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);
2029 bio->bi_iter.bi_size = 0;
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_iter.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_iter.bi_sector = failrec->logical >> 9;
2315 bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
2316 bio->bi_iter.bi_size = 0;
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;
2383 bio_for_each_segment_all(bvec, bio, i) {
2384 struct page *page = bvec->bv_page;
2386 /* We always issue full-page reads, but if some block
2387 * in a page fails to read, blk_update_request() will
2388 * advance bv_offset and adjust bv_len to compensate.
2389 * Print a warning for nonzero offsets, and an error
2390 * if they don't add up to a full page. */
2391 if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE) {
2392 if (bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE)
2393 btrfs_err(BTRFS_I(page->mapping->host)->root->fs_info,
2394 "partial page write in btrfs with offset %u and length %u",
2395 bvec->bv_offset, bvec->bv_len);
2397 btrfs_info(BTRFS_I(page->mapping->host)->root->fs_info,
2398 "incomplete page write in btrfs with offset %u and "
2400 bvec->bv_offset, bvec->bv_len);
2403 start = page_offset(page);
2404 end = start + bvec->bv_offset + bvec->bv_len - 1;
2406 if (end_extent_writepage(page, err, start, end))
2409 end_page_writeback(page);
2416 endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len,
2419 struct extent_state *cached = NULL;
2420 u64 end = start + len - 1;
2422 if (uptodate && tree->track_uptodate)
2423 set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC);
2424 unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC);
2428 * after a readpage IO is done, we need to:
2429 * clear the uptodate bits on error
2430 * set the uptodate bits if things worked
2431 * set the page up to date if all extents in the tree are uptodate
2432 * clear the lock bit in the extent tree
2433 * unlock the page if there are no other extents locked for it
2435 * Scheduling is not allowed, so the extent state tree is expected
2436 * to have one and only one object corresponding to this IO.
2438 static void end_bio_extent_readpage(struct bio *bio, int err)
2440 struct bio_vec *bvec;
2441 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
2442 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
2443 struct extent_io_tree *tree;
2448 u64 extent_start = 0;
2457 bio_for_each_segment_all(bvec, bio, i) {
2458 struct page *page = bvec->bv_page;
2459 struct inode *inode = page->mapping->host;
2461 pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, "
2462 "mirror=%lu\n", (u64)bio->bi_iter.bi_sector, err,
2463 io_bio->mirror_num);
2464 tree = &BTRFS_I(inode)->io_tree;
2466 /* We always issue full-page reads, but if some block
2467 * in a page fails to read, blk_update_request() will
2468 * advance bv_offset and adjust bv_len to compensate.
2469 * Print a warning for nonzero offsets, and an error
2470 * if they don't add up to a full page. */
2471 if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE) {
2472 if (bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE)
2473 btrfs_err(BTRFS_I(page->mapping->host)->root->fs_info,
2474 "partial page read in btrfs with offset %u and length %u",
2475 bvec->bv_offset, bvec->bv_len);
2477 btrfs_info(BTRFS_I(page->mapping->host)->root->fs_info,
2478 "incomplete page read in btrfs with offset %u and "
2480 bvec->bv_offset, bvec->bv_len);
2483 start = page_offset(page);
2484 end = start + bvec->bv_offset + bvec->bv_len - 1;
2487 mirror = io_bio->mirror_num;
2488 if (likely(uptodate && tree->ops &&
2489 tree->ops->readpage_end_io_hook)) {
2490 ret = tree->ops->readpage_end_io_hook(io_bio, offset,
2496 clean_io_failure(start, page);
2499 if (likely(uptodate))
2502 if (tree->ops && tree->ops->readpage_io_failed_hook) {
2503 ret = tree->ops->readpage_io_failed_hook(page, mirror);
2505 test_bit(BIO_UPTODATE, &bio->bi_flags))
2509 * The generic bio_readpage_error handles errors the
2510 * following way: If possible, new read requests are
2511 * created and submitted and will end up in
2512 * end_bio_extent_readpage as well (if we're lucky, not
2513 * in the !uptodate case). In that case it returns 0 and
2514 * we just go on with the next page in our bio. If it
2515 * can't handle the error it will return -EIO and we
2516 * remain responsible for that page.
2518 ret = bio_readpage_error(bio, offset, page, start, end,
2522 test_bit(BIO_UPTODATE, &bio->bi_flags);
2529 if (likely(uptodate)) {
2530 loff_t i_size = i_size_read(inode);
2531 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
2534 /* Zero out the end if this page straddles i_size */
2535 offset = i_size & (PAGE_CACHE_SIZE-1);
2536 if (page->index == end_index && offset)
2537 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
2538 SetPageUptodate(page);
2540 ClearPageUptodate(page);
2546 if (unlikely(!uptodate)) {
2548 endio_readpage_release_extent(tree,
2554 endio_readpage_release_extent(tree, start,
2555 end - start + 1, 0);
2556 } else if (!extent_len) {
2557 extent_start = start;
2558 extent_len = end + 1 - start;
2559 } else if (extent_start + extent_len == start) {
2560 extent_len += end + 1 - start;
2562 endio_readpage_release_extent(tree, extent_start,
2563 extent_len, uptodate);
2564 extent_start = start;
2565 extent_len = end + 1 - start;
2570 endio_readpage_release_extent(tree, extent_start, extent_len,
2573 io_bio->end_io(io_bio, err);
2578 * this allocates from the btrfs_bioset. We're returning a bio right now
2579 * but you can call btrfs_io_bio for the appropriate container_of magic
2582 btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
2585 struct btrfs_io_bio *btrfs_bio;
2588 bio = bio_alloc_bioset(gfp_flags, nr_vecs, btrfs_bioset);
2590 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
2591 while (!bio && (nr_vecs /= 2)) {
2592 bio = bio_alloc_bioset(gfp_flags,
2593 nr_vecs, btrfs_bioset);
2598 bio->bi_bdev = bdev;
2599 bio->bi_iter.bi_sector = first_sector;
2600 btrfs_bio = btrfs_io_bio(bio);
2601 btrfs_bio->csum = NULL;
2602 btrfs_bio->csum_allocated = NULL;
2603 btrfs_bio->end_io = NULL;
2608 struct bio *btrfs_bio_clone(struct bio *bio, gfp_t gfp_mask)
2610 return bio_clone_bioset(bio, gfp_mask, btrfs_bioset);
2614 /* this also allocates from the btrfs_bioset */
2615 struct bio *btrfs_io_bio_alloc(gfp_t gfp_mask, unsigned int nr_iovecs)
2617 struct btrfs_io_bio *btrfs_bio;
2620 bio = bio_alloc_bioset(gfp_mask, nr_iovecs, btrfs_bioset);
2622 btrfs_bio = btrfs_io_bio(bio);
2623 btrfs_bio->csum = NULL;
2624 btrfs_bio->csum_allocated = NULL;
2625 btrfs_bio->end_io = NULL;
2631 static int __must_check submit_one_bio(int rw, struct bio *bio,
2632 int mirror_num, unsigned long bio_flags)
2635 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2636 struct page *page = bvec->bv_page;
2637 struct extent_io_tree *tree = bio->bi_private;
2640 start = page_offset(page) + bvec->bv_offset;
2642 bio->bi_private = NULL;
2646 if (tree->ops && tree->ops->submit_bio_hook)
2647 ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
2648 mirror_num, bio_flags, start);
2650 btrfsic_submit_bio(rw, bio);
2652 if (bio_flagged(bio, BIO_EOPNOTSUPP))
2658 static int merge_bio(int rw, struct extent_io_tree *tree, struct page *page,
2659 unsigned long offset, size_t size, struct bio *bio,
2660 unsigned long bio_flags)
2663 if (tree->ops && tree->ops->merge_bio_hook)
2664 ret = tree->ops->merge_bio_hook(rw, page, offset, size, bio,
2671 static int submit_extent_page(int rw, struct extent_io_tree *tree,
2672 struct page *page, sector_t sector,
2673 size_t size, unsigned long offset,
2674 struct block_device *bdev,
2675 struct bio **bio_ret,
2676 unsigned long max_pages,
2677 bio_end_io_t end_io_func,
