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 = page_offset(page);
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 = page_offset(page);
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 = page_offset(page) + bvec->bv_offset;
2301 end = start + bvec->bv_len - 1;
2303 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
2308 if (--bvec >= bio->bi_io_vec)
2309 prefetchw(&bvec->bv_page->flags);
2311 if (end_extent_writepage(page, err, start, end))
2315 end_page_writeback(page);
2317 check_page_writeback(tree, page);
2318 } while (bvec >= bio->bi_io_vec);
2324 * after a readpage IO is done, we need to:
2325 * clear the uptodate bits on error
2326 * set the uptodate bits if things worked
2327 * set the page up to date if all extents in the tree are uptodate
2328 * clear the lock bit in the extent tree
2329 * unlock the page if there are no other extents locked for it
2331 * Scheduling is not allowed, so the extent state tree is expected
2332 * to have one and only one object corresponding to this IO.
2334 static void end_bio_extent_readpage(struct bio *bio, int err)
2336 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
2337 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
2338 struct bio_vec *bvec = bio->bi_io_vec;
2339 struct extent_io_tree *tree;
2350 struct page *page = bvec->bv_page;
2351 struct extent_state *cached = NULL;
2352 struct extent_state *state;
2354 pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, "
2355 "mirror=%ld\n", (u64)bio->bi_sector, err,
2356 (long int)bio->bi_bdev);
2357 tree = &BTRFS_I(page->mapping->host)->io_tree;
2359 start = page_offset(page) + bvec->bv_offset;
2360 end = start + bvec->bv_len - 1;
2362 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
2367 if (++bvec <= bvec_end)
2368 prefetchw(&bvec->bv_page->flags);
2370 spin_lock(&tree->lock);
2371 state = find_first_extent_bit_state(tree, start, EXTENT_LOCKED);
2372 if (state && state->start == start) {
2374 * take a reference on the state, unlock will drop
2377 cache_state(state, &cached);
2379 spin_unlock(&tree->lock);
2381 mirror = (int)(unsigned long)bio->bi_bdev;
2382 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
2383 ret = tree->ops->readpage_end_io_hook(page, start, end,
2388 clean_io_failure(start, page);
2391 if (!uptodate && tree->ops && tree->ops->readpage_io_failed_hook) {
2392 ret = tree->ops->readpage_io_failed_hook(page, mirror);
2394 test_bit(BIO_UPTODATE, &bio->bi_flags))
2396 } else if (!uptodate) {
2398 * The generic bio_readpage_error handles errors the
2399 * following way: If possible, new read requests are
2400 * created and submitted and will end up in
2401 * end_bio_extent_readpage as well (if we're lucky, not
2402 * in the !uptodate case). In that case it returns 0 and
2403 * we just go on with the next page in our bio. If it
2404 * can't handle the error it will return -EIO and we
2405 * remain responsible for that page.
2407 ret = bio_readpage_error(bio, page, start, end, mirror, NULL);
2410 test_bit(BIO_UPTODATE, &bio->bi_flags);
2413 uncache_state(&cached);
2418 if (uptodate && tree->track_uptodate) {
2419 set_extent_uptodate(tree, start, end, &cached,
2422 unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC);
2426 SetPageUptodate(page);
2428 ClearPageUptodate(page);
2434 check_page_uptodate(tree, page);
2436 ClearPageUptodate(page);
2439 check_page_locked(tree, page);
2441 } while (bvec <= bvec_end);
2447 btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
2452 bio = bio_alloc(gfp_flags, nr_vecs);
2454 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
2455 while (!bio && (nr_vecs /= 2))
2456 bio = bio_alloc(gfp_flags, nr_vecs);
2461 bio->bi_bdev = bdev;
2462 bio->bi_sector = first_sector;
2467 static int __must_check submit_one_bio(int rw, struct bio *bio,
2468 int mirror_num, unsigned long bio_flags)
2471 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2472 struct page *page = bvec->bv_page;
2473 struct extent_io_tree *tree = bio->bi_private;
2476 start = page_offset(page) + bvec->bv_offset;
2478 bio->bi_private = NULL;
2482 if (tree->ops && tree->ops->submit_bio_hook)
2483 ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
2484 mirror_num, bio_flags, start);
2486 btrfsic_submit_bio(rw, bio);
2488 if (bio_flagged(bio, BIO_EOPNOTSUPP))
2494 static int merge_bio(int rw, struct extent_io_tree *tree, struct page *page,
2495 unsigned long offset, size_t size, struct bio *bio,
2496 unsigned long bio_flags)
2499 if (tree->ops && tree->ops->merge_bio_hook)
2500 ret = tree->ops->merge_bio_hook(rw, page, offset, size, bio,
2507 static int submit_extent_page(int rw, struct extent_io_tree *tree,
2508 struct page *page, sector_t sector,
2509 size_t size, unsigned long offset,
2510 struct block_device *bdev,
2511 struct bio **bio_ret,
2512 unsigned long max_pages,
2513 bio_end_io_t end_io_func,
2515 unsigned long prev_bio_flags,
2516 unsigned long bio_flags)
2522 int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED;
2523 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
2524 size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE);
2526 if (bio_ret && *bio_ret) {
2529 contig = bio->bi_sector == sector;
2531 contig = bio->bi_sector + (bio->bi_size >> 9) ==
2534 if (prev_bio_flags != bio_flags || !contig ||
2535 merge_bio(rw, tree, page, offset, page_size, bio, bio_flags) ||
2536 bio_add_page(bio, page, page_size, offset) < page_size) {
2537 ret = submit_one_bio(rw, bio, mirror_num,
2546 if (this_compressed)
2549 nr = bio_get_nr_vecs(bdev);
2551 bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
2555 bio_add_page(bio, page, page_size, offset);
2556 bio->bi_end_io = end_io_func;
2557 bio->bi_private = tree;
2562 ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
2567 void attach_extent_buffer_page(struct extent_buffer *eb, struct page *page)
2569 if (!PagePrivate(page)) {
2570 SetPagePrivate(page);
2571 page_cache_get(page);
2572 set_page_private(page, (unsigned long)eb);
2574 WARN_ON(page->private != (unsigned long)eb);
2578 void set_page_extent_mapped(struct page *page)
2580 if (!PagePrivate(page)) {
2581 SetPagePrivate(page);
2582 page_cache_get(page);
2583 set_page_private(page, EXTENT_PAGE_PRIVATE);
2588 * basic readpage implementation. Locked extent state structs are inserted
2589 * into the tree that are removed when the IO is done (by the end_io
2591 * XXX JDM: This needs looking at to ensure proper page locking
2593 static int __extent_read_full_page(struct extent_io_tree *tree,
2595 get_extent_t *get_extent,
2596 struct bio **bio, int mirror_num,
2597 unsigned long *bio_flags)
2599 struct inode *inode = page->mapping->host;
2600 u64 start = page_offset(page);
2601 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2605 u64 last_byte = i_size_read(inode);
2609 struct extent_map *em;
2610 struct block_device *bdev;
2611 struct btrfs_ordered_extent *ordered;
2614 size_t pg_offset = 0;
2616 size_t disk_io_size;
