1 #include <linux/bitops.h>
2 #include <linux/slab.h>
5 #include <linux/pagemap.h>
6 #include <linux/page-flags.h>
7 #include <linux/module.h>
8 #include <linux/spinlock.h>
9 #include <linux/blkdev.h>
10 #include <linux/swap.h>
11 #include <linux/writeback.h>
12 #include <linux/pagevec.h>
13 #include <linux/prefetch.h>
14 #include <linux/cleancache.h>
15 #include "extent_io.h"
16 #include "extent_map.h"
19 #include "btrfs_inode.h"
21 #include "check-integrity.h"
23 #include "rcu-string.h"
25 static struct kmem_cache *extent_state_cache;
26 static struct kmem_cache *extent_buffer_cache;
28 static LIST_HEAD(buffers);
29 static LIST_HEAD(states);
33 static DEFINE_SPINLOCK(leak_lock);
36 #define BUFFER_LRU_MAX 64
41 struct rb_node rb_node;
44 struct extent_page_data {
46 struct extent_io_tree *tree;
47 get_extent_t *get_extent;
49 /* tells writepage not to lock the state bits for this range
50 * it still does the unlocking
52 unsigned int extent_locked:1;
54 /* tells the submit_bio code to use a WRITE_SYNC */
55 unsigned int sync_io:1;
58 static noinline void flush_write_bio(void *data);
59 static inline struct btrfs_fs_info *
60 tree_fs_info(struct extent_io_tree *tree)
62 return btrfs_sb(tree->mapping->host->i_sb);
65 int __init extent_io_init(void)
67 extent_state_cache = kmem_cache_create("extent_state",
68 sizeof(struct extent_state), 0,
69 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
70 if (!extent_state_cache)
73 extent_buffer_cache = kmem_cache_create("extent_buffers",
74 sizeof(struct extent_buffer), 0,
75 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
76 if (!extent_buffer_cache)
77 goto free_state_cache;
81 kmem_cache_destroy(extent_state_cache);
85 void extent_io_exit(void)
87 struct extent_state *state;
88 struct extent_buffer *eb;
90 while (!list_empty(&states)) {
91 state = list_entry(states.next, struct extent_state, leak_list);
92 printk(KERN_ERR "btrfs state leak: start %llu end %llu "
93 "state %lu in tree %p refs %d\n",
94 (unsigned long long)state->start,
95 (unsigned long long)state->end,
96 state->state, state->tree, atomic_read(&state->refs));
97 list_del(&state->leak_list);
98 kmem_cache_free(extent_state_cache, state);
102 while (!list_empty(&buffers)) {
103 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
104 printk(KERN_ERR "btrfs buffer leak start %llu len %lu "
105 "refs %d\n", (unsigned long long)eb->start,
106 eb->len, atomic_read(&eb->refs));
107 list_del(&eb->leak_list);
108 kmem_cache_free(extent_buffer_cache, eb);
110 if (extent_state_cache)
111 kmem_cache_destroy(extent_state_cache);
112 if (extent_buffer_cache)
113 kmem_cache_destroy(extent_buffer_cache);
116 void extent_io_tree_init(struct extent_io_tree *tree,
117 struct address_space *mapping)
119 tree->state = RB_ROOT;
120 INIT_RADIX_TREE(&tree->buffer, GFP_ATOMIC);
122 tree->dirty_bytes = 0;
123 spin_lock_init(&tree->lock);
124 spin_lock_init(&tree->buffer_lock);
125 tree->mapping = mapping;
128 static struct extent_state *alloc_extent_state(gfp_t mask)
130 struct extent_state *state;
135 state = kmem_cache_alloc(extent_state_cache, mask);
142 spin_lock_irqsave(&leak_lock, flags);
143 list_add(&state->leak_list, &states);
144 spin_unlock_irqrestore(&leak_lock, flags);
146 atomic_set(&state->refs, 1);
147 init_waitqueue_head(&state->wq);
148 trace_alloc_extent_state(state, mask, _RET_IP_);
152 void free_extent_state(struct extent_state *state)
156 if (atomic_dec_and_test(&state->refs)) {
160 WARN_ON(state->tree);
162 spin_lock_irqsave(&leak_lock, flags);
163 list_del(&state->leak_list);
164 spin_unlock_irqrestore(&leak_lock, flags);
166 trace_free_extent_state(state, _RET_IP_);
167 kmem_cache_free(extent_state_cache, state);
171 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
172 struct rb_node *node)
174 struct rb_node **p = &root->rb_node;
175 struct rb_node *parent = NULL;
176 struct tree_entry *entry;
180 entry = rb_entry(parent, struct tree_entry, rb_node);
182 if (offset < entry->start)
184 else if (offset > entry->end)
190 rb_link_node(node, parent, p);
191 rb_insert_color(node, root);
195 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
196 struct rb_node **prev_ret,
197 struct rb_node **next_ret)
199 struct rb_root *root = &tree->state;
200 struct rb_node *n = root->rb_node;
201 struct rb_node *prev = NULL;
202 struct rb_node *orig_prev = NULL;
203 struct tree_entry *entry;
204 struct tree_entry *prev_entry = NULL;
207 entry = rb_entry(n, struct tree_entry, rb_node);
211 if (offset < entry->start)
213 else if (offset > entry->end)
221 while (prev && offset > prev_entry->end) {
222 prev = rb_next(prev);
223 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
230 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
231 while (prev && offset < prev_entry->start) {
232 prev = rb_prev(prev);
233 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
240 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
243 struct rb_node *prev = NULL;
246 ret = __etree_search(tree, offset, &prev, NULL);
252 static void merge_cb(struct extent_io_tree *tree, struct extent_state *new,
253 struct extent_state *other)
255 if (tree->ops && tree->ops->merge_extent_hook)
256 tree->ops->merge_extent_hook(tree->mapping->host, new,
261 * utility function to look for merge candidates inside a given range.
262 * Any extents with matching state are merged together into a single
263 * extent in the tree. Extents with EXTENT_IO in their state field
264 * are not merged because the end_io handlers need to be able to do
265 * operations on them without sleeping (or doing allocations/splits).
267 * This should be called with the tree lock held.
269 static void merge_state(struct extent_io_tree *tree,
270 struct extent_state *state)
272 struct extent_state *other;
273 struct rb_node *other_node;
275 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
278 other_node = rb_prev(&state->rb_node);
280 other = rb_entry(other_node, struct extent_state, rb_node);
281 if (other->end == state->start - 1 &&
282 other->state == state->state) {
283 merge_cb(tree, state, other);
284 state->start = other->start;
286 rb_erase(&other->rb_node, &tree->state);
287 free_extent_state(other);
290 other_node = rb_next(&state->rb_node);
292 other = rb_entry(other_node, struct extent_state, rb_node);
293 if (other->start == state->end + 1 &&
294 other->state == state->state) {
295 merge_cb(tree, state, other);
296 state->end = other->end;
298 rb_erase(&other->rb_node, &tree->state);
299 free_extent_state(other);
304 static void set_state_cb(struct extent_io_tree *tree,
305 struct extent_state *state, int *bits)
307 if (tree->ops && tree->ops->set_bit_hook)
308 tree->ops->set_bit_hook(tree->mapping->host, state, bits);
311 static void clear_state_cb(struct extent_io_tree *tree,
312 struct extent_state *state, int *bits)
314 if (tree->ops && tree->ops->clear_bit_hook)
315 tree->ops->clear_bit_hook(tree->mapping->host, state, bits);
318 static void set_state_bits(struct extent_io_tree *tree,
319 struct extent_state *state, int *bits);
322 * insert an extent_state struct into the tree. 'bits' are set on the
323 * struct before it is inserted.
325 * This may return -EEXIST if the extent is already there, in which case the
326 * state struct is freed.
328 * The tree lock is not taken internally. This is a utility function and
329 * probably isn't what you want to call (see set/clear_extent_bit).
331 static int insert_state(struct extent_io_tree *tree,
332 struct extent_state *state, u64 start, u64 end,
335 struct rb_node *node;
338 printk(KERN_ERR "btrfs end < start %llu %llu\n",
339 (unsigned long long)end,
340 (unsigned long long)start);
343 state->start = start;
346 set_state_bits(tree, state, bits);
348 node = tree_insert(&tree->state, end, &state->rb_node);
350 struct extent_state *found;
351 found = rb_entry(node, struct extent_state, rb_node);
352 printk(KERN_ERR "btrfs found node %llu %llu on insert of "
353 "%llu %llu\n", (unsigned long long)found->start,
354 (unsigned long long)found->end,
355 (unsigned long long)start, (unsigned long long)end);
359 merge_state(tree, state);
363 static void split_cb(struct extent_io_tree *tree, struct extent_state *orig,
366 if (tree->ops && tree->ops->split_extent_hook)
367 tree->ops->split_extent_hook(tree->mapping->host, orig, split);
371 * split a given extent state struct in two, inserting the preallocated
372 * struct 'prealloc' as the newly created second half. 'split' indicates an
373 * offset inside 'orig' where it should be split.
376 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
377 * are two extent state structs in the tree:
378 * prealloc: [orig->start, split - 1]
379 * orig: [ split, orig->end ]
381 * The tree locks are not taken by this function. They need to be held
384 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
385 struct extent_state *prealloc, u64 split)
387 struct rb_node *node;
389 split_cb(tree, orig, split);
391 prealloc->start = orig->start;
392 prealloc->end = split - 1;
393 prealloc->state = orig->state;
396 node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
398 free_extent_state(prealloc);
401 prealloc->tree = tree;
405 static struct extent_state *next_state(struct extent_state *state)
407 struct rb_node *next = rb_next(&state->rb_node);
409 return rb_entry(next, struct extent_state, rb_node);
415 * utility function to clear some bits in an extent state struct.
416 * it will optionally wake up any one waiting on this state (wake == 1).
418 * If no bits are set on the state struct after clearing things, the
419 * struct is freed and removed from the tree
421 static struct extent_state *clear_state_bit(struct extent_io_tree *tree,
422 struct extent_state *state,
425 struct extent_state *next;
426 int bits_to_clear = *bits & ~EXTENT_CTLBITS;
428 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
429 u64 range = state->end - state->start + 1;
430 WARN_ON(range > tree->dirty_bytes);
431 tree->dirty_bytes -= range;
433 clear_state_cb(tree, state, bits);
434 state->state &= ~bits_to_clear;
437 if (state->state == 0) {
438 next = next_state(state);
440 rb_erase(&state->rb_node, &tree->state);
442 free_extent_state(state);
447 merge_state(tree, state);
448 next = next_state(state);
453 static struct extent_state *
454 alloc_extent_state_atomic(struct extent_state *prealloc)
457 prealloc = alloc_extent_state(GFP_ATOMIC);
462 void extent_io_tree_panic(struct extent_io_tree *tree, int err)
464 btrfs_panic(tree_fs_info(tree), err, "Locking error: "
465 "Extent tree was modified by another "
466 "thread while locked.");
470 * clear some bits on a range in the tree. This may require splitting
471 * or inserting elements in the tree, so the gfp mask is used to
472 * indicate which allocations or sleeping are allowed.
474 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
475 * the given range from the tree regardless of state (ie for truncate).
477 * the range [start, end] is inclusive.
479 * This takes the tree lock, and returns 0 on success and < 0 on error.
481 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
482 int bits, int wake, int delete,
483 struct extent_state **cached_state,
486 struct extent_state *state;
487 struct extent_state *cached;
488 struct extent_state *prealloc = NULL;
489 struct rb_node *node;
495 bits |= ~EXTENT_CTLBITS;
496 bits |= EXTENT_FIRST_DELALLOC;
498 if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY))
501 if (!prealloc && (mask & __GFP_WAIT)) {
502 prealloc = alloc_extent_state(mask);
507 spin_lock(&tree->lock);
509 cached = *cached_state;
512 *cached_state = NULL;
516 if (cached && cached->tree && cached->start <= start &&
517 cached->end > start) {
519 atomic_dec(&cached->refs);
524 free_extent_state(cached);
527 * this search will find the extents that end after
530 node = tree_search(tree, start);
533 state = rb_entry(node, struct extent_state, rb_node);
535 if (state->start > end)
537 WARN_ON(state->end < start);
538 last_end = state->end;
540 /* the state doesn't have the wanted bits, go ahead */
541 if (!(state->state & bits)) {
542 state = next_state(state);
547 * | ---- desired range ---- |
549 * | ------------- state -------------- |
551 * We need to split the extent we found, and may flip
552 * bits on second half.
554 * If the extent we found extends past our range, we
555 * just split and search again. It'll get split again
556 * the next time though.
558 * If the extent we found is inside our range, we clear
559 * the desired bit on it.
562 if (state->start < start) {
563 prealloc = alloc_extent_state_atomic(prealloc);
565 err = split_state(tree, state, prealloc, start);
567 extent_io_tree_panic(tree, err);
572 if (state->end <= end) {
573 state = clear_state_bit(tree, state, &bits, wake);
579 * | ---- desired range ---- |
581 * We need to split the extent, and clear the bit
584 if (state->start <= end && state->end > end) {
585 prealloc = alloc_extent_state_atomic(prealloc);
587 err = split_state(tree, state, prealloc, end + 1);
589 extent_io_tree_panic(tree, err);
594 clear_state_bit(tree, prealloc, &bits, wake);
600 state = clear_state_bit(tree, state, &bits, wake);
602 if (last_end == (u64)-1)
604 start = last_end + 1;
605 if (start <= end && state && !need_resched())
610 spin_unlock(&tree->lock);
612 free_extent_state(prealloc);
619 spin_unlock(&tree->lock);
620 if (mask & __GFP_WAIT)
625 static void wait_on_state(struct extent_io_tree *tree,
626 struct extent_state *state)
627 __releases(tree->lock)
628 __acquires(tree->lock)
631 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
632 spin_unlock(&tree->lock);
634 spin_lock(&tree->lock);
635 finish_wait(&state->wq, &wait);
639 * waits for one or more bits to clear on a range in the state tree.
