1 #include <linux/bitops.h>
2 #include <linux/slab.h>
5 #include <linux/pagemap.h>
6 #include <linux/page-flags.h>
7 #include <linux/spinlock.h>
8 #include <linux/blkdev.h>
9 #include <linux/swap.h>
10 #include <linux/writeback.h>
11 #include <linux/pagevec.h>
12 #include <linux/prefetch.h>
13 #include <linux/cleancache.h>
14 #include "extent_io.h"
15 #include "extent_map.h"
17 #include "btrfs_inode.h"
19 #include "check-integrity.h"
21 #include "rcu-string.h"
24 static struct kmem_cache *extent_state_cache;
25 static struct kmem_cache *extent_buffer_cache;
26 static struct bio_set *btrfs_bioset;
28 #ifdef CONFIG_BTRFS_DEBUG
29 static LIST_HEAD(buffers);
30 static LIST_HEAD(states);
32 static DEFINE_SPINLOCK(leak_lock);
35 void btrfs_leak_debug_add(struct list_head *new, struct list_head *head)
39 spin_lock_irqsave(&leak_lock, flags);
41 spin_unlock_irqrestore(&leak_lock, flags);
45 void btrfs_leak_debug_del(struct list_head *entry)
49 spin_lock_irqsave(&leak_lock, flags);
51 spin_unlock_irqrestore(&leak_lock, flags);
55 void btrfs_leak_debug_check(void)
57 struct extent_state *state;
58 struct extent_buffer *eb;
60 while (!list_empty(&states)) {
61 state = list_entry(states.next, struct extent_state, leak_list);
62 printk(KERN_ERR "BTRFS: state leak: start %llu end %llu "
63 "state %lu in tree %p refs %d\n",
64 state->start, state->end, state->state, state->tree,
65 atomic_read(&state->refs));
66 list_del(&state->leak_list);
67 kmem_cache_free(extent_state_cache, state);
70 while (!list_empty(&buffers)) {
71 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
72 printk(KERN_ERR "BTRFS: buffer leak start %llu len %lu "
74 eb->start, eb->len, atomic_read(&eb->refs));
75 list_del(&eb->leak_list);
76 kmem_cache_free(extent_buffer_cache, eb);
80 #define btrfs_debug_check_extent_io_range(tree, start, end) \
81 __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end))
82 static inline void __btrfs_debug_check_extent_io_range(const char *caller,
83 struct extent_io_tree *tree, u64 start, u64 end)
91 inode = tree->mapping->host;
92 isize = i_size_read(inode);
93 if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) {
94 printk_ratelimited(KERN_DEBUG
95 "BTRFS: %s: ino %llu isize %llu odd range [%llu,%llu]\n",
96 caller, btrfs_ino(inode), isize, start, end);
100 #define btrfs_leak_debug_add(new, head) do {} while (0)
101 #define btrfs_leak_debug_del(entry) do {} while (0)
102 #define btrfs_leak_debug_check() do {} while (0)
103 #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0)
106 #define BUFFER_LRU_MAX 64
111 struct rb_node rb_node;
114 struct extent_page_data {
116 struct extent_io_tree *tree;
117 get_extent_t *get_extent;
118 unsigned long bio_flags;
120 /* tells writepage not to lock the state bits for this range
121 * it still does the unlocking
123 unsigned int extent_locked:1;
125 /* tells the submit_bio code to use a WRITE_SYNC */
126 unsigned int sync_io:1;
129 static noinline void flush_write_bio(void *data);
130 static inline struct btrfs_fs_info *
131 tree_fs_info(struct extent_io_tree *tree)
135 return btrfs_sb(tree->mapping->host->i_sb);
138 int __init extent_io_init(void)
140 extent_state_cache = kmem_cache_create("btrfs_extent_state",
141 sizeof(struct extent_state), 0,
142 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
143 if (!extent_state_cache)
146 extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer",
147 sizeof(struct extent_buffer), 0,
148 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
149 if (!extent_buffer_cache)
150 goto free_state_cache;
152 btrfs_bioset = bioset_create(BIO_POOL_SIZE,
153 offsetof(struct btrfs_io_bio, bio));
155 goto free_buffer_cache;
157 if (bioset_integrity_create(btrfs_bioset, BIO_POOL_SIZE))
163 bioset_free(btrfs_bioset);
167 kmem_cache_destroy(extent_buffer_cache);
168 extent_buffer_cache = NULL;
171 kmem_cache_destroy(extent_state_cache);
172 extent_state_cache = NULL;
176 void extent_io_exit(void)
178 btrfs_leak_debug_check();
181 * Make sure all delayed rcu free are flushed before we
185 if (extent_state_cache)
186 kmem_cache_destroy(extent_state_cache);
187 if (extent_buffer_cache)
188 kmem_cache_destroy(extent_buffer_cache);
190 bioset_free(btrfs_bioset);
193 void extent_io_tree_init(struct extent_io_tree *tree,
194 struct address_space *mapping)
196 tree->state = RB_ROOT;
198 tree->dirty_bytes = 0;
199 spin_lock_init(&tree->lock);
200 tree->mapping = mapping;
203 static struct extent_state *alloc_extent_state(gfp_t mask)
205 struct extent_state *state;
207 state = kmem_cache_alloc(extent_state_cache, mask);
213 btrfs_leak_debug_add(&state->leak_list, &states);
214 atomic_set(&state->refs, 1);
215 init_waitqueue_head(&state->wq);
216 trace_alloc_extent_state(state, mask, _RET_IP_);
220 void free_extent_state(struct extent_state *state)
224 if (atomic_dec_and_test(&state->refs)) {
225 WARN_ON(state->tree);
226 btrfs_leak_debug_del(&state->leak_list);
227 trace_free_extent_state(state, _RET_IP_);
228 kmem_cache_free(extent_state_cache, state);
232 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
233 struct rb_node *node,
234 struct rb_node ***p_in,
235 struct rb_node **parent_in)
237 struct rb_node **p = &root->rb_node;
238 struct rb_node *parent = NULL;
239 struct tree_entry *entry;
241 if (p_in && parent_in) {
249 entry = rb_entry(parent, struct tree_entry, rb_node);
251 if (offset < entry->start)
253 else if (offset > entry->end)
260 rb_link_node(node, parent, p);
261 rb_insert_color(node, root);
265 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
266 struct rb_node **prev_ret,
267 struct rb_node **next_ret,
268 struct rb_node ***p_ret,
269 struct rb_node **parent_ret)
271 struct rb_root *root = &tree->state;
272 struct rb_node **n = &root->rb_node;
273 struct rb_node *prev = NULL;
274 struct rb_node *orig_prev = NULL;
275 struct tree_entry *entry;
276 struct tree_entry *prev_entry = NULL;
280 entry = rb_entry(prev, struct tree_entry, rb_node);
283 if (offset < entry->start)
285 else if (offset > entry->end)
298 while (prev && offset > prev_entry->end) {
299 prev = rb_next(prev);
300 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
307 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
308 while (prev && offset < prev_entry->start) {
309 prev = rb_prev(prev);
310 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
317 static inline struct rb_node *
318 tree_search_for_insert(struct extent_io_tree *tree,
320 struct rb_node ***p_ret,
321 struct rb_node **parent_ret)
323 struct rb_node *prev = NULL;
326 ret = __etree_search(tree, offset, &prev, NULL, p_ret, parent_ret);
332 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
335 return tree_search_for_insert(tree, offset, NULL, NULL);
338 static void merge_cb(struct extent_io_tree *tree, struct extent_state *new,
339 struct extent_state *other)
341 if (tree->ops && tree->ops->merge_extent_hook)
342 tree->ops->merge_extent_hook(tree->mapping->host, new,
347 * utility function to look for merge candidates inside a given range.
348 * Any extents with matching state are merged together into a single
349 * extent in the tree. Extents with EXTENT_IO in their state field
350 * are not merged because the end_io handlers need to be able to do
351 * operations on them without sleeping (or doing allocations/splits).
353 * This should be called with the tree lock held.
355 static void merge_state(struct extent_io_tree *tree,
356 struct extent_state *state)
358 struct extent_state *other;
359 struct rb_node *other_node;
361 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
364 other_node = rb_prev(&state->rb_node);
366 other = rb_entry(other_node, struct extent_state, rb_node);
367 if (other->end == state->start - 1 &&
368 other->state == state->state) {
369 merge_cb(tree, state, other);
370 state->start = other->start;
372 rb_erase(&other->rb_node, &tree->state);
373 free_extent_state(other);
376 other_node = rb_next(&state->rb_node);
378 other = rb_entry(other_node, struct extent_state, rb_node);
379 if (other->start == state->end + 1 &&
380 other->state == state->state) {
381 merge_cb(tree, state, other);
382 state->end = other->end;
384 rb_erase(&other->rb_node, &tree->state);
385 free_extent_state(other);
390 static void set_state_cb(struct extent_io_tree *tree,
391 struct extent_state *state, unsigned long *bits)
393 if (tree->ops && tree->ops->set_bit_hook)
394 tree->ops->set_bit_hook(tree->mapping->host, state, bits);
397 static void clear_state_cb(struct extent_io_tree *tree,
398 struct extent_state *state, unsigned long *bits)
400 if (tree->ops && tree->ops->clear_bit_hook)
401 tree->ops->clear_bit_hook(tree->mapping->host, state, bits);
404 static void set_state_bits(struct extent_io_tree *tree,
405 struct extent_state *state, unsigned long *bits);
408 * insert an extent_state struct into the tree. 'bits' are set on the
409 * struct before it is inserted.
411 * This may return -EEXIST if the extent is already there, in which case the
412 * state struct is freed.
414 * The tree lock is not taken internally. This is a utility function and
415 * probably isn't what you want to call (see set/clear_extent_bit).
417 static int insert_state(struct extent_io_tree *tree,
418 struct extent_state *state, u64 start, u64 end,
420 struct rb_node **parent,
423 struct rb_node *node;
426 WARN(1, KERN_ERR "BTRFS: end < start %llu %llu\n",
428 state->start = start;
431 set_state_bits(tree, state, bits);
433 node = tree_insert(&tree->state, end, &state->rb_node, p, parent);
435 struct extent_state *found;
436 found = rb_entry(node, struct extent_state, rb_node);
437 printk(KERN_ERR "BTRFS: found node %llu %llu on insert of "
439 found->start, found->end, start, end);
443 merge_state(tree, state);
447 static void split_cb(struct extent_io_tree *tree, struct extent_state *orig,
450 if (tree->ops && tree->ops->split_extent_hook)
451 tree->ops->split_extent_hook(tree->mapping->host, orig, split);
455 * split a given extent state struct in two, inserting the preallocated
456 * struct 'prealloc' as the newly created second half. 'split' indicates an
457 * offset inside 'orig' where it should be split.
460 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
461 * are two extent state structs in the tree:
462 * prealloc: [orig->start, split - 1]
463 * orig: [ split, orig->end ]
465 * The tree locks are not taken by this function. They need to be held
468 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
469 struct extent_state *prealloc, u64 split)
471 struct rb_node *node;
473 split_cb(tree, orig, split);
475 prealloc->start = orig->start;
476 prealloc->end = split - 1;
477 prealloc->state = orig->state;
480 node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node,
483 free_extent_state(prealloc);
486 prealloc->tree = tree;
490 static struct extent_state *next_state(struct extent_state *state)
492 struct rb_node *next = rb_next(&state->rb_node);
494 return rb_entry(next, struct extent_state, rb_node);
500 * utility function to clear some bits in an extent state struct.
501 * it will optionally wake up any one waiting on this state (wake == 1).
503 * If no bits are set on the state struct after clearing things, the
504 * struct is freed and removed from the tree
506 static struct extent_state *clear_state_bit(struct extent_io_tree *tree,
507 struct extent_state *state,
508 unsigned long *bits, int wake)
510 struct extent_state *next;
511 unsigned long bits_to_clear = *bits & ~EXTENT_CTLBITS;
513 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
514 u64 range = state->end - state->start + 1;
515 WARN_ON(range > tree->dirty_bytes);
516 tree->dirty_bytes -= range;
518 clear_state_cb(tree, state, bits);
519 state->state &= ~bits_to_clear;
522 if (state->state == 0) {
523 next = next_state(state);
525 rb_erase(&state->rb_node, &tree->state);
527 free_extent_state(state);
532 merge_state(tree, state);
533 next = next_state(state);
538 static struct extent_state *
539 alloc_extent_state_atomic(struct extent_state *prealloc)
542 prealloc = alloc_extent_state(GFP_ATOMIC);
547 static void extent_io_tree_panic(struct extent_io_tree *tree, int err)
549 btrfs_panic(tree_fs_info(tree), err, "Locking error: "
550 "Extent tree was modified by another "
551 "thread while locked.");
555 * clear some bits on a range in the tree. This may require splitting
556 * or inserting elements in the tree, so the gfp mask is used to
557 * indicate which allocations or sleeping are allowed.
559 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
560 * the given range from the tree regardless of state (ie for truncate).
562 * the range [start, end] is inclusive.
564 * This takes the tree lock, and returns 0 on success and < 0 on error.
566 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
567 unsigned long bits, int wake, int delete,
568 struct extent_state **cached_state,
571 struct extent_state *state;
572 struct extent_state *cached;
573 struct extent_state *prealloc = NULL;
574 struct rb_node *node;
579 btrfs_debug_check_extent_io_range(tree, start, end);
581 if (bits & EXTENT_DELALLOC)
582 bits |= EXTENT_NORESERVE;
585 bits |= ~EXTENT_CTLBITS;
586 bits |= EXTENT_FIRST_DELALLOC;
588 if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY))
591 if (!prealloc && (mask & __GFP_WAIT)) {
592 prealloc = alloc_extent_state(mask);
597 spin_lock(&tree->lock);
599 cached = *cached_state;
602 *cached_state = NULL;
606 if (cached && cached->tree && cached->start <= start &&
607 cached->end > start) {
609 atomic_dec(&cached->refs);
614 free_extent_state(cached);
617 * this search will find the extents that end after
620 node = tree_search(tree, start);
623 state = rb_entry(node, struct extent_state, rb_node);
625 if (state->start > end)
627 WARN_ON(state->end < start);
628 last_end = state->end;
630 /* the state doesn't have the wanted bits, go ahead */
631 if (!(state->state & bits)) {
632 state = next_state(state);
637 * | ---- desired range ---- |
639 * | ------------- state -------------- |
641 * We need to split the extent we found, and may flip
642 * bits on second half.
644 * If the extent we found extends past our range, we
645 * just split and search again. It'll get split again
646 * the next time though.
648 * If the extent we found is inside our range, we clear
649 * the desired bit on it.