2679 unsigned long prev_bio_flags,
2680 unsigned long bio_flags)
2686 int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED;
2687 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
2688 size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE);
2690 if (bio_ret && *bio_ret) {
2693 contig = bio->bi_iter.bi_sector == sector;
2695 contig = bio_end_sector(bio) == sector;
2697 if (prev_bio_flags != bio_flags || !contig ||
2698 merge_bio(rw, tree, page, offset, page_size, bio, bio_flags) ||
2699 bio_add_page(bio, page, page_size, offset) < page_size) {
2700 ret = submit_one_bio(rw, bio, mirror_num,
2709 if (this_compressed)
2712 nr = bio_get_nr_vecs(bdev);
2714 bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
2718 bio_add_page(bio, page, page_size, offset);
2719 bio->bi_end_io = end_io_func;
2720 bio->bi_private = tree;
2725 ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
2730 static void attach_extent_buffer_page(struct extent_buffer *eb,
2733 if (!PagePrivate(page)) {
2734 SetPagePrivate(page);
2735 page_cache_get(page);
2736 set_page_private(page, (unsigned long)eb);
2738 WARN_ON(page->private != (unsigned long)eb);
2742 void set_page_extent_mapped(struct page *page)
2744 if (!PagePrivate(page)) {
2745 SetPagePrivate(page);
2746 page_cache_get(page);
2747 set_page_private(page, EXTENT_PAGE_PRIVATE);
2751 static struct extent_map *
2752 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
2753 u64 start, u64 len, get_extent_t *get_extent,
2754 struct extent_map **em_cached)
2756 struct extent_map *em;
2758 if (em_cached && *em_cached) {
2760 if (em->in_tree && start >= em->start &&
2761 start < extent_map_end(em)) {
2762 atomic_inc(&em->refs);
2766 free_extent_map(em);
2770 em = get_extent(inode, page, pg_offset, start, len, 0);
2771 if (em_cached && !IS_ERR_OR_NULL(em)) {
2773 atomic_inc(&em->refs);
2779 * basic readpage implementation. Locked extent state structs are inserted
2780 * into the tree that are removed when the IO is done (by the end_io
2782 * XXX JDM: This needs looking at to ensure proper page locking
2784 static int __do_readpage(struct extent_io_tree *tree,
2786 get_extent_t *get_extent,
2787 struct extent_map **em_cached,
2788 struct bio **bio, int mirror_num,
2789 unsigned long *bio_flags, int rw)
2791 struct inode *inode = page->mapping->host;
2792 u64 start = page_offset(page);
2793 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2797 u64 last_byte = i_size_read(inode);
2801 struct extent_map *em;
2802 struct block_device *bdev;
2805 int parent_locked = *bio_flags & EXTENT_BIO_PARENT_LOCKED;
2806 size_t pg_offset = 0;
2808 size_t disk_io_size;
2809 size_t blocksize = inode->i_sb->s_blocksize;
2810 unsigned long this_bio_flag = *bio_flags & EXTENT_BIO_PARENT_LOCKED;
2812 set_page_extent_mapped(page);
2815 if (!PageUptodate(page)) {
2816 if (cleancache_get_page(page) == 0) {
2817 BUG_ON(blocksize != PAGE_SIZE);
2818 unlock_extent(tree, start, end);
2823 if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
2825 size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
2828 iosize = PAGE_CACHE_SIZE - zero_offset;
2829 userpage = kmap_atomic(page);
2830 memset(userpage + zero_offset, 0, iosize);
2831 flush_dcache_page(page);
2832 kunmap_atomic(userpage);
2835 while (cur <= end) {
2836 unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
2838 if (cur >= last_byte) {
2840 struct extent_state *cached = NULL;
2842 iosize = PAGE_CACHE_SIZE - pg_offset;
2843 userpage = kmap_atomic(page);
2844 memset(userpage + pg_offset, 0, iosize);
2845 flush_dcache_page(page);
2846 kunmap_atomic(userpage);
2847 set_extent_uptodate(tree, cur, cur + iosize - 1,
2850 unlock_extent_cached(tree, cur,
2855 em = __get_extent_map(inode, page, pg_offset, cur,
2856 end - cur + 1, get_extent, em_cached);
2857 if (IS_ERR_OR_NULL(em)) {
2860 unlock_extent(tree, cur, end);
2863 extent_offset = cur - em->start;
2864 BUG_ON(extent_map_end(em) <= cur);
2867 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2868 this_bio_flag |= EXTENT_BIO_COMPRESSED;
2869 extent_set_compress_type(&this_bio_flag,
2873 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2874 cur_end = min(extent_map_end(em) - 1, end);
2875 iosize = ALIGN(iosize, blocksize);
2876 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2877 disk_io_size = em->block_len;
2878 sector = em->block_start >> 9;
2880 sector = (em->block_start + extent_offset) >> 9;
2881 disk_io_size = iosize;
2884 block_start = em->block_start;
2885 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2886 block_start = EXTENT_MAP_HOLE;
2887 free_extent_map(em);
2890 /* we've found a hole, just zero and go on */
2891 if (block_start == EXTENT_MAP_HOLE) {
2893 struct extent_state *cached = NULL;
2895 userpage = kmap_atomic(page);
2896 memset(userpage + pg_offset, 0, iosize);
2897 flush_dcache_page(page);
2898 kunmap_atomic(userpage);
2900 set_extent_uptodate(tree, cur, cur + iosize - 1,
2902 unlock_extent_cached(tree, cur, cur + iosize - 1,
2905 pg_offset += iosize;
2908 /* the get_extent function already copied into the page */
2909 if (test_range_bit(tree, cur, cur_end,
2910 EXTENT_UPTODATE, 1, NULL)) {
2911 check_page_uptodate(tree, page);
2913 unlock_extent(tree, cur, cur + iosize - 1);
2915 pg_offset += iosize;
2918 /* we have an inline extent but it didn't get marked up
2919 * to date. Error out
2921 if (block_start == EXTENT_MAP_INLINE) {
2924 unlock_extent(tree, cur, cur + iosize - 1);
2926 pg_offset += iosize;
2931 ret = submit_extent_page(rw, tree, page,
2932 sector, disk_io_size, pg_offset,
2934 end_bio_extent_readpage, mirror_num,
2939 *bio_flags = this_bio_flag;
2943 unlock_extent(tree, cur, cur + iosize - 1);
2946 pg_offset += iosize;
2950 if (!PageError(page))
2951 SetPageUptodate(page);
2957 static inline void __do_contiguous_readpages(struct extent_io_tree *tree,
2958 struct page *pages[], int nr_pages,
2960 get_extent_t *get_extent,
2961 struct extent_map **em_cached,
2962 struct bio **bio, int mirror_num,
2963 unsigned long *bio_flags, int rw)
2965 struct inode *inode;
2966 struct btrfs_ordered_extent *ordered;
2969 inode = pages[0]->mapping->host;
2971 lock_extent(tree, start, end);
2972 ordered = btrfs_lookup_ordered_range(inode, start,
2976 unlock_extent(tree, start, end);
2977 btrfs_start_ordered_extent(inode, ordered, 1);
2978 btrfs_put_ordered_extent(ordered);
2981 for (index = 0; index < nr_pages; index++) {
2982 __do_readpage(tree, pages[index], get_extent, em_cached, bio,
2983 mirror_num, bio_flags, rw);
2984 page_cache_release(pages[index]);
2988 static void __extent_readpages(struct extent_io_tree *tree,
2989 struct page *pages[],
2990 int nr_pages, get_extent_t *get_extent,
2991 struct extent_map **em_cached,
2992 struct bio **bio, int mirror_num,
2993 unsigned long *bio_flags, int rw)
2999 int first_index = 0;
3001 for (index = 0; index < nr_pages; index++) {
3002 page_start = page_offset(pages[index]);
3005 end = start + PAGE_CACHE_SIZE - 1;
3006 first_index = index;
3007 } else if (end + 1 == page_start) {
3008 end += PAGE_CACHE_SIZE;
3010 __do_contiguous_readpages(tree, &pages[first_index],
3011 index - first_index, start,
3012 end, get_extent, em_cached,
3013 bio, mirror_num, bio_flags,
3016 end = start + PAGE_CACHE_SIZE - 1;
3017 first_index = index;
3022 __do_contiguous_readpages(tree, &pages[first_index],
3023 index - first_index, start,
3024 end, get_extent, em_cached, bio,
3025 mirror_num, bio_flags, rw);
3028 static int __extent_read_full_page(struct extent_io_tree *tree,
3030 get_extent_t *get_extent,
3031 struct bio **bio, int mirror_num,
3032 unsigned long *bio_flags, int rw)
3034 struct inode *inode = page->mapping->host;
3035 struct btrfs_ordered_extent *ordered;
3036 u64 start = page_offset(page);