2617 size_t blocksize = inode->i_sb->s_blocksize;
2618 unsigned long this_bio_flag = 0;
2620 set_page_extent_mapped(page);
2622 if (!PageUptodate(page)) {
2623 if (cleancache_get_page(page) == 0) {
2624 BUG_ON(blocksize != PAGE_SIZE);
2631 lock_extent(tree, start, end);
2632 ordered = btrfs_lookup_ordered_extent(inode, start);
2635 unlock_extent(tree, start, end);
2636 btrfs_start_ordered_extent(inode, ordered, 1);
2637 btrfs_put_ordered_extent(ordered);
2640 if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
2642 size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
2645 iosize = PAGE_CACHE_SIZE - zero_offset;
2646 userpage = kmap_atomic(page);
2647 memset(userpage + zero_offset, 0, iosize);
2648 flush_dcache_page(page);
2649 kunmap_atomic(userpage);
2652 while (cur <= end) {
2653 unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
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 = ALIGN(iosize, blocksize);
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 ret = submit_extent_page(READ, tree, page,
2744 sector, disk_io_size, pg_offset,
2746 end_bio_extent_readpage, mirror_num,
2751 *bio_flags = this_bio_flag;
2754 unlock_extent(tree, cur, cur + iosize - 1);
2757 pg_offset += iosize;
2761 if (!PageError(page))
2762 SetPageUptodate(page);
2768 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
2769 get_extent_t *get_extent, int mirror_num)
2771 struct bio *bio = NULL;
2772 unsigned long bio_flags = 0;
2775 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
2778 ret = submit_one_bio(READ, bio, mirror_num, bio_flags);
2782 static noinline void update_nr_written(struct page *page,
2783 struct writeback_control *wbc,
2784 unsigned long nr_written)
2786 wbc->nr_to_write -= nr_written;
2787 if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
2788 wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
2789 page->mapping->writeback_index = page->index + nr_written;
2793 * the writepage semantics are similar to regular writepage. extent
2794 * records are inserted to lock ranges in the tree, and as dirty areas
2795 * are found, they are marked writeback. Then the lock bits are removed
2796 * and the end_io handler clears the writeback ranges
2798 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2801 struct inode *inode = page->mapping->host;
2802 struct extent_page_data *epd = data;
2803 struct extent_io_tree *tree = epd->tree;
2804 u64 start = page_offset(page);
2806 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2810 u64 last_byte = i_size_read(inode);
2814 struct extent_state *cached_state = NULL;
2815 struct extent_map *em;
2816 struct block_device *bdev;
2819 size_t pg_offset = 0;
2821 loff_t i_size = i_size_read(inode);
2822 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
2828 unsigned long nr_written = 0;
2829 bool fill_delalloc = true;
2831 if (wbc->sync_mode == WB_SYNC_ALL)
2832 write_flags = WRITE_SYNC;
2834 write_flags = WRITE;
2836 trace___extent_writepage(page, inode, wbc);
2838 WARN_ON(!PageLocked(page));
2840 ClearPageError(page);
2842 pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
2843 if (page->index > end_index ||
2844 (page->index == end_index && !pg_offset)) {
2845 page->mapping->a_ops->invalidatepage(page, 0);
2850 if (page->index == end_index) {
2853 userpage = kmap_atomic(page);
2854 memset(userpage + pg_offset, 0,
2855 PAGE_CACHE_SIZE - pg_offset);
2856 kunmap_atomic(userpage);
2857 flush_dcache_page(page);
2861 set_page_extent_mapped(page);
2863 if (!tree->ops || !tree->ops->fill_delalloc)
2864 fill_delalloc = false;
2866 delalloc_start = start;
2869 if (!epd->extent_locked && fill_delalloc) {
2870 u64 delalloc_to_write = 0;
2872 * make sure the wbc mapping index is at least updated
2875 update_nr_written(page, wbc, 0);
2877 while (delalloc_end < page_end) {
2878 nr_delalloc = find_lock_delalloc_range(inode, tree,
2883 if (nr_delalloc == 0) {
2884 delalloc_start = delalloc_end + 1;
2887 ret = tree->ops->fill_delalloc(inode, page,
2892 /* File system has been set read-only */
2898 * delalloc_end is already one less than the total
2899 * length, so we don't subtract one from
2902 delalloc_to_write += (delalloc_end - delalloc_start +
2905 delalloc_start = delalloc_end + 1;
2907 if (wbc->nr_to_write < delalloc_to_write) {
2910 if (delalloc_to_write < thresh * 2)
2911 thresh = delalloc_to_write;
2912 wbc->nr_to_write = min_t(u64, delalloc_to_write,
2916 /* did the fill delalloc function already unlock and start
2922 * we've unlocked the page, so we can't update
2923 * the mapping's writeback index, just update
2926 wbc->nr_to_write -= nr_written;
2930 if (tree->ops && tree->ops->writepage_start_hook) {
2931 ret = tree->ops->writepage_start_hook(page, start,
2934 /* Fixup worker will requeue */
2936 wbc->pages_skipped++;
2938 redirty_page_for_writepage(wbc, page);
2939 update_nr_written(page, wbc, nr_written);
2947 * we don't want to touch the inode after unlocking the page,
2948 * so we update the mapping writeback index now
2950 update_nr_written(page, wbc, nr_written + 1);
2953 if (last_byte <= start) {
2954 if (tree->ops && tree->ops->writepage_end_io_hook)
2955 tree->ops->writepage_end_io_hook(page, start,
2960 blocksize = inode->i_sb->s_blocksize;
2962 while (cur <= end) {
2963 if (cur >= last_byte) {
2964 if (tree->ops && tree->ops->writepage_end_io_hook)
2965 tree->ops->writepage_end_io_hook(page, cur,
2969 em = epd->get_extent(inode, page, pg_offset, cur,
2971 if (IS_ERR_OR_NULL(em)) {
2976 extent_offset = cur - em->start;
2977 BUG_ON(extent_map_end(em) <= cur);
2979 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2980 iosize = ALIGN(iosize, blocksize);
2981 sector = (em->block_start + extent_offset) >> 9;
2983 block_start = em->block_start;
2984 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
2985 free_extent_map(em);
2989 * compressed and inline extents are written through other
2992 if (compressed || block_start == EXTENT_MAP_HOLE ||
2993 block_start == EXTENT_MAP_INLINE) {
2995 * end_io notification does not happen here for
2996 * compressed extents
2998 if (!compressed && tree->ops &&
2999 tree->ops->writepage_end_io_hook)
3000 tree->ops->writepage_end_io_hook(page, cur,
3003 else if (compressed) {
3004 /* we don't want to end_page_writeback on
3005 * a compressed extent. this happens
3012 pg_offset += iosize;
3015 /* leave this out until we have a page_mkwrite call */
3016 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
3017 EXTENT_DIRTY, 0, NULL)) {
3019 pg_offset += iosize;
3023 if (tree->ops && tree->ops->writepage_io_hook) {
3024 ret = tree->ops->writepage_io_hook(page, cur,
3032 unsigned long max_nr = end_index + 1;
3034 set_range_writeback(tree, cur, cur + iosize - 1);
3035 if (!PageWriteback(page)) {
3036 printk(KERN_ERR "btrfs warning page %lu not "
3037 "writeback, cur %llu end %llu\n",
3038 page->index, (unsigned long long)cur,
3039 (unsigned long long)end);
3042 ret = submit_extent_page(write_flags, tree, page,