652 if (state->start < start) {
653 prealloc = alloc_extent_state_atomic(prealloc);
655 err = split_state(tree, state, prealloc, start);
657 extent_io_tree_panic(tree, err);
662 if (state->end <= end) {
663 state = clear_state_bit(tree, state, &bits, wake);
669 * | ---- desired range ---- |
671 * We need to split the extent, and clear the bit
674 if (state->start <= end && state->end > end) {
675 prealloc = alloc_extent_state_atomic(prealloc);
677 err = split_state(tree, state, prealloc, end + 1);
679 extent_io_tree_panic(tree, err);
684 clear_state_bit(tree, prealloc, &bits, wake);
690 state = clear_state_bit(tree, state, &bits, wake);
692 if (last_end == (u64)-1)
694 start = last_end + 1;
695 if (start <= end && state && !need_resched())
700 spin_unlock(&tree->lock);
702 free_extent_state(prealloc);
709 spin_unlock(&tree->lock);
710 if (mask & __GFP_WAIT)
715 static void wait_on_state(struct extent_io_tree *tree,
716 struct extent_state *state)
717 __releases(tree->lock)
718 __acquires(tree->lock)
721 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
722 spin_unlock(&tree->lock);
724 spin_lock(&tree->lock);
725 finish_wait(&state->wq, &wait);
729 * waits for one or more bits to clear on a range in the state tree.
730 * The range [start, end] is inclusive.
731 * The tree lock is taken by this function
733 static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
736 struct extent_state *state;
737 struct rb_node *node;
739 btrfs_debug_check_extent_io_range(tree, start, end);
741 spin_lock(&tree->lock);
745 * this search will find all the extents that end after
748 node = tree_search(tree, start);
752 state = rb_entry(node, struct extent_state, rb_node);
754 if (state->start > end)
757 if (state->state & bits) {
758 start = state->start;
759 atomic_inc(&state->refs);
760 wait_on_state(tree, state);
761 free_extent_state(state);
764 start = state->end + 1;
769 cond_resched_lock(&tree->lock);
772 spin_unlock(&tree->lock);
775 static void set_state_bits(struct extent_io_tree *tree,
776 struct extent_state *state,
779 unsigned long bits_to_set = *bits & ~EXTENT_CTLBITS;
781 set_state_cb(tree, state, bits);
782 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
783 u64 range = state->end - state->start + 1;
784 tree->dirty_bytes += range;
786 state->state |= bits_to_set;
789 static void cache_state(struct extent_state *state,
790 struct extent_state **cached_ptr)
792 if (cached_ptr && !(*cached_ptr)) {
793 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) {
795 atomic_inc(&state->refs);
801 * set some bits on a range in the tree. This may require allocations or
802 * sleeping, so the gfp mask is used to indicate what is allowed.
804 * If any of the exclusive bits are set, this will fail with -EEXIST if some
805 * part of the range already has the desired bits set. The start of the
806 * existing range is returned in failed_start in this case.
808 * [start, end] is inclusive This takes the tree lock.
811 static int __must_check
812 __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
813 unsigned long bits, unsigned long exclusive_bits,
814 u64 *failed_start, struct extent_state **cached_state,
817 struct extent_state *state;
818 struct extent_state *prealloc = NULL;
819 struct rb_node *node;
821 struct rb_node *parent;
826 btrfs_debug_check_extent_io_range(tree, start, end);
828 bits |= EXTENT_FIRST_DELALLOC;
830 if (!prealloc && (mask & __GFP_WAIT)) {
831 prealloc = alloc_extent_state(mask);
835 spin_lock(&tree->lock);
836 if (cached_state && *cached_state) {
837 state = *cached_state;
838 if (state->start <= start && state->end > start &&
840 node = &state->rb_node;
845 * this search will find all the extents that end after
848 node = tree_search_for_insert(tree, start, &p, &parent);
850 prealloc = alloc_extent_state_atomic(prealloc);
852 err = insert_state(tree, prealloc, start, end,
855 extent_io_tree_panic(tree, err);
857 cache_state(prealloc, cached_state);
861 state = rb_entry(node, struct extent_state, rb_node);
863 last_start = state->start;
864 last_end = state->end;
867 * | ---- desired range ---- |
870 * Just lock what we found and keep going
872 if (state->start == start && state->end <= end) {
873 if (state->state & exclusive_bits) {
874 *failed_start = state->start;
879 set_state_bits(tree, state, &bits);
880 cache_state(state, cached_state);
881 merge_state(tree, state);
882 if (last_end == (u64)-1)
884 start = last_end + 1;
885 state = next_state(state);
886 if (start < end && state && state->start == start &&
893 * | ---- desired range ---- |
896 * | ------------- state -------------- |
898 * We need to split the extent we found, and may flip bits on
901 * If the extent we found extends past our
902 * range, we just split and search again. It'll get split
903 * again the next time though.
905 * If the extent we found is inside our range, we set the
908 if (state->start < start) {
909 if (state->state & exclusive_bits) {
910 *failed_start = start;
915 prealloc = alloc_extent_state_atomic(prealloc);
917 err = split_state(tree, state, prealloc, start);
919 extent_io_tree_panic(tree, err);
924 if (state->end <= end) {
925 set_state_bits(tree, state, &bits);
926 cache_state(state, cached_state);
927 merge_state(tree, state);
928 if (last_end == (u64)-1)
930 start = last_end + 1;
931 state = next_state(state);
932 if (start < end && state && state->start == start &&
939 * | ---- desired range ---- |
940 * | state | or | state |
942 * There's a hole, we need to insert something in it and
943 * ignore the extent we found.
945 if (state->start > start) {
947 if (end < last_start)
950 this_end = last_start - 1;
952 prealloc = alloc_extent_state_atomic(prealloc);
956 * Avoid to free 'prealloc' if it can be merged with
959 err = insert_state(tree, prealloc, start, this_end,
962 extent_io_tree_panic(tree, err);
964 cache_state(prealloc, cached_state);
966 start = this_end + 1;
970 * | ---- desired range ---- |
972 * We need to split the extent, and set the bit
975 if (state->start <= end && state->end > end) {
976 if (state->state & exclusive_bits) {
977 *failed_start = start;
982 prealloc = alloc_extent_state_atomic(prealloc);
984 err = split_state(tree, state, prealloc, end + 1);
986 extent_io_tree_panic(tree, err);
988 set_state_bits(tree, prealloc, &bits);
989 cache_state(prealloc, cached_state);
990 merge_state(tree, prealloc);
998 spin_unlock(&tree->lock);
1000 free_extent_state(prealloc);
1007 spin_unlock(&tree->lock);
1008 if (mask & __GFP_WAIT)
1013 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1014 unsigned long bits, u64 * failed_start,
1015 struct extent_state **cached_state, gfp_t mask)
1017 return __set_extent_bit(tree, start, end, bits, 0, failed_start,
1018 cached_state, mask);
1023 * convert_extent_bit - convert all bits in a given range from one bit to
1025 * @tree: the io tree to search
1026 * @start: the start offset in bytes
1027 * @end: the end offset in bytes (inclusive)
1028 * @bits: the bits to set in this range
1029 * @clear_bits: the bits to clear in this range
1030 * @cached_state: state that we're going to cache
1031 * @mask: the allocation mask
1033 * This will go through and set bits for the given range. If any states exist
1034 * already in this range they are set with the given bit and cleared of the
1035 * clear_bits. This is only meant to be used by things that are mergeable, ie
1036 * converting from say DELALLOC to DIRTY. This is not meant to be used with
1037 * boundary bits like LOCK.
1039 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1040 unsigned long bits, unsigned long clear_bits,
1041 struct extent_state **cached_state, gfp_t mask)
1043 struct extent_state *state;
1044 struct extent_state *prealloc = NULL;
1045 struct rb_node *node;
1047 struct rb_node *parent;
1052 btrfs_debug_check_extent_io_range(tree, start, end);
1055 if (!prealloc && (mask & __GFP_WAIT)) {
1056 prealloc = alloc_extent_state(mask);
1061 spin_lock(&tree->lock);
1062 if (cached_state && *cached_state) {
1063 state = *cached_state;
1064 if (state->start <= start && state->end > start &&
1066 node = &state->rb_node;
1072 * this search will find all the extents that end after
1075 node = tree_search_for_insert(tree, start, &p, &parent);
1077 prealloc = alloc_extent_state_atomic(prealloc);
1082 err = insert_state(tree, prealloc, start, end,
1083 &p, &parent, &bits);
1085 extent_io_tree_panic(tree, err);
1086 cache_state(prealloc, cached_state);
1090 state = rb_entry(node, struct extent_state, rb_node);
1092 last_start = state->start;
1093 last_end = state->end;
1096 * | ---- desired range ---- |
1099 * Just lock what we found and keep going
1101 if (state->start == start && state->end <= end) {
1102 set_state_bits(tree, state, &bits);
1103 cache_state(state, cached_state);
1104 state = clear_state_bit(tree, state, &clear_bits, 0);
1105 if (last_end == (u64)-1)
1107 start = last_end + 1;
1108 if (start < end && state && state->start == start &&
1115 * | ---- desired range ---- |
1118 * | ------------- state -------------- |
1120 * We need to split the extent we found, and may flip bits on
1123 * If the extent we found extends past our
1124 * range, we just split and search again. It'll get split
1125 * again the next time though.
1127 * If the extent we found is inside our range, we set the
1128 * desired bit on it.
1130 if (state->start < start) {
1131 prealloc = alloc_extent_state_atomic(prealloc);
1136 err = split_state(tree, state, prealloc, start);
1138 extent_io_tree_panic(tree, err);
1142 if (state->end <= end) {
1143 set_state_bits(tree, state, &bits);
1144 cache_state(state, cached_state);
1145 state = clear_state_bit(tree, state, &clear_bits, 0);
1146 if (last_end == (u64)-1)
1148 start = last_end + 1;
1149 if (start < end && state && state->start == start &&
1156 * | ---- desired range ---- |
1157 * | state | or | state |
1159 * There's a hole, we need to insert something in it and
1160 * ignore the extent we found.
1162 if (state->start > start) {
1164 if (end < last_start)
1167 this_end = last_start - 1;
1169 prealloc = alloc_extent_state_atomic(prealloc);
1176 * Avoid to free 'prealloc' if it can be merged with
1179 err = insert_state(tree, prealloc, start, this_end,
1182 extent_io_tree_panic(tree, err);
1183 cache_state(prealloc, cached_state);
1185 start = this_end + 1;
1189 * | ---- desired range ---- |
1191 * We need to split the extent, and set the bit
1194 if (state->start <= end && state->end > end) {
1195 prealloc = alloc_extent_state_atomic(prealloc);
1201 err = split_state(tree, state, prealloc, end + 1);
1203 extent_io_tree_panic(tree, err);
1205 set_state_bits(tree, prealloc, &bits);
1206 cache_state(prealloc, cached_state);
1207 clear_state_bit(tree, prealloc, &clear_bits, 0);
1215 spin_unlock(&tree->lock);
1217 free_extent_state(prealloc);
1224 spin_unlock(&tree->lock);
1225 if (mask & __GFP_WAIT)
1230 /* wrappers around set/clear extent bit */
1231 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1234 return set_extent_bit(tree, start, end, EXTENT_DIRTY, NULL,
1238 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1239 unsigned long bits, gfp_t mask)
1241 return set_extent_bit(tree, start, end, bits, NULL,
1245 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1246 unsigned long bits, gfp_t mask)
1248 return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask);
1251 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
1252 struct extent_state **cached_state, gfp_t mask)
1254 return set_extent_bit(tree, start, end,
1255 EXTENT_DELALLOC | EXTENT_UPTODATE,
1256 NULL, cached_state, mask);
1259 int set_extent_defrag(struct extent_io_tree *tree, u64 start, u64 end,
1260 struct extent_state **cached_state, gfp_t mask)
1262 return set_extent_bit(tree, start, end,
1263 EXTENT_DELALLOC | EXTENT_UPTODATE | EXTENT_DEFRAG,
1264 NULL, cached_state, mask);
1267 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1270 return clear_extent_bit(tree, start, end,
1271 EXTENT_DIRTY | EXTENT_DELALLOC |
1272 EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask);
1275 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
1278 return set_extent_bit(tree, start, end, EXTENT_NEW, NULL,
1282 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
1283 struct extent_state **cached_state, gfp_t mask)
1285 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, NULL,
1286 cached_state, mask);
1289 int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
1290 struct extent_state **cached_state, gfp_t mask)
1292 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0,
1293 cached_state, mask);
1297 * either insert or lock state struct between start and end use mask to tell
1298 * us if waiting is desired.
1300 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1301 unsigned long bits, struct extent_state **cached_state)
1306 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED | bits,
1307 EXTENT_LOCKED, &failed_start,
1308 cached_state, GFP_NOFS);
1309 if (err == -EEXIST) {
1310 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1311 start = failed_start;
1314 WARN_ON(start > end);
1319 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1321 return lock_extent_bits(tree, start, end, 0, NULL);
1324 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1329 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1330 &failed_start, NULL, GFP_NOFS);
1331 if (err == -EEXIST) {
1332 if (failed_start > start)
1333 clear_extent_bit(tree, start, failed_start - 1,
1334 EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS);
1340 int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end,
1341 struct extent_state **cached, gfp_t mask)
1343 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached,
1347 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1349 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL,
1353 int extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
1355 unsigned long index = start >> PAGE_CACHE_SHIFT;
1356 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1359 while (index <= end_index) {
1360 page = find_get_page(inode->i_mapping, index);
1361 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1362 clear_page_dirty_for_io(page);
1363 page_cache_release(page);
1369 int extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
1371 unsigned long index = start >> PAGE_CACHE_SHIFT;
1372 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1375 while (index <= end_index) {
1376 page = find_get_page(inode->i_mapping, index);
1377 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1378 account_page_redirty(page);
1379 __set_page_dirty_nobuffers(page);
1380 page_cache_release(page);
1387 * helper function to set both pages and extents in the tree writeback
1389 static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
1391 unsigned long index = start >> PAGE_CACHE_SHIFT;
1392 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1395 while (index <= end_index) {
1396 page = find_get_page(tree->mapping, index);
1397 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1398 set_page_writeback(page);
1399 page_cache_release(page);
1405 /* find the first state struct with 'bits' set after 'start', and
1406 * return it. tree->lock must be held. NULL will returned if
1407 * nothing was found after 'start'
1409 static struct extent_state *
1410 find_first_extent_bit_state(struct extent_io_tree *tree,
1411 u64 start, unsigned long bits)
1413 struct rb_node *node;
1414 struct extent_state *state;
1417 * this search will find all the extents that end after
1420 node = tree_search(tree, start);
1425 state = rb_entry(node, struct extent_state, rb_node);
1426 if (state->end >= start && (state->state & bits))
1429 node = rb_next(node);
1438 * find the first offset in the io tree with 'bits' set. zero is
1439 * returned if we find something, and *start_ret and *end_ret are
1440 * set to reflect the state struct that was found.
