (state && state->start == failrec->start) {
2049 fs_info = BTRFS_I(inode)->root->fs_info;
2050 num_copies = btrfs_num_copies(fs_info, failrec->logical,
2052 if (num_copies > 1) {
2053 ret = repair_io_failure(fs_info, start, failrec->len,
2054 failrec->logical, page,
2055 failrec->failed_mirror);
2063 ret = free_io_failure(inode, failrec, did_repair);
2069 * this is a generic handler for readpage errors (default
2070 * readpage_io_failed_hook). if other copies exist, read those and write back
2071 * good data to the failed position. does not investigate in remapping the
2072 * failed extent elsewhere, hoping the device will be smart enough to do this as
2076 static int bio_readpage_error(struct bio *failed_bio, struct page *page,
2077 u64 start, u64 end, int failed_mirror,
2078 struct extent_state *state)
2080 struct io_failure_record *failrec = NULL;
2082 struct extent_map *em;
2083 struct inode *inode = page->mapping->host;
2084 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2085 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2086 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2093 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
2095 ret = get_state_private(failure_tree, start, &private);
2097 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2100 failrec->start = start;
2101 failrec->len = end - start + 1;
2102 failrec->this_mirror = 0;
2103 failrec->bio_flags = 0;
2104 failrec->in_validation = 0;
2106 read_lock(&em_tree->lock);
2107 em = lookup_extent_mapping(em_tree, start, failrec->len);
2109 read_unlock(&em_tree->lock);
2114 if (em->start > start || em->start + em->len < start) {
2115 free_extent_map(em);
2118 read_unlock(&em_tree->lock);
2124 logical = start - em->start;
2125 logical = em->block_start + logical;
2126 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2127 logical = em->block_start;
2128 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2129 extent_set_compress_type(&failrec->bio_flags,
2132 pr_debug("bio_readpage_error: (new) logical=%llu, start=%llu, "
2133 "len=%llu\n", logical, start, failrec->len);
2134 failrec->logical = logical;
2135 free_extent_map(em);
2137 /* set the bits in the private failure tree */
2138 ret = set_extent_bits(failure_tree, start, end,
2139 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
2141 ret = set_state_private(failure_tree, start,
2142 (u64)(unsigned long)failrec);
2143 /* set the bits in the inode's tree */
2145 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED,
2152 failrec = (struct io_failure_record *)(unsigned long)private;
2153 pr_debug("bio_readpage_error: (found) logical=%llu, "
2154 "start=%llu, len=%llu, validation=%d\n",
2155 failrec->logical, failrec->start, failrec->len,
2156 failrec->in_validation);
2158 * when data can be on disk more than twice, add to failrec here
2159 * (e.g. with a list for failed_mirror) to make
2160 * clean_io_failure() clean all those errors at once.
2163 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
2164 failrec->logical, failrec->len);
2165 if (num_copies == 1) {
2167 * we only have a single copy of the data, so don't bother with
2168 * all the retry and error correction code that follows. no
2169 * matter what the error is, it is very likely to persist.
2171 pr_debug("bio_readpage_error: cannot repair, num_copies == 1. "
2172 "state=%p, num_copies=%d, next_mirror %d, "
2173 "failed_mirror %d\n", state, num_copies,
2174 failrec->this_mirror, failed_mirror);
2175 free_io_failure(inode, failrec, 0);
2180 spin_lock(&tree->lock);
2181 state = find_first_extent_bit_state(tree, failrec->start,
2183 if (state && state->start != failrec->start)
2185 spin_unlock(&tree->lock);
2189 * there are two premises:
2190 * a) deliver good data to the caller
2191 * b) correct the bad sectors on disk
2193 if (failed_bio->bi_vcnt > 1) {
2195 * to fulfill b), we need to know the exact failing sectors, as
2196 * we don't want to rewrite any more than the failed ones. thus,
2197 * we need separate read requests for the failed bio
2199 * if the following BUG_ON triggers, our validation request got
2200 * merged. we need separate requests for our algorithm to work.
2202 BUG_ON(failrec->in_validation);
2203 failrec->in_validation = 1;
2204 failrec->this_mirror = failed_mirror;
2205 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
2208 * we're ready to fulfill a) and b) alongside. get a good copy
2209 * of the failed sector and if we succeed, we have setup
2210 * everything for repair_io_failure to do the rest for us.
2212 if (failrec->in_validation) {
2213 BUG_ON(failrec->this_mirror != failed_mirror);
2214 failrec->in_validation = 0;
2215 failrec->this_mirror = 0;
2217 failrec->failed_mirror = failed_mirror;
2218 failrec->this_mirror++;
2219 if (failrec->this_mirror == failed_mirror)
2220 failrec->this_mirror++;
2221 read_mode = READ_SYNC;
2224 if (!state || failrec->this_mirror > num_copies) {
2225 pr_debug("bio_readpage_error: (fail) state=%p, num_copies=%d, "
2226 "next_mirror %d, failed_mirror %d\n", state,
2227 num_copies, failrec->this_mirror, failed_mirror);
2228 free_io_failure(inode, failrec, 0);
2232 bio = bio_alloc(GFP_NOFS, 1);
2234 free_io_failure(inode, failrec, 0);
2237 bio->bi_private = state;
2238 bio->bi_end_io = failed_bio->bi_end_io;
2239 bio->bi_sector = failrec->logical >> 9;
2240 bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
2243 bio_add_page(bio, page, failrec->len, start - page_offset(page));
2245 pr_debug("bio_readpage_error: submitting new read[%#x] to "
2246 "this_mirror=%d, num_copies=%d, in_validation=%d\n", read_mode,
2247 failrec->this_mirror, num_copies, failrec->in_validation);
2249 ret = tree->ops->submit_bio_hook(inode, read_mode, bio,
2250 failrec->this_mirror,
2251 failrec->bio_flags, 0);
2255 /* lots and lots of room for performance fixes in the end_bio funcs */
2257 int end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2259 int uptodate = (err == 0);
2260 struct extent_io_tree *tree;
2263 tree = &BTRFS_I(page->mapping->host)->io_tree;
2265 if (tree->ops && tree->ops->writepage_end_io_hook) {
2266 ret = tree->ops->writepage_end_io_hook(page, start,
2267 end, NULL, uptodate);
2273 ClearPageUptodate(page);
2280 * after a writepage IO is done, we need to:
2281 * clear the uptodate bits on error
2282 * clear the writeback bits in the extent tree for this IO
2283 * end_page_writeback if the page has no more pending IO
2285 * Scheduling is not allowed, so the extent state tree is expected
2286 * to have one and only one object corresponding to this IO.
2288 static void end_bio_extent_writepage(struct bio *bio, int err)
2290 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2291 struct extent_io_tree *tree;
2297 struct page *page = bvec->bv_page;
2298 tree = &BTRFS_I(page->mapping->host)->io_tree;
2300 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
2302 end = start + bvec->bv_len - 1;
2304 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
2309 if (--bvec >= bio->bi_io_vec)
2310 prefetchw(&bvec->bv_page->flags);
2312 if (end_extent_writepage(page, err, start, end))
2316 end_page_writeback(page);
2318 check_page_writeback(tree, page);
2319 } while (bvec >= bio->bi_io_vec);
2325 * after a readpage IO is done, we need to:
2326 * clear the uptodate bits on error
2327 * set the uptodate bits if things worked
2328 * set the page up to date if all extents in the tree are uptodate
2329 * clear the lock bit in the extent tree
2330 * unlock the page if there are no other extents locked for it
2332 * Scheduling is not allowed, so the extent state tree is expected
2333 * to have one and only one object corresponding to this IO.