640 * The range [start, end] is inclusive.
641 * The tree lock is taken by this function
643 void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits)
645 struct extent_state *state;
646 struct rb_node *node;
648 spin_lock(&tree->lock);
652 * this search will find all the extents that end after
655 node = tree_search(tree, start);
659 state = rb_entry(node, struct extent_state, rb_node);
661 if (state->start > end)
664 if (state->state & bits) {
665 start = state->start;
666 atomic_inc(&state->refs);
667 wait_on_state(tree, state);
668 free_extent_state(state);
671 start = state->end + 1;
676 cond_resched_lock(&tree->lock);
679 spin_unlock(&tree->lock);
682 static void set_state_bits(struct extent_io_tree *tree,
683 struct extent_state *state,
686 int bits_to_set = *bits & ~EXTENT_CTLBITS;
688 set_state_cb(tree, state, bits);
689 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
690 u64 range = state->end - state->start + 1;
691 tree->dirty_bytes += range;
693 state->state |= bits_to_set;
696 static void cache_state(struct extent_state *state,
697 struct extent_state **cached_ptr)
699 if (cached_ptr && !(*cached_ptr)) {
700 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) {
702 atomic_inc(&state->refs);
707 static void uncache_state(struct extent_state **cached_ptr)
709 if (cached_ptr && (*cached_ptr)) {
710 struct extent_state *state = *cached_ptr;
712 free_extent_state(state);
717 * set some bits on a range in the tree. This may require allocations or
718 * sleeping, so the gfp mask is used to indicate what is allowed.
720 * If any of the exclusive bits are set, this will fail with -EEXIST if some
721 * part of the range already has the desired bits set. The start of the
722 * existing range is returned in failed_start in this case.
724 * [start, end] is inclusive This takes the tree lock.
727 static int __must_check
728 __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
729 int bits, int exclusive_bits, u64 *failed_start,
730 struct extent_state **cached_state, gfp_t mask)
732 struct extent_state *state;
733 struct extent_state *prealloc = NULL;
734 struct rb_node *node;
739 bits |= EXTENT_FIRST_DELALLOC;
741 if (!prealloc && (mask & __GFP_WAIT)) {
742 prealloc = alloc_extent_state(mask);
746 spin_lock(&tree->lock);
747 if (cached_state && *cached_state) {
748 state = *cached_state;
749 if (state->start <= start && state->end > start &&
751 node = &state->rb_node;
756 * this search will find all the extents that end after
759 node = tree_search(tree, start);
761 prealloc = alloc_extent_state_atomic(prealloc);
763 err = insert_state(tree, prealloc, start, end, &bits);
765 extent_io_tree_panic(tree, err);
770 state = rb_entry(node, struct extent_state, rb_node);
772 last_start = state->start;
773 last_end = state->end;
776 * | ---- desired range ---- |
779 * Just lock what we found and keep going
781 if (state->start == start && state->end <= end) {
782 if (state->state & exclusive_bits) {
783 *failed_start = state->start;
788 set_state_bits(tree, state, &bits);
789 cache_state(state, cached_state);
790 merge_state(tree, state);
791 if (last_end == (u64)-1)
793 start = last_end + 1;
794 state = next_state(state);
795 if (start < end && state && state->start == start &&
802 * | ---- desired range ---- |
805 * | ------------- state -------------- |
807 * We need to split the extent we found, and may flip bits on
810 * If the extent we found extends past our
811 * range, we just split and search again. It'll get split
812 * again the next time though.
814 * If the extent we found is inside our range, we set the
817 if (state->start < start) {
818 if (state->state & exclusive_bits) {
819 *failed_start = start;
824 prealloc = alloc_extent_state_atomic(prealloc);
826 err = split_state(tree, state, prealloc, start);
828 extent_io_tree_panic(tree, err);
833 if (state->end <= end) {
834 set_state_bits(tree, state, &bits);
835 cache_state(state, cached_state);
836 merge_state(tree, state);
837 if (last_end == (u64)-1)
839 start = last_end + 1;
840 state = next_state(state);
841 if (start < end && state && state->start == start &&
848 * | ---- desired range ---- |
849 * | state | or | state |
851 * There's a hole, we need to insert something in it and
852 * ignore the extent we found.
854 if (state->start > start) {
856 if (end < last_start)
859 this_end = last_start - 1;
861 prealloc = alloc_extent_state_atomic(prealloc);
865 * Avoid to free 'prealloc' if it can be merged with
868 err = insert_state(tree, prealloc, start, this_end,
871 extent_io_tree_panic(tree, err);
873 cache_state(prealloc, cached_state);
875 start = this_end + 1;
879 * | ---- desired range ---- |
881 * We need to split the extent, and set the bit
884 if (state->start <= end && state->end > end) {
885 if (state->state & exclusive_bits) {
886 *failed_start = start;
891 prealloc = alloc_extent_state_atomic(prealloc);
893 err = split_state(tree, state, prealloc, end + 1);
895 extent_io_tree_panic(tree, err);
897 set_state_bits(tree, prealloc, &bits);
898 cache_state(prealloc, cached_state);
899 merge_state(tree, prealloc);
907 spin_unlock(&tree->lock);
909 free_extent_state(prealloc);
916 spin_unlock(&tree->lock);
917 if (mask & __GFP_WAIT)
922 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits,
923 u64 *failed_start, struct extent_state **cached_state,
926 return __set_extent_bit(tree, start, end, bits, 0, failed_start,
932 * convert_extent - convert all bits in a given range from one bit to another
933 * @tree: the io tree to search
934 * @start: the start offset in bytes
935 * @end: the end offset in bytes (inclusive)
936 * @bits: the bits to set in this range
937 * @clear_bits: the bits to clear in this range
938 * @mask: the allocation mask
940 * This will go through and set bits for the given range. If any states exist
941 * already in this range they are set with the given bit and cleared of the
942 * clear_bits. This is only meant to be used by things that are mergeable, ie
943 * converting from say DELALLOC to DIRTY. This is not meant to be used with
944 * boundary bits like LOCK.
946 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
947 int bits, int clear_bits, gfp_t mask)
949 struct extent_state *state;
950 struct extent_state *prealloc = NULL;
951 struct rb_node *node;
957 if (!prealloc && (mask & __GFP_WAIT)) {
958 prealloc = alloc_extent_state(mask);
963 spin_lock(&tree->lock);
965 * this search will find all the extents that end after
968 node = tree_search(tree, start);
970 prealloc = alloc_extent_state_atomic(prealloc);
975 err = insert_state(tree, prealloc, start, end, &bits);
978 extent_io_tree_panic(tree, err);
981 state = rb_entry(node, struct extent_state, rb_node);
983 last_start = state->start;
984 last_end = state->end;
987 * | ---- desired range ---- |
990 * Just lock what we found and keep going
992 if (state->start == start && state->end <= end) {
993 set_state_bits(tree, state, &bits);
994 state = clear_state_bit(tree, state, &clear_bits, 0);
995 if (last_end == (u64)-1)
997 start = last_end + 1;
998 if (start < end && state && state->start == start &&
1005 * | ---- desired range ---- |
1008 * | ------------- state -------------- |
1010 * We need to split the extent we found, and may flip bits on
1013 * If the extent we found extends past our
1014 * range, we just split and search again. It'll get split
1015 * again the next time though.
1017 * If the extent we found is inside our range, we set the
1018 * desired bit on it.
1020 if (state->start < start) {
1021 prealloc = alloc_extent_state_atomic(prealloc);
1026 err = split_state(tree, state, prealloc, start);
1028 extent_io_tree_panic(tree, err);
1032 if (state->end <= end) {
1033 set_state_bits(tree, state, &bits);
1034 state = clear_state_bit(tree, state, &clear_bits, 0);
1035 if (last_end == (u64)-1)
1037 start = last_end + 1;
1038 if (start < end && state && state->start == start &&
1045 * | ---- desired range ---- |
1046 * | state | or | state |
1048 * There's a hole, we need to insert something in it and
1049 * ignore the extent we found.
1051 if (state->start > start) {
1053 if (end < last_start)
1056 this_end = last_start - 1;
1058 prealloc = alloc_extent_state_atomic(prealloc);
1065 * Avoid to free 'prealloc' if it can be merged with
1068 err = insert_state(tree, prealloc, start, this_end,
1071 extent_io_tree_panic(tree, err);
1073 start = this_end + 1;
1077 * | ---- desired range ---- |
1079 * We need to split the extent, and set the bit
1082 if (state->start <= end && state->end > end) {
1083 prealloc = alloc_extent_state_atomic(prealloc);
1089 err = split_state(tree, state, prealloc, end + 1);
1091 extent_io_tree_panic(tree, err);
1093 set_state_bits(tree, prealloc, &bits);
1094 clear_state_bit(tree, prealloc, &clear_bits, 0);
1102 spin_unlock(&tree->lock);
1104 free_extent_state(prealloc);
1111 spin_unlock(&tree->lock);
1112 if (mask & __GFP_WAIT)
1117 /* wrappers around set/clear extent bit */
1118 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1121 return set_extent_bit(tree, start, end, EXTENT_DIRTY, NULL,
1125 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1126 int bits, gfp_t mask)
1128 return set_extent_bit(tree, start, end, bits, NULL,
1132 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1133 int bits, gfp_t mask)
1135 return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask);
1138 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
1139 struct extent_state **cached_state, gfp_t mask)
1141 return set_extent_bit(tree, start, end,
1142 EXTENT_DELALLOC | EXTENT_UPTODATE,
1143 NULL, cached_state, mask);
1146 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1149 return clear_extent_bit(tree, start, end,
1150 EXTENT_DIRTY | EXTENT_DELALLOC |
1151 EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask);
1154 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
1157 return set_extent_bit(tree, start, end, EXTENT_NEW, NULL,
1161 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
1162 struct extent_state **cached_state, gfp_t mask)
1164 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0,
1165 cached_state, mask);
1168 int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
1169 struct extent_state **cached_state, gfp_t mask)
1171 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0,
1172 cached_state, mask);
1176 * either insert or lock state struct between start and end use mask to tell
1177 * us if waiting is desired.
1179 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1180 int bits, struct extent_state **cached_state)
1185 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED | bits,
1186 EXTENT_LOCKED, &failed_start,
1187 cached_state, GFP_NOFS);
1188 if (err == -EEXIST) {
1189 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1190 start = failed_start;
1193 WARN_ON(start > end);
1198 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1200 return lock_extent_bits(tree, start, end, 0, NULL);
1203 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1208 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1209 &failed_start, NULL, GFP_NOFS);
1210 if (err == -EEXIST) {
1211 if (failed_start > start)
1212 clear_extent_bit(tree, start, failed_start - 1,
1213 EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS);
1219 int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end,
1220 struct extent_state **cached, gfp_t mask)
1222 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached,
1226 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1228 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL,
1233 * helper function to set both pages and extents in the tree writeback
1235 static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
1237 unsigned long index = start >> PAGE_CACHE_SHIFT;
1238 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1241 while (index <= end_index) {
1242 page = find_get_page(tree->mapping, index);
1243 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1244 set_page_writeback(page);
1245 page_cache_release(page);
1251 /* find the first state struct with 'bits' set after 'start', and
1252 * return it. tree->lock must be held. NULL will returned if
1253 * nothing was found after 'start'
1255 struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree,
1256 u64 start, int bits)
1258 struct rb_node *node;
1259 struct extent_state *state;
1262 * this search will find all the extents that end after
1265 node = tree_search(tree, start);
1270 state = rb_entry(node, struct extent_state, rb_node);
1271 if (state->end >= start && (state->state & bits))
1274 node = rb_next(node);
1283 * find the first offset in the io tree with 'bits' set. zero is
1284 * returned if we find something, and *start_ret and *end_ret are
1285 * set to reflect the state struct that was found.
1287 * If nothing was found, 1 is returned. If found something, return 0.
