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 static struct kmem_cache *extent_state_cache;
24 static struct kmem_cache *extent_buffer_cache;
26 static LIST_HEAD(buffers);
27 static LIST_HEAD(states);
31 static DEFINE_SPINLOCK(leak_lock);
34 #define BUFFER_LRU_MAX 64
39 struct rb_node rb_node;
42 struct extent_page_data {
44 struct extent_io_tree *tree;
45 get_extent_t *get_extent;
47 /* tells writepage not to lock the state bits for this range
48 * it still does the unlocking
50 unsigned int extent_locked:1;
52 /* tells the submit_bio code to use a WRITE_SYNC */
53 unsigned int sync_io:1;
56 static inline struct btrfs_fs_info *
57 tree_fs_info(struct extent_io_tree *tree)
59 return btrfs_sb(tree->mapping->host->i_sb);
62 int __init extent_io_init(void)
64 extent_state_cache = kmem_cache_create("extent_state",
65 sizeof(struct extent_state), 0,
66 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
67 if (!extent_state_cache)
70 extent_buffer_cache = kmem_cache_create("extent_buffers",
71 sizeof(struct extent_buffer), 0,
72 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
73 if (!extent_buffer_cache)
74 goto free_state_cache;
78 kmem_cache_destroy(extent_state_cache);
82 void extent_io_exit(void)
84 struct extent_state *state;
85 struct extent_buffer *eb;
87 while (!list_empty(&states)) {
88 state = list_entry(states.next, struct extent_state, leak_list);
89 printk(KERN_ERR "btrfs state leak: start %llu end %llu "
90 "state %lu in tree %p refs %d\n",
91 (unsigned long long)state->start,
92 (unsigned long long)state->end,
93 state->state, state->tree, atomic_read(&state->refs));
94 list_del(&state->leak_list);
95 kmem_cache_free(extent_state_cache, state);
99 while (!list_empty(&buffers)) {
100 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
101 printk(KERN_ERR "btrfs buffer leak start %llu len %lu "
102 "refs %d\n", (unsigned long long)eb->start,
103 eb->len, atomic_read(&eb->refs));
104 list_del(&eb->leak_list);
105 kmem_cache_free(extent_buffer_cache, eb);
107 if (extent_state_cache)
108 kmem_cache_destroy(extent_state_cache);
109 if (extent_buffer_cache)
110 kmem_cache_destroy(extent_buffer_cache);
113 void extent_io_tree_init(struct extent_io_tree *tree,
114 struct address_space *mapping)
116 tree->state = RB_ROOT;
117 INIT_RADIX_TREE(&tree->buffer, GFP_ATOMIC);
119 tree->dirty_bytes = 0;
120 spin_lock_init(&tree->lock);
121 spin_lock_init(&tree->buffer_lock);
122 tree->mapping = mapping;
125 static struct extent_state *alloc_extent_state(gfp_t mask)
127 struct extent_state *state;
132 state = kmem_cache_alloc(extent_state_cache, mask);
139 spin_lock_irqsave(&leak_lock, flags);
140 list_add(&state->leak_list, &states);
141 spin_unlock_irqrestore(&leak_lock, flags);
143 atomic_set(&state->refs, 1);
144 init_waitqueue_head(&state->wq);
145 trace_alloc_extent_state(state, mask, _RET_IP_);
149 void free_extent_state(struct extent_state *state)
153 if (atomic_dec_and_test(&state->refs)) {
157 WARN_ON(state->tree);
159 spin_lock_irqsave(&leak_lock, flags);
160 list_del(&state->leak_list);
161 spin_unlock_irqrestore(&leak_lock, flags);
163 trace_free_extent_state(state, _RET_IP_);
164 kmem_cache_free(extent_state_cache, state);
168 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
169 struct rb_node *node)
171 struct rb_node **p = &root->rb_node;
172 struct rb_node *parent = NULL;
173 struct tree_entry *entry;
177 entry = rb_entry(parent, struct tree_entry, rb_node);
179 if (offset < entry->start)
181 else if (offset > entry->end)
187 entry = rb_entry(node, struct tree_entry, rb_node);
188 rb_link_node(node, parent, p);
189 rb_insert_color(node, root);
193 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
194 struct rb_node **prev_ret,
195 struct rb_node **next_ret)
197 struct rb_root *root = &tree->state;
198 struct rb_node *n = root->rb_node;
199 struct rb_node *prev = NULL;
200 struct rb_node *orig_prev = NULL;
201 struct tree_entry *entry;
202 struct tree_entry *prev_entry = NULL;
205 entry = rb_entry(n, struct tree_entry, rb_node);
209 if (offset < entry->start)
211 else if (offset > entry->end)
219 while (prev && offset > prev_entry->end) {
220 prev = rb_next(prev);
221 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
228 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
229 while (prev && offset < prev_entry->start) {
230 prev = rb_prev(prev);
231 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
238 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
241 struct rb_node *prev = NULL;
244 ret = __etree_search(tree, offset, &prev, NULL);
250 static void merge_cb(struct extent_io_tree *tree, struct extent_state *new,
251 struct extent_state *other)
253 if (tree->ops && tree->ops->merge_extent_hook)
254 tree->ops->merge_extent_hook(tree->mapping->host, new,
259 * utility function to look for merge candidates inside a given range.
260 * Any extents with matching state are merged together into a single
261 * extent in the tree. Extents with EXTENT_IO in their state field
262 * are not merged because the end_io handlers need to be able to do
263 * operations on them without sleeping (or doing allocations/splits).
265 * This should be called with the tree lock held.
267 static void merge_state(struct extent_io_tree *tree,
268 struct extent_state *state)
270 struct extent_state *other;
271 struct rb_node *other_node;
273 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
276 other_node = rb_prev(&state->rb_node);
278 other = rb_entry(other_node, struct extent_state, rb_node);
279 if (other->end == state->start - 1 &&
280 other->state == state->state) {
281 merge_cb(tree, state, other);
282 state->start = other->start;
284 rb_erase(&other->rb_node, &tree->state);
285 free_extent_state(other);
288 other_node = rb_next(&state->rb_node);
290 other = rb_entry(other_node, struct extent_state, rb_node);
291 if (other->start == state->end + 1 &&
292 other->state == state->state) {
293 merge_cb(tree, state, other);
294 state->end = other->end;
296 rb_erase(&other->rb_node, &tree->state);
297 free_extent_state(other);
302 static void set_state_cb(struct extent_io_tree *tree,
303 struct extent_state *state, int *bits)
305 if (tree->ops && tree->ops->set_bit_hook)
306 tree->ops->set_bit_hook(tree->mapping->host, state, bits);
309 static void clear_state_cb(struct extent_io_tree *tree,
310 struct extent_state *state, int *bits)
312 if (tree->ops && tree->ops->clear_bit_hook)
313 tree->ops->clear_bit_hook(tree->mapping->host, state, bits);
316 static void set_state_bits(struct extent_io_tree *tree,
317 struct extent_state *state, int *bits);
320 * insert an extent_state struct into the tree. 'bits' are set on the
321 * struct before it is inserted.
323 * This may return -EEXIST if the extent is already there, in which case the
324 * state struct is freed.
326 * The tree lock is not taken internally. This is a utility function and
327 * probably isn't what you want to call (see set/clear_extent_bit).
329 static int insert_state(struct extent_io_tree *tree,
330 struct extent_state *state, u64 start, u64 end,
333 struct rb_node *node;
336 printk(KERN_ERR "btrfs end < start %llu %llu\n",
337 (unsigned long long)end,
338 (unsigned long long)start);
341 state->start = start;
344 set_state_bits(tree, state, bits);
346 node = tree_insert(&tree->state, end, &state->rb_node);
348 struct extent_state *found;
349 found = rb_entry(node, struct extent_state, rb_node);
350 printk(KERN_ERR "btrfs found node %llu %llu on insert of "
351 "%llu %llu\n", (unsigned long long)found->start,
352 (unsigned long long)found->end,
353 (unsigned long long)start, (unsigned long long)end);
357 merge_state(tree, state);
361 static void split_cb(struct extent_io_tree *tree, struct extent_state *orig,
364 if (tree->ops && tree->ops->split_extent_hook)
365 tree->ops->split_extent_hook(tree->mapping->host, orig, split);
369 * split a given extent state struct in two, inserting the preallocated
370 * struct 'prealloc' as the newly created second half. 'split' indicates an
371 * offset inside 'orig' where it should be split.
374 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
375 * are two extent state structs in the tree:
376 * prealloc: [orig->start, split - 1]
377 * orig: [ split, orig->end ]
379 * The tree locks are not taken by this function. They need to be held
382 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
383 struct extent_state *prealloc, u64 split)
385 struct rb_node *node;
387 split_cb(tree, orig, split);
389 prealloc->start = orig->start;
390 prealloc->end = split - 1;
391 prealloc->state = orig->state;
394 node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
396 free_extent_state(prealloc);
399 prealloc->tree = tree;
404 * utility function to clear some bits in an extent state struct.
405 * it will optionally wake up any one waiting on this state (wake == 1), or
406 * forcibly remove the state from the tree (delete == 1).
408 * If no bits are set on the state struct after clearing things, the
409 * struct is freed and removed from the tree
411 static int clear_state_bit(struct extent_io_tree *tree,
412 struct extent_state *state,
415 int bits_to_clear = *bits & ~EXTENT_CTLBITS;
416 int ret = state->state & bits_to_clear;
418 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
419 u64 range = state->end - state->start + 1;
420 WARN_ON(range > tree->dirty_bytes);
421 tree->dirty_bytes -= range;
423 clear_state_cb(tree, state, bits);
424 state->state &= ~bits_to_clear;
427 if (state->state == 0) {
429 rb_erase(&state->rb_node, &tree->state);
431 free_extent_state(state);
436 merge_state(tree, state);
441 static struct extent_state *
442 alloc_extent_state_atomic(struct extent_state *prealloc)
445 prealloc = alloc_extent_state(GFP_ATOMIC);
450 void extent_io_tree_panic(struct extent_io_tree *tree, int err)
452 btrfs_panic(tree_fs_info(tree), err, "Locking error: "
453 "Extent tree was modified by another "
454 "thread while locked.");
458 * clear some bits on a range in the tree. This may require splitting
459 * or inserting elements in the tree, so the gfp mask is used to
460 * indicate which allocations or sleeping are allowed.
462 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
463 * the given range from the tree regardless of state (ie for truncate).
465 * the range [start, end] is inclusive.
467 * This takes the tree lock, and returns 0 on success and < 0 on error.
469 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
470 int bits, int wake, int delete,
471 struct extent_state **cached_state,
474 struct extent_state *state;
475 struct extent_state *cached;
476 struct extent_state *prealloc = NULL;
477 struct rb_node *next_node;
478 struct rb_node *node;
484 bits |= ~EXTENT_CTLBITS;
485 bits |= EXTENT_FIRST_DELALLOC;
487 if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY))
490 if (!prealloc && (mask & __GFP_WAIT)) {
491 prealloc = alloc_extent_state(mask);
496 spin_lock(&tree->lock);
498 cached = *cached_state;
501 *cached_state = NULL;
505 if (cached && cached->tree && cached->start <= start &&
506 cached->end > start) {
508 atomic_dec(&cached->refs);
513 free_extent_state(cached);
516 * this search will find the extents that end after
519 node = tree_search(tree, start);
522 state = rb_entry(node, struct extent_state, rb_node);
524 if (state->start > end)
526 WARN_ON(state->end < start);
527 last_end = state->end;
529 if (state->end < end && !need_resched())
530 next_node = rb_next(&state->rb_node);
534 /* the state doesn't have the wanted bits, go ahead */
535 if (!(state->state & bits))
539 * | ---- desired range ---- |
541 * | ------------- state -------------- |
543 * We need to split the extent we found, and may flip
544 * bits on second half.
546 * If the extent we found extends past our range, we
547 * just split and search again. It'll get split again
548 * the next time though.
