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 static inline bool extent_state_in_tree(const struct extent_state *state)
30 return !RB_EMPTY_NODE(&state->rb_node);
33 #ifdef CONFIG_BTRFS_DEBUG
34 static LIST_HEAD(buffers);
35 static LIST_HEAD(states);
37 static DEFINE_SPINLOCK(leak_lock);
40 void btrfs_leak_debug_add(struct list_head *new, struct list_head *head)
44 spin_lock_irqsave(&leak_lock, flags);
46 spin_unlock_irqrestore(&leak_lock, flags);
50 void btrfs_leak_debug_del(struct list_head *entry)
54 spin_lock_irqsave(&leak_lock, flags);
56 spin_unlock_irqrestore(&leak_lock, flags);
60 void btrfs_leak_debug_check(void)
62 struct extent_state *state;
63 struct extent_buffer *eb;
65 while (!list_empty(&states)) {
66 state = list_entry(states.next, struct extent_state, leak_list);
67 pr_err("BTRFS: state leak: start %llu end %llu state %u in tree %d refs %d\n",
68 state->start, state->end, state->state,
69 extent_state_in_tree(state),
70 atomic_read(&state->refs));
71 list_del(&state->leak_list);
72 kmem_cache_free(extent_state_cache, state);
75 while (!list_empty(&buffers)) {
76 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
77 printk(KERN_ERR "BTRFS: buffer leak start %llu len %lu "
79 eb->start, eb->len, atomic_read(&eb->refs));
80 list_del(&eb->leak_list);
81 kmem_cache_free(extent_buffer_cache, eb);
85 #define btrfs_debug_check_extent_io_range(tree, start, end) \
86 __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end))
87 static inline void __btrfs_debug_check_extent_io_range(const char *caller,
88 struct extent_io_tree *tree, u64 start, u64 end)
96 inode = tree->mapping->host;
97 isize = i_size_read(inode);
98 if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) {
99 btrfs_debug_rl(BTRFS_I(inode)->root->fs_info,
100 "%s: ino %llu isize %llu odd range [%llu,%llu]",
101 caller, btrfs_ino(inode), isize, start, end);
105 #define btrfs_leak_debug_add(new, head) do {} while (0)
106 #define btrfs_leak_debug_del(entry) do {} while (0)
107 #define btrfs_leak_debug_check() do {} while (0)
108 #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0)
111 #define BUFFER_LRU_MAX 64
116 struct rb_node rb_node;
119 struct extent_page_data {
121 struct extent_io_tree *tree;
122 get_extent_t *get_extent;
123 unsigned long bio_flags;
125 /* tells writepage not to lock the state bits for this range
126 * it still does the unlocking
128 unsigned int extent_locked:1;
130 /* tells the submit_bio code to use a WRITE_SYNC */
131 unsigned int sync_io:1;
134 static void add_extent_changeset(struct extent_state *state, unsigned bits,
135 struct extent_changeset *changeset,
142 if (set && (state->state & bits) == bits)
144 if (!set && (state->state & bits) == 0)
146 changeset->bytes_changed += state->end - state->start + 1;
147 ret = ulist_add(changeset->range_changed, state->start, state->end,
153 static noinline void flush_write_bio(void *data);
154 static inline struct btrfs_fs_info *
155 tree_fs_info(struct extent_io_tree *tree)
159 return btrfs_sb(tree->mapping->host->i_sb);
162 int __init extent_io_init(void)
164 extent_state_cache = kmem_cache_create("btrfs_extent_state",
165 sizeof(struct extent_state), 0,
166 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
167 if (!extent_state_cache)
170 extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer",
171 sizeof(struct extent_buffer), 0,
172 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
173 if (!extent_buffer_cache)
174 goto free_state_cache;
176 btrfs_bioset = bioset_create(BIO_POOL_SIZE,
177 offsetof(struct btrfs_io_bio, bio));
179 goto free_buffer_cache;
181 if (bioset_integrity_create(btrfs_bioset, BIO_POOL_SIZE))
187 bioset_free(btrfs_bioset);