3037 u64 end = start + PAGE_CACHE_SIZE - 1;
3041 lock_extent(tree, start, end);
3042 ordered = btrfs_lookup_ordered_extent(inode, start);
3045 unlock_extent(tree, start, end);
3046 btrfs_start_ordered_extent(inode, ordered, 1);
3047 btrfs_put_ordered_extent(ordered);
3050 ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num,
3055 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
3056 get_extent_t *get_extent, int mirror_num)
3058 struct bio *bio = NULL;
3059 unsigned long bio_flags = 0;
3062 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
3065 ret = submit_one_bio(READ, bio, mirror_num, bio_flags);
3069 int extent_read_full_page_nolock(struct extent_io_tree *tree, struct page *page,
3070 get_extent_t *get_extent, int mirror_num)
3072 struct bio *bio = NULL;
3073 unsigned long bio_flags = EXTENT_BIO_PARENT_LOCKED;
3076 ret = __do_readpage(tree, page, get_extent, NULL, &bio, mirror_num,
3079 ret = submit_one_bio(READ, bio, mirror_num, bio_flags);
3083 static noinline void update_nr_written(struct page *page,
3084 struct writeback_control *wbc,
3085 unsigned long nr_written)
3087 wbc->nr_to_write -= nr_written;
3088 if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
3089 wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
3090 page->mapping->writeback_index = page->index + nr_written;
3094 * the writepage semantics are similar to regular writepage. extent
3095 * records are inserted to lock ranges in the tree, and as dirty areas
3096 * are found, they are marked writeback. Then the lock bits are removed
3097 * and the end_io handler clears the writeback ranges
3099 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
3102 struct inode *inode = page->mapping->host;
3103 struct extent_page_data *epd = data;
3104 struct extent_io_tree *tree = epd->tree;
3105 u64 start = page_offset(page);
3107 u64 page_end = start + PAGE_CACHE_SIZE - 1;
3111 u64 last_byte = i_size_read(inode);
3115 struct extent_state *cached_state = NULL;
3116 struct extent_map *em;
3117 struct block_device *bdev;
3120 size_t pg_offset = 0;
3122 loff_t i_size = i_size_read(inode);
3123 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
3129 unsigned long nr_written = 0;
3130 bool fill_delalloc = true;
3132 if (wbc->sync_mode == WB_SYNC_ALL)
3133 write_flags = WRITE_SYNC;
3135 write_flags = WRITE;
3137 trace___extent_writepage(page, inode, wbc);
3139 WARN_ON(!PageLocked(page));
3141 ClearPageError(page);
3143 pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
3144 if (page->index > end_index ||
3145 (page->index == end_index && !pg_offset)) {
3146 page->mapping->a_ops->invalidatepage(page, 0, PAGE_CACHE_SIZE);
3151 if (page->index == end_index) {
3154 userpage = kmap_atomic(page);
3155 memset(userpage + pg_offset, 0,
3156 PAGE_CACHE_SIZE - pg_offset);
3157 kunmap_atomic(userpage);
3158 flush_dcache_page(page);
3162 set_page_extent_mapped(page);
3164 if (!tree->ops || !tree->ops->fill_delalloc)
3165 fill_delalloc = false;
3167 delalloc_start = start;
3170 if (!epd->extent_locked && fill_delalloc) {
3171 u64 delalloc_to_write = 0;
3173 * make sure the wbc mapping index is at least updated
3176 update_nr_written(page, wbc, 0);
3178 while (delalloc_end < page_end) {
3179 nr_delalloc = find_lock_delalloc_range(inode, tree,
3184 if (nr_delalloc == 0) {
3185 delalloc_start = delalloc_end + 1;
3188 ret = tree->ops->fill_delalloc(inode, page,
3193 /* File system has been set read-only */
3199 * delalloc_end is already one less than the total
3200 * length, so we don't subtract one from
3203 delalloc_to_write += (delalloc_end - delalloc_start +
3206 delalloc_start = delalloc_end + 1;
3208 if (wbc->nr_to_write < delalloc_to_write) {
3211 if (delalloc_to_write < thresh * 2)
3212 thresh = delalloc_to_write;
3213 wbc->nr_to_write = min_t(u64, delalloc_to_write,
3217 /* did the fill delalloc function already unlock and start
3223 * we've unlocked the page, so we can't update
3224 * the mapping's writeback index, just update
3227 wbc->nr_to_write -= nr_written;
3231 if (tree->ops && tree->ops->writepage_start_hook) {
3232 ret = tree->ops->writepage_start_hook(page, start,
3235 /* Fixup worker will requeue */
3237 wbc->pages_skipped++;
3239 redirty_page_for_writepage(wbc, page);
3240 update_nr_written(page, wbc, nr_written);
3248 * we don't want to touch the inode after unlocking the page,
3249 * so we update the mapping writeback index now
3251 update_nr_written(page, wbc, nr_written + 1);
3254 if (last_byte <= start) {
3255 if (tree->ops && tree->ops->writepage_end_io_hook)
3256 tree->ops->writepage_end_io_hook(page, start,
3261 blocksize = inode->i_sb->s_blocksize;
3263 while (cur <= end) {
3264 if (cur >= last_byte) {
3265 if (tree->ops && tree->ops->writepage_end_io_hook)
3266 tree->ops->writepage_end_io_hook(page, cur,
3270 em = epd->get_extent(inode, page, pg_offset, cur,
3272 if (IS_ERR_OR_NULL(em)) {
3277 extent_offset = cur - em->start;
3278 BUG_ON(extent_map_end(em) <= cur);
3280 iosize = min(extent_map_end(em) - cur, end - cur + 1);
3281 iosize = ALIGN(iosize, blocksize);
3282 sector = (em->block_start + extent_offset) >> 9;
3284 block_start = em->block_start;
3285 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3286 free_extent_map(em);
3290 * compressed and inline extents are written through other
3293 if (compressed || block_start == EXTENT_MAP_HOLE ||
3294 block_start == EXTENT_MAP_INLINE) {
3296 * end_io notification does not happen here for
3297 * compressed extents
3299 if (!compressed && tree->ops &&
3300 tree->ops->writepage_end_io_hook)
3301 tree->ops->writepage_end_io_hook(page, cur,
3304 else if (compressed) {
3305 /* we don't want to end_page_writeback on
3306 * a compressed extent. this happens
3313 pg_offset += iosize;
3316 /* leave this out until we have a page_mkwrite call */
3317 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
3318 EXTENT_DIRTY, 0, NULL)) {
3320 pg_offset += iosize;
3324 if (tree->ops && tree->ops->writepage_io_hook) {
3325 ret = tree->ops->writepage_io_hook(page, cur,
3333 unsigned long max_nr = end_index + 1;
3335 set_range_writeback(tree, cur, cur + iosize - 1);
3336 if (!PageWriteback(page)) {
3337 btrfs_err(BTRFS_I(inode)->root->fs_info,
3338 "page %lu not writeback, cur %llu end %llu",
3339 page->index, cur, end);
3342 ret = submit_extent_page(write_flags, tree, page,
3343 sector, iosize, pg_offset,
3344 bdev, &epd->bio, max_nr,
3345 end_bio_extent_writepage,
3351 pg_offset += iosize;
3356 /* make sure the mapping tag for page dirty gets cleared */
3357 set_page_writeback(page);
3358 end_page_writeback(page);
3364 /* drop our reference on any cached states */
3365 free_extent_state(cached_state);
3369 static int eb_wait(void *word)
3375 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3377 wait_on_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK, eb_wait,
3378 TASK_UNINTERRUPTIBLE);
3381 static int lock_extent_buffer_for_io(struct extent_buffer *eb,
3382 struct btrfs_fs_info *fs_info,
3383 struct extent_page_data *epd)
3385 unsigned long i, num_pages;
3389 if (!btrfs_try_tree_write_lock(eb)) {
3391 flush_write_bio(epd);
3392 btrfs_tree_lock(eb);
3395 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3396 btrfs_tree_unlock(eb);
3400 flush_write_bio(epd);
3404 wait_on_extent_buffer_writeback(eb);
3405 btrfs_tree_lock(eb);
3406 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3408 btrfs_tree_unlock(eb);
3413 * We need to do this to prevent races in people who check if the eb is
3414 * under IO since we can end up having no IO bits set for a short period
3417 spin_lock(&eb->refs_lock);
3418 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3419 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3420 spin_unlock(&eb->refs_lock);