1289 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1290 u64 *start_ret, u64 *end_ret, int bits)
1292 struct extent_state *state;
1295 spin_lock(&tree->lock);
1296 state = find_first_extent_bit_state(tree, start, bits);
1298 *start_ret = state->start;
1299 *end_ret = state->end;
1302 spin_unlock(&tree->lock);
1307 * find a contiguous range of bytes in the file marked as delalloc, not
1308 * more than 'max_bytes'. start and end are used to return the range,
1310 * 1 is returned if we find something, 0 if nothing was in the tree
1312 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1313 u64 *start, u64 *end, u64 max_bytes,
1314 struct extent_state **cached_state)
1316 struct rb_node *node;
1317 struct extent_state *state;
1318 u64 cur_start = *start;
1320 u64 total_bytes = 0;
1322 spin_lock(&tree->lock);
1325 * this search will find all the extents that end after
1328 node = tree_search(tree, cur_start);
1336 state = rb_entry(node, struct extent_state, rb_node);
1337 if (found && (state->start != cur_start ||
1338 (state->state & EXTENT_BOUNDARY))) {
1341 if (!(state->state & EXTENT_DELALLOC)) {
1347 *start = state->start;
1348 *cached_state = state;
1349 atomic_inc(&state->refs);
1353 cur_start = state->end + 1;
1354 node = rb_next(node);
1357 total_bytes += state->end - state->start + 1;
1358 if (total_bytes >= max_bytes)
1362 spin_unlock(&tree->lock);
1366 static noinline void __unlock_for_delalloc(struct inode *inode,
1367 struct page *locked_page,
1371 struct page *pages[16];
1372 unsigned long index = start >> PAGE_CACHE_SHIFT;
1373 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1374 unsigned long nr_pages = end_index - index + 1;
1377 if (index == locked_page->index && end_index == index)
1380 while (nr_pages > 0) {
1381 ret = find_get_pages_contig(inode->i_mapping, index,
1382 min_t(unsigned long, nr_pages,
1383 ARRAY_SIZE(pages)), pages);
1384 for (i = 0; i < ret; i++) {
1385 if (pages[i] != locked_page)
1386 unlock_page(pages[i]);
1387 page_cache_release(pages[i]);
1395 static noinline int lock_delalloc_pages(struct inode *inode,
1396 struct page *locked_page,
1400 unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT;
1401 unsigned long start_index = index;
1402 unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT;
1403 unsigned long pages_locked = 0;
1404 struct page *pages[16];
1405 unsigned long nrpages;
1409 /* the caller is responsible for locking the start index */
1410 if (index == locked_page->index && index == end_index)
1413 /* skip the page at the start index */
1414 nrpages = end_index - index + 1;
1415 while (nrpages > 0) {
1416 ret = find_get_pages_contig(inode->i_mapping, index,
1417 min_t(unsigned long,
1418 nrpages, ARRAY_SIZE(pages)), pages);
1423 /* now we have an array of pages, lock them all */
1424 for (i = 0; i < ret; i++) {
1426 * the caller is taking responsibility for
1429 if (pages[i] != locked_page) {
1430 lock_page(pages[i]);
1431 if (!PageDirty(pages[i]) ||
1432 pages[i]->mapping != inode->i_mapping) {
1434 unlock_page(pages[i]);
1435 page_cache_release(pages[i]);
1439 page_cache_release(pages[i]);
1448 if (ret && pages_locked) {
1449 __unlock_for_delalloc(inode, locked_page,
1451 ((u64)(start_index + pages_locked - 1)) <<
1458 * find a contiguous range of bytes in the file marked as delalloc, not
1459 * more than 'max_bytes'. start and end are used to return the range,
1461 * 1 is returned if we find something, 0 if nothing was in the tree
1463 static noinline u64 find_lock_delalloc_range(struct inode *inode,
1464 struct extent_io_tree *tree,
1465 struct page *locked_page,
1466 u64 *start, u64 *end,
1472 struct extent_state *cached_state = NULL;
1477 /* step one, find a bunch of delalloc bytes starting at start */
1478 delalloc_start = *start;
1480 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1481 max_bytes, &cached_state);
1482 if (!found || delalloc_end <= *start) {
1483 *start = delalloc_start;
1484 *end = delalloc_end;
1485 free_extent_state(cached_state);
1490 * start comes from the offset of locked_page. We have to lock
1491 * pages in order, so we can't process delalloc bytes before
1494 if (delalloc_start < *start)
1495 delalloc_start = *start;
1498 * make sure to limit the number of pages we try to lock down
1501 if (delalloc_end + 1 - delalloc_start > max_bytes && loops)
1502 delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1;
1504 /* step two, lock all the pages after the page that has start */
1505 ret = lock_delalloc_pages(inode, locked_page,
1506 delalloc_start, delalloc_end);
1507 if (ret == -EAGAIN) {
1508 /* some of the pages are gone, lets avoid looping by
1509 * shortening the size of the delalloc range we're searching
1511 free_extent_state(cached_state);
1513 unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1);
1514 max_bytes = PAGE_CACHE_SIZE - offset;
1522 BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */
1524 /* step three, lock the state bits for the whole range */
1525 lock_extent_bits(tree, delalloc_start, delalloc_end, 0, &cached_state);
1527 /* then test to make sure it is all still delalloc */
1528 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1529 EXTENT_DELALLOC, 1, cached_state);
1531 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1532 &cached_state, GFP_NOFS);
1533 __unlock_for_delalloc(inode, locked_page,
1534 delalloc_start, delalloc_end);
1538 free_extent_state(cached_state);
1539 *start = delalloc_start;
1540 *end = delalloc_end;
1545 int extent_clear_unlock_delalloc(struct inode *inode,
1546 struct extent_io_tree *tree,
1547 u64 start, u64 end, struct page *locked_page,
1551 struct page *pages[16];
1552 unsigned long index = start >> PAGE_CACHE_SHIFT;
1553 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1554 unsigned long nr_pages = end_index - index + 1;
1558 if (op & EXTENT_CLEAR_UNLOCK)
1559 clear_bits |= EXTENT_LOCKED;
1560 if (op & EXTENT_CLEAR_DIRTY)
1561 clear_bits |= EXTENT_DIRTY;
1563 if (op & EXTENT_CLEAR_DELALLOC)
1564 clear_bits |= EXTENT_DELALLOC;
1566 clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS);
1567 if (!(op & (EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
1568 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK |
1569 EXTENT_SET_PRIVATE2)))
1572 while (nr_pages > 0) {
1573 ret = find_get_pages_contig(inode->i_mapping, index,
1574 min_t(unsigned long,
1575 nr_pages, ARRAY_SIZE(pages)), pages);
1576 for (i = 0; i < ret; i++) {
1578 if (op & EXTENT_SET_PRIVATE2)
1579 SetPagePrivate2(pages[i]);
1581 if (pages[i] == locked_page) {
1582 page_cache_release(pages[i]);
1585 if (op & EXTENT_CLEAR_DIRTY)
1586 clear_page_dirty_for_io(pages[i]);
1587 if (op & EXTENT_SET_WRITEBACK)
1588 set_page_writeback(pages[i]);
1589 if (op & EXTENT_END_WRITEBACK)
1590 end_page_writeback(pages[i]);
1591 if (op & EXTENT_CLEAR_UNLOCK_PAGE)
1592 unlock_page(pages[i]);
1593 page_cache_release(pages[i]);
1603 * count the number of bytes in the tree that have a given bit(s)
1604 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1605 * cached. The total number found is returned.
1607 u64 count_range_bits(struct extent_io_tree *tree,
1608 u64 *start, u64 search_end, u64 max_bytes,
1609 unsigned long bits, int contig)
1611 struct rb_node *node;
1612 struct extent_state *state;
1613 u64 cur_start = *start;
1614 u64 total_bytes = 0;
1618 if (search_end <= cur_start) {
1623 spin_lock(&tree->lock);
1624 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1625 total_bytes = tree->dirty_bytes;
1629 * this search will find all the extents that end after
1632 node = tree_search(tree, cur_start);
1637 state = rb_entry(node, struct extent_state, rb_node);
1638 if (state->start > search_end)
1640 if (contig && found && state->start > last + 1)
1642 if (state->end >= cur_start && (state->state & bits) == bits) {
1643 total_bytes += min(search_end, state->end) + 1 -
1644 max(cur_start, state->start);
1645 if (total_bytes >= max_bytes)
1648 *start = max(cur_start, state->start);
1652 } else if (contig && found) {
1655 node = rb_next(node);
1660 spin_unlock(&tree->lock);
1665 * set the private field for a given byte offset in the tree. If there isn't
1666 * an extent_state there already, this does nothing.
1668 int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1670 struct rb_node *node;
1671 struct extent_state *state;
1674 spin_lock(&tree->lock);
1676 * this search will find all the extents that end after
1679 node = tree_search(tree, start);
1684 state = rb_entry(node, struct extent_state, rb_node);
1685 if (state->start != start) {
1689 state->private = private;
1691 spin_unlock(&tree->lock);
1695 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1697 struct rb_node *node;
1698 struct extent_state *state;
1701 spin_lock(&tree->lock);
1703 * this search will find all the extents that end after
1706 node = tree_search(tree, start);
1711 state = rb_entry(node, struct extent_state, rb_node);
1712 if (state->start != start) {
1716 *private = state->private;
1718 spin_unlock(&tree->lock);
1723 * searches a range in the state tree for a given mask.
1724 * If 'filled' == 1, this returns 1 only if every extent in the tree
1725 * has the bits set. Otherwise, 1 is returned if any bit in the
1726 * range is found set.
1728 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1729 int bits, int filled, struct extent_state *cached)
1731 struct extent_state *state = NULL;
1732 struct rb_node *node;
1735 spin_lock(&tree->lock);
1736 if (cached && cached->tree && cached->start <= start &&
1737 cached->end > start)
1738 node = &cached->rb_node;
1740 node = tree_search(tree, start);
1741 while (node && start <= end) {
1742 state = rb_entry(node, struct extent_state, rb_node);
1744 if (filled && state->start > start) {
1749 if (state->start > end)
1752 if (state->state & bits) {
1756 } else if (filled) {
1761 if (state->end == (u64)-1)
1764 start = state->end + 1;
1767 node = rb_next(node);
1774 spin_unlock(&tree->lock);
1779 * helper function to set a given page up to date if all the
1780 * extents in the tree for that page are up to date
1782 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
1784 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1785 u64 end = start + PAGE_CACHE_SIZE - 1;
1786 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1787 SetPageUptodate(page);
1791 * helper function to unlock a page if all the extents in the tree
1792 * for that page are unlocked
1794 static void check_page_locked(struct extent_io_tree *tree, struct page *page)
1796 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1797 u64 end = start + PAGE_CACHE_SIZE - 1;
1798 if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL))
1803 * helper function to end page writeback if all the extents
1804 * in the tree for that page are done with writeback
1806 static void check_page_writeback(struct extent_io_tree *tree,
1809 end_page_writeback(page);
1813 * When IO fails, either with EIO or csum verification fails, we
1814 * try other mirrors that might have a good copy of the data. This
1815 * io_failure_record is used to record state as we go through all the
1816 * mirrors. If another mirror has good data, the page is set up to date
1817 * and things continue. If a good mirror can't be found, the original
1818 * bio end_io callback is called to indicate things have failed.
1820 struct io_failure_record {
1825 unsigned long bio_flags;
1831 static int free_io_failure(struct inode *inode, struct io_failure_record *rec,
1836 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1838 set_state_private(failure_tree, rec->start, 0);
1839 ret = clear_extent_bits(failure_tree, rec->start,
1840 rec->start + rec->len - 1,
1841 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1846 ret = clear_extent_bits(&BTRFS_I(inode)->io_tree, rec->start,
1847 rec->start + rec->len - 1,
1848 EXTENT_DAMAGED, GFP_NOFS);
1857 static void repair_io_failure_callback(struct bio *bio, int err)
1859 complete(bio->bi_private);
1863 * this bypasses the standard btrfs submit functions deliberately, as
1864 * the standard behavior is to write all copies in a raid setup. here we only
1865 * want to write the one bad copy. so we do the mapping for ourselves and issue
1866 * submit_bio directly.
1867 * to avoid any synchonization issues, wait for the data after writing, which
1868 * actually prevents the read that triggered the error from finishing.
1869 * currently, there can be no more than two copies of every data bit. thus,
1870 * exactly one rewrite is required.