3043 sector, iosize, pg_offset,
3044 bdev, &epd->bio, max_nr,
3045 end_bio_extent_writepage,
3051 pg_offset += iosize;
3056 /* make sure the mapping tag for page dirty gets cleared */
3057 set_page_writeback(page);
3058 end_page_writeback(page);
3064 /* drop our reference on any cached states */
3065 free_extent_state(cached_state);
3069 static int eb_wait(void *word)
3075 static void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3077 wait_on_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK, eb_wait,
3078 TASK_UNINTERRUPTIBLE);
3081 static int lock_extent_buffer_for_io(struct extent_buffer *eb,
3082 struct btrfs_fs_info *fs_info,
3083 struct extent_page_data *epd)
3085 unsigned long i, num_pages;
3089 if (!btrfs_try_tree_write_lock(eb)) {
3091 flush_write_bio(epd);
3092 btrfs_tree_lock(eb);
3095 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3096 btrfs_tree_unlock(eb);
3100 flush_write_bio(epd);
3104 wait_on_extent_buffer_writeback(eb);
3105 btrfs_tree_lock(eb);
3106 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3108 btrfs_tree_unlock(eb);
3113 * We need to do this to prevent races in people who check if the eb is
3114 * under IO since we can end up having no IO bits set for a short period
3117 spin_lock(&eb->refs_lock);
3118 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3119 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3120 spin_unlock(&eb->refs_lock);
3121 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3122 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
3124 fs_info->dirty_metadata_batch);
3127 spin_unlock(&eb->refs_lock);
3130 btrfs_tree_unlock(eb);
3135 num_pages = num_extent_pages(eb->start, eb->len);
3136 for (i = 0; i < num_pages; i++) {
3137 struct page *p = extent_buffer_page(eb, i);
3139 if (!trylock_page(p)) {
3141 flush_write_bio(epd);
3151 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3153 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3154 smp_mb__after_clear_bit();
3155 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3158 static void end_bio_extent_buffer_writepage(struct bio *bio, int err)
3160 int uptodate = err == 0;
3161 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
3162 struct extent_buffer *eb;
3166 struct page *page = bvec->bv_page;
3169 eb = (struct extent_buffer *)page->private;
3171 done = atomic_dec_and_test(&eb->io_pages);
3173 if (!uptodate || test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
3174 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3175 ClearPageUptodate(page);
3179 end_page_writeback(page);
3184 end_extent_buffer_writeback(eb);
3185 } while (bvec >= bio->bi_io_vec);
3191 static int write_one_eb(struct extent_buffer *eb,
3192 struct btrfs_fs_info *fs_info,
3193 struct writeback_control *wbc,
3194 struct extent_page_data *epd)
3196 struct block_device *bdev = fs_info->fs_devices->latest_bdev;
3197 u64 offset = eb->start;
3198 unsigned long i, num_pages;
3199 unsigned long bio_flags = 0;
3200 int rw = (epd->sync_io ? WRITE_SYNC : WRITE);
3203 clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3204 num_pages = num_extent_pages(eb->start, eb->len);
3205 atomic_set(&eb->io_pages, num_pages);
3206 if (btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID)
3207 bio_flags = EXTENT_BIO_TREE_LOG;
3209 for (i = 0; i < num_pages; i++) {
3210 struct page *p = extent_buffer_page(eb, i);
3212 clear_page_dirty_for_io(p);
3213 set_page_writeback(p);
3214 ret = submit_extent_page(rw, eb->tree, p, offset >> 9,
3215 PAGE_CACHE_SIZE, 0, bdev, &epd->bio,
3216 -1, end_bio_extent_buffer_writepage,
3217 0, epd->bio_flags, bio_flags);
3218 epd->bio_flags = bio_flags;
3220 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3222 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3223 end_extent_buffer_writeback(eb);
3227 offset += PAGE_CACHE_SIZE;
3228 update_nr_written(p, wbc, 1);
3232 if (unlikely(ret)) {
3233 for (; i < num_pages; i++) {
3234 struct page *p = extent_buffer_page(eb, i);
3242 int btree_write_cache_pages(struct address_space *mapping,
3243 struct writeback_control *wbc)
3245 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
3246 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
3247 struct extent_buffer *eb, *prev_eb = NULL;
3248 struct extent_page_data epd = {
3252 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3257 int nr_to_write_done = 0;
3258 struct pagevec pvec;
3261 pgoff_t end; /* Inclusive */
3265 pagevec_init(&pvec, 0);
3266 if (wbc->range_cyclic) {
3267 index = mapping->writeback_index; /* Start from prev offset */
3270 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3271 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3274 if (wbc->sync_mode == WB_SYNC_ALL)
3275 tag = PAGECACHE_TAG_TOWRITE;
3277 tag = PAGECACHE_TAG_DIRTY;
3279 if (wbc->sync_mode == WB_SYNC_ALL)
3280 tag_pages_for_writeback(mapping, index, end);
3281 while (!done && !nr_to_write_done && (index <= end) &&
3282 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3283 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3287 for (i = 0; i < nr_pages; i++) {
3288 struct page *page = pvec.pages[i];
3290 if (!PagePrivate(page))
3293 if (!wbc->range_cyclic && page->index > end) {
3298 spin_lock(&mapping->private_lock);
3299 if (!PagePrivate(page)) {
3300 spin_unlock(&mapping->private_lock);
3304 eb = (struct extent_buffer *)page->private;
3307 * Shouldn't happen and normally this would be a BUG_ON
3308 * but no sense in crashing the users box for something
3309 * we can survive anyway.
3312 spin_unlock(&mapping->private_lock);
3317 if (eb == prev_eb) {
3318 spin_unlock(&mapping->private_lock);
3322 ret = atomic_inc_not_zero(&eb->refs);
3323 spin_unlock(&mapping->private_lock);
3328 ret = lock_extent_buffer_for_io(eb, fs_info, &epd);
3330 free_extent_buffer(eb);
3334 ret = write_one_eb(eb, fs_info, wbc, &epd);
3337 free_extent_buffer(eb);
3340 free_extent_buffer(eb);
3343 * the filesystem may choose to bump up nr_to_write.
3344 * We have to make sure to honor the new nr_to_write
3347 nr_to_write_done = wbc->nr_to_write <= 0;
3349 pagevec_release(&pvec);
3352 if (!scanned && !done) {
3354 * We hit the last page and there is more work to be done: wrap
3355 * back to the start of the file
3361 flush_write_bio(&epd);
3366 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3367 * @mapping: address space structure to write
3368 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3369 * @writepage: function called for each page
3370 * @data: data passed to writepage function
3372 * If a page is already under I/O, write_cache_pages() skips it, even
3373 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3374 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3375 * and msync() need to guarantee that all the data which was dirty at the time
3376 * the call was made get new I/O started against them. If wbc->sync_mode is
3377 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3378 * existing IO to complete.