3421 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3422 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
3424 fs_info->dirty_metadata_batch);
3427 spin_unlock(&eb->refs_lock);
3430 btrfs_tree_unlock(eb);
3435 num_pages = num_extent_pages(eb->start, eb->len);
3436 for (i = 0; i < num_pages; i++) {
3437 struct page *p = extent_buffer_page(eb, i);
3439 if (!trylock_page(p)) {
3441 flush_write_bio(epd);
3451 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3453 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3454 smp_mb__after_clear_bit();
3455 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3458 static void end_bio_extent_buffer_writepage(struct bio *bio, int err)
3460 struct bio_vec *bvec;
3461 struct extent_buffer *eb;
3464 bio_for_each_segment_all(bvec, bio, i) {
3465 struct page *page = bvec->bv_page;
3467 eb = (struct extent_buffer *)page->private;
3469 done = atomic_dec_and_test(&eb->io_pages);
3471 if (err || test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
3472 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3473 ClearPageUptodate(page);
3477 end_page_writeback(page);
3482 end_extent_buffer_writeback(eb);
3488 static int write_one_eb(struct extent_buffer *eb,
3489 struct btrfs_fs_info *fs_info,
3490 struct writeback_control *wbc,
3491 struct extent_page_data *epd)
3493 struct block_device *bdev = fs_info->fs_devices->latest_bdev;
3494 struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
3495 u64 offset = eb->start;
3496 unsigned long i, num_pages;
3497 unsigned long bio_flags = 0;
3498 int rw = (epd->sync_io ? WRITE_SYNC : WRITE) | REQ_META;
3501 clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3502 num_pages = num_extent_pages(eb->start, eb->len);
3503 atomic_set(&eb->io_pages, num_pages);
3504 if (btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID)
3505 bio_flags = EXTENT_BIO_TREE_LOG;
3507 for (i = 0; i < num_pages; i++) {
3508 struct page *p = extent_buffer_page(eb, i);
3510 clear_page_dirty_for_io(p);
3511 set_page_writeback(p);
3512 ret = submit_extent_page(rw, tree, p, offset >> 9,
3513 PAGE_CACHE_SIZE, 0, bdev, &epd->bio,
3514 -1, end_bio_extent_buffer_writepage,
3515 0, epd->bio_flags, bio_flags);
3516 epd->bio_flags = bio_flags;
3518 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3520 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3521 end_extent_buffer_writeback(eb);
3525 offset += PAGE_CACHE_SIZE;
3526 update_nr_written(p, wbc, 1);
3530 if (unlikely(ret)) {
3531 for (; i < num_pages; i++) {
3532 struct page *p = extent_buffer_page(eb, i);
3540 int btree_write_cache_pages(struct address_space *mapping,
3541 struct writeback_control *wbc)
3543 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
3544 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
3545 struct extent_buffer *eb, *prev_eb = NULL;
3546 struct extent_page_data epd = {
3550 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3555 int nr_to_write_done = 0;
3556 struct pagevec pvec;
3559 pgoff_t end; /* Inclusive */
3563 pagevec_init(&pvec, 0);
3564 if (wbc->range_cyclic) {
3565 index = mapping->writeback_index; /* Start from prev offset */
3568 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3569 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3572 if (wbc->sync_mode == WB_SYNC_ALL)
3573 tag = PAGECACHE_TAG_TOWRITE;
3575 tag = PAGECACHE_TAG_DIRTY;
3577 if (wbc->sync_mode == WB_SYNC_ALL)
3578 tag_pages_for_writeback(mapping, index, end);
3579 while (!done && !nr_to_write_done && (index <= end) &&
3580 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3581 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3585 for (i = 0; i < nr_pages; i++) {
3586 struct page *page = pvec.pages[i];
3588 if (!PagePrivate(page))
3591 if (!wbc->range_cyclic && page->index > end) {
3596 spin_lock(&mapping->private_lock);
3597 if (!PagePrivate(page)) {
3598 spin_unlock(&mapping->private_lock);
3602 eb = (struct extent_buffer *)page->private;
3605 * Shouldn't happen and normally this would be a BUG_ON
3606 * but no sense in crashing the users box for something
3607 * we can survive anyway.
3610 spin_unlock(&mapping->private_lock);
3614 if (eb == prev_eb) {
3615 spin_unlock(&mapping->private_lock);
3619 ret = atomic_inc_not_zero(&eb->refs);
3620 spin_unlock(&mapping->private_lock);
3625 ret = lock_extent_buffer_for_io(eb, fs_info, &epd);
3627 free_extent_buffer(eb);
3631 ret = write_one_eb(eb, fs_info, wbc, &epd);
3634 free_extent_buffer(eb);
3637 free_extent_buffer(eb);
3640 * the filesystem may choose to bump up nr_to_write.
3641 * We have to make sure to honor the new nr_to_write
3644 nr_to_write_done = wbc->nr_to_write <= 0;
3646 pagevec_release(&pvec);
3649 if (!scanned && !done) {
3651 * We hit the last page and there is more work to be done: wrap
3652 * back to the start of the file
3658 flush_write_bio(&epd);
3663 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3664 * @mapping: address space structure to write
3665 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3666 * @writepage: function called for each page
3667 * @data: data passed to writepage function
3669 * If a page is already under I/O, write_cache_pages() skips it, even
3670 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3671 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3672 * and msync() need to guarantee that all the data which was dirty at the time
3673 * the call was made get new I/O started against them. If wbc->sync_mode is
3674 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3675 * existing IO to complete.
3677 static int extent_write_cache_pages(struct extent_io_tree *tree,
3678 struct address_space *mapping,
3679 struct writeback_control *wbc,
3680 writepage_t writepage, void *data,
3681 void (*flush_fn)(void *))
3683 struct inode *inode = mapping->host;
3686 int nr_to_write_done = 0;
3687 struct pagevec pvec;
3690 pgoff_t end; /* Inclusive */
3695 * We have to hold onto the inode so that ordered extents can do their
3696 * work when the IO finishes. The alternative to this is failing to add
3697 * an ordered extent if the igrab() fails there and that is a huge pain
3698 * to deal with, so instead just hold onto the inode throughout the
3699 * writepages operation. If it fails here we are freeing up the inode
3700 * anyway and we'd rather not waste our time writing out stuff that is
3701 * going to be truncated anyway.
3706 pagevec_init(&pvec, 0);
3707 if (wbc->range_cyclic) {
3708 index = mapping->writeback_index; /* Start from prev offset */
3711 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3712 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3715 if (wbc->sync_mode == WB_SYNC_ALL)
3716 tag = PAGECACHE_TAG_TOWRITE;
3718 tag = PAGECACHE_TAG_DIRTY;
3720 if (wbc->sync_mode == WB_SYNC_ALL)
3721 tag_pages_for_writeback(mapping, index, end);
3722 while (!done && !nr_to_write_done && (index <= end) &&
3723 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3724 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3728 for (i = 0; i < nr_pages; i++) {
3729 struct page *page = pvec.pages[i];
3732 * At this point we hold neither mapping->tree_lock nor
3733 * lock on the page itself: the page may be truncated or
3734 * invalidated (changing page->mapping to NULL), or even
3735 * swizzled back from swapper_space to tmpfs file
3738 if (!trylock_page(page)) {
3743 if (unlikely(page->mapping != mapping)) {
3748 if (!wbc->range_cyclic && page->index > end) {
3754 if (wbc->sync_mode != WB_SYNC_NONE) {
3755 if (PageWriteback(page))
3757 wait_on_page_writeback(page);
3760 if (PageWriteback(page) ||
3761 !clear_page_dirty_for_io(page)) {
3766 ret = (*writepage)(page, wbc, data);
3768 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
3776 * the filesystem may choose to bump up nr_to_write.