191 kmem_cache_destroy(extent_buffer_cache);
192 extent_buffer_cache = NULL;
195 kmem_cache_destroy(extent_state_cache);
196 extent_state_cache = NULL;
200 void extent_io_exit(void)
202 btrfs_leak_debug_check();
205 * Make sure all delayed rcu free are flushed before we
209 if (extent_state_cache)
210 kmem_cache_destroy(extent_state_cache);
211 if (extent_buffer_cache)
212 kmem_cache_destroy(extent_buffer_cache);
214 bioset_free(btrfs_bioset);
217 void extent_io_tree_init(struct extent_io_tree *tree,
218 struct address_space *mapping)
220 tree->state = RB_ROOT;
222 tree->dirty_bytes = 0;
223 spin_lock_init(&tree->lock);
224 tree->mapping = mapping;
227 static struct extent_state *alloc_extent_state(gfp_t mask)
229 struct extent_state *state;
231 state = kmem_cache_alloc(extent_state_cache, mask);
236 RB_CLEAR_NODE(&state->rb_node);
237 btrfs_leak_debug_add(&state->leak_list, &states);
238 atomic_set(&state->refs, 1);
239 init_waitqueue_head(&state->wq);
240 trace_alloc_extent_state(state, mask, _RET_IP_);
244 void free_extent_state(struct extent_state *state)
248 if (atomic_dec_and_test(&state->refs)) {
249 WARN_ON(extent_state_in_tree(state));
250 btrfs_leak_debug_del(&state->leak_list);
251 trace_free_extent_state(state, _RET_IP_);
252 kmem_cache_free(extent_state_cache, state);
256 static struct rb_node *tree_insert(struct rb_root *root,
257 struct rb_node *search_start,
259 struct rb_node *node,
260 struct rb_node ***p_in,
261 struct rb_node **parent_in)
264 struct rb_node *parent = NULL;
265 struct tree_entry *entry;
267 if (p_in && parent_in) {
273 p = search_start ? &search_start : &root->rb_node;
276 entry = rb_entry(parent, struct tree_entry, rb_node);
278 if (offset < entry->start)
280 else if (offset > entry->end)
287 rb_link_node(node, parent, p);
288 rb_insert_color(node, root);
292 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
293 struct rb_node **prev_ret,
294 struct rb_node **next_ret,
295 struct rb_node ***p_ret,
296 struct rb_node **parent_ret)
298 struct rb_root *root = &tree->state;
299 struct rb_node **n = &root->rb_node;
300 struct rb_node *prev = NULL;
301 struct rb_node *orig_prev = NULL;
302 struct tree_entry *entry;
303 struct tree_entry *prev_entry = NULL;
307 entry = rb_entry(prev, struct tree_entry, rb_node);
310 if (offset < entry->start)
312 else if (offset > entry->end)
325 while (prev && offset > prev_entry->end) {
326 prev = rb_next(prev);
327 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
334 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
335 while (prev && offset < prev_entry->start) {
336 prev = rb_prev(prev);
337 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
344 static inline struct rb_node *
345 tree_search_for_insert(struct extent_io_tree *tree,
347 struct rb_node ***p_ret,
348 struct rb_node **parent_ret)
350 struct rb_node *prev = NULL;
353 ret = __etree_search(tree, offset, &prev, NULL, p_ret, parent_ret);
359 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
362 return tree_search_for_insert(tree, offset, NULL, NULL);
365 static void merge_cb(struct extent_io_tree *tree, struct extent_state *new,
366 struct extent_state *other)
368 if (tree->ops && tree->ops->merge_extent_hook)
369 tree->ops->merge_extent_hook(tree->mapping->host, new,
374 * utility function to look for merge candidates inside a given range.
375 * Any extents with matching state are merged together into a single
376 * extent in the tree. Extents with EXTENT_IO in their state field
377 * are not merged because the end_io handlers need to be able to do
378 * operations on them without sleeping (or doing allocations/splits).
380 * This should be called with the tree lock held.