3380 static int extent_write_cache_pages(struct extent_io_tree *tree,
3381 struct address_space *mapping,
3382 struct writeback_control *wbc,
3383 writepage_t writepage, void *data,
3384 void (*flush_fn)(void *))
3386 struct inode *inode = mapping->host;
3389 int nr_to_write_done = 0;
3390 struct pagevec pvec;
3393 pgoff_t end; /* Inclusive */
3398 * We have to hold onto the inode so that ordered extents can do their
3399 * work when the IO finishes. The alternative to this is failing to add
3400 * an ordered extent if the igrab() fails there and that is a huge pain
3401 * to deal with, so instead just hold onto the inode throughout the
3402 * writepages operation. If it fails here we are freeing up the inode
3403 * anyway and we'd rather not waste our time writing out stuff that is
3404 * going to be truncated anyway.
3409 pagevec_init(&pvec, 0);
3410 if (wbc->range_cyclic) {
3411 index = mapping->writeback_index; /* Start from prev offset */
3414 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3415 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3418 if (wbc->sync_mode == WB_SYNC_ALL)
3419 tag = PAGECACHE_TAG_TOWRITE;
3421 tag = PAGECACHE_TAG_DIRTY;
3423 if (wbc->sync_mode == WB_SYNC_ALL)
3424 tag_pages_for_writeback(mapping, index, end);
3425 while (!done && !nr_to_write_done && (index <= end) &&
3426 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3427 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3431 for (i = 0; i < nr_pages; i++) {
3432 struct page *page = pvec.pages[i];
3435 * At this point we hold neither mapping->tree_lock nor
3436 * lock on the page itself: the page may be truncated or
3437 * invalidated (changing page->mapping to NULL), or even
3438 * swizzled back from swapper_space to tmpfs file
3441 if (!trylock_page(page)) {
3446 if (unlikely(page->mapping != mapping)) {
3451 if (!wbc->range_cyclic && page->index > end) {
3457 if (wbc->sync_mode != WB_SYNC_NONE) {
3458 if (PageWriteback(page))
3460 wait_on_page_writeback(page);
3463 if (PageWriteback(page) ||
3464 !clear_page_dirty_for_io(page)) {
3469 ret = (*writepage)(page, wbc, data);
3471 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
3479 * the filesystem may choose to bump up nr_to_write.
3480 * We have to make sure to honor the new nr_to_write
3483 nr_to_write_done = wbc->nr_to_write <= 0;
3485 pagevec_release(&pvec);
3488 if (!scanned && !done) {
3490 * We hit the last page and there is more work to be done: wrap
3491 * back to the start of the file
3497 btrfs_add_delayed_iput(inode);
3501 static void flush_epd_write_bio(struct extent_page_data *epd)
3510 ret = submit_one_bio(rw, epd->bio, 0, epd->bio_flags);
3511 BUG_ON(ret < 0); /* -ENOMEM */
3516 static noinline void flush_write_bio(void *data)
3518 struct extent_page_data *epd = data;
3519 flush_epd_write_bio(epd);
3522 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
3523 get_extent_t *get_extent,
3524 struct writeback_control *wbc)
3527 struct extent_page_data epd = {
3530 .get_extent = get_extent,
3532 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3536 ret = __extent_writepage(page, wbc, &epd);
3538 flush_epd_write_bio(&epd);
3542 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
3543 u64 start, u64 end, get_extent_t *get_extent,
3547 struct address_space *mapping = inode->i_mapping;
3549 unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
3552 struct extent_page_data epd = {
3555 .get_extent = get_extent,
3557 .sync_io = mode == WB_SYNC_ALL,
3560 struct writeback_control wbc_writepages = {
3562 .nr_to_write = nr_pages * 2,
3563 .range_start = start,
3564 .range_end = end + 1,
3567 while (start <= end) {
3568 page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
3569 if (clear_page_dirty_for_io(page))
3570 ret = __extent_writepage(page, &wbc_writepages, &epd);
3572 if (tree->ops && tree->ops->writepage_end_io_hook)
3573 tree->ops->writepage_end_io_hook(page, start,
3574 start + PAGE_CACHE_SIZE - 1,
3578 page_cache_release(page);
3579 start += PAGE_CACHE_SIZE;
3582 flush_epd_write_bio(&epd);
3586 int extent_writepages(struct extent_io_tree *tree,
3587 struct address_space *mapping,
3588 get_extent_t *get_extent,
3589 struct writeback_control *wbc)
3592 struct extent_page_data epd = {
3595 .get_extent = get_extent,
3597 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3601 ret = extent_write_cache_pages(tree, mapping, wbc,
3602 __extent_writepage, &epd,
3604 flush_epd_write_bio(&epd);
3608 int extent_readpages(struct extent_io_tree *tree,
3609 struct address_space *mapping,
3610 struct list_head *pages, unsigned nr_pages,
3611 get_extent_t get_extent)
3613 struct bio *bio = NULL;
3615 unsigned long bio_flags = 0;
3616 struct page *pagepool[16];
3621 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
3622 page = list_entry(pages->prev, struct page, lru);
3624 prefetchw(&page->flags);
3625 list_del(&page->lru);
3626 if (add_to_page_cache_lru(page, mapping,
3627 page->index, GFP_NOFS)) {
3628 page_cache_release(page);
3632 pagepool[nr++] = page;
3633 if (nr < ARRAY_SIZE(pagepool))
3635 for (i = 0; i < nr; i++) {
3636 __extent_read_full_page(tree, pagepool[i], get_extent,
3637 &bio, 0, &bio_flags);
3638 page_cache_release(pagepool[i]);
3642 for (i = 0; i < nr; i++) {
3643 __extent_read_full_page(tree, pagepool[i], get_extent,
3644 &bio, 0, &bio_flags);
3645 page_cache_release(pagepool[i]);
3648 BUG_ON(!list_empty(pages));
3650 return submit_one_bio(READ, bio, 0, bio_flags);
3655 * basic invalidatepage code, this waits on any locked or writeback
3656 * ranges corresponding to the page, and then deletes any extent state
3657 * records from the tree
3659 int extent_invalidatepage(struct extent_io_tree *tree,
3660 struct page *page, unsigned long offset)
3662 struct extent_state *cached_state = NULL;
3663 u64 start = page_offset(page);
3664 u64 end = start + PAGE_CACHE_SIZE - 1;
3665 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
3667 start += ALIGN(offset, blocksize);
3671 lock_extent_bits(tree, start, end, 0, &cached_state);
3672 wait_on_page_writeback(page);
3673 clear_extent_bit(tree, start, end,
3674 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
3675 EXTENT_DO_ACCOUNTING,
3676 1, 1, &cached_state, GFP_NOFS);
3681 * a helper for releasepage, this tests for areas of the page that
3682 * are locked or under IO and drops the related state bits if it is safe
3685 int try_release_extent_state(struct extent_map_tree *map,
3686 struct extent_io_tree *tree, struct page *page,
3689 u64 start = page_offset(page);
3690 u64 end = start + PAGE_CACHE_SIZE - 1;
3693 if (test_range_bit(tree, start, end,
3694 EXTENT_IOBITS, 0, NULL))
3697 if ((mask & GFP_NOFS) == GFP_NOFS)
3700 * at this point we can safely clear everything except the
3701 * locked bit and the nodatasum bit
3703 ret = clear_extent_bit(tree, start, end,
3704 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
3707 /* if clear_extent_bit failed for enomem reasons,
3708 * we can't allow the release to continue.