2335 static void end_bio_extent_readpage(struct bio *bio, int err)
2337 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
2338 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
2339 struct bio_vec *bvec = bio->bi_io_vec;
2340 struct extent_io_tree *tree;
2351 struct page *page = bvec->bv_page;
2352 struct extent_state *cached = NULL;
2353 struct extent_state *state;
2355 pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, "
2356 "mirror=%ld\n", (u64)bio->bi_sector, err,
2357 (long int)bio->bi_bdev);
2358 tree = &BTRFS_I(page->mapping->host)->io_tree;
2360 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
2362 end = start + bvec->bv_len - 1;
2364 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
2369 if (++bvec <= bvec_end)
2370 prefetchw(&bvec->bv_page->flags);
2372 spin_lock(&tree->lock);
2373 state = find_first_extent_bit_state(tree, start, EXTENT_LOCKED);
2374 if (state && state->start == start) {
2376 * take a reference on the state, unlock will drop
2379 cache_state(state, &cached);
2381 spin_unlock(&tree->lock);
2383 mirror = (int)(unsigned long)bio->bi_bdev;
2384 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
2385 ret = tree->ops->readpage_end_io_hook(page, start, end,
2390 clean_io_failure(start, page);
2393 if (!uptodate && tree->ops && tree->ops->readpage_io_failed_hook) {
2394 ret = tree->ops->readpage_io_failed_hook(page, mirror);
2396 test_bit(BIO_UPTODATE, &bio->bi_flags))
2398 } else if (!uptodate) {
2400 * The generic bio_readpage_error handles errors the
2401 * following way: If possible, new read requests are
2402 * created and submitted and will end up in
2403 * end_bio_extent_readpage as well (if we're lucky, not
2404 * in the !uptodate case). In that case it returns 0 and
2405 * we just go on with the next page in our bio. If it
2406 * can't handle the error it will return -EIO and we
2407 * remain responsible for that page.
2409 ret = bio_readpage_error(bio, page, start, end, mirror, NULL);
2412 test_bit(BIO_UPTODATE, &bio->bi_flags);
2415 uncache_state(&cached);
2420 if (uptodate && tree->track_uptodate) {
2421 set_extent_uptodate(tree, start, end, &cached,
2424 unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC);
2428 SetPageUptodate(page);
2430 ClearPageUptodate(page);
2436 check_page_uptodate(tree, page);
2438 ClearPageUptodate(page);
2441 check_page_locked(tree, page);
2443 } while (bvec <= bvec_end);
2449 btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
2454 bio = bio_alloc(gfp_flags, nr_vecs);
2456 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
2457 while (!bio && (nr_vecs /= 2))
2458 bio = bio_alloc(gfp_flags, nr_vecs);
2463 bio->bi_bdev = bdev;
2464 bio->bi_sector = first_sector;
2469 static int __must_check submit_one_bio(int rw, struct bio *bio,
2470 int mirror_num, unsigned long bio_flags)
2473 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2474 struct page *page = bvec->bv_page;
2475 struct extent_io_tree *tree = bio->bi_private;
2478 start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
2480 bio->bi_private = NULL;
2484 if (tree->ops && tree->ops->submit_bio_hook)
2485 ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
2486 mirror_num, bio_flags, start);
2488 btrfsic_submit_bio(rw, bio);
2490 if (bio_flagged(bio, BIO_EOPNOTSUPP))
2496 static int merge_bio(int rw, struct extent_io_tree *tree, struct page *page,
2497 unsigned long offset, size_t size, struct bio *bio,
2498 unsigned long bio_flags)
2501 if (tree->ops && tree->ops->merge_bio_hook)
2502 ret = tree->ops->merge_bio_hook(rw, page, offset, size, bio,
2509 static int submit_extent_page(int rw, struct extent_io_tree *tree,
2510 struct page *page, sector_t sector,
2511 size_t size, unsigned long offset,
2512 struct block_device *bdev,
2513 struct bio **bio_ret,
2514 unsigned long max_pages,
2515 bio_end_io_t end_io_func,
2517 unsigned long prev_bio_flags,
2518 unsigned long bio_flags)
2524 int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED;
2525 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
2526 size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE);
2528 if (bio_ret && *bio_ret) {
2531 contig = bio->bi_sector == sector;
2533 contig = bio->bi_sector + (bio->bi_size >> 9) ==
2536 if (prev_bio_flags != bio_flags || !contig ||
2537 merge_bio(rw, tree, page, offset, page_size, bio, bio_flags) ||
2538 bio_add_page(bio, page, page_size, offset) < page_size) {
2539 ret = submit_one_bio(rw, bio, mirror_num,
2548 if (this_compressed)
2551 nr = bio_get_nr_vecs(bdev);
2553 bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
2557 bio_add_page(bio, page, page_size, offset);
2558 bio->bi_end_io = end_io_func;
2559 bio->bi_private = tree;
2564 ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
2569 void attach_extent_buffer_page(struct extent_buffer *eb, struct page *page)
2571 if (!PagePrivate(page)) {
2572 SetPagePrivate(page);
2573 page_cache_get(page);
2574 set_page_private(page, (unsigned long)eb);
2576 WARN_ON(page->private != (unsigned long)eb);
2580 void set_page_extent_mapped(struct page *page)
2582 if (!PagePrivate(page)) {
2583 SetPagePrivate(page);
2584 page_cache_get(page);
2585 set_page_private(page, EXTENT_PAGE_PRIVATE);
2590 * basic readpage implementation. Locked extent state structs are inserted
2591 * into the tree that are removed when the IO is done (by the end_io
2593 * XXX JDM: This needs looking at to ensure proper page locking
2595 static int __extent_read_full_page(struct extent_io_tree *tree,
2597 get_extent_t *get_extent,
2598 struct bio **bio, int mirror_num,
2599 unsigned long *bio_flags)
2601 struct inode *inode = page->mapping->host;
2602 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2603 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2607 u64 last_byte = i_size_read(inode);
2611 struct extent_map *em;
2612 struct block_device *bdev;
2613 struct btrfs_ordered_extent *ordered;
2616 size_t pg_offset = 0;
2618 size_t disk_io_size;
2619 size_t blocksize = inode->i_sb->s_blocksize;
2620 unsigned long this_bio_flag = 0;
2622 set_page_extent_mapped(page);
2624 if (!PageUptodate(page)) {
2625 if (cleancache_get_page(page) == 0) {
2626 BUG_ON(blocksize != PAGE_SIZE);
2633 lock_extent(tree, start, end);
2634 ordered = btrfs_lookup_ordered_extent(inode, start);
2637 unlock_extent(tree, start, end);
2638 btrfs_start_ordered_extent(inode, ordered, 1);
2639 btrfs_put_ordered_extent(ordered);
2642 if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
2644 size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
2647 iosize = PAGE_CACHE_SIZE - zero_offset;
2648 userpage = kmap_atomic(page);
2649 memset(userpage + zero_offset, 0, iosize);
2650 flush_dcache_page(page);
2651 kunmap_atomic(userpage);
2654 while (cur <= end) {
2655 if (cur >= last_byte) {
2657 struct extent_state *cached = NULL;
2659 iosize = PAGE_CACHE_SIZE - pg_offset;
2660 userpage = kmap_atomic(page);
2661 memset(userpage + pg_offset, 0, iosize);
2662 flush_dcache_page(page);
2663 kunmap_atomic(userpage);
2664 set_extent_uptodate(tree, cur, cur + iosize - 1,
2666 unlock_extent_cached(tree, cur, cur + iosize - 1,
2670 em = get_extent(inode, page, pg_offset, cur,
2672 if (IS_ERR_OR_NULL(em)) {
2674 unlock_extent(tree, cur, end);
2677 extent_offset = cur - em->start;
2678 BUG_ON(extent_map_end(em) <= cur);
2681 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2682 this_bio_flag = EXTENT_BIO_COMPRESSED;
2683 extent_set_compress_type(&this_bio_flag,
2687 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2688 cur_end = min(extent_map_end(em) - 1, end);
2689 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2690 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2691 disk_io_size = em->block_len;
2692 sector = em->block_start >> 9;
2694 sector = (em->block_start + extent_offset) >> 9;
2695 disk_io_size = iosize;
2698 block_start = em->block_start;
2699 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2700 block_start = EXTENT_MAP_HOLE;
2701 free_extent_map(em);
2704 /* we've found a hole, just zero and go on */
2705 if (block_start == EXTENT_MAP_HOLE) {
2707 struct extent_state *cached = NULL;