1872 int repair_io_failure(struct btrfs_mapping_tree *map_tree, u64 start,
1873 u64 length, u64 logical, struct page *page,
1877 struct btrfs_device *dev;
1878 DECLARE_COMPLETION_ONSTACK(compl);
1881 struct btrfs_bio *bbio = NULL;
1884 BUG_ON(!mirror_num);
1886 bio = bio_alloc(GFP_NOFS, 1);
1889 bio->bi_private = &compl;
1890 bio->bi_end_io = repair_io_failure_callback;
1892 map_length = length;
1894 ret = btrfs_map_block(map_tree, WRITE, logical,
1895 &map_length, &bbio, mirror_num);
1900 BUG_ON(mirror_num != bbio->mirror_num);
1901 sector = bbio->stripes[mirror_num-1].physical >> 9;
1902 bio->bi_sector = sector;
1903 dev = bbio->stripes[mirror_num-1].dev;
1905 if (!dev || !dev->bdev || !dev->writeable) {
1909 bio->bi_bdev = dev->bdev;
1910 bio_add_page(bio, page, length, start-page_offset(page));
1911 btrfsic_submit_bio(WRITE_SYNC, bio);
1912 wait_for_completion(&compl);
1914 if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) {
1915 /* try to remap that extent elsewhere? */
1917 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
1921 printk_ratelimited_in_rcu(KERN_INFO "btrfs read error corrected: ino %lu off %llu "
1922 "(dev %s sector %llu)\n", page->mapping->host->i_ino,
1923 start, rcu_str_deref(dev->name), sector);
1929 int repair_eb_io_failure(struct btrfs_root *root, struct extent_buffer *eb,
1932 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1933 u64 start = eb->start;
1934 unsigned long i, num_pages = num_extent_pages(eb->start, eb->len);
1937 for (i = 0; i < num_pages; i++) {
1938 struct page *p = extent_buffer_page(eb, i);
1939 ret = repair_io_failure(map_tree, start, PAGE_CACHE_SIZE,
1940 start, p, mirror_num);
1943 start += PAGE_CACHE_SIZE;
1950 * each time an IO finishes, we do a fast check in the IO failure tree
1951 * to see if we need to process or clean up an io_failure_record
1953 static int clean_io_failure(u64 start, struct page *page)
1956 u64 private_failure;
1957 struct io_failure_record *failrec;
1958 struct btrfs_mapping_tree *map_tree;
1959 struct extent_state *state;
1963 struct inode *inode = page->mapping->host;
1966 ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1967 (u64)-1, 1, EXTENT_DIRTY, 0);
1971 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, start,
1976 failrec = (struct io_failure_record *)(unsigned long) private_failure;
1977 BUG_ON(!failrec->this_mirror);
1979 if (failrec->in_validation) {
1980 /* there was no real error, just free the record */
1981 pr_debug("clean_io_failure: freeing dummy error at %llu\n",
1987 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1988 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1991 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1993 if (state && state->start == failrec->start) {
1994 map_tree = &BTRFS_I(inode)->root->fs_info->mapping_tree;
1995 num_copies = btrfs_num_copies(map_tree, failrec->logical,
1997 if (num_copies > 1) {
1998 ret = repair_io_failure(map_tree, start, failrec->len,
1999 failrec->logical, page,
2000 failrec->failed_mirror);
2007 ret = free_io_failure(inode, failrec, did_repair);
2013 * this is a generic handler for readpage errors (default
2014 * readpage_io_failed_hook). if other copies exist, read those and write back
2015 * good data to the failed position. does not investigate in remapping the
2016 * failed extent elsewhere, hoping the device will be smart enough to do this as
2020 static int bio_readpage_error(struct bio *failed_bio, struct page *page,
2021 u64 start, u64 end, int failed_mirror,
2022 struct extent_state *state)
2024 struct io_failure_record *failrec = NULL;
2026 struct extent_map *em;
2027 struct inode *inode = page->mapping->host;
2028 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2029 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2030 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2037 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
2039 ret = get_state_private(failure_tree, start, &private);
2041 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2044 failrec->start = start;
2045 failrec->len = end - start + 1;
2046 failrec->this_mirror = 0;
2047 failrec->bio_flags = 0;
2048 failrec->in_validation = 0;
2050 read_lock(&em_tree->lock);
2051 em = lookup_extent_mapping(em_tree, start, failrec->len);
2053 read_unlock(&em_tree->lock);
2058 if (em->start > start || em->start + em->len < start) {
2059 free_extent_map(em);
2062 read_unlock(&em_tree->lock);
2064 if (!em || IS_ERR(em)) {
2068 logical = start - em->start;
2069 logical = em->block_start + logical;
2070 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2071 logical = em->block_start;
2072 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2073 extent_set_compress_type(&failrec->bio_flags,
2076 pr_debug("bio_readpage_error: (new) logical=%llu, start=%llu, "
2077 "len=%llu\n", logical, start, failrec->len);
2078 failrec->logical = logical;
2079 free_extent_map(em);
2081 /* set the bits in the private failure tree */
2082 ret = set_extent_bits(failure_tree, start, end,
2083 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
2085 ret = set_state_private(failure_tree, start,
2086 (u64)(unsigned long)failrec);
2087 /* set the bits in the inode's tree */
2089 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED,
2096 failrec = (struct io_failure_record *)(unsigned long)private;
2097 pr_debug("bio_readpage_error: (found) logical=%llu, "
2098 "start=%llu, len=%llu, validation=%d\n",
2099 failrec->logical, failrec->start, failrec->len,
2100 failrec->in_validation);
2102 * when data can be on disk more than twice, add to failrec here
2103 * (e.g. with a list for failed_mirror) to make
2104 * clean_io_failure() clean all those errors at once.
2107 num_copies = btrfs_num_copies(
2108 &BTRFS_I(inode)->root->fs_info->mapping_tree,
2109 failrec->logical, failrec->len);
2110 if (num_copies == 1) {
2112 * we only have a single copy of the data, so don't bother with
2113 * all the retry and error correction code that follows. no
2114 * matter what the error is, it is very likely to persist.
2116 pr_debug("bio_readpage_error: cannot repair, num_copies == 1. "
2117 "state=%p, num_copies=%d, next_mirror %d, "
2118 "failed_mirror %d\n", state, num_copies,
2119 failrec->this_mirror, failed_mirror);
2120 free_io_failure(inode, failrec, 0);
2125 spin_lock(&tree->lock);
2126 state = find_first_extent_bit_state(tree, failrec->start,
2128 if (state && state->start != failrec->start)
2130 spin_unlock(&tree->lock);
2134 * there are two premises:
2135 * a) deliver good data to the caller
2136 * b) correct the bad sectors on disk
2138 if (failed_bio->bi_vcnt > 1) {
2140 * to fulfill b), we need to know the exact failing sectors, as
2141 * we don't want to rewrite any more than the failed ones. thus,
2142 * we need separate read requests for the failed bio
2144 * if the following BUG_ON triggers, our validation request got
2145 * merged. we need separate requests for our algorithm to work.
2147 BUG_ON(failrec->in_validation);
2148 failrec->in_validation = 1;
2149 failrec->this_mirror = failed_mirror;
2150 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
2153 * we're ready to fulfill a) and b) alongside. get a good copy
2154 * of the failed sector and if we succeed, we have setup
2155 * everything for repair_io_failure to do the rest for us.
2157 if (failrec->in_validation) {
2158 BUG_ON(failrec->this_mirror != failed_mirror);
2159 failrec->in_validation = 0;
2160 failrec->this_mirror = 0;
2162 failrec->failed_mirror = failed_mirror;
2163 failrec->this_mirror++;
2164 if (failrec->this_mirror == failed_mirror)
2165 failrec->this_mirror++;
2166 read_mode = READ_SYNC;
2169 if (!state || failrec->this_mirror > num_copies) {
2170 pr_debug("bio_readpage_error: (fail) state=%p, num_copies=%d, "
2171 "next_mirror %d, failed_mirror %d\n", state,
2172 num_copies, failrec->this_mirror, failed_mirror);
2173 free_io_failure(inode, failrec, 0);
2177 bio = bio_alloc(GFP_NOFS, 1);
2179 free_io_failure(inode, failrec, 0);
2182 bio->bi_private = state;
2183 bio->bi_end_io = failed_bio->bi_end_io;
2184 bio->bi_sector = failrec->logical >> 9;
2185 bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
2188 bio_add_page(bio, page, failrec->len, start - page_offset(page));
2190 pr_debug("bio_readpage_error: submitting new read[%#x] to "
2191 "this_mirror=%d, num_copies=%d, in_validation=%d\n", read_mode,
2192 failrec->this_mirror, num_copies, failrec->in_validation);
2194 ret = tree->ops->submit_bio_hook(inode, read_mode, bio,
2195 failrec->this_mirror,
2196 failrec->bio_flags, 0);
2200 /* lots and lots of room for performance fixes in the end_bio funcs */
2202 int end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2204 int uptodate = (err == 0);
2205 struct extent_io_tree *tree;
2208 tree = &BTRFS_I(page->mapping->host)->io_tree;
2210 if (tree->ops && tree->ops->writepage_end_io_hook) {
2211 ret = tree->ops->writepage_end_io_hook(page, start,
2212 end, NULL, uptodate);
2218 ClearPageUptodate(page);
2225 * after a writepage IO is done, we need to:
2226 * clear the uptodate bits on error
2227 * clear the writeback bits in the extent tree for this IO
2228 * end_page_writeback if the page has no more pending IO
2230 * Scheduling is not allowed, so the extent state tree is expected
2231 * to have one and only one object corresponding to this IO.
2233 static void end_bio_extent_writepage(struct bio *bio, int err)
2235 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2236 struct extent_io_tree *tree;
2242 struct page *page = bvec->bv_page;
2243 tree = &BTRFS_I(page->mapping->host)->io_tree;
2245 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
2247 end = start + bvec->bv_len - 1;
2249 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
2254 if (--bvec >= bio->bi_io_vec)
2255 prefetchw(&bvec->bv_page->flags);
2257 if (end_extent_writepage(page, err, start, end))
2261 end_page_writeback(page);
2263 check_page_writeback(tree, page);
2264 } while (bvec >= bio->bi_io_vec);
2270 * after a readpage IO is done, we need to:
2271 * clear the uptodate bits on error
2272 * set the uptodate bits if things worked
2273 * set the page up to date if all extents in the tree are uptodate
2274 * clear the lock bit in the extent tree
2275 * unlock the page if there are no other extents locked for it
2277 * Scheduling is not allowed, so the extent state tree is expected
2278 * to have one and only one object corresponding to this IO.
2280 static void end_bio_extent_readpage(struct bio *bio, int err)
2282 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
2283 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
2284 struct bio_vec *bvec = bio->bi_io_vec;
2285 struct extent_io_tree *tree;
2296 struct page *page = bvec->bv_page;
2297 struct extent_state *cached = NULL;
2298 struct extent_state *state;
2300 pr_debug("end_bio_extent_readpage: bi_vcnt=%d, idx=%d, err=%d, "
2301 "mirror=%ld\n", bio->bi_vcnt, bio->bi_idx, err,
2302 (long int)bio->bi_bdev);
2303 tree = &BTRFS_I(page->mapping->host)->io_tree;
2305 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
2307 end = start + bvec->bv_len - 1;
2309 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
2314 if (++bvec <= bvec_end)
2315 prefetchw(&bvec->bv_page->flags);
2317 spin_lock(&tree->lock);
2318 state = find_first_extent_bit_state(tree, start, EXTENT_LOCKED);
2319 if (state && state->start == start) {
2321 * take a reference on the state, unlock will drop
2324 cache_state(state, &cached);
2326 spin_unlock(&tree->lock);
2328 mirror = (int)(unsigned long)bio->bi_bdev;
2329 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
2330 ret = tree->ops->readpage_end_io_hook(page, start, end,
2335 clean_io_failure(start, page);
2338 if (!uptodate && tree->ops && tree->ops->readpage_io_failed_hook) {
2339 ret = tree->ops->readpage_io_failed_hook(page, mirror);
2341 test_bit(BIO_UPTODATE, &bio->bi_flags))
2343 } else if (!uptodate) {
2345 * The generic bio_readpage_error handles errors the
2346 * following way: If possible, new read requests are
2347 * created and submitted and will end up in
2348 * end_bio_extent_readpage as well (if we're lucky, not
2349 * in the !uptodate case). In that case it returns 0 and
2350 * we just go on with the next page in our bio. If it
2351 * can't handle the error it will return -EIO and we
2352 * remain responsible for that page.
2354 ret = bio_readpage_error(bio, page, start, end, mirror, NULL);
2357 test_bit(BIO_UPTODATE, &bio->bi_flags);
2360 uncache_state(&cached);
2365 if (uptodate && tree->track_uptodate) {
2366 set_extent_uptodate(tree, start, end, &cached,
2369 unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC);
2373 SetPageUptodate(page);
2375 ClearPageUptodate(page);
2381 check_page_uptodate(tree, page);
2383 ClearPageUptodate(page);
2386 check_page_locked(tree, page);
2388 } while (bvec <= bvec_end);
2394 btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
2399 bio = bio_alloc(gfp_flags, nr_vecs);
2401 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
2402 while (!bio && (nr_vecs /= 2))
2403 bio = bio_alloc(gfp_flags, nr_vecs);
2408 bio->bi_bdev = bdev;
2409 bio->bi_sector = first_sector;
2415 * Since writes are async, they will only return -ENOMEM.
2416 * Reads can return the full range of I/O error conditions.