3719 * a helper for releasepage. As long as there are no locked extents
3720 * in the range corresponding to the page, both state records and extent
3721 * map records are removed
3723 int try_release_extent_mapping(struct extent_map_tree *map,
3724 struct extent_io_tree *tree, struct page *page,
3727 struct extent_map *em;
3728 u64 start = page_offset(page);
3729 u64 end = start + PAGE_CACHE_SIZE - 1;
3731 if ((mask & __GFP_WAIT) &&
3732 page->mapping->host->i_size > 16 * 1024 * 1024) {
3734 while (start <= end) {
3735 len = end - start + 1;
3736 write_lock(&map->lock);
3737 em = lookup_extent_mapping(map, start, len);
3739 write_unlock(&map->lock);
3742 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
3743 em->start != start) {
3744 write_unlock(&map->lock);
3745 free_extent_map(em);
3748 if (!test_range_bit(tree, em->start,
3749 extent_map_end(em) - 1,
3750 EXTENT_LOCKED | EXTENT_WRITEBACK,
3752 remove_extent_mapping(map, em);
3753 /* once for the rb tree */
3754 free_extent_map(em);
3756 start = extent_map_end(em);
3757 write_unlock(&map->lock);
3760 free_extent_map(em);
3763 return try_release_extent_state(map, tree, page, mask);
3767 * helper function for fiemap, which doesn't want to see any holes.
3768 * This maps until we find something past 'last'
3770 static struct extent_map *get_extent_skip_holes(struct inode *inode,
3773 get_extent_t *get_extent)
3775 u64 sectorsize = BTRFS_I(inode)->root->sectorsize;
3776 struct extent_map *em;
3783 len = last - offset;
3786 len = ALIGN(len, sectorsize);
3787 em = get_extent(inode, NULL, 0, offset, len, 0);
3788 if (IS_ERR_OR_NULL(em))
3791 /* if this isn't a hole return it */
3792 if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) &&
3793 em->block_start != EXTENT_MAP_HOLE) {
3797 /* this is a hole, advance to the next extent */
3798 offset = extent_map_end(em);
3799 free_extent_map(em);
3806 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
3807 __u64 start, __u64 len, get_extent_t *get_extent)
3811 u64 max = start + len;
3815 u64 last_for_get_extent = 0;
3817 u64 isize = i_size_read(inode);
3818 struct btrfs_key found_key;
3819 struct extent_map *em = NULL;
3820 struct extent_state *cached_state = NULL;
3821 struct btrfs_path *path;
3822 struct btrfs_file_extent_item *item;
3827 unsigned long emflags;
3832 path = btrfs_alloc_path();
3835 path->leave_spinning = 1;
3837 start = ALIGN(start, BTRFS_I(inode)->root->sectorsize);
3838 len = ALIGN(len, BTRFS_I(inode)->root->sectorsize);
3841 * lookup the last file extent. We're not using i_size here
3842 * because there might be preallocation past i_size
3844 ret = btrfs_lookup_file_extent(NULL, BTRFS_I(inode)->root,
3845 path, btrfs_ino(inode), -1, 0);
3847 btrfs_free_path(path);
3852 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3853 struct btrfs_file_extent_item);
3854 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
3855 found_type = btrfs_key_type(&found_key);
3857 /* No extents, but there might be delalloc bits */
3858 if (found_key.objectid != btrfs_ino(inode) ||
3859 found_type != BTRFS_EXTENT_DATA_KEY) {
3860 /* have to trust i_size as the end */
3862 last_for_get_extent = isize;
3865 * remember the start of the last extent. There are a
3866 * bunch of different factors that go into the length of the
3867 * extent, so its much less complex to remember where it started
3869 last = found_key.offset;
3870 last_for_get_extent = last + 1;
3872 btrfs_free_path(path);
3875 * we might have some extents allocated but more delalloc past those
3876 * extents. so, we trust isize unless the start of the last extent is
3881 last_for_get_extent = isize;
3884 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len, 0,
3887 em = get_extent_skip_holes(inode, start, last_for_get_extent,
3897 u64 offset_in_extent;
3899 /* break if the extent we found is outside the range */
3900 if (em->start >= max || extent_map_end(em) < off)
3904 * get_extent may return an extent that starts before our
3905 * requested range. We have to make sure the ranges
3906 * we return to fiemap always move forward and don't
3907 * overlap, so adjust the offsets here
3909 em_start = max(em->start, off);
3912 * record the offset from the start of the extent
3913 * for adjusting the disk offset below
3915 offset_in_extent = em_start - em->start;
3916 em_end = extent_map_end(em);
3917 em_len = em_end - em_start;
3918 emflags = em->flags;
3923 * bump off for our next call to get_extent
3925 off = extent_map_end(em);
3929 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
3931 flags |= FIEMAP_EXTENT_LAST;
3932 } else if (em->block_start == EXTENT_MAP_INLINE) {
3933 flags |= (FIEMAP_EXTENT_DATA_INLINE |
3934 FIEMAP_EXTENT_NOT_ALIGNED);
3935 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
3936 flags |= (FIEMAP_EXTENT_DELALLOC |
3937 FIEMAP_EXTENT_UNKNOWN);
3939 disko = em->block_start + offset_in_extent;
3941 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
3942 flags |= FIEMAP_EXTENT_ENCODED;
3944 free_extent_map(em);
3946 if ((em_start >= last) || em_len == (u64)-1 ||
3947 (last == (u64)-1 && isize <= em_end)) {
3948 flags |= FIEMAP_EXTENT_LAST;
3952 /* now scan forward to see if this is really the last extent. */
3953 em = get_extent_skip_holes(inode, off, last_for_get_extent,
3960 flags |= FIEMAP_EXTENT_LAST;
3963 ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
3969 free_extent_map(em);
3971 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len,
3972 &cached_state, GFP_NOFS);
3976 static void __free_extent_buffer(struct extent_buffer *eb)
3979 unsigned long flags;
3980 spin_lock_irqsave(&leak_lock, flags);
3981 list_del(&eb->leak_list);
3982 spin_unlock_irqrestore(&leak_lock, flags);
3984 kmem_cache_free(extent_buffer_cache, eb);
3987 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
3992 struct extent_buffer *eb = NULL;
3994 unsigned long flags;
3997 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
4004 rwlock_init(&eb->lock);
4005 atomic_set(&eb->write_locks, 0);
4006 atomic_set(&eb->read_locks, 0);
4007 atomic_set(&eb->blocking_readers, 0);
4008 atomic_set(&eb->blocking_writers, 0);
4009 atomic_set(&eb->spinning_readers, 0);
4010 atomic_set(&eb->spinning_writers, 0);
4011 eb->lock_nested = 0;
4012 init_waitqueue_head(&eb->write_lock_wq);
4013 init_waitqueue_head(&eb->read_lock_wq);
4016 spin_lock_irqsave(&leak_lock, flags);
4017 list_add(&eb->leak_list, &buffers);
4018 spin_unlock_irqrestore(&leak_lock, flags);
4020 spin_lock_init(&eb->refs_lock);
4021 atomic_set(&eb->refs, 1);
4022 atomic_set(&eb->io_pages, 0);
4025 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4027 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4028 > MAX_INLINE_EXTENT_BUFFER_SIZE);
4029 BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE);
4034 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src)
4038 struct extent_buffer *new;
4039 unsigned long num_pages = num_extent_pages(src->start, src->len);
4041 new = __alloc_extent_buffer(NULL, src->start, src->len, GFP_ATOMIC);
4045 for (i = 0; i < num_pages; i++) {
4046 p = alloc_page(GFP_ATOMIC);
4048 attach_extent_buffer_page(new, p);
4049 WARN_ON(PageDirty(p));
4054 copy_extent_buffer(new, src, 0, 0, src->len);
4055 set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
4056 set_bit(EXTENT_BUFFER_DUMMY, &new->bflags);
4061 struct extent_buffer *alloc_dummy_extent_buffer(u64 start, unsigned long len)
4063 struct extent_buffer *eb;
4064 unsigned long num_pages = num_extent_pages(0, len);
4067 eb = __alloc_extent_buffer(NULL, start, len, GFP_ATOMIC);
4071 for (i = 0; i < num_pages; i++) {
4072 eb->pages[i] = alloc_page(GFP_ATOMIC);
4076 set_extent_buffer_uptodate(eb);
4077 btrfs_set_header_nritems(eb, 0);
4078 set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4083 __free_page(eb->pages[i - 1]);
4084 __free_extent_buffer(eb);
4088 static int extent_buffer_under_io(struct extent_buffer *eb)
4090 return (atomic_read(&eb->io_pages) ||
4091 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4092 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4096 * Helper for releasing extent buffer page.