550 * If the extent we found is inside our range, we clear
551 * the desired bit on it.
554 if (state->start < start) {
555 prealloc = alloc_extent_state_atomic(prealloc);
557 err = split_state(tree, state, prealloc, start);
559 extent_io_tree_panic(tree, err);
564 if (state->end <= end) {
565 clear_state_bit(tree, state, &bits, wake);
566 if (last_end == (u64)-1)
568 start = last_end + 1;
573 * | ---- desired range ---- |
575 * We need to split the extent, and clear the bit
578 if (state->start <= end && state->end > end) {
579 prealloc = alloc_extent_state_atomic(prealloc);
581 err = split_state(tree, state, prealloc, end + 1);
583 extent_io_tree_panic(tree, err);
588 clear_state_bit(tree, prealloc, &bits, wake);
594 clear_state_bit(tree, state, &bits, wake);
596 if (last_end == (u64)-1)
598 start = last_end + 1;
599 if (start <= end && next_node) {
600 state = rb_entry(next_node, struct extent_state,
607 spin_unlock(&tree->lock);
609 free_extent_state(prealloc);
616 spin_unlock(&tree->lock);
617 if (mask & __GFP_WAIT)
622 static void wait_on_state(struct extent_io_tree *tree,
623 struct extent_state *state)
624 __releases(tree->lock)
625 __acquires(tree->lock)
628 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
629 spin_unlock(&tree->lock);
631 spin_lock(&tree->lock);
632 finish_wait(&state->wq, &wait);
636 * waits for one or more bits to clear on a range in the state tree.
637 * The range [start, end] is inclusive.
638 * The tree lock is taken by this function
640 void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits)
642 struct extent_state *state;
643 struct rb_node *node;
645 spin_lock(&tree->lock);
649 * this search will find all the extents that end after
652 node = tree_search(tree, start);
656 state = rb_entry(node, struct extent_state, rb_node);
658 if (state->start > end)
661 if (state->state & bits) {
662 start = state->start;
663 atomic_inc(&state->refs);
664 wait_on_state(tree, state);
665 free_extent_state(state);
668 start = state->end + 1;
673 cond_resched_lock(&tree->lock);
676 spin_unlock(&tree->lock);
679 static void set_state_bits(struct extent_io_tree *tree,
680 struct extent_state *state,
683 int bits_to_set = *bits & ~EXTENT_CTLBITS;
685 set_state_cb(tree, state, bits);
686 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
687 u64 range = state->end - state->start + 1;
688 tree->dirty_bytes += range;
690 state->state |= bits_to_set;
693 static void cache_state(struct extent_state *state,
694 struct extent_state **cached_ptr)
696 if (cached_ptr && !(*cached_ptr)) {
697 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) {
699 atomic_inc(&state->refs);
704 static void uncache_state(struct extent_state **cached_ptr)
706 if (cached_ptr && (*cached_ptr)) {
707 struct extent_state *state = *cached_ptr;
709 free_extent_state(state);
714 * set some bits on a range in the tree. This may require allocations or
715 * sleeping, so the gfp mask is used to indicate what is allowed.
717 * If any of the exclusive bits are set, this will fail with -EEXIST if some
718 * part of the range already has the desired bits set. The start of the
719 * existing range is returned in failed_start in this case.
721 * [start, end] is inclusive This takes the tree lock.
724 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
725 int bits, int exclusive_bits, u64 *failed_start,
726 struct extent_state **cached_state, gfp_t mask)
728 struct extent_state *state;
729 struct extent_state *prealloc = NULL;
730 struct rb_node *node;
735 bits |= EXTENT_FIRST_DELALLOC;
737 if (!prealloc && (mask & __GFP_WAIT)) {
738 prealloc = alloc_extent_state(mask);
742 spin_lock(&tree->lock);
743 if (cached_state && *cached_state) {
744 state = *cached_state;
745 if (state->start <= start && state->end > start &&
747 node = &state->rb_node;
752 * this search will find all the extents that end after
755 node = tree_search(tree, start);
757 prealloc = alloc_extent_state_atomic(prealloc);
759 err = insert_state(tree, prealloc, start, end, &bits);
761 extent_io_tree_panic(tree, err);
766 state = rb_entry(node, struct extent_state, rb_node);
768 last_start = state->start;
769 last_end = state->end;
772 * | ---- desired range ---- |
775 * Just lock what we found and keep going
777 if (state->start == start && state->end <= end) {
778 struct rb_node *next_node;
779 if (state->state & exclusive_bits) {
780 *failed_start = state->start;
785 set_state_bits(tree, state, &bits);
787 cache_state(state, cached_state);
788 merge_state(tree, state);
789 if (last_end == (u64)-1)
792 start = last_end + 1;
793 next_node = rb_next(&state->rb_node);
794 if (next_node && start < end && prealloc && !need_resched()) {
795 state = rb_entry(next_node, struct extent_state,
797 if (state->start == start)
804 * | ---- desired range ---- |
807 * | ------------- state -------------- |
809 * We need to split the extent we found, and may flip bits on
812 * If the extent we found extends past our
813 * range, we just split and search again. It'll get split
814 * again the next time though.
816 * If the extent we found is inside our range, we set the
819 if (state->start < start) {
820 if (state->state & exclusive_bits) {
821 *failed_start = start;
826 prealloc = alloc_extent_state_atomic(prealloc);
828 err = split_state(tree, state, prealloc, start);
830 extent_io_tree_panic(tree, err);
835 if (state->end <= end) {
836 set_state_bits(tree, state, &bits);
837 cache_state(state, cached_state);
838 merge_state(tree, state);
839 if (last_end == (u64)-1)
841 start = last_end + 1;
846 * | ---- desired range ---- |
847 * | state | or | state |
849 * There's a hole, we need to insert something in it and
850 * ignore the extent we found.
852 if (state->start > start) {
854 if (end < last_start)
857 this_end = last_start - 1;
859 prealloc = alloc_extent_state_atomic(prealloc);
863 * Avoid to free 'prealloc' if it can be merged with
866 err = insert_state(tree, prealloc, start, this_end,
869 extent_io_tree_panic(tree, err);
871 cache_state(prealloc, cached_state);
873 start = this_end + 1;
877 * | ---- desired range ---- |
879 * We need to split the extent, and set the bit
882 if (state->start <= end && state->end > end) {
883 if (state->state & exclusive_bits) {
884 *failed_start = start;
889 prealloc = alloc_extent_state_atomic(prealloc);
891 err = split_state(tree, state, prealloc, end + 1);
893 extent_io_tree_panic(tree, err);
895 set_state_bits(tree, prealloc, &bits);
896 cache_state(prealloc, cached_state);
897 merge_state(tree, prealloc);
905 spin_unlock(&tree->lock);
907 free_extent_state(prealloc);
914 spin_unlock(&tree->lock);
915 if (mask & __GFP_WAIT)
921 * convert_extent - convert all bits in a given range from one bit to another
922 * @tree: the io tree to search
923 * @start: the start offset in bytes
924 * @end: the end offset in bytes (inclusive)
925 * @bits: the bits to set in this range
926 * @clear_bits: the bits to clear in this range
927 * @mask: the allocation mask
929 * This will go through and set bits for the given range. If any states exist
930 * already in this range they are set with the given bit and cleared of the
931 * clear_bits. This is only meant to be used by things that are mergeable, ie
932 * converting from say DELALLOC to DIRTY. This is not meant to be used with
933 * boundary bits like LOCK.
935 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
936 int bits, int clear_bits, gfp_t mask)
938 struct extent_state *state;
939 struct extent_state *prealloc = NULL;
940 struct rb_node *node;
946 if (!prealloc && (mask & __GFP_WAIT)) {
947 prealloc = alloc_extent_state(mask);
952 spin_lock(&tree->lock);
954 * this search will find all the extents that end after
957 node = tree_search(tree, start);
959 prealloc = alloc_extent_state_atomic(prealloc);
964 err = insert_state(tree, prealloc, start, end, &bits);
967 extent_io_tree_panic(tree, err);
970 state = rb_entry(node, struct extent_state, rb_node);
972 last_start = state->start;
973 last_end = state->end;
976 * | ---- desired range ---- |
979 * Just lock what we found and keep going
981 if (state->start == start && state->end <= end) {
982 struct rb_node *next_node;
984 set_state_bits(tree, state, &bits);
985 clear_state_bit(tree, state, &clear_bits, 0);
986 if (last_end == (u64)-1)
989 start = last_end + 1;
990 next_node = rb_next(&state->rb_node);
991 if (next_node && start < end && prealloc && !need_resched()) {
992 state = rb_entry(next_node, struct extent_state,
994 if (state->start == start)
1001 * | ---- desired range ---- |
1004 * | ------------- state -------------- |
1006 * We need to split the extent we found, and may flip bits on
1009 * If the extent we found extends past our
1010 * range, we just split and search again. It'll get split
1011 * again the next time though.
1013 * If the extent we found is inside our range, we set the
1014 * desired bit on it.
1016 if (state->start < start) {
1017 prealloc = alloc_extent_state_atomic(prealloc);
1022 err = split_state(tree, state, prealloc, start);
1024 extent_io_tree_panic(tree, err);
1028 if (state->end <= end) {
1029 set_state_bits(tree, state, &bits);
1030 clear_state_bit(tree, state, &clear_bits, 0);
1031 if (last_end == (u64)-1)
1033 start = last_end + 1;
1038 * | ---- desired range ---- |
1039 * | state | or | state |
1041 * There's a hole, we need to insert something in it and
1042 * ignore the extent we found.
1044 if (state->start > start) {
1046 if (end < last_start)
1049 this_end = last_start - 1;
1051 prealloc = alloc_extent_state_atomic(prealloc);
1058 * Avoid to free 'prealloc' if it can be merged with
1061 err = insert_state(tree, prealloc, start, this_end,
1064 extent_io_tree_panic(tree, err);
1066 start = this_end + 1;
1070 * | ---- desired range ---- |
1072 * We need to split the extent, and set the bit
1075 if (state->start <= end && state->end > end) {
1076 prealloc = alloc_extent_state_atomic(prealloc);
1082 err = split_state(tree, state, prealloc, end + 1);
1084 extent_io_tree_panic(tree, err);
1086 set_state_bits(tree, prealloc, &bits);
1087 clear_state_bit(tree, prealloc, &clear_bits, 0);
1095 spin_unlock(&tree->lock);
1097 free_extent_state(prealloc);
1104 spin_unlock(&tree->lock);
1105 if (mask & __GFP_WAIT)
1110 /* wrappers around set/clear extent bit */
1111 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1114 return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL,
1118 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1119 int bits, gfp_t mask)
1121 return set_extent_bit(tree, start, end, bits, 0, NULL,
1125 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1126 int bits, gfp_t mask)
1128 return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask);
1131 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
1132 struct extent_state **cached_state, gfp_t mask)
1134 return set_extent_bit(tree, start, end,
1135 EXTENT_DELALLOC | EXTENT_UPTODATE,
1136 0, NULL, cached_state, mask);
1139 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1142 return clear_extent_bit(tree, start, end,
1143 EXTENT_DIRTY | EXTENT_DELALLOC |
1144 EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask);
1147 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
1150 return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL,
1154 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
1155 struct extent_state **cached_state, gfp_t mask)
1157 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0,
1158 NULL, cached_state, mask);
1161 static int clear_extent_uptodate(struct extent_io_tree *tree, u64 start,
1162 u64 end, struct extent_state **cached_state,
1165 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0,
1166 cached_state, mask);
1170 * either insert or lock state struct between start and end use mask to tell
1171 * us if waiting is desired.