2709 userpage = kmap_atomic(page);
2710 memset(userpage + pg_offset, 0, iosize);
2711 flush_dcache_page(page);
2712 kunmap_atomic(userpage);
2714 set_extent_uptodate(tree, cur, cur + iosize - 1,
2716 unlock_extent_cached(tree, cur, cur + iosize - 1,
2719 pg_offset += iosize;
2722 /* the get_extent function already copied into the page */
2723 if (test_range_bit(tree, cur, cur_end,
2724 EXTENT_UPTODATE, 1, NULL)) {
2725 check_page_uptodate(tree, page);
2726 unlock_extent(tree, cur, cur + iosize - 1);
2728 pg_offset += iosize;
2731 /* we have an inline extent but it didn't get marked up
2732 * to date. Error out
2734 if (block_start == EXTENT_MAP_INLINE) {
2736 unlock_extent(tree, cur, cur + iosize - 1);
2738 pg_offset += iosize;
2743 if (tree->ops && tree->ops->readpage_io_hook) {
2744 ret = tree->ops->readpage_io_hook(page, cur,
2748 unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
2750 ret = submit_extent_page(READ, tree, page,
2751 sector, disk_io_size, pg_offset,
2753 end_bio_extent_readpage, mirror_num,
2758 *bio_flags = this_bio_flag;
2763 unlock_extent(tree, cur, cur + iosize - 1);
2766 pg_offset += iosize;
2770 if (!PageError(page))
2771 SetPageUptodate(page);
2777 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
2778 get_extent_t *get_extent, int mirror_num)
2780 struct bio *bio = NULL;
2781 unsigned long bio_flags = 0;
2784 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
2787 ret = submit_one_bio(READ, bio, mirror_num, bio_flags);
2791 static noinline void update_nr_written(struct page *page,
2792 struct writeback_control *wbc,
2793 unsigned long nr_written)
2795 wbc->nr_to_write -= nr_written;
2796 if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
2797 wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
2798 page->mapping->writeback_index = page->index + nr_written;
2802 * the writepage semantics are similar to regular writepage. extent
2803 * records are inserted to lock ranges in the tree, and as dirty areas
2804 * are found, they are marked writeback. Then the lock bits are removed
2805 * and the end_io handler clears the writeback ranges
2807 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2810 struct inode *inode = page->mapping->host;
2811 struct extent_page_data *epd = data;
2812 struct extent_io_tree *tree = epd->tree;
2813 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2815 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2819 u64 last_byte = i_size_read(inode);
2823 struct extent_state *cached_state = NULL;
2824 struct extent_map *em;
2825 struct block_device *bdev;
2828 size_t pg_offset = 0;
2830 loff_t i_size = i_size_read(inode);
2831 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
2837 unsigned long nr_written = 0;
2838 bool fill_delalloc = true;
2840 if (wbc->sync_mode == WB_SYNC_ALL)
2841 write_flags = WRITE_SYNC;
2843 write_flags = WRITE;
2845 trace___extent_writepage(page, inode, wbc);
2847 WARN_ON(!PageLocked(page));
2849 ClearPageError(page);
2851 pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
2852 if (page->index > end_index ||
2853 (page->index == end_index && !pg_offset)) {
2854 page->mapping->a_ops->invalidatepage(page, 0);
2859 if (page->index == end_index) {
2862 userpage = kmap_atomic(page);
2863 memset(userpage + pg_offset, 0,
2864 PAGE_CACHE_SIZE - pg_offset);
2865 kunmap_atomic(userpage);
2866 flush_dcache_page(page);
2870 set_page_extent_mapped(page);
2872 if (!tree->ops || !tree->ops->fill_delalloc)
2873 fill_delalloc = false;
2875 delalloc_start = start;
2878 if (!epd->extent_locked && fill_delalloc) {
2879 u64 delalloc_to_write = 0;
2881 * make sure the wbc mapping index is at least updated
2884 update_nr_written(page, wbc, 0);
2886 while (delalloc_end < page_end) {
2887 nr_delalloc = find_lock_delalloc_range(inode, tree,
2892 if (nr_delalloc == 0) {
2893 delalloc_start = delalloc_end + 1;
2896 ret = tree->ops->fill_delalloc(inode, page,
2901 /* File system has been set read-only */
2907 * delalloc_end is already one less than the total
2908 * length, so we don't subtract one from
2911 delalloc_to_write += (delalloc_end - delalloc_start +
2914 delalloc_start = delalloc_end + 1;
2916 if (wbc->nr_to_write < delalloc_to_write) {
2919 if (delalloc_to_write < thresh * 2)
2920 thresh = delalloc_to_write;
2921 wbc->nr_to_write = min_t(u64, delalloc_to_write,
2925 /* did the fill delalloc function already unlock and start
2931 * we've unlocked the page, so we can't update
2932 * the mapping's writeback index, just update
2935 wbc->nr_to_write -= nr_written;
2939 if (tree->ops && tree->ops->writepage_start_hook) {
2940 ret = tree->ops->writepage_start_hook(page, start,
2943 /* Fixup worker will requeue */
2945 wbc->pages_skipped++;
2947 redirty_page_for_writepage(wbc, page);
2948 update_nr_written(page, wbc, nr_written);
2956 * we don't want to touch the inode after unlocking the page,
2957 * so we update the mapping writeback index now
2959 update_nr_written(page, wbc, nr_written + 1);
2962 if (last_byte <= start) {
2963 if (tree->ops && tree->ops->writepage_end_io_hook)
2964 tree->ops->writepage_end_io_hook(page, start,
2969 blocksize = inode->i_sb->s_blocksize;
2971 while (cur <= end) {
2972 if (cur >= last_byte) {
2973 if (tree->ops && tree->ops->writepage_end_io_hook)
2974 tree->ops->writepage_end_io_hook(page, cur,
2978 em = epd->get_extent(inode, page, pg_offset, cur,
2980 if (IS_ERR_OR_NULL(em)) {
2985 extent_offset = cur - em->start;
2986 BUG_ON(extent_map_end(em) <= cur);
2988 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2989 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2990 sector = (em->block_start + extent_offset) >> 9;
2992 block_start = em->block_start;
2993 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
2994 free_extent_map(em);
2998 * compressed and inline extents are written through other
3001 if (compressed || block_start == EXTENT_MAP_HOLE ||
3002 block_start == EXTENT_MAP_INLINE) {
3004 * end_io notification does not happen here for
3005 * compressed extents
3007 if (!compressed && tree->ops &&
3008 tree->ops->writepage_end_io_hook)
3009 tree->ops->writepage_end_io_hook(page, cur,
3012 else if (compressed) {
3013 /* we don't want to end_page_writeback on
3014 * a compressed extent. this happens
3021 pg_offset += iosize;
3024 /* leave this out until we have a page_mkwrite call */
3025 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
3026 EXTENT_DIRTY, 0, NULL)) {
3028 pg_offset += iosize;
3032 if (tree->ops && tree->ops->writepage_io_hook) {
3033 ret = tree->ops->writepage_io_hook(page, cur,
3041 unsigned long max_nr = end_index + 1;
3043 set_range_writeback(tree, cur, cur + iosize - 1);
3044 if (!PageWriteback(page)) {
3045 printk(KERN_ERR "btrfs warning page %lu not "
3046 "writeback, cur %llu end %llu\n",
3047 page->index, (unsigned long long)cur,
3048 (unsigned long long)end);
3051 ret = submit_extent_page(write_flags, tree, page,
3052 sector, iosize, pg_offset,
3053 bdev, &epd->bio, max_nr,
3054 end_bio_extent_writepage,
3060 pg_offset += iosize;
3065 /* make sure the mapping tag for page dirty gets cleared */
3066 set_page_writeback(page);
3067 end_page_writeback(page);
3073 /* drop our reference on any cached states */
3074 free_extent_state(cached_state);
3078 static int eb_wait(void *word)
3084 static void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3086 wait_on_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK, eb_wait,
3087 TASK_UNINTERRUPTIBLE);
3090 static int lock_extent_buffer_for_io(struct extent_buffer *eb,
3091 struct btrfs_fs_info *fs_info,
3092 struct extent_page_data *epd)
3094 unsigned long i, num_pages;
3098 if (!btrfs_try_tree_write_lock(eb)) {
3100 flush_write_bio(epd);
3101 btrfs_tree_lock(eb);
3104 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3105 btrfs_tree_unlock(eb);
3109 flush_write_bio(epd);
3113 wait_on_extent_buffer_writeback(eb);
3114 btrfs_tree_lock(eb);
3115 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3117 btrfs_tree_unlock(eb);
3122 * We need to do this to prevent races in people who check if the eb is
3123 * under IO since we can end up having no IO bits set for a short period
