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,
233 struct rb_node *search_start,
235 struct rb_node *node,
236 struct rb_node ***p_in,
237 struct rb_node **parent_in)
240 struct rb_node *parent = NULL;
241 struct tree_entry *entry;
243 if (p_in && parent_in) {
249 p = search_start ? &search_start : &root->rb_node;
252 entry = rb_entry(parent, struct tree_entry, rb_node);
254 if (offset < entry->start)
256 else if (offset > entry->end)
263 rb_link_node(node, parent, p);
264 rb_insert_color(node, root);
268 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
269 struct rb_node **prev_ret,
270 struct rb_node **next_ret,
271 struct rb_node ***p_ret,
272 struct rb_node **parent_ret)
274 struct rb_root *root = &tree->state;
275 struct rb_node **n = &root->rb_node;
276 struct rb_node *prev = NULL;
277 struct rb_node *orig_prev = NULL;
278 struct tree_entry *entry;
279 struct tree_entry *prev_entry = NULL;
283 entry = rb_entry(prev, struct tree_entry, rb_node);
286 if (offset < entry->start)
288 else if (offset > entry->end)
301 while (prev && offset > prev_entry->end) {
302 prev = rb_next(prev);
303 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
310 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
311 while (prev && offset < prev_entry->start) {
312 prev = rb_prev(prev);
313 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
320 static inline struct rb_node *
321 tree_search_for_insert(struct extent_io_tree *tree,
323 struct rb_node ***p_ret,
324 struct rb_node **parent_ret)
326 struct rb_node *prev = NULL;
329 ret = __etree_search(tree, offset, &prev, NULL, p_ret, parent_ret);
335 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
338 return tree_search_for_insert(tree, offset, NULL, NULL);
341 static void merge_cb(struct extent_io_tree *tree, struct extent_state *new,
342 struct extent_state *other)
344 if (tree->ops && tree->ops->merge_extent_hook)
345 tree->ops->merge_extent_hook(tree->mapping->host, new,
350 * utility function to look for merge candidates inside a given range.
351 * Any extents with matching state are merged together into a single
352 * extent in the tree. Extents with EXTENT_IO in their state field
353 * are not merged because the end_io handlers need to be able to do
354 * operations on them without sleeping (or doing allocations/splits).
356 * This should be called with the tree lock held.
358 static void merge_state(struct extent_io_tree *tree,
359 struct extent_state *state)
361 struct extent_state *other;
362 struct rb_node *other_node;
364 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
367 other_node = rb_prev(&state->rb_node);
369 other = rb_entry(other_node, struct extent_state, rb_node);
370 if (other->end == state->start - 1 &&
371 other->state == state->state) {
372 merge_cb(tree, state, other);
373 state->start = other->start;
375 rb_erase(&other->rb_node, &tree->state);
376 free_extent_state(other);
379 other_node = rb_next(&state->rb_node);
381 other = rb_entry(other_node, struct extent_state, rb_node);
382 if (other->start == state->end + 1 &&
383 other->state == state->state) {
384 merge_cb(tree, state, other);
385 state->end = other->end;
387 rb_erase(&other->rb_node, &tree->state);
388 free_extent_state(other);
393 static void set_state_cb(struct extent_io_tree *tree,
394 struct extent_state *state, unsigned long *bits)
396 if (tree->ops && tree->ops->set_bit_hook)
397 tree->ops->set_bit_hook(tree->mapping->host, state, bits);
400 static void clear_state_cb(struct extent_io_tree *tree,
401 struct extent_state *state, unsigned long *bits)
403 if (tree->ops && tree->ops->clear_bit_hook)
404 tree->ops->clear_bit_hook(tree->mapping->host, state, bits);
407 static void set_state_bits(struct extent_io_tree *tree,
408 struct extent_state *state, unsigned long *bits);
411 * insert an extent_state struct into the tree. 'bits' are set on the
412 * struct before it is inserted.
414 * This may return -EEXIST if the extent is already there, in which case the
415 * state struct is freed.
417 * The tree lock is not taken internally. This is a utility function and
418 * probably isn't what you want to call (see set/clear_extent_bit).
420 static int insert_state(struct extent_io_tree *tree,
421 struct extent_state *state, u64 start, u64 end,
423 struct rb_node **parent,
426 struct rb_node *node;
429 WARN(1, KERN_ERR "BTRFS: end < start %llu %llu\n",
431 state->start = start;
434 set_state_bits(tree, state, bits);
436 node = tree_insert(&tree->state, NULL, end, &state->rb_node, p, parent);
438 struct extent_state *found;
439 found = rb_entry(node, struct extent_state, rb_node);
440 printk(KERN_ERR "BTRFS: found node %llu %llu on insert of "
442 found->start, found->end, start, end);
446 merge_state(tree, state);
450 static void split_cb(struct extent_io_tree *tree, struct extent_state *orig,
453 if (tree->ops && tree->ops->split_extent_hook)
454 tree->ops->split_extent_hook(tree->mapping->host, orig, split);
458 * split a given extent state struct in two, inserting the preallocated
459 * struct 'prealloc' as the newly created second half. 'split' indicates an
460 * offset inside 'orig' where it should be split.
463 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
464 * are two extent state structs in the tree:
465 * prealloc: [orig->start, split - 1]
466 * orig: [ split, orig->end ]
468 * The tree locks are not taken by this function. They need to be held
471 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
472 struct extent_state *prealloc, u64 split)
474 struct rb_node *node;
476 split_cb(tree, orig, split);
478 prealloc->start = orig->start;
479 prealloc->end = split - 1;
480 prealloc->state = orig->state;
483 node = tree_insert(&tree->state, &orig->rb_node, prealloc->end,
484 &prealloc->rb_node, NULL, NULL);
486 free_extent_state(prealloc);
489 prealloc->tree = tree;
493 static struct extent_state *next_state(struct extent_state *state)
495 struct rb_node *next = rb_next(&state->rb_node);
497 return rb_entry(next, struct extent_state, rb_node);
503 * utility function to clear some bits in an extent state struct.
504 * it will optionally wake up any one waiting on this state (wake == 1).
506 * If no bits are set on the state struct after clearing things, the
507 * struct is freed and removed from the tree
509 static struct extent_state *clear_state_bit(struct extent_io_tree *tree,
510 struct extent_state *state,
511 unsigned long *bits, int wake)
513 struct extent_state *next;
514 unsigned long bits_to_clear = *bits & ~EXTENT_CTLBITS;
516 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
517 u64 range = state->end - state->start + 1;
518 WARN_ON(range > tree->dirty_bytes);
519 tree->dirty_bytes -= range;
521 clear_state_cb(tree, state, bits);
522 state->state &= ~bits_to_clear;
525 if (state->state == 0) {
526 next = next_state(state);
528 rb_erase(&state->rb_node, &tree->state);
530 free_extent_state(state);
535 merge_state(tree, state);
536 next = next_state(state);
541 static struct extent_state *
542 alloc_extent_state_atomic(struct extent_state *prealloc)
545 prealloc = alloc_extent_state(GFP_ATOMIC);
550 static void extent_io_tree_panic(struct extent_io_tree *tree, int err)
552 btrfs_panic(tree_fs_info(tree), err, "Locking error: "
553 "Extent tree was modified by another "
554 "thread while locked.");
558 * clear some bits on a range in the tree. This may require splitting
559 * or inserting elements in the tree, so the gfp mask is used to
560 * indicate which allocations or sleeping are allowed.
562 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
563 * the given range from the tree regardless of state (ie for truncate).
565 * the range [start, end] is inclusive.
567 * This takes the tree lock, and returns 0 on success and < 0 on error.
569 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
570 unsigned long bits, int wake, int delete,
571 struct extent_state **cached_state,
574 struct extent_state *state;
575 struct extent_state *cached;
576 struct extent_state *prealloc = NULL;
577 struct rb_node *node;
582 btrfs_debug_check_extent_io_range(tree, start, end);
584 if (bits & EXTENT_DELALLOC)
585 bits |= EXTENT_NORESERVE;
588 bits |= ~EXTENT_CTLBITS;
589 bits |= EXTENT_FIRST_DELALLOC;
591 if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY))
594 if (!prealloc && (mask & __GFP_WAIT)) {
595 prealloc = alloc_extent_state(mask);
600 spin_lock(&tree->lock);
602 cached = *cached_state;
605 *cached_state = NULL;
609 if (cached && cached->tree && cached->start <= start &&
610 cached->end > start) {
612 atomic_dec(&cached->refs);
617 free_extent_state(cached);
620 * this search will find the extents that end after
623 node = tree_search(tree, start);
626 state = rb_entry(node, struct extent_state, rb_node);
628 if (state->start > end)
630 WARN_ON(state->end < start);
631 last_end = state->end;
633 /* the state doesn't have the wanted bits, go ahead */
634 if (!(state->state & bits)) {
635 state = next_state(state);
640 * | ---- desired range ---- |
642 * | ------------- state -------------- |
644 * We need to split the extent we found, and may flip
645 * bits on second half.
647 * If the extent we found extends past our range, we
648 * just split and search again. It'll get split again
649 * the next time though.
651 * If the extent we found is inside our range, we clear
652 * the desired bit on it.
655 if (state->start < start) {
656 prealloc = alloc_extent_state_atomic(prealloc);
658 err = split_state(tree, state, prealloc, start);
660 extent_io_tree_panic(tree, err);
665 if (state->end <= end) {
666 state = clear_state_bit(tree, state, &bits, wake);
672 * | ---- desired range ---- |
674 * We need to split the extent, and clear the bit
677 if (state->start <= end && state->end > end) {
678 prealloc = alloc_extent_state_atomic(prealloc);
680 err = split_state(tree, state, prealloc, end + 1);
682 extent_io_tree_panic(tree, err);
687 clear_state_bit(tree, prealloc, &bits, wake);
693 state = clear_state_bit(tree, state, &bits, wake);
695 if (last_end == (u64)-1)
697 start = last_end + 1;
698 if (start <= end && state && !need_resched())
703 spin_unlock(&tree->lock);
705 free_extent_state(prealloc);
712 spin_unlock(&tree->lock);
713 if (mask & __GFP_WAIT)
718 static void wait_on_state(struct extent_io_tree *tree,
719 struct extent_state *state)
720 __releases(tree->lock)
721 __acquires(tree->lock)
724 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
725 spin_unlock(&tree->lock);
727 spin_lock(&tree->lock);
728 finish_wait(&state->wq, &wait);
732 * waits for one or more bits to clear on a range in the state tree.
733 * The range [start, end] is inclusive.
734 * The tree lock is taken by this function
736 static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
739 struct extent_state *state;
740 struct rb_node *node;
742 btrfs_debug_check_extent_io_range(tree, start, end);
744 spin_lock(&tree->lock);
748 * this search will find all the extents that end after
751 node = tree_search(tree, start);
756 state = rb_entry(node, struct extent_state, rb_node);
758 if (state->start > end)
761 if (state->state & bits) {
762 start = state->start;
763 atomic_inc(&state->refs);
764 wait_on_state(tree, state);
765 free_extent_state(state);
768 start = state->end + 1;
773 if (!cond_resched_lock(&tree->lock)) {
774 node = rb_next(node);
779 spin_unlock(&tree->lock);
782 static void set_state_bits(struct extent_io_tree *tree,
783 struct extent_state *state,
786 unsigned long bits_to_set = *bits & ~EXTENT_CTLBITS;
788 set_state_cb(tree, state, bits);
789 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
790 u64 range = state->end - state->start + 1;
791 tree->dirty_bytes += range;
793 state->state |= bits_to_set;
796 static void cache_state(struct extent_state *state,
797 struct extent_state **cached_ptr)
799 if (cached_ptr && !(*cached_ptr)) {
800 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) {
802 atomic_inc(&state->refs);
808 * set some bits on a range in the tree. This may require allocations or
809 * sleeping, so the gfp mask is used to indicate what is allowed.
811 * If any of the exclusive bits are set, this will fail with -EEXIST if some
812 * part of the range already has the desired bits set. The start of the
813 * existing range is returned in failed_start in this case.
815 * [start, end] is inclusive This takes the tree lock.
818 static int __must_check
819 __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
820 unsigned long bits, unsigned long exclusive_bits,
821 u64 *failed_start, struct extent_state **cached_state,
824 struct extent_state *state;
825 struct extent_state *prealloc = NULL;
826 struct rb_node *node;
828 struct rb_node *parent;
833 btrfs_debug_check_extent_io_range(tree, start, end);
835 bits |= EXTENT_FIRST_DELALLOC;
837 if (!prealloc && (mask & __GFP_WAIT)) {
838 prealloc = alloc_extent_state(mask);
842 spin_lock(&tree->lock);
843 if (cached_state && *cached_state) {
844 state = *cached_state;
845 if (state->start <= start && state->end > start &&
847 node = &state->rb_node;
852 * this search will find all the extents that end after
855 node = tree_search_for_insert(tree, start, &p, &parent);
857 prealloc = alloc_extent_state_atomic(prealloc);
859 err = insert_state(tree, prealloc, start, end,
862 extent_io_tree_panic(tree, err);
864 cache_state(prealloc, cached_state);
868 state = rb_entry(node, struct extent_state, rb_node);
870 last_start = state->start;
871 last_end = state->end;
874 * | ---- desired range ---- |
877 * Just lock what we found and keep going
879 if (state->start == start && state->end <= end) {
880 if (state->state & exclusive_bits) {
881 *failed_start = state->start;
886 set_state_bits(tree, state, &bits);
887 cache_state(state, cached_state);
888 merge_state(tree, state);
889 if (last_end == (u64)-1)
891 start = last_end + 1;
892 state = next_state(state);
893 if (start < end && state && state->start == start &&
900 * | ---- desired range ---- |
903 * | ------------- state -------------- |
905 * We need to split the extent we found, and may flip bits on
908 * If the extent we found extends past our
909 * range, we just split and search again. It'll get split
910 * again the next time though.