382 static void merge_state(struct extent_io_tree *tree,
383 struct extent_state *state)
385 struct extent_state *other;
386 struct rb_node *other_node;
388 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
391 other_node = rb_prev(&state->rb_node);
393 other = rb_entry(other_node, struct extent_state, rb_node);
394 if (other->end == state->start - 1 &&
395 other->state == state->state) {
396 merge_cb(tree, state, other);
397 state->start = other->start;
398 rb_erase(&other->rb_node, &tree->state);
399 RB_CLEAR_NODE(&other->rb_node);
400 free_extent_state(other);
403 other_node = rb_next(&state->rb_node);
405 other = rb_entry(other_node, struct extent_state, rb_node);
406 if (other->start == state->end + 1 &&
407 other->state == state->state) {
408 merge_cb(tree, state, other);
409 state->end = other->end;
410 rb_erase(&other->rb_node, &tree->state);
411 RB_CLEAR_NODE(&other->rb_node);
412 free_extent_state(other);
417 static void set_state_cb(struct extent_io_tree *tree,
418 struct extent_state *state, unsigned *bits)
420 if (tree->ops && tree->ops->set_bit_hook)
421 tree->ops->set_bit_hook(tree->mapping->host, state, bits);
424 static void clear_state_cb(struct extent_io_tree *tree,
425 struct extent_state *state, unsigned *bits)
427 if (tree->ops && tree->ops->clear_bit_hook)
428 tree->ops->clear_bit_hook(tree->mapping->host, state, bits);
431 static void set_state_bits(struct extent_io_tree *tree,
432 struct extent_state *state, unsigned *bits,
433 struct extent_changeset *changeset);
436 * insert an extent_state struct into the tree. 'bits' are set on the
437 * struct before it is inserted.
439 * This may return -EEXIST if the extent is already there, in which case the
440 * state struct is freed.
442 * The tree lock is not taken internally. This is a utility function and
443 * probably isn't what you want to call (see set/clear_extent_bit).
445 static int insert_state(struct extent_io_tree *tree,
446 struct extent_state *state, u64 start, u64 end,
448 struct rb_node **parent,
449 unsigned *bits, struct extent_changeset *changeset)
451 struct rb_node *node;
454 WARN(1, KERN_ERR "BTRFS: end < start %llu %llu\n",
456 state->start = start;
459 set_state_bits(tree, state, bits, changeset);
461 node = tree_insert(&tree->state, NULL, end, &state->rb_node, p, parent);
463 struct extent_state *found;
464 found = rb_entry(node, struct extent_state, rb_node);
465 printk(KERN_ERR "BTRFS: found node %llu %llu on insert of "
467 found->start, found->end, start, end);
470 merge_state(tree, state);
474 static void split_cb(struct extent_io_tree *tree, struct extent_state *orig,
477 if (tree->ops && tree->ops->split_extent_hook)
478 tree->ops->split_extent_hook(tree->mapping->host, orig, split);
482 * split a given extent state struct in two, inserting the preallocated
483 * struct 'prealloc' as the newly created second half. 'split' indicates an
484 * offset inside 'orig' where it should be split.
487 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
488 * are two extent state structs in the tree:
489 * prealloc: [orig->start, split - 1]
490 * orig: [ split, orig->end ]
492 * The tree locks are not taken by this function. They need to be held
495 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
496 struct extent_state *prealloc, u64 split)
498 struct rb_node *node;
500 split_cb(tree, orig, split);
502 prealloc->start = orig->start;
503 prealloc->end = split - 1;
504 prealloc->state = orig->state;
507 node = tree_insert(&tree->state, &orig->rb_node, prealloc->end,
508 &prealloc->rb_node, NULL, NULL);
510 free_extent_state(prealloc);
516 static struct extent_state *next_state(struct extent_state *state)
518 struct rb_node *next = rb_next(&state->rb_node);
520 return rb_entry(next, struct extent_state, rb_node);
526 * utility function to clear some bits in an extent state struct.
527 * it will optionally wake up any one waiting on this state (wake == 1).