1173 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1174 int bits, struct extent_state **cached_state, gfp_t mask)
1179 err = set_extent_bit(tree, start, end, EXTENT_LOCKED | bits,
1180 EXTENT_LOCKED, &failed_start,
1181 cached_state, mask);
1182 if (err == -EEXIST && (mask & __GFP_WAIT)) {
1183 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1184 start = failed_start;
1188 WARN_ON(start > end);
1193 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
1195 return lock_extent_bits(tree, start, end, 0, NULL, mask);
1198 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end,
1204 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1205 &failed_start, NULL, mask);
1206 if (err == -EEXIST) {
1207 if (failed_start > start)
1208 clear_extent_bit(tree, start, failed_start - 1,
1209 EXTENT_LOCKED, 1, 0, NULL, mask);
1215 int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end,
1216 struct extent_state **cached, gfp_t mask)
1218 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached,
1222 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
1224 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL,
1229 * helper function to set both pages and extents in the tree writeback
1231 static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
1233 unsigned long index = start >> PAGE_CACHE_SHIFT;
1234 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1237 while (index <= end_index) {
1238 page = find_get_page(tree->mapping, index);
1240 set_page_writeback(page);
1241 page_cache_release(page);
1247 /* find the first state struct with 'bits' set after 'start', and
1248 * return it. tree->lock must be held. NULL will returned if
1249 * nothing was found after 'start'
1251 struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree,
1252 u64 start, int bits)
1254 struct rb_node *node;
1255 struct extent_state *state;
1258 * this search will find all the extents that end after
1261 node = tree_search(tree, start);
1266 state = rb_entry(node, struct extent_state, rb_node);
1267 if (state->end >= start && (state->state & bits))
1270 node = rb_next(node);
1279 * find the first offset in the io tree with 'bits' set. zero is
1280 * returned if we find something, and *start_ret and *end_ret are
1281 * set to reflect the state struct that was found.
1283 * If nothing was found, 1 is returned, < 0 on error
1285 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1286 u64 *start_ret, u64 *end_ret, int bits)
1288 struct extent_state *state;
1291 spin_lock(&tree->lock);
1292 state = find_first_extent_bit_state(tree, start, bits);
1294 *start_ret = state->start;
1295 *end_ret = state->end;
1298 spin_unlock(&tree->lock);
1303 * find a contiguous range of bytes in the file marked as delalloc, not
1304 * more than 'max_bytes'. start and end are used to return the range,
1306 * 1 is returned if we find something, 0 if nothing was in the tree
1308 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1309 u64 *start, u64 *end, u64 max_bytes,
1310 struct extent_state **cached_state)
1312 struct rb_node *node;
1313 struct extent_state *state;
1314 u64 cur_start = *start;
1316 u64 total_bytes = 0;
1318 spin_lock(&tree->lock);
1321 * this search will find all the extents that end after
1324 node = tree_search(tree, cur_start);
1332 state = rb_entry(node, struct extent_state, rb_node);
1333 if (found && (state->start != cur_start ||
1334 (state->state & EXTENT_BOUNDARY))) {
1337 if (!(state->state & EXTENT_DELALLOC)) {
1343 *start = state->start;
1344 *cached_state = state;
1345 atomic_inc(&state->refs);
1349 cur_start = state->end + 1;
1350 node = rb_next(node);
1353 total_bytes += state->end - state->start + 1;
1354 if (total_bytes >= max_bytes)
1358 spin_unlock(&tree->lock);
1362 static noinline void __unlock_for_delalloc(struct inode *inode,
1363 struct page *locked_page,
1367 struct page *pages[16];
1368 unsigned long index = start >> PAGE_CACHE_SHIFT;
1369 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1370 unsigned long nr_pages = end_index - index + 1;
1373 if (index == locked_page->index && end_index == index)
1376 while (nr_pages > 0) {
1377 ret = find_get_pages_contig(inode->i_mapping, index,
1378 min_t(unsigned long, nr_pages,
1379 ARRAY_SIZE(pages)), pages);
1380 for (i = 0; i < ret; i++) {
1381 if (pages[i] != locked_page)
1382 unlock_page(pages[i]);
1383 page_cache_release(pages[i]);
1391 static noinline int lock_delalloc_pages(struct inode *inode,
1392 struct page *locked_page,
1396 unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT;
1397 unsigned long start_index = index;
1398 unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT;
1399 unsigned long pages_locked = 0;
1400 struct page *pages[16];
1401 unsigned long nrpages;
1405 /* the caller is responsible for locking the start index */
1406 if (index == locked_page->index && index == end_index)
1409 /* skip the page at the start index */
1410 nrpages = end_index - index + 1;
1411 while (nrpages > 0) {
1412 ret = find_get_pages_contig(inode->i_mapping, index,
1413 min_t(unsigned long,
1414 nrpages, ARRAY_SIZE(pages)), pages);
1419 /* now we have an array of pages, lock them all */
1420 for (i = 0; i < ret; i++) {
1422 * the caller is taking responsibility for
1425 if (pages[i] != locked_page) {
1426 lock_page(pages[i]);
1427 if (!PageDirty(pages[i]) ||
1428 pages[i]->mapping != inode->i_mapping) {
1430 unlock_page(pages[i]);
1431 page_cache_release(pages[i]);
1435 page_cache_release(pages[i]);
1444 if (ret && pages_locked) {
1445 __unlock_for_delalloc(inode, locked_page,
1447 ((u64)(start_index + pages_locked - 1)) <<
1454 * find a contiguous range of bytes in the file marked as delalloc, not
1455 * more than 'max_bytes'. start and end are used to return the range,
1457 * 1 is returned if we find something, 0 if nothing was in the tree
1459 static noinline u64 find_lock_delalloc_range(struct inode *inode,
1460 struct extent_io_tree *tree,
1461 struct page *locked_page,
1462 u64 *start, u64 *end,
1468 struct extent_state *cached_state = NULL;
1473 /* step one, find a bunch of delalloc bytes starting at start */
1474 delalloc_start = *start;
1476 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1477 max_bytes, &cached_state);
1478 if (!found || delalloc_end <= *start) {
1479 *start = delalloc_start;
1480 *end = delalloc_end;
1481 free_extent_state(cached_state);
1486 * start comes from the offset of locked_page. We have to lock
1487 * pages in order, so we can't process delalloc bytes before
1490 if (delalloc_start < *start)
1491 delalloc_start = *start;
1494 * make sure to limit the number of pages we try to lock down
1497 if (delalloc_end + 1 - delalloc_start > max_bytes && loops)
1498 delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1;
1500 /* step two, lock all the pages after the page that has start */
1501 ret = lock_delalloc_pages(inode, locked_page,
1502 delalloc_start, delalloc_end);
1503 if (ret == -EAGAIN) {
1504 /* some of the pages are gone, lets avoid looping by
1505 * shortening the size of the delalloc range we're searching
1507 free_extent_state(cached_state);
1509 unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1);
1510 max_bytes = PAGE_CACHE_SIZE - offset;
1520 /* step three, lock the state bits for the whole range */
1521 lock_extent_bits(tree, delalloc_start, delalloc_end,
1522 0, &cached_state, GFP_NOFS);
1524 /* then test to make sure it is all still delalloc */
1525 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1526 EXTENT_DELALLOC, 1, cached_state);
1528 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1529 &cached_state, GFP_NOFS);
1530 __unlock_for_delalloc(inode, locked_page,
1531 delalloc_start, delalloc_end);
1535 free_extent_state(cached_state);
1536 *start = delalloc_start;
1537 *end = delalloc_end;
1542 int extent_clear_unlock_delalloc(struct inode *inode,
1543 struct extent_io_tree *tree,
1544 u64 start, u64 end, struct page *locked_page,
1548 struct page *pages[16];
1549 unsigned long index = start >> PAGE_CACHE_SHIFT;
1550 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1551 unsigned long nr_pages = end_index - index + 1;
1555 if (op & EXTENT_CLEAR_UNLOCK)
1556 clear_bits |= EXTENT_LOCKED;
1557 if (op & EXTENT_CLEAR_DIRTY)
1558 clear_bits |= EXTENT_DIRTY;
1560 if (op & EXTENT_CLEAR_DELALLOC)
1561 clear_bits |= EXTENT_DELALLOC;
1563 clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS);
1564 if (!(op & (EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
1565 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK |
1566 EXTENT_SET_PRIVATE2)))
1569 while (nr_pages > 0) {
1570 ret = find_get_pages_contig(inode->i_mapping, index,
1571 min_t(unsigned long,
1572 nr_pages, ARRAY_SIZE(pages)), pages);
1573 for (i = 0; i < ret; i++) {
1575 if (op & EXTENT_SET_PRIVATE2)
1576 SetPagePrivate2(pages[i]);
1578 if (pages[i] == locked_page) {
1579 page_cache_release(pages[i]);
1582 if (op & EXTENT_CLEAR_DIRTY)
1583 clear_page_dirty_for_io(pages[i]);
1584 if (op & EXTENT_SET_WRITEBACK)
1585 set_page_writeback(pages[i]);
1586 if (op & EXTENT_END_WRITEBACK)
1587 end_page_writeback(pages[i]);
1588 if (op & EXTENT_CLEAR_UNLOCK_PAGE)
1589 unlock_page(pages[i]);
1590 page_cache_release(pages[i]);
1600 * count the number of bytes in the tree that have a given bit(s)
1601 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1602 * cached. The total number found is returned.
1604 u64 count_range_bits(struct extent_io_tree *tree,
1605 u64 *start, u64 search_end, u64 max_bytes,
1606 unsigned long bits, int contig)
1608 struct rb_node *node;
1609 struct extent_state *state;
1610 u64 cur_start = *start;
1611 u64 total_bytes = 0;
1615 if (search_end <= cur_start) {
1620 spin_lock(&tree->lock);
1621 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1622 total_bytes = tree->dirty_bytes;
1626 * this search will find all the extents that end after
1629 node = tree_search(tree, cur_start);
1634 state = rb_entry(node, struct extent_state, rb_node);
1635 if (state->start > search_end)
1637 if (contig && found && state->start > last + 1)
1639 if (state->end >= cur_start && (state->state & bits) == bits) {
1640 total_bytes += min(search_end, state->end) + 1 -
1641 max(cur_start, state->start);
1642 if (total_bytes >= max_bytes)
1645 *start = max(cur_start, state->start);
1649 } else if (contig && found) {
1652 node = rb_next(node);
1657 spin_unlock(&tree->lock);
1662 * set the private field for a given byte offset in the tree. If there isn't
1663 * an extent_state there already, this does nothing.
1665 int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1667 struct rb_node *node;
1668 struct extent_state *state;
1671 spin_lock(&tree->lock);
1673 * this search will find all the extents that end after
1676 node = tree_search(tree, start);
1681 state = rb_entry(node, struct extent_state, rb_node);
1682 if (state->start != start) {
1686 state->private = private;
1688 spin_unlock(&tree->lock);
1692 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1694 struct rb_node *node;
1695 struct extent_state *state;
1698 spin_lock(&tree->lock);
1700 * this search will find all the extents that end after
1703 node = tree_search(tree, start);
1708 state = rb_entry(node, struct extent_state, rb_node);
1709 if (state->start != start) {
1713 *private = state->private;
1715 spin_unlock(&tree->lock);
1720 * searches a range in the state tree for a given mask.
1721 * If 'filled' == 1, this returns 1 only if every extent in the tree
1722 * has the bits set. Otherwise, 1 is returned if any bit in the
1723 * range is found set.