912 * If the extent we found is inside our range, we set the
915 if (state->start < start) {
916 if (state->state & exclusive_bits) {
917 *failed_start = start;
922 prealloc = alloc_extent_state_atomic(prealloc);
924 err = split_state(tree, state, prealloc, start);
926 extent_io_tree_panic(tree, err);
931 if (state->end <= end) {
932 set_state_bits(tree, state, &bits);
933 cache_state(state, cached_state);
934 merge_state(tree, state);
935 if (last_end == (u64)-1)
937 start = last_end + 1;
938 state = next_state(state);
939 if (start < end && state && state->start == start &&
946 * | ---- desired range ---- |
947 * | state | or | state |
949 * There's a hole, we need to insert something in it and
950 * ignore the extent we found.
952 if (state->start > start) {
954 if (end < last_start)
957 this_end = last_start - 1;
959 prealloc = alloc_extent_state_atomic(prealloc);
963 * Avoid to free 'prealloc' if it can be merged with
966 err = insert_state(tree, prealloc, start, this_end,
969 extent_io_tree_panic(tree, err);
971 cache_state(prealloc, cached_state);
973 start = this_end + 1;
977 * | ---- desired range ---- |
979 * We need to split the extent, and set the bit
982 if (state->start <= end && state->end > end) {
983 if (state->state & exclusive_bits) {
984 *failed_start = start;
989 prealloc = alloc_extent_state_atomic(prealloc);
991 err = split_state(tree, state, prealloc, end + 1);
993 extent_io_tree_panic(tree, err);
995 set_state_bits(tree, prealloc, &bits);
996 cache_state(prealloc, cached_state);
997 merge_state(tree, prealloc);
1005 spin_unlock(&tree->lock);
1007 free_extent_state(prealloc);
1014 spin_unlock(&tree->lock);
1015 if (mask & __GFP_WAIT)
1020 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1021 unsigned long bits, u64 * failed_start,
1022 struct extent_state **cached_state, gfp_t mask)
1024 return __set_extent_bit(tree, start, end, bits, 0, failed_start,
1025 cached_state, mask);
1030 * convert_extent_bit - convert all bits in a given range from one bit to
1032 * @tree: the io tree to search
1033 * @start: the start offset in bytes
1034 * @end: the end offset in bytes (inclusive)
1035 * @bits: the bits to set in this range
1036 * @clear_bits: the bits to clear in this range
1037 * @cached_state: state that we're going to cache
1038 * @mask: the allocation mask
1040 * This will go through and set bits for the given range. If any states exist
1041 * already in this range they are set with the given bit and cleared of the
1042 * clear_bits. This is only meant to be used by things that are mergeable, ie
1043 * converting from say DELALLOC to DIRTY. This is not meant to be used with
1044 * boundary bits like LOCK.
1046 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1047 unsigned long bits, unsigned long clear_bits,
1048 struct extent_state **cached_state, gfp_t mask)
1050 struct extent_state *state;
1051 struct extent_state *prealloc = NULL;
1052 struct rb_node *node;
1054 struct rb_node *parent;
1059 btrfs_debug_check_extent_io_range(tree, start, end);
1062 if (!prealloc && (mask & __GFP_WAIT)) {
1063 prealloc = alloc_extent_state(mask);
1068 spin_lock(&tree->lock);
1069 if (cached_state && *cached_state) {
1070 state = *cached_state;
1071 if (state->start <= start && state->end > start &&
1073 node = &state->rb_node;
1079 * this search will find all the extents that end after
1082 node = tree_search_for_insert(tree, start, &p, &parent);
1084 prealloc = alloc_extent_state_atomic(prealloc);
1089 err = insert_state(tree, prealloc, start, end,
1090 &p, &parent, &bits);
1092 extent_io_tree_panic(tree, err);
1093 cache_state(prealloc, cached_state);
1097 state = rb_entry(node, struct extent_state, rb_node);
1099 last_start = state->start;
1100 last_end = state->end;
1103 * | ---- desired range ---- |
1106 * Just lock what we found and keep going
1108 if (state->start == start && state->end <= end) {
1109 set_state_bits(tree, state, &bits);
1110 cache_state(state, cached_state);
1111 state = clear_state_bit(tree, state, &clear_bits, 0);
1112 if (last_end == (u64)-1)
1114 start = last_end + 1;
1115 if (start < end && state && state->start == start &&
1122 * | ---- desired range ---- |
1125 * | ------------- state -------------- |
1127 * We need to split the extent we found, and may flip bits on
1130 * If the extent we found extends past our
1131 * range, we just split and search again. It'll get split
1132 * again the next time though.
1134 * If the extent we found is inside our range, we set the
1135 * desired bit on it.
1137 if (state->start < start) {
1138 prealloc = alloc_extent_state_atomic(prealloc);
1143 err = split_state(tree, state, prealloc, start);
1145 extent_io_tree_panic(tree, err);
1149 if (state->end <= end) {
1150 set_state_bits(tree, state, &bits);
1151 cache_state(state, cached_state);
1152 state = clear_state_bit(tree, state, &clear_bits, 0);
1153 if (last_end == (u64)-1)
1155 start = last_end + 1;
1156 if (start < end && state && state->start == start &&
1163 * | ---- desired range ---- |
1164 * | state | or | state |
1166 * There's a hole, we need to insert something in it and
1167 * ignore the extent we found.
1169 if (state->start > start) {
1171 if (end < last_start)
1174 this_end = last_start - 1;
1176 prealloc = alloc_extent_state_atomic(prealloc);
1183 * Avoid to free 'prealloc' if it can be merged with
1186 err = insert_state(tree, prealloc, start, this_end,
1189 extent_io_tree_panic(tree, err);
1190 cache_state(prealloc, cached_state);
1192 start = this_end + 1;
1196 * | ---- desired range ---- |
1198 * We need to split the extent, and set the bit
1201 if (state->start <= end && state->end > end) {
1202 prealloc = alloc_extent_state_atomic(prealloc);
1208 err = split_state(tree, state, prealloc, end + 1);
1210 extent_io_tree_panic(tree, err);
1212 set_state_bits(tree, prealloc, &bits);
1213 cache_state(prealloc, cached_state);
1214 clear_state_bit(tree, prealloc, &clear_bits, 0);
1222 spin_unlock(&tree->lock);
1224 free_extent_state(prealloc);
1231 spin_unlock(&tree->lock);
1232 if (mask & __GFP_WAIT)
1237 /* wrappers around set/clear extent bit */
1238 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1241 return set_extent_bit(tree, start, end, EXTENT_DIRTY, NULL,
1245 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1246 unsigned long bits, gfp_t mask)
1248 return set_extent_bit(tree, start, end, bits, NULL,
1252 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1253 unsigned long bits, gfp_t mask)
1255 return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask);
1258 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
1259 struct extent_state **cached_state, gfp_t mask)
1261 return set_extent_bit(tree, start, end,
1262 EXTENT_DELALLOC | EXTENT_UPTODATE,
1263 NULL, cached_state, mask);
1266 int set_extent_defrag(struct extent_io_tree *tree, u64 start, u64 end,
1267 struct extent_state **cached_state, gfp_t mask)
1269 return set_extent_bit(tree, start, end,
1270 EXTENT_DELALLOC | EXTENT_UPTODATE | EXTENT_DEFRAG,
1271 NULL, cached_state, mask);
1274 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1277 return clear_extent_bit(tree, start, end,
1278 EXTENT_DIRTY | EXTENT_DELALLOC |
1279 EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask);
1282 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
1285 return set_extent_bit(tree, start, end, EXTENT_NEW, NULL,
1289 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
1290 struct extent_state **cached_state, gfp_t mask)
1292 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, NULL,
1293 cached_state, mask);
1296 int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
1297 struct extent_state **cached_state, gfp_t mask)
1299 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0,
1300 cached_state, mask);
1304 * either insert or lock state struct between start and end use mask to tell
1305 * us if waiting is desired.
1307 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1308 unsigned long bits, struct extent_state **cached_state)
1313 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED | bits,
1314 EXTENT_LOCKED, &failed_start,
1315 cached_state, GFP_NOFS);
1316 if (err == -EEXIST) {
1317 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1318 start = failed_start;
1321 WARN_ON(start > end);
1326 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1328 return lock_extent_bits(tree, start, end, 0, NULL);
1331 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1336 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1337 &failed_start, NULL, GFP_NOFS);
1338 if (err == -EEXIST) {
1339 if (failed_start > start)
1340 clear_extent_bit(tree, start, failed_start - 1,
1341 EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS);
1347 int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end,
1348 struct extent_state **cached, gfp_t mask)
1350 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached,
1354 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1356 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL,
1360 int extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
1362 unsigned long index = start >> PAGE_CACHE_SHIFT;
1363 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1366 while (index <= end_index) {
1367 page = find_get_page(inode->i_mapping, index);
1368 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1369 clear_page_dirty_for_io(page);
1370 page_cache_release(page);
1376 int extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
1378 unsigned long index = start >> PAGE_CACHE_SHIFT;
1379 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1382 while (index <= end_index) {
1383 page = find_get_page(inode->i_mapping, index);
1384 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1385 account_page_redirty(page);
1386 __set_page_dirty_nobuffers(page);
1387 page_cache_release(page);
1394 * helper function to set both pages and extents in the tree writeback
1396 static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
1398 unsigned long index = start >> PAGE_CACHE_SHIFT;
1399 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1402 while (index <= end_index) {
1403 page = find_get_page(tree->mapping, index);
1404 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1405 set_page_writeback(page);
1406 page_cache_release(page);
1412 /* find the first state struct with 'bits' set after 'start', and
1413 * return it. tree->lock must be held. NULL will returned if
1414 * nothing was found after 'start'
1416 static struct extent_state *
1417 find_first_extent_bit_state(struct extent_io_tree *tree,
1418 u64 start, unsigned long bits)
1420 struct rb_node *node;
1421 struct extent_state *state;
1424 * this search will find all the extents that end after
1427 node = tree_search(tree, start);
1432 state = rb_entry(node, struct extent_state, rb_node);
1433 if (state->end >= start && (state->state & bits))
1436 node = rb_next(node);
1445 * find the first offset in the io tree with 'bits' set. zero is
1446 * returned if we find something, and *start_ret and *end_ret are
1447 * set to reflect the state struct that was found.
1449 * If nothing was found, 1 is returned. If found something, return 0.
1451 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1452 u64 *start_ret, u64 *end_ret, unsigned long bits,
1453 struct extent_state **cached_state)
1455 struct extent_state *state;
1459 spin_lock(&tree->lock);
1460 if (cached_state && *cached_state) {
1461 state = *cached_state;
1462 if (state->end == start - 1 && state->tree) {
1463 n = rb_next(&state->rb_node);
1465 state = rb_entry(n, struct extent_state,
1467 if (state->state & bits)
1471 free_extent_state(*cached_state);
1472 *cached_state = NULL;
1475 free_extent_state(*cached_state);
1476 *cached_state = NULL;
1479 state = find_first_extent_bit_state(tree, start, bits);
1482 cache_state(state, cached_state);
1483 *start_ret = state->start;
1484 *end_ret = state->end;
1488 spin_unlock(&tree->lock);
1493 * find a contiguous range of bytes in the file marked as delalloc, not
1494 * more than 'max_bytes'. start and end are used to return the range,
1496 * 1 is returned if we find something, 0 if nothing was in the tree
1498 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1499 u64 *start, u64 *end, u64 max_bytes,
1500 struct extent_state **cached_state)
1502 struct rb_node *node;
1503 struct extent_state *state;
1504 u64 cur_start = *start;
1506 u64 total_bytes = 0;
1508 spin_lock(&tree->lock);
1511 * this search will find all the extents that end after
1514 node = tree_search(tree, cur_start);
1522 state = rb_entry(node, struct extent_state, rb_node);
1523 if (found && (state->start != cur_start ||
1524 (state->state & EXTENT_BOUNDARY))) {
1527 if (!(state->state & EXTENT_DELALLOC)) {
1533 *start = state->start;
1534 *cached_state = state;
1535 atomic_inc(&state->refs);
1539 cur_start = state->end + 1;
1540 node = rb_next(node);
1541 total_bytes += state->end - state->start + 1;
1542 if (total_bytes >= max_bytes)
1548 spin_unlock(&tree->lock);
1552 static noinline void __unlock_for_delalloc(struct inode *inode,
1553 struct page *locked_page,
1557 struct page *pages[16];
1558 unsigned long index = start >> PAGE_CACHE_SHIFT;
1559 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1560 unsigned long nr_pages = end_index - index + 1;
1563 if (index == locked_page->index && end_index == index)
1566 while (nr_pages > 0) {
1567 ret = find_get_pages_contig(inode->i_mapping, index,
1568 min_t(unsigned long, nr_pages,
1569 ARRAY_SIZE(pages)), pages);
1570 for (i = 0; i < ret; i++) {
1571 if (pages[i] != locked_page)
1572 unlock_page(pages[i]);
1573 page_cache_release(pages[i]);
1581 static noinline int lock_delalloc_pages(struct inode *inode,
1582 struct page *locked_page,
1586 unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT;
1587 unsigned long start_index = index;
1588 unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT;
1589 unsigned long pages_locked = 0;
1590 struct page *pages[16];
1591 unsigned long nrpages;
1595 /* the caller is responsible for locking the start index */
1596 if (index == locked_page->index && index == end_index)
1599 /* skip the page at the start index */
1600 nrpages = end_index - index + 1;
1601 while (nrpages > 0) {
1602 ret = find_get_pages_contig(inode->i_mapping, index,
1603 min_t(unsigned long,
1604 nrpages, ARRAY_SIZE(pages)), pages);
1609 /* now we have an array of pages, lock them all */
1610 for (i = 0; i < ret; i++) {
1612 * the caller is taking responsibility for
1615 if (pages[i] != locked_page) {
1616 lock_page(pages[i]);
1617 if (!PageDirty(pages[i]) ||
1618 pages[i]->mapping != inode->i_mapping) {
1620 unlock_page(pages[i]);
1621 page_cache_release(pages[i]);
1625 page_cache_release(pages[i]);
1634 if (ret && pages_locked) {
1635 __unlock_for_delalloc(inode, locked_page,
1637 ((u64)(start_index + pages_locked - 1)) <<
1644 * find a contiguous range of bytes in the file marked as delalloc, not
1645 * more than 'max_bytes'. start and end are used to return the range,
1647 * 1 is returned if we find something, 0 if nothing was in the tree
1649 STATIC u64 find_lock_delalloc_range(struct inode *inode,
1650 struct extent_io_tree *tree,
1651 struct page *locked_page, u64 *start,
1652 u64 *end, u64 max_bytes)
1657 struct extent_state *cached_state = NULL;
1662 /* step one, find a bunch of delalloc bytes starting at start */
1663 delalloc_start = *start;
1665 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1666 max_bytes, &cached_state);
1667 if (!found || delalloc_end <= *start) {
1668 *start = delalloc_start;
1669 *end = delalloc_end;
1670 free_extent_state(cached_state);
1675 * start comes from the offset of locked_page. We have to lock
1676 * pages in order, so we can't process delalloc bytes before
1679 if (delalloc_start < *start)
1680 delalloc_start = *start;
1683 * make sure to limit the number of pages we try to lock down
1685 if (delalloc_end + 1 - delalloc_start > max_bytes)
1686 delalloc_end = delalloc_start + max_bytes - 1;
1688 /* step two, lock all the pages after the page that has start */
1689 ret = lock_delalloc_pages(inode, locked_page,
1690 delalloc_start, delalloc_end);
1691 if (ret == -EAGAIN) {
1692 /* some of the pages are gone, lets avoid looping by
1693 * shortening the size of the delalloc range we're searching
1695 free_extent_state(cached_state);
1697 max_bytes = PAGE_CACHE_SIZE;
1705 BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */
1707 /* step three, lock the state bits for the whole range */
1708 lock_extent_bits(tree, delalloc_start, delalloc_end, 0, &cached_state);
1710 /* then test to make sure it is all still delalloc */
1711 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1712 EXTENT_DELALLOC, 1, cached_state);
1714 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1715 &cached_state, GFP_NOFS);
1716 __unlock_for_delalloc(inode, locked_page,
1717 delalloc_start, delalloc_end);
1721 free_extent_state(cached_state);
1722 *start = delalloc_start;
1723 *end = delalloc_end;
1728 int extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end,
1729 struct page *locked_page,
1730 unsigned long clear_bits,
1731 unsigned long page_ops)
1733 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
1735 struct page *pages[16];
1736 unsigned long index = start >> PAGE_CACHE_SHIFT;
1737 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1738 unsigned long nr_pages = end_index - index + 1;
1741 clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS);
1745 while (nr_pages > 0) {
1746 ret = find_get_pages_contig(inode->i_mapping, index,
1747 min_t(unsigned long,
1748 nr_pages, ARRAY_SIZE(pages)), pages);
1749 for (i = 0; i < ret; i++) {
1751 if (page_ops & PAGE_SET_PRIVATE2)
1752 SetPagePrivate2(pages[i]);
1754 if (pages[i] == locked_page) {
1755 page_cache_release(pages[i]);
1758 if (page_ops & PAGE_CLEAR_DIRTY)
1759 clear_page_dirty_for_io(pages[i]);
1760 if (page_ops & PAGE_SET_WRITEBACK)
1761 set_page_writeback(pages[i]);
1762 if (page_ops & PAGE_END_WRITEBACK)
1763 end_page_writeback(pages[i]);
1764 if (page_ops & PAGE_UNLOCK)
1765 unlock_page(pages[i]);
1766 page_cache_release(pages[i]);
1776 * count the number of bytes in the tree that have a given bit(s)
1777 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1778 * cached. The total number found is returned.