1442 * If nothing was found, 1 is returned. If found something, return 0.
1444 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1445 u64 *start_ret, u64 *end_ret, unsigned long bits,
1446 struct extent_state **cached_state)
1448 struct extent_state *state;
1452 spin_lock(&tree->lock);
1453 if (cached_state && *cached_state) {
1454 state = *cached_state;
1455 if (state->end == start - 1 && state->tree) {
1456 n = rb_next(&state->rb_node);
1458 state = rb_entry(n, struct extent_state,
1460 if (state->state & bits)
1464 free_extent_state(*cached_state);
1465 *cached_state = NULL;
1468 free_extent_state(*cached_state);
1469 *cached_state = NULL;
1472 state = find_first_extent_bit_state(tree, start, bits);
1475 cache_state(state, cached_state);
1476 *start_ret = state->start;
1477 *end_ret = state->end;
1481 spin_unlock(&tree->lock);
1486 * find a contiguous range of bytes in the file marked as delalloc, not
1487 * more than 'max_bytes'. start and end are used to return the range,
1489 * 1 is returned if we find something, 0 if nothing was in the tree
1491 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1492 u64 *start, u64 *end, u64 max_bytes,
1493 struct extent_state **cached_state)
1495 struct rb_node *node;
1496 struct extent_state *state;
1497 u64 cur_start = *start;
1499 u64 total_bytes = 0;
1501 spin_lock(&tree->lock);
1504 * this search will find all the extents that end after
1507 node = tree_search(tree, cur_start);
1515 state = rb_entry(node, struct extent_state, rb_node);
1516 if (found && (state->start != cur_start ||
1517 (state->state & EXTENT_BOUNDARY))) {
1520 if (!(state->state & EXTENT_DELALLOC)) {
1526 *start = state->start;
1527 *cached_state = state;
1528 atomic_inc(&state->refs);
1532 cur_start = state->end + 1;
1533 node = rb_next(node);
1534 total_bytes += state->end - state->start + 1;
1535 if (total_bytes >= max_bytes)
1541 spin_unlock(&tree->lock);
1545 static noinline void __unlock_for_delalloc(struct inode *inode,
1546 struct page *locked_page,
1550 struct page *pages[16];
1551 unsigned long index = start >> PAGE_CACHE_SHIFT;
1552 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1553 unsigned long nr_pages = end_index - index + 1;
1556 if (index == locked_page->index && end_index == index)
1559 while (nr_pages > 0) {
1560 ret = find_get_pages_contig(inode->i_mapping, index,
1561 min_t(unsigned long, nr_pages,
1562 ARRAY_SIZE(pages)), pages);
1563 for (i = 0; i < ret; i++) {
1564 if (pages[i] != locked_page)
1565 unlock_page(pages[i]);
1566 page_cache_release(pages[i]);
1574 static noinline int lock_delalloc_pages(struct inode *inode,
1575 struct page *locked_page,
1579 unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT;
1580 unsigned long start_index = index;
1581 unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT;
1582 unsigned long pages_locked = 0;
1583 struct page *pages[16];
1584 unsigned long nrpages;
1588 /* the caller is responsible for locking the start index */
1589 if (index == locked_page->index && index == end_index)
1592 /* skip the page at the start index */
1593 nrpages = end_index - index + 1;
1594 while (nrpages > 0) {
1595 ret = find_get_pages_contig(inode->i_mapping, index,
1596 min_t(unsigned long,
1597 nrpages, ARRAY_SIZE(pages)), pages);
1602 /* now we have an array of pages, lock them all */
1603 for (i = 0; i < ret; i++) {
1605 * the caller is taking responsibility for
1608 if (pages[i] != locked_page) {
1609 lock_page(pages[i]);
1610 if (!PageDirty(pages[i]) ||
1611 pages[i]->mapping != inode->i_mapping) {
1613 unlock_page(pages[i]);
1614 page_cache_release(pages[i]);
1618 page_cache_release(pages[i]);
1627 if (ret && pages_locked) {
1628 __unlock_for_delalloc(inode, locked_page,
1630 ((u64)(start_index + pages_locked - 1)) <<
1637 * find a contiguous range of bytes in the file marked as delalloc, not
1638 * more than 'max_bytes'. start and end are used to return the range,
1640 * 1 is returned if we find something, 0 if nothing was in the tree
1642 STATIC u64 find_lock_delalloc_range(struct inode *inode,
1643 struct extent_io_tree *tree,
1644 struct page *locked_page, u64 *start,
1645 u64 *end, u64 max_bytes)
1650 struct extent_state *cached_state = NULL;
1655 /* step one, find a bunch of delalloc bytes starting at start */
1656 delalloc_start = *start;
1658 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1659 max_bytes, &cached_state);
1660 if (!found || delalloc_end <= *start) {
1661 *start = delalloc_start;
1662 *end = delalloc_end;
1663 free_extent_state(cached_state);
1668 * start comes from the offset of locked_page. We have to lock
1669 * pages in order, so we can't process delalloc bytes before
1672 if (delalloc_start < *start)
1673 delalloc_start = *start;
1676 * make sure to limit the number of pages we try to lock down
1678 if (delalloc_end + 1 - delalloc_start > max_bytes)
1679 delalloc_end = delalloc_start + max_bytes - 1;
1681 /* step two, lock all the pages after the page that has start */
1682 ret = lock_delalloc_pages(inode, locked_page,
1683 delalloc_start, delalloc_end);
1684 if (ret == -EAGAIN) {
1685 /* some of the pages are gone, lets avoid looping by
1686 * shortening the size of the delalloc range we're searching
1688 free_extent_state(cached_state);
1690 max_bytes = PAGE_CACHE_SIZE;
1698 BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */
1700 /* step three, lock the state bits for the whole range */
1701 lock_extent_bits(tree, delalloc_start, delalloc_end, 0, &cached_state);
1703 /* then test to make sure it is all still delalloc */
1704 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1705 EXTENT_DELALLOC, 1, cached_state);
1707 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1708 &cached_state, GFP_NOFS);
1709 __unlock_for_delalloc(inode, locked_page,
1710 delalloc_start, delalloc_end);
1714 free_extent_state(cached_state);
1715 *start = delalloc_start;
1716 *end = delalloc_end;
1721 int extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end,
1722 struct page *locked_page,
1723 unsigned long clear_bits,
1724 unsigned long page_ops)
1726 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
1728 struct page *pages[16];
1729 unsigned long index = start >> PAGE_CACHE_SHIFT;
1730 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1731 unsigned long nr_pages = end_index - index + 1;
1734 clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS);
1738 while (nr_pages > 0) {
1739 ret = find_get_pages_contig(inode->i_mapping, index,
1740 min_t(unsigned long,
1741 nr_pages, ARRAY_SIZE(pages)), pages);
1742 for (i = 0; i < ret; i++) {
1744 if (page_ops & PAGE_SET_PRIVATE2)
1745 SetPagePrivate2(pages[i]);
1747 if (pages[i] == locked_page) {
1748 page_cache_release(pages[i]);
1751 if (page_ops & PAGE_CLEAR_DIRTY)
1752 clear_page_dirty_for_io(pages[i]);
1753 if (page_ops & PAGE_SET_WRITEBACK)
1754 set_page_writeback(pages[i]);
1755 if (page_ops & PAGE_END_WRITEBACK)
1756 end_page_writeback(pages[i]);
1757 if (page_ops & PAGE_UNLOCK)
1758 unlock_page(pages[i]);
1759 page_cache_release(pages[i]);
1769 * count the number of bytes in the tree that have a given bit(s)
1770 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1771 * cached. The total number found is returned.
1773 u64 count_range_bits(struct extent_io_tree *tree,
1774 u64 *start, u64 search_end, u64 max_bytes,
1775 unsigned long bits, int contig)
1777 struct rb_node *node;
1778 struct extent_state *state;
1779 u64 cur_start = *start;
1780 u64 total_bytes = 0;
1784 if (WARN_ON(search_end <= cur_start))
1787 spin_lock(&tree->lock);
1788 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1789 total_bytes = tree->dirty_bytes;
1793 * this search will find all the extents that end after
1796 node = tree_search(tree, cur_start);
1801 state = rb_entry(node, struct extent_state, rb_node);
1802 if (state->start > search_end)
1804 if (contig && found && state->start > last + 1)
1806 if (state->end >= cur_start && (state->state & bits) == bits) {
1807 total_bytes += min(search_end, state->end) + 1 -
1808 max(cur_start, state->start);
1809 if (total_bytes >= max_bytes)
1812 *start = max(cur_start, state->start);
1816 } else if (contig && found) {
1819 node = rb_next(node);
1824 spin_unlock(&tree->lock);
1829 * set the private field for a given byte offset in the tree. If there isn't
1830 * an extent_state there already, this does nothing.
1832 static int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1834 struct rb_node *node;
1835 struct extent_state *state;
1838 spin_lock(&tree->lock);
1840 * this search will find all the extents that end after
1843 node = tree_search(tree, start);
1848 state = rb_entry(node, struct extent_state, rb_node);
1849 if (state->start != start) {
1853 state->private = private;
1855 spin_unlock(&tree->lock);
1859 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1861 struct rb_node *node;
1862 struct extent_state *state;
1865 spin_lock(&tree->lock);
1867 * this search will find all the extents that end after
1870 node = tree_search(tree, start);
1875 state = rb_entry(node, struct extent_state, rb_node);
1876 if (state->start != start) {
1880 *private = state->private;
1882 spin_unlock(&tree->lock);
1887 * searches a range in the state tree for a given mask.
1888 * If 'filled' == 1, this returns 1 only if every extent in the tree
1889 * has the bits set. Otherwise, 1 is returned if any bit in the
1890 * range is found set.
1892 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1893 unsigned long bits, int filled, struct extent_state *cached)
1895 struct extent_state *state = NULL;
1896 struct rb_node *node;
1899 spin_lock(&tree->lock);
1900 if (cached && cached->tree && cached->start <= start &&
1901 cached->end > start)
1902 node = &cached->rb_node;
1904 node = tree_search(tree, start);
1905 while (node && start <= end) {
1906 state = rb_entry(node, struct extent_state, rb_node);
1908 if (filled && state->start > start) {
1913 if (state->start > end)
1916 if (state->state & bits) {
1920 } else if (filled) {
1925 if (state->end == (u64)-1)
1928 start = state->end + 1;
1931 node = rb_next(node);
1938 spin_unlock(&tree->lock);
1943 * helper function to set a given page up to date if all the
1944 * extents in the tree for that page are up to date
1946 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
1948 u64 start = page_offset(page);
1949 u64 end = start + PAGE_CACHE_SIZE - 1;
1950 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1951 SetPageUptodate(page);
1955 * When IO fails, either with EIO or csum verification fails, we
1956 * try other mirrors that might have a good copy of the data. This
1957 * io_failure_record is used to record state as we go through all the
1958 * mirrors. If another mirror has good data, the page is set up to date
1959 * and things continue. If a good mirror can't be found, the original
1960 * bio end_io callback is called to indicate things have failed.
1962 struct io_failure_record {
1967 unsigned long bio_flags;
1973 static int free_io_failure(struct inode *inode, struct io_failure_record *rec,
1978 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1980 set_state_private(failure_tree, rec->start, 0);
1981 ret = clear_extent_bits(failure_tree, rec->start,
1982 rec->start + rec->len - 1,
1983 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1987 ret = clear_extent_bits(&BTRFS_I(inode)->io_tree, rec->start,
1988 rec->start + rec->len - 1,
1989 EXTENT_DAMAGED, GFP_NOFS);
1998 * this bypasses the standard btrfs submit functions deliberately, as
1999 * the standard behavior is to write all copies in a raid setup. here we only
2000 * want to write the one bad copy. so we do the mapping for ourselves and issue
2001 * submit_bio directly.
2002 * to avoid any synchronization issues, wait for the data after writing, which
2003 * actually prevents the read that triggered the error from finishing.
2004 * currently, there can be no more than two copies of every data bit. thus,
2005 * exactly one rewrite is required.