3126 spin_lock(&eb->refs_lock);
3127 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3128 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3129 spin_unlock(&eb->refs_lock);
3130 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3131 spin_lock(&fs_info->delalloc_lock);
3132 if (fs_info->dirty_metadata_bytes >= eb->len)
3133 fs_info->dirty_metadata_bytes -= eb->len;
3136 spin_unlock(&fs_info->delalloc_lock);
3139 spin_unlock(&eb->refs_lock);
3142 btrfs_tree_unlock(eb);
3147 num_pages = num_extent_pages(eb->start, eb->len);
3148 for (i = 0; i < num_pages; i++) {
3149 struct page *p = extent_buffer_page(eb, i);
3151 if (!trylock_page(p)) {
3153 flush_write_bio(epd);
3163 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3165 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3166 smp_mb__after_clear_bit();
3167 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3170 static void end_bio_extent_buffer_writepage(struct bio *bio, int err)
3172 int uptodate = err == 0;
3173 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
3174 struct extent_buffer *eb;
3178 struct page *page = bvec->bv_page;
3181 eb = (struct extent_buffer *)page->private;
3183 done = atomic_dec_and_test(&eb->io_pages);
3185 if (!uptodate || test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
3186 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3187 ClearPageUptodate(page);
3191 end_page_writeback(page);
3196 end_extent_buffer_writeback(eb);
3197 } while (bvec >= bio->bi_io_vec);
3203 static int write_one_eb(struct extent_buffer *eb,
3204 struct btrfs_fs_info *fs_info,
3205 struct writeback_control *wbc,
3206 struct extent_page_data *epd)
3208 struct block_device *bdev = fs_info->fs_devices->latest_bdev;
3209 u64 offset = eb->start;
3210 unsigned long i, num_pages;
3211 unsigned long bio_flags = 0;
3212 int rw = (epd->sync_io ? WRITE_SYNC : WRITE);
3215 clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3216 num_pages = num_extent_pages(eb->start, eb->len);
3217 atomic_set(&eb->io_pages, num_pages);
3218 if (btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID)
3219 bio_flags = EXTENT_BIO_TREE_LOG;
3221 for (i = 0; i < num_pages; i++) {
3222 struct page *p = extent_buffer_page(eb, i);
3224 clear_page_dirty_for_io(p);
3225 set_page_writeback(p);
3226 ret = submit_extent_page(rw, eb->tree, p, offset >> 9,
3227 PAGE_CACHE_SIZE, 0, bdev, &epd->bio,
3228 -1, end_bio_extent_buffer_writepage,
3229 0, epd->bio_flags, bio_flags);
3230 epd->bio_flags = bio_flags;
3232 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3234 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3235 end_extent_buffer_writeback(eb);
3239 offset += PAGE_CACHE_SIZE;
3240 update_nr_written(p, wbc, 1);
3244 if (unlikely(ret)) {
3245 for (; i < num_pages; i++) {
3246 struct page *p = extent_buffer_page(eb, i);
3254 int btree_write_cache_pages(struct address_space *mapping,
3255 struct writeback_control *wbc)
3257 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
3258 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
3259 struct extent_buffer *eb, *prev_eb = NULL;
3260 struct extent_page_data epd = {
3264 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3269 int nr_to_write_done = 0;
3270 struct pagevec pvec;
3273 pgoff_t end; /* Inclusive */
3277 pagevec_init(&pvec, 0);
3278 if (wbc->range_cyclic) {
3279 index = mapping->writeback_index; /* Start from prev offset */
3282 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3283 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3286 if (wbc->sync_mode == WB_SYNC_ALL)
3287 tag = PAGECACHE_TAG_TOWRITE;
3289 tag = PAGECACHE_TAG_DIRTY;
3291 if (wbc->sync_mode == WB_SYNC_ALL)
3292 tag_pages_for_writeback(mapping, index, end);
3293 while (!done && !nr_to_write_done && (index <= end) &&
3294 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3295 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3299 for (i = 0; i < nr_pages; i++) {
3300 struct page *page = pvec.pages[i];
3302 if (!PagePrivate(page))
3305 if (!wbc->range_cyclic && page->index > end) {
3310 spin_lock(&mapping->private_lock);
3311 if (!PagePrivate(page)) {
3312 spin_unlock(&mapping->private_lock);
3316 eb = (struct extent_buffer *)page->private;
3319 * Shouldn't happen and normally this would be a BUG_ON
3320 * but no sense in crashing the users box for something
3321 * we can survive anyway.
3324 spin_unlock(&mapping->private_lock);
3329 if (eb == prev_eb) {
3330 spin_unlock(&mapping->private_lock);
3334 ret = atomic_inc_not_zero(&eb->refs);
3335 spin_unlock(&mapping->private_lock);
3340 ret = lock_extent_buffer_for_io(eb, fs_info, &epd);
3342 free_extent_buffer(eb);
3346 ret = write_one_eb(eb, fs_info, wbc, &epd);
3349 free_extent_buffer(eb);
3352 free_extent_buffer(eb);
3355 * the filesystem may choose to bump up nr_to_write.
3356 * We have to make sure to honor the new nr_to_write
3359 nr_to_write_done = wbc->nr_to_write <= 0;
3361 pagevec_release(&pvec);
3364 if (!scanned && !done) {
3366 * We hit the last page and there is more work to be done: wrap
3367 * back to the start of the file
3373 flush_write_bio(&epd);
3378 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3379 * @mapping: address space structure to write
3380 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3381 * @writepage: function called for each page
3382 * @data: data passed to writepage function
3384 * If a page is already under I/O, write_cache_pages() skips it, even
3385 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3386 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3387 * and msync() need to guarantee that all the data which was dirty at the time
3388 * the call was made get new I/O started against them. If wbc->sync_mode is
3389 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3390 * existing IO to complete.
3392 static int extent_write_cache_pages(struct extent_io_tree *tree,
3393 struct address_space *mapping,
3394 struct writeback_control *wbc,
3395 writepage_t writepage, void *data,
3396 void (*flush_fn)(void *))
3398 struct inode *inode = mapping->host;
3401 int nr_to_write_done = 0;
3402 struct pagevec pvec;
3405 pgoff_t end; /* Inclusive */
3410 * We have to hold onto the inode so that ordered extents can do their
3411 * work when the IO finishes. The alternative to this is failing to add
3412 * an ordered extent if the igrab() fails there and that is a huge pain
3413 * to deal with, so instead just hold onto the inode throughout the
3414 * writepages operation. If it fails here we are freeing up the inode
3415 * anyway and we'd rather not waste our time writing out stuff that is
3416 * going to be truncated anyway.
3421 pagevec_init(&pvec, 0);
3422 if (wbc->range_cyclic) {
3423 index = mapping->writeback_index; /* Start from prev offset */
3426 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3427 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3430 if (wbc->sync_mode == WB_SYNC_ALL)
3431 tag = PAGECACHE_TAG_TOWRITE;
3433 tag = PAGECACHE_TAG_DIRTY;
3435 if (wbc->sync_mode == WB_SYNC_ALL)
3436 tag_pages_for_writeback(mapping, index, end);
3437 while (!done && !nr_to_write_done && (index <= end) &&
3438 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3439 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3443 for (i = 0; i < nr_pages; i++) {
3444 struct page *page = pvec.pages[i];
3447 * At this point we hold neither mapping->tree_lock nor
3448 * lock on the page itself: the page may be truncated or
3449 * invalidated (changing page->mapping to NULL), or even
3450 * swizzled back from swapper_space to tmpfs file
3454 tree->ops->write_cache_pages_lock_hook) {
3455 tree->ops->write_cache_pages_lock_hook(page,
3458 if (!trylock_page(page)) {
3464 if (unlikely(page->mapping != mapping)) {
3469 if (!wbc->range_cyclic && page->index > end) {
3475 if (wbc->sync_mode != WB_SYNC_NONE) {
3476 if (PageWriteback(page))
3478 wait_on_page_writeback(page);
3481 if (PageWriteback(page) ||
3482 !clear_page_dirty_for_io(page)) {
3487 ret = (*writepage)(page, wbc, data);
3489 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
3497 * the filesystem may choose to bump up nr_to_write.