1725 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1726 int bits, int filled, struct extent_state *cached)
1728 struct extent_state *state = NULL;
1729 struct rb_node *node;
1732 spin_lock(&tree->lock);
1733 if (cached && cached->tree && cached->start <= start &&
1734 cached->end > start)
1735 node = &cached->rb_node;
1737 node = tree_search(tree, start);
1738 while (node && start <= end) {
1739 state = rb_entry(node, struct extent_state, rb_node);
1741 if (filled && state->start > start) {
1746 if (state->start > end)
1749 if (state->state & bits) {
1753 } else if (filled) {
1758 if (state->end == (u64)-1)
1761 start = state->end + 1;
1764 node = rb_next(node);
1771 spin_unlock(&tree->lock);
1776 * helper function to set a given page up to date if all the
1777 * extents in the tree for that page are up to date
1779 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
1781 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1782 u64 end = start + PAGE_CACHE_SIZE - 1;
1783 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1784 SetPageUptodate(page);
1788 * helper function to unlock a page if all the extents in the tree
1789 * for that page are unlocked
1791 static void check_page_locked(struct extent_io_tree *tree, struct page *page)
1793 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1794 u64 end = start + PAGE_CACHE_SIZE - 1;
1795 if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL))
1800 * helper function to end page writeback if all the extents
1801 * in the tree for that page are done with writeback
1803 static void check_page_writeback(struct extent_io_tree *tree,
1806 end_page_writeback(page);
1810 * When IO fails, either with EIO or csum verification fails, we
1811 * try other mirrors that might have a good copy of the data. This
1812 * io_failure_record is used to record state as we go through all the
1813 * mirrors. If another mirror has good data, the page is set up to date
1814 * and things continue. If a good mirror can't be found, the original
1815 * bio end_io callback is called to indicate things have failed.
1817 struct io_failure_record {
1822 unsigned long bio_flags;
1828 static int free_io_failure(struct inode *inode, struct io_failure_record *rec,
1833 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1835 set_state_private(failure_tree, rec->start, 0);
1836 ret = clear_extent_bits(failure_tree, rec->start,
1837 rec->start + rec->len - 1,
1838 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1843 ret = clear_extent_bits(&BTRFS_I(inode)->io_tree, rec->start,
1844 rec->start + rec->len - 1,
1845 EXTENT_DAMAGED, GFP_NOFS);
1854 static void repair_io_failure_callback(struct bio *bio, int err)
1856 complete(bio->bi_private);
1860 * this bypasses the standard btrfs submit functions deliberately, as
1861 * the standard behavior is to write all copies in a raid setup. here we only
1862 * want to write the one bad copy. so we do the mapping for ourselves and issue
1863 * submit_bio directly.
1864 * to avoid any synchonization issues, wait for the data after writing, which
1865 * actually prevents the read that triggered the error from finishing.
1866 * currently, there can be no more than two copies of every data bit. thus,
1867 * exactly one rewrite is required.
1869 int repair_io_failure(struct btrfs_mapping_tree *map_tree, u64 start,
1870 u64 length, u64 logical, struct page *page,
1874 struct btrfs_device *dev;
1875 DECLARE_COMPLETION_ONSTACK(compl);
1878 struct btrfs_bio *bbio = NULL;
1881 BUG_ON(!mirror_num);
1883 bio = bio_alloc(GFP_NOFS, 1);
1886 bio->bi_private = &compl;
1887 bio->bi_end_io = repair_io_failure_callback;
1889 map_length = length;
1891 ret = btrfs_map_block(map_tree, WRITE, logical,
1892 &map_length, &bbio, mirror_num);
1897 BUG_ON(mirror_num != bbio->mirror_num);
1898 sector = bbio->stripes[mirror_num-1].physical >> 9;
1899 bio->bi_sector = sector;
1900 dev = bbio->stripes[mirror_num-1].dev;
1902 if (!dev || !dev->bdev || !dev->writeable) {
1906 bio->bi_bdev = dev->bdev;
1907 bio_add_page(bio, page, length, start-page_offset(page));
1908 btrfsic_submit_bio(WRITE_SYNC, bio);
1909 wait_for_completion(&compl);
1911 if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) {
1912 /* try to remap that extent elsewhere? */
1917 printk(KERN_INFO "btrfs read error corrected: ino %lu off %llu (dev %s "
1918 "sector %llu)\n", page->mapping->host->i_ino, start,
1926 * each time an IO finishes, we do a fast check in the IO failure tree
1927 * to see if we need to process or clean up an io_failure_record
1929 static int clean_io_failure(u64 start, struct page *page)
1932 u64 private_failure;
1933 struct io_failure_record *failrec;
1934 struct btrfs_mapping_tree *map_tree;
1935 struct extent_state *state;
1939 struct inode *inode = page->mapping->host;
1942 ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1943 (u64)-1, 1, EXTENT_DIRTY, 0);
1947 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, start,
1952 failrec = (struct io_failure_record *)(unsigned long) private_failure;
1953 BUG_ON(!failrec->this_mirror);
1955 if (failrec->in_validation) {
1956 /* there was no real error, just free the record */
1957 pr_debug("clean_io_failure: freeing dummy error at %llu\n",
1963 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1964 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1967 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1969 if (state && state->start == failrec->start) {
1970 map_tree = &BTRFS_I(inode)->root->fs_info->mapping_tree;
1971 num_copies = btrfs_num_copies(map_tree, failrec->logical,
1973 if (num_copies > 1) {
1974 ret = repair_io_failure(map_tree, start, failrec->len,
1975 failrec->logical, page,
1976 failrec->failed_mirror);
1983 ret = free_io_failure(inode, failrec, did_repair);
1989 * this is a generic handler for readpage errors (default
1990 * readpage_io_failed_hook). if other copies exist, read those and write back
1991 * good data to the failed position. does not investigate in remapping the
1992 * failed extent elsewhere, hoping the device will be smart enough to do this as
1996 static int bio_readpage_error(struct bio *failed_bio, struct page *page,
1997 u64 start, u64 end, int failed_mirror,
1998 struct extent_state *state)
2000 struct io_failure_record *failrec = NULL;
2002 struct extent_map *em;
2003 struct inode *inode = page->mapping->host;
2004 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2005 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2006 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2013 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
2015 ret = get_state_private(failure_tree, start, &private);
2017 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2020 failrec->start = start;
2021 failrec->len = end - start + 1;
2022 failrec->this_mirror = 0;
2023 failrec->bio_flags = 0;
2024 failrec->in_validation = 0;
2026 read_lock(&em_tree->lock);
2027 em = lookup_extent_mapping(em_tree, start, failrec->len);
2029 read_unlock(&em_tree->lock);
2034 if (em->start > start || em->start + em->len < start) {
2035 free_extent_map(em);
2038 read_unlock(&em_tree->lock);
2040 if (!em || IS_ERR(em)) {
2044 logical = start - em->start;
2045 logical = em->block_start + logical;
2046 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2047 logical = em->block_start;
2048 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2049 extent_set_compress_type(&failrec->bio_flags,
2052 pr_debug("bio_readpage_error: (new) logical=%llu, start=%llu, "
2053 "len=%llu\n", logical, start, failrec->len);
2054 failrec->logical = logical;
2055 free_extent_map(em);
2057 /* set the bits in the private failure tree */
2058 ret = set_extent_bits(failure_tree, start, end,
2059 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
2061 ret = set_state_private(failure_tree, start,
2062 (u64)(unsigned long)failrec);
2063 /* set the bits in the inode's tree */
2065 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED,
2072 failrec = (struct io_failure_record *)(unsigned long)private;
2073 pr_debug("bio_readpage_error: (found) logical=%llu, "
2074 "start=%llu, len=%llu, validation=%d\n",
2075 failrec->logical, failrec->start, failrec->len,
2076 failrec->in_validation);
2078 * when data can be on disk more than twice, add to failrec here
2079 * (e.g. with a list for failed_mirror) to make
2080 * clean_io_failure() clean all those errors at once.
2083 num_copies = btrfs_num_copies(
2084 &BTRFS_I(inode)->root->fs_info->mapping_tree,
2085 failrec->logical, failrec->len);
2086 if (num_copies == 1) {
2088 * we only have a single copy of the data, so don't bother with
2089 * all the retry and error correction code that follows. no
2090 * matter what the error is, it is very likely to persist.
2092 pr_debug("bio_readpage_error: cannot repair, num_copies == 1. "
2093 "state=%p, num_copies=%d, next_mirror %d, "
2094 "failed_mirror %d\n", state, num_copies,
2095 failrec->this_mirror, failed_mirror);
2096 free_io_failure(inode, failrec, 0);
2101 spin_lock(&tree->lock);
2102 state = find_first_extent_bit_state(tree, failrec->start,
2104 if (state && state->start != failrec->start)
2106 spin_unlock(&tree->lock);
2110 * there are two premises:
2111 * a) deliver good data to the caller
2112 * b) correct the bad sectors on disk
2114 if (failed_bio->bi_vcnt > 1) {
2116 * to fulfill b), we need to know the exact failing sectors, as
2117 * we don't want to rewrite any more than the failed ones. thus,
2118 * we need separate read requests for the failed bio
2120 * if the following BUG_ON triggers, our validation request got
2121 * merged. we need separate requests for our algorithm to work.
2123 BUG_ON(failrec->in_validation);
2124 failrec->in_validation = 1;
2125 failrec->this_mirror = failed_mirror;
2126 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
2129 * we're ready to fulfill a) and b) alongside. get a good copy
2130 * of the failed sector and if we succeed, we have setup
2131 * everything for repair_io_failure to do the rest for us.
2133 if (failrec->in_validation) {
2134 BUG_ON(failrec->this_mirror != failed_mirror);
2135 failrec->in_validation = 0;
2136 failrec->this_mirror = 0;
2138 failrec->failed_mirror = failed_mirror;
2139 failrec->this_mirror++;
2140 if (failrec->this_mirror == failed_mirror)
2141 failrec->this_mirror++;
2142 read_mode = READ_SYNC;
2145 if (!state || failrec->this_mirror > num_copies) {
2146 pr_debug("bio_readpage_error: (fail) state=%p, num_copies=%d, "
2147 "next_mirror %d, failed_mirror %d\n", state,
2148 num_copies, failrec->this_mirror, failed_mirror);
2149 free_io_failure(inode, failrec, 0);
2153 bio = bio_alloc(GFP_NOFS, 1);
2154 bio->bi_private = state;
2155 bio->bi_end_io = failed_bio->bi_end_io;
2156 bio->bi_sector = failrec->logical >> 9;
2157 bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
2160 bio_add_page(bio, page, failrec->len, start - page_offset(page));
2162 pr_debug("bio_readpage_error: submitting new read[%#x] to "
2163 "this_mirror=%d, num_copies=%d, in_validation=%d\n", read_mode,
2164 failrec->this_mirror, num_copies, failrec->in_validation);
2166 ret = tree->ops->submit_bio_hook(inode, read_mode, bio,
2167 failrec->this_mirror,
2168 failrec->bio_flags, 0);
2172 /* lots and lots of room for performance fixes in the end_bio funcs */
2174 int end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2176 int uptodate = (err == 0);
2177 struct extent_io_tree *tree;
2180 tree = &BTRFS_I(page->mapping->host)->io_tree;
2182 if (tree->ops && tree->ops->writepage_end_io_hook) {
2183 ret = tree->ops->writepage_end_io_hook(page, start,
2184 end, NULL, uptodate);
2189 if (!uptodate && tree->ops &&
2190 tree->ops->writepage_io_failed_hook) {
2191 ret = tree->ops->writepage_io_failed_hook(NULL, page,
2193 /* Writeback already completed */
2200 clear_extent_uptodate(tree, start, end, NULL, GFP_NOFS);
2201 ClearPageUptodate(page);
2208 * after a writepage IO is done, we need to:
2209 * clear the uptodate bits on error
2210 * clear the writeback bits in the extent tree for this IO
2211 * end_page_writeback if the page has no more pending IO
2213 * Scheduling is not allowed, so the extent state tree is expected
2214 * to have one and only one object corresponding to this IO.
2216 static void end_bio_extent_writepage(struct bio *bio, int err)
2218 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2219 struct extent_io_tree *tree;
2225 struct page *page = bvec->bv_page;
2226 tree = &BTRFS_I(page->mapping->host)->io_tree;
2228 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
2230 end = start + bvec->bv_len - 1;
2232 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
2237 if (--bvec >= bio->bi_io_vec)
2238 prefetchw(&bvec->bv_page->flags);
2240 if (end_extent_writepage(page, err, start, end))
2244 end_page_writeback(page);
2246 check_page_writeback(tree, page);
2247 } while (bvec >= bio->bi_io_vec);
2253 * after a readpage IO is done, we need to:
2254 * clear the uptodate bits on error
2255 * set the uptodate bits if things worked
2256 * set the page up to date if all extents in the tree are uptodate
2257 * clear the lock bit in the extent tree
2258 * unlock the page if there are no other extents locked for it
2260 * Scheduling is not allowed, so the extent state tree is expected
2261 * to have one and only one object corresponding to this IO.