2007 int repair_io_failure(struct btrfs_fs_info *fs_info, u64 start,
2008 u64 length, u64 logical, struct page *page,
2012 struct btrfs_device *dev;
2015 struct btrfs_bio *bbio = NULL;
2016 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
2019 ASSERT(!(fs_info->sb->s_flags & MS_RDONLY));
2020 BUG_ON(!mirror_num);
2022 /* we can't repair anything in raid56 yet */
2023 if (btrfs_is_parity_mirror(map_tree, logical, length, mirror_num))
2026 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
2030 map_length = length;
2032 ret = btrfs_map_block(fs_info, WRITE, logical,
2033 &map_length, &bbio, mirror_num);
2038 BUG_ON(mirror_num != bbio->mirror_num);
2039 sector = bbio->stripes[mirror_num-1].physical >> 9;
2040 bio->bi_sector = sector;
2041 dev = bbio->stripes[mirror_num-1].dev;
2043 if (!dev || !dev->bdev || !dev->writeable) {
2047 bio->bi_bdev = dev->bdev;
2048 bio_add_page(bio, page, length, start - page_offset(page));
2050 if (btrfsic_submit_bio_wait(WRITE_SYNC, bio)) {
2051 /* try to remap that extent elsewhere? */
2053 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
2057 printk_ratelimited_in_rcu(KERN_INFO
2058 "BTRFS: read error corrected: ino %lu off %llu "
2059 "(dev %s sector %llu)\n", page->mapping->host->i_ino,
2060 start, rcu_str_deref(dev->name), sector);
2066 int repair_eb_io_failure(struct btrfs_root *root, struct extent_buffer *eb,
2069 u64 start = eb->start;
2070 unsigned long i, num_pages = num_extent_pages(eb->start, eb->len);
2073 if (root->fs_info->sb->s_flags & MS_RDONLY)
2076 for (i = 0; i < num_pages; i++) {
2077 struct page *p = extent_buffer_page(eb, i);
2078 ret = repair_io_failure(root->fs_info, start, PAGE_CACHE_SIZE,
2079 start, p, mirror_num);
2082 start += PAGE_CACHE_SIZE;
2089 * each time an IO finishes, we do a fast check in the IO failure tree
2090 * to see if we need to process or clean up an io_failure_record
2092 static int clean_io_failure(u64 start, struct page *page)
2095 u64 private_failure;
2096 struct io_failure_record *failrec;
2097 struct inode *inode = page->mapping->host;
2098 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2099 struct extent_state *state;
2105 ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
2106 (u64)-1, 1, EXTENT_DIRTY, 0);
2110 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, start,
2115 failrec = (struct io_failure_record *)(unsigned long) private_failure;
2116 BUG_ON(!failrec->this_mirror);
2118 if (failrec->in_validation) {
2119 /* there was no real error, just free the record */
2120 pr_debug("clean_io_failure: freeing dummy error at %llu\n",
2125 if (fs_info->sb->s_flags & MS_RDONLY)
2128 spin_lock(&BTRFS_I(inode)->io_tree.lock);
2129 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
2132 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
2134 if (state && state->start <= failrec->start &&
2135 state->end >= failrec->start + failrec->len - 1) {
2136 num_copies = btrfs_num_copies(fs_info, failrec->logical,
2138 if (num_copies > 1) {
2139 ret = repair_io_failure(fs_info, start, failrec->len,
2140 failrec->logical, page,
2141 failrec->failed_mirror);
2149 ret = free_io_failure(inode, failrec, did_repair);
2155 * this is a generic handler for readpage errors (default
2156 * readpage_io_failed_hook). if other copies exist, read those and write back
2157 * good data to the failed position. does not investigate in remapping the
2158 * failed extent elsewhere, hoping the device will be smart enough to do this as
2162 static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset,
2163 struct page *page, u64 start, u64 end,
2166 struct io_failure_record *failrec = NULL;
2168 struct extent_map *em;
2169 struct inode *inode = page->mapping->host;
2170 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2171 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2172 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2174 struct btrfs_io_bio *btrfs_failed_bio;
2175 struct btrfs_io_bio *btrfs_bio;
2181 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
2183 ret = get_state_private(failure_tree, start, &private);
2185 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2188 failrec->start = start;
2189 failrec->len = end - start + 1;
2190 failrec->this_mirror = 0;
2191 failrec->bio_flags = 0;
2192 failrec->in_validation = 0;
2194 read_lock(&em_tree->lock);
2195 em = lookup_extent_mapping(em_tree, start, failrec->len);
2197 read_unlock(&em_tree->lock);
2202 if (em->start > start || em->start + em->len <= start) {
2203 free_extent_map(em);
2206 read_unlock(&em_tree->lock);
2212 logical = start - em->start;
2213 logical = em->block_start + logical;
2214 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2215 logical = em->block_start;
2216 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2217 extent_set_compress_type(&failrec->bio_flags,
2220 pr_debug("bio_readpage_error: (new) logical=%llu, start=%llu, "
2221 "len=%llu\n", logical, start, failrec->len);
2222 failrec->logical = logical;
2223 free_extent_map(em);
2225 /* set the bits in the private failure tree */
2226 ret = set_extent_bits(failure_tree, start, end,
2227 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
2229 ret = set_state_private(failure_tree, start,
2230 (u64)(unsigned long)failrec);
2231 /* set the bits in the inode's tree */
2233 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED,
2240 failrec = (struct io_failure_record *)(unsigned long)private;
2241 pr_debug("bio_readpage_error: (found) logical=%llu, "
2242 "start=%llu, len=%llu, validation=%d\n",
2243 failrec->logical, failrec->start, failrec->len,
2244 failrec->in_validation);
2246 * when data can be on disk more than twice, add to failrec here
2247 * (e.g. with a list for failed_mirror) to make
2248 * clean_io_failure() clean all those errors at once.
2251 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
2252 failrec->logical, failrec->len);
2253 if (num_copies == 1) {
2255 * we only have a single copy of the data, so don't bother with
2256 * all the retry and error correction code that follows. no
2257 * matter what the error is, it is very likely to persist.
2259 pr_debug("bio_readpage_error: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
2260 num_copies, failrec->this_mirror, failed_mirror);
2261 free_io_failure(inode, failrec, 0);
2266 * there are two premises:
2267 * a) deliver good data to the caller
2268 * b) correct the bad sectors on disk
2270 if (failed_bio->bi_vcnt > 1) {
2272 * to fulfill b), we need to know the exact failing sectors, as
2273 * we don't want to rewrite any more than the failed ones. thus,
2274 * we need separate read requests for the failed bio
2276 * if the following BUG_ON triggers, our validation request got
2277 * merged. we need separate requests for our algorithm to work.
2279 BUG_ON(failrec->in_validation);
2280 failrec->in_validation = 1;
2281 failrec->this_mirror = failed_mirror;
2282 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
2285 * we're ready to fulfill a) and b) alongside. get a good copy
2286 * of the failed sector and if we succeed, we have setup
2287 * everything for repair_io_failure to do the rest for us.
2289 if (failrec->in_validation) {
2290 BUG_ON(failrec->this_mirror != failed_mirror);
2291 failrec->in_validation = 0;
2292 failrec->this_mirror = 0;
2294 failrec->failed_mirror = failed_mirror;
2295 failrec->this_mirror++;
2296 if (failrec->this_mirror == failed_mirror)
2297 failrec->this_mirror++;
2298 read_mode = READ_SYNC;
2301 if (failrec->this_mirror > num_copies) {
2302 pr_debug("bio_readpage_error: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
2303 num_copies, failrec->this_mirror, failed_mirror);
2304 free_io_failure(inode, failrec, 0);
2308 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
2310 free_io_failure(inode, failrec, 0);
2313 bio->bi_end_io = failed_bio->bi_end_io;
2314 bio->bi_sector = failrec->logical >> 9;
2315 bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
2318 btrfs_failed_bio = btrfs_io_bio(failed_bio);
2319 if (btrfs_failed_bio->csum) {
2320 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2321 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
2323 btrfs_bio = btrfs_io_bio(bio);
2324 btrfs_bio->csum = btrfs_bio->csum_inline;
2325 phy_offset >>= inode->i_sb->s_blocksize_bits;
2326 phy_offset *= csum_size;
2327 memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + phy_offset,
2331 bio_add_page(bio, page, failrec->len, start - page_offset(page));
2333 pr_debug("bio_readpage_error: submitting new read[%#x] to "
2334 "this_mirror=%d, num_copies=%d, in_validation=%d\n", read_mode,
2335 failrec->this_mirror, num_copies, failrec->in_validation);
2337 ret = tree->ops->submit_bio_hook(inode, read_mode, bio,
2338 failrec->this_mirror,
2339 failrec->bio_flags, 0);
2343 /* lots and lots of room for performance fixes in the end_bio funcs */
2345 int end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2347 int uptodate = (err == 0);
2348 struct extent_io_tree *tree;
2351 tree = &BTRFS_I(page->mapping->host)->io_tree;
2353 if (tree->ops && tree->ops->writepage_end_io_hook) {
2354 ret = tree->ops->writepage_end_io_hook(page, start,
2355 end, NULL, uptodate);
2361 ClearPageUptodate(page);
2368 * after a writepage IO is done, we need to:
2369 * clear the uptodate bits on error
2370 * clear the writeback bits in the extent tree for this IO
2371 * end_page_writeback if the page has no more pending IO
2373 * Scheduling is not allowed, so the extent state tree is expected
2374 * to have one and only one object corresponding to this IO.
2376 static void end_bio_extent_writepage(struct bio *bio, int err)
2378 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2383 struct page *page = bvec->bv_page;
2385 /* We always issue full-page reads, but if some block
2386 * in a page fails to read, blk_update_request() will
2387 * advance bv_offset and adjust bv_len to compensate.
2388 * Print a warning for nonzero offsets, and an error
2389 * if they don't add up to a full page. */
2390 if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE) {
2391 if (bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE)
2392 btrfs_err(BTRFS_I(page->mapping->host)->root->fs_info,
2393 "partial page write in btrfs with offset %u and length %u",
2394 bvec->bv_offset, bvec->bv_len);
2396 btrfs_info(BTRFS_I(page->mapping->host)->root->fs_info,
2397 "incomplete page write in btrfs with offset %u and "
2399 bvec->bv_offset, bvec->bv_len);
2402 start = page_offset(page);
2403 end = start + bvec->bv_offset + bvec->bv_len - 1;
2405 if (--bvec >= bio->bi_io_vec)
2406 prefetchw(&bvec->bv_page->flags);
2408 if (end_extent_writepage(page, err, start, end))
2411 end_page_writeback(page);
2412 } while (bvec >= bio->bi_io_vec);
2418 endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len,
2421 struct extent_state *cached = NULL;
2422 u64 end = start + len - 1;
2424 if (uptodate && tree->track_uptodate)
2425 set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC);
2426 unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC);
2430 * after a readpage IO is done, we need to:
2431 * clear the uptodate bits on error
2432 * set the uptodate bits if things worked
2433 * set the page up to date if all extents in the tree are uptodate
2434 * clear the lock bit in the extent tree
2435 * unlock the page if there are no other extents locked for it
2437 * Scheduling is not allowed, so the extent state tree is expected
2438 * to have one and only one object corresponding to this IO.
2440 static void end_bio_extent_readpage(struct bio *bio, int err)
2442 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
2443 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
2444 struct bio_vec *bvec = bio->bi_io_vec;
2445 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
2446 struct extent_io_tree *tree;
2451 u64 extent_start = 0;
2460 struct page *page = bvec->bv_page;
2461 struct inode *inode = page->mapping->host;
2463 pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, "
2464 "mirror=%lu\n", (u64)bio->bi_sector, err,
2465 io_bio->mirror_num);
2466 tree = &BTRFS_I(inode)->io_tree;
2468 /* We always issue full-page reads, but if some block
2469 * in a page fails to read, blk_update_request() will
2470 * advance bv_offset and adjust bv_len to compensate.
2471 * Print a warning for nonzero offsets, and an error
2472 * if they don't add up to a full page. */
2473 if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE) {
2474 if (bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE)
2475 btrfs_err(BTRFS_I(page->mapping->host)->root->fs_info,
2476 "partial page read in btrfs with offset %u and length %u",
2477 bvec->bv_offset, bvec->bv_len);
2479 btrfs_info(BTRFS_I(page->mapping->host)->root->fs_info,
2480 "incomplete page read in btrfs with offset %u and "
2482 bvec->bv_offset, bvec->bv_len);
2485 start = page_offset(page);
2486 end = start + bvec->bv_offset + bvec->bv_len - 1;
2489 if (++bvec <= bvec_end)
2490 prefetchw(&bvec->bv_page->flags);
2492 mirror = io_bio->mirror_num;
2493 if (likely(uptodate && tree->ops &&
2494 tree->ops->readpage_end_io_hook)) {
2495 ret = tree->ops->readpage_end_io_hook(io_bio, offset,
2501 clean_io_failure(start, page);
2504 if (likely(uptodate))
2507 if (tree->ops && tree->ops->readpage_io_failed_hook) {
2508 ret = tree->ops->readpage_io_failed_hook(page, mirror);
2510 test_bit(BIO_UPTODATE, &bio->bi_flags))
2514 * The generic bio_readpage_error handles errors the
2515 * following way: If possible, new read requests are
2516 * created and submitted and will end up in
2517 * end_bio_extent_readpage as well (if we're lucky, not
2518 * in the !uptodate case). In that case it returns 0 and
2519 * we just go on with the next page in our bio. If it
2520 * can't handle the error it will return -EIO and we
2521 * remain responsible for that page.