4098 static void btrfs_release_extent_buffer_page(struct extent_buffer *eb,
4099 unsigned long start_idx)
4101 unsigned long index;
4102 unsigned long num_pages;
4104 int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4106 BUG_ON(extent_buffer_under_io(eb));
4108 num_pages = num_extent_pages(eb->start, eb->len);
4109 index = start_idx + num_pages;
4110 if (start_idx >= index)
4115 page = extent_buffer_page(eb, index);
4116 if (page && mapped) {
4117 spin_lock(&page->mapping->private_lock);
4119 * We do this since we'll remove the pages after we've
4120 * removed the eb from the radix tree, so we could race
4121 * and have this page now attached to the new eb. So
4122 * only clear page_private if it's still connected to
4125 if (PagePrivate(page) &&
4126 page->private == (unsigned long)eb) {
4127 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4128 BUG_ON(PageDirty(page));
4129 BUG_ON(PageWriteback(page));
4131 * We need to make sure we haven't be attached
4134 ClearPagePrivate(page);
4135 set_page_private(page, 0);
4136 /* One for the page private */
4137 page_cache_release(page);
4139 spin_unlock(&page->mapping->private_lock);
4143 /* One for when we alloced the page */
4144 page_cache_release(page);
4146 } while (index != start_idx);
4150 * Helper for releasing the extent buffer.
4152 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4154 btrfs_release_extent_buffer_page(eb, 0);
4155 __free_extent_buffer(eb);
4158 static void check_buffer_tree_ref(struct extent_buffer *eb)
4161 /* the ref bit is tricky. We have to make sure it is set
4162 * if we have the buffer dirty. Otherwise the
4163 * code to free a buffer can end up dropping a dirty
4166 * Once the ref bit is set, it won't go away while the
4167 * buffer is dirty or in writeback, and it also won't
4168 * go away while we have the reference count on the
4171 * We can't just set the ref bit without bumping the
4172 * ref on the eb because free_extent_buffer might
4173 * see the ref bit and try to clear it. If this happens
4174 * free_extent_buffer might end up dropping our original
4175 * ref by mistake and freeing the page before we are able
4176 * to add one more ref.
4178 * So bump the ref count first, then set the bit. If someone
4179 * beat us to it, drop the ref we added.
4181 refs = atomic_read(&eb->refs);
4182 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4185 spin_lock(&eb->refs_lock);
4186 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4187 atomic_inc(&eb->refs);
4188 spin_unlock(&eb->refs_lock);
4191 static void mark_extent_buffer_accessed(struct extent_buffer *eb)
4193 unsigned long num_pages, i;
4195 check_buffer_tree_ref(eb);
4197 num_pages = num_extent_pages(eb->start, eb->len);
4198 for (i = 0; i < num_pages; i++) {
4199 struct page *p = extent_buffer_page(eb, i);
4200 mark_page_accessed(p);
4204 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
4205 u64 start, unsigned long len)
4207 unsigned long num_pages = num_extent_pages(start, len);
4209 unsigned long index = start >> PAGE_CACHE_SHIFT;
4210 struct extent_buffer *eb;
4211 struct extent_buffer *exists = NULL;
4213 struct address_space *mapping = tree->mapping;
4218 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
4219 if (eb && atomic_inc_not_zero(&eb->refs)) {
4221 mark_extent_buffer_accessed(eb);
4226 eb = __alloc_extent_buffer(tree, start, len, GFP_NOFS);
4230 for (i = 0; i < num_pages; i++, index++) {
4231 p = find_or_create_page(mapping, index, GFP_NOFS);
4235 spin_lock(&mapping->private_lock);
4236 if (PagePrivate(p)) {
4238 * We could have already allocated an eb for this page
4239 * and attached one so lets see if we can get a ref on
4240 * the existing eb, and if we can we know it's good and
4241 * we can just return that one, else we know we can just
4242 * overwrite page->private.
4244 exists = (struct extent_buffer *)p->private;
4245 if (atomic_inc_not_zero(&exists->refs)) {
4246 spin_unlock(&mapping->private_lock);
4248 page_cache_release(p);
4249 mark_extent_buffer_accessed(exists);
4254 * Do this so attach doesn't complain and we need to
4255 * drop the ref the old guy had.
4257 ClearPagePrivate(p);
4258 WARN_ON(PageDirty(p));
4259 page_cache_release(p);
4261 attach_extent_buffer_page(eb, p);
4262 spin_unlock(&mapping->private_lock);
4263 WARN_ON(PageDirty(p));
4264 mark_page_accessed(p);
4266 if (!PageUptodate(p))
4270 * see below about how we avoid a nasty race with release page
4271 * and why we unlock later
4275 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4277 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
4281 spin_lock(&tree->buffer_lock);
4282 ret = radix_tree_insert(&tree->buffer, start >> PAGE_CACHE_SHIFT, eb);
4283 if (ret == -EEXIST) {
4284 exists = radix_tree_lookup(&tree->buffer,
4285 start >> PAGE_CACHE_SHIFT);
4286 if (!atomic_inc_not_zero(&exists->refs)) {
4287 spin_unlock(&tree->buffer_lock);
4288 radix_tree_preload_end();
4292 spin_unlock(&tree->buffer_lock);
4293 radix_tree_preload_end();
4294 mark_extent_buffer_accessed(exists);
4297 /* add one reference for the tree */
4298 check_buffer_tree_ref(eb);
4299 spin_unlock(&tree->buffer_lock);
4300 radix_tree_preload_end();
4303 * there is a race where release page may have
4304 * tried to find this extent buffer in the radix
4305 * but failed. It will tell the VM it is safe to
4306 * reclaim the, and it will clear the page private bit.