1780 u64 count_range_bits(struct extent_io_tree *tree,
1781 u64 *start, u64 search_end, u64 max_bytes,
1782 unsigned long bits, int contig)
1784 struct rb_node *node;
1785 struct extent_state *state;
1786 u64 cur_start = *start;
1787 u64 total_bytes = 0;
1791 if (WARN_ON(search_end <= cur_start))
1794 spin_lock(&tree->lock);
1795 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1796 total_bytes = tree->dirty_bytes;
1800 * this search will find all the extents that end after
1803 node = tree_search(tree, cur_start);
1808 state = rb_entry(node, struct extent_state, rb_node);
1809 if (state->start > search_end)
1811 if (contig && found && state->start > last + 1)
1813 if (state->end >= cur_start && (state->state & bits) == bits) {
1814 total_bytes += min(search_end, state->end) + 1 -
1815 max(cur_start, state->start);
1816 if (total_bytes >= max_bytes)
1819 *start = max(cur_start, state->start);
1823 } else if (contig && found) {
1826 node = rb_next(node);
1831 spin_unlock(&tree->lock);
1836 * set the private field for a given byte offset in the tree. If there isn't
1837 * an extent_state there already, this does nothing.
1839 static int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1841 struct rb_node *node;
1842 struct extent_state *state;
1845 spin_lock(&tree->lock);
1847 * this search will find all the extents that end after
1850 node = tree_search(tree, start);
1855 state = rb_entry(node, struct extent_state, rb_node);
1856 if (state->start != start) {
1860 state->private = private;
1862 spin_unlock(&tree->lock);
1866 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1868 struct rb_node *node;
1869 struct extent_state *state;
1872 spin_lock(&tree->lock);
1874 * this search will find all the extents that end after
1877 node = tree_search(tree, start);
1882 state = rb_entry(node, struct extent_state, rb_node);
1883 if (state->start != start) {
1887 *private = state->private;
1889 spin_unlock(&tree->lock);
1894 * searches a range in the state tree for a given mask.
1895 * If 'filled' == 1, this returns 1 only if every extent in the tree
1896 * has the bits set. Otherwise, 1 is returned if any bit in the
1897 * range is found set.
1899 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1900 unsigned long bits, int filled, struct extent_state *cached)
1902 struct extent_state *state = NULL;
1903 struct rb_node *node;
1906 spin_lock(&tree->lock);
1907 if (cached && cached->tree && cached->start <= start &&
1908 cached->end > start)
1909 node = &cached->rb_node;
1911 node = tree_search(tree, start);
1912 while (node && start <= end) {
1913 state = rb_entry(node, struct extent_state, rb_node);
1915 if (filled && state->start > start) {
1920 if (state->start > end)
1923 if (state->state & bits) {
1927 } else if (filled) {
1932 if (state->end == (u64)-1)
1935 start = state->end + 1;
1938 node = rb_next(node);
1945 spin_unlock(&tree->lock);
1950 * helper function to set a given page up to date if all the
1951 * extents in the tree for that page are up to date
1953 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
1955 u64 start = page_offset(page);
1956 u64 end = start + PAGE_CACHE_SIZE - 1;
1957 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1958 SetPageUptodate(page);
1962 * When IO fails, either with EIO or csum verification fails, we
1963 * try other mirrors that might have a good copy of the data. This
1964 * io_failure_record is used to record state as we go through all the
1965 * mirrors. If another mirror has good data, the page is set up to date
1966 * and things continue. If a good mirror can't be found, the original
1967 * bio end_io callback is called to indicate things have failed.
1969 struct io_failure_record {
1974 unsigned long bio_flags;
1980 static int free_io_failure(struct inode *inode, struct io_failure_record *rec,
1985 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1987 set_state_private(failure_tree, rec->start, 0);
1988 ret = clear_extent_bits(failure_tree, rec->start,
1989 rec->start + rec->len - 1,
1990 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1994 ret = clear_extent_bits(&BTRFS_I(inode)->io_tree, rec->start,
1995 rec->start + rec->len - 1,
1996 EXTENT_DAMAGED, GFP_NOFS);
2005 * this bypasses the standard btrfs submit functions deliberately, as
2006 * the standard behavior is to write all copies in a raid setup. here we only
2007 * want to write the one bad copy. so we do the mapping for ourselves and issue
2008 * submit_bio directly.
2009 * to avoid any synchronization issues, wait for the data after writing, which
2010 * actually prevents the read that triggered the error from finishing.
2011 * currently, there can be no more than two copies of every data bit. thus,
2012 * exactly one rewrite is required.
2014 int repair_io_failure(struct btrfs_fs_info *fs_info, u64 start,
2015 u64 length, u64 logical, struct page *page,
2019 struct btrfs_device *dev;
2022 struct btrfs_bio *bbio = NULL;
2023 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
2026 ASSERT(!(fs_info->sb->s_flags & MS_RDONLY));
2027 BUG_ON(!mirror_num);
2029 /* we can't repair anything in raid56 yet */
2030 if (btrfs_is_parity_mirror(map_tree, logical, length, mirror_num))
2033 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
2036 bio->bi_iter.bi_size = 0;
2037 map_length = length;
2039 ret = btrfs_map_block(fs_info, WRITE, logical,
2040 &map_length, &bbio, mirror_num);
2045 BUG_ON(mirror_num != bbio->mirror_num);
2046 sector = bbio->stripes[mirror_num-1].physical >> 9;
2047 bio->bi_iter.bi_sector = sector;
2048 dev = bbio->stripes[mirror_num-1].dev;
2050 if (!dev || !dev->bdev || !dev->writeable) {
2054 bio->bi_bdev = dev->bdev;
2055 bio_add_page(bio, page, length, start - page_offset(page));
2057 if (btrfsic_submit_bio_wait(WRITE_SYNC, bio)) {
2058 /* try to remap that extent elsewhere? */
2060 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
2064 printk_ratelimited_in_rcu(KERN_INFO
2065 "BTRFS: read error corrected: ino %lu off %llu "
2066 "(dev %s sector %llu)\n", page->mapping->host->i_ino,
2067 start, rcu_str_deref(dev->name), sector);
2073 int repair_eb_io_failure(struct btrfs_root *root, struct extent_buffer *eb,
2076 u64 start = eb->start;
2077 unsigned long i, num_pages = num_extent_pages(eb->start, eb->len);
2080 if (root->fs_info->sb->s_flags & MS_RDONLY)
2083 for (i = 0; i < num_pages; i++) {
2084 struct page *p = extent_buffer_page(eb, i);
2085 ret = repair_io_failure(root->fs_info, start, PAGE_CACHE_SIZE,
2086 start, p, mirror_num);
2089 start += PAGE_CACHE_SIZE;
2096 * each time an IO finishes, we do a fast check in the IO failure tree
2097 * to see if we need to process or clean up an io_failure_record
2099 static int clean_io_failure(u64 start, struct page *page)
2102 u64 private_failure;
2103 struct io_failure_record *failrec;
2104 struct inode *inode = page->mapping->host;
2105 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2106 struct extent_state *state;
2112 ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
2113 (u64)-1, 1, EXTENT_DIRTY, 0);
2117 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, start,
2122 failrec = (struct io_failure_record *)(unsigned long) private_failure;
2123 BUG_ON(!failrec->this_mirror);
2125 if (failrec->in_validation) {
2126 /* there was no real error, just free the record */
2127 pr_debug("clean_io_failure: freeing dummy error at %llu\n",
2132 if (fs_info->sb->s_flags & MS_RDONLY)
2135 spin_lock(&BTRFS_I(inode)->io_tree.lock);
2136 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
2139 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
2141 if (state && state->start <= failrec->start &&
2142 state->end >= failrec->start + failrec->len - 1) {
2143 num_copies = btrfs_num_copies(fs_info, failrec->logical,
2145 if (num_copies > 1) {
2146 ret = repair_io_failure(fs_info, start, failrec->len,
2147 failrec->logical, page,
2148 failrec->failed_mirror);
2156 ret = free_io_failure(inode, failrec, did_repair);
2162 * this is a generic handler for readpage errors (default
2163 * readpage_io_failed_hook). if other copies exist, read those and write back
2164 * good data to the failed position. does not investigate in remapping the
2165 * failed extent elsewhere, hoping the device will be smart enough to do this as
2169 static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset,
2170 struct page *page, u64 start, u64 end,
2173 struct io_failure_record *failrec = NULL;
2175 struct extent_map *em;
2176 struct inode *inode = page->mapping->host;
2177 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2178 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2179 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2181 struct btrfs_io_bio *btrfs_failed_bio;
2182 struct btrfs_io_bio *btrfs_bio;
2188 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
2190 ret = get_state_private(failure_tree, start, &private);
2192 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2195 failrec->start = start;
2196 failrec->len = end - start + 1;
2197 failrec->this_mirror = 0;
2198 failrec->bio_flags = 0;
2199 failrec->in_validation = 0;
2201 read_lock(&em_tree->lock);
2202 em = lookup_extent_mapping(em_tree, start, failrec->len);
2204 read_unlock(&em_tree->lock);
2209 if (em->start > start || em->start + em->len <= start) {
2210 free_extent_map(em);
2213 read_unlock(&em_tree->lock);
2219 logical = start - em->start;
2220 logical = em->block_start + logical;
2221 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2222 logical = em->block_start;
2223 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2224 extent_set_compress_type(&failrec->bio_flags,
2227 pr_debug("bio_readpage_error: (new) logical=%llu, start=%llu, "
2228 "len=%llu\n", logical, start, failrec->len);
2229 failrec->logical = logical;
2230 free_extent_map(em);
2232 /* set the bits in the private failure tree */
2233 ret = set_extent_bits(failure_tree, start, end,
2234 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
2236 ret = set_state_private(failure_tree, start,
2237 (u64)(unsigned long)failrec);
2238 /* set the bits in the inode's tree */
2240 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED,
2247 failrec = (struct io_failure_record *)(unsigned long)private;
2248 pr_debug("bio_readpage_error: (found) logical=%llu, "
2249 "start=%llu, len=%llu, validation=%d\n",
2250 failrec->logical, failrec->start, failrec->len,
2251 failrec->in_validation);
2253 * when data can be on disk more than twice, add to failrec here
2254 * (e.g. with a list for failed_mirror) to make
2255 * clean_io_failure() clean all those errors at once.
2258 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
2259 failrec->logical, failrec->len);
2260 if (num_copies == 1) {
2262 * we only have a single copy of the data, so don't bother with
2263 * all the retry and error correction code that follows. no
2264 * matter what the error is, it is very likely to persist.