2263 static void end_bio_extent_readpage(struct bio *bio, int err)
2265 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
2266 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
2267 struct bio_vec *bvec = bio->bi_io_vec;
2268 struct extent_io_tree *tree;
2278 struct page *page = bvec->bv_page;
2279 struct extent_state *cached = NULL;
2280 struct extent_state *state;
2282 pr_debug("end_bio_extent_readpage: bi_vcnt=%d, idx=%d, err=%d, "
2283 "mirror=%ld\n", bio->bi_vcnt, bio->bi_idx, err,
2284 (long int)bio->bi_bdev);
2285 tree = &BTRFS_I(page->mapping->host)->io_tree;
2287 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
2289 end = start + bvec->bv_len - 1;
2291 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
2296 if (++bvec <= bvec_end)
2297 prefetchw(&bvec->bv_page->flags);
2299 spin_lock(&tree->lock);
2300 state = find_first_extent_bit_state(tree, start, EXTENT_LOCKED);
2301 if (state && state->start == start) {
2303 * take a reference on the state, unlock will drop
2306 cache_state(state, &cached);
2308 spin_unlock(&tree->lock);
2310 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
2311 ret = tree->ops->readpage_end_io_hook(page, start, end,
2316 clean_io_failure(start, page);
2320 failed_mirror = (int)(unsigned long)bio->bi_bdev;
2322 * The generic bio_readpage_error handles errors the
2323 * following way: If possible, new read requests are
2324 * created and submitted and will end up in
2325 * end_bio_extent_readpage as well (if we're lucky, not
2326 * in the !uptodate case). In that case it returns 0 and
2327 * we just go on with the next page in our bio. If it
2328 * can't handle the error it will return -EIO and we
2329 * remain responsible for that page.
2331 ret = bio_readpage_error(bio, page, start, end,
2332 failed_mirror, NULL);
2336 test_bit(BIO_UPTODATE, &bio->bi_flags);
2339 uncache_state(&cached);
2342 if (tree->ops && tree->ops->readpage_io_failed_hook) {
2343 ret = tree->ops->readpage_io_failed_hook(
2344 bio, page, start, end,
2345 failed_mirror, state);
2353 set_extent_uptodate(tree, start, end, &cached,
2356 unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC);
2360 SetPageUptodate(page);
2362 ClearPageUptodate(page);
2368 check_page_uptodate(tree, page);
2370 ClearPageUptodate(page);
2373 check_page_locked(tree, page);
2375 } while (bvec <= bvec_end);
2381 btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
2386 bio = bio_alloc(gfp_flags, nr_vecs);
2388 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
2389 while (!bio && (nr_vecs /= 2))
2390 bio = bio_alloc(gfp_flags, nr_vecs);
2395 bio->bi_bdev = bdev;
2396 bio->bi_sector = first_sector;
2401 static int __must_check submit_one_bio(int rw, struct bio *bio,
2402 int mirror_num, unsigned long bio_flags)
2405 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2406 struct page *page = bvec->bv_page;
2407 struct extent_io_tree *tree = bio->bi_private;
2410 start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
2412 bio->bi_private = NULL;
2416 if (tree->ops && tree->ops->submit_bio_hook)
2417 ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
2418 mirror_num, bio_flags, start);
2420 btrfsic_submit_bio(rw, bio);
2422 if (bio_flagged(bio, BIO_EOPNOTSUPP))
2428 static int merge_bio(struct extent_io_tree *tree, struct page *page,
2429 unsigned long offset, size_t size, struct bio *bio,
2430 unsigned long bio_flags)
2433 if (tree->ops && tree->ops->merge_bio_hook)
2434 ret = tree->ops->merge_bio_hook(page, offset, size, bio,
2441 static int submit_extent_page(int rw, struct extent_io_tree *tree,
2442 struct page *page, sector_t sector,
2443 size_t size, unsigned long offset,
2444 struct block_device *bdev,
2445 struct bio **bio_ret,
2446 unsigned long max_pages,
2447 bio_end_io_t end_io_func,
2449 unsigned long prev_bio_flags,
2450 unsigned long bio_flags)
2456 int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED;
2457 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
2458 size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE);
2460 if (bio_ret && *bio_ret) {
2463 contig = bio->bi_sector == sector;
2465 contig = bio->bi_sector + (bio->bi_size >> 9) ==
2468 if (prev_bio_flags != bio_flags || !contig ||
2469 merge_bio(tree, page, offset, page_size, bio, bio_flags) ||
2470 bio_add_page(bio, page, page_size, offset) < page_size) {
2471 ret = submit_one_bio(rw, bio, mirror_num,
2479 if (this_compressed)
2482 nr = bio_get_nr_vecs(bdev);
2484 bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
2488 bio_add_page(bio, page, page_size, offset);
2489 bio->bi_end_io = end_io_func;
2490 bio->bi_private = tree;
2495 ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
2502 void set_page_extent_mapped(struct page *page)
2504 if (!PagePrivate(page)) {
2505 SetPagePrivate(page);
2506 page_cache_get(page);
2507 set_page_private(page, EXTENT_PAGE_PRIVATE);
2511 static void set_page_extent_head(struct page *page, unsigned long len)
2513 WARN_ON(!PagePrivate(page));
2514 set_page_private(page, EXTENT_PAGE_PRIVATE_FIRST_PAGE | len << 2);
2518 * basic readpage implementation. Locked extent state structs are inserted
2519 * into the tree that are removed when the IO is done (by the end_io
2522 static int __extent_read_full_page(struct extent_io_tree *tree,
2524 get_extent_t *get_extent,
2525 struct bio **bio, int mirror_num,
2526 unsigned long *bio_flags)
2528 struct inode *inode = page->mapping->host;
2529 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2530 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2534 u64 last_byte = i_size_read(inode);
2538 struct extent_map *em;
2539 struct block_device *bdev;
2540 struct btrfs_ordered_extent *ordered;
2543 size_t pg_offset = 0;
2545 size_t disk_io_size;
2546 size_t blocksize = inode->i_sb->s_blocksize;
2547 unsigned long this_bio_flag = 0;
2549 set_page_extent_mapped(page);
2551 if (!PageUptodate(page)) {
2552 if (cleancache_get_page(page) == 0) {
2553 BUG_ON(blocksize != PAGE_SIZE);
2560 lock_extent(tree, start, end, GFP_NOFS);
2561 ordered = btrfs_lookup_ordered_extent(inode, start);
2564 unlock_extent(tree, start, end, GFP_NOFS);
2565 btrfs_start_ordered_extent(inode, ordered, 1);
2566 btrfs_put_ordered_extent(ordered);
2569 if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
2571 size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
2574 iosize = PAGE_CACHE_SIZE - zero_offset;
2575 userpage = kmap_atomic(page, KM_USER0);
2576 memset(userpage + zero_offset, 0, iosize);
2577 flush_dcache_page(page);
2578 kunmap_atomic(userpage, KM_USER0);
2581 while (cur <= end) {
2582 if (cur >= last_byte) {
2584 struct extent_state *cached = NULL;
2586 iosize = PAGE_CACHE_SIZE - pg_offset;
2587 userpage = kmap_atomic(page, KM_USER0);
2588 memset(userpage + pg_offset, 0, iosize);
2589 flush_dcache_page(page);
2590 kunmap_atomic(userpage, KM_USER0);
2591 set_extent_uptodate(tree, cur, cur + iosize - 1,
2593 unlock_extent_cached(tree, cur, cur + iosize - 1,
2597 em = get_extent(inode, page, pg_offset, cur,
2599 if (IS_ERR_OR_NULL(em)) {
2601 unlock_extent(tree, cur, end, GFP_NOFS);
2604 extent_offset = cur - em->start;
2605 BUG_ON(extent_map_end(em) <= cur);
2608 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2609 this_bio_flag = EXTENT_BIO_COMPRESSED;
2610 extent_set_compress_type(&this_bio_flag,
2614 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2615 cur_end = min(extent_map_end(em) - 1, end);
2616 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2617 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2618 disk_io_size = em->block_len;
2619 sector = em->block_start >> 9;
2621 sector = (em->block_start + extent_offset) >> 9;
2622 disk_io_size = iosize;
2625 block_start = em->block_start;
2626 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2627 block_start = EXTENT_MAP_HOLE;
2628 free_extent_map(em);
2631 /* we've found a hole, just zero and go on */
2632 if (block_start == EXTENT_MAP_HOLE) {
2634 struct extent_state *cached = NULL;
2636 userpage = kmap_atomic(page, KM_USER0);
2637 memset(userpage + pg_offset, 0, iosize);
2638 flush_dcache_page(page);
2639 kunmap_atomic(userpage, KM_USER0);
2641 set_extent_uptodate(tree, cur, cur + iosize - 1,
2643 unlock_extent_cached(tree, cur, cur + iosize - 1,
2646 pg_offset += iosize;
2649 /* the get_extent function already copied into the page */
2650 if (test_range_bit(tree, cur, cur_end,
2651 EXTENT_UPTODATE, 1, NULL)) {
2652 check_page_uptodate(tree, page);
2653 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
2655 pg_offset += iosize;
2658 /* we have an inline extent but it didn't get marked up
2659 * to date. Error out
2661 if (block_start == EXTENT_MAP_INLINE) {
2663 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
2665 pg_offset += iosize;
2670 if (tree->ops && tree->ops->readpage_io_hook) {
2671 ret = tree->ops->readpage_io_hook(page, cur,
2675 unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
2677 ret = submit_extent_page(READ, tree, page,
2678 sector, disk_io_size, pg_offset,
2680 end_bio_extent_readpage, mirror_num,
2684 *bio_flags = this_bio_flag;
2689 pg_offset += iosize;
2693 if (!PageError(page))
2694 SetPageUptodate(page);
2700 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
2701 get_extent_t *get_extent, int mirror_num)
2703 struct bio *bio = NULL;
2704 unsigned long bio_flags = 0;
2707 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
2710 ret = submit_one_bio(READ, bio, mirror_num, bio_flags);
2716 static noinline void update_nr_written(struct page *page,
2717 struct writeback_control *wbc,
2718 unsigned long nr_written)
2720 wbc->nr_to_write -= nr_written;
2721 if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
2722 wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
2723 page->mapping->writeback_index = page->index + nr_written;
2727 * the writepage semantics are similar to regular writepage. extent
2728 * records are inserted to lock ranges in the tree, and as dirty areas
2729 * are found, they are marked writeback. Then the lock bits are removed
2730 * and the end_io handler clears the writeback ranges
2732 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2735 struct inode *inode = page->mapping->host;
2736 struct extent_page_data *epd = data;
2737 struct extent_io_tree *tree = epd->tree;
2738 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2740 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2744 u64 last_byte = i_size_read(inode);
2748 struct extent_state *cached_state = NULL;
2749 struct extent_map *em;
2750 struct block_device *bdev;
2753 size_t pg_offset = 0;
2755 loff_t i_size = i_size_read(inode);
2756 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
2762 unsigned long nr_written = 0;
2763 bool fill_delalloc = true;
2765 if (wbc->sync_mode == WB_SYNC_ALL)
2766 write_flags = WRITE_SYNC;
2768 write_flags = WRITE;
2770 trace___extent_writepage(page, inode, wbc);
2772 WARN_ON(!PageLocked(page));
2774 ClearPageError(page);
2776 pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
2777 if (page->index > end_index ||
2778 (page->index == end_index && !pg_offset)) {