4307 * We must make sure to set the page private bit properly
4308 * after the extent buffer is in the radix tree so
4309 * it doesn't get lost
4311 SetPageChecked(eb->pages[0]);
4312 for (i = 1; i < num_pages; i++) {
4313 p = extent_buffer_page(eb, i);
4314 ClearPageChecked(p);
4317 unlock_page(eb->pages[0]);
4321 for (i = 0; i < num_pages; i++) {
4323 unlock_page(eb->pages[i]);
4326 WARN_ON(!atomic_dec_and_test(&eb->refs));
4327 btrfs_release_extent_buffer(eb);
4331 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
4332 u64 start, unsigned long len)
4334 struct extent_buffer *eb;
4337 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
4338 if (eb && atomic_inc_not_zero(&eb->refs)) {
4340 mark_extent_buffer_accessed(eb);
4348 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
4350 struct extent_buffer *eb =
4351 container_of(head, struct extent_buffer, rcu_head);
4353 __free_extent_buffer(eb);
4356 /* Expects to have eb->eb_lock already held */
4357 static int release_extent_buffer(struct extent_buffer *eb, gfp_t mask)
4359 WARN_ON(atomic_read(&eb->refs) == 0);
4360 if (atomic_dec_and_test(&eb->refs)) {
4361 if (test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags)) {
4362 spin_unlock(&eb->refs_lock);
4364 struct extent_io_tree *tree = eb->tree;
4366 spin_unlock(&eb->refs_lock);
4368 spin_lock(&tree->buffer_lock);
4369 radix_tree_delete(&tree->buffer,
4370 eb->start >> PAGE_CACHE_SHIFT);
4371 spin_unlock(&tree->buffer_lock);
4374 /* Should be safe to release our pages at this point */
4375 btrfs_release_extent_buffer_page(eb, 0);
4376 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
4379 spin_unlock(&eb->refs_lock);
4384 void free_extent_buffer(struct extent_buffer *eb)
4392 refs = atomic_read(&eb->refs);
4395 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
4400 spin_lock(&eb->refs_lock);
4401 if (atomic_read(&eb->refs) == 2 &&
4402 test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))
4403 atomic_dec(&eb->refs);
4405 if (atomic_read(&eb->refs) == 2 &&
4406 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
4407 !extent_buffer_under_io(eb) &&
4408 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4409 atomic_dec(&eb->refs);
4412 * I know this is terrible, but it's temporary until we stop tracking
4413 * the uptodate bits and such for the extent buffers.
4415 release_extent_buffer(eb, GFP_ATOMIC);
4418 void free_extent_buffer_stale(struct extent_buffer *eb)
4423 spin_lock(&eb->refs_lock);
4424 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
4426 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
4427 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4428 atomic_dec(&eb->refs);
4429 release_extent_buffer(eb, GFP_NOFS);
4432 void clear_extent_buffer_dirty(struct extent_buffer *eb)
4435 unsigned long num_pages;
4438 num_pages = num_extent_pages(eb->start, eb->len);
4440 for (i = 0; i < num_pages; i++) {
4441 page = extent_buffer_page(eb, i);
4442 if (!PageDirty(page))
4446 WARN_ON(!PagePrivate(page));
4448 clear_page_dirty_for_io(page);
4449 spin_lock_irq(&page->mapping->tree_lock);
4450 if (!PageDirty(page)) {
4451 radix_tree_tag_clear(&page->mapping->page_tree,
4453 PAGECACHE_TAG_DIRTY);
4455 spin_unlock_irq(&page->mapping->tree_lock);
4456 ClearPageError(page);
4459 WARN_ON(atomic_read(&eb->refs) == 0);
4462 int set_extent_buffer_dirty(struct extent_buffer *eb)
4465 unsigned long num_pages;
4468 check_buffer_tree_ref(eb);
4470 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
4472 num_pages = num_extent_pages(eb->start, eb->len);
4473 WARN_ON(atomic_read(&eb->refs) == 0);
4474 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
4476 for (i = 0; i < num_pages; i++)
4477 set_page_dirty(extent_buffer_page(eb, i));
4481 static int range_straddles_pages(u64 start, u64 len)
4483 if (len < PAGE_CACHE_SIZE)
4485 if (start & (PAGE_CACHE_SIZE - 1))
4487 if ((start + len) & (PAGE_CACHE_SIZE - 1))
4492 int clear_extent_buffer_uptodate(struct extent_buffer *eb)
4496 unsigned long num_pages;
4498 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4499 num_pages = num_extent_pages(eb->start, eb->len);
4500 for (i = 0; i < num_pages; i++) {
4501 page = extent_buffer_page(eb, i);
4503 ClearPageUptodate(page);
4508 int set_extent_buffer_uptodate(struct extent_buffer *eb)
4512 unsigned long num_pages;
4514 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4515 num_pages = num_extent_pages(eb->start, eb->len);
4516 for (i = 0; i < num_pages; i++) {
4517 page = extent_buffer_page(eb, i);
4518 SetPageUptodate(page);
4523 int extent_range_uptodate(struct extent_io_tree *tree,
4528 int pg_uptodate = 1;
4530 unsigned long index;
4532 if (range_straddles_pages(start, end - start + 1)) {
4533 ret = test_range_bit(tree, start, end,
4534 EXTENT_UPTODATE, 1, NULL);
4538 while (start <= end) {
4539 index = start >> PAGE_CACHE_SHIFT;
4540 page = find_get_page(tree->mapping, index);
4543 uptodate = PageUptodate(page);
4544 page_cache_release(page);
4549 start += PAGE_CACHE_SIZE;
4554 int extent_buffer_uptodate(struct extent_buffer *eb)
4556 return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4559 int read_extent_buffer_pages(struct extent_io_tree *tree,
4560 struct extent_buffer *eb, u64 start, int wait,
4561 get_extent_t *get_extent, int mirror_num)
4564 unsigned long start_i;
4568 int locked_pages = 0;
4569 int all_uptodate = 1;
4570 unsigned long num_pages;
4571 unsigned long num_reads = 0;
4572 struct bio *bio = NULL;
4573 unsigned long bio_flags = 0;
4575 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
4579 WARN_ON(start < eb->start);
4580 start_i = (start >> PAGE_CACHE_SHIFT) -
4581 (eb->start >> PAGE_CACHE_SHIFT);
4586 num_pages = num_extent_pages(eb->start, eb->len);
4587 for (i = start_i; i < num_pages; i++) {
4588 page = extent_buffer_page(eb, i);
4589 if (wait == WAIT_NONE) {
4590 if (!trylock_page(page))
4596 if (!PageUptodate(page)) {
4603 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4607 clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
4608 eb->read_mirror = 0;
4609 atomic_set(&eb->io_pages, num_reads);
4610 for (i = start_i; i < num_pages; i++) {
4611 page = extent_buffer_page(eb, i);
4612 if (!PageUptodate(page)) {
4613 ClearPageError(page);
4614 err = __extent_read_full_page(tree, page,
4616 mirror_num, &bio_flags);
4625 err = submit_one_bio(READ, bio, mirror_num, bio_flags);
4630 if (ret || wait != WAIT_COMPLETE)
4633 for (i = start_i; i < num_pages; i++) {
4634 page = extent_buffer_page(eb, i);
4635 wait_on_page_locked(page);
4636 if (!PageUptodate(page))
4644 while (locked_pages > 0) {
4645 page = extent_buffer_page(eb, i);
4653 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
4654 unsigned long start,
4661 char *dst = (char *)dstv;
4662 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4663 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4665 WARN_ON(start > eb->len);
4666 WARN_ON(start + len > eb->start + eb->len);
4668 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4671 page = extent_buffer_page(eb, i);
4673 cur = min(len, (PAGE_CACHE_SIZE - offset));
4674 kaddr = page_address(page);
4675 memcpy(dst, kaddr + offset, cur);
4684 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
4685 unsigned long min_len, char **map,
4686 unsigned long *map_start,
4687 unsigned long *map_len)
4689 size_t offset = start & (PAGE_CACHE_SIZE - 1);