2266 pr_debug("bio_readpage_error: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
2267 num_copies, failrec->this_mirror, failed_mirror);
2268 free_io_failure(inode, failrec, 0);
2273 * there are two premises:
2274 * a) deliver good data to the caller
2275 * b) correct the bad sectors on disk
2277 if (failed_bio->bi_vcnt > 1) {
2279 * to fulfill b), we need to know the exact failing sectors, as
2280 * we don't want to rewrite any more than the failed ones. thus,
2281 * we need separate read requests for the failed bio
2283 * if the following BUG_ON triggers, our validation request got
2284 * merged. we need separate requests for our algorithm to work.
2286 BUG_ON(failrec->in_validation);
2287 failrec->in_validation = 1;
2288 failrec->this_mirror = failed_mirror;
2289 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
2292 * we're ready to fulfill a) and b) alongside. get a good copy
2293 * of the failed sector and if we succeed, we have setup
2294 * everything for repair_io_failure to do the rest for us.
2296 if (failrec->in_validation) {
2297 BUG_ON(failrec->this_mirror != failed_mirror);
2298 failrec->in_validation = 0;
2299 failrec->this_mirror = 0;
2301 failrec->failed_mirror = failed_mirror;
2302 failrec->this_mirror++;
2303 if (failrec->this_mirror == failed_mirror)
2304 failrec->this_mirror++;
2305 read_mode = READ_SYNC;
2308 if (failrec->this_mirror > num_copies) {
2309 pr_debug("bio_readpage_error: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
2310 num_copies, failrec->this_mirror, failed_mirror);
2311 free_io_failure(inode, failrec, 0);
2315 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
2317 free_io_failure(inode, failrec, 0);
2320 bio->bi_end_io = failed_bio->bi_end_io;
2321 bio->bi_iter.bi_sector = failrec->logical >> 9;
2322 bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
2323 bio->bi_iter.bi_size = 0;
2325 btrfs_failed_bio = btrfs_io_bio(failed_bio);
2326 if (btrfs_failed_bio->csum) {
2327 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2328 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
2330 btrfs_bio = btrfs_io_bio(bio);
2331 btrfs_bio->csum = btrfs_bio->csum_inline;
2332 phy_offset >>= inode->i_sb->s_blocksize_bits;
2333 phy_offset *= csum_size;
2334 memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + phy_offset,
2338 bio_add_page(bio, page, failrec->len, start - page_offset(page));
2340 pr_debug("bio_readpage_error: submitting new read[%#x] to "
2341 "this_mirror=%d, num_copies=%d, in_validation=%d\n", read_mode,
2342 failrec->this_mirror, num_copies, failrec->in_validation);
2344 ret = tree->ops->submit_bio_hook(inode, read_mode, bio,
2345 failrec->this_mirror,
2346 failrec->bio_flags, 0);
2350 /* lots and lots of room for performance fixes in the end_bio funcs */
2352 int end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2354 int uptodate = (err == 0);
2355 struct extent_io_tree *tree;
2358 tree = &BTRFS_I(page->mapping->host)->io_tree;
2360 if (tree->ops && tree->ops->writepage_end_io_hook) {
2361 ret = tree->ops->writepage_end_io_hook(page, start,
2362 end, NULL, uptodate);
2368 ClearPageUptodate(page);
2370 ret = ret < 0 ? ret : -EIO;
2371 mapping_set_error(page->mapping, ret);
2377 * after a writepage IO is done, we need to:
2378 * clear the uptodate bits on error
2379 * clear the writeback bits in the extent tree for this IO
2380 * end_page_writeback if the page has no more pending IO
2382 * Scheduling is not allowed, so the extent state tree is expected
2383 * to have one and only one object corresponding to this IO.
2385 static void end_bio_extent_writepage(struct bio *bio, int err)
2387 struct bio_vec *bvec;
2392 bio_for_each_segment_all(bvec, bio, i) {
2393 struct page *page = bvec->bv_page;
2395 /* We always issue full-page reads, but if some block
2396 * in a page fails to read, blk_update_request() will
2397 * advance bv_offset and adjust bv_len to compensate.
2398 * Print a warning for nonzero offsets, and an error
2399 * if they don't add up to a full page. */
2400 if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE) {
2401 if (bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE)
2402 btrfs_err(BTRFS_I(page->mapping->host)->root->fs_info,
2403 "partial page write in btrfs with offset %u and length %u",
2404 bvec->bv_offset, bvec->bv_len);
2406 btrfs_info(BTRFS_I(page->mapping->host)->root->fs_info,
2407 "incomplete page write in btrfs with offset %u and "
2409 bvec->bv_offset, bvec->bv_len);
2412 start = page_offset(page);
2413 end = start + bvec->bv_offset + bvec->bv_len - 1;
2415 if (end_extent_writepage(page, err, start, end))
2418 end_page_writeback(page);
2425 endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len,
2428 struct extent_state *cached = NULL;
2429 u64 end = start + len - 1;
2431 if (uptodate && tree->track_uptodate)
2432 set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC);
2433 unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC);
2437 * after a readpage IO is done, we need to:
2438 * clear the uptodate bits on error
2439 * set the uptodate bits if things worked
2440 * set the page up to date if all extents in the tree are uptodate
2441 * clear the lock bit in the extent tree
2442 * unlock the page if there are no other extents locked for it
2444 * Scheduling is not allowed, so the extent state tree is expected
2445 * to have one and only one object corresponding to this IO.
2447 static void end_bio_extent_readpage(struct bio *bio, int err)
2449 struct bio_vec *bvec;
2450 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
2451 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
2452 struct extent_io_tree *tree;
2457 u64 extent_start = 0;
2466 bio_for_each_segment_all(bvec, bio, i) {
2467 struct page *page = bvec->bv_page;
2468 struct inode *inode = page->mapping->host;
2470 pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, "
2471 "mirror=%lu\n", (u64)bio->bi_iter.bi_sector, err,
2472 io_bio->mirror_num);
2473 tree = &BTRFS_I(inode)->io_tree;
2475 /* We always issue full-page reads, but if some block
2476 * in a page fails to read, blk_update_request() will
2477 * advance bv_offset and adjust bv_len to compensate.
2478 * Print a warning for nonzero offsets, and an error
2479 * if they don't add up to a full page. */
2480 if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE) {
2481 if (bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE)
2482 btrfs_err(BTRFS_I(page->mapping->host)->root->fs_info,
2483 "partial page read in btrfs with offset %u and length %u",
2484 bvec->bv_offset, bvec->bv_len);
2486 btrfs_info(BTRFS_I(page->mapping->host)->root->fs_info,
2487 "incomplete page read in btrfs with offset %u and "
2489 bvec->bv_offset, bvec->bv_len);
2492 start = page_offset(page);
2493 end = start + bvec->bv_offset + bvec->bv_len - 1;
2496 mirror = io_bio->mirror_num;
2497 if (likely(uptodate && tree->ops &&
2498 tree->ops->readpage_end_io_hook)) {
2499 ret = tree->ops->readpage_end_io_hook(io_bio, offset,
2505 clean_io_failure(start, page);
2508 if (likely(uptodate))
2511 if (tree->ops && tree->ops->readpage_io_failed_hook) {
2512 ret = tree->ops->readpage_io_failed_hook(page, mirror);
2514 test_bit(BIO_UPTODATE, &bio->bi_flags))
2518 * The generic bio_readpage_error handles errors the
2519 * following way: If possible, new read requests are
2520 * created and submitted and will end up in
2521 * end_bio_extent_readpage as well (if we're lucky, not
2522 * in the !uptodate case). In that case it returns 0 and
2523 * we just go on with the next page in our bio. If it
2524 * can't handle the error it will return -EIO and we
2525 * remain responsible for that page.
2527 ret = bio_readpage_error(bio, offset, page, start, end,
2531 test_bit(BIO_UPTODATE, &bio->bi_flags);
2538 if (likely(uptodate)) {
2539 loff_t i_size = i_size_read(inode);
2540 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
2543 /* Zero out the end if this page straddles i_size */
2544 offset = i_size & (PAGE_CACHE_SIZE-1);
2545 if (page->index == end_index && offset)
2546 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
2547 SetPageUptodate(page);
2549 ClearPageUptodate(page);
2555 if (unlikely(!uptodate)) {
2557 endio_readpage_release_extent(tree,
2563 endio_readpage_release_extent(tree, start,
2564 end - start + 1, 0);
2565 } else if (!extent_len) {
2566 extent_start = start;
2567 extent_len = end + 1 - start;
2568 } else if (extent_start + extent_len == start) {
2569 extent_len += end + 1 - start;
2571 endio_readpage_release_extent(tree, extent_start,
2572 extent_len, uptodate);
2573 extent_start = start;
2574 extent_len = end + 1 - start;
2579 endio_readpage_release_extent(tree, extent_start, extent_len,
2582 io_bio->end_io(io_bio, err);
2587 * this allocates from the btrfs_bioset. We're returning a bio right now
2588 * but you can call btrfs_io_bio for the appropriate container_of magic
2591 btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
2594 struct btrfs_io_bio *btrfs_bio;
2597 bio = bio_alloc_bioset(gfp_flags, nr_vecs, btrfs_bioset);
2599 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
2600 while (!bio && (nr_vecs /= 2)) {
2601 bio = bio_alloc_bioset(gfp_flags,
2602 nr_vecs, btrfs_bioset);
2607 bio->bi_bdev = bdev;
2608 bio->bi_iter.bi_sector = first_sector;
2609 btrfs_bio = btrfs_io_bio(bio);
2610 btrfs_bio->csum = NULL;
2611 btrfs_bio->csum_allocated = NULL;
2612 btrfs_bio->end_io = NULL;
2617 struct bio *btrfs_bio_clone(struct bio *bio, gfp_t gfp_mask)
2619 return bio_clone_bioset(bio, gfp_mask, btrfs_bioset);
2623 /* this also allocates from the btrfs_bioset */
2624 struct bio *btrfs_io_bio_alloc(gfp_t gfp_mask, unsigned int nr_iovecs)
2626 struct btrfs_io_bio *btrfs_bio;
2629 bio = bio_alloc_bioset(gfp_mask, nr_iovecs, btrfs_bioset);
2631 btrfs_bio = btrfs_io_bio(bio);
2632 btrfs_bio->csum = NULL;
2633 btrfs_bio->csum_allocated = NULL;
2634 btrfs_bio->end_io = NULL;
2640 static int __must_check submit_one_bio(int rw, struct bio *bio,
2641 int mirror_num, unsigned long bio_flags)
2644 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2645 struct page *page = bvec->bv_page;
2646 struct extent_io_tree *tree = bio->bi_private;
2649 start = page_offset(page) + bvec->bv_offset;
2651 bio->bi_private = NULL;
2655 if (tree->ops && tree->ops->submit_bio_hook)
2656 ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
2657 mirror_num, bio_flags, start);
2659 btrfsic_submit_bio(rw, bio);
2661 if (bio_flagged(bio, BIO_EOPNOTSUPP))
2667 static int merge_bio(int rw, struct extent_io_tree *tree, struct page *page,
2668 unsigned long offset, size_t size, struct bio *bio,
2669 unsigned long bio_flags)
2672 if (tree->ops && tree->ops->merge_bio_hook)
2673 ret = tree->ops->merge_bio_hook(rw, page, offset, size, bio,
2680 static int submit_extent_page(int rw, struct extent_io_tree *tree,
2681 struct page *page, sector_t sector,
2682 size_t size, unsigned long offset,
2683 struct block_device *bdev,
2684 struct bio **bio_ret,
2685 unsigned long max_pages,
2686 bio_end_io_t end_io_func,
2688 unsigned long prev_bio_flags,
2689 unsigned long bio_flags)
2695 int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED;
2696 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
2697 size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE);
2699 if (bio_ret && *bio_ret) {
2702 contig = bio->bi_iter.bi_sector == sector;
2704 contig = bio_end_sector(bio) == sector;
2706 if (prev_bio_flags != bio_flags || !contig ||
2707 merge_bio(rw, tree, page, offset, page_size, bio, bio_flags) ||
2708 bio_add_page(bio, page, page_size, offset) < page_size) {
2709 ret = submit_one_bio(rw, bio, mirror_num,
2718 if (this_compressed)
2721 nr = bio_get_nr_vecs(bdev);
2723 bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
2727 bio_add_page(bio, page, page_size, offset);
2728 bio->bi_end_io = end_io_func;
2729 bio->bi_private = tree;
2734 ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
2739 static void attach_extent_buffer_page(struct extent_buffer *eb,
2742 if (!PagePrivate(page)) {
2743 SetPagePrivate(page);
2744 page_cache_get(page);
2745 set_page_private(page, (unsigned long)eb);
2747 WARN_ON(page->private != (unsigned long)eb);
2751 void set_page_extent_mapped(struct page *page)
2753 if (!PagePrivate(page)) {
2754 SetPagePrivate(page);
2755 page_cache_get(page);
2756 set_page_private(page, EXTENT_PAGE_PRIVATE);
2760 static struct extent_map *
2761 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
2762 u64 start, u64 len, get_extent_t *get_extent,
2763 struct extent_map **em_cached)
2765 struct extent_map *em;
2767 if (em_cached && *em_cached) {
2769 if (extent_map_in_tree(em) && start >= em->start &&
2770 start < extent_map_end(em)) {
2771 atomic_inc(&em->refs);
2775 free_extent_map(em);
2779 em = get_extent(inode, page, pg_offset, start, len, 0);
2780 if (em_cached && !IS_ERR_OR_NULL(em)) {
2782 atomic_inc(&em->refs);
2788 * basic readpage implementation. Locked extent state structs are inserted
2789 * into the tree that are removed when the IO is done (by the end_io
2791 * XXX JDM: This needs looking at to ensure proper page locking
2793 static int __do_readpage(struct extent_io_tree *tree,
2795 get_extent_t *get_extent,
2796 struct extent_map **em_cached,
2797 struct bio **bio, int mirror_num,
2798 unsigned long *bio_flags, int rw)
2800 struct inode *inode = page->mapping->host;
2801 u64 start = page_offset(page);
2802 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2806 u64 last_byte = i_size_read(inode);