3777 * We have to make sure to honor the new nr_to_write
3780 nr_to_write_done = wbc->nr_to_write <= 0;
3782 pagevec_release(&pvec);
3785 if (!scanned && !done) {
3787 * We hit the last page and there is more work to be done: wrap
3788 * back to the start of the file
3794 btrfs_add_delayed_iput(inode);
3798 static void flush_epd_write_bio(struct extent_page_data *epd)
3807 ret = submit_one_bio(rw, epd->bio, 0, epd->bio_flags);
3808 BUG_ON(ret < 0); /* -ENOMEM */
3813 static noinline void flush_write_bio(void *data)
3815 struct extent_page_data *epd = data;
3816 flush_epd_write_bio(epd);
3819 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
3820 get_extent_t *get_extent,
3821 struct writeback_control *wbc)
3824 struct extent_page_data epd = {
3827 .get_extent = get_extent,
3829 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3833 ret = __extent_writepage(page, wbc, &epd);
3835 flush_epd_write_bio(&epd);
3839 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
3840 u64 start, u64 end, get_extent_t *get_extent,
3844 struct address_space *mapping = inode->i_mapping;
3846 unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
3849 struct extent_page_data epd = {
3852 .get_extent = get_extent,
3854 .sync_io = mode == WB_SYNC_ALL,
3857 struct writeback_control wbc_writepages = {
3859 .nr_to_write = nr_pages * 2,
3860 .range_start = start,
3861 .range_end = end + 1,
3864 while (start <= end) {
3865 page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
3866 if (clear_page_dirty_for_io(page))
3867 ret = __extent_writepage(page, &wbc_writepages, &epd);
3869 if (tree->ops && tree->ops->writepage_end_io_hook)
3870 tree->ops->writepage_end_io_hook(page, start,
3871 start + PAGE_CACHE_SIZE - 1,
3875 page_cache_release(page);
3876 start += PAGE_CACHE_SIZE;
3879 flush_epd_write_bio(&epd);
3883 int extent_writepages(struct extent_io_tree *tree,
3884 struct address_space *mapping,
3885 get_extent_t *get_extent,
3886 struct writeback_control *wbc)
3889 struct extent_page_data epd = {
3892 .get_extent = get_extent,
3894 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3898 ret = extent_write_cache_pages(tree, mapping, wbc,
3899 __extent_writepage, &epd,
3901 flush_epd_write_bio(&epd);
3905 int extent_readpages(struct extent_io_tree *tree,
3906 struct address_space *mapping,
3907 struct list_head *pages, unsigned nr_pages,
3908 get_extent_t get_extent)
3910 struct bio *bio = NULL;
3912 unsigned long bio_flags = 0;
3913 struct page *pagepool[16];
3915 struct extent_map *em_cached = NULL;
3918 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
3919 page = list_entry(pages->prev, struct page, lru);
3921 prefetchw(&page->flags);
3922 list_del(&page->lru);
3923 if (add_to_page_cache_lru(page, mapping,
3924 page->index, GFP_NOFS)) {
3925 page_cache_release(page);
3929 pagepool[nr++] = page;
3930 if (nr < ARRAY_SIZE(pagepool))
3932 __extent_readpages(tree, pagepool, nr, get_extent, &em_cached,
3933 &bio, 0, &bio_flags, READ);
3937 __extent_readpages(tree, pagepool, nr, get_extent, &em_cached,
3938 &bio, 0, &bio_flags, READ);
3941 free_extent_map(em_cached);
3943 BUG_ON(!list_empty(pages));
3945 return submit_one_bio(READ, bio, 0, bio_flags);
3950 * basic invalidatepage code, this waits on any locked or writeback
3951 * ranges corresponding to the page, and then deletes any extent state
3952 * records from the tree
3954 int extent_invalidatepage(struct extent_io_tree *tree,
3955 struct page *page, unsigned long offset)
3957 struct extent_state *cached_state = NULL;
3958 u64 start = page_offset(page);
3959 u64 end = start + PAGE_CACHE_SIZE - 1;
3960 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
3962 start += ALIGN(offset, blocksize);
3966 lock_extent_bits(tree, start, end, 0, &cached_state);
3967 wait_on_page_writeback(page);
3968 clear_extent_bit(tree, start, end,
3969 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
3970 EXTENT_DO_ACCOUNTING,
3971 1, 1, &cached_state, GFP_NOFS);
3976 * a helper for releasepage, this tests for areas of the page that
3977 * are locked or under IO and drops the related state bits if it is safe
3980 static int try_release_extent_state(struct extent_map_tree *map,
3981 struct extent_io_tree *tree,
3982 struct page *page, gfp_t mask)
3984 u64 start = page_offset(page);
3985 u64 end = start + PAGE_CACHE_SIZE - 1;
3988 if (test_range_bit(tree, start, end,
3989 EXTENT_IOBITS, 0, NULL))
3992 if ((mask & GFP_NOFS) == GFP_NOFS)
3995 * at this point we can safely clear everything except the
3996 * locked bit and the nodatasum bit
3998 ret = clear_extent_bit(tree, start, end,
3999 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
4002 /* if clear_extent_bit failed for enomem reasons,
4003 * we can't allow the release to continue.
4014 * a helper for releasepage. As long as there are no locked extents
4015 * in the range corresponding to the page, both state records and extent
4016 * map records are removed
4018 int try_release_extent_mapping(struct extent_map_tree *map,
4019 struct extent_io_tree *tree, struct page *page,
4022 struct extent_map *em;
4023 u64 start = page_offset(page);
4024 u64 end = start + PAGE_CACHE_SIZE - 1;
4026 if ((mask & __GFP_WAIT) &&
4027 page->mapping->host->i_size > 16 * 1024 * 1024) {
4029 while (start <= end) {
4030 len = end - start + 1;
4031 write_lock(&map->lock);
4032 em = lookup_extent_mapping(map, start, len);
4034 write_unlock(&map->lock);
4037 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
4038 em->start != start) {
4039 write_unlock(&map->lock);
4040 free_extent_map(em);
4043 if (!test_range_bit(tree, em->start,
4044 extent_map_end(em) - 1,
4045 EXTENT_LOCKED | EXTENT_WRITEBACK,
4047 remove_extent_mapping(map, em);
4048 /* once for the rb tree */
4049 free_extent_map(em);
4051 start = extent_map_end(em);
4052 write_unlock(&map->lock);
4055 free_extent_map(em);
4058 return try_release_extent_state(map, tree, page, mask);
4062 * helper function for fiemap, which doesn't want to see any holes.
4063 * This maps until we find something past 'last'
4065 static struct extent_map *get_extent_skip_holes(struct inode *inode,
4068 get_extent_t *get_extent)
4070 u64 sectorsize = BTRFS_I(inode)->root->sectorsize;
4071 struct extent_map *em;
4078 len = last - offset;
4081 len = ALIGN(len, sectorsize);
4082 em = get_extent(inode, NULL, 0, offset, len, 0);
4083 if (IS_ERR_OR_NULL(em))
4086 /* if this isn't a hole return it */
4087 if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) &&
4088 em->block_start != EXTENT_MAP_HOLE) {
4092 /* this is a hole, advance to the next extent */
4093 offset = extent_map_end(em);
4094 free_extent_map(em);
4101 static noinline int count_ext_ref(u64 inum, u64 offset, u64 root_id, void *ctx)
4103 unsigned long cnt = *((unsigned long *)ctx);
4106 *((unsigned long *)ctx) = cnt;
4108 /* Now we're sure that the extent is shared. */
4114 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4115 __u64 start, __u64 len, get_extent_t *get_extent)
4119 u64 max = start + len;
4123 u64 last_for_get_extent = 0;
4125 u64 isize = i_size_read(inode);
4126 struct btrfs_key found_key;
4127 struct extent_map *em = NULL;
4128 struct extent_state *cached_state = NULL;
4129 struct btrfs_path *path;
4138 path = btrfs_alloc_path();
4141 path->leave_spinning = 1;
4143 start = ALIGN(start, BTRFS_I(inode)->root->sectorsize);
4144 len = ALIGN(len, BTRFS_I(inode)->root->sectorsize);
4147 * lookup the last file extent. We're not using i_size here
4148 * because there might be preallocation past i_size
4150 ret = btrfs_lookup_file_extent(NULL, BTRFS_I(inode)->root,
4151 path, btrfs_ino(inode), -1, 0);
4153 btrfs_free_path(path);
4158 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
4159 found_type = btrfs_key_type(&found_key);
4161 /* No extents, but there might be delalloc bits */
4162 if (found_key.objectid != btrfs_ino(inode) ||
4163 found_type != BTRFS_EXTENT_DATA_KEY) {
4164 /* have to trust i_size as the end */
4166 last_for_get_extent = isize;
4169 * remember the start of the last extent. There are a
4170 * bunch of different factors that go into the length of the
4171 * extent, so its much less complex to remember where it started
4173 last = found_key.offset;
4174 last_for_get_extent = last + 1;
4176 btrfs_release_path(path);
4179 * we might have some extents allocated but more delalloc past those
4180 * extents. so, we trust isize unless the start of the last extent is
4185 last_for_get_extent = isize;
4188 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1, 0,
4191 em = get_extent_skip_holes(inode, start, last_for_get_extent,
4201 u64 offset_in_extent = 0;
4203 /* break if the extent we found is outside the range */
4204 if (em->start >= max || extent_map_end(em) < off)
4208 * get_extent may return an extent that starts before our
4209 * requested range. We have to make sure the ranges
4210 * we return to fiemap always move forward and don't
4211 * overlap, so adjust the offsets here
4213 em_start = max(em->start, off);
4216 * record the offset from the start of the extent
4217 * for adjusting the disk offset below. Only do this if the
4218 * extent isn't compressed since our in ram offset may be past
4219 * what we have actually allocated on disk.