3498 * We have to make sure to honor the new nr_to_write
3501 nr_to_write_done = wbc->nr_to_write <= 0;
3503 pagevec_release(&pvec);
3506 if (!scanned && !done) {
3508 * We hit the last page and there is more work to be done: wrap
3509 * back to the start of the file
3515 btrfs_add_delayed_iput(inode);
3519 static void flush_epd_write_bio(struct extent_page_data *epd)
3528 ret = submit_one_bio(rw, epd->bio, 0, epd->bio_flags);
3529 BUG_ON(ret < 0); /* -ENOMEM */
3534 static noinline void flush_write_bio(void *data)
3536 struct extent_page_data *epd = data;
3537 flush_epd_write_bio(epd);
3540 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
3541 get_extent_t *get_extent,
3542 struct writeback_control *wbc)
3545 struct extent_page_data epd = {
3548 .get_extent = get_extent,
3550 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3554 ret = __extent_writepage(page, wbc, &epd);
3556 flush_epd_write_bio(&epd);
3560 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
3561 u64 start, u64 end, get_extent_t *get_extent,
3565 struct address_space *mapping = inode->i_mapping;
3567 unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
3570 struct extent_page_data epd = {
3573 .get_extent = get_extent,
3575 .sync_io = mode == WB_SYNC_ALL,
3578 struct writeback_control wbc_writepages = {
3580 .nr_to_write = nr_pages * 2,
3581 .range_start = start,
3582 .range_end = end + 1,
3585 while (start <= end) {
3586 page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
3587 if (clear_page_dirty_for_io(page))
3588 ret = __extent_writepage(page, &wbc_writepages, &epd);
3590 if (tree->ops && tree->ops->writepage_end_io_hook)
3591 tree->ops->writepage_end_io_hook(page, start,
3592 start + PAGE_CACHE_SIZE - 1,
3596 page_cache_release(page);
3597 start += PAGE_CACHE_SIZE;
3600 flush_epd_write_bio(&epd);
3604 int extent_writepages(struct extent_io_tree *tree,
3605 struct address_space *mapping,
3606 get_extent_t *get_extent,
3607 struct writeback_control *wbc)
3610 struct extent_page_data epd = {
3613 .get_extent = get_extent,
3615 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3619 ret = extent_write_cache_pages(tree, mapping, wbc,
3620 __extent_writepage, &epd,
3622 flush_epd_write_bio(&epd);
3626 int extent_readpages(struct extent_io_tree *tree,
3627 struct address_space *mapping,
3628 struct list_head *pages, unsigned nr_pages,
3629 get_extent_t get_extent)
3631 struct bio *bio = NULL;
3633 unsigned long bio_flags = 0;
3634 struct page *pagepool[16];
3639 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
3640 page = list_entry(pages->prev, struct page, lru);
3642 prefetchw(&page->flags);
3643 list_del(&page->lru);
3644 if (add_to_page_cache_lru(page, mapping,
3645 page->index, GFP_NOFS)) {
3646 page_cache_release(page);
3650 pagepool[nr++] = page;
3651 if (nr < ARRAY_SIZE(pagepool))
3653 for (i = 0; i < nr; i++) {
3654 __extent_read_full_page(tree, pagepool[i], get_extent,
3655 &bio, 0, &bio_flags);
3656 page_cache_release(pagepool[i]);
3660 for (i = 0; i < nr; i++) {
3661 __extent_read_full_page(tree, pagepool[i], get_extent,
3662 &bio, 0, &bio_flags);
3663 page_cache_release(pagepool[i]);
3666 BUG_ON(!list_empty(pages));
3668 return submit_one_bio(READ, bio, 0, bio_flags);
3673 * basic invalidatepage code, this waits on any locked or writeback
3674 * ranges corresponding to the page, and then deletes any extent state
3675 * records from the tree
3677 int extent_invalidatepage(struct extent_io_tree *tree,
3678 struct page *page, unsigned long offset)
3680 struct extent_state *cached_state = NULL;
3681 u64 start = ((u64)page->index << PAGE_CACHE_SHIFT);
3682 u64 end = start + PAGE_CACHE_SIZE - 1;
3683 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
3685 start += (offset + blocksize - 1) & ~(blocksize - 1);
3689 lock_extent_bits(tree, start, end, 0, &cached_state);
3690 wait_on_page_writeback(page);
3691 clear_extent_bit(tree, start, end,
3692 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
3693 EXTENT_DO_ACCOUNTING,
3694 1, 1, &cached_state, GFP_NOFS);
3699 * a helper for releasepage, this tests for areas of the page that
3700 * are locked or under IO and drops the related state bits if it is safe
3703 int try_release_extent_state(struct extent_map_tree *map,
3704 struct extent_io_tree *tree, struct page *page,
3707 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
3708 u64 end = start + PAGE_CACHE_SIZE - 1;
3711 if (test_range_bit(tree, start, end,
3712 EXTENT_IOBITS, 0, NULL))
3715 if ((mask & GFP_NOFS) == GFP_NOFS)
3718 * at this point we can safely clear everything except the
3719 * locked bit and the nodatasum bit
3721 ret = clear_extent_bit(tree, start, end,
3722 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
3725 /* if clear_extent_bit failed for enomem reasons,
3726 * we can't allow the release to continue.
3737 * a helper for releasepage. As long as there are no locked extents
3738 * in the range corresponding to the page, both state records and extent
3739 * map records are removed
3741 int try_release_extent_mapping(struct extent_map_tree *map,
3742 struct extent_io_tree *tree, struct page *page,
3745 struct extent_map *em;
3746 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
3747 u64 end = start + PAGE_CACHE_SIZE - 1;
3749 if ((mask & __GFP_WAIT) &&
3750 page->mapping->host->i_size > 16 * 1024 * 1024) {
3752 while (start <= end) {
3753 len = end - start + 1;
3754 write_lock(&map->lock);
3755 em = lookup_extent_mapping(map, start, len);
3757 write_unlock(&map->lock);
3760 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
3761 em->start != start) {
3762 write_unlock(&map->lock);
3763 free_extent_map(em);
3766 if (!test_range_bit(tree, em->start,
3767 extent_map_end(em) - 1,
3768 EXTENT_LOCKED | EXTENT_WRITEBACK,
3770 remove_extent_mapping(map, em);
3771 /* once for the rb tree */
3772 free_extent_map(em);
3774 start = extent_map_end(em);
3775 write_unlock(&map->lock);
3778 free_extent_map(em);
3781 return try_release_extent_state(map, tree, page, mask);
3785 * helper function for fiemap, which doesn't want to see any holes.
3786 * This maps until we find something past 'last'
3788 static struct extent_map *get_extent_skip_holes(struct inode *inode,
3791 get_extent_t *get_extent)
3793 u64 sectorsize = BTRFS_I(inode)->root->sectorsize;
3794 struct extent_map *em;
3801 len = last - offset;
3804 len = (len + sectorsize - 1) & ~(sectorsize - 1);
3805 em = get_extent(inode, NULL, 0, offset, len, 0);
3806 if (IS_ERR_OR_NULL(em))
3809 /* if this isn't a hole return it */
3810 if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) &&
3811 em->block_start != EXTENT_MAP_HOLE) {
3815 /* this is a hole, advance to the next extent */
3816 offset = extent_map_end(em);
3817 free_extent_map(em);
3824 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
3825 __u64 start, __u64 len, get_extent_t *get_extent)
3829 u64 max = start + len;
3833 u64 last_for_get_extent = 0;
3835 u64 isize = i_size_read(inode);
3836 struct btrfs_key found_key;
3837 struct extent_map *em = NULL;
3838 struct extent_state *cached_state = NULL;
3839 struct btrfs_path *path;
3840 struct btrfs_file_extent_item *item;
3845 unsigned long emflags;
3850 path = btrfs_alloc_path();
3853 path->leave_spinning = 1;
3855 start = ALIGN(start, BTRFS_I(inode)->root->sectorsize);
3856 len = ALIGN(len, BTRFS_I(inode)->root->sectorsize);
3859 * lookup the last file extent. We're not using i_size here
3860 * because there might be preallocation past i_size
3862 ret = btrfs_lookup_file_extent(NULL, BTRFS_I(inode)->root,
3863 path, btrfs_ino(inode), -1, 0);
3865 btrfs_free_path(path);
3870 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3871 struct btrfs_file_extent_item);
3872 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
3873 found_type = btrfs_key_type(&found_key);
3875 /* No extents, but there might be delalloc bits */
3876 if (found_key.objectid != btrfs_ino(inode) ||
3877 found_type != BTRFS_EXTENT_DATA_KEY) {
3878 /* have to trust i_size as the end */
3880 last_for_get_extent = isize;
3883 * remember the start of the last extent. There are a
3884 * bunch of different factors that go into the length of the
3885 * extent, so its much less complex to remember where it started
3887 last = found_key.offset;
3888 last_for_get_extent = last + 1;
3890 btrfs_free_path(path);
3893 * we might have some extents allocated but more delalloc past those