2779 page->mapping->a_ops->invalidatepage(page, 0);
2784 if (page->index == end_index) {
2787 userpage = kmap_atomic(page, KM_USER0);
2788 memset(userpage + pg_offset, 0,
2789 PAGE_CACHE_SIZE - pg_offset);
2790 kunmap_atomic(userpage, KM_USER0);
2791 flush_dcache_page(page);
2795 set_page_extent_mapped(page);
2797 if (!tree->ops || !tree->ops->fill_delalloc)
2798 fill_delalloc = false;
2800 delalloc_start = start;
2803 if (!epd->extent_locked && fill_delalloc) {
2804 u64 delalloc_to_write = 0;
2806 * make sure the wbc mapping index is at least updated
2809 update_nr_written(page, wbc, 0);
2811 while (delalloc_end < page_end) {
2812 nr_delalloc = find_lock_delalloc_range(inode, tree,
2817 if (nr_delalloc == 0) {
2818 delalloc_start = delalloc_end + 1;
2821 ret = tree->ops->fill_delalloc(inode, page,
2828 * delalloc_end is already one less than the total
2829 * length, so we don't subtract one from
2832 delalloc_to_write += (delalloc_end - delalloc_start +
2835 delalloc_start = delalloc_end + 1;
2837 if (wbc->nr_to_write < delalloc_to_write) {
2840 if (delalloc_to_write < thresh * 2)
2841 thresh = delalloc_to_write;
2842 wbc->nr_to_write = min_t(u64, delalloc_to_write,
2846 /* did the fill delalloc function already unlock and start
2852 * we've unlocked the page, so we can't update
2853 * the mapping's writeback index, just update
2856 wbc->nr_to_write -= nr_written;
2860 if (tree->ops && tree->ops->writepage_start_hook) {
2861 ret = tree->ops->writepage_start_hook(page, start,
2864 /* Fixup worker will requeue */
2866 wbc->pages_skipped++;
2868 redirty_page_for_writepage(wbc, page);
2869 update_nr_written(page, wbc, nr_written);
2877 * we don't want to touch the inode after unlocking the page,
2878 * so we update the mapping writeback index now
2880 update_nr_written(page, wbc, nr_written + 1);
2883 if (last_byte <= start) {
2884 if (tree->ops && tree->ops->writepage_end_io_hook)
2885 tree->ops->writepage_end_io_hook(page, start,
2890 blocksize = inode->i_sb->s_blocksize;
2892 while (cur <= end) {
2893 if (cur >= last_byte) {
2894 if (tree->ops && tree->ops->writepage_end_io_hook)
2895 tree->ops->writepage_end_io_hook(page, cur,
2899 em = epd->get_extent(inode, page, pg_offset, cur,
2901 if (IS_ERR_OR_NULL(em)) {
2906 extent_offset = cur - em->start;
2907 BUG_ON(extent_map_end(em) <= cur);
2909 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2910 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2911 sector = (em->block_start + extent_offset) >> 9;
2913 block_start = em->block_start;
2914 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
2915 free_extent_map(em);
2919 * compressed and inline extents are written through other
2922 if (compressed || block_start == EXTENT_MAP_HOLE ||
2923 block_start == EXTENT_MAP_INLINE) {
2925 * end_io notification does not happen here for
2926 * compressed extents
2928 if (!compressed && tree->ops &&
2929 tree->ops->writepage_end_io_hook)
2930 tree->ops->writepage_end_io_hook(page, cur,
2933 else if (compressed) {
2934 /* we don't want to end_page_writeback on
2935 * a compressed extent. this happens
2942 pg_offset += iosize;
2945 /* leave this out until we have a page_mkwrite call */
2946 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
2947 EXTENT_DIRTY, 0, NULL)) {
2949 pg_offset += iosize;
2953 if (tree->ops && tree->ops->writepage_io_hook) {
2954 ret = tree->ops->writepage_io_hook(page, cur,
2962 unsigned long max_nr = end_index + 1;
2964 set_range_writeback(tree, cur, cur + iosize - 1);
2965 if (!PageWriteback(page)) {
2966 printk(KERN_ERR "btrfs warning page %lu not "
2967 "writeback, cur %llu end %llu\n",
2968 page->index, (unsigned long long)cur,
2969 (unsigned long long)end);
2972 ret = submit_extent_page(write_flags, tree, page,
2973 sector, iosize, pg_offset,
2974 bdev, &epd->bio, max_nr,
2975 end_bio_extent_writepage,
2981 pg_offset += iosize;
2986 /* make sure the mapping tag for page dirty gets cleared */
2987 set_page_writeback(page);
2988 end_page_writeback(page);
2994 /* drop our reference on any cached states */
2995 free_extent_state(cached_state);
3000 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3001 * @mapping: address space structure to write
3002 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3003 * @writepage: function called for each page
3004 * @data: data passed to writepage function
3006 * If a page is already under I/O, write_cache_pages() skips it, even
3007 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3008 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3009 * and msync() need to guarantee that all the data which was dirty at the time
3010 * the call was made get new I/O started against them. If wbc->sync_mode is
3011 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3012 * existing IO to complete.
3014 static int extent_write_cache_pages(struct extent_io_tree *tree,
3015 struct address_space *mapping,
3016 struct writeback_control *wbc,
3017 writepage_t writepage, void *data,
3018 void (*flush_fn)(void *))
3022 int nr_to_write_done = 0;
3023 struct pagevec pvec;
3026 pgoff_t end; /* Inclusive */
3030 pagevec_init(&pvec, 0);
3031 if (wbc->range_cyclic) {
3032 index = mapping->writeback_index; /* Start from prev offset */
3035 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3036 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3039 if (wbc->sync_mode == WB_SYNC_ALL)
3040 tag = PAGECACHE_TAG_TOWRITE;
3042 tag = PAGECACHE_TAG_DIRTY;
3044 if (wbc->sync_mode == WB_SYNC_ALL)
3045 tag_pages_for_writeback(mapping, index, end);
3046 while (!done && !nr_to_write_done && (index <= end) &&
3047 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3048 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3052 for (i = 0; i < nr_pages; i++) {
3053 struct page *page = pvec.pages[i];
3056 * At this point we hold neither mapping->tree_lock nor
3057 * lock on the page itself: the page may be truncated or
3058 * invalidated (changing page->mapping to NULL), or even
3059 * swizzled back from swapper_space to tmpfs file
3063 tree->ops->write_cache_pages_lock_hook) {
3064 tree->ops->write_cache_pages_lock_hook(page,
3067 if (!trylock_page(page)) {
3073 if (unlikely(page->mapping != mapping)) {
3078 if (!wbc->range_cyclic && page->index > end) {
3084 if (wbc->sync_mode != WB_SYNC_NONE) {
3085 if (PageWriteback(page))
3087 wait_on_page_writeback(page);
3090 if (PageWriteback(page) ||
3091 !clear_page_dirty_for_io(page)) {
3096 ret = (*writepage)(page, wbc, data);
3098 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
3106 * the filesystem may choose to bump up nr_to_write.
3107 * We have to make sure to honor the new nr_to_write
3110 nr_to_write_done = wbc->nr_to_write <= 0;
3112 pagevec_release(&pvec);
3115 if (!scanned && !done) {
3117 * We hit the last page and there is more work to be done: wrap
3118 * back to the start of the file
3127 static void flush_epd_write_bio(struct extent_page_data *epd)
3136 ret = submit_one_bio(rw, epd->bio, 0, 0);
3142 static noinline void flush_write_bio(void *data)
3144 struct extent_page_data *epd = data;
3145 flush_epd_write_bio(epd);
3148 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
3149 get_extent_t *get_extent,
3150 struct writeback_control *wbc)
3153 struct extent_page_data epd = {
3156 .get_extent = get_extent,
3158 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3161 ret = __extent_writepage(page, wbc, &epd);
3163 flush_epd_write_bio(&epd);
3167 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
3168 u64 start, u64 end, get_extent_t *get_extent,
3172 struct address_space *mapping = inode->i_mapping;
3174 unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
3177 struct extent_page_data epd = {
3180 .get_extent = get_extent,
3182 .sync_io = mode == WB_SYNC_ALL,
3184 struct writeback_control wbc_writepages = {
3186 .nr_to_write = nr_pages * 2,
3187 .range_start = start,
3188 .range_end = end + 1,
3191 while (start <= end) {
3192 page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
3193 if (clear_page_dirty_for_io(page))
3194 ret = __extent_writepage(page, &wbc_writepages, &epd);
3196 if (tree->ops && tree->ops->writepage_end_io_hook)
3197 tree->ops->writepage_end_io_hook(page, start,
3198 start + PAGE_CACHE_SIZE - 1,
3202 page_cache_release(page);
3203 start += PAGE_CACHE_SIZE;
3206 flush_epd_write_bio(&epd);
3210 int extent_writepages(struct extent_io_tree *tree,
3211 struct address_space *mapping,
3212 get_extent_t *get_extent,
3213 struct writeback_control *wbc)
3216 struct extent_page_data epd = {
3219 .get_extent = get_extent,
3221 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3224 ret = extent_write_cache_pages(tree, mapping, wbc,
3225 __extent_writepage, &epd,
3227 flush_epd_write_bio(&epd);
3231 int extent_readpages(struct extent_io_tree *tree,
3232 struct address_space *mapping,
3233 struct list_head *pages, unsigned nr_pages,
3234 get_extent_t get_extent)
3236 struct bio *bio = NULL;
3238 unsigned long bio_flags = 0;
3240 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
3241 struct page *page = list_entry(pages->prev, struct page, lru);
3243 prefetchw(&page->flags);
3244 list_del(&page->lru);
3245 if (!add_to_page_cache_lru(page, mapping,
3246 page->index, GFP_NOFS)) {
3247 __extent_read_full_page(tree, page, get_extent,
3248 &bio, 0, &bio_flags);
3250 page_cache_release(page);
3252 BUG_ON(!list_empty(pages));
3254 int ret = submit_one_bio(READ, bio, 0, bio_flags);
3261 * basic invalidatepage code, this waits on any locked or writeback
3262 * ranges corresponding to the page, and then deletes any extent state
3263 * records from the tree
3265 int extent_invalidatepage(struct extent_io_tree *tree,
3266 struct page *page, unsigned long offset)
3268 struct extent_state *cached_state = NULL;
3269 u64 start = ((u64)page->index << PAGE_CACHE_SHIFT);
3270 u64 end = start + PAGE_CACHE_SIZE - 1;
3271 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
3273 start += (offset + blocksize - 1) & ~(blocksize - 1);
3277 lock_extent_bits(tree, start, end, 0, &cached_state, GFP_NOFS);
3278 wait_on_page_writeback(page);
3279 clear_extent_bit(tree, start, end,
3280 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
3281 EXTENT_DO_ACCOUNTING,
3282 1, 1, &cached_state, GFP_NOFS);
3287 * a helper for releasepage, this tests for areas of the page that
3288 * are locked or under IO and drops the related state bits if it is safe
3291 int try_release_extent_state(struct extent_map_tree *map,
3292 struct extent_io_tree *tree, struct page *page,
3295 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
3296 u64 end = start + PAGE_CACHE_SIZE - 1;
3299 if (test_range_bit(tree, start, end,
3300 EXTENT_IOBITS, 0, NULL))
3303 if ((mask & GFP_NOFS) == GFP_NOFS)
3306 * at this point we can safely clear everything except the
3307 * locked bit and the nodatasum bit
3309 ret = clear_extent_bit(tree, start, end,
3310 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
3313 /* if clear_extent_bit failed for enomem reasons,
3314 * we can't allow the release to continue.