2810 struct extent_map *em;
2811 struct block_device *bdev;
2814 int parent_locked = *bio_flags & EXTENT_BIO_PARENT_LOCKED;
2815 size_t pg_offset = 0;
2817 size_t disk_io_size;
2818 size_t blocksize = inode->i_sb->s_blocksize;
2819 unsigned long this_bio_flag = *bio_flags & EXTENT_BIO_PARENT_LOCKED;
2821 set_page_extent_mapped(page);
2824 if (!PageUptodate(page)) {
2825 if (cleancache_get_page(page) == 0) {
2826 BUG_ON(blocksize != PAGE_SIZE);
2827 unlock_extent(tree, start, end);
2832 if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
2834 size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
2837 iosize = PAGE_CACHE_SIZE - zero_offset;
2838 userpage = kmap_atomic(page);
2839 memset(userpage + zero_offset, 0, iosize);
2840 flush_dcache_page(page);
2841 kunmap_atomic(userpage);
2844 while (cur <= end) {
2845 unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
2847 if (cur >= last_byte) {
2849 struct extent_state *cached = NULL;
2851 iosize = PAGE_CACHE_SIZE - pg_offset;
2852 userpage = kmap_atomic(page);
2853 memset(userpage + pg_offset, 0, iosize);
2854 flush_dcache_page(page);
2855 kunmap_atomic(userpage);
2856 set_extent_uptodate(tree, cur, cur + iosize - 1,
2859 unlock_extent_cached(tree, cur,
2864 em = __get_extent_map(inode, page, pg_offset, cur,
2865 end - cur + 1, get_extent, em_cached);
2866 if (IS_ERR_OR_NULL(em)) {
2869 unlock_extent(tree, cur, end);
2872 extent_offset = cur - em->start;
2873 BUG_ON(extent_map_end(em) <= cur);
2876 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2877 this_bio_flag |= EXTENT_BIO_COMPRESSED;
2878 extent_set_compress_type(&this_bio_flag,
2882 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2883 cur_end = min(extent_map_end(em) - 1, end);
2884 iosize = ALIGN(iosize, blocksize);
2885 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2886 disk_io_size = em->block_len;
2887 sector = em->block_start >> 9;
2889 sector = (em->block_start + extent_offset) >> 9;
2890 disk_io_size = iosize;
2893 block_start = em->block_start;
2894 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2895 block_start = EXTENT_MAP_HOLE;
2896 free_extent_map(em);
2899 /* we've found a hole, just zero and go on */
2900 if (block_start == EXTENT_MAP_HOLE) {
2902 struct extent_state *cached = NULL;
2904 userpage = kmap_atomic(page);
2905 memset(userpage + pg_offset, 0, iosize);
2906 flush_dcache_page(page);
2907 kunmap_atomic(userpage);
2909 set_extent_uptodate(tree, cur, cur + iosize - 1,
2911 unlock_extent_cached(tree, cur, cur + iosize - 1,
2914 pg_offset += iosize;
2917 /* the get_extent function already copied into the page */
2918 if (test_range_bit(tree, cur, cur_end,
2919 EXTENT_UPTODATE, 1, NULL)) {
2920 check_page_uptodate(tree, page);
2922 unlock_extent(tree, cur, cur + iosize - 1);
2924 pg_offset += iosize;
2927 /* we have an inline extent but it didn't get marked up
2928 * to date. Error out
2930 if (block_start == EXTENT_MAP_INLINE) {
2933 unlock_extent(tree, cur, cur + iosize - 1);
2935 pg_offset += iosize;
2940 ret = submit_extent_page(rw, tree, page,
2941 sector, disk_io_size, pg_offset,
2943 end_bio_extent_readpage, mirror_num,
2948 *bio_flags = this_bio_flag;
2952 unlock_extent(tree, cur, cur + iosize - 1);
2955 pg_offset += iosize;
2959 if (!PageError(page))
2960 SetPageUptodate(page);
2966 static inline void __do_contiguous_readpages(struct extent_io_tree *tree,
2967 struct page *pages[], int nr_pages,
2969 get_extent_t *get_extent,
2970 struct extent_map **em_cached,
2971 struct bio **bio, int mirror_num,
2972 unsigned long *bio_flags, int rw)
2974 struct inode *inode;
2975 struct btrfs_ordered_extent *ordered;
2978 inode = pages[0]->mapping->host;
2980 lock_extent(tree, start, end);
2981 ordered = btrfs_lookup_ordered_range(inode, start,
2985 unlock_extent(tree, start, end);
2986 btrfs_start_ordered_extent(inode, ordered, 1);
2987 btrfs_put_ordered_extent(ordered);
2990 for (index = 0; index < nr_pages; index++) {
2991 __do_readpage(tree, pages[index], get_extent, em_cached, bio,
2992 mirror_num, bio_flags, rw);
2993 page_cache_release(pages[index]);
2997 static void __extent_readpages(struct extent_io_tree *tree,
2998 struct page *pages[],
2999 int nr_pages, get_extent_t *get_extent,
3000 struct extent_map **em_cached,
3001 struct bio **bio, int mirror_num,
3002 unsigned long *bio_flags, int rw)
3008 int first_index = 0;
3010 for (index = 0; index < nr_pages; index++) {
3011 page_start = page_offset(pages[index]);
3014 end = start + PAGE_CACHE_SIZE - 1;
3015 first_index = index;
3016 } else if (end + 1 == page_start) {
3017 end += PAGE_CACHE_SIZE;
3019 __do_contiguous_readpages(tree, &pages[first_index],
3020 index - first_index, start,
3021 end, get_extent, em_cached,
3022 bio, mirror_num, bio_flags,
3025 end = start + PAGE_CACHE_SIZE - 1;
3026 first_index = index;
3031 __do_contiguous_readpages(tree, &pages[first_index],
3032 index - first_index, start,
3033 end, get_extent, em_cached, bio,
3034 mirror_num, bio_flags, rw);
3037 static int __extent_read_full_page(struct extent_io_tree *tree,
3039 get_extent_t *get_extent,
3040 struct bio **bio, int mirror_num,
3041 unsigned long *bio_flags, int rw)
3043 struct inode *inode = page->mapping->host;
3044 struct btrfs_ordered_extent *ordered;
3045 u64 start = page_offset(page);
3046 u64 end = start + PAGE_CACHE_SIZE - 1;
3050 lock_extent(tree, start, end);
3051 ordered = btrfs_lookup_ordered_extent(inode, start);
3054 unlock_extent(tree, start, end);
3055 btrfs_start_ordered_extent(inode, ordered, 1);
3056 btrfs_put_ordered_extent(ordered);
3059 ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num,
3064 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
3065 get_extent_t *get_extent, int mirror_num)
3067 struct bio *bio = NULL;
3068 unsigned long bio_flags = 0;
3071 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
3074 ret = submit_one_bio(READ, bio, mirror_num, bio_flags);
3078 int extent_read_full_page_nolock(struct extent_io_tree *tree, struct page *page,
3079 get_extent_t *get_extent, int mirror_num)
3081 struct bio *bio = NULL;
3082 unsigned long bio_flags = EXTENT_BIO_PARENT_LOCKED;
3085 ret = __do_readpage(tree, page, get_extent, NULL, &bio, mirror_num,
3088 ret = submit_one_bio(READ, bio, mirror_num, bio_flags);
3092 static noinline void update_nr_written(struct page *page,
3093 struct writeback_control *wbc,
3094 unsigned long nr_written)
3096 wbc->nr_to_write -= nr_written;
3097 if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
3098 wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
3099 page->mapping->writeback_index = page->index + nr_written;
3103 * the writepage semantics are similar to regular writepage. extent
3104 * records are inserted to lock ranges in the tree, and as dirty areas
3105 * are found, they are marked writeback. Then the lock bits are removed
3106 * and the end_io handler clears the writeback ranges
3108 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
3111 struct inode *inode = page->mapping->host;
3112 struct extent_page_data *epd = data;
3113 struct extent_io_tree *tree = epd->tree;
3114 u64 start = page_offset(page);
3116 u64 page_end = start + PAGE_CACHE_SIZE - 1;
3120 u64 last_byte = i_size_read(inode);
3124 struct extent_state *cached_state = NULL;
3125 struct extent_map *em;
3126 struct block_device *bdev;
3129 size_t pg_offset = 0;
3131 loff_t i_size = i_size_read(inode);
3132 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
3138 unsigned long nr_written = 0;
3139 bool fill_delalloc = true;
3141 if (wbc->sync_mode == WB_SYNC_ALL)
3142 write_flags = WRITE_SYNC;
3144 write_flags = WRITE;
3146 trace___extent_writepage(page, inode, wbc);
3148 WARN_ON(!PageLocked(page));
3150 ClearPageError(page);
3152 pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
3153 if (page->index > end_index ||
3154 (page->index == end_index && !pg_offset)) {
3155 page->mapping->a_ops->invalidatepage(page, 0, PAGE_CACHE_SIZE);
3160 if (page->index == end_index) {
3163 userpage = kmap_atomic(page);
3164 memset(userpage + pg_offset, 0,
3165 PAGE_CACHE_SIZE - pg_offset);
3166 kunmap_atomic(userpage);
3167 flush_dcache_page(page);
3171 set_page_extent_mapped(page);
3173 if (!tree->ops || !tree->ops->fill_delalloc)
3174 fill_delalloc = false;
3176 delalloc_start = start;
3179 if (!epd->extent_locked && fill_delalloc) {
3180 u64 delalloc_to_write = 0;
3182 * make sure the wbc mapping index is at least updated
3185 update_nr_written(page, wbc, 0);
3187 while (delalloc_end < page_end) {
3188 nr_delalloc = find_lock_delalloc_range(inode, tree,
3193 if (nr_delalloc == 0) {
3194 delalloc_start = delalloc_end + 1;
3197 ret = tree->ops->fill_delalloc(inode, page,
3202 /* File system has been set read-only */
3208 * delalloc_end is already one less than the total
3209 * length, so we don't subtract one from
3212 delalloc_to_write += (delalloc_end - delalloc_start +
3215 delalloc_start = delalloc_end + 1;
3217 if (wbc->nr_to_write < delalloc_to_write) {
3220 if (delalloc_to_write < thresh * 2)
3221 thresh = delalloc_to_write;
3222 wbc->nr_to_write = min_t(u64, delalloc_to_write,
3226 /* did the fill delalloc function already unlock and start
3232 * we've unlocked the page, so we can't update
3233 * the mapping's writeback index, just update
3236 wbc->nr_to_write -= nr_written;
3240 if (tree->ops && tree->ops->writepage_start_hook) {
3241 ret = tree->ops->writepage_start_hook(page, start,
3244 /* Fixup worker will requeue */
3246 wbc->pages_skipped++;
3248 redirty_page_for_writepage(wbc, page);
3249 update_nr_written(page, wbc, nr_written);
3257 * we don't want to touch the inode after unlocking the page,
3258 * so we update the mapping writeback index now
3260 update_nr_written(page, wbc, nr_written + 1);
3263 if (last_byte <= start) {
3264 if (tree->ops && tree->ops->writepage_end_io_hook)
3265 tree->ops->writepage_end_io_hook(page, start,
3270 blocksize = inode->i_sb->s_blocksize;
3272 while (cur <= end) {
3273 if (cur >= last_byte) {
3274 if (tree->ops && tree->ops->writepage_end_io_hook)
3275 tree->ops->writepage_end_io_hook(page, cur,
3279 em = epd->get_extent(inode, page, pg_offset, cur,
3281 if (IS_ERR_OR_NULL(em)) {
3283 ret = PTR_ERR_OR_ZERO(em);
3287 extent_offset = cur - em->start;
3288 BUG_ON(extent_map_end(em) <= cur);
3290 iosize = min(extent_map_end(em) - cur, end - cur + 1);
3291 iosize = ALIGN(iosize, blocksize);
3292 sector = (em->block_start + extent_offset) >> 9;
3294 block_start = em->block_start;
3295 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3296 free_extent_map(em);
3300 * compressed and inline extents are written through other
3303 if (compressed || block_start == EXTENT_MAP_HOLE ||
3304 block_start == EXTENT_MAP_INLINE) {
3306 * end_io notification does not happen here for
3307 * compressed extents
3309 if (!compressed && tree->ops &&
3310 tree->ops->writepage_end_io_hook)
3311 tree->ops->writepage_end_io_hook(page, cur,
3314 else if (compressed) {
3315 /* we don't want to end_page_writeback on
3316 * a compressed extent. this happens
3323 pg_offset += iosize;
3326 /* leave this out until we have a page_mkwrite call */
3327 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
3328 EXTENT_DIRTY, 0, NULL)) {
3330 pg_offset += iosize;
3334 if (tree->ops && tree->ops->writepage_io_hook) {
3335 ret = tree->ops->writepage_io_hook(page, cur,
3343 unsigned long max_nr = end_index + 1;
3345 set_range_writeback(tree, cur, cur + iosize - 1);
3346 if (!PageWriteback(page)) {
3347 btrfs_err(BTRFS_I(inode)->root->fs_info,
3348 "page %lu not writeback, cur %llu end %llu",
3349 page->index, cur, end);
3352 ret = submit_extent_page(write_flags, tree, page,
3353 sector, iosize, pg_offset,
3354 bdev, &epd->bio, max_nr,
3355 end_bio_extent_writepage,
3361 pg_offset += iosize;
3366 /* make sure the mapping tag for page dirty gets cleared */
3367 set_page_writeback(page);
3368 end_page_writeback(page);
3370 if (PageError(page)) {
3371 ret = ret < 0 ? ret : -EIO;
3372 end_extent_writepage(page, ret, start, page_end);
3378 /* drop our reference on any cached states */
3379 free_extent_state(cached_state);
3383 static int eb_wait(void *word)
3389 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3391 wait_on_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK, eb_wait,
3392 TASK_UNINTERRUPTIBLE);
3395 static int lock_extent_buffer_for_io(struct extent_buffer *eb,
3396 struct btrfs_fs_info *fs_info,
3397 struct extent_page_data *epd)
3399 unsigned long i, num_pages;
3403 if (!btrfs_try_tree_write_lock(eb)) {
3405 flush_write_bio(epd);
3406 btrfs_tree_lock(eb);
3409 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3410 btrfs_tree_unlock(eb);
3414 flush_write_bio(epd);
3418 wait_on_extent_buffer_writeback(eb);
3419 btrfs_tree_lock(eb);
3420 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3422 btrfs_tree_unlock(eb);
3427 * We need to do this to prevent races in people who check if the eb is
3428 * under IO since we can end up having no IO bits set for a short period
3431 spin_lock(&eb->refs_lock);
3432 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3433 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3434 spin_unlock(&eb->refs_lock);
3435 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3436 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
3438 fs_info->dirty_metadata_batch);
3441 spin_unlock(&eb->refs_lock);
3444 btrfs_tree_unlock(eb);
3449 num_pages = num_extent_pages(eb->start, eb->len);
3450 for (i = 0; i < num_pages; i++) {