3894 * extents. so, we trust isize unless the start of the last extent is
3899 last_for_get_extent = isize;
3902 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len, 0,
3905 em = get_extent_skip_holes(inode, start, last_for_get_extent,
3915 u64 offset_in_extent;
3917 /* break if the extent we found is outside the range */
3918 if (em->start >= max || extent_map_end(em) < off)
3922 * get_extent may return an extent that starts before our
3923 * requested range. We have to make sure the ranges
3924 * we return to fiemap always move forward and don't
3925 * overlap, so adjust the offsets here
3927 em_start = max(em->start, off);
3930 * record the offset from the start of the extent
3931 * for adjusting the disk offset below
3933 offset_in_extent = em_start - em->start;
3934 em_end = extent_map_end(em);
3935 em_len = em_end - em_start;
3936 emflags = em->flags;
3941 * bump off for our next call to get_extent
3943 off = extent_map_end(em);
3947 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
3949 flags |= FIEMAP_EXTENT_LAST;
3950 } else if (em->block_start == EXTENT_MAP_INLINE) {
3951 flags |= (FIEMAP_EXTENT_DATA_INLINE |
3952 FIEMAP_EXTENT_NOT_ALIGNED);
3953 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
3954 flags |= (FIEMAP_EXTENT_DELALLOC |
3955 FIEMAP_EXTENT_UNKNOWN);
3957 disko = em->block_start + offset_in_extent;
3959 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
3960 flags |= FIEMAP_EXTENT_ENCODED;
3962 free_extent_map(em);
3964 if ((em_start >= last) || em_len == (u64)-1 ||
3965 (last == (u64)-1 && isize <= em_end)) {
3966 flags |= FIEMAP_EXTENT_LAST;
3970 /* now scan forward to see if this is really the last extent. */
3971 em = get_extent_skip_holes(inode, off, last_for_get_extent,
3978 flags |= FIEMAP_EXTENT_LAST;
3981 ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
3987 free_extent_map(em);
3989 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len,
3990 &cached_state, GFP_NOFS);
3994 static void __free_extent_buffer(struct extent_buffer *eb)
3997 unsigned long flags;
3998 spin_lock_irqsave(&leak_lock, flags);
3999 list_del(&eb->leak_list);
4000 spin_unlock_irqrestore(&leak_lock, flags);
4002 if (eb->pages && eb->pages != eb->inline_pages)
4004 kmem_cache_free(extent_buffer_cache, eb);
4007 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
4012 struct extent_buffer *eb = NULL;
4014 unsigned long flags;
4017 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
4024 rwlock_init(&eb->lock);
4025 atomic_set(&eb->write_locks, 0);
4026 atomic_set(&eb->read_locks, 0);
4027 atomic_set(&eb->blocking_readers, 0);
4028 atomic_set(&eb->blocking_writers, 0);
4029 atomic_set(&eb->spinning_readers, 0);
4030 atomic_set(&eb->spinning_writers, 0);
4031 eb->lock_nested = 0;
4032 init_waitqueue_head(&eb->write_lock_wq);
4033 init_waitqueue_head(&eb->read_lock_wq);
4036 spin_lock_irqsave(&leak_lock, flags);
4037 list_add(&eb->leak_list, &buffers);
4038 spin_unlock_irqrestore(&leak_lock, flags);
4040 spin_lock_init(&eb->refs_lock);
4041 atomic_set(&eb->refs, 1);
4042 atomic_set(&eb->io_pages, 0);
4044 if (len > MAX_INLINE_EXTENT_BUFFER_SIZE) {
4045 struct page **pages;
4046 int num_pages = (len + PAGE_CACHE_SIZE - 1) >>
4048 pages = kzalloc(num_pages, mask);
4050 __free_extent_buffer(eb);
4055 eb->pages = eb->inline_pages;
4061 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src)
4065 struct extent_buffer *new;
4066 unsigned long num_pages = num_extent_pages(src->start, src->len);
4068 new = __alloc_extent_buffer(NULL, src->start, src->len, GFP_ATOMIC);
4072 for (i = 0; i < num_pages; i++) {
4073 p = alloc_page(GFP_ATOMIC);
4075 attach_extent_buffer_page(new, p);
4076 WARN_ON(PageDirty(p));
4081 copy_extent_buffer(new, src, 0, 0, src->len);
4082 set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
4083 set_bit(EXTENT_BUFFER_DUMMY, &new->bflags);
4088 struct extent_buffer *alloc_dummy_extent_buffer(u64 start, unsigned long len)
4090 struct extent_buffer *eb;
4091 unsigned long num_pages = num_extent_pages(0, len);
4094 eb = __alloc_extent_buffer(NULL, start, len, GFP_ATOMIC);
4098 for (i = 0; i < num_pages; i++) {
4099 eb->pages[i] = alloc_page(GFP_ATOMIC);
4103 set_extent_buffer_uptodate(eb);
4104 btrfs_set_header_nritems(eb, 0);
4105 set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4110 __free_page(eb->pages[i - 1]);
4111 __free_extent_buffer(eb);
4115 static int extent_buffer_under_io(struct extent_buffer *eb)
4117 return (atomic_read(&eb->io_pages) ||
4118 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4119 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4123 * Helper for releasing extent buffer page.
4125 static void btrfs_release_extent_buffer_page(struct extent_buffer *eb,
4126 unsigned long start_idx)
4128 unsigned long index;
4129 unsigned long num_pages;
4131 int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4133 BUG_ON(extent_buffer_under_io(eb));
4135 num_pages = num_extent_pages(eb->start, eb->len);
4136 index = start_idx + num_pages;
4137 if (start_idx >= index)
4142 page = extent_buffer_page(eb, index);
4143 if (page && mapped) {
4144 spin_lock(&page->mapping->private_lock);
4146 * We do this since we'll remove the pages after we've
4147 * removed the eb from the radix tree, so we could race
4148 * and have this page now attached to the new eb. So
4149 * only clear page_private if it's still connected to
4152 if (PagePrivate(page) &&
4153 page->private == (unsigned long)eb) {
4154 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4155 BUG_ON(PageDirty(page));
4156 BUG_ON(PageWriteback(page));
4158 * We need to make sure we haven't be attached
4161 ClearPagePrivate(page);
4162 set_page_private(page, 0);
4163 /* One for the page private */
4164 page_cache_release(page);
4166 spin_unlock(&page->mapping->private_lock);
4170 /* One for when we alloced the page */
4171 page_cache_release(page);
4173 } while (index != start_idx);
4177 * Helper for releasing the extent buffer.
4179 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4181 btrfs_release_extent_buffer_page(eb, 0);
4182 __free_extent_buffer(eb);
4185 static void check_buffer_tree_ref(struct extent_buffer *eb)
4187 /* the ref bit is tricky. We have to make sure it is set
4188 * if we have the buffer dirty. Otherwise the
4189 * code to free a buffer can end up dropping a dirty
4192 * Once the ref bit is set, it won't go away while the
4193 * buffer is dirty or in writeback, and it also won't
4194 * go away while we have the reference count on the
4197 * We can't just set the ref bit without bumping the
4198 * ref on the eb because free_extent_buffer might
4199 * see the ref bit and try to clear it. If this happens
4200 * free_extent_buffer might end up dropping our original
4201 * ref by mistake and freeing the page before we are able
4202 * to add one more ref.
4204 * So bump the ref count first, then set the bit. If someone
4205 * beat us to it, drop the ref we added.
4207 spin_lock(&eb->refs_lock);
4208 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4209 atomic_inc(&eb->refs);
4210 spin_unlock(&eb->refs_lock);
4213 static void mark_extent_buffer_accessed(struct extent_buffer *eb)
4215 unsigned long num_pages, i;
4217 check_buffer_tree_ref(eb);
4219 num_pages = num_extent_pages(eb->start, eb->len);
4220 for (i = 0; i < num_pages; i++) {
4221 struct page *p = extent_buffer_page(eb, i);
4222 mark_page_accessed(p);
4226 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
4227 u64 start, unsigned long len)
4229 unsigned long num_pages = num_extent_pages(start, len);
4231 unsigned long index = start >> PAGE_CACHE_SHIFT;
4232 struct extent_buffer *eb;
4233 struct extent_buffer *exists = NULL;
4235 struct address_space *mapping = tree->mapping;
4240 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
4241 if (eb && atomic_inc_not_zero(&eb->refs)) {
4243 mark_extent_buffer_accessed(eb);
4248 eb = __alloc_extent_buffer(tree, start, len, GFP_NOFS);
4252 for (i = 0; i < num_pages; i++, index++) {
4253 p = find_or_create_page(mapping, index, GFP_NOFS);
4257 spin_lock(&mapping->private_lock);
4258 if (PagePrivate(p)) {
4260 * We could have already allocated an eb for this page
4261 * and attached one so lets see if we can get a ref on
4262 * the existing eb, and if we can we know it's good and
4263 * we can just return that one, else we know we can just
4264 * overwrite page->private.