4692 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4693 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4694 unsigned long end_i = (start_offset + start + min_len - 1) >>
4701 offset = start_offset;
4705 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
4708 if (start + min_len > eb->len) {
4709 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, "
4710 "wanted %lu %lu\n", (unsigned long long)eb->start,
4711 eb->len, start, min_len);
4715 p = extent_buffer_page(eb, i);
4716 kaddr = page_address(p);
4717 *map = kaddr + offset;
4718 *map_len = PAGE_CACHE_SIZE - offset;
4722 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
4723 unsigned long start,
4730 char *ptr = (char *)ptrv;
4731 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4732 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4735 WARN_ON(start > eb->len);
4736 WARN_ON(start + len > eb->start + eb->len);
4738 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4741 page = extent_buffer_page(eb, i);
4743 cur = min(len, (PAGE_CACHE_SIZE - offset));
4745 kaddr = page_address(page);
4746 ret = memcmp(ptr, kaddr + offset, cur);
4758 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
4759 unsigned long start, unsigned long len)
4765 char *src = (char *)srcv;
4766 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4767 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4769 WARN_ON(start > eb->len);
4770 WARN_ON(start + len > eb->start + eb->len);
4772 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4775 page = extent_buffer_page(eb, i);
4776 WARN_ON(!PageUptodate(page));
4778 cur = min(len, PAGE_CACHE_SIZE - offset);
4779 kaddr = page_address(page);
4780 memcpy(kaddr + offset, src, cur);
4789 void memset_extent_buffer(struct extent_buffer *eb, char c,
4790 unsigned long start, unsigned long len)
4796 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4797 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4799 WARN_ON(start > eb->len);
4800 WARN_ON(start + len > eb->start + eb->len);
4802 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4805 page = extent_buffer_page(eb, i);
4806 WARN_ON(!PageUptodate(page));
4808 cur = min(len, PAGE_CACHE_SIZE - offset);
4809 kaddr = page_address(page);
4810 memset(kaddr + offset, c, cur);
4818 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
4819 unsigned long dst_offset, unsigned long src_offset,
4822 u64 dst_len = dst->len;
4827 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4828 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
4830 WARN_ON(src->len != dst_len);
4832 offset = (start_offset + dst_offset) &
4833 ((unsigned long)PAGE_CACHE_SIZE - 1);
4836 page = extent_buffer_page(dst, i);
4837 WARN_ON(!PageUptodate(page));
4839 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
4841 kaddr = page_address(page);
4842 read_extent_buffer(src, kaddr + offset, src_offset, cur);
4851 static void move_pages(struct page *dst_page, struct page *src_page,
4852 unsigned long dst_off, unsigned long src_off,
4855 char *dst_kaddr = page_address(dst_page);
4856 if (dst_page == src_page) {
4857 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
4859 char *src_kaddr = page_address(src_page);
4860 char *p = dst_kaddr + dst_off + len;
4861 char *s = src_kaddr + src_off + len;
4868 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
4870 unsigned long distance = (src > dst) ? src - dst : dst - src;
4871 return distance < len;
4874 static void copy_pages(struct page *dst_page, struct page *src_page,
4875 unsigned long dst_off, unsigned long src_off,
4878 char *dst_kaddr = page_address(dst_page);
4880 int must_memmove = 0;
4882 if (dst_page != src_page) {
4883 src_kaddr = page_address(src_page);
4885 src_kaddr = dst_kaddr;
4886 if (areas_overlap(src_off, dst_off, len))
4891 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
4893 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
4896 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
4897 unsigned long src_offset, unsigned long len)
4900 size_t dst_off_in_page;
4901 size_t src_off_in_page;
4902 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4903 unsigned long dst_i;
4904 unsigned long src_i;
4906 if (src_offset + len > dst->len) {
4907 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
4908 "len %lu dst len %lu\n", src_offset, len, dst->len);
4911 if (dst_offset + len > dst->len) {
4912 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
4913 "len %lu dst len %lu\n", dst_offset, len, dst->len);
4918 dst_off_in_page = (start_offset + dst_offset) &
4919 ((unsigned long)PAGE_CACHE_SIZE - 1);
4920 src_off_in_page = (start_offset + src_offset) &
4921 ((unsigned long)PAGE_CACHE_SIZE - 1);
4923 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
4924 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
4926 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
4928 cur = min_t(unsigned long, cur,
4929 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
4931 copy_pages(extent_buffer_page(dst, dst_i),
4932 extent_buffer_page(dst, src_i),
4933 dst_off_in_page, src_off_in_page, cur);
4941 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
4942 unsigned long src_offset, unsigned long len)
4945 size_t dst_off_in_page;
4946 size_t src_off_in_page;
4947 unsigned long dst_end = dst_offset + len - 1;
4948 unsigned long src_end = src_offset + len - 1;
4949 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4950 unsigned long dst_i;
4951 unsigned long src_i;
4953 if (src_offset + len > dst->len) {
4954 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
4955 "len %lu len %lu\n", src_offset, len, dst->len);
4958 if (dst_offset + len > dst->len) {
4959 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
4960 "len %lu len %lu\n", dst_offset, len, dst->len);
4963 if (dst_offset < src_offset) {
4964 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
4968 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
4969 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
4971 dst_off_in_page = (start_offset + dst_end) &
4972 ((unsigned long)PAGE_CACHE_SIZE - 1);
4973 src_off_in_page = (start_offset + src_end) &
4974 ((unsigned long)PAGE_CACHE_SIZE - 1);
4976 cur = min_t(unsigned long, len, src_off_in_page + 1);
4977 cur = min(cur, dst_off_in_page + 1);
4978 move_pages(extent_buffer_page(dst, dst_i),
4979 extent_buffer_page(dst, src_i),
4980 dst_off_in_page - cur + 1,
4981 src_off_in_page - cur + 1, cur);
4989 int try_release_extent_buffer(struct page *page, gfp_t mask)
4991 struct extent_buffer *eb;
4994 * We need to make sure noboody is attaching this page to an eb right
4997 spin_lock(&page->mapping->private_lock);
4998 if (!PagePrivate(page)) {
4999 spin_unlock(&page->mapping->private_lock);
5003 eb = (struct extent_buffer *)page->private;
5007 * This is a little awful but should be ok, we need to make sure that
5008 * the eb doesn't disappear out from under us while we're looking at
5011 spin_lock(&eb->refs_lock);
5012 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
5013 spin_unlock(&eb->refs_lock);
5014 spin_unlock(&page->mapping->private_lock);
5017 spin_unlock(&page->mapping->private_lock);
5019 if ((mask & GFP_NOFS) == GFP_NOFS)
5023 * If tree ref isn't set then we know the ref on this eb is a real ref,
5024 * so just return, this page will likely be freed soon anyway.
5026 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
5027 spin_unlock(&eb->refs_lock);
5031 return release_extent_buffer(eb, mask);