2418 static int __must_check submit_one_bio(int rw, struct bio *bio,
2419 int mirror_num, unsigned long bio_flags)
2422 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2423 struct page *page = bvec->bv_page;
2424 struct extent_io_tree *tree = bio->bi_private;
2427 start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
2429 bio->bi_private = NULL;
2433 if (tree->ops && tree->ops->submit_bio_hook)
2434 ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
2435 mirror_num, bio_flags, start);
2437 btrfsic_submit_bio(rw, bio);
2439 if (bio_flagged(bio, BIO_EOPNOTSUPP))
2445 static int merge_bio(struct extent_io_tree *tree, struct page *page,
2446 unsigned long offset, size_t size, struct bio *bio,
2447 unsigned long bio_flags)
2450 if (tree->ops && tree->ops->merge_bio_hook)
2451 ret = tree->ops->merge_bio_hook(page, offset, size, bio,
2458 static int submit_extent_page(int rw, struct extent_io_tree *tree,
2459 struct page *page, sector_t sector,
2460 size_t size, unsigned long offset,
2461 struct block_device *bdev,
2462 struct bio **bio_ret,
2463 unsigned long max_pages,
2464 bio_end_io_t end_io_func,
2466 unsigned long prev_bio_flags,
2467 unsigned long bio_flags)
2473 int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED;
2474 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
2475 size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE);
2477 if (bio_ret && *bio_ret) {
2480 contig = bio->bi_sector == sector;
2482 contig = bio->bi_sector + (bio->bi_size >> 9) ==
2485 if (prev_bio_flags != bio_flags || !contig ||
2486 merge_bio(tree, page, offset, page_size, bio, bio_flags) ||
2487 bio_add_page(bio, page, page_size, offset) < page_size) {
2488 ret = submit_one_bio(rw, bio, mirror_num,
2497 if (this_compressed)
2500 nr = bio_get_nr_vecs(bdev);
2502 bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
2506 bio_add_page(bio, page, page_size, offset);
2507 bio->bi_end_io = end_io_func;
2508 bio->bi_private = tree;
2513 ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
2518 void attach_extent_buffer_page(struct extent_buffer *eb, struct page *page)
2520 if (!PagePrivate(page)) {
2521 SetPagePrivate(page);
2522 page_cache_get(page);
2523 set_page_private(page, (unsigned long)eb);
2525 WARN_ON(page->private != (unsigned long)eb);
2529 void set_page_extent_mapped(struct page *page)
2531 if (!PagePrivate(page)) {
2532 SetPagePrivate(page);
2533 page_cache_get(page);
2534 set_page_private(page, EXTENT_PAGE_PRIVATE);
2539 * basic readpage implementation. Locked extent state structs are inserted
2540 * into the tree that are removed when the IO is done (by the end_io
2542 * XXX JDM: This needs looking at to ensure proper page locking
2544 static int __extent_read_full_page(struct extent_io_tree *tree,
2546 get_extent_t *get_extent,
2547 struct bio **bio, int mirror_num,
2548 unsigned long *bio_flags)
2550 struct inode *inode = page->mapping->host;
2551 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2552 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2556 u64 last_byte = i_size_read(inode);
2560 struct extent_map *em;
2561 struct block_device *bdev;
2562 struct btrfs_ordered_extent *ordered;
2565 size_t pg_offset = 0;
2567 size_t disk_io_size;
2568 size_t blocksize = inode->i_sb->s_blocksize;
2569 unsigned long this_bio_flag = 0;
2571 set_page_extent_mapped(page);
2573 if (!PageUptodate(page)) {
2574 if (cleancache_get_page(page) == 0) {
2575 BUG_ON(blocksize != PAGE_SIZE);
2582 lock_extent(tree, start, end);
2583 ordered = btrfs_lookup_ordered_extent(inode, start);
2586 unlock_extent(tree, start, end);
2587 btrfs_start_ordered_extent(inode, ordered, 1);
2588 btrfs_put_ordered_extent(ordered);
2591 if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
2593 size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
2596 iosize = PAGE_CACHE_SIZE - zero_offset;
2597 userpage = kmap_atomic(page);
2598 memset(userpage + zero_offset, 0, iosize);
2599 flush_dcache_page(page);
2600 kunmap_atomic(userpage);
2603 while (cur <= end) {
2604 if (cur >= last_byte) {
2606 struct extent_state *cached = NULL;
2608 iosize = PAGE_CACHE_SIZE - pg_offset;
2609 userpage = kmap_atomic(page);
2610 memset(userpage + pg_offset, 0, iosize);
2611 flush_dcache_page(page);
2612 kunmap_atomic(userpage);
2613 set_extent_uptodate(tree, cur, cur + iosize - 1,
2615 unlock_extent_cached(tree, cur, cur + iosize - 1,
2619 em = get_extent(inode, page, pg_offset, cur,
2621 if (IS_ERR_OR_NULL(em)) {
2623 unlock_extent(tree, cur, end);
2626 extent_offset = cur - em->start;
2627 BUG_ON(extent_map_end(em) <= cur);
2630 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2631 this_bio_flag = EXTENT_BIO_COMPRESSED;
2632 extent_set_compress_type(&this_bio_flag,
2636 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2637 cur_end = min(extent_map_end(em) - 1, end);
2638 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2639 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2640 disk_io_size = em->block_len;
2641 sector = em->block_start >> 9;
2643 sector = (em->block_start + extent_offset) >> 9;
2644 disk_io_size = iosize;
2647 block_start = em->block_start;
2648 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2649 block_start = EXTENT_MAP_HOLE;
2650 free_extent_map(em);
2653 /* we've found a hole, just zero and go on */
2654 if (block_start == EXTENT_MAP_HOLE) {
2656 struct extent_state *cached = NULL;
2658 userpage = kmap_atomic(page);
2659 memset(userpage + pg_offset, 0, iosize);
2660 flush_dcache_page(page);
2661 kunmap_atomic(userpage);
2663 set_extent_uptodate(tree, cur, cur + iosize - 1,
2665 unlock_extent_cached(tree, cur, cur + iosize - 1,
2668 pg_offset += iosize;
2671 /* the get_extent function already copied into the page */
2672 if (test_range_bit(tree, cur, cur_end,
2673 EXTENT_UPTODATE, 1, NULL)) {
2674 check_page_uptodate(tree, page);
2675 unlock_extent(tree, cur, cur + iosize - 1);
2677 pg_offset += iosize;
2680 /* we have an inline extent but it didn't get marked up
2681 * to date. Error out
2683 if (block_start == EXTENT_MAP_INLINE) {
2685 unlock_extent(tree, cur, cur + iosize - 1);
2687 pg_offset += iosize;
2692 if (tree->ops && tree->ops->readpage_io_hook) {
2693 ret = tree->ops->readpage_io_hook(page, cur,
2697 unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
2699 ret = submit_extent_page(READ, tree, page,
2700 sector, disk_io_size, pg_offset,
2702 end_bio_extent_readpage, mirror_num,
2705 BUG_ON(ret == -ENOMEM);
2707 *bio_flags = this_bio_flag;
2712 pg_offset += iosize;
2716 if (!PageError(page))
2717 SetPageUptodate(page);
2723 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
2724 get_extent_t *get_extent, int mirror_num)
2726 struct bio *bio = NULL;
2727 unsigned long bio_flags = 0;
2730 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
2733 ret = submit_one_bio(READ, bio, mirror_num, bio_flags);
2737 static noinline void update_nr_written(struct page *page,
2738 struct writeback_control *wbc,
2739 unsigned long nr_written)
2741 wbc->nr_to_write -= nr_written;
2742 if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
2743 wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
2744 page->mapping->writeback_index = page->index + nr_written;
2748 * the writepage semantics are similar to regular writepage. extent
2749 * records are inserted to lock ranges in the tree, and as dirty areas
2750 * are found, they are marked writeback. Then the lock bits are removed
2751 * and the end_io handler clears the writeback ranges
2753 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2756 struct inode *inode = page->mapping->host;
2757 struct extent_page_data *epd = data;
2758 struct extent_io_tree *tree = epd->tree;
2759 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2761 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2765 u64 last_byte = i_size_read(inode);
2769 struct extent_state *cached_state = NULL;
2770 struct extent_map *em;
2771 struct block_device *bdev;
2774 size_t pg_offset = 0;
2776 loff_t i_size = i_size_read(inode);
2777 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
2783 unsigned long nr_written = 0;
2784 bool fill_delalloc = true;
2786 if (wbc->sync_mode == WB_SYNC_ALL)
2787 write_flags = WRITE_SYNC;
2789 write_flags = WRITE;
2791 trace___extent_writepage(page, inode, wbc);
2793 WARN_ON(!PageLocked(page));
2795 ClearPageError(page);
2797 pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
2798 if (page->index > end_index ||
2799 (page->index == end_index && !pg_offset)) {
2800 page->mapping->a_ops->invalidatepage(page, 0);
2805 if (page->index == end_index) {
2808 userpage = kmap_atomic(page);
2809 memset(userpage + pg_offset, 0,
2810 PAGE_CACHE_SIZE - pg_offset);
2811 kunmap_atomic(userpage);
2812 flush_dcache_page(page);
2816 set_page_extent_mapped(page);
2818 if (!tree->ops || !tree->ops->fill_delalloc)
2819 fill_delalloc = false;
2821 delalloc_start = start;
2824 if (!epd->extent_locked && fill_delalloc) {
2825 u64 delalloc_to_write = 0;
2827 * make sure the wbc mapping index is at least updated
2830 update_nr_written(page, wbc, 0);
2832 while (delalloc_end < page_end) {
2833 nr_delalloc = find_lock_delalloc_range(inode, tree,
2838 if (nr_delalloc == 0) {
2839 delalloc_start = delalloc_end + 1;
2842 ret = tree->ops->fill_delalloc(inode, page,
2847 /* File system has been set read-only */
2853 * delalloc_end is already one less than the total
2854 * length, so we don't subtract one from
2857 delalloc_to_write += (delalloc_end - delalloc_start +
2860 delalloc_start = delalloc_end + 1;
2862 if (wbc->nr_to_write < delalloc_to_write) {
2865 if (delalloc_to_write < thresh * 2)
2866 thresh = delalloc_to_write;
2867 wbc->nr_to_write = min_t(u64, delalloc_to_write,
2871 /* did the fill delalloc function already unlock and start
2877 * we've unlocked the page, so we can't update
2878 * the mapping's writeback index, just update
2881 wbc->nr_to_write -= nr_written;
2885 if (tree->ops && tree->ops->writepage_start_hook) {
2886 ret = tree->ops->writepage_start_hook(page, start,
2889 /* Fixup worker will requeue */
2891 wbc->pages_skipped++;
2893 redirty_page_for_writepage(wbc, page);
2894 update_nr_written(page, wbc, nr_written);
2902 * we don't want to touch the inode after unlocking the page,
2903 * so we update the mapping writeback index now
2905 update_nr_written(page, wbc, nr_written + 1);
2908 if (last_byte <= start) {
2909 if (tree->ops && tree->ops->writepage_end_io_hook)
2910 tree->ops->writepage_end_io_hook(page, start,
2915 blocksize = inode->i_sb->s_blocksize;
2917 while (cur <= end) {
2918 if (cur >= last_byte) {
2919 if (tree->ops && tree->ops->writepage_end_io_hook)
2920 tree->ops->writepage_end_io_hook(page, cur,
2924 em = epd->get_extent(inode, page, pg_offset, cur,
2926 if (IS_ERR_OR_NULL(em)) {
2931 extent_offset = cur - em->start;
2932 BUG_ON(extent_map_end(em) <= cur);
2934 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2935 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2936 sector = (em->block_start + extent_offset) >> 9;
2938 block_start = em->block_start;
2939 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
2940 free_extent_map(em);
2944 * compressed and inline extents are written through other
2947 if (compressed || block_start == EXTENT_MAP_HOLE ||
2948 block_start == EXTENT_MAP_INLINE) {
2950 * end_io notification does not happen here for
2951 * compressed extents
2953 if (!compressed && tree->ops &&
2954 tree->ops->writepage_end_io_hook)
2955 tree->ops->writepage_end_io_hook(page, cur,
2958 else if (compressed) {
2959 /* we don't want to end_page_writeback on
2960 * a compressed extent. this happens
2967 pg_offset += iosize;
2970 /* leave this out until we have a page_mkwrite call */
2971 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
2972 EXTENT_DIRTY, 0, NULL)) {
2974 pg_offset += iosize;
2978 if (tree->ops && tree->ops->writepage_io_hook) {
2979 ret = tree->ops->writepage_io_hook(page, cur,
2987 unsigned long max_nr = end_index + 1;
2989 set_range_writeback(tree, cur, cur + iosize - 1);
2990 if (!PageWriteback(page)) {
2991 printk(KERN_ERR "btrfs warning page %lu not "
2992 "writeback, cur %llu end %llu\n",
2993 page->index, (unsigned long long)cur,
2994 (unsigned long long)end);
2997 ret = submit_extent_page(write_flags, tree, page,
2998 sector, iosize, pg_offset,
2999 bdev, &epd->bio, max_nr,
3000 end_bio_extent_writepage,
3006 pg_offset += iosize;
3011 /* make sure the mapping tag for page dirty gets cleared */
3012 set_page_writeback(page);
3013 end_page_writeback(page);
3019 /* drop our reference on any cached states */
3020 free_extent_state(cached_state);
3024 static int eb_wait(void *word)
3030 static void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3032 wait_on_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK, eb_wait,
3033 TASK_UNINTERRUPTIBLE);
3036 static int lock_extent_buffer_for_io(struct extent_buffer *eb,
3037 struct btrfs_fs_info *fs_info,
3038 struct extent_page_data *epd)
3040 unsigned long i, num_pages;
3044 if (!btrfs_try_tree_write_lock(eb)) {
3046 flush_write_bio(epd);
3047 btrfs_tree_lock(eb);
3050 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3051 btrfs_tree_unlock(eb);
3055 flush_write_bio(epd);
3059 wait_on_extent_buffer_writeback(eb);
3060 btrfs_tree_lock(eb);
3061 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3063 btrfs_tree_unlock(eb);
3068 * We need to do this to prevent races in people who check if the eb is
3069 * under IO since we can end up having no IO bits set for a short period
3072 spin_lock(&eb->refs_lock);
3073 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3074 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3075 spin_unlock(&eb->refs_lock);
3076 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3077 spin_lock(&fs_info->delalloc_lock);
3078 if (fs_info->dirty_metadata_bytes >= eb->len)
3079 fs_info->dirty_metadata_bytes -= eb->len;
3082 spin_unlock(&fs_info->delalloc_lock);
3085 spin_unlock(&eb->refs_lock);
3088 btrfs_tree_unlock(eb);
3093 num_pages = num_extent_pages(eb->start, eb->len);
3094 for (i = 0; i < num_pages; i++) {
3095 struct page *p = extent_buffer_page(eb, i);
3097 if (!trylock_page(p)) {
3099 flush_write_bio(epd);
3109 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3111 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3112 smp_mb__after_clear_bit();
3113 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3116 static void end_bio_extent_buffer_writepage(struct bio *bio, int err)
3118 int uptodate = err == 0;
3119 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
3120 struct extent_buffer *eb;
3124 struct page *page = bvec->bv_page;
3127 eb = (struct extent_buffer *)page->private;
3129 done = atomic_dec_and_test(&eb->io_pages);
3131 if (!uptodate || test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
3132 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3133 ClearPageUptodate(page);
3137 end_page_writeback(page);
3142 end_extent_buffer_writeback(eb);
3143 } while (bvec >= bio->bi_io_vec);
3149 static int write_one_eb(struct extent_buffer *eb,
3150 struct btrfs_fs_info *fs_info,
3151 struct writeback_control *wbc,
3152 struct extent_page_data *epd)
3154 struct block_device *bdev = fs_info->fs_devices->latest_bdev;
3155 u64 offset = eb->start;
3156 unsigned long i, num_pages;
3157 int rw = (epd->sync_io ? WRITE_SYNC : WRITE);
3160 clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3161 num_pages = num_extent_pages(eb->start, eb->len);
3162 atomic_set(&eb->io_pages, num_pages);
3163 for (i = 0; i < num_pages; i++) {
3164 struct page *p = extent_buffer_page(eb, i);
3166 clear_page_dirty_for_io(p);
3167 set_page_writeback(p);
3168 ret = submit_extent_page(rw, eb->tree, p, offset >> 9,
3169 PAGE_CACHE_SIZE, 0, bdev, &epd->bio,
3170 -1, end_bio_extent_buffer_writepage,
3173 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3175 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3176 end_extent_buffer_writeback(eb);
3180 offset += PAGE_CACHE_SIZE;
3181 update_nr_written(p, wbc, 1);
3185 if (unlikely(ret)) {
3186 for (; i < num_pages; i++) {
3187 struct page *p = extent_buffer_page(eb, i);
3195 int btree_write_cache_pages(struct address_space *mapping,
3196 struct writeback_control *wbc)
3198 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
3199 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
3200 struct extent_buffer *eb, *prev_eb = NULL;
3201 struct extent_page_data epd = {
3205 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3209 int nr_to_write_done = 0;
3210 struct pagevec pvec;
3213 pgoff_t end; /* Inclusive */
3217 pagevec_init(&pvec, 0);
3218 if (wbc->range_cyclic) {
3219 index = mapping->writeback_index; /* Start from prev offset */
3222 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3223 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3226 if (wbc->sync_mode == WB_SYNC_ALL)
3227 tag = PAGECACHE_TAG_TOWRITE;
3229 tag = PAGECACHE_TAG_DIRTY;
3231 if (wbc->sync_mode == WB_SYNC_ALL)
3232 tag_pages_for_writeback(mapping, index, end);
3233 while (!done && !nr_to_write_done && (index <= end) &&
3234 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3235 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3239 for (i = 0; i < nr_pages; i++) {
3240 struct page *page = pvec.pages[i];
3242 if (!PagePrivate(page))
3245 if (!wbc->range_cyclic && page->index > end) {
3250 eb = (struct extent_buffer *)page->private;
3259 if (!atomic_inc_not_zero(&eb->refs)) {
3265 ret = lock_extent_buffer_for_io(eb, fs_info, &epd);
3267 free_extent_buffer(eb);
3271 ret = write_one_eb(eb, fs_info, wbc, &epd);
3274 free_extent_buffer(eb);
3277 free_extent_buffer(eb);
3280 * the filesystem may choose to bump up nr_to_write.