3451 struct page *p = extent_buffer_page(eb, i);
3453 if (!trylock_page(p)) {
3455 flush_write_bio(epd);
3465 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3467 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3468 smp_mb__after_clear_bit();
3469 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3472 static void end_bio_extent_buffer_writepage(struct bio *bio, int err)
3474 struct bio_vec *bvec;
3475 struct extent_buffer *eb;
3478 bio_for_each_segment_all(bvec, bio, i) {
3479 struct page *page = bvec->bv_page;
3481 eb = (struct extent_buffer *)page->private;
3483 done = atomic_dec_and_test(&eb->io_pages);
3485 if (err || test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
3486 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3487 ClearPageUptodate(page);
3491 end_page_writeback(page);
3496 end_extent_buffer_writeback(eb);
3502 static int write_one_eb(struct extent_buffer *eb,
3503 struct btrfs_fs_info *fs_info,
3504 struct writeback_control *wbc,
3505 struct extent_page_data *epd)
3507 struct block_device *bdev = fs_info->fs_devices->latest_bdev;
3508 struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
3509 u64 offset = eb->start;
3510 unsigned long i, num_pages;
3511 unsigned long bio_flags = 0;
3512 int rw = (epd->sync_io ? WRITE_SYNC : WRITE) | REQ_META;
3515 clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3516 num_pages = num_extent_pages(eb->start, eb->len);
3517 atomic_set(&eb->io_pages, num_pages);
3518 if (btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID)
3519 bio_flags = EXTENT_BIO_TREE_LOG;
3521 for (i = 0; i < num_pages; i++) {
3522 struct page *p = extent_buffer_page(eb, i);
3524 clear_page_dirty_for_io(p);
3525 set_page_writeback(p);
3526 ret = submit_extent_page(rw, tree, p, offset >> 9,
3527 PAGE_CACHE_SIZE, 0, bdev, &epd->bio,
3528 -1, end_bio_extent_buffer_writepage,
3529 0, epd->bio_flags, bio_flags);
3530 epd->bio_flags = bio_flags;
3532 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3534 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3535 end_extent_buffer_writeback(eb);
3539 offset += PAGE_CACHE_SIZE;
3540 update_nr_written(p, wbc, 1);
3544 if (unlikely(ret)) {
3545 for (; i < num_pages; i++) {
3546 struct page *p = extent_buffer_page(eb, i);
3554 int btree_write_cache_pages(struct address_space *mapping,
3555 struct writeback_control *wbc)
3557 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
3558 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
3559 struct extent_buffer *eb, *prev_eb = NULL;
3560 struct extent_page_data epd = {
3564 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3569 int nr_to_write_done = 0;
3570 struct pagevec pvec;
3573 pgoff_t end; /* Inclusive */
3577 pagevec_init(&pvec, 0);
3578 if (wbc->range_cyclic) {
3579 index = mapping->writeback_index; /* Start from prev offset */
3582 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3583 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3586 if (wbc->sync_mode == WB_SYNC_ALL)
3587 tag = PAGECACHE_TAG_TOWRITE;
3589 tag = PAGECACHE_TAG_DIRTY;
3591 if (wbc->sync_mode == WB_SYNC_ALL)
3592 tag_pages_for_writeback(mapping, index, end);
3593 while (!done && !nr_to_write_done && (index <= end) &&
3594 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3595 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3599 for (i = 0; i < nr_pages; i++) {
3600 struct page *page = pvec.pages[i];
3602 if (!PagePrivate(page))
3605 if (!wbc->range_cyclic && page->index > end) {
3610 spin_lock(&mapping->private_lock);
3611 if (!PagePrivate(page)) {
3612 spin_unlock(&mapping->private_lock);
3616 eb = (struct extent_buffer *)page->private;
3619 * Shouldn't happen and normally this would be a BUG_ON
3620 * but no sense in crashing the users box for something
3621 * we can survive anyway.
3624 spin_unlock(&mapping->private_lock);
3628 if (eb == prev_eb) {
3629 spin_unlock(&mapping->private_lock);
3633 ret = atomic_inc_not_zero(&eb->refs);
3634 spin_unlock(&mapping->private_lock);
3639 ret = lock_extent_buffer_for_io(eb, fs_info, &epd);
3641 free_extent_buffer(eb);
3645 ret = write_one_eb(eb, fs_info, wbc, &epd);
3648 free_extent_buffer(eb);
3651 free_extent_buffer(eb);
3654 * the filesystem may choose to bump up nr_to_write.
3655 * We have to make sure to honor the new nr_to_write
3658 nr_to_write_done = wbc->nr_to_write <= 0;
3660 pagevec_release(&pvec);
3663 if (!scanned && !done) {
3665 * We hit the last page and there is more work to be done: wrap
3666 * back to the start of the file
3672 flush_write_bio(&epd);
3677 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3678 * @mapping: address space structure to write
3679 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3680 * @writepage: function called for each page
3681 * @data: data passed to writepage function
3683 * If a page is already under I/O, write_cache_pages() skips it, even
3684 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3685 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3686 * and msync() need to guarantee that all the data which was dirty at the time
3687 * the call was made get new I/O started against them. If wbc->sync_mode is
3688 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3689 * existing IO to complete.
3691 static int extent_write_cache_pages(struct extent_io_tree *tree,
3692 struct address_space *mapping,
3693 struct writeback_control *wbc,
3694 writepage_t writepage, void *data,
3695 void (*flush_fn)(void *))
3697 struct inode *inode = mapping->host;
3701 int nr_to_write_done = 0;
3702 struct pagevec pvec;
3705 pgoff_t end; /* Inclusive */
3710 * We have to hold onto the inode so that ordered extents can do their
3711 * work when the IO finishes. The alternative to this is failing to add
3712 * an ordered extent if the igrab() fails there and that is a huge pain
3713 * to deal with, so instead just hold onto the inode throughout the
3714 * writepages operation. If it fails here we are freeing up the inode
3715 * anyway and we'd rather not waste our time writing out stuff that is
3716 * going to be truncated anyway.
3721 pagevec_init(&pvec, 0);
3722 if (wbc->range_cyclic) {
3723 index = mapping->writeback_index; /* Start from prev offset */
3726 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3727 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3730 if (wbc->sync_mode == WB_SYNC_ALL)
3731 tag = PAGECACHE_TAG_TOWRITE;
3733 tag = PAGECACHE_TAG_DIRTY;
3735 if (wbc->sync_mode == WB_SYNC_ALL)
3736 tag_pages_for_writeback(mapping, index, end);
3737 while (!done && !nr_to_write_done && (index <= end) &&
3738 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3739 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3743 for (i = 0; i < nr_pages; i++) {
3744 struct page *page = pvec.pages[i];
3747 * At this point we hold neither mapping->tree_lock nor
3748 * lock on the page itself: the page may be truncated or
3749 * invalidated (changing page->mapping to NULL), or even
3750 * swizzled back from swapper_space to tmpfs file
3753 if (!trylock_page(page)) {
3758 if (unlikely(page->mapping != mapping)) {
3763 if (!wbc->range_cyclic && page->index > end) {
3769 if (wbc->sync_mode != WB_SYNC_NONE) {
3770 if (PageWriteback(page))
3772 wait_on_page_writeback(page);
3775 if (PageWriteback(page) ||
3776 !clear_page_dirty_for_io(page)) {
3781 ret = (*writepage)(page, wbc, data);
3783 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
3787 if (!err && ret < 0)
3791 * the filesystem may choose to bump up nr_to_write.
3792 * We have to make sure to honor the new nr_to_write
3795 nr_to_write_done = wbc->nr_to_write <= 0;
3797 pagevec_release(&pvec);
3800 if (!scanned && !done && !err) {
3802 * We hit the last page and there is more work to be done: wrap
3803 * back to the start of the file
3809 btrfs_add_delayed_iput(inode);
3813 static void flush_epd_write_bio(struct extent_page_data *epd)
3822 ret = submit_one_bio(rw, epd->bio, 0, epd->bio_flags);
3823 BUG_ON(ret < 0); /* -ENOMEM */
3828 static noinline void flush_write_bio(void *data)
3830 struct extent_page_data *epd = data;
3831 flush_epd_write_bio(epd);
3834 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
3835 get_extent_t *get_extent,
3836 struct writeback_control *wbc)
3839 struct extent_page_data epd = {
3842 .get_extent = get_extent,
3844 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3848 ret = __extent_writepage(page, wbc, &epd);
3850 flush_epd_write_bio(&epd);
3854 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
3855 u64 start, u64 end, get_extent_t *get_extent,
3859 struct address_space *mapping = inode->i_mapping;
3861 unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
3864 struct extent_page_data epd = {
3867 .get_extent = get_extent,
3869 .sync_io = mode == WB_SYNC_ALL,
3872 struct writeback_control wbc_writepages = {
3874 .nr_to_write = nr_pages * 2,
3875 .range_start = start,
3876 .range_end = end + 1,
3879 while (start <= end) {
3880 page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
3881 if (clear_page_dirty_for_io(page))
3882 ret = __extent_writepage(page, &wbc_writepages, &epd);
3884 if (tree->ops && tree->ops->writepage_end_io_hook)
3885 tree->ops->writepage_end_io_hook(page, start,
3886 start + PAGE_CACHE_SIZE - 1,
3890 page_cache_release(page);
3891 start += PAGE_CACHE_SIZE;
3894 flush_epd_write_bio(&epd);
3898 int extent_writepages(struct extent_io_tree *tree,
3899 struct address_space *mapping,
3900 get_extent_t *get_extent,
3901 struct writeback_control *wbc)
3904 struct extent_page_data epd = {
3907 .get_extent = get_extent,
3909 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3913 ret = extent_write_cache_pages(tree, mapping, wbc,
3914 __extent_writepage, &epd,
3916 flush_epd_write_bio(&epd);
3920 int extent_readpages(struct extent_io_tree *tree,
3921 struct address_space *mapping,
3922 struct list_head *pages, unsigned nr_pages,
3923 get_extent_t get_extent)
3925 struct bio *bio = NULL;
3927 unsigned long bio_flags = 0;
3928 struct page *pagepool[16];
3930 struct extent_map *em_cached = NULL;
3933 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
3934 page = list_entry(pages->prev, struct page, lru);
3936 prefetchw(&page->flags);
3937 list_del(&page->lru);
3938 if (add_to_page_cache_lru(page, mapping,
3939 page->index, GFP_NOFS)) {
3940 page_cache_release(page);
3944 pagepool[nr++] = page;
3945 if (nr < ARRAY_SIZE(pagepool))
3947 __extent_readpages(tree, pagepool, nr, get_extent, &em_cached,
3948 &bio, 0, &bio_flags, READ);
3952 __extent_readpages(tree, pagepool, nr, get_extent, &em_cached,
3953 &bio, 0, &bio_flags, READ);
3956 free_extent_map(em_cached);
3958 BUG_ON(!list_empty(pages));
3960 return submit_one_bio(READ, bio, 0, bio_flags);
3965 * basic invalidatepage code, this waits on any locked or writeback
3966 * ranges corresponding to the page, and then deletes any extent state
3967 * records from the tree
3969 int extent_invalidatepage(struct extent_io_tree *tree,
3970 struct page *page, unsigned long offset)
3972 struct extent_state *cached_state = NULL;
3973 u64 start = page_offset(page);
3974 u64 end = start + PAGE_CACHE_SIZE - 1;
3975 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
3977 start += ALIGN(offset, blocksize);
3981 lock_extent_bits(tree, start, end, 0, &cached_state);
3982 wait_on_page_writeback(page);
3983 clear_extent_bit(tree, start, end,
3984 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
3985 EXTENT_DO_ACCOUNTING,
3986 1, 1, &cached_state, GFP_NOFS);
3991 * a helper for releasepage, this tests for areas of the page that
3992 * are locked or under IO and drops the related state bits if it is safe
3995 static int try_release_extent_state(struct extent_map_tree *map,
3996 struct extent_io_tree *tree,
3997 struct page *page, gfp_t mask)
3999 u64 start = page_offset(page);
4000 u64 end = start + PAGE_CACHE_SIZE - 1;
4003 if (test_range_bit(tree, start, end,
4004 EXTENT_IOBITS, 0, NULL))
4007 if ((mask & GFP_NOFS) == GFP_NOFS)
4010 * at this point we can safely clear everything except the
4011 * locked bit and the nodatasum bit
4013 ret = clear_extent_bit(tree, start, end,
4014 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
4017 /* if clear_extent_bit failed for enomem reasons,
4018 * we can't allow the release to continue.
4029 * a helper for releasepage. As long as there are no locked extents
4030 * in the range corresponding to the page, both state records and extent
4031 * map records are removed
4033 int try_release_extent_mapping(struct extent_map_tree *map,
4034 struct extent_io_tree *tree, struct page *page,
4037 struct extent_map *em;
4038 u64 start = page_offset(page);
4039 u64 end = start + PAGE_CACHE_SIZE - 1;
4041 if ((mask & __GFP_WAIT) &&
4042 page->mapping->host->i_size > 16 * 1024 * 1024) {
4044 while (start <= end) {
4045 len = end - start + 1;
4046 write_lock(&map->lock);
4047 em = lookup_extent_mapping(map, start, len);
4049 write_unlock(&map->lock);
4052 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
4053 em->start != start) {
4054 write_unlock(&map->lock);
4055 free_extent_map(em);
4058 if (!test_range_bit(tree, em->start,
4059 extent_map_end(em) - 1,
4060 EXTENT_LOCKED | EXTENT_WRITEBACK,
4062 remove_extent_mapping(map, em);
4063 /* once for the rb tree */
4064 free_extent_map(em);
4066 start = extent_map_end(em);
4067 write_unlock(&map->lock);
4070 free_extent_map(em);
4073 return try_release_extent_state(map, tree, page, mask);
4077 * helper function for fiemap, which doesn't want to see any holes.