529 * If no bits are set on the state struct after clearing things, the
530 * struct is freed and removed from the tree
532 static struct extent_state *clear_state_bit(struct extent_io_tree *tree,
533 struct extent_state *state,
534 unsigned *bits, int wake,
535 struct extent_changeset *changeset)
537 struct extent_state *next;
538 unsigned bits_to_clear = *bits & ~EXTENT_CTLBITS;
540 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
541 u64 range = state->end - state->start + 1;
542 WARN_ON(range > tree->dirty_bytes);
543 tree->dirty_bytes -= range;
545 clear_state_cb(tree, state, bits);
546 add_extent_changeset(state, bits_to_clear, changeset, 0);
547 state->state &= ~bits_to_clear;
550 if (state->state == 0) {
551 next = next_state(state);
552 if (extent_state_in_tree(state)) {
553 rb_erase(&state->rb_node, &tree->state);
554 RB_CLEAR_NODE(&state->rb_node);
555 free_extent_state(state);
560 merge_state(tree, state);
561 next = next_state(state);
566 static struct extent_state *
567 alloc_extent_state_atomic(struct extent_state *prealloc)
570 prealloc = alloc_extent_state(GFP_ATOMIC);
575 static void extent_io_tree_panic(struct extent_io_tree *tree, int err)
577 btrfs_panic(tree_fs_info(tree), err, "Locking error: "
578 "Extent tree was modified by another "
579 "thread while locked.");
583 * clear some bits on a range in the tree. This may require splitting
584 * or inserting elements in the tree, so the gfp mask is used to
585 * indicate which allocations or sleeping are allowed.
587 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
588 * the given range from the tree regardless of state (ie for truncate).
590 * the range [start, end] is inclusive.
592 * This takes the tree lock, and returns 0 on success and < 0 on error.
594 static int __clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
595 unsigned bits, int wake, int delete,
596 struct extent_state **cached_state,
597 gfp_t mask, struct extent_changeset *changeset)
599 struct extent_state *state;
600 struct extent_state *cached;
601 struct extent_state *prealloc = NULL;
602 struct rb_node *node;
607 btrfs_debug_check_extent_io_range(tree, start, end);
609 if (bits & EXTENT_DELALLOC)
610 bits |= EXTENT_NORESERVE;
613 bits |= ~EXTENT_CTLBITS;
614 bits |= EXTENT_FIRST_DELALLOC;
616 if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY))
619 if (!prealloc && gfpflags_allow_blocking(mask)) {
621 * Don't care for allocation failure here because we might end
622 * up not needing the pre-allocated extent state at all, which
623 * is the case if we only have in the tree extent states that
624 * cover our input range and don't cover too any other range.
625 * If we end up needing a new extent state we allocate it later.
627 prealloc = alloc_extent_state(mask);
630 spin_lock(&tree->lock);
632 cached = *cached_state;
635 *cached_state = NULL;
639 if (cached && extent_state_in_tree(cached) &&
640 cached->start <= start && cached->end > start) {
642 atomic_dec(&cached->refs);
647 free_extent_state(cached);
650 * this search will find the extents that end after
653 node = tree_search(tree, start);
656 state = rb_entry(node, struct extent_state, rb_node);
658 if (state->start > end)
660 WARN_ON(state->end < start);
661 last_end = state->end;
663 /* the state doesn't have the wanted bits, go ahead */
664 if (!(state->state & bits)) {
665 state = next_state(state);
670 * | ---- desired range ---- |
672 * | ------------- state -------------- |
674 * We need to split the extent we found, and may flip
675 * bits on second half.
677 * If the extent we found extends past our range, we
678 * just split and search again. It'll get split again
679 * the next time though.
681 * If the extent we found is inside our range, we clear
682 * the desired bit on it.