4221 if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4222 offset_in_extent = em_start - em->start;
4223 em_end = extent_map_end(em);
4224 em_len = em_end - em_start;
4229 * bump off for our next call to get_extent
4231 off = extent_map_end(em);
4235 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
4237 flags |= FIEMAP_EXTENT_LAST;
4238 } else if (em->block_start == EXTENT_MAP_INLINE) {
4239 flags |= (FIEMAP_EXTENT_DATA_INLINE |
4240 FIEMAP_EXTENT_NOT_ALIGNED);
4241 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
4242 flags |= (FIEMAP_EXTENT_DELALLOC |
4243 FIEMAP_EXTENT_UNKNOWN);
4245 unsigned long ref_cnt = 0;
4247 disko = em->block_start + offset_in_extent;
4250 * As btrfs supports shared space, this information
4251 * can be exported to userspace tools via
4252 * flag FIEMAP_EXTENT_SHARED.
4254 ret = iterate_inodes_from_logical(
4256 BTRFS_I(inode)->root->fs_info,
4257 path, count_ext_ref, &ref_cnt);
4258 if (ret < 0 && ret != -ENOENT)
4262 flags |= FIEMAP_EXTENT_SHARED;
4264 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4265 flags |= FIEMAP_EXTENT_ENCODED;
4267 free_extent_map(em);
4269 if ((em_start >= last) || em_len == (u64)-1 ||
4270 (last == (u64)-1 && isize <= em_end)) {
4271 flags |= FIEMAP_EXTENT_LAST;
4275 /* now scan forward to see if this is really the last extent. */
4276 em = get_extent_skip_holes(inode, off, last_for_get_extent,
4283 flags |= FIEMAP_EXTENT_LAST;
4286 ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
4292 free_extent_map(em);
4294 btrfs_free_path(path);
4295 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4296 &cached_state, GFP_NOFS);
4300 static void __free_extent_buffer(struct extent_buffer *eb)
4302 btrfs_leak_debug_del(&eb->leak_list);
4303 kmem_cache_free(extent_buffer_cache, eb);
4306 static int extent_buffer_under_io(struct extent_buffer *eb)
4308 return (atomic_read(&eb->io_pages) ||
4309 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4310 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4314 * Helper for releasing extent buffer page.
4316 static void btrfs_release_extent_buffer_page(struct extent_buffer *eb,
4317 unsigned long start_idx)
4319 unsigned long index;
4320 unsigned long num_pages;
4322 int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4324 BUG_ON(extent_buffer_under_io(eb));
4326 num_pages = num_extent_pages(eb->start, eb->len);
4327 index = start_idx + num_pages;
4328 if (start_idx >= index)
4333 page = extent_buffer_page(eb, index);
4334 if (page && mapped) {
4335 spin_lock(&page->mapping->private_lock);
4337 * We do this since we'll remove the pages after we've
4338 * removed the eb from the radix tree, so we could race
4339 * and have this page now attached to the new eb. So
4340 * only clear page_private if it's still connected to
4343 if (PagePrivate(page) &&
4344 page->private == (unsigned long)eb) {
4345 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4346 BUG_ON(PageDirty(page));
4347 BUG_ON(PageWriteback(page));
4349 * We need to make sure we haven't be attached
4352 ClearPagePrivate(page);
4353 set_page_private(page, 0);
4354 /* One for the page private */
4355 page_cache_release(page);
4357 spin_unlock(&page->mapping->private_lock);
4361 /* One for when we alloced the page */
4362 page_cache_release(page);
4364 } while (index != start_idx);
4368 * Helper for releasing the extent buffer.
4370 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4372 btrfs_release_extent_buffer_page(eb, 0);
4373 __free_extent_buffer(eb);
4376 static struct extent_buffer *
4377 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
4378 unsigned long len, gfp_t mask)
4380 struct extent_buffer *eb = NULL;
4382 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
4387 eb->fs_info = fs_info;
4389 rwlock_init(&eb->lock);
4390 atomic_set(&eb->write_locks, 0);
4391 atomic_set(&eb->read_locks, 0);
4392 atomic_set(&eb->blocking_readers, 0);
4393 atomic_set(&eb->blocking_writers, 0);
4394 atomic_set(&eb->spinning_readers, 0);
4395 atomic_set(&eb->spinning_writers, 0);
4396 eb->lock_nested = 0;
4397 init_waitqueue_head(&eb->write_lock_wq);
4398 init_waitqueue_head(&eb->read_lock_wq);
4400 btrfs_leak_debug_add(&eb->leak_list, &buffers);
4402 spin_lock_init(&eb->refs_lock);
4403 atomic_set(&eb->refs, 1);
4404 atomic_set(&eb->io_pages, 0);
4407 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4409 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4410 > MAX_INLINE_EXTENT_BUFFER_SIZE);
4411 BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE);
4416 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src)
4420 struct extent_buffer *new;
4421 unsigned long num_pages = num_extent_pages(src->start, src->len);
4423 new = __alloc_extent_buffer(NULL, src->start, src->len, GFP_NOFS);
4427 for (i = 0; i < num_pages; i++) {
4428 p = alloc_page(GFP_NOFS);
4430 btrfs_release_extent_buffer(new);
4433 attach_extent_buffer_page(new, p);
4434 WARN_ON(PageDirty(p));
4439 copy_extent_buffer(new, src, 0, 0, src->len);
4440 set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
4441 set_bit(EXTENT_BUFFER_DUMMY, &new->bflags);
4446 struct extent_buffer *alloc_dummy_extent_buffer(u64 start, unsigned long len)
4448 struct extent_buffer *eb;
4449 unsigned long num_pages = num_extent_pages(0, len);
4452 eb = __alloc_extent_buffer(NULL, start, len, GFP_NOFS);
4456 for (i = 0; i < num_pages; i++) {
4457 eb->pages[i] = alloc_page(GFP_NOFS);
4461 set_extent_buffer_uptodate(eb);
4462 btrfs_set_header_nritems(eb, 0);
4463 set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4468 __free_page(eb->pages[i - 1]);
4469 __free_extent_buffer(eb);
4473 static void check_buffer_tree_ref(struct extent_buffer *eb)
4476 /* the ref bit is tricky. We have to make sure it is set
4477 * if we have the buffer dirty. Otherwise the
4478 * code to free a buffer can end up dropping a dirty
4481 * Once the ref bit is set, it won't go away while the
4482 * buffer is dirty or in writeback, and it also won't
4483 * go away while we have the reference count on the
4486 * We can't just set the ref bit without bumping the
4487 * ref on the eb because free_extent_buffer might
4488 * see the ref bit and try to clear it. If this happens
4489 * free_extent_buffer might end up dropping our original
4490 * ref by mistake and freeing the page before we are able
4491 * to add one more ref.
4493 * So bump the ref count first, then set the bit. If someone
4494 * beat us to it, drop the ref we added.