3281 * We have to make sure to honor the new nr_to_write
3284 nr_to_write_done = wbc->nr_to_write <= 0;
3286 pagevec_release(&pvec);
3289 if (!scanned && !done) {
3291 * We hit the last page and there is more work to be done: wrap
3292 * back to the start of the file
3298 flush_write_bio(&epd);
3303 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3304 * @mapping: address space structure to write
3305 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3306 * @writepage: function called for each page
3307 * @data: data passed to writepage function
3309 * If a page is already under I/O, write_cache_pages() skips it, even
3310 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3311 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3312 * and msync() need to guarantee that all the data which was dirty at the time
3313 * the call was made get new I/O started against them. If wbc->sync_mode is
3314 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3315 * existing IO to complete.
3317 static int extent_write_cache_pages(struct extent_io_tree *tree,
3318 struct address_space *mapping,
3319 struct writeback_control *wbc,
3320 writepage_t writepage, void *data,
3321 void (*flush_fn)(void *))
3323 struct inode *inode = mapping->host;
3326 int nr_to_write_done = 0;
3327 struct pagevec pvec;
3330 pgoff_t end; /* Inclusive */
3335 * We have to hold onto the inode so that ordered extents can do their
3336 * work when the IO finishes. The alternative to this is failing to add
3337 * an ordered extent if the igrab() fails there and that is a huge pain
3338 * to deal with, so instead just hold onto the inode throughout the
3339 * writepages operation. If it fails here we are freeing up the inode
3340 * anyway and we'd rather not waste our time writing out stuff that is
3341 * going to be truncated anyway.
3346 pagevec_init(&pvec, 0);
3347 if (wbc->range_cyclic) {
3348 index = mapping->writeback_index; /* Start from prev offset */
3351 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3352 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3355 if (wbc->sync_mode == WB_SYNC_ALL)
3356 tag = PAGECACHE_TAG_TOWRITE;
3358 tag = PAGECACHE_TAG_DIRTY;
3360 if (wbc->sync_mode == WB_SYNC_ALL)
3361 tag_pages_for_writeback(mapping, index, end);
3362 while (!done && !nr_to_write_done && (index <= end) &&
3363 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3364 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3368 for (i = 0; i < nr_pages; i++) {
3369 struct page *page = pvec.pages[i];
3372 * At this point we hold neither mapping->tree_lock nor
3373 * lock on the page itself: the page may be truncated or
3374 * invalidated (changing page->mapping to NULL), or even
3375 * swizzled back from swapper_space to tmpfs file
3379 tree->ops->write_cache_pages_lock_hook) {
3380 tree->ops->write_cache_pages_lock_hook(page,
3383 if (!trylock_page(page)) {
3389 if (unlikely(page->mapping != mapping)) {
3394 if (!wbc->range_cyclic && page->index > end) {
3400 if (wbc->sync_mode != WB_SYNC_NONE) {
3401 if (PageWriteback(page))
3403 wait_on_page_writeback(page);
3406 if (PageWriteback(page) ||
3407 !clear_page_dirty_for_io(page)) {
3412 ret = (*writepage)(page, wbc, data);
3414 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
3422 * the filesystem may choose to bump up nr_to_write.
3423 * We have to make sure to honor the new nr_to_write
3426 nr_to_write_done = wbc->nr_to_write <= 0;
3428 pagevec_release(&pvec);
3431 if (!scanned && !done) {
3433 * We hit the last page and there is more work to be done: wrap
3434 * back to the start of the file
3440 btrfs_add_delayed_iput(inode);
3444 static void flush_epd_write_bio(struct extent_page_data *epd)
3453 ret = submit_one_bio(rw, epd->bio, 0, 0);
3454 BUG_ON(ret < 0); /* -ENOMEM */
3459 static noinline void flush_write_bio(void *data)
3461 struct extent_page_data *epd = data;
3462 flush_epd_write_bio(epd);
3465 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
3466 get_extent_t *get_extent,
3467 struct writeback_control *wbc)
3470 struct extent_page_data epd = {
3473 .get_extent = get_extent,
3475 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3478 ret = __extent_writepage(page, wbc, &epd);
3480 flush_epd_write_bio(&epd);
3484 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
3485 u64 start, u64 end, get_extent_t *get_extent,
3489 struct address_space *mapping = inode->i_mapping;
3491 unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
3494 struct extent_page_data epd = {
3497 .get_extent = get_extent,
3499 .sync_io = mode == WB_SYNC_ALL,
3501 struct writeback_control wbc_writepages = {
3503 .nr_to_write = nr_pages * 2,
3504 .range_start = start,
3505 .range_end = end + 1,
3508 while (start <= end) {
3509 page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
3510 if (clear_page_dirty_for_io(page))
3511 ret = __extent_writepage(page, &wbc_writepages, &epd);
3513 if (tree->ops && tree->ops->writepage_end_io_hook)
3514 tree->ops->writepage_end_io_hook(page, start,
3515 start + PAGE_CACHE_SIZE - 1,
3519 page_cache_release(page);
3520 start += PAGE_CACHE_SIZE;
3523 flush_epd_write_bio(&epd);
3527 int extent_writepages(struct extent_io_tree *tree,
3528 struct address_space *mapping,
3529 get_extent_t *get_extent,
3530 struct writeback_control *wbc)
3533 struct extent_page_data epd = {
3536 .get_extent = get_extent,
3538 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3541 ret = extent_write_cache_pages(tree, mapping, wbc,
3542 __extent_writepage, &epd,
3544 flush_epd_write_bio(&epd);
3548 int extent_readpages(struct extent_io_tree *tree,
3549 struct address_space *mapping,
3550 struct list_head *pages, unsigned nr_pages,
3551 get_extent_t get_extent)
3553 struct bio *bio = NULL;
3555 unsigned long bio_flags = 0;
3556 struct page *pagepool[16];
3561 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
3562 page = list_entry(pages->prev, struct page, lru);
3564 prefetchw(&page->flags);
3565 list_del(&page->lru);
3566 if (add_to_page_cache_lru(page, mapping,
3567 page->index, GFP_NOFS)) {
3568 page_cache_release(page);
3572 pagepool[nr++] = page;
3573 if (nr < ARRAY_SIZE(pagepool))
3575 for (i = 0; i < nr; i++) {
3576 __extent_read_full_page(tree, pagepool[i], get_extent,
3577 &bio, 0, &bio_flags);
3578 page_cache_release(pagepool[i]);
3582 for (i = 0; i < nr; i++) {
3583 __extent_read_full_page(tree, pagepool[i], get_extent,
3584 &bio, 0, &bio_flags);
3585 page_cache_release(pagepool[i]);
3588 BUG_ON(!list_empty(pages));
3590 return submit_one_bio(READ, bio, 0, bio_flags);
3595 * basic invalidatepage code, this waits on any locked or writeback
3596 * ranges corresponding to the page, and then deletes any extent state
3597 * records from the tree
3599 int extent_invalidatepage(struct extent_io_tree *tree,
3600 struct page *page, unsigned long offset)
3602 struct extent_state *cached_state = NULL;
3603 u64 start = ((u64)page->index << PAGE_CACHE_SHIFT);
3604 u64 end = start + PAGE_CACHE_SIZE - 1;
3605 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
3607 start += (offset + blocksize - 1) & ~(blocksize - 1);
3611 lock_extent_bits(tree, start, end, 0, &cached_state);
3612 wait_on_page_writeback(page);
3613 clear_extent_bit(tree, start, end,
3614 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
3615 EXTENT_DO_ACCOUNTING,
3616 1, 1, &cached_state, GFP_NOFS);
3621 * a helper for releasepage, this tests for areas of the page that
3622 * are locked or under IO and drops the related state bits if it is safe
3625 int try_release_extent_state(struct extent_map_tree *map,
3626 struct extent_io_tree *tree, struct page *page,
3629 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
3630 u64 end = start + PAGE_CACHE_SIZE - 1;
3633 if (test_range_bit(tree, start, end,
3634 EXTENT_IOBITS, 0, NULL))
3637 if ((mask & GFP_NOFS) == GFP_NOFS)
3640 * at this point we can safely clear everything except the
3641 * locked bit and the nodatasum bit
3643 ret = clear_extent_bit(tree, start, end,
3644 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
3647 /* if clear_extent_bit failed for enomem reasons,
3648 * we can't allow the release to continue.
3659 * a helper for releasepage. As long as there are no locked extents
3660 * in the range corresponding to the page, both state records and extent
3661 * map records are removed
3663 int try_release_extent_mapping(struct extent_map_tree *map,
3664 struct extent_io_tree *tree, struct page *page,
3667 struct extent_map *em;
3668 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
3669 u64 end = start + PAGE_CACHE_SIZE - 1;
3671 if ((mask & __GFP_WAIT) &&
3672 page->mapping->host->i_size > 16 * 1024 * 1024) {
3674 while (start <= end) {
3675 len = end - start + 1;
3676 write_lock(&map->lock);
3677 em = lookup_extent_mapping(map, start, len);
3679 write_unlock(&map->lock);
3682 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
3683 em->start != start) {
3684 write_unlock(&map->lock);
3685 free_extent_map(em);
3688 if (!test_range_bit(tree, em->start,
3689 extent_map_end(em) - 1,
3690 EXTENT_LOCKED | EXTENT_WRITEBACK,
3692 remove_extent_mapping(map, em);
3693 /* once for the rb tree */
3694 free_extent_map(em);
3696 start = extent_map_end(em);
3697 write_unlock(&map->lock);
3700 free_extent_map(em);
3703 return try_release_extent_state(map, tree, page, mask);
3707 * helper function for fiemap, which doesn't want to see any holes.