4078 * This maps until we find something past 'last'
4080 static struct extent_map *get_extent_skip_holes(struct inode *inode,
4083 get_extent_t *get_extent)
4085 u64 sectorsize = BTRFS_I(inode)->root->sectorsize;
4086 struct extent_map *em;
4093 len = last - offset;
4096 len = ALIGN(len, sectorsize);
4097 em = get_extent(inode, NULL, 0, offset, len, 0);
4098 if (IS_ERR_OR_NULL(em))
4101 /* if this isn't a hole return it */
4102 if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) &&
4103 em->block_start != EXTENT_MAP_HOLE) {
4107 /* this is a hole, advance to the next extent */
4108 offset = extent_map_end(em);
4109 free_extent_map(em);
4116 static noinline int count_ext_ref(u64 inum, u64 offset, u64 root_id, void *ctx)
4118 unsigned long cnt = *((unsigned long *)ctx);
4121 *((unsigned long *)ctx) = cnt;
4123 /* Now we're sure that the extent is shared. */
4129 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4130 __u64 start, __u64 len, get_extent_t *get_extent)
4134 u64 max = start + len;
4138 u64 last_for_get_extent = 0;
4140 u64 isize = i_size_read(inode);
4141 struct btrfs_key found_key;
4142 struct extent_map *em = NULL;
4143 struct extent_state *cached_state = NULL;
4144 struct btrfs_path *path;
4153 path = btrfs_alloc_path();
4156 path->leave_spinning = 1;
4158 start = ALIGN(start, BTRFS_I(inode)->root->sectorsize);
4159 len = ALIGN(len, BTRFS_I(inode)->root->sectorsize);
4162 * lookup the last file extent. We're not using i_size here
4163 * because there might be preallocation past i_size
4165 ret = btrfs_lookup_file_extent(NULL, BTRFS_I(inode)->root,
4166 path, btrfs_ino(inode), -1, 0);
4168 btrfs_free_path(path);
4173 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
4174 found_type = btrfs_key_type(&found_key);
4176 /* No extents, but there might be delalloc bits */
4177 if (found_key.objectid != btrfs_ino(inode) ||
4178 found_type != BTRFS_EXTENT_DATA_KEY) {
4179 /* have to trust i_size as the end */
4181 last_for_get_extent = isize;
4184 * remember the start of the last extent. There are a
4185 * bunch of different factors that go into the length of the
4186 * extent, so its much less complex to remember where it started
4188 last = found_key.offset;
4189 last_for_get_extent = last + 1;
4191 btrfs_release_path(path);
4194 * we might have some extents allocated but more delalloc past those
4195 * extents. so, we trust isize unless the start of the last extent is
4200 last_for_get_extent = isize;
4203 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1, 0,
4206 em = get_extent_skip_holes(inode, start, last_for_get_extent,
4216 u64 offset_in_extent = 0;
4218 /* break if the extent we found is outside the range */
4219 if (em->start >= max || extent_map_end(em) < off)
4223 * get_extent may return an extent that starts before our
4224 * requested range. We have to make sure the ranges
4225 * we return to fiemap always move forward and don't
4226 * overlap, so adjust the offsets here
4228 em_start = max(em->start, off);
4231 * record the offset from the start of the extent
4232 * for adjusting the disk offset below. Only do this if the
4233 * extent isn't compressed since our in ram offset may be past
4234 * what we have actually allocated on disk.
4236 if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4237 offset_in_extent = em_start - em->start;
4238 em_end = extent_map_end(em);
4239 em_len = em_end - em_start;
4244 * bump off for our next call to get_extent
4246 off = extent_map_end(em);
4250 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
4252 flags |= FIEMAP_EXTENT_LAST;
4253 } else if (em->block_start == EXTENT_MAP_INLINE) {
4254 flags |= (FIEMAP_EXTENT_DATA_INLINE |
4255 FIEMAP_EXTENT_NOT_ALIGNED);
4256 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
4257 flags |= (FIEMAP_EXTENT_DELALLOC |
4258 FIEMAP_EXTENT_UNKNOWN);
4260 unsigned long ref_cnt = 0;
4262 disko = em->block_start + offset_in_extent;
4265 * As btrfs supports shared space, this information
4266 * can be exported to userspace tools via
4267 * flag FIEMAP_EXTENT_SHARED.
4269 ret = iterate_inodes_from_logical(
4271 BTRFS_I(inode)->root->fs_info,
4272 path, count_ext_ref, &ref_cnt);
4273 if (ret < 0 && ret != -ENOENT)
4277 flags |= FIEMAP_EXTENT_SHARED;
4279 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4280 flags |= FIEMAP_EXTENT_ENCODED;
4282 free_extent_map(em);
4284 if ((em_start >= last) || em_len == (u64)-1 ||
4285 (last == (u64)-1 && isize <= em_end)) {
4286 flags |= FIEMAP_EXTENT_LAST;
4290 /* now scan forward to see if this is really the last extent. */
4291 em = get_extent_skip_holes(inode, off, last_for_get_extent,
4298 flags |= FIEMAP_EXTENT_LAST;
4301 ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
4307 free_extent_map(em);
4309 btrfs_free_path(path);
4310 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4311 &cached_state, GFP_NOFS);
4315 static void __free_extent_buffer(struct extent_buffer *eb)
4317 btrfs_leak_debug_del(&eb->leak_list);
4318 kmem_cache_free(extent_buffer_cache, eb);
4321 int extent_buffer_under_io(struct extent_buffer *eb)
4323 return (atomic_read(&eb->io_pages) ||
4324 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4325 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4329 * Helper for releasing extent buffer page.
4331 static void btrfs_release_extent_buffer_page(struct extent_buffer *eb,
4332 unsigned long start_idx)
4334 unsigned long index;
4335 unsigned long num_pages;
4337 int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4339 BUG_ON(extent_buffer_under_io(eb));
4341 num_pages = num_extent_pages(eb->start, eb->len);
4342 index = start_idx + num_pages;
4343 if (start_idx >= index)
4348 page = extent_buffer_page(eb, index);
4349 if (page && mapped) {
4350 spin_lock(&page->mapping->private_lock);
4352 * We do this since we'll remove the pages after we've
4353 * removed the eb from the radix tree, so we could race
4354 * and have this page now attached to the new eb. So
4355 * only clear page_private if it's still connected to
4358 if (PagePrivate(page) &&
4359 page->private == (unsigned long)eb) {
4360 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4361 BUG_ON(PageDirty(page));
4362 BUG_ON(PageWriteback(page));
4364 * We need to make sure we haven't be attached
4367 ClearPagePrivate(page);
4368 set_page_private(page, 0);
4369 /* One for the page private */
4370 page_cache_release(page);
4372 spin_unlock(&page->mapping->private_lock);
4376 /* One for when we alloced the page */
4377 page_cache_release(page);
4379 } while (index != start_idx);
4383 * Helper for releasing the extent buffer.
4385 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4387 btrfs_release_extent_buffer_page(eb, 0);
4388 __free_extent_buffer(eb);
4391 static struct extent_buffer *
4392 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
4393 unsigned long len, gfp_t mask)
4395 struct extent_buffer *eb = NULL;
4397 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
4402 eb->fs_info = fs_info;
4404 rwlock_init(&eb->lock);
4405 atomic_set(&eb->write_locks, 0);
4406 atomic_set(&eb->read_locks, 0);
4407 atomic_set(&eb->blocking_readers, 0);
4408 atomic_set(&eb->blocking_writers, 0);
4409 atomic_set(&eb->spinning_readers, 0);
4410 atomic_set(&eb->spinning_writers, 0);
4411 eb->lock_nested = 0;
4412 init_waitqueue_head(&eb->write_lock_wq);
4413 init_waitqueue_head(&eb->read_lock_wq);
4415 btrfs_leak_debug_add(&eb->leak_list, &buffers);
4417 spin_lock_init(&eb->refs_lock);
4418 atomic_set(&eb->refs, 1);
4419 atomic_set(&eb->io_pages, 0);
4422 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4424 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4425 > MAX_INLINE_EXTENT_BUFFER_SIZE);
4426 BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE);
4431 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src)
4435 struct extent_buffer *new;
4436 unsigned long num_pages = num_extent_pages(src->start, src->len);
4438 new = __alloc_extent_buffer(NULL, src->start, src->len, GFP_NOFS);
4442 for (i = 0; i < num_pages; i++) {
4443 p = alloc_page(GFP_NOFS);
4445 btrfs_release_extent_buffer(new);
4448 attach_extent_buffer_page(new, p);
4449 WARN_ON(PageDirty(p));
4454 copy_extent_buffer(new, src, 0, 0, src->len);
4455 set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
4456 set_bit(EXTENT_BUFFER_DUMMY, &new->bflags);
4461 struct extent_buffer *alloc_dummy_extent_buffer(u64 start, unsigned long len)
4463 struct extent_buffer *eb;
4464 unsigned long num_pages = num_extent_pages(0, len);
4467 eb = __alloc_extent_buffer(NULL, start, len, GFP_NOFS);
4471 for (i = 0; i < num_pages; i++) {
4472 eb->pages[i] = alloc_page(GFP_NOFS);
4476 set_extent_buffer_uptodate(eb);
4477 btrfs_set_header_nritems(eb, 0);
4478 set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4483 __free_page(eb->pages[i - 1]);
4484 __free_extent_buffer(eb);
4488 static void check_buffer_tree_ref(struct extent_buffer *eb)
4491 /* the ref bit is tricky. We have to make sure it is set
4492 * if we have the buffer dirty. Otherwise the
4493 * code to free a buffer can end up dropping a dirty
4496 * Once the ref bit is set, it won't go away while the
4497 * buffer is dirty or in writeback, and it also won't
4498 * go away while we have the reference count on the
4501 * We can't just set the ref bit without bumping the
4502 * ref on the eb because free_extent_buffer might
4503 * see the ref bit and try to clear it. If this happens
4504 * free_extent_buffer might end up dropping our original
4505 * ref by mistake and freeing the page before we are able
4506 * to add one more ref.
4508 * So bump the ref count first, then set the bit. If someone
4509 * beat us to it, drop the ref we added.
4511 refs = atomic_read(&eb->refs);
4512 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4515 spin_lock(&eb->refs_lock);
4516 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4517 atomic_inc(&eb->refs);
4518 spin_unlock(&eb->refs_lock);
4521 static void mark_extent_buffer_accessed(struct extent_buffer *eb)
4523 unsigned long num_pages, i;
4525 check_buffer_tree_ref(eb);
4527 num_pages = num_extent_pages(eb->start, eb->len);
4528 for (i = 0; i < num_pages; i++) {
4529 struct page *p = extent_buffer_page(eb, i);
4530 mark_page_accessed(p);
4534 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
4537 struct extent_buffer *eb;
4540 eb = radix_tree_lookup(&fs_info->buffer_radix,
4541 start >> PAGE_CACHE_SHIFT);
4542 if (eb && atomic_inc_not_zero(&eb->refs)) {
4544 mark_extent_buffer_accessed(eb);
4552 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4553 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
4554 u64 start, unsigned long len)
4556 struct extent_buffer *eb, *exists = NULL;
4559 eb = find_extent_buffer(fs_info, start);
4562 eb = alloc_dummy_extent_buffer(start, len);
4565 eb->fs_info = fs_info;
4567 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
4570 spin_lock(&fs_info->buffer_lock);
4571 ret = radix_tree_insert(&fs_info->buffer_radix,
4572 start >> PAGE_CACHE_SHIFT, eb);
4573 spin_unlock(&fs_info->buffer_lock);
4574 radix_tree_preload_end();
4575 if (ret == -EEXIST) {
4576 exists = find_extent_buffer(fs_info, start);
4582 check_buffer_tree_ref(eb);
4583 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
4586 * We will free dummy extent buffer's if they come into
4587 * free_extent_buffer with a ref count of 2, but if we are using this we
4588 * want the buffers to stay in memory until we're done with them, so
4589 * bump the ref count again.
4591 atomic_inc(&eb->refs);
4594 btrfs_release_extent_buffer(eb);
4599 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
4600 u64 start, unsigned long len)
4602 unsigned long num_pages = num_extent_pages(start, len);
4604 unsigned long index = start >> PAGE_CACHE_SHIFT;
4605 struct extent_buffer *eb;
4606 struct extent_buffer *exists = NULL;
4608 struct address_space *mapping = fs_info->btree_inode->i_mapping;
4612 eb = find_extent_buffer(fs_info, start);
4616 eb = __alloc_extent_buffer(fs_info, start, len, GFP_NOFS);
4620 for (i = 0; i < num_pages; i++, index++) {
4621 p = find_or_create_page(mapping, index, GFP_NOFS);
4625 spin_lock(&mapping->private_lock);
4626 if (PagePrivate(p)) {
4628 * We could have already allocated an eb for this page
4629 * and attached one so lets see if we can get a ref on
4630 * the existing eb, and if we can we know it's good and
4631 * we can just return that one, else we know we can just
4632 * overwrite page->private.
4634 exists = (struct extent_buffer *)p->private;
4635 if (atomic_inc_not_zero(&exists->refs)) {
4636 spin_unlock(&mapping->private_lock);
4638 page_cache_release(p);
4639 mark_extent_buffer_accessed(exists);
4644 * Do this so attach doesn't complain and we need to
4645 * drop the ref the old guy had.