4496 refs = atomic_read(&eb->refs);
4497 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4500 spin_lock(&eb->refs_lock);
4501 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4502 atomic_inc(&eb->refs);
4503 spin_unlock(&eb->refs_lock);
4506 static void mark_extent_buffer_accessed(struct extent_buffer *eb)
4508 unsigned long num_pages, i;
4510 check_buffer_tree_ref(eb);
4512 num_pages = num_extent_pages(eb->start, eb->len);
4513 for (i = 0; i < num_pages; i++) {
4514 struct page *p = extent_buffer_page(eb, i);
4515 mark_page_accessed(p);
4519 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
4522 struct extent_buffer *eb;
4525 eb = radix_tree_lookup(&fs_info->buffer_radix,
4526 start >> PAGE_CACHE_SHIFT);
4527 if (eb && atomic_inc_not_zero(&eb->refs)) {
4529 mark_extent_buffer_accessed(eb);
4537 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
4538 u64 start, unsigned long len)
4540 unsigned long num_pages = num_extent_pages(start, len);
4542 unsigned long index = start >> PAGE_CACHE_SHIFT;
4543 struct extent_buffer *eb;
4544 struct extent_buffer *exists = NULL;
4546 struct address_space *mapping = fs_info->btree_inode->i_mapping;
4550 eb = find_extent_buffer(fs_info, start);
4554 eb = __alloc_extent_buffer(fs_info, start, len, GFP_NOFS);
4558 for (i = 0; i < num_pages; i++, index++) {
4559 p = find_or_create_page(mapping, index, GFP_NOFS);
4563 spin_lock(&mapping->private_lock);
4564 if (PagePrivate(p)) {
4566 * We could have already allocated an eb for this page
4567 * and attached one so lets see if we can get a ref on
4568 * the existing eb, and if we can we know it's good and
4569 * we can just return that one, else we know we can just
4570 * overwrite page->private.
4572 exists = (struct extent_buffer *)p->private;
4573 if (atomic_inc_not_zero(&exists->refs)) {
4574 spin_unlock(&mapping->private_lock);
4576 page_cache_release(p);
4577 mark_extent_buffer_accessed(exists);
4582 * Do this so attach doesn't complain and we need to
4583 * drop the ref the old guy had.
4585 ClearPagePrivate(p);
4586 WARN_ON(PageDirty(p));
4587 page_cache_release(p);
4589 attach_extent_buffer_page(eb, p);
4590 spin_unlock(&mapping->private_lock);
4591 WARN_ON(PageDirty(p));
4592 mark_page_accessed(p);
4594 if (!PageUptodate(p))
4598 * see below about how we avoid a nasty race with release page
4599 * and why we unlock later
4603 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4605 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
4609 spin_lock(&fs_info->buffer_lock);
4610 ret = radix_tree_insert(&fs_info->buffer_radix,
4611 start >> PAGE_CACHE_SHIFT, eb);
4612 spin_unlock(&fs_info->buffer_lock);
4613 radix_tree_preload_end();
4614 if (ret == -EEXIST) {
4615 exists = find_extent_buffer(fs_info, start);
4621 /* add one reference for the tree */
4622 check_buffer_tree_ref(eb);
4623 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
4626 * there is a race where release page may have
4627 * tried to find this extent buffer in the radix
4628 * but failed. It will tell the VM it is safe to
4629 * reclaim the, and it will clear the page private bit.
4630 * We must make sure to set the page private bit properly
4631 * after the extent buffer is in the radix tree so
4632 * it doesn't get lost
4634 SetPageChecked(eb->pages[0]);
4635 for (i = 1; i < num_pages; i++) {
4636 p = extent_buffer_page(eb, i);
4637 ClearPageChecked(p);
4640 unlock_page(eb->pages[0]);
4644 for (i = 0; i < num_pages; i++) {
4646 unlock_page(eb->pages[i]);
4649 WARN_ON(!atomic_dec_and_test(&eb->refs));
4650 btrfs_release_extent_buffer(eb);
4654 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
4656 struct extent_buffer *eb =
4657 container_of(head, struct extent_buffer, rcu_head);
4659 __free_extent_buffer(eb);
4662 /* Expects to have eb->eb_lock already held */
4663 static int release_extent_buffer(struct extent_buffer *eb)
4665 WARN_ON(atomic_read(&eb->refs) == 0);
4666 if (atomic_dec_and_test(&eb->refs)) {
4667 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
4668 struct btrfs_fs_info *fs_info = eb->fs_info;
4670 spin_unlock(&eb->refs_lock);
4672 spin_lock(&fs_info->buffer_lock);
4673 radix_tree_delete(&fs_info->buffer_radix,
4674 eb->start >> PAGE_CACHE_SHIFT);
4675 spin_unlock(&fs_info->buffer_lock);
4677 spin_unlock(&eb->refs_lock);
4680 /* Should be safe to release our pages at this point */
4681 btrfs_release_extent_buffer_page(eb, 0);
4682 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
4685 spin_unlock(&eb->refs_lock);
4690 void free_extent_buffer(struct extent_buffer *eb)
4698 refs = atomic_read(&eb->refs);
4701 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
4706 spin_lock(&eb->refs_lock);
4707 if (atomic_read(&eb->refs) == 2 &&
4708 test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))
4709 atomic_dec(&eb->refs);
4711 if (atomic_read(&eb->refs) == 2 &&
4712 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
4713 !extent_buffer_under_io(eb) &&
4714 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4715 atomic_dec(&eb->refs);
4718 * I know this is terrible, but it's temporary until we stop tracking
4719 * the uptodate bits and such for the extent buffers.
4721 release_extent_buffer(eb);
4724 void free_extent_buffer_stale(struct extent_buffer *eb)
4729 spin_lock(&eb->refs_lock);
4730 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
4732 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
4733 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4734 atomic_dec(&eb->refs);
4735 release_extent_buffer(eb);
4738 void clear_extent_buffer_dirty(struct extent_buffer *eb)
4741 unsigned long num_pages;
4744 num_pages = num_extent_pages(eb->start, eb->len);
4746 for (i = 0; i < num_pages; i++) {
4747 page = extent_buffer_page(eb, i);
4748 if (!PageDirty(page))
4752 WARN_ON(!PagePrivate(page));
4754 clear_page_dirty_for_io(page);
4755 spin_lock_irq(&page->mapping->tree_lock);
4756 if (!PageDirty(page)) {
4757 radix_tree_tag_clear(&page->mapping->page_tree,
4759 PAGECACHE_TAG_DIRTY);
4761 spin_unlock_irq(&page->mapping->tree_lock);
4762 ClearPageError(page);
4765 WARN_ON(atomic_read(&eb->refs) == 0);
4768 int set_extent_buffer_dirty(struct extent_buffer *eb)
4771 unsigned long num_pages;
4774 check_buffer_tree_ref(eb);
4776 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
4778 num_pages = num_extent_pages(eb->start, eb->len);
4779 WARN_ON(atomic_read(&eb->refs) == 0);
4780 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
4782 for (i = 0; i < num_pages; i++)
4783 set_page_dirty(extent_buffer_page(eb, i));
4787 int clear_extent_buffer_uptodate(struct extent_buffer *eb)
4791 unsigned long num_pages;
4793 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4794 num_pages = num_extent_pages(eb->start, eb->len);
4795 for (i = 0; i < num_pages; i++) {
4796 page = extent_buffer_page(eb, i);
4798 ClearPageUptodate(page);
4803 int set_extent_buffer_uptodate(struct extent_buffer *eb)
4807 unsigned long num_pages;
4809 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4810 num_pages = num_extent_pages(eb->start, eb->len);
4811 for (i = 0; i < num_pages; i++) {
4812 page = extent_buffer_page(eb, i);
4813 SetPageUptodate(page);
4818 int extent_buffer_uptodate(struct extent_buffer *eb)
4820 return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4823 int read_extent_buffer_pages(struct extent_io_tree *tree,
4824 struct extent_buffer *eb, u64 start, int wait,
4825 get_extent_t *get_extent, int mirror_num)
4828 unsigned long start_i;
4832 int locked_pages = 0;
4833 int all_uptodate = 1;
4834 unsigned long num_pages;
4835 unsigned long num_reads = 0;
4836 struct bio *bio = NULL;
4837 unsigned long bio_flags = 0;
4839 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
4843 WARN_ON(start < eb->start);
4844 start_i = (start >> PAGE_CACHE_SHIFT) -
4845 (eb->start >> PAGE_CACHE_SHIFT);
4850 num_pages = num_extent_pages(eb->start, eb->len);
4851 for (i = start_i; i < num_pages; i++) {
4852 page = extent_buffer_page(eb, i);
4853 if (wait == WAIT_NONE) {
4854 if (!trylock_page(page))
4860 if (!PageUptodate(page)) {
4867 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4871 clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
4872 eb->read_mirror = 0;