2523 ret = bio_readpage_error(bio, offset, page, start, end,
2527 test_bit(BIO_UPTODATE, &bio->bi_flags);
2534 if (likely(uptodate)) {
2535 loff_t i_size = i_size_read(inode);
2536 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
2539 /* Zero out the end if this page straddles i_size */
2540 offset = i_size & (PAGE_CACHE_SIZE-1);
2541 if (page->index == end_index && offset)
2542 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
2543 SetPageUptodate(page);
2545 ClearPageUptodate(page);
2551 if (unlikely(!uptodate)) {
2553 endio_readpage_release_extent(tree,
2559 endio_readpage_release_extent(tree, start,
2560 end - start + 1, 0);
2561 } else if (!extent_len) {
2562 extent_start = start;
2563 extent_len = end + 1 - start;
2564 } else if (extent_start + extent_len == start) {
2565 extent_len += end + 1 - start;
2567 endio_readpage_release_extent(tree, extent_start,
2568 extent_len, uptodate);
2569 extent_start = start;
2570 extent_len = end + 1 - start;
2572 } while (bvec <= bvec_end);
2575 endio_readpage_release_extent(tree, extent_start, extent_len,
2578 io_bio->end_io(io_bio, err);
2583 * this allocates from the btrfs_bioset. We're returning a bio right now
2584 * but you can call btrfs_io_bio for the appropriate container_of magic
2587 btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
2590 struct btrfs_io_bio *btrfs_bio;
2593 bio = bio_alloc_bioset(gfp_flags, nr_vecs, btrfs_bioset);
2595 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
2596 while (!bio && (nr_vecs /= 2)) {
2597 bio = bio_alloc_bioset(gfp_flags,
2598 nr_vecs, btrfs_bioset);
2604 bio->bi_bdev = bdev;
2605 bio->bi_sector = first_sector;
2606 btrfs_bio = btrfs_io_bio(bio);
2607 btrfs_bio->csum = NULL;
2608 btrfs_bio->csum_allocated = NULL;
2609 btrfs_bio->end_io = NULL;
2614 struct bio *btrfs_bio_clone(struct bio *bio, gfp_t gfp_mask)
2616 return bio_clone_bioset(bio, gfp_mask, btrfs_bioset);
2620 /* this also allocates from the btrfs_bioset */
2621 struct bio *btrfs_io_bio_alloc(gfp_t gfp_mask, unsigned int nr_iovecs)
2623 struct btrfs_io_bio *btrfs_bio;
2626 bio = bio_alloc_bioset(gfp_mask, nr_iovecs, btrfs_bioset);
2628 btrfs_bio = btrfs_io_bio(bio);
2629 btrfs_bio->csum = NULL;
2630 btrfs_bio->csum_allocated = NULL;
2631 btrfs_bio->end_io = NULL;
2637 static int __must_check submit_one_bio(int rw, struct bio *bio,
2638 int mirror_num, unsigned long bio_flags)
2641 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2642 struct page *page = bvec->bv_page;
2643 struct extent_io_tree *tree = bio->bi_private;
2646 start = page_offset(page) + bvec->bv_offset;
2648 bio->bi_private = NULL;
2652 if (tree->ops && tree->ops->submit_bio_hook)
2653 ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
2654 mirror_num, bio_flags, start);
2656 btrfsic_submit_bio(rw, bio);
2658 if (bio_flagged(bio, BIO_EOPNOTSUPP))
2664 static int merge_bio(int rw, struct extent_io_tree *tree, struct page *page,
2665 unsigned long offset, size_t size, struct bio *bio,
2666 unsigned long bio_flags)
2669 if (tree->ops && tree->ops->merge_bio_hook)
2670 ret = tree->ops->merge_bio_hook(rw, page, offset, size, bio,
2677 static int submit_extent_page(int rw, struct extent_io_tree *tree,
2678 struct page *page, sector_t sector,
2679 size_t size, unsigned long offset,
2680 struct block_device *bdev,
2681 struct bio **bio_ret,
2682 unsigned long max_pages,
2683 bio_end_io_t end_io_func,
2685 unsigned long prev_bio_flags,
2686 unsigned long bio_flags)
2692 int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED;
2693 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
2694 size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE);
2696 if (bio_ret && *bio_ret) {
2699 contig = bio->bi_sector == sector;
2701 contig = bio_end_sector(bio) == sector;
2703 if (prev_bio_flags != bio_flags || !contig ||
2704 merge_bio(rw, tree, page, offset, page_size, bio, bio_flags) ||
2705 bio_add_page(bio, page, page_size, offset) < page_size) {
2706 ret = submit_one_bio(rw, bio, mirror_num,
2715 if (this_compressed)
2718 nr = bio_get_nr_vecs(bdev);
2720 bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
2724 bio_add_page(bio, page, page_size, offset);
2725 bio->bi_end_io = end_io_func;
2726 bio->bi_private = tree;
2731 ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
2736 static void attach_extent_buffer_page(struct extent_buffer *eb,
2739 if (!PagePrivate(page)) {
2740 SetPagePrivate(page);
2741 page_cache_get(page);
2742 set_page_private(page, (unsigned long)eb);
2744 WARN_ON(page->private != (unsigned long)eb);
2748 void set_page_extent_mapped(struct page *page)
2750 if (!PagePrivate(page)) {
2751 SetPagePrivate(page);
2752 page_cache_get(page);
2753 set_page_private(page, EXTENT_PAGE_PRIVATE);
2757 static struct extent_map *
2758 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
2759 u64 start, u64 len, get_extent_t *get_extent,
2760 struct extent_map **em_cached)
2762 struct extent_map *em;
2764 if (em_cached && *em_cached) {
2766 if (em->in_tree && start >= em->start &&
2767 start < extent_map_end(em)) {
2768 atomic_inc(&em->refs);
2772 free_extent_map(em);
2776 em = get_extent(inode, page, pg_offset, start, len, 0);
2777 if (em_cached && !IS_ERR_OR_NULL(em)) {
2779 atomic_inc(&em->refs);
2785 * basic readpage implementation. Locked extent state structs are inserted
2786 * into the tree that are removed when the IO is done (by the end_io
2788 * XXX JDM: This needs looking at to ensure proper page locking
2790 static int __do_readpage(struct extent_io_tree *tree,
2792 get_extent_t *get_extent,
2793 struct extent_map **em_cached,
2794 struct bio **bio, int mirror_num,
2795 unsigned long *bio_flags, int rw)
2797 struct inode *inode = page->mapping->host;
2798 u64 start = page_offset(page);
2799 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2803 u64 last_byte = i_size_read(inode);
2807 struct extent_map *em;
2808 struct block_device *bdev;
2811 int parent_locked = *bio_flags & EXTENT_BIO_PARENT_LOCKED;
2812 size_t pg_offset = 0;
2814 size_t disk_io_size;
2815 size_t blocksize = inode->i_sb->s_blocksize;
2816 unsigned long this_bio_flag = *bio_flags & EXTENT_BIO_PARENT_LOCKED;
2818 set_page_extent_mapped(page);
2821 if (!PageUptodate(page)) {
2822 if (cleancache_get_page(page) == 0) {
2823 BUG_ON(blocksize != PAGE_SIZE);
2824 unlock_extent(tree, start, end);
2829 if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
2831 size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
2834 iosize = PAGE_CACHE_SIZE - zero_offset;
2835 userpage = kmap_atomic(page);
2836 memset(userpage + zero_offset, 0, iosize);
2837 flush_dcache_page(page);
2838 kunmap_atomic(userpage);
2841 while (cur <= end) {
2842 unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
2844 if (cur >= last_byte) {
2846 struct extent_state *cached = NULL;
2848 iosize = PAGE_CACHE_SIZE - pg_offset;
2849 userpage = kmap_atomic(page);
2850 memset(userpage + pg_offset, 0, iosize);
2851 flush_dcache_page(page);
2852 kunmap_atomic(userpage);
2853 set_extent_uptodate(tree, cur, cur + iosize - 1,
2856 unlock_extent_cached(tree, cur,
2861 em = __get_extent_map(inode, page, pg_offset, cur,
2862 end - cur + 1, get_extent, em_cached);
2863 if (IS_ERR_OR_NULL(em)) {
2866 unlock_extent(tree, cur, end);
2869 extent_offset = cur - em->start;
2870 BUG_ON(extent_map_end(em) <= cur);
2873 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2874 this_bio_flag |= EXTENT_BIO_COMPRESSED;
2875 extent_set_compress_type(&this_bio_flag,
2879 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2880 cur_end = min(extent_map_end(em) - 1, end);
2881 iosize = ALIGN(iosize, blocksize);
2882 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2883 disk_io_size = em->block_len;
2884 sector = em->block_start >> 9;
2886 sector = (em->block_start + extent_offset) >> 9;
2887 disk_io_size = iosize;
2890 block_start = em->block_start;
2891 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2892 block_start = EXTENT_MAP_HOLE;
2893 free_extent_map(em);
2896 /* we've found a hole, just zero and go on */
2897 if (block_start == EXTENT_MAP_HOLE) {
2899 struct extent_state *cached = NULL;
2901 userpage = kmap_atomic(page);
2902 memset(userpage + pg_offset, 0, iosize);
2903 flush_dcache_page(page);
2904 kunmap_atomic(userpage);
2906 set_extent_uptodate(tree, cur, cur + iosize - 1,
2908 unlock_extent_cached(tree, cur, cur + iosize - 1,
2911 pg_offset += iosize;
2914 /* the get_extent function already copied into the page */
2915 if (test_range_bit(tree, cur, cur_end,
2916 EXTENT_UPTODATE, 1, NULL)) {
2917 check_page_uptodate(tree, page);
2919 unlock_extent(tree, cur, cur + iosize - 1);
2921 pg_offset += iosize;
2924 /* we have an inline extent but it didn't get marked up
2925 * to date. Error out
2927 if (block_start == EXTENT_MAP_INLINE) {
2930 unlock_extent(tree, cur, cur + iosize - 1);
2932 pg_offset += iosize;
2937 ret = submit_extent_page(rw, tree, page,
2938 sector, disk_io_size, pg_offset,
2940 end_bio_extent_readpage, mirror_num,
2945 *bio_flags = this_bio_flag;
2949 unlock_extent(tree, cur, cur + iosize - 1);
2952 pg_offset += iosize;
2956 if (!PageError(page))
2957 SetPageUptodate(page);
2963 static inline void __do_contiguous_readpages(struct extent_io_tree *tree,
2964 struct page *pages[], int nr_pages,
2966 get_extent_t *get_extent,
2967 struct extent_map **em_cached,
2968 struct bio **bio, int mirror_num,
2969 unsigned long *bio_flags, int rw)
2971 struct inode *inode;
2972 struct btrfs_ordered_extent *ordered;
2975 inode = pages[0]->mapping->host;
2977 lock_extent(tree, start, end);
2978 ordered = btrfs_lookup_ordered_range(inode, start,
2982 unlock_extent(tree, start, end);
2983 btrfs_start_ordered_extent(inode, ordered, 1);
2984 btrfs_put_ordered_extent(ordered);
2987 for (index = 0; index < nr_pages; index++) {
2988 __do_readpage(tree, pages[index], get_extent, em_cached, bio,
2989 mirror_num, bio_flags, rw);
2990 page_cache_release(pages[index]);
2994 static void __extent_readpages(struct extent_io_tree *tree,
2995 struct page *pages[],
2996 int nr_pages, get_extent_t *get_extent,
2997 struct extent_map **em_cached,
2998 struct bio **bio, int mirror_num,
2999 unsigned long *bio_flags, int rw)
3005 int first_index = 0;
3007 for (index = 0; index < nr_pages; index++) {
3008 page_start = page_offset(pages[index]);
3011 end = start + PAGE_CACHE_SIZE - 1;
3012 first_index = index;
3013 } else if (end + 1 == page_start) {
3014 end += PAGE_CACHE_SIZE;
3016 __do_contiguous_readpages(tree, &pages[first_index],
3017 index - first_index, start,
3018 end, get_extent, em_cached,
3019 bio, mirror_num, bio_flags,
3022 end = start + PAGE_CACHE_SIZE - 1;
3023 first_index = index;
3028 __do_contiguous_readpages(tree, &pages[first_index],
3029 index - first_index, start,
3030 end, get_extent, em_cached, bio,
3031 mirror_num, bio_flags, rw);
3034 static int __extent_read_full_page(struct extent_io_tree *tree,
3036 get_extent_t *get_extent,
3037 struct bio **bio, int mirror_num,
3038 unsigned long *bio_flags, int rw)
3040 struct inode *inode = page->mapping->host;
3041 struct btrfs_ordered_extent *ordered;
3042 u64 start = page_offset(page);
3043 u64 end = start + PAGE_CACHE_SIZE - 1;
3047 lock_extent(tree, start, end);
3048 ordered = btrfs_lookup_ordered_extent(inode, start);
3051 unlock_extent(tree, start, end);
3052 btrfs_start_ordered_extent(inode, ordered, 1);
3053 btrfs_put_ordered_extent(ordered);
3056 ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num,
3061 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
3062 get_extent_t *get_extent, int mirror_num)
3064 struct bio *bio = NULL;
3065 unsigned long bio_flags = 0;
3068 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
3071 ret = submit_one_bio(READ, bio, mirror_num, bio_flags);
3075 int extent_read_full_page_nolock(struct extent_io_tree *tree, struct page *page,
3076 get_extent_t *get_extent, int mirror_num)
3078 struct bio *bio = NULL;
3079 unsigned long bio_flags = EXTENT_BIO_PARENT_LOCKED;
3082 ret = __do_readpage(tree, page, get_extent, NULL, &bio, mirror_num,
3085 ret = submit_one_bio(READ, bio, mirror_num, bio_flags);
3089 static noinline void update_nr_written(struct page *page,
3090 struct writeback_control *wbc,
3091 unsigned long nr_written)
3093 wbc->nr_to_write -= nr_written;
3094 if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
3095 wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
3096 page->mapping->writeback_index = page->index + nr_written;
3100 * the writepage semantics are similar to regular writepage. extent
3101 * records are inserted to lock ranges in the tree, and as dirty areas
3102 * are found, they are marked writeback. Then the lock bits are removed
3103 * and the end_io handler clears the writeback ranges
3105 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
3108 struct inode *inode = page->mapping->host;
3109 struct extent_page_data *epd = data;
3110 struct extent_io_tree *tree = epd->tree;
3111 u64 start = page_offset(page);
3113 u64 page_end = start + PAGE_CACHE_SIZE - 1;
3117 u64 last_byte = i_size_read(inode);
3121 struct extent_state *cached_state = NULL;
3122 struct extent_map *em;
3123 struct block_device *bdev;
3126 size_t pg_offset = 0;
3128 loff_t i_size = i_size_read(inode);
3129 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
3135 unsigned long nr_written = 0;
3136 bool fill_delalloc = true;
3138 if (wbc->sync_mode == WB_SYNC_ALL)
3139 write_flags = WRITE_SYNC;
3141 write_flags = WRITE;
3143 trace___extent_writepage(page, inode, wbc);
3145 WARN_ON(!PageLocked(page));
3147 ClearPageError(page);
3149 pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
3150 if (page->index > end_index ||
3151 (page->index == end_index && !pg_offset)) {
3152 page->mapping->a_ops->invalidatepage(page, 0, PAGE_CACHE_SIZE);
3157 if (page->index == end_index) {
3160 userpage = kmap_atomic(page);
3161 memset(userpage + pg_offset, 0,
3162 PAGE_CACHE_SIZE - pg_offset);
3163 kunmap_atomic(userpage);
3164 flush_dcache_page(page);
3168 set_page_extent_mapped(page);
3170 if (!tree->ops || !tree->ops->fill_delalloc)
3171 fill_delalloc = false;
3173 delalloc_start = start;
3176 if (!epd->extent_locked && fill_delalloc) {
3177 u64 delalloc_to_write = 0;
3179 * make sure the wbc mapping index is at least updated
3182 update_nr_written(page, wbc, 0);
3184 while (delalloc_end < page_end) {
3185 nr_delalloc = find_lock_delalloc_range(inode, tree,
3190 if (nr_delalloc == 0) {
3191 delalloc_start = delalloc_end + 1;
3194 ret = tree->ops->fill_delalloc(inode, page,
3199 /* File system has been set read-only */
3205 * delalloc_end is already one less than the total
3206 * length, so we don't subtract one from
3209 delalloc_to_write += (delalloc_end - delalloc_start +
3212 delalloc_start = delalloc_end + 1;
3214 if (wbc->nr_to_write < delalloc_to_write) {
3217 if (delalloc_to_write < thresh * 2)
3218 thresh = delalloc_to_write;
3219 wbc->nr_to_write = min_t(u64, delalloc_to_write,
3223 /* did the fill delalloc function already unlock and start
3229 * we've unlocked the page, so we can't update
3230 * the mapping's writeback index, just update
3233 wbc->nr_to_write -= nr_written;
3237 if (tree->ops && tree->ops->writepage_start_hook) {
3238 ret = tree->ops->writepage_start_hook(page, start,
3241 /* Fixup worker will requeue */
3243 wbc->pages_skipped++;
3245 redirty_page_for_writepage(wbc, page);
3246 update_nr_written(page, wbc, nr_written);
3254 * we don't want to touch the inode after unlocking the page,
3255 * so we update the mapping writeback index now
3257 update_nr_written(page, wbc, nr_written + 1);
3260 if (last_byte <= start) {
3261 if (tree->ops && tree->ops->writepage_end_io_hook)
3262 tree->ops->writepage_end_io_hook(page, start,