4266 exists = (struct extent_buffer *)p->private;
4267 if (atomic_inc_not_zero(&exists->refs)) {
4268 spin_unlock(&mapping->private_lock);
4270 page_cache_release(p);
4271 mark_extent_buffer_accessed(exists);
4276 * Do this so attach doesn't complain and we need to
4277 * drop the ref the old guy had.
4279 ClearPagePrivate(p);
4280 WARN_ON(PageDirty(p));
4281 page_cache_release(p);
4283 attach_extent_buffer_page(eb, p);
4284 spin_unlock(&mapping->private_lock);
4285 WARN_ON(PageDirty(p));
4286 mark_page_accessed(p);
4288 if (!PageUptodate(p))
4292 * see below about how we avoid a nasty race with release page
4293 * and why we unlock later
4297 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4299 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
4303 spin_lock(&tree->buffer_lock);
4304 ret = radix_tree_insert(&tree->buffer, start >> PAGE_CACHE_SHIFT, eb);
4305 if (ret == -EEXIST) {
4306 exists = radix_tree_lookup(&tree->buffer,
4307 start >> PAGE_CACHE_SHIFT);
4308 if (!atomic_inc_not_zero(&exists->refs)) {
4309 spin_unlock(&tree->buffer_lock);
4310 radix_tree_preload_end();
4314 spin_unlock(&tree->buffer_lock);
4315 radix_tree_preload_end();
4316 mark_extent_buffer_accessed(exists);
4319 /* add one reference for the tree */
4320 check_buffer_tree_ref(eb);
4321 spin_unlock(&tree->buffer_lock);
4322 radix_tree_preload_end();
4325 * there is a race where release page may have
4326 * tried to find this extent buffer in the radix
4327 * but failed. It will tell the VM it is safe to
4328 * reclaim the, and it will clear the page private bit.
4329 * We must make sure to set the page private bit properly
4330 * after the extent buffer is in the radix tree so
4331 * it doesn't get lost
4333 SetPageChecked(eb->pages[0]);
4334 for (i = 1; i < num_pages; i++) {
4335 p = extent_buffer_page(eb, i);
4336 ClearPageChecked(p);
4339 unlock_page(eb->pages[0]);
4343 for (i = 0; i < num_pages; i++) {
4345 unlock_page(eb->pages[i]);
4348 WARN_ON(!atomic_dec_and_test(&eb->refs));
4349 btrfs_release_extent_buffer(eb);
4353 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
4354 u64 start, unsigned long len)
4356 struct extent_buffer *eb;
4359 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
4360 if (eb && atomic_inc_not_zero(&eb->refs)) {
4362 mark_extent_buffer_accessed(eb);
4370 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
4372 struct extent_buffer *eb =
4373 container_of(head, struct extent_buffer, rcu_head);
4375 __free_extent_buffer(eb);
4378 /* Expects to have eb->eb_lock already held */
4379 static int release_extent_buffer(struct extent_buffer *eb, gfp_t mask)
4381 WARN_ON(atomic_read(&eb->refs) == 0);
4382 if (atomic_dec_and_test(&eb->refs)) {
4383 if (test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags)) {
4384 spin_unlock(&eb->refs_lock);
4386 struct extent_io_tree *tree = eb->tree;
4388 spin_unlock(&eb->refs_lock);
4390 spin_lock(&tree->buffer_lock);
4391 radix_tree_delete(&tree->buffer,
4392 eb->start >> PAGE_CACHE_SHIFT);
4393 spin_unlock(&tree->buffer_lock);
4396 /* Should be safe to release our pages at this point */
4397 btrfs_release_extent_buffer_page(eb, 0);
4398 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
4401 spin_unlock(&eb->refs_lock);
4406 void free_extent_buffer(struct extent_buffer *eb)
4411 spin_lock(&eb->refs_lock);
4412 if (atomic_read(&eb->refs) == 2 &&
4413 test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))
4414 atomic_dec(&eb->refs);
4416 if (atomic_read(&eb->refs) == 2 &&
4417 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
4418 !extent_buffer_under_io(eb) &&
4419 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4420 atomic_dec(&eb->refs);
4423 * I know this is terrible, but it's temporary until we stop tracking
4424 * the uptodate bits and such for the extent buffers.
4426 release_extent_buffer(eb, GFP_ATOMIC);
4429 void free_extent_buffer_stale(struct extent_buffer *eb)
4434 spin_lock(&eb->refs_lock);
4435 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
4437 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
4438 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4439 atomic_dec(&eb->refs);
4440 release_extent_buffer(eb, GFP_NOFS);
4443 void clear_extent_buffer_dirty(struct extent_buffer *eb)
4446 unsigned long num_pages;
4449 num_pages = num_extent_pages(eb->start, eb->len);
4451 for (i = 0; i < num_pages; i++) {
4452 page = extent_buffer_page(eb, i);
4453 if (!PageDirty(page))
4457 WARN_ON(!PagePrivate(page));
4459 clear_page_dirty_for_io(page);
4460 spin_lock_irq(&page->mapping->tree_lock);
4461 if (!PageDirty(page)) {
4462 radix_tree_tag_clear(&page->mapping->page_tree,
4464 PAGECACHE_TAG_DIRTY);
4466 spin_unlock_irq(&page->mapping->tree_lock);
4467 ClearPageError(page);
4470 WARN_ON(atomic_read(&eb->refs) == 0);
4473 int set_extent_buffer_dirty(struct extent_buffer *eb)
4476 unsigned long num_pages;
4479 check_buffer_tree_ref(eb);
4481 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
4483 num_pages = num_extent_pages(eb->start, eb->len);
4484 WARN_ON(atomic_read(&eb->refs) == 0);
4485 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
4487 for (i = 0; i < num_pages; i++)
4488 set_page_dirty(extent_buffer_page(eb, i));
4492 static int range_straddles_pages(u64 start, u64 len)
4494 if (len < PAGE_CACHE_SIZE)
4496 if (start & (PAGE_CACHE_SIZE - 1))
4498 if ((start + len) & (PAGE_CACHE_SIZE - 1))
4503 int clear_extent_buffer_uptodate(struct extent_buffer *eb)
4507 unsigned long num_pages;
4509 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4510 num_pages = num_extent_pages(eb->start, eb->len);
4511 for (i = 0; i < num_pages; i++) {
4512 page = extent_buffer_page(eb, i);
4514 ClearPageUptodate(page);
4519 int set_extent_buffer_uptodate(struct extent_buffer *eb)
4523 unsigned long num_pages;
4525 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4526 num_pages = num_extent_pages(eb->start, eb->len);
4527 for (i = 0; i < num_pages; i++) {
4528 page = extent_buffer_page(eb, i);
4529 SetPageUptodate(page);
4534 int extent_range_uptodate(struct extent_io_tree *tree,
4539 int pg_uptodate = 1;
4541 unsigned long index;
4543 if (range_straddles_pages(start, end - start + 1)) {
4544 ret = test_range_bit(tree, start, end,
4545 EXTENT_UPTODATE, 1, NULL);
4549 while (start <= end) {
4550 index = start >> PAGE_CACHE_SHIFT;
4551 page = find_get_page(tree->mapping, index);
4554 uptodate = PageUptodate(page);
4555 page_cache_release(page);
4560 start += PAGE_CACHE_SIZE;
4565 int extent_buffer_uptodate(struct extent_buffer *eb)
4567 return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4570 int read_extent_buffer_pages(struct extent_io_tree *tree,
4571 struct extent_buffer *eb, u64 start, int wait,
4572 get_extent_t *get_extent, int mirror_num)
4575 unsigned long start_i;
4579 int locked_pages = 0;
4580 int all_uptodate = 1;
4581 unsigned long num_pages;
4582 unsigned long num_reads = 0;
4583 struct bio *bio = NULL;
4584 unsigned long bio_flags = 0;
4586 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
4590 WARN_ON(start < eb->start);
4591 start_i = (start >> PAGE_CACHE_SHIFT) -
4592 (eb->start >> PAGE_CACHE_SHIFT);
4597 num_pages = num_extent_pages(eb->start, eb->len);
4598 for (i = start_i; i < num_pages; i++) {
4599 page = extent_buffer_page(eb, i);
4600 if (wait == WAIT_NONE) {
4601 if (!trylock_page(page))
4607 if (!PageUptodate(page)) {
4614 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4618 clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
4619 eb->read_mirror = 0;
4620 atomic_set(&eb->io_pages, num_reads);
4621 for (i = start_i; i < num_pages; i++) {
4622 page = extent_buffer_page(eb, i);
4623 if (!PageUptodate(page)) {
4624 ClearPageError(page);
4625 err = __extent_read_full_page(tree, page,
4627 mirror_num, &bio_flags);
4636 err = submit_one_bio(READ, bio, mirror_num, bio_flags);
4641 if (ret || wait != WAIT_COMPLETE)
4644 for (i = start_i; i < num_pages; i++) {
4645 page = extent_buffer_page(eb, i);
4646 wait_on_page_locked(page);
4647 if (!PageUptodate(page))
4655 while (locked_pages > 0) {
4656 page = extent_buffer_page(eb, i);
4664 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