4873 atomic_set(&eb->io_pages, num_reads);
4874 for (i = start_i; i < num_pages; i++) {
4875 page = extent_buffer_page(eb, i);
4876 if (!PageUptodate(page)) {
4877 ClearPageError(page);
4878 err = __extent_read_full_page(tree, page,
4880 mirror_num, &bio_flags,
4890 err = submit_one_bio(READ | REQ_META, bio, mirror_num,
4896 if (ret || wait != WAIT_COMPLETE)
4899 for (i = start_i; i < num_pages; i++) {
4900 page = extent_buffer_page(eb, i);
4901 wait_on_page_locked(page);
4902 if (!PageUptodate(page))
4910 while (locked_pages > 0) {
4911 page = extent_buffer_page(eb, i);
4919 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
4920 unsigned long start,
4927 char *dst = (char *)dstv;
4928 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4929 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4931 WARN_ON(start > eb->len);
4932 WARN_ON(start + len > eb->start + eb->len);
4934 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
4937 page = extent_buffer_page(eb, i);
4939 cur = min(len, (PAGE_CACHE_SIZE - offset));
4940 kaddr = page_address(page);
4941 memcpy(dst, kaddr + offset, cur);
4950 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
4951 unsigned long min_len, char **map,
4952 unsigned long *map_start,
4953 unsigned long *map_len)
4955 size_t offset = start & (PAGE_CACHE_SIZE - 1);
4958 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4959 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4960 unsigned long end_i = (start_offset + start + min_len - 1) >>
4967 offset = start_offset;
4971 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
4974 if (start + min_len > eb->len) {
4975 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, "
4977 eb->start, eb->len, start, min_len);
4981 p = extent_buffer_page(eb, i);
4982 kaddr = page_address(p);
4983 *map = kaddr + offset;
4984 *map_len = PAGE_CACHE_SIZE - offset;
4988 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
4989 unsigned long start,
4996 char *ptr = (char *)ptrv;
4997 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4998 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5001 WARN_ON(start > eb->len);
5002 WARN_ON(start + len > eb->start + eb->len);
5004 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5007 page = extent_buffer_page(eb, i);
5009 cur = min(len, (PAGE_CACHE_SIZE - offset));
5011 kaddr = page_address(page);
5012 ret = memcmp(ptr, kaddr + offset, cur);
5024 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
5025 unsigned long start, unsigned long len)
5031 char *src = (char *)srcv;
5032 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5033 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5035 WARN_ON(start > eb->len);
5036 WARN_ON(start + len > eb->start + eb->len);
5038 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5041 page = extent_buffer_page(eb, i);
5042 WARN_ON(!PageUptodate(page));
5044 cur = min(len, PAGE_CACHE_SIZE - offset);
5045 kaddr = page_address(page);
5046 memcpy(kaddr + offset, src, cur);
5055 void memset_extent_buffer(struct extent_buffer *eb, char c,
5056 unsigned long start, unsigned long len)
5062 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5063 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5065 WARN_ON(start > eb->len);
5066 WARN_ON(start + len > eb->start + eb->len);
5068 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5071 page = extent_buffer_page(eb, i);
5072 WARN_ON(!PageUptodate(page));
5074 cur = min(len, PAGE_CACHE_SIZE - offset);
5075 kaddr = page_address(page);
5076 memset(kaddr + offset, c, cur);
5084 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
5085 unsigned long dst_offset, unsigned long src_offset,
5088 u64 dst_len = dst->len;
5093 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5094 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
5096 WARN_ON(src->len != dst_len);
5098 offset = (start_offset + dst_offset) &
5099 (PAGE_CACHE_SIZE - 1);
5102 page = extent_buffer_page(dst, i);
5103 WARN_ON(!PageUptodate(page));
5105 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
5107 kaddr = page_address(page);
5108 read_extent_buffer(src, kaddr + offset, src_offset, cur);
5117 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
5119 unsigned long distance = (src > dst) ? src - dst : dst - src;
5120 return distance < len;
5123 static void copy_pages(struct page *dst_page, struct page *src_page,
5124 unsigned long dst_off, unsigned long src_off,
5127 char *dst_kaddr = page_address(dst_page);
5129 int must_memmove = 0;
5131 if (dst_page != src_page) {
5132 src_kaddr = page_address(src_page);
5134 src_kaddr = dst_kaddr;
5135 if (areas_overlap(src_off, dst_off, len))
5140 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
5142 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
5145 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5146 unsigned long src_offset, unsigned long len)
5149 size_t dst_off_in_page;
5150 size_t src_off_in_page;
5151 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5152 unsigned long dst_i;
5153 unsigned long src_i;
5155 if (src_offset + len > dst->len) {
5156 printk(KERN_ERR "BTRFS: memmove bogus src_offset %lu move "
5157 "len %lu dst len %lu\n", src_offset, len, dst->len);
5160 if (dst_offset + len > dst->len) {
5161 printk(KERN_ERR "BTRFS: memmove bogus dst_offset %lu move "
5162 "len %lu dst len %lu\n", dst_offset, len, dst->len);
5167 dst_off_in_page = (start_offset + dst_offset) &
5168 (PAGE_CACHE_SIZE - 1);
5169 src_off_in_page = (start_offset + src_offset) &
5170 (PAGE_CACHE_SIZE - 1);
5172 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
5173 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
5175 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
5177 cur = min_t(unsigned long, cur,
5178 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
5180 copy_pages(extent_buffer_page(dst, dst_i),
5181 extent_buffer_page(dst, src_i),
5182 dst_off_in_page, src_off_in_page, cur);
5190 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5191 unsigned long src_offset, unsigned long len)
5194 size_t dst_off_in_page;
5195 size_t src_off_in_page;
5196 unsigned long dst_end = dst_offset + len - 1;
5197 unsigned long src_end = src_offset + len - 1;
5198 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5199 unsigned long dst_i;
5200 unsigned long src_i;
5202 if (src_offset + len > dst->len) {
5203 printk(KERN_ERR "BTRFS: memmove bogus src_offset %lu move "
5204 "len %lu len %lu\n", src_offset, len, dst->len);
5207 if (dst_offset + len > dst->len) {
5208 printk(KERN_ERR "BTRFS: memmove bogus dst_offset %lu move "
5209 "len %lu len %lu\n", dst_offset, len, dst->len);
5212 if (dst_offset < src_offset) {
5213 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
5217 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
5218 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
5220 dst_off_in_page = (start_offset + dst_end) &
5221 (PAGE_CACHE_SIZE - 1);
5222 src_off_in_page = (start_offset + src_end) &
5223 (PAGE_CACHE_SIZE - 1);
5225 cur = min_t(unsigned long, len, src_off_in_page + 1);
5226 cur = min(cur, dst_off_in_page + 1);
5227 copy_pages(extent_buffer_page(dst, dst_i),
5228 extent_buffer_page(dst, src_i),
5229 dst_off_in_page - cur + 1,
5230 src_off_in_page - cur + 1, cur);
5238 int try_release_extent_buffer(struct page *page)
5240 struct extent_buffer *eb;
5243 * We need to make sure noboody is attaching this page to an eb right
5246 spin_lock(&page->mapping->private_lock);
5247 if (!PagePrivate(page)) {
5248 spin_unlock(&page->mapping->private_lock);
5252 eb = (struct extent_buffer *)page->private;
5256 * This is a little awful but should be ok, we need to make sure that
5257 * the eb doesn't disappear out from under us while we're looking at
5260 spin_lock(&eb->refs_lock);
5261 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
5262 spin_unlock(&eb->refs_lock);
5263 spin_unlock(&page->mapping->private_lock);
5266 spin_unlock(&page->mapping->private_lock);
5269 * If tree ref isn't set then we know the ref on this eb is a real ref,
5270 * so just return, this page will likely be freed soon anyway.
5272 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
5273 spin_unlock(&eb->refs_lock);
5277 return release_extent_buffer(eb);