3267 blocksize = inode->i_sb->s_blocksize;
3269 while (cur <= end) {
3270 if (cur >= last_byte) {
3271 if (tree->ops && tree->ops->writepage_end_io_hook)
3272 tree->ops->writepage_end_io_hook(page, cur,
3276 em = epd->get_extent(inode, page, pg_offset, cur,
3278 if (IS_ERR_OR_NULL(em)) {
3283 extent_offset = cur - em->start;
3284 BUG_ON(extent_map_end(em) <= cur);
3286 iosize = min(extent_map_end(em) - cur, end - cur + 1);
3287 iosize = ALIGN(iosize, blocksize);
3288 sector = (em->block_start + extent_offset) >> 9;
3290 block_start = em->block_start;
3291 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3292 free_extent_map(em);
3296 * compressed and inline extents are written through other
3299 if (compressed || block_start == EXTENT_MAP_HOLE ||
3300 block_start == EXTENT_MAP_INLINE) {
3302 * end_io notification does not happen here for
3303 * compressed extents
3305 if (!compressed && tree->ops &&
3306 tree->ops->writepage_end_io_hook)
3307 tree->ops->writepage_end_io_hook(page, cur,
3310 else if (compressed) {
3311 /* we don't want to end_page_writeback on
3312 * a compressed extent. this happens
3319 pg_offset += iosize;
3322 /* leave this out until we have a page_mkwrite call */
3323 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
3324 EXTENT_DIRTY, 0, NULL)) {
3326 pg_offset += iosize;
3330 if (tree->ops && tree->ops->writepage_io_hook) {
3331 ret = tree->ops->writepage_io_hook(page, cur,
3339 unsigned long max_nr = end_index + 1;
3341 set_range_writeback(tree, cur, cur + iosize - 1);
3342 if (!PageWriteback(page)) {
3343 btrfs_err(BTRFS_I(inode)->root->fs_info,
3344 "page %lu not writeback, cur %llu end %llu",
3345 page->index, cur, end);
3348 ret = submit_extent_page(write_flags, tree, page,
3349 sector, iosize, pg_offset,
3350 bdev, &epd->bio, max_nr,
3351 end_bio_extent_writepage,
3357 pg_offset += iosize;
3362 /* make sure the mapping tag for page dirty gets cleared */
3363 set_page_writeback(page);
3364 end_page_writeback(page);
3370 /* drop our reference on any cached states */
3371 free_extent_state(cached_state);
3375 static int eb_wait(void *word)
3381 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3383 wait_on_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK, eb_wait,
3384 TASK_UNINTERRUPTIBLE);
3387 static int lock_extent_buffer_for_io(struct extent_buffer *eb,
3388 struct btrfs_fs_info *fs_info,
3389 struct extent_page_data *epd)
3391 unsigned long i, num_pages;
3395 if (!btrfs_try_tree_write_lock(eb)) {
3397 flush_write_bio(epd);
3398 btrfs_tree_lock(eb);
3401 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3402 btrfs_tree_unlock(eb);
3406 flush_write_bio(epd);
3410 wait_on_extent_buffer_writeback(eb);
3411 btrfs_tree_lock(eb);
3412 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3414 btrfs_tree_unlock(eb);
3419 * We need to do this to prevent races in people who check if the eb is
3420 * under IO since we can end up having no IO bits set for a short period
3423 spin_lock(&eb->refs_lock);
3424 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3425 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3426 spin_unlock(&eb->refs_lock);
3427 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3428 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
3430 fs_info->dirty_metadata_batch);
3433 spin_unlock(&eb->refs_lock);
3436 btrfs_tree_unlock(eb);
3441 num_pages = num_extent_pages(eb->start, eb->len);
3442 for (i = 0; i < num_pages; i++) {
3443 struct page *p = extent_buffer_page(eb, i);
3445 if (!trylock_page(p)) {
3447 flush_write_bio(epd);
3457 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3459 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3460 smp_mb__after_clear_bit();
3461 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3464 static void end_bio_extent_buffer_writepage(struct bio *bio, int err)
3466 int uptodate = err == 0;
3467 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
3468 struct extent_buffer *eb;
3472 struct page *page = bvec->bv_page;
3475 eb = (struct extent_buffer *)page->private;
3477 done = atomic_dec_and_test(&eb->io_pages);
3479 if (!uptodate || test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
3480 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3481 ClearPageUptodate(page);
3485 end_page_writeback(page);
3490 end_extent_buffer_writeback(eb);
3491 } while (bvec >= bio->bi_io_vec);
3497 static int write_one_eb(struct extent_buffer *eb,
3498 struct btrfs_fs_info *fs_info,
3499 struct writeback_control *wbc,
3500 struct extent_page_data *epd)
3502 struct block_device *bdev = fs_info->fs_devices->latest_bdev;
3503 struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
3504 u64 offset = eb->start;
3505 unsigned long i, num_pages;
3506 unsigned long bio_flags = 0;
3507 int rw = (epd->sync_io ? WRITE_SYNC : WRITE) | REQ_META;
3510 clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3511 num_pages = num_extent_pages(eb->start, eb->len);
3512 atomic_set(&eb->io_pages, num_pages);
3513 if (btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID)
3514 bio_flags = EXTENT_BIO_TREE_LOG;
3516 for (i = 0; i < num_pages; i++) {
3517 struct page *p = extent_buffer_page(eb, i);
3519 clear_page_dirty_for_io(p);
3520 set_page_writeback(p);
3521 ret = submit_extent_page(rw, tree, p, offset >> 9,
3522 PAGE_CACHE_SIZE, 0, bdev, &epd->bio,
3523 -1, end_bio_extent_buffer_writepage,
3524 0, epd->bio_flags, bio_flags);
3525 epd->bio_flags = bio_flags;
3527 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3529 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3530 end_extent_buffer_writeback(eb);
3534 offset += PAGE_CACHE_SIZE;
3535 update_nr_written(p, wbc, 1);
3539 if (unlikely(ret)) {
3540 for (; i < num_pages; i++) {
3541 struct page *p = extent_buffer_page(eb, i);
3549 int btree_write_cache_pages(struct address_space *mapping,
3550 struct writeback_control *wbc)
3552 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
3553 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
3554 struct extent_buffer *eb, *prev_eb = NULL;
3555 struct extent_page_data epd = {
3559 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3564 int nr_to_write_done = 0;
3565 struct pagevec pvec;
3568 pgoff_t end; /* Inclusive */
3572 pagevec_init(&pvec, 0);
3573 if (wbc->range_cyclic) {
3574 index = mapping->writeback_index; /* Start from prev offset */
3577 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3578 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3581 if (wbc->sync_mode == WB_SYNC_ALL)
3582 tag = PAGECACHE_TAG_TOWRITE;
3584 tag = PAGECACHE_TAG_DIRTY;
3586 if (wbc->sync_mode == WB_SYNC_ALL)
3587 tag_pages_for_writeback(mapping, index, end);
3588 while (!done && !nr_to_write_done && (index <= end) &&
3589 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3590 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3594 for (i = 0; i < nr_pages; i++) {
3595 struct page *page = pvec.pages[i];
3597 if (!PagePrivate(page))
3600 if (!wbc->range_cyclic && page->index > end) {
3605 spin_lock(&mapping->private_lock);
3606 if (!PagePrivate(page)) {
3607 spin_unlock(&mapping->private_lock);
3611 eb = (struct extent_buffer *)page->private;
3614 * Shouldn't happen and normally this would be a BUG_ON
3615 * but no sense in crashing the users box for something
3616 * we can survive anyway.
3619 spin_unlock(&mapping->private_lock);
3623 if (eb == prev_eb) {
3624 spin_unlock(&mapping->private_lock);
3628 ret = atomic_inc_not_zero(&eb->refs);
3629 spin_unlock(&mapping->private_lock);
3634 ret = lock_extent_buffer_for_io(eb, fs_info, &epd);
3636 free_extent_buffer(eb);
3640 ret = write_one_eb(eb, fs_info, wbc, &epd);
3643 free_extent_buffer(eb);
3646 free_extent_buffer(eb);
3649 * the filesystem may choose to bump up nr_to_write.
3650 * We have to make sure to honor the new nr_to_write
3653 nr_to_write_done = wbc->nr_to_write <= 0;
3655 pagevec_release(&pvec);
3658 if (!scanned && !done) {
3660 * We hit the last page and there is more work to be done: wrap
3661 * back to the start of the file
3667 flush_write_bio(&epd);
3672 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3673 * @mapping: address space structure to write
3674 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3675 * @writepage: function called for each page
3676 * @data: data passed to writepage function
3678 * If a page is already under I/O, write_cache_pages() skips it, even
3679 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3680 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3681 * and msync() need to guarantee that all the data which was dirty at the time
3682 * the call was made get new I/O started against them. If wbc->sync_mode is
3683 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3684 * existing IO to complete.
3686 static int extent_write_cache_pages(struct extent_io_tree *tree,
3687 struct address_space *mapping,
3688 struct writeback_control *wbc,
3689 writepage_t writepage, void *data,
3690 void (*flush_fn)(void *))
3692 struct inode *inode = mapping->host;
3695 int nr_to_write_done = 0;
3696 struct pagevec pvec;
3699 pgoff_t end; /* Inclusive */
3704 * We have to hold onto the inode so that ordered extents can do their
3705 * work when the IO finishes. The alternative to this is failing to add
3706 * an ordered extent if the igrab() fails there and that is a huge pain
3707 * to deal with, so instead just hold onto the inode throughout the
3708 * writepages operation. If it fails here we are freeing up the inode
3709 * anyway and we'd rather not waste our time writing out stuff that is
3710 * going to be truncated anyway.
3715 pagevec_init(&pvec, 0);
3716 if (wbc->range_cyclic) {
3717 index = mapping->writeback_index; /* Start from prev offset */
3720 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3721 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3724 if (wbc->sync_mode == WB_SYNC_ALL)
3725 tag = PAGECACHE_TAG_TOWRITE;
3727 tag = PAGECACHE_TAG_DIRTY;
3729 if (wbc->sync_mode == WB_SYNC_ALL)
3730 tag_pages_for_writeback(mapping, index, end);
3731 while (!done && !nr_to_write_done && (index <= end) &&
3732 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3733 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3737 for (i = 0; i < nr_pages; i++) {
3738 struct page *page = pvec.pages[i];
3741 * At this point we hold neither mapping->tree_lock nor
3742 * lock on the page itself: the page may be truncated or
3743 * invalidated (changing page->mapping to NULL), or even
3744 * swizzled back from swapper_space to tmpfs file
3747 if (!trylock_page(page)) {
3752 if (unlikely(page->mapping != mapping)) {
3757 if (!wbc->range_cyclic && page->index > end) {
3763 if (wbc->sync_mode != WB_SYNC_NONE) {
3764 if (PageWriteback(page))
3766 wait_on_page_writeback(page);
3769 if (PageWriteback(page) ||
3770 !clear_page_dirty_for_io(page)) {
3775 ret = (*writepage)(page, wbc, data);
3777 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
3785 * the filesystem may choose to bump up nr_to_write.
3786 * We have to make sure to honor the new nr_to_write
3789 nr_to_write_done = wbc->nr_to_write <= 0;
3791 pagevec_release(&pvec);
3794 if (!scanned && !done) {
3796 * We hit the last page and there is more work to be done: wrap
3797 * back to the start of the file
3803 btrfs_add_delayed_iput(inode);
3807 static void flush_epd_write_bio(struct extent_page_data *epd)
3816 ret = submit_one_bio(rw, epd->bio, 0, epd->bio_flags);
3817 BUG_ON(ret < 0); /* -ENOMEM */
3822 static noinline void flush_write_bio(void *data)
3824 struct extent_page_data *epd = data;
3825 flush_epd_write_bio(epd);
3828 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
3829 get_extent_t *get_extent,
3830 struct writeback_control *wbc)
3833 struct extent_page_data epd = {
3836 .get_extent = get_extent,
3838 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3842 ret = __extent_writepage(page, wbc, &epd);
3844 flush_epd_write_bio(&epd);
3848 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
3849 u64 start, u64 end, get_extent_t *get_extent,
3853 struct address_space *mapping = inode->i_mapping;
3855 unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
3858 struct extent_page_data epd = {
3861 .get_extent = get_extent,
3863 .sync_io = mode == WB_SYNC_ALL,
3866 struct writeback_control wbc_writepages = {
3868 .nr_to_write = nr_pages * 2,
3869 .range_start = start,
3870 .range_end = end + 1,
3873 while (start <= end) {
3874 page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
3875 if (clear_page_dirty_for_io(page))
3876 ret = __extent_writepage(page, &wbc_writepages, &epd);
3878 if (tree->ops && tree->ops->writepage_end_io_hook)
3879 tree->ops->writepage_end_io_hook(page, start,
3880 start + PAGE_CACHE_SIZE - 1,
3884 page_cache_release(page);
3885 start += PAGE_CACHE_SIZE;
3888 flush_epd_write_bio(&epd);
3892 int extent_writepages(struct extent_io_tree *tree,
3893 struct address_space *mapping,
3894 get_extent_t *get_extent,
3895 struct writeback_control *wbc)
3898 struct extent_page_data epd = {
3901 .get_extent = get_extent,
3903 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3907 ret = extent_write_cache_pages(tree, mapping, wbc,
3908 __extent_writepage, &epd,
3910 flush_epd_write_bio(&epd);
3914 int extent_readpages(struct extent_io_tree *tree,
3915 struct address_space *mapping,
3916 struct list_head *pages, unsigned nr_pages,
3917 get_extent_t get_extent)
3919 struct bio *bio = NULL;
3921 unsigned long bio_flags = 0;
3922 struct page *pagepool[16];
3924 struct extent_map *em_cached = NULL;
3927 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
3928 page = list_entry(pages->prev, struct page, lru);
3930 prefetchw(&page->flags);
3931 list_del(&page->lru);
3932 if (add_to_page_cache_lru(page, mapping,
3933 page->index, GFP_NOFS)) {
3934 page_cache_release(page);
3938 pagepool[nr++] = page;
3939 if (nr < ARRAY_SIZE(pagepool))
3941 __extent_readpages(tree, pagepool, nr, get_extent, &em_cached,
3942 &bio, 0, &bio_flags, READ);
3946 __extent_readpages(tree, pagepool, nr, get_extent, &em_cached,
3947 &bio, 0, &bio_flags, READ);
3950 free_extent_map(em_cached);
3952 BUG_ON(!list_empty(pages));
3954 return submit_one_bio(READ, bio, 0, bio_flags);
3959 * basic invalidatepage code, this waits on any locked or writeback
3960 * ranges corresponding to the page, and then deletes any extent state
3961 * records from the tree
3963 int extent_invalidatepage(struct extent_io_tree *tree,
3964 struct page *page, unsigned long offset)
3966 struct extent_state *cached_state = NULL;
3967 u64 start = page_offset(page);
3968 u64 end = start + PAGE_CACHE_SIZE - 1;
3969 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
3971 start += ALIGN(offset, blocksize);
3975 lock_extent_bits(tree, start, end, 0, &cached_state);
3976 wait_on_page_writeback(page);
3977 clear_extent_bit(tree, start, end,
3978 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
3979 EXTENT_DO_ACCOUNTING,
3980 1, 1, &cached_state, GFP_NOFS);
3985 * a helper for releasepage, this tests for areas of the page that
3986 * are locked or under IO and drops the related state bits if it is safe
3989 static int try_release_extent_state(struct extent_map_tree *map,
3990 struct extent_io_tree *tree,
3991 struct page *page, gfp_t mask)
3993 u64 start = page_offset(page);
3994 u64 end = start + PAGE_CACHE_SIZE - 1;
3997 if (test_range_bit(tree, start, end,
3998 EXTENT_IOBITS, 0, NULL))
4001 if ((mask & GFP_NOFS) == GFP_NOFS)
4004 * at this point we can safely clear everything except the
4005 * locked bit and the nodatasum bit
4007 ret = clear_extent_bit(tree, start, end,
4008 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
4011 /* if clear_extent_bit failed for enomem reasons,
4012 * we can't allow the release to continue.