4665 unsigned long start,
4672 char *dst = (char *)dstv;
4673 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4674 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4676 WARN_ON(start > eb->len);
4677 WARN_ON(start + len > eb->start + eb->len);
4679 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4682 page = extent_buffer_page(eb, i);
4684 cur = min(len, (PAGE_CACHE_SIZE - offset));
4685 kaddr = page_address(page);
4686 memcpy(dst, kaddr + offset, cur);
4695 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
4696 unsigned long min_len, char **map,
4697 unsigned long *map_start,
4698 unsigned long *map_len)
4700 size_t offset = start & (PAGE_CACHE_SIZE - 1);
4703 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4704 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4705 unsigned long end_i = (start_offset + start + min_len - 1) >>
4712 offset = start_offset;
4716 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
4719 if (start + min_len > eb->len) {
4720 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, "
4721 "wanted %lu %lu\n", (unsigned long long)eb->start,
4722 eb->len, start, min_len);
4726 p = extent_buffer_page(eb, i);
4727 kaddr = page_address(p);
4728 *map = kaddr + offset;
4729 *map_len = PAGE_CACHE_SIZE - offset;
4733 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
4734 unsigned long start,
4741 char *ptr = (char *)ptrv;
4742 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4743 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4746 WARN_ON(start > eb->len);
4747 WARN_ON(start + len > eb->start + eb->len);
4749 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4752 page = extent_buffer_page(eb, i);
4754 cur = min(len, (PAGE_CACHE_SIZE - offset));
4756 kaddr = page_address(page);
4757 ret = memcmp(ptr, kaddr + offset, cur);
4769 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
4770 unsigned long start, unsigned long len)
4776 char *src = (char *)srcv;
4777 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4778 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4780 WARN_ON(start > eb->len);
4781 WARN_ON(start + len > eb->start + eb->len);
4783 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4786 page = extent_buffer_page(eb, i);
4787 WARN_ON(!PageUptodate(page));
4789 cur = min(len, PAGE_CACHE_SIZE - offset);
4790 kaddr = page_address(page);
4791 memcpy(kaddr + offset, src, cur);
4800 void memset_extent_buffer(struct extent_buffer *eb, char c,
4801 unsigned long start, unsigned long len)
4807 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4808 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4810 WARN_ON(start > eb->len);
4811 WARN_ON(start + len > eb->start + eb->len);
4813 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4816 page = extent_buffer_page(eb, i);
4817 WARN_ON(!PageUptodate(page));
4819 cur = min(len, PAGE_CACHE_SIZE - offset);
4820 kaddr = page_address(page);
4821 memset(kaddr + offset, c, cur);
4829 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
4830 unsigned long dst_offset, unsigned long src_offset,
4833 u64 dst_len = dst->len;
4838 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4839 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
4841 WARN_ON(src->len != dst_len);
4843 offset = (start_offset + dst_offset) &
4844 ((unsigned long)PAGE_CACHE_SIZE - 1);
4847 page = extent_buffer_page(dst, i);
4848 WARN_ON(!PageUptodate(page));
4850 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
4852 kaddr = page_address(page);
4853 read_extent_buffer(src, kaddr + offset, src_offset, cur);
4862 static void move_pages(struct page *dst_page, struct page *src_page,
4863 unsigned long dst_off, unsigned long src_off,
4866 char *dst_kaddr = page_address(dst_page);
4867 if (dst_page == src_page) {
4868 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
4870 char *src_kaddr = page_address(src_page);
4871 char *p = dst_kaddr + dst_off + len;
4872 char *s = src_kaddr + src_off + len;
4879 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
4881 unsigned long distance = (src > dst) ? src - dst : dst - src;
4882 return distance < len;
4885 static void copy_pages(struct page *dst_page, struct page *src_page,
4886 unsigned long dst_off, unsigned long src_off,
4889 char *dst_kaddr = page_address(dst_page);
4891 int must_memmove = 0;
4893 if (dst_page != src_page) {
4894 src_kaddr = page_address(src_page);
4896 src_kaddr = dst_kaddr;
4897 if (areas_overlap(src_off, dst_off, len))
4902 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
4904 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
4907 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
4908 unsigned long src_offset, unsigned long len)
4911 size_t dst_off_in_page;
4912 size_t src_off_in_page;
4913 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4914 unsigned long dst_i;
4915 unsigned long src_i;
4917 if (src_offset + len > dst->len) {
4918 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
4919 "len %lu dst len %lu\n", src_offset, len, dst->len);
4922 if (dst_offset + len > dst->len) {
4923 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
4924 "len %lu dst len %lu\n", dst_offset, len, dst->len);
4929 dst_off_in_page = (start_offset + dst_offset) &
4930 ((unsigned long)PAGE_CACHE_SIZE - 1);
4931 src_off_in_page = (start_offset + src_offset) &
4932 ((unsigned long)PAGE_CACHE_SIZE - 1);
4934 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
4935 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
4937 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
4939 cur = min_t(unsigned long, cur,
4940 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
4942 copy_pages(extent_buffer_page(dst, dst_i),
4943 extent_buffer_page(dst, src_i),
4944 dst_off_in_page, src_off_in_page, cur);
4952 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
4953 unsigned long src_offset, unsigned long len)
4956 size_t dst_off_in_page;
4957 size_t src_off_in_page;
4958 unsigned long dst_end = dst_offset + len - 1;
4959 unsigned long src_end = src_offset + len - 1;
4960 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4961 unsigned long dst_i;
4962 unsigned long src_i;
4964 if (src_offset + len > dst->len) {
4965 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
4966 "len %lu len %lu\n", src_offset, len, dst->len);
4969 if (dst_offset + len > dst->len) {
4970 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
4971 "len %lu len %lu\n", dst_offset, len, dst->len);
4974 if (dst_offset < src_offset) {
4975 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
4979 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
4980 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
4982 dst_off_in_page = (start_offset + dst_end) &
4983 ((unsigned long)PAGE_CACHE_SIZE - 1);
4984 src_off_in_page = (start_offset + src_end) &
4985 ((unsigned long)PAGE_CACHE_SIZE - 1);
4987 cur = min_t(unsigned long, len, src_off_in_page + 1);
4988 cur = min(cur, dst_off_in_page + 1);
4989 move_pages(extent_buffer_page(dst, dst_i),
4990 extent_buffer_page(dst, src_i),
4991 dst_off_in_page - cur + 1,
4992 src_off_in_page - cur + 1, cur);
5000 int try_release_extent_buffer(struct page *page, gfp_t mask)
5002 struct extent_buffer *eb;
5005 * We need to make sure noboody is attaching this page to an eb right
5008 spin_lock(&page->mapping->private_lock);
5009 if (!PagePrivate(page)) {
5010 spin_unlock(&page->mapping->private_lock);
5014 eb = (struct extent_buffer *)page->private;
5018 * This is a little awful but should be ok, we need to make sure that
5019 * the eb doesn't disappear out from under us while we're looking at
5022 spin_lock(&eb->refs_lock);
5023 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
5024 spin_unlock(&eb->refs_lock);
5025 spin_unlock(&page->mapping->private_lock);
5028 spin_unlock(&page->mapping->private_lock);
5030 if ((mask & GFP_NOFS) == GFP_NOFS)
5034 * If tree ref isn't set then we know the ref on this eb is a real ref,
5035 * so just return, this page will likely be freed soon anyway.
5037 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
5038 spin_unlock(&eb->refs_lock);
5042 return release_extent_buffer(eb, mask);