3708 * This maps until we find something past 'last'
3710 static struct extent_map *get_extent_skip_holes(struct inode *inode,
3713 get_extent_t *get_extent)
3715 u64 sectorsize = BTRFS_I(inode)->root->sectorsize;
3716 struct extent_map *em;
3723 len = last - offset;
3726 len = (len + sectorsize - 1) & ~(sectorsize - 1);
3727 em = get_extent(inode, NULL, 0, offset, len, 0);
3728 if (IS_ERR_OR_NULL(em))
3731 /* if this isn't a hole return it */
3732 if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) &&
3733 em->block_start != EXTENT_MAP_HOLE) {
3737 /* this is a hole, advance to the next extent */
3738 offset = extent_map_end(em);
3739 free_extent_map(em);
3746 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
3747 __u64 start, __u64 len, get_extent_t *get_extent)
3751 u64 max = start + len;
3755 u64 last_for_get_extent = 0;
3757 u64 isize = i_size_read(inode);
3758 struct btrfs_key found_key;
3759 struct extent_map *em = NULL;
3760 struct extent_state *cached_state = NULL;
3761 struct btrfs_path *path;
3762 struct btrfs_file_extent_item *item;
3767 unsigned long emflags;
3772 path = btrfs_alloc_path();
3775 path->leave_spinning = 1;
3777 start = ALIGN(start, BTRFS_I(inode)->root->sectorsize);
3778 len = ALIGN(len, BTRFS_I(inode)->root->sectorsize);
3781 * lookup the last file extent. We're not using i_size here
3782 * because there might be preallocation past i_size
3784 ret = btrfs_lookup_file_extent(NULL, BTRFS_I(inode)->root,
3785 path, btrfs_ino(inode), -1, 0);
3787 btrfs_free_path(path);
3792 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3793 struct btrfs_file_extent_item);
3794 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
3795 found_type = btrfs_key_type(&found_key);
3797 /* No extents, but there might be delalloc bits */
3798 if (found_key.objectid != btrfs_ino(inode) ||
3799 found_type != BTRFS_EXTENT_DATA_KEY) {
3800 /* have to trust i_size as the end */
3802 last_for_get_extent = isize;
3805 * remember the start of the last extent. There are a
3806 * bunch of different factors that go into the length of the
3807 * extent, so its much less complex to remember where it started
3809 last = found_key.offset;
3810 last_for_get_extent = last + 1;
3812 btrfs_free_path(path);
3815 * we might have some extents allocated but more delalloc past those
3816 * extents. so, we trust isize unless the start of the last extent is
3821 last_for_get_extent = isize;
3824 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len, 0,
3827 em = get_extent_skip_holes(inode, start, last_for_get_extent,
3837 u64 offset_in_extent;
3839 /* break if the extent we found is outside the range */
3840 if (em->start >= max || extent_map_end(em) < off)
3844 * get_extent may return an extent that starts before our
3845 * requested range. We have to make sure the ranges
3846 * we return to fiemap always move forward and don't
3847 * overlap, so adjust the offsets here
3849 em_start = max(em->start, off);
3852 * record the offset from the start of the extent
3853 * for adjusting the disk offset below
3855 offset_in_extent = em_start - em->start;
3856 em_end = extent_map_end(em);
3857 em_len = em_end - em_start;
3858 emflags = em->flags;
3863 * bump off for our next call to get_extent
3865 off = extent_map_end(em);
3869 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
3871 flags |= FIEMAP_EXTENT_LAST;
3872 } else if (em->block_start == EXTENT_MAP_INLINE) {
3873 flags |= (FIEMAP_EXTENT_DATA_INLINE |
3874 FIEMAP_EXTENT_NOT_ALIGNED);
3875 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
3876 flags |= (FIEMAP_EXTENT_DELALLOC |
3877 FIEMAP_EXTENT_UNKNOWN);
3879 disko = em->block_start + offset_in_extent;
3881 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
3882 flags |= FIEMAP_EXTENT_ENCODED;
3884 free_extent_map(em);
3886 if ((em_start >= last) || em_len == (u64)-1 ||
3887 (last == (u64)-1 && isize <= em_end)) {
3888 flags |= FIEMAP_EXTENT_LAST;
3892 /* now scan forward to see if this is really the last extent. */
3893 em = get_extent_skip_holes(inode, off, last_for_get_extent,
3900 flags |= FIEMAP_EXTENT_LAST;
3903 ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
3909 free_extent_map(em);
3911 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len,
3912 &cached_state, GFP_NOFS);
3916 inline struct page *extent_buffer_page(struct extent_buffer *eb,
3919 return eb->pages[i];
3922 inline unsigned long num_extent_pages(u64 start, u64 len)
3924 return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
3925 (start >> PAGE_CACHE_SHIFT);
3928 static void __free_extent_buffer(struct extent_buffer *eb)
3931 unsigned long flags;
3932 spin_lock_irqsave(&leak_lock, flags);
3933 list_del(&eb->leak_list);
3934 spin_unlock_irqrestore(&leak_lock, flags);
3936 if (eb->pages && eb->pages != eb->inline_pages)
3938 kmem_cache_free(extent_buffer_cache, eb);
3941 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
3946 struct extent_buffer *eb = NULL;
3948 unsigned long flags;
3951 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
3958 rwlock_init(&eb->lock);
3959 atomic_set(&eb->write_locks, 0);
3960 atomic_set(&eb->read_locks, 0);
3961 atomic_set(&eb->blocking_readers, 0);
3962 atomic_set(&eb->blocking_writers, 0);
3963 atomic_set(&eb->spinning_readers, 0);
3964 atomic_set(&eb->spinning_writers, 0);
3965 eb->lock_nested = 0;
3966 init_waitqueue_head(&eb->write_lock_wq);
3967 init_waitqueue_head(&eb->read_lock_wq);
3970 spin_lock_irqsave(&leak_lock, flags);
3971 list_add(&eb->leak_list, &buffers);
3972 spin_unlock_irqrestore(&leak_lock, flags);
3974 spin_lock_init(&eb->refs_lock);
3975 atomic_set(&eb->refs, 1);
3976 atomic_set(&eb->io_pages, 0);
3978 if (len > MAX_INLINE_EXTENT_BUFFER_SIZE) {
3979 struct page **pages;
3980 int num_pages = (len + PAGE_CACHE_SIZE - 1) >>
3982 pages = kzalloc(num_pages, mask);
3984 __free_extent_buffer(eb);
3989 eb->pages = eb->inline_pages;
3995 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src)
3999 struct extent_buffer *new;
4000 unsigned long num_pages = num_extent_pages(src->start, src->len);
4002 new = __alloc_extent_buffer(NULL, src->start, src->len, GFP_ATOMIC);
4006 for (i = 0; i < num_pages; i++) {
4007 p = alloc_page(GFP_ATOMIC);
4009 attach_extent_buffer_page(new, p);
4010 WARN_ON(PageDirty(p));
4015 copy_extent_buffer(new, src, 0, 0, src->len);
4016 set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
4017 set_bit(EXTENT_BUFFER_DUMMY, &new->bflags);
4022 struct extent_buffer *alloc_dummy_extent_buffer(u64 start, unsigned long len)
4024 struct extent_buffer *eb;
4025 unsigned long num_pages = num_extent_pages(0, len);
4028 eb = __alloc_extent_buffer(NULL, start, len, GFP_ATOMIC);
4032 for (i = 0; i < num_pages; i++) {
4033 eb->pages[i] = alloc_page(GFP_ATOMIC);
4037 set_extent_buffer_uptodate(eb);
4038 btrfs_set_header_nritems(eb, 0);
4039 set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4043 for (i--; i > 0; i--)
4044 __free_page(eb->pages[i]);
4045 __free_extent_buffer(eb);
4049 static int extent_buffer_under_io(struct extent_buffer *eb)
4051 return (atomic_read(&eb->io_pages) ||
4052 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4053 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4057 * Helper for releasing extent buffer page.
4059 static void btrfs_release_extent_buffer_page(struct extent_buffer *eb,
4060 unsigned long start_idx)
4062 unsigned long index;
4063 unsigned long num_pages;
4065 int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4067 BUG_ON(extent_buffer_under_io(eb));
4069 num_pages = num_extent_pages(eb->start, eb->len);
4070 index = start_idx + num_pages;
4071 if (start_idx >= index)
4076 page = extent_buffer_page(eb, index);
4077 if (page && mapped) {
4078 spin_lock(&page->mapping->private_lock);
4080 * We do this since we'll remove the pages after we've
4081 * removed the eb from the radix tree, so we could race
4082 * and have this page now attached to the new eb. So
4083 * only clear page_private if it's still connected to
4086 if (PagePrivate(page) &&
4087 page->private == (unsigned long)eb) {
4088 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4089 BUG_ON(PageDirty(page));
4090 BUG_ON(PageWriteback(page));
4092 * We need to make sure we haven't be attached
4095 ClearPagePrivate(page);
4096 set_page_private(page, 0);
4097 /* One for the page private */
4098 page_cache_release(page);
4100 spin_unlock(&page->mapping->private_lock);
4104 /* One for when we alloced the page */
4105 page_cache_release(page);
4107 } while (index != start_idx);
4111 * Helper for releasing the extent buffer.
4113 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4115 btrfs_release_extent_buffer_page(eb, 0);
4116 __free_extent_buffer(eb);
4119 static void check_buffer_tree_ref(struct extent_buffer *eb)
4121 /* the ref bit is tricky. We have to make sure it is set
4122 * if we have the buffer dirty. Otherwise the
4123 * code to free a buffer can end up dropping a dirty
4126 * Once the ref bit is set, it won't go away while the
4127 * buffer is dirty or in writeback, and it also won't
4128 * go away while we have the reference count on the
4131 * We can't just set the ref bit without bumping the
4132 * ref on the eb because free_extent_buffer might
4133 * see the ref bit and try to clear it. If this happens
4134 * free_extent_buffer might end up dropping our original
4135 * ref by mistake and freeing the page before we are able
4136 * to add one more ref.
4138 * So bump the ref count first, then set the bit. If someone
4139 * beat us to it, drop the ref we added.
4141 spin_lock(&eb->refs_lock);
4142 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4143 atomic_inc(&eb->refs);
4144 spin_unlock(&eb->refs_lock);
4147 static void mark_extent_buffer_accessed(struct extent_buffer *eb)
4149 unsigned long num_pages, i;
4151 check_buffer_tree_ref(eb);
4153 num_pages = num_extent_pages(eb->start, eb->len);
4154 for (i = 0; i < num_pages; i++) {
4155 struct page *p = extent_buffer_page(eb, i);
4156 mark_page_accessed(p);
4160 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
4161 u64 start, unsigned long len)
4163 unsigned long num_pages = num_extent_pages(start, len);
4165 unsigned long index = start >> PAGE_CACHE_SHIFT;
4166 struct extent_buffer *eb;
4167 struct extent_buffer *exists = NULL;
4169 struct address_space *mapping = tree->mapping;
4174 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
4175 if (eb && atomic_inc_not_zero(&eb->refs)) {
4177 mark_extent_buffer_accessed(eb);
4182 eb = __alloc_extent_buffer(tree, start, len, GFP_NOFS);
4186 for (i = 0; i < num_pages; i++, index++) {
4187 p = find_or_create_page(mapping, index, GFP_NOFS);
4193 spin_lock(&mapping->private_lock);
4194 if (PagePrivate(p)) {
4196 * We could have already allocated an eb for this page
4197 * and attached one so lets see if we can get a ref on
4198 * the existing eb, and if we can we know it's good and
4199 * we can just return that one, else we know we can just
4200 * overwrite page->private.
4202 exists = (struct extent_buffer *)p->private;
4203 if (atomic_inc_not_zero(&exists->refs)) {
4204 spin_unlock(&mapping->private_lock);
4206 page_cache_release(p);
4207 mark_extent_buffer_accessed(exists);
4212 * Do this so attach doesn't complain and we need to
4213 * drop the ref the old guy had.
4215 ClearPagePrivate(p);
4216 WARN_ON(PageDirty(p));
4217 page_cache_release(p);
4219 attach_extent_buffer_page(eb, p);
4220 spin_unlock(&mapping->private_lock);
4221 WARN_ON(PageDirty(p));
4222 mark_page_accessed(p);
4224 if (!PageUptodate(p))
4228 * see below about how we avoid a nasty race with release page
4229 * and why we unlock later
4233 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4235 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
4239 spin_lock(&tree->buffer_lock);
4240 ret = radix_tree_insert(&tree->buffer, start >> PAGE_CACHE_SHIFT, eb);
4241 if (ret == -EEXIST) {
4242 exists = radix_tree_lookup(&tree->buffer,
4243 start >> PAGE_CACHE_SHIFT);
4244 if (!atomic_inc_not_zero(&exists->refs)) {
4245 spin_unlock(&tree->buffer_lock);
4246 radix_tree_preload_end();
4250 spin_unlock(&tree->buffer_lock);
4251 radix_tree_preload_end();
4252 mark_extent_buffer_accessed(exists);
4255 /* add one reference for the tree */
4256 check_buffer_tree_ref(eb);
4257 spin_unlock(&tree->buffer_lock);
4258 radix_tree_preload_end();
4261 * there is a race where release page may have
4262 * tried to find this extent buffer in the radix
4263 * but failed. It will tell the VM it is safe to
4264 * reclaim the, and it will clear the page private bit.
4265 * We must make sure to set the page private bit properly
4266 * after the extent buffer is in the radix tree so
4267 * it doesn't get lost
4269 SetPageChecked(eb->pages[0]);
4270 for (i = 1; i < num_pages; i++) {
4271 p = extent_buffer_page(eb, i);
4272 ClearPageChecked(p);
4275 unlock_page(eb->pages[0]);
4279 for (i = 0; i < num_pages; i++) {
4281 unlock_page(eb->pages[i]);
4284 WARN_ON(!atomic_dec_and_test(&eb->refs));
4285 btrfs_release_extent_buffer(eb);
4289 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
4290 u64 start, unsigned long len)
4292 struct extent_buffer *eb;
4295 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
4296 if (eb && atomic_inc_not_zero(&eb->refs)) {
4298 mark_extent_buffer_accessed(eb);
4306 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
4308 struct extent_buffer *eb =
4309 container_of(head, struct extent_buffer, rcu_head);
4311 __free_extent_buffer(eb);
4314 /* Expects to have eb->eb_lock already held */
4315 static int release_extent_buffer(struct extent_buffer *eb, gfp_t mask)
4317 WARN_ON(atomic_read(&eb->refs) == 0);
4318 if (atomic_dec_and_test(&eb->refs)) {
4319 if (test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags)) {
4320 spin_unlock(&eb->refs_lock);
4322 struct extent_io_tree *tree = eb->tree;
4324 spin_unlock(&eb->refs_lock);
4326 spin_lock(&tree->buffer_lock);
4327 radix_tree_delete(&tree->buffer,
4328 eb->start >> PAGE_CACHE_SHIFT);
4329 spin_unlock(&tree->buffer_lock);
4332 /* Should be safe to release our pages at this point */
4333 btrfs_release_extent_buffer_page(eb, 0);
4335 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
4338 spin_unlock(&eb->refs_lock);
4343 void free_extent_buffer(struct extent_buffer *eb)
4348 spin_lock(&eb->refs_lock);
4349 if (atomic_read(&eb->refs) == 2 &&
4350 test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))
4351 atomic_dec(&eb->refs);
4353 if (atomic_read(&eb->refs) == 2 &&
4354 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
4355 !extent_buffer_under_io(eb) &&
4356 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4357 atomic_dec(&eb->refs);
4360 * I know this is terrible, but it's temporary until we stop tracking
4361 * the uptodate bits and such for the extent buffers.