4023 * a helper for releasepage. As long as there are no locked extents
4024 * in the range corresponding to the page, both state records and extent
4025 * map records are removed
4027 int try_release_extent_mapping(struct extent_map_tree *map,
4028 struct extent_io_tree *tree, struct page *page,
4031 struct extent_map *em;
4032 u64 start = page_offset(page);
4033 u64 end = start + PAGE_CACHE_SIZE - 1;
4035 if ((mask & __GFP_WAIT) &&
4036 page->mapping->host->i_size > 16 * 1024 * 1024) {
4038 while (start <= end) {
4039 len = end - start + 1;
4040 write_lock(&map->lock);
4041 em = lookup_extent_mapping(map, start, len);
4043 write_unlock(&map->lock);
4046 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
4047 em->start != start) {
4048 write_unlock(&map->lock);
4049 free_extent_map(em);
4052 if (!test_range_bit(tree, em->start,
4053 extent_map_end(em) - 1,
4054 EXTENT_LOCKED | EXTENT_WRITEBACK,
4056 remove_extent_mapping(map, em);
4057 /* once for the rb tree */
4058 free_extent_map(em);
4060 start = extent_map_end(em);
4061 write_unlock(&map->lock);
4064 free_extent_map(em);
4067 return try_release_extent_state(map, tree, page, mask);
4071 * helper function for fiemap, which doesn't want to see any holes.
4072 * This maps until we find something past 'last'
4074 static struct extent_map *get_extent_skip_holes(struct inode *inode,
4077 get_extent_t *get_extent)
4079 u64 sectorsize = BTRFS_I(inode)->root->sectorsize;
4080 struct extent_map *em;
4087 len = last - offset;
4090 len = ALIGN(len, sectorsize);
4091 em = get_extent(inode, NULL, 0, offset, len, 0);
4092 if (IS_ERR_OR_NULL(em))
4095 /* if this isn't a hole return it */
4096 if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) &&
4097 em->block_start != EXTENT_MAP_HOLE) {
4101 /* this is a hole, advance to the next extent */
4102 offset = extent_map_end(em);
4103 free_extent_map(em);
4110 static noinline int count_ext_ref(u64 inum, u64 offset, u64 root_id, void *ctx)
4112 unsigned long cnt = *((unsigned long *)ctx);
4115 *((unsigned long *)ctx) = cnt;
4117 /* Now we're sure that the extent is shared. */
4123 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4124 __u64 start, __u64 len, get_extent_t *get_extent)
4128 u64 max = start + len;
4132 u64 last_for_get_extent = 0;
4134 u64 isize = i_size_read(inode);
4135 struct btrfs_key found_key;
4136 struct extent_map *em = NULL;
4137 struct extent_state *cached_state = NULL;
4138 struct btrfs_path *path;
4147 path = btrfs_alloc_path();
4150 path->leave_spinning = 1;
4152 start = ALIGN(start, BTRFS_I(inode)->root->sectorsize);
4153 len = ALIGN(len, BTRFS_I(inode)->root->sectorsize);
4156 * lookup the last file extent. We're not using i_size here
4157 * because there might be preallocation past i_size
4159 ret = btrfs_lookup_file_extent(NULL, BTRFS_I(inode)->root,
4160 path, btrfs_ino(inode), -1, 0);
4162 btrfs_free_path(path);
4167 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
4168 found_type = btrfs_key_type(&found_key);
4170 /* No extents, but there might be delalloc bits */
4171 if (found_key.objectid != btrfs_ino(inode) ||
4172 found_type != BTRFS_EXTENT_DATA_KEY) {
4173 /* have to trust i_size as the end */
4175 last_for_get_extent = isize;
4178 * remember the start of the last extent. There are a
4179 * bunch of different factors that go into the length of the
4180 * extent, so its much less complex to remember where it started
4182 last = found_key.offset;
4183 last_for_get_extent = last + 1;
4185 btrfs_release_path(path);
4188 * we might have some extents allocated but more delalloc past those
4189 * extents. so, we trust isize unless the start of the last extent is
4194 last_for_get_extent = isize;
4197 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1, 0,
4200 em = get_extent_skip_holes(inode, start, last_for_get_extent,
4210 u64 offset_in_extent = 0;
4212 /* break if the extent we found is outside the range */
4213 if (em->start >= max || extent_map_end(em) < off)
4217 * get_extent may return an extent that starts before our
4218 * requested range. We have to make sure the ranges
4219 * we return to fiemap always move forward and don't
4220 * overlap, so adjust the offsets here
4222 em_start = max(em->start, off);
4225 * record the offset from the start of the extent
4226 * for adjusting the disk offset below. Only do this if the
4227 * extent isn't compressed since our in ram offset may be past
4228 * what we have actually allocated on disk.
4230 if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4231 offset_in_extent = em_start - em->start;
4232 em_end = extent_map_end(em);
4233 em_len = em_end - em_start;
4238 * bump off for our next call to get_extent
4240 off = extent_map_end(em);
4244 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
4246 flags |= FIEMAP_EXTENT_LAST;
4247 } else if (em->block_start == EXTENT_MAP_INLINE) {
4248 flags |= (FIEMAP_EXTENT_DATA_INLINE |
4249 FIEMAP_EXTENT_NOT_ALIGNED);
4250 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
4251 flags |= (FIEMAP_EXTENT_DELALLOC |
4252 FIEMAP_EXTENT_UNKNOWN);
4254 unsigned long ref_cnt = 0;
4256 disko = em->block_start + offset_in_extent;
4259 * As btrfs supports shared space, this information
4260 * can be exported to userspace tools via
4261 * flag FIEMAP_EXTENT_SHARED.
4263 ret = iterate_inodes_from_logical(
4265 BTRFS_I(inode)->root->fs_info,
4266 path, count_ext_ref, &ref_cnt);
4267 if (ret < 0 && ret != -ENOENT)
4271 flags |= FIEMAP_EXTENT_SHARED;
4273 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4274 flags |= FIEMAP_EXTENT_ENCODED;
4276 free_extent_map(em);
4278 if ((em_start >= last) || em_len == (u64)-1 ||
4279 (last == (u64)-1 && isize <= em_end)) {
4280 flags |= FIEMAP_EXTENT_LAST;
4284 /* now scan forward to see if this is really the last extent. */
4285 em = get_extent_skip_holes(inode, off, last_for_get_extent,
4292 flags |= FIEMAP_EXTENT_LAST;
4295 ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
4301 free_extent_map(em);
4303 btrfs_free_path(path);
4304 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4305 &cached_state, GFP_NOFS);
4309 static void __free_extent_buffer(struct extent_buffer *eb)
4311 btrfs_leak_debug_del(&eb->leak_list);
4312 kmem_cache_free(extent_buffer_cache, eb);
4315 static int extent_buffer_under_io(struct extent_buffer *eb)
4317 return (atomic_read(&eb->io_pages) ||
4318 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4319 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4323 * Helper for releasing extent buffer page.
4325 static void btrfs_release_extent_buffer_page(struct extent_buffer *eb,
4326 unsigned long start_idx)
4328 unsigned long index;
4329 unsigned long num_pages;
4331 int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4333 BUG_ON(extent_buffer_under_io(eb));
4335 num_pages = num_extent_pages(eb->start, eb->len);
4336 index = start_idx + num_pages;
4337 if (start_idx >= index)
4342 page = extent_buffer_page(eb, index);
4343 if (page && mapped) {
4344 spin_lock(&page->mapping->private_lock);
4346 * We do this since we'll remove the pages after we've
4347 * removed the eb from the radix tree, so we could race
4348 * and have this page now attached to the new eb. So
4349 * only clear page_private if it's still connected to
4352 if (PagePrivate(page) &&
4353 page->private == (unsigned long)eb) {
4354 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4355 BUG_ON(PageDirty(page));
4356 BUG_ON(PageWriteback(page));
4358 * We need to make sure we haven't be attached
4361 ClearPagePrivate(page);
4362 set_page_private(page, 0);
4363 /* One for the page private */
4364 page_cache_release(page);
4366 spin_unlock(&page->mapping->private_lock);
4370 /* One for when we alloced the page */
4371 page_cache_release(page);
4373 } while (index != start_idx);
4377 * Helper for releasing the extent buffer.
4379 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4381 btrfs_release_extent_buffer_page(eb, 0);
4382 __free_extent_buffer(eb);
4385 static struct extent_buffer *
4386 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
4387 unsigned long len, gfp_t mask)
4389 struct extent_buffer *eb = NULL;
4391 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
4396 eb->fs_info = fs_info;
4398 rwlock_init(&eb->lock);
4399 atomic_set(&eb->write_locks, 0);
4400 atomic_set(&eb->read_locks, 0);
4401 atomic_set(&eb->blocking_readers, 0);
4402 atomic_set(&eb->blocking_writers, 0);
4403 atomic_set(&eb->spinning_readers, 0);
4404 atomic_set(&eb->spinning_writers, 0);
4405 eb->lock_nested = 0;
4406 init_waitqueue_head(&eb->write_lock_wq);
4407 init_waitqueue_head(&eb->read_lock_wq);
4409 btrfs_leak_debug_add(&eb->leak_list, &buffers);
4411 spin_lock_init(&eb->refs_lock);
4412 atomic_set(&eb->refs, 1);
4413 atomic_set(&eb->io_pages, 0);
4416 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4418 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4419 > MAX_INLINE_EXTENT_BUFFER_SIZE);
4420 BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE);
4425 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src)
4429 struct extent_buffer *new;
4430 unsigned long num_pages = num_extent_pages(src->start, src->len);
4432 new = __alloc_extent_buffer(NULL, src->start, src->len, GFP_NOFS);
4436 for (i = 0; i < num_pages; i++) {
4437 p = alloc_page(GFP_NOFS);
4439 btrfs_release_extent_buffer(new);
4442 attach_extent_buffer_page(new, p);
4443 WARN_ON(PageDirty(p));
4448 copy_extent_buffer(new, src, 0, 0, src->len);
4449 set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
4450 set_bit(EXTENT_BUFFER_DUMMY, &new->bflags);
4455 struct extent_buffer *alloc_dummy_extent_buffer(u64 start, unsigned long len)
4457 struct extent_buffer *eb;
4458 unsigned long num_pages = num_extent_pages(0, len);
4461 eb = __alloc_extent_buffer(NULL, start, len, GFP_NOFS);
4465 for (i = 0; i < num_pages; i++) {
4466 eb->pages[i] = alloc_page(GFP_NOFS);
4470 set_extent_buffer_uptodate(eb);
4471 btrfs_set_header_nritems(eb, 0);
4472 set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4477 __free_page(eb->pages[i - 1]);
4478 __free_extent_buffer(eb);
4482 static void check_buffer_tree_ref(struct extent_buffer *eb)
4485 /* the ref bit is tricky. We have to make sure it is set
4486 * if we have the buffer dirty. Otherwise the
4487 * code to free a buffer can end up dropping a dirty
4490 * Once the ref bit is set, it won't go away while the
4491 * buffer is dirty or in writeback, and it also won't
4492 * go away while we have the reference count on the
4495 * We can't just set the ref bit without bumping the
4496 * ref on the eb because free_extent_buffer might
4497 * see the ref bit and try to clear it. If this happens
4498 * free_extent_buffer might end up dropping our original
4499 * ref by mistake and freeing the page before we are able
4500 * to add one more ref.
4502 * So bump the ref count first, then set the bit. If someone
4503 * beat us to it, drop the ref we added.
4505 refs = atomic_read(&eb->refs);
4506 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4509 spin_lock(&eb->refs_lock);
4510 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4511 atomic_inc(&eb->refs);
4512 spin_unlock(&eb->refs_lock);
4515 static void mark_extent_buffer_accessed(struct extent_buffer *eb)
4517 unsigned long num_pages, i;
4519 check_buffer_tree_ref(eb);
4521 num_pages = num_extent_pages(eb->start, eb->len);
4522 for (i = 0; i < num_pages; i++) {
4523 struct page *p = extent_buffer_page(eb, i);
4524 mark_page_accessed(p);
4528 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
4531 struct extent_buffer *eb;
4534 eb = radix_tree_lookup(&fs_info->buffer_radix,
4535 start >> PAGE_CACHE_SHIFT);
4536 if (eb && atomic_inc_not_zero(&eb->refs)) {
4538 mark_extent_buffer_accessed(eb);
4546 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
4547 u64 start, unsigned long len)
4549 unsigned long num_pages = num_extent_pages(start, len);
4551 unsigned long index = start >> PAGE_CACHE_SHIFT;
4552 struct extent_buffer *eb;
4553 struct extent_buffer *exists = NULL;
4555 struct address_space *mapping = fs_info->btree_inode->i_mapping;
4559 eb = find_extent_buffer(fs_info, start);
4563 eb = __alloc_extent_buffer(fs_info, start, len, GFP_NOFS);
4567 for (i = 0; i < num_pages; i++, index++) {
4568 p = find_or_create_page(mapping, index, GFP_NOFS);
4572 spin_lock(&mapping->private_lock);
4573 if (PagePrivate(p)) {
4575 * We could have already allocated an eb for this page
4576 * and attached one so lets see if we can get a ref on
4577 * the existing eb, and if we can we know it's good and
4578 * we can just return that one, else we know we can just
4579 * overwrite page->private.