3325 * a helper for releasepage. As long as there are no locked extents
3326 * in the range corresponding to the page, both state records and extent
3327 * map records are removed
3329 int try_release_extent_mapping(struct extent_map_tree *map,
3330 struct extent_io_tree *tree, struct page *page,
3333 struct extent_map *em;
3334 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
3335 u64 end = start + PAGE_CACHE_SIZE - 1;
3337 if ((mask & __GFP_WAIT) &&
3338 page->mapping->host->i_size > 16 * 1024 * 1024) {
3340 while (start <= end) {
3341 len = end - start + 1;
3342 write_lock(&map->lock);
3343 em = lookup_extent_mapping(map, start, len);
3345 write_unlock(&map->lock);
3348 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
3349 em->start != start) {
3350 write_unlock(&map->lock);
3351 free_extent_map(em);
3354 if (!test_range_bit(tree, em->start,
3355 extent_map_end(em) - 1,
3356 EXTENT_LOCKED | EXTENT_WRITEBACK,
3358 remove_extent_mapping(map, em);
3359 /* once for the rb tree */
3360 free_extent_map(em);
3362 start = extent_map_end(em);
3363 write_unlock(&map->lock);
3366 free_extent_map(em);
3369 return try_release_extent_state(map, tree, page, mask);
3373 * helper function for fiemap, which doesn't want to see any holes.
3374 * This maps until we find something past 'last'
3376 static struct extent_map *get_extent_skip_holes(struct inode *inode,
3379 get_extent_t *get_extent)
3381 u64 sectorsize = BTRFS_I(inode)->root->sectorsize;
3382 struct extent_map *em;
3389 len = last - offset;
3392 len = (len + sectorsize - 1) & ~(sectorsize - 1);
3393 em = get_extent(inode, NULL, 0, offset, len, 0);
3394 if (IS_ERR_OR_NULL(em))
3397 /* if this isn't a hole return it */
3398 if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) &&
3399 em->block_start != EXTENT_MAP_HOLE) {
3403 /* this is a hole, advance to the next extent */
3404 offset = extent_map_end(em);
3405 free_extent_map(em);
3412 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
3413 __u64 start, __u64 len, get_extent_t *get_extent)
3417 u64 max = start + len;
3421 u64 last_for_get_extent = 0;
3423 u64 isize = i_size_read(inode);
3424 struct btrfs_key found_key;
3425 struct extent_map *em = NULL;
3426 struct extent_state *cached_state = NULL;
3427 struct btrfs_path *path;
3428 struct btrfs_file_extent_item *item;
3433 unsigned long emflags;
3438 path = btrfs_alloc_path();
3441 path->leave_spinning = 1;
3443 start = ALIGN(start, BTRFS_I(inode)->root->sectorsize);
3444 len = ALIGN(len, BTRFS_I(inode)->root->sectorsize);
3447 * lookup the last file extent. We're not using i_size here
3448 * because there might be preallocation past i_size
3450 ret = btrfs_lookup_file_extent(NULL, BTRFS_I(inode)->root,
3451 path, btrfs_ino(inode), -1, 0);
3453 btrfs_free_path(path);
3458 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3459 struct btrfs_file_extent_item);
3460 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
3461 found_type = btrfs_key_type(&found_key);
3463 /* No extents, but there might be delalloc bits */
3464 if (found_key.objectid != btrfs_ino(inode) ||
3465 found_type != BTRFS_EXTENT_DATA_KEY) {
3466 /* have to trust i_size as the end */
3468 last_for_get_extent = isize;
3471 * remember the start of the last extent. There are a
3472 * bunch of different factors that go into the length of the
3473 * extent, so its much less complex to remember where it started
3475 last = found_key.offset;
3476 last_for_get_extent = last + 1;
3478 btrfs_free_path(path);
3481 * we might have some extents allocated but more delalloc past those
3482 * extents. so, we trust isize unless the start of the last extent is
3487 last_for_get_extent = isize;
3490 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len, 0,
3491 &cached_state, GFP_NOFS);
3493 em = get_extent_skip_holes(inode, start, last_for_get_extent,
3503 u64 offset_in_extent;
3505 /* break if the extent we found is outside the range */
3506 if (em->start >= max || extent_map_end(em) < off)
3510 * get_extent may return an extent that starts before our
3511 * requested range. We have to make sure the ranges
3512 * we return to fiemap always move forward and don't
3513 * overlap, so adjust the offsets here
3515 em_start = max(em->start, off);
3518 * record the offset from the start of the extent
3519 * for adjusting the disk offset below
3521 offset_in_extent = em_start - em->start;
3522 em_end = extent_map_end(em);
3523 em_len = em_end - em_start;
3524 emflags = em->flags;
3529 * bump off for our next call to get_extent
3531 off = extent_map_end(em);
3535 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
3537 flags |= FIEMAP_EXTENT_LAST;
3538 } else if (em->block_start == EXTENT_MAP_INLINE) {
3539 flags |= (FIEMAP_EXTENT_DATA_INLINE |
3540 FIEMAP_EXTENT_NOT_ALIGNED);
3541 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
3542 flags |= (FIEMAP_EXTENT_DELALLOC |
3543 FIEMAP_EXTENT_UNKNOWN);
3545 disko = em->block_start + offset_in_extent;
3547 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
3548 flags |= FIEMAP_EXTENT_ENCODED;
3550 free_extent_map(em);
3552 if ((em_start >= last) || em_len == (u64)-1 ||
3553 (last == (u64)-1 && isize <= em_end)) {
3554 flags |= FIEMAP_EXTENT_LAST;
3558 /* now scan forward to see if this is really the last extent. */
3559 em = get_extent_skip_holes(inode, off, last_for_get_extent,
3566 flags |= FIEMAP_EXTENT_LAST;
3569 ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
3575 free_extent_map(em);
3577 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len,
3578 &cached_state, GFP_NOFS);
3582 inline struct page *extent_buffer_page(struct extent_buffer *eb,
3586 struct address_space *mapping;
3589 return eb->first_page;
3590 i += eb->start >> PAGE_CACHE_SHIFT;
3591 mapping = eb->first_page->mapping;
3596 * extent_buffer_page is only called after pinning the page
3597 * by increasing the reference count. So we know the page must
3598 * be in the radix tree.
3601 p = radix_tree_lookup(&mapping->page_tree, i);
3607 inline unsigned long num_extent_pages(u64 start, u64 len)
3609 return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
3610 (start >> PAGE_CACHE_SHIFT);
3613 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
3618 struct extent_buffer *eb = NULL;
3620 unsigned long flags;
3623 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
3628 rwlock_init(&eb->lock);
3629 atomic_set(&eb->write_locks, 0);
3630 atomic_set(&eb->read_locks, 0);
3631 atomic_set(&eb->blocking_readers, 0);
3632 atomic_set(&eb->blocking_writers, 0);
3633 atomic_set(&eb->spinning_readers, 0);
3634 atomic_set(&eb->spinning_writers, 0);
3635 eb->lock_nested = 0;
3636 init_waitqueue_head(&eb->write_lock_wq);
3637 init_waitqueue_head(&eb->read_lock_wq);
3640 spin_lock_irqsave(&leak_lock, flags);
3641 list_add(&eb->leak_list, &buffers);
3642 spin_unlock_irqrestore(&leak_lock, flags);
3644 atomic_set(&eb->refs, 1);
3649 static void __free_extent_buffer(struct extent_buffer *eb)
3652 unsigned long flags;
3653 spin_lock_irqsave(&leak_lock, flags);
3654 list_del(&eb->leak_list);
3655 spin_unlock_irqrestore(&leak_lock, flags);
3657 kmem_cache_free(extent_buffer_cache, eb);
3661 * Helper for releasing extent buffer page.
3663 static void btrfs_release_extent_buffer_page(struct extent_buffer *eb,
3664 unsigned long start_idx)
3666 unsigned long index;
3669 if (!eb->first_page)
3672 index = num_extent_pages(eb->start, eb->len);
3673 if (start_idx >= index)
3678 page = extent_buffer_page(eb, index);
3680 page_cache_release(page);
3681 } while (index != start_idx);
3685 * Helper for releasing the extent buffer.
3687 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
3689 btrfs_release_extent_buffer_page(eb, 0);
3690 __free_extent_buffer(eb);
3693 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
3694 u64 start, unsigned long len,
3697 unsigned long num_pages = num_extent_pages(start, len);
3699 unsigned long index = start >> PAGE_CACHE_SHIFT;
3700 struct extent_buffer *eb;
3701 struct extent_buffer *exists = NULL;
3703 struct address_space *mapping = tree->mapping;
3708 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
3709 if (eb && atomic_inc_not_zero(&eb->refs)) {
3711 mark_page_accessed(eb->first_page);
3716 eb = __alloc_extent_buffer(tree, start, len, GFP_NOFS);
3721 eb->first_page = page0;
3724 page_cache_get(page0);
3725 mark_page_accessed(page0);
3726 set_page_extent_mapped(page0);
3727 set_page_extent_head(page0, len);
3728 uptodate = PageUptodate(page0);
3732 for (; i < num_pages; i++, index++) {
3733 p = find_or_create_page(mapping, index, GFP_NOFS);
3738 set_page_extent_mapped(p);
3739 mark_page_accessed(p);
3742 set_page_extent_head(p, len);
3744 set_page_private(p, EXTENT_PAGE_PRIVATE);
3746 if (!PageUptodate(p))
3750 * see below about how we avoid a nasty race with release page
3751 * and why we unlock later
3757 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3759 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
3763 spin_lock(&tree->buffer_lock);
3764 ret = radix_tree_insert(&tree->buffer, start >> PAGE_CACHE_SHIFT, eb);
3765 if (ret == -EEXIST) {
3766 exists = radix_tree_lookup(&tree->buffer,
3767 start >> PAGE_CACHE_SHIFT);
3768 /* add one reference for the caller */
3769 atomic_inc(&exists->refs);
3770 spin_unlock(&tree->buffer_lock);
3771 radix_tree_preload_end();
3774 /* add one reference for the tree */
3775 atomic_inc(&eb->refs);
3776 spin_unlock(&tree->buffer_lock);
3777 radix_tree_preload_end();
3780 * there is a race where release page may have
3781 * tried to find this extent buffer in the radix
3782 * but failed. It will tell the VM it is safe to
3783 * reclaim the, and it will clear the page private bit.