4363 release_extent_buffer(eb, GFP_ATOMIC);
4366 void free_extent_buffer_stale(struct extent_buffer *eb)
4371 spin_lock(&eb->refs_lock);
4372 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
4374 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
4375 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4376 atomic_dec(&eb->refs);
4377 release_extent_buffer(eb, GFP_NOFS);
4380 void clear_extent_buffer_dirty(struct extent_buffer *eb)
4383 unsigned long num_pages;
4386 num_pages = num_extent_pages(eb->start, eb->len);
4388 for (i = 0; i < num_pages; i++) {
4389 page = extent_buffer_page(eb, i);
4390 if (!PageDirty(page))
4394 WARN_ON(!PagePrivate(page));
4396 clear_page_dirty_for_io(page);
4397 spin_lock_irq(&page->mapping->tree_lock);
4398 if (!PageDirty(page)) {
4399 radix_tree_tag_clear(&page->mapping->page_tree,
4401 PAGECACHE_TAG_DIRTY);
4403 spin_unlock_irq(&page->mapping->tree_lock);
4404 ClearPageError(page);
4407 WARN_ON(atomic_read(&eb->refs) == 0);
4410 int set_extent_buffer_dirty(struct extent_buffer *eb)
4413 unsigned long num_pages;
4416 check_buffer_tree_ref(eb);
4418 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
4420 num_pages = num_extent_pages(eb->start, eb->len);
4421 WARN_ON(atomic_read(&eb->refs) == 0);
4422 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
4424 for (i = 0; i < num_pages; i++)
4425 set_page_dirty(extent_buffer_page(eb, i));
4429 static int range_straddles_pages(u64 start, u64 len)
4431 if (len < PAGE_CACHE_SIZE)
4433 if (start & (PAGE_CACHE_SIZE - 1))
4435 if ((start + len) & (PAGE_CACHE_SIZE - 1))
4440 int clear_extent_buffer_uptodate(struct extent_buffer *eb)
4444 unsigned long num_pages;
4446 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4447 num_pages = num_extent_pages(eb->start, eb->len);
4448 for (i = 0; i < num_pages; i++) {
4449 page = extent_buffer_page(eb, i);
4451 ClearPageUptodate(page);
4456 int set_extent_buffer_uptodate(struct extent_buffer *eb)
4460 unsigned long num_pages;
4462 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4463 num_pages = num_extent_pages(eb->start, eb->len);
4464 for (i = 0; i < num_pages; i++) {
4465 page = extent_buffer_page(eb, i);
4466 SetPageUptodate(page);
4471 int extent_range_uptodate(struct extent_io_tree *tree,
4476 int pg_uptodate = 1;
4478 unsigned long index;
4480 if (range_straddles_pages(start, end - start + 1)) {
4481 ret = test_range_bit(tree, start, end,
4482 EXTENT_UPTODATE, 1, NULL);
4486 while (start <= end) {
4487 index = start >> PAGE_CACHE_SHIFT;
4488 page = find_get_page(tree->mapping, index);
4491 uptodate = PageUptodate(page);
4492 page_cache_release(page);
4497 start += PAGE_CACHE_SIZE;
4502 int extent_buffer_uptodate(struct extent_buffer *eb)
4504 return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4507 int read_extent_buffer_pages(struct extent_io_tree *tree,
4508 struct extent_buffer *eb, u64 start, int wait,
4509 get_extent_t *get_extent, int mirror_num)
4512 unsigned long start_i;
4516 int locked_pages = 0;
4517 int all_uptodate = 1;
4518 unsigned long num_pages;
4519 unsigned long num_reads = 0;
4520 struct bio *bio = NULL;
4521 unsigned long bio_flags = 0;
4523 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
4527 WARN_ON(start < eb->start);
4528 start_i = (start >> PAGE_CACHE_SHIFT) -
4529 (eb->start >> PAGE_CACHE_SHIFT);
4534 num_pages = num_extent_pages(eb->start, eb->len);
4535 for (i = start_i; i < num_pages; i++) {
4536 page = extent_buffer_page(eb, i);
4537 if (wait == WAIT_NONE) {
4538 if (!trylock_page(page))
4544 if (!PageUptodate(page)) {
4551 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4555 clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
4556 eb->read_mirror = 0;
4557 atomic_set(&eb->io_pages, num_reads);
4558 for (i = start_i; i < num_pages; i++) {
4559 page = extent_buffer_page(eb, i);
4560 if (!PageUptodate(page)) {
4561 ClearPageError(page);
4562 err = __extent_read_full_page(tree, page,
4564 mirror_num, &bio_flags);
4573 err = submit_one_bio(READ, bio, mirror_num, bio_flags);
4578 if (ret || wait != WAIT_COMPLETE)
4581 for (i = start_i; i < num_pages; i++) {
4582 page = extent_buffer_page(eb, i);
4583 wait_on_page_locked(page);
4584 if (!PageUptodate(page))
4592 while (locked_pages > 0) {
4593 page = extent_buffer_page(eb, i);
4601 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
4602 unsigned long start,
4609 char *dst = (char *)dstv;
4610 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4611 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4613 WARN_ON(start > eb->len);
4614 WARN_ON(start + len > eb->start + eb->len);
4616 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4619 page = extent_buffer_page(eb, i);
4621 cur = min(len, (PAGE_CACHE_SIZE - offset));
4622 kaddr = page_address(page);
4623 memcpy(dst, kaddr + offset, cur);
4632 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
4633 unsigned long min_len, char **map,
4634 unsigned long *map_start,
4635 unsigned long *map_len)
4637 size_t offset = start & (PAGE_CACHE_SIZE - 1);
4640 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4641 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4642 unsigned long end_i = (start_offset + start + min_len - 1) >>
4649 offset = start_offset;
4653 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
4656 if (start + min_len > eb->len) {
4657 printk(KERN_ERR "btrfs bad mapping eb start %llu len %lu, "
4658 "wanted %lu %lu\n", (unsigned long long)eb->start,
4659 eb->len, start, min_len);
4664 p = extent_buffer_page(eb, i);
4665 kaddr = page_address(p);
4666 *map = kaddr + offset;
4667 *map_len = PAGE_CACHE_SIZE - offset;
4671 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
4672 unsigned long start,
4679 char *ptr = (char *)ptrv;
4680 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4681 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4684 WARN_ON(start > eb->len);
4685 WARN_ON(start + len > eb->start + eb->len);
4687 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4690 page = extent_buffer_page(eb, i);
4692 cur = min(len, (PAGE_CACHE_SIZE - offset));
4694 kaddr = page_address(page);
4695 ret = memcmp(ptr, kaddr + offset, cur);
4707 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
4708 unsigned long start, unsigned long len)
4714 char *src = (char *)srcv;
4715 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4716 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4718 WARN_ON(start > eb->len);
4719 WARN_ON(start + len > eb->start + eb->len);
4721 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4724 page = extent_buffer_page(eb, i);
4725 WARN_ON(!PageUptodate(page));
4727 cur = min(len, PAGE_CACHE_SIZE - offset);
4728 kaddr = page_address(page);
4729 memcpy(kaddr + offset, src, cur);
4738 void memset_extent_buffer(struct extent_buffer *eb, char c,
4739 unsigned long start, unsigned long len)
4745 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4746 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4748 WARN_ON(start > eb->len);
4749 WARN_ON(start + len > eb->start + eb->len);
4751 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4754 page = extent_buffer_page(eb, i);
4755 WARN_ON(!PageUptodate(page));
4757 cur = min(len, PAGE_CACHE_SIZE - offset);
4758 kaddr = page_address(page);
4759 memset(kaddr + offset, c, cur);
4767 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
4768 unsigned long dst_offset, unsigned long src_offset,
4771 u64 dst_len = dst->len;
4776 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4777 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
4779 WARN_ON(src->len != dst_len);
4781 offset = (start_offset + dst_offset) &
4782 ((unsigned long)PAGE_CACHE_SIZE - 1);
4785 page = extent_buffer_page(dst, i);
4786 WARN_ON(!PageUptodate(page));
4788 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
4790 kaddr = page_address(page);
4791 read_extent_buffer(src, kaddr + offset, src_offset, cur);
4800 static void move_pages(struct page *dst_page, struct page *src_page,
4801 unsigned long dst_off, unsigned long src_off,
4804 char *dst_kaddr = page_address(dst_page);
4805 if (dst_page == src_page) {
4806 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
4808 char *src_kaddr = page_address(src_page);
4809 char *p = dst_kaddr + dst_off + len;
4810 char *s = src_kaddr + src_off + len;
4817 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
4819 unsigned long distance = (src > dst) ? src - dst : dst - src;
4820 return distance < len;
4823 static void copy_pages(struct page *dst_page, struct page *src_page,
4824 unsigned long dst_off, unsigned long src_off,
4827 char *dst_kaddr = page_address(dst_page);
4829 int must_memmove = 0;
4831 if (dst_page != src_page) {
4832 src_kaddr = page_address(src_page);
4834 src_kaddr = dst_kaddr;
4835 if (areas_overlap(src_off, dst_off, len))
4840 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
4842 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
4845 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
4846 unsigned long src_offset, unsigned long len)
4849 size_t dst_off_in_page;
4850 size_t src_off_in_page;
4851 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4852 unsigned long dst_i;
4853 unsigned long src_i;
4855 if (src_offset + len > dst->len) {
4856 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
4857 "len %lu dst len %lu\n", src_offset, len, dst->len);
4860 if (dst_offset + len > dst->len) {
4861 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
4862 "len %lu dst len %lu\n", dst_offset, len, dst->len);
4867 dst_off_in_page = (start_offset + dst_offset) &
4868 ((unsigned long)PAGE_CACHE_SIZE - 1);
4869 src_off_in_page = (start_offset + src_offset) &
4870 ((unsigned long)PAGE_CACHE_SIZE - 1);
4872 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
4873 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
4875 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
4877 cur = min_t(unsigned long, cur,
4878 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
4880 copy_pages(extent_buffer_page(dst, dst_i),
4881 extent_buffer_page(dst, src_i),
4882 dst_off_in_page, src_off_in_page, cur);
4890 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
4891 unsigned long src_offset, unsigned long len)
4894 size_t dst_off_in_page;
4895 size_t src_off_in_page;
4896 unsigned long dst_end = dst_offset + len - 1;
4897 unsigned long src_end = src_offset + len - 1;
4898 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4899 unsigned long dst_i;
4900 unsigned long src_i;
4902 if (src_offset + len > dst->len) {
4903 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
4904 "len %lu len %lu\n", src_offset, len, dst->len);
4907 if (dst_offset + len > dst->len) {
4908 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
4909 "len %lu len %lu\n", dst_offset, len, dst->len);
4912 if (dst_offset < src_offset) {
4913 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
4917 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
4918 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
4920 dst_off_in_page = (start_offset + dst_end) &
4921 ((unsigned long)PAGE_CACHE_SIZE - 1);
4922 src_off_in_page = (start_offset + src_end) &
4923 ((unsigned long)PAGE_CACHE_SIZE - 1);
4925 cur = min_t(unsigned long, len, src_off_in_page + 1);
4926 cur = min(cur, dst_off_in_page + 1);
4927 move_pages(extent_buffer_page(dst, dst_i),
4928 extent_buffer_page(dst, src_i),
4929 dst_off_in_page - cur + 1,
4930 src_off_in_page - cur + 1, cur);
4938 int try_release_extent_buffer(struct page *page, gfp_t mask)
4940 struct extent_buffer *eb;
4943 * We need to make sure noboody is attaching this page to an eb right
4946 spin_lock(&page->mapping->private_lock);
4947 if (!PagePrivate(page)) {
4948 spin_unlock(&page->mapping->private_lock);
4952 eb = (struct extent_buffer *)page->private;
4956 * This is a little awful but should be ok, we need to make sure that
4957 * the eb doesn't disappear out from under us while we're looking at
4960 spin_lock(&eb->refs_lock);
4961 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
4962 spin_unlock(&eb->refs_lock);
4963 spin_unlock(&page->mapping->private_lock);
4966 spin_unlock(&page->mapping->private_lock);
4968 if ((mask & GFP_NOFS) == GFP_NOFS)
4972 * If tree ref isn't set then we know the ref on this eb is a real ref,
4973 * so just return, this page will likely be freed soon anyway.
4975 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
4976 spin_unlock(&eb->refs_lock);
4980 return release_extent_buffer(eb, mask);