4581 exists = (struct extent_buffer *)p->private;
4582 if (atomic_inc_not_zero(&exists->refs)) {
4583 spin_unlock(&mapping->private_lock);
4585 page_cache_release(p);
4586 mark_extent_buffer_accessed(exists);
4591 * Do this so attach doesn't complain and we need to
4592 * drop the ref the old guy had.
4594 ClearPagePrivate(p);
4595 WARN_ON(PageDirty(p));
4596 page_cache_release(p);
4598 attach_extent_buffer_page(eb, p);
4599 spin_unlock(&mapping->private_lock);
4600 WARN_ON(PageDirty(p));
4601 mark_page_accessed(p);
4603 if (!PageUptodate(p))
4607 * see below about how we avoid a nasty race with release page
4608 * and why we unlock later
4612 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4614 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
4618 spin_lock(&fs_info->buffer_lock);
4619 ret = radix_tree_insert(&fs_info->buffer_radix,
4620 start >> PAGE_CACHE_SHIFT, eb);
4621 spin_unlock(&fs_info->buffer_lock);
4622 radix_tree_preload_end();
4623 if (ret == -EEXIST) {
4624 exists = find_extent_buffer(fs_info, start);
4630 /* add one reference for the tree */
4631 check_buffer_tree_ref(eb);
4632 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
4635 * there is a race where release page may have
4636 * tried to find this extent buffer in the radix
4637 * but failed. It will tell the VM it is safe to
4638 * reclaim the, and it will clear the page private bit.
4639 * We must make sure to set the page private bit properly
4640 * after the extent buffer is in the radix tree so
4641 * it doesn't get lost
4643 SetPageChecked(eb->pages[0]);
4644 for (i = 1; i < num_pages; i++) {
4645 p = extent_buffer_page(eb, i);
4646 ClearPageChecked(p);
4649 unlock_page(eb->pages[0]);
4653 for (i = 0; i < num_pages; i++) {
4655 unlock_page(eb->pages[i]);
4658 WARN_ON(!atomic_dec_and_test(&eb->refs));
4659 btrfs_release_extent_buffer(eb);
4663 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
4665 struct extent_buffer *eb =
4666 container_of(head, struct extent_buffer, rcu_head);
4668 __free_extent_buffer(eb);
4671 /* Expects to have eb->eb_lock already held */
4672 static int release_extent_buffer(struct extent_buffer *eb)
4674 WARN_ON(atomic_read(&eb->refs) == 0);
4675 if (atomic_dec_and_test(&eb->refs)) {
4676 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
4677 struct btrfs_fs_info *fs_info = eb->fs_info;
4679 spin_unlock(&eb->refs_lock);
4681 spin_lock(&fs_info->buffer_lock);
4682 radix_tree_delete(&fs_info->buffer_radix,
4683 eb->start >> PAGE_CACHE_SHIFT);
4684 spin_unlock(&fs_info->buffer_lock);
4686 spin_unlock(&eb->refs_lock);
4689 /* Should be safe to release our pages at this point */
4690 btrfs_release_extent_buffer_page(eb, 0);
4691 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
4694 spin_unlock(&eb->refs_lock);
4699 void free_extent_buffer(struct extent_buffer *eb)
4707 refs = atomic_read(&eb->refs);
4710 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
4715 spin_lock(&eb->refs_lock);
4716 if (atomic_read(&eb->refs) == 2 &&
4717 test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))
4718 atomic_dec(&eb->refs);
4720 if (atomic_read(&eb->refs) == 2 &&
4721 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
4722 !extent_buffer_under_io(eb) &&
4723 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4724 atomic_dec(&eb->refs);
4727 * I know this is terrible, but it's temporary until we stop tracking
4728 * the uptodate bits and such for the extent buffers.
4730 release_extent_buffer(eb);
4733 void free_extent_buffer_stale(struct extent_buffer *eb)
4738 spin_lock(&eb->refs_lock);
4739 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
4741 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
4742 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4743 atomic_dec(&eb->refs);
4744 release_extent_buffer(eb);
4747 void clear_extent_buffer_dirty(struct extent_buffer *eb)
4750 unsigned long num_pages;
4753 num_pages = num_extent_pages(eb->start, eb->len);
4755 for (i = 0; i < num_pages; i++) {
4756 page = extent_buffer_page(eb, i);
4757 if (!PageDirty(page))
4761 WARN_ON(!PagePrivate(page));
4763 clear_page_dirty_for_io(page);
4764 spin_lock_irq(&page->mapping->tree_lock);
4765 if (!PageDirty(page)) {
4766 radix_tree_tag_clear(&page->mapping->page_tree,
4768 PAGECACHE_TAG_DIRTY);
4770 spin_unlock_irq(&page->mapping->tree_lock);
4771 ClearPageError(page);
4774 WARN_ON(atomic_read(&eb->refs) == 0);
4777 int set_extent_buffer_dirty(struct extent_buffer *eb)
4780 unsigned long num_pages;
4783 check_buffer_tree_ref(eb);
4785 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
4787 num_pages = num_extent_pages(eb->start, eb->len);
4788 WARN_ON(atomic_read(&eb->refs) == 0);
4789 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
4791 for (i = 0; i < num_pages; i++)
4792 set_page_dirty(extent_buffer_page(eb, i));
4796 int clear_extent_buffer_uptodate(struct extent_buffer *eb)
4800 unsigned long num_pages;
4802 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4803 num_pages = num_extent_pages(eb->start, eb->len);
4804 for (i = 0; i < num_pages; i++) {
4805 page = extent_buffer_page(eb, i);
4807 ClearPageUptodate(page);
4812 int set_extent_buffer_uptodate(struct extent_buffer *eb)
4816 unsigned long num_pages;
4818 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4819 num_pages = num_extent_pages(eb->start, eb->len);
4820 for (i = 0; i < num_pages; i++) {
4821 page = extent_buffer_page(eb, i);
4822 SetPageUptodate(page);
4827 int extent_buffer_uptodate(struct extent_buffer *eb)
4829 return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4832 int read_extent_buffer_pages(struct extent_io_tree *tree,
4833 struct extent_buffer *eb, u64 start, int wait,
4834 get_extent_t *get_extent, int mirror_num)
4837 unsigned long start_i;
4841 int locked_pages = 0;
4842 int all_uptodate = 1;
4843 unsigned long num_pages;
4844 unsigned long num_reads = 0;
4845 struct bio *bio = NULL;
4846 unsigned long bio_flags = 0;
4848 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
4852 WARN_ON(start < eb->start);
4853 start_i = (start >> PAGE_CACHE_SHIFT) -
4854 (eb->start >> PAGE_CACHE_SHIFT);
4859 num_pages = num_extent_pages(eb->start, eb->len);
4860 for (i = start_i; i < num_pages; i++) {
4861 page = extent_buffer_page(eb, i);
4862 if (wait == WAIT_NONE) {
4863 if (!trylock_page(page))
4869 if (!PageUptodate(page)) {
4876 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4880 clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
4881 eb->read_mirror = 0;
4882 atomic_set(&eb->io_pages, num_reads);
4883 for (i = start_i; i < num_pages; i++) {
4884 page = extent_buffer_page(eb, i);
4885 if (!PageUptodate(page)) {
4886 ClearPageError(page);
4887 err = __extent_read_full_page(tree, page,
4889 mirror_num, &bio_flags,
4899 err = submit_one_bio(READ | REQ_META, bio, mirror_num,
4905 if (ret || wait != WAIT_COMPLETE)
4908 for (i = start_i; i < num_pages; i++) {
4909 page = extent_buffer_page(eb, i);
4910 wait_on_page_locked(page);
4911 if (!PageUptodate(page))
4919 while (locked_pages > 0) {
4920 page = extent_buffer_page(eb, i);
4928 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
4929 unsigned long start,
4936 char *dst = (char *)dstv;
4937 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4938 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4940 WARN_ON(start > eb->len);
4941 WARN_ON(start + len > eb->start + eb->len);
4943 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
4946 page = extent_buffer_page(eb, i);
4948 cur = min(len, (PAGE_CACHE_SIZE - offset));
4949 kaddr = page_address(page);
4950 memcpy(dst, kaddr + offset, cur);
4959 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
4960 unsigned long min_len, char **map,
4961 unsigned long *map_start,
4962 unsigned long *map_len)
4964 size_t offset = start & (PAGE_CACHE_SIZE - 1);
4967 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4968 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4969 unsigned long end_i = (start_offset + start + min_len - 1) >>
4976 offset = start_offset;
4980 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
4983 if (start + min_len > eb->len) {
4984 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, "
4986 eb->start, eb->len, start, min_len);
4990 p = extent_buffer_page(eb, i);
4991 kaddr = page_address(p);
4992 *map = kaddr + offset;
4993 *map_len = PAGE_CACHE_SIZE - offset;
4997 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
4998 unsigned long start,
5005 char *ptr = (char *)ptrv;
5006 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5007 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5010 WARN_ON(start > eb->len);
5011 WARN_ON(start + len > eb->start + eb->len);
5013 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5016 page = extent_buffer_page(eb, i);
5018 cur = min(len, (PAGE_CACHE_SIZE - offset));
5020 kaddr = page_address(page);
5021 ret = memcmp(ptr, kaddr + offset, cur);
5033 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
5034 unsigned long start, unsigned long len)
5040 char *src = (char *)srcv;
5041 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5042 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5044 WARN_ON(start > eb->len);
5045 WARN_ON(start + len > eb->start + eb->len);
5047 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5050 page = extent_buffer_page(eb, i);
5051 WARN_ON(!PageUptodate(page));
5053 cur = min(len, PAGE_CACHE_SIZE - offset);
5054 kaddr = page_address(page);
5055 memcpy(kaddr + offset, src, cur);
5064 void memset_extent_buffer(struct extent_buffer *eb, char c,
5065 unsigned long start, unsigned long len)
5071 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5072 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5074 WARN_ON(start > eb->len);
5075 WARN_ON(start + len > eb->start + eb->len);
5077 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5080 page = extent_buffer_page(eb, i);
5081 WARN_ON(!PageUptodate(page));
5083 cur = min(len, PAGE_CACHE_SIZE - offset);
5084 kaddr = page_address(page);
5085 memset(kaddr + offset, c, cur);
5093 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
5094 unsigned long dst_offset, unsigned long src_offset,
5097 u64 dst_len = dst->len;
5102 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5103 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
5105 WARN_ON(src->len != dst_len);
5107 offset = (start_offset + dst_offset) &
5108 (PAGE_CACHE_SIZE - 1);
5111 page = extent_buffer_page(dst, i);
5112 WARN_ON(!PageUptodate(page));
5114 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
5116 kaddr = page_address(page);
5117 read_extent_buffer(src, kaddr + offset, src_offset, cur);
5126 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
5128 unsigned long distance = (src > dst) ? src - dst : dst - src;
5129 return distance < len;
5132 static void copy_pages(struct page *dst_page, struct page *src_page,
5133 unsigned long dst_off, unsigned long src_off,
5136 char *dst_kaddr = page_address(dst_page);
5138 int must_memmove = 0;
5140 if (dst_page != src_page) {
5141 src_kaddr = page_address(src_page);
5143 src_kaddr = dst_kaddr;
5144 if (areas_overlap(src_off, dst_off, len))
5149 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
5151 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
5154 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5155 unsigned long src_offset, unsigned long len)
5158 size_t dst_off_in_page;
5159 size_t src_off_in_page;
5160 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5161 unsigned long dst_i;
5162 unsigned long src_i;
5164 if (src_offset + len > dst->len) {
5165 printk(KERN_ERR "BTRFS: memmove bogus src_offset %lu move "
5166 "len %lu dst len %lu\n", src_offset, len, dst->len);
5169 if (dst_offset + len > dst->len) {
5170 printk(KERN_ERR "BTRFS: memmove bogus dst_offset %lu move "
5171 "len %lu dst len %lu\n", dst_offset, len, dst->len);
5176 dst_off_in_page = (start_offset + dst_offset) &
5177 (PAGE_CACHE_SIZE - 1);
5178 src_off_in_page = (start_offset + src_offset) &
5179 (PAGE_CACHE_SIZE - 1);
5181 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
5182 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
5184 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
5186 cur = min_t(unsigned long, cur,
5187 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
5189 copy_pages(extent_buffer_page(dst, dst_i),
5190 extent_buffer_page(dst, src_i),
5191 dst_off_in_page, src_off_in_page, cur);
5199 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5200 unsigned long src_offset, unsigned long len)
5203 size_t dst_off_in_page;
5204 size_t src_off_in_page;
5205 unsigned long dst_end = dst_offset + len - 1;
5206 unsigned long src_end = src_offset + len - 1;
5207 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5208 unsigned long dst_i;
5209 unsigned long src_i;
5211 if (src_offset + len > dst->len) {
5212 printk(KERN_ERR "BTRFS: memmove bogus src_offset %lu move "
5213 "len %lu len %lu\n", src_offset, len, dst->len);
5216 if (dst_offset + len > dst->len) {
5217 printk(KERN_ERR "BTRFS: memmove bogus dst_offset %lu move "
5218 "len %lu len %lu\n", dst_offset, len, dst->len);
5221 if (dst_offset < src_offset) {
5222 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
5226 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
5227 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
5229 dst_off_in_page = (start_offset + dst_end) &
5230 (PAGE_CACHE_SIZE - 1);
5231 src_off_in_page = (start_offset + src_end) &
5232 (PAGE_CACHE_SIZE - 1);
5234 cur = min_t(unsigned long, len, src_off_in_page + 1);
5235 cur = min(cur, dst_off_in_page + 1);
5236 copy_pages(extent_buffer_page(dst, dst_i),
5237 extent_buffer_page(dst, src_i),
5238 dst_off_in_page - cur + 1,
5239 src_off_in_page - cur + 1, cur);
5247 int try_release_extent_buffer(struct page *page)
5249 struct extent_buffer *eb;
5252 * We need to make sure noboody is attaching this page to an eb right
5255 spin_lock(&page->mapping->private_lock);
5256 if (!PagePrivate(page)) {
5257 spin_unlock(&page->mapping->private_lock);
5261 eb = (struct extent_buffer *)page->private;
5265 * This is a little awful but should be ok, we need to make sure that
5266 * the eb doesn't disappear out from under us while we're looking at
5269 spin_lock(&eb->refs_lock);
5270 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
5271 spin_unlock(&eb->refs_lock);
5272 spin_unlock(&page->mapping->private_lock);
5275 spin_unlock(&page->mapping->private_lock);
5278 * If tree ref isn't set then we know the ref on this eb is a real ref,
5279 * so just return, this page will likely be freed soon anyway.
5281 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
5282 spin_unlock(&eb->refs_lock);
5286 return release_extent_buffer(eb);