3784 * We must make sure to set the page private bit properly
3785 * after the extent buffer is in the radix tree so
3786 * it doesn't get lost
3788 set_page_extent_mapped(eb->first_page);
3789 set_page_extent_head(eb->first_page, eb->len);
3791 unlock_page(eb->first_page);
3795 if (eb->first_page && !page0)
3796 unlock_page(eb->first_page);
3798 if (!atomic_dec_and_test(&eb->refs))
3800 btrfs_release_extent_buffer(eb);
3804 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
3805 u64 start, unsigned long len)
3807 struct extent_buffer *eb;
3810 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
3811 if (eb && atomic_inc_not_zero(&eb->refs)) {
3813 mark_page_accessed(eb->first_page);
3821 void free_extent_buffer(struct extent_buffer *eb)
3826 if (!atomic_dec_and_test(&eb->refs))
3832 void clear_extent_buffer_dirty(struct extent_io_tree *tree,
3833 struct extent_buffer *eb)
3836 unsigned long num_pages;
3839 num_pages = num_extent_pages(eb->start, eb->len);
3841 for (i = 0; i < num_pages; i++) {
3842 page = extent_buffer_page(eb, i);
3843 if (!PageDirty(page))
3847 WARN_ON(!PagePrivate(page));
3849 set_page_extent_mapped(page);
3851 set_page_extent_head(page, eb->len);
3853 clear_page_dirty_for_io(page);
3854 spin_lock_irq(&page->mapping->tree_lock);
3855 if (!PageDirty(page)) {
3856 radix_tree_tag_clear(&page->mapping->page_tree,
3858 PAGECACHE_TAG_DIRTY);
3860 spin_unlock_irq(&page->mapping->tree_lock);
3861 ClearPageError(page);
3866 int set_extent_buffer_dirty(struct extent_io_tree *tree,
3867 struct extent_buffer *eb)
3870 unsigned long num_pages;
3873 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
3874 num_pages = num_extent_pages(eb->start, eb->len);
3875 for (i = 0; i < num_pages; i++)
3876 __set_page_dirty_nobuffers(extent_buffer_page(eb, i));
3880 static int __eb_straddles_pages(u64 start, u64 len)
3882 if (len < PAGE_CACHE_SIZE)
3884 if (start & (PAGE_CACHE_SIZE - 1))
3886 if ((start + len) & (PAGE_CACHE_SIZE - 1))
3891 static int eb_straddles_pages(struct extent_buffer *eb)
3893 return __eb_straddles_pages(eb->start, eb->len);
3896 int clear_extent_buffer_uptodate(struct extent_io_tree *tree,
3897 struct extent_buffer *eb,
3898 struct extent_state **cached_state)
3902 unsigned long num_pages;
3904 num_pages = num_extent_pages(eb->start, eb->len);
3905 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3907 clear_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
3908 cached_state, GFP_NOFS);
3910 for (i = 0; i < num_pages; i++) {
3911 page = extent_buffer_page(eb, i);
3913 ClearPageUptodate(page);
3918 int set_extent_buffer_uptodate(struct extent_io_tree *tree,
3919 struct extent_buffer *eb)
3923 unsigned long num_pages;
3925 num_pages = num_extent_pages(eb->start, eb->len);
3927 if (eb_straddles_pages(eb)) {
3928 set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
3931 for (i = 0; i < num_pages; i++) {
3932 page = extent_buffer_page(eb, i);
3933 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
3934 ((i == num_pages - 1) &&
3935 ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
3936 check_page_uptodate(tree, page);
3939 SetPageUptodate(page);
3944 int extent_range_uptodate(struct extent_io_tree *tree,
3949 int pg_uptodate = 1;
3951 unsigned long index;
3953 if (__eb_straddles_pages(start, end - start + 1)) {
3954 ret = test_range_bit(tree, start, end,
3955 EXTENT_UPTODATE, 1, NULL);
3959 while (start <= end) {
3960 index = start >> PAGE_CACHE_SHIFT;
3961 page = find_get_page(tree->mapping, index);
3964 uptodate = PageUptodate(page);
3965 page_cache_release(page);
3970 start += PAGE_CACHE_SIZE;
3975 int extent_buffer_uptodate(struct extent_io_tree *tree,
3976 struct extent_buffer *eb,
3977 struct extent_state *cached_state)
3980 unsigned long num_pages;
3983 int pg_uptodate = 1;
3985 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
3988 if (eb_straddles_pages(eb)) {
3989 ret = test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3990 EXTENT_UPTODATE, 1, cached_state);
3995 num_pages = num_extent_pages(eb->start, eb->len);
3996 for (i = 0; i < num_pages; i++) {
3997 page = extent_buffer_page(eb, i);
3998 if (!PageUptodate(page)) {
4006 int read_extent_buffer_pages(struct extent_io_tree *tree,
4007 struct extent_buffer *eb, u64 start, int wait,
4008 get_extent_t *get_extent, int mirror_num)
4011 unsigned long start_i;
4015 int locked_pages = 0;
4016 int all_uptodate = 1;
4017 int inc_all_pages = 0;
4018 unsigned long num_pages;
4019 struct bio *bio = NULL;
4020 unsigned long bio_flags = 0;
4022 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
4025 if (eb_straddles_pages(eb)) {
4026 if (test_range_bit(tree, eb->start, eb->start + eb->len - 1,
4027 EXTENT_UPTODATE, 1, NULL)) {
4033 WARN_ON(start < eb->start);
4034 start_i = (start >> PAGE_CACHE_SHIFT) -
4035 (eb->start >> PAGE_CACHE_SHIFT);
4040 num_pages = num_extent_pages(eb->start, eb->len);
4041 for (i = start_i; i < num_pages; i++) {
4042 page = extent_buffer_page(eb, i);
4043 if (wait == WAIT_NONE) {
4044 if (!trylock_page(page))
4050 if (!PageUptodate(page))
4055 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4059 for (i = start_i; i < num_pages; i++) {
4060 page = extent_buffer_page(eb, i);
4062 WARN_ON(!PagePrivate(page));
4064 set_page_extent_mapped(page);
4066 set_page_extent_head(page, eb->len);
4069 page_cache_get(page);
4070 if (!PageUptodate(page)) {
4073 ClearPageError(page);
4074 err = __extent_read_full_page(tree, page,
4076 mirror_num, &bio_flags);
4085 err = submit_one_bio(READ, bio, mirror_num, bio_flags);
4089 if (ret || wait != WAIT_COMPLETE)
4092 for (i = start_i; i < num_pages; i++) {
4093 page = extent_buffer_page(eb, i);
4094 wait_on_page_locked(page);
4095 if (!PageUptodate(page))
4100 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4105 while (locked_pages > 0) {
4106 page = extent_buffer_page(eb, i);
4114 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
4115 unsigned long start,
4122 char *dst = (char *)dstv;
4123 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4124 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4126 WARN_ON(start > eb->len);
4127 WARN_ON(start + len > eb->start + eb->len);
4129 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4132 page = extent_buffer_page(eb, i);
4134 cur = min(len, (PAGE_CACHE_SIZE - offset));
4135 kaddr = page_address(page);
4136 memcpy(dst, kaddr + offset, cur);
4145 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
4146 unsigned long min_len, char **map,
4147 unsigned long *map_start,
4148 unsigned long *map_len)
4150 size_t offset = start & (PAGE_CACHE_SIZE - 1);
4153 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4154 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4155 unsigned long end_i = (start_offset + start + min_len - 1) >>
4162 offset = start_offset;
4166 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
4169 if (start + min_len > eb->len) {
4170 printk(KERN_ERR "btrfs bad mapping eb start %llu len %lu, "
4171 "wanted %lu %lu\n", (unsigned long long)eb->start,
4172 eb->len, start, min_len);
4177 p = extent_buffer_page(eb, i);
4178 kaddr = page_address(p);
4179 *map = kaddr + offset;
4180 *map_len = PAGE_CACHE_SIZE - offset;
4184 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
4185 unsigned long start,
4192 char *ptr = (char *)ptrv;
4193 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4194 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4197 WARN_ON(start > eb->len);
4198 WARN_ON(start + len > eb->start + eb->len);
4200 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4203 page = extent_buffer_page(eb, i);
4205 cur = min(len, (PAGE_CACHE_SIZE - offset));
4207 kaddr = page_address(page);
4208 ret = memcmp(ptr, kaddr + offset, cur);
4220 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
4221 unsigned long start, unsigned long len)
4227 char *src = (char *)srcv;
4228 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4229 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4231 WARN_ON(start > eb->len);
4232 WARN_ON(start + len > eb->start + eb->len);
4234 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4237 page = extent_buffer_page(eb, i);
4238 WARN_ON(!PageUptodate(page));
4240 cur = min(len, PAGE_CACHE_SIZE - offset);
4241 kaddr = page_address(page);
4242 memcpy(kaddr + offset, src, cur);
4251 void memset_extent_buffer(struct extent_buffer *eb, char c,
4252 unsigned long start, unsigned long len)
4258 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4259 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4261 WARN_ON(start > eb->len);
4262 WARN_ON(start + len > eb->start + eb->len);
4264 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4267 page = extent_buffer_page(eb, i);
4268 WARN_ON(!PageUptodate(page));
4270 cur = min(len, PAGE_CACHE_SIZE - offset);
4271 kaddr = page_address(page);
4272 memset(kaddr + offset, c, cur);
4280 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
4281 unsigned long dst_offset, unsigned long src_offset,
4284 u64 dst_len = dst->len;
4289 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4290 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
4292 WARN_ON(src->len != dst_len);
4294 offset = (start_offset + dst_offset) &
4295 ((unsigned long)PAGE_CACHE_SIZE - 1);
4298 page = extent_buffer_page(dst, i);
4299 WARN_ON(!PageUptodate(page));
4301 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
4303 kaddr = page_address(page);
4304 read_extent_buffer(src, kaddr + offset, src_offset, cur);
4313 static void move_pages(struct page *dst_page, struct page *src_page,
4314 unsigned long dst_off, unsigned long src_off,
4317 char *dst_kaddr = page_address(dst_page);
4318 if (dst_page == src_page) {
4319 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
4321 char *src_kaddr = page_address(src_page);
4322 char *p = dst_kaddr + dst_off + len;
4323 char *s = src_kaddr + src_off + len;
4330 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
4332 unsigned long distance = (src > dst) ? src - dst : dst - src;
4333 return distance < len;
4336 static void copy_pages(struct page *dst_page, struct page *src_page,
4337 unsigned long dst_off, unsigned long src_off,
4340 char *dst_kaddr = page_address(dst_page);
4343 if (dst_page != src_page) {
4344 src_kaddr = page_address(src_page);
4346 src_kaddr = dst_kaddr;
4347 BUG_ON(areas_overlap(src_off, dst_off, len));
4350 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
4353 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
4354 unsigned long src_offset, unsigned long len)
4357 size_t dst_off_in_page;
4358 size_t src_off_in_page;
4359 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4360 unsigned long dst_i;
4361 unsigned long src_i;
4363 if (src_offset + len > dst->len) {
4364 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
4365 "len %lu dst len %lu\n", src_offset, len, dst->len);
4368 if (dst_offset + len > dst->len) {
4369 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
4370 "len %lu dst len %lu\n", dst_offset, len, dst->len);
4375 dst_off_in_page = (start_offset + dst_offset) &
4376 ((unsigned long)PAGE_CACHE_SIZE - 1);
4377 src_off_in_page = (start_offset + src_offset) &
4378 ((unsigned long)PAGE_CACHE_SIZE - 1);
4380 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
4381 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
4383 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
4385 cur = min_t(unsigned long, cur,
4386 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
4388 copy_pages(extent_buffer_page(dst, dst_i),
4389 extent_buffer_page(dst, src_i),
4390 dst_off_in_page, src_off_in_page, cur);
4398 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
4399 unsigned long src_offset, unsigned long len)
4402 size_t dst_off_in_page;
4403 size_t src_off_in_page;
4404 unsigned long dst_end = dst_offset + len - 1;
4405 unsigned long src_end = src_offset + len - 1;
4406 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4407 unsigned long dst_i;
4408 unsigned long src_i;
4410 if (src_offset + len > dst->len) {
4411 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
4412 "len %lu len %lu\n", src_offset, len, dst->len);
4415 if (dst_offset + len > dst->len) {
4416 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
4417 "len %lu len %lu\n", dst_offset, len, dst->len);
4420 if (!areas_overlap(src_offset, dst_offset, len)) {
4421 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
4425 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
4426 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
4428 dst_off_in_page = (start_offset + dst_end) &
4429 ((unsigned long)PAGE_CACHE_SIZE - 1);
4430 src_off_in_page = (start_offset + src_end) &
4431 ((unsigned long)PAGE_CACHE_SIZE - 1);
4433 cur = min_t(unsigned long, len, src_off_in_page + 1);
4434 cur = min(cur, dst_off_in_page + 1);
4435 move_pages(extent_buffer_page(dst, dst_i),
4436 extent_buffer_page(dst, src_i),
4437 dst_off_in_page - cur + 1,
4438 src_off_in_page - cur + 1, cur);
4446 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
4448 struct extent_buffer *eb =
4449 container_of(head, struct extent_buffer, rcu_head);
4451 btrfs_release_extent_buffer(eb);
4454 int try_release_extent_buffer(struct extent_io_tree *tree, struct page *page)
4456 u64 start = page_offset(page);
4457 struct extent_buffer *eb;
4460 spin_lock(&tree->buffer_lock);
4461 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
4463 spin_unlock(&tree->buffer_lock);
4467 if (test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
4473 * set @eb->refs to 0 if it is already 1, and then release the @eb.
4476 if (atomic_cmpxchg(&eb->refs, 1, 0) != 1) {
4481 radix_tree_delete(&tree->buffer, start >> PAGE_CACHE_SHIFT);
4483 spin_unlock(&tree->buffer_lock);
4485 /* at this point we can safely release the extent buffer */
4486 if (atomic_read(&eb->refs) == 0)
4487 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);