4647 ClearPagePrivate(p);
4648 WARN_ON(PageDirty(p));
4649 page_cache_release(p);
4651 attach_extent_buffer_page(eb, p);
4652 spin_unlock(&mapping->private_lock);
4653 WARN_ON(PageDirty(p));
4654 mark_page_accessed(p);
4656 if (!PageUptodate(p))
4660 * see below about how we avoid a nasty race with release page
4661 * and why we unlock later
4665 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4667 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
4671 spin_lock(&fs_info->buffer_lock);
4672 ret = radix_tree_insert(&fs_info->buffer_radix,
4673 start >> PAGE_CACHE_SHIFT, eb);
4674 spin_unlock(&fs_info->buffer_lock);
4675 radix_tree_preload_end();
4676 if (ret == -EEXIST) {
4677 exists = find_extent_buffer(fs_info, start);
4683 /* add one reference for the tree */
4684 check_buffer_tree_ref(eb);
4685 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
4688 * there is a race where release page may have
4689 * tried to find this extent buffer in the radix
4690 * but failed. It will tell the VM it is safe to
4691 * reclaim the, and it will clear the page private bit.
4692 * We must make sure to set the page private bit properly
4693 * after the extent buffer is in the radix tree so
4694 * it doesn't get lost
4696 SetPageChecked(eb->pages[0]);
4697 for (i = 1; i < num_pages; i++) {
4698 p = extent_buffer_page(eb, i);
4699 ClearPageChecked(p);
4702 unlock_page(eb->pages[0]);
4706 for (i = 0; i < num_pages; i++) {
4708 unlock_page(eb->pages[i]);
4711 WARN_ON(!atomic_dec_and_test(&eb->refs));
4712 btrfs_release_extent_buffer(eb);
4716 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
4718 struct extent_buffer *eb =
4719 container_of(head, struct extent_buffer, rcu_head);
4721 __free_extent_buffer(eb);
4724 /* Expects to have eb->eb_lock already held */
4725 static int release_extent_buffer(struct extent_buffer *eb)
4727 WARN_ON(atomic_read(&eb->refs) == 0);
4728 if (atomic_dec_and_test(&eb->refs)) {
4729 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
4730 struct btrfs_fs_info *fs_info = eb->fs_info;
4732 spin_unlock(&eb->refs_lock);
4734 spin_lock(&fs_info->buffer_lock);
4735 radix_tree_delete(&fs_info->buffer_radix,
4736 eb->start >> PAGE_CACHE_SHIFT);
4737 spin_unlock(&fs_info->buffer_lock);
4739 spin_unlock(&eb->refs_lock);
4742 /* Should be safe to release our pages at this point */
4743 btrfs_release_extent_buffer_page(eb, 0);
4744 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
4747 spin_unlock(&eb->refs_lock);
4752 void free_extent_buffer(struct extent_buffer *eb)
4760 refs = atomic_read(&eb->refs);
4763 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
4768 spin_lock(&eb->refs_lock);
4769 if (atomic_read(&eb->refs) == 2 &&
4770 test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))
4771 atomic_dec(&eb->refs);
4773 if (atomic_read(&eb->refs) == 2 &&
4774 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
4775 !extent_buffer_under_io(eb) &&
4776 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4777 atomic_dec(&eb->refs);
4780 * I know this is terrible, but it's temporary until we stop tracking
4781 * the uptodate bits and such for the extent buffers.
4783 release_extent_buffer(eb);
4786 void free_extent_buffer_stale(struct extent_buffer *eb)
4791 spin_lock(&eb->refs_lock);
4792 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
4794 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
4795 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4796 atomic_dec(&eb->refs);
4797 release_extent_buffer(eb);
4800 void clear_extent_buffer_dirty(struct extent_buffer *eb)
4803 unsigned long num_pages;
4806 num_pages = num_extent_pages(eb->start, eb->len);
4808 for (i = 0; i < num_pages; i++) {
4809 page = extent_buffer_page(eb, i);
4810 if (!PageDirty(page))
4814 WARN_ON(!PagePrivate(page));
4816 clear_page_dirty_for_io(page);
4817 spin_lock_irq(&page->mapping->tree_lock);
4818 if (!PageDirty(page)) {
4819 radix_tree_tag_clear(&page->mapping->page_tree,
4821 PAGECACHE_TAG_DIRTY);
4823 spin_unlock_irq(&page->mapping->tree_lock);
4824 ClearPageError(page);
4827 WARN_ON(atomic_read(&eb->refs) == 0);
4830 int set_extent_buffer_dirty(struct extent_buffer *eb)
4833 unsigned long num_pages;
4836 check_buffer_tree_ref(eb);
4838 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
4840 num_pages = num_extent_pages(eb->start, eb->len);
4841 WARN_ON(atomic_read(&eb->refs) == 0);
4842 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
4844 for (i = 0; i < num_pages; i++)
4845 set_page_dirty(extent_buffer_page(eb, i));
4849 int clear_extent_buffer_uptodate(struct extent_buffer *eb)
4853 unsigned long num_pages;
4855 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4856 num_pages = num_extent_pages(eb->start, eb->len);
4857 for (i = 0; i < num_pages; i++) {
4858 page = extent_buffer_page(eb, i);
4860 ClearPageUptodate(page);
4865 int set_extent_buffer_uptodate(struct extent_buffer *eb)
4869 unsigned long num_pages;
4871 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4872 num_pages = num_extent_pages(eb->start, eb->len);
4873 for (i = 0; i < num_pages; i++) {
4874 page = extent_buffer_page(eb, i);
4875 SetPageUptodate(page);
4880 int extent_buffer_uptodate(struct extent_buffer *eb)
4882 return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4885 int read_extent_buffer_pages(struct extent_io_tree *tree,
4886 struct extent_buffer *eb, u64 start, int wait,
4887 get_extent_t *get_extent, int mirror_num)
4890 unsigned long start_i;
4894 int locked_pages = 0;
4895 int all_uptodate = 1;
4896 unsigned long num_pages;
4897 unsigned long num_reads = 0;
4898 struct bio *bio = NULL;
4899 unsigned long bio_flags = 0;
4901 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
4905 WARN_ON(start < eb->start);
4906 start_i = (start >> PAGE_CACHE_SHIFT) -
4907 (eb->start >> PAGE_CACHE_SHIFT);
4912 num_pages = num_extent_pages(eb->start, eb->len);
4913 for (i = start_i; i < num_pages; i++) {
4914 page = extent_buffer_page(eb, i);
4915 if (wait == WAIT_NONE) {
4916 if (!trylock_page(page))
4922 if (!PageUptodate(page)) {
4929 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4933 clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
4934 eb->read_mirror = 0;
4935 atomic_set(&eb->io_pages, num_reads);
4936 for (i = start_i; i < num_pages; i++) {
4937 page = extent_buffer_page(eb, i);
4938 if (!PageUptodate(page)) {
4939 ClearPageError(page);
4940 err = __extent_read_full_page(tree, page,
4942 mirror_num, &bio_flags,
4952 err = submit_one_bio(READ | REQ_META, bio, mirror_num,
4958 if (ret || wait != WAIT_COMPLETE)
4961 for (i = start_i; i < num_pages; i++) {
4962 page = extent_buffer_page(eb, i);
4963 wait_on_page_locked(page);
4964 if (!PageUptodate(page))
4972 while (locked_pages > 0) {
4973 page = extent_buffer_page(eb, i);
4981 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
4982 unsigned long start,
4989 char *dst = (char *)dstv;
4990 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4991 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4993 WARN_ON(start > eb->len);
4994 WARN_ON(start + len > eb->start + eb->len);
4996 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
4999 page = extent_buffer_page(eb, i);
5001 cur = min(len, (PAGE_CACHE_SIZE - offset));
5002 kaddr = page_address(page);
5003 memcpy(dst, kaddr + offset, cur);
5012 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
5013 unsigned long min_len, char **map,
5014 unsigned long *map_start,
5015 unsigned long *map_len)
5017 size_t offset = start & (PAGE_CACHE_SIZE - 1);
5020 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5021 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5022 unsigned long end_i = (start_offset + start + min_len - 1) >>
5029 offset = start_offset;
5033 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
5036 if (start + min_len > eb->len) {
5037 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, "
5039 eb->start, eb->len, start, min_len);
5043 p = extent_buffer_page(eb, i);
5044 kaddr = page_address(p);
5045 *map = kaddr + offset;
5046 *map_len = PAGE_CACHE_SIZE - offset;
5050 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
5051 unsigned long start,
5058 char *ptr = (char *)ptrv;
5059 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5060 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5063 WARN_ON(start > eb->len);
5064 WARN_ON(start + len > eb->start + eb->len);
5066 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5069 page = extent_buffer_page(eb, i);
5071 cur = min(len, (PAGE_CACHE_SIZE - offset));
5073 kaddr = page_address(page);
5074 ret = memcmp(ptr, kaddr + offset, cur);
5086 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
5087 unsigned long start, unsigned long len)
5093 char *src = (char *)srcv;
5094 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5095 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5097 WARN_ON(start > eb->len);
5098 WARN_ON(start + len > eb->start + eb->len);
5100 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5103 page = extent_buffer_page(eb, i);
5104 WARN_ON(!PageUptodate(page));
5106 cur = min(len, PAGE_CACHE_SIZE - offset);
5107 kaddr = page_address(page);
5108 memcpy(kaddr + offset, src, cur);
5117 void memset_extent_buffer(struct extent_buffer *eb, char c,
5118 unsigned long start, unsigned long len)
5124 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5125 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5127 WARN_ON(start > eb->len);
5128 WARN_ON(start + len > eb->start + eb->len);
5130 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5133 page = extent_buffer_page(eb, i);
5134 WARN_ON(!PageUptodate(page));
5136 cur = min(len, PAGE_CACHE_SIZE - offset);
5137 kaddr = page_address(page);
5138 memset(kaddr + offset, c, cur);
5146 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
5147 unsigned long dst_offset, unsigned long src_offset,
5150 u64 dst_len = dst->len;
5155 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5156 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
5158 WARN_ON(src->len != dst_len);
5160 offset = (start_offset + dst_offset) &
5161 (PAGE_CACHE_SIZE - 1);
5164 page = extent_buffer_page(dst, i);
5165 WARN_ON(!PageUptodate(page));
5167 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
5169 kaddr = page_address(page);
5170 read_extent_buffer(src, kaddr + offset, src_offset, cur);
5179 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
5181 unsigned long distance = (src > dst) ? src - dst : dst - src;
5182 return distance < len;
5185 static void copy_pages(struct page *dst_page, struct page *src_page,
5186 unsigned long dst_off, unsigned long src_off,
5189 char *dst_kaddr = page_address(dst_page);
5191 int must_memmove = 0;
5193 if (dst_page != src_page) {
5194 src_kaddr = page_address(src_page);
5196 src_kaddr = dst_kaddr;
5197 if (areas_overlap(src_off, dst_off, len))
5202 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
5204 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
5207 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5208 unsigned long src_offset, unsigned long len)
5211 size_t dst_off_in_page;
5212 size_t src_off_in_page;
5213 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5214 unsigned long dst_i;
5215 unsigned long src_i;
5217 if (src_offset + len > dst->len) {
5218 printk(KERN_ERR "BTRFS: memmove bogus src_offset %lu move "
5219 "len %lu dst len %lu\n", src_offset, len, dst->len);
5222 if (dst_offset + len > dst->len) {
5223 printk(KERN_ERR "BTRFS: memmove bogus dst_offset %lu move "
5224 "len %lu dst len %lu\n", dst_offset, len, dst->len);
5229 dst_off_in_page = (start_offset + dst_offset) &
5230 (PAGE_CACHE_SIZE - 1);
5231 src_off_in_page = (start_offset + src_offset) &
5232 (PAGE_CACHE_SIZE - 1);
5234 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
5235 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
5237 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
5239 cur = min_t(unsigned long, cur,
5240 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
5242 copy_pages(extent_buffer_page(dst, dst_i),
5243 extent_buffer_page(dst, src_i),
5244 dst_off_in_page, src_off_in_page, cur);
5252 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5253 unsigned long src_offset, unsigned long len)
5256 size_t dst_off_in_page;
5257 size_t src_off_in_page;
5258 unsigned long dst_end = dst_offset + len - 1;
5259 unsigned long src_end = src_offset + len - 1;
5260 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5261 unsigned long dst_i;
5262 unsigned long src_i;
5264 if (src_offset + len > dst->len) {
5265 printk(KERN_ERR "BTRFS: memmove bogus src_offset %lu move "
5266 "len %lu len %lu\n", src_offset, len, dst->len);
5269 if (dst_offset + len > dst->len) {
5270 printk(KERN_ERR "BTRFS: memmove bogus dst_offset %lu move "
5271 "len %lu len %lu\n", dst_offset, len, dst->len);
5274 if (dst_offset < src_offset) {
5275 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
5279 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
5280 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
5282 dst_off_in_page = (start_offset + dst_end) &
5283 (PAGE_CACHE_SIZE - 1);
5284 src_off_in_page = (start_offset + src_end) &
5285 (PAGE_CACHE_SIZE - 1);
5287 cur = min_t(unsigned long, len, src_off_in_page + 1);
5288 cur = min(cur, dst_off_in_page + 1);
5289 copy_pages(extent_buffer_page(dst, dst_i),
5290 extent_buffer_page(dst, src_i),
5291 dst_off_in_page - cur + 1,
5292 src_off_in_page - cur + 1, cur);
5300 int try_release_extent_buffer(struct page *page)
5302 struct extent_buffer *eb;
5305 * We need to make sure noboody is attaching this page to an eb right
5308 spin_lock(&page->mapping->private_lock);
5309 if (!PagePrivate(page)) {
5310 spin_unlock(&page->mapping->private_lock);
5314 eb = (struct extent_buffer *)page->private;
5318 * This is a little awful but should be ok, we need to make sure that
5319 * the eb doesn't disappear out from under us while we're looking at
5322 spin_lock(&eb->refs_lock);
5323 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
5324 spin_unlock(&eb->refs_lock);
5325 spin_unlock(&page->mapping->private_lock);
5328 spin_unlock(&page->mapping->private_lock);
5331 * If tree ref isn't set then we know the ref on this eb is a real ref,
5332 * so just return, this page will likely be freed soon anyway.
5334 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
5335 spin_unlock(&eb->refs_lock);
5339 return release_extent_buffer(eb);