685 if (state->start < start) {
686 prealloc = alloc_extent_state_atomic(prealloc);
688 err = split_state(tree, state, prealloc, start);
690 extent_io_tree_panic(tree, err);
695 if (state->end <= end) {
696 state = clear_state_bit(tree, state, &bits, wake,
703 * | ---- desired range ---- |
705 * We need to split the extent, and clear the bit
708 if (state->start <= end && state->end > end) {
709 prealloc = alloc_extent_state_atomic(prealloc);
711 err = split_state(tree, state, prealloc, end + 1);
713 extent_io_tree_panic(tree, err);
718 clear_state_bit(tree, prealloc, &bits, wake, changeset);
724 state = clear_state_bit(tree, state, &bits, wake, changeset);
726 if (last_end == (u64)-1)
728 start = last_end + 1;
729 if (start <= end && state && !need_resched())
734 spin_unlock(&tree->lock);
736 free_extent_state(prealloc);
743 spin_unlock(&tree->lock);
744 if (gfpflags_allow_blocking(mask))
749 static void wait_on_state(struct extent_io_tree *tree,
750 struct extent_state *state)
751 __releases(tree->lock)
752 __acquires(tree->lock)
755 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
756 spin_unlock(&tree->lock);
758 spin_lock(&tree->lock);
759 finish_wait(&state->wq, &wait);
763 * waits for one or more bits to clear on a range in the state tree.
764 * The range [start, end] is inclusive.
765 * The tree lock is taken by this function
767 static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
770 struct extent_state *state;
771 struct rb_node *node;
773 btrfs_debug_check_extent_io_range(tree, start, end);
775 spin_lock(&tree->lock);
779 * this search will find all the extents that end after
782 node = tree_search(tree, start);
787 state = rb_entry(node, struct extent_state, rb_node);
789 if (state->start > end)
792 if (state->state & bits) {
793 start = state->start;
794 atomic_inc(&state->refs);
795 wait_on_state(tree, state);
796 free_extent_state(state);
799 start = state->end + 1;
804 if (!cond_resched_lock(&tree->lock)) {
805 node = rb_next(node);
810 spin_unlock(&tree->lock);
813 static void set_state_bits(struct extent_io_tree *tree,
814 struct extent_state *state,
815 unsigned *bits, struct extent_changeset *changeset)
817 unsigned bits_to_set = *bits & ~EXTENT_CTLBITS;
819 set_state_cb(tree, state, bits);
820 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
821 u64 range = state->end - state->start + 1;
822 tree->dirty_bytes += range;
824 add_extent_changeset(state, bits_to_set, changeset, 1);
825 state->state |= bits_to_set;
828 static void cache_state_if_flags(struct extent_state *state,
829 struct extent_state **cached_ptr,
832 if (cached_ptr && !(*cached_ptr)) {
833 if (!flags || (state->state & flags)) {
835 atomic_inc(&state->refs);
840 static void cache_state(struct extent_state *state,
841 struct extent_state **cached_ptr)
843 return cache_state_if_flags(state, cached_ptr,
844 EXTENT_IOBITS | EXTENT_BOUNDARY);
848 * set some bits on a range in the tree. This may require allocations or
849 * sleeping, so the gfp mask is used to indicate what is allowed.
851 * If any of the exclusive bits are set, this will fail with -EEXIST if some
852 * part of the range already has the desired bits set. The start of the
853 * existing range is returned in failed_start in this case.
855 * [start, end] is inclusive This takes the tree lock.
858 static int __must_check
859 __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
860 unsigned bits, unsigned exclusive_bits,
861 u64 *failed_start, struct extent_state **cached_state,
862 gfp_t mask, struct extent_changeset *changeset)
864 struct extent_state *state;
865 struct extent_state *prealloc = NULL;
866 struct rb_node *node;
868 struct rb_node *parent;
873 btrfs_debug_check_extent_io_range(tree, start, end);
875 bits |= EXTENT_FIRST_DELALLOC;
877 if (!prealloc && gfpflags_allow_blocking(mask)) {
878 prealloc = alloc_extent_state(mask);
882 spin_lock(&tree->lock);
883 if (cached_state && *cached_state) {
884 state = *cached_state;
885 if (state->start <= start && state->end > start &&
886 extent_state_in_tree(state)) {
887 node = &state->rb_node;
892 * this search will find all the extents that end after
895 node = tree_search_for_insert(tree, start, &p, &parent);
897 prealloc = alloc_extent_state_atomic(prealloc);
899 err = insert_state(tree, prealloc, start, end,
900 &p, &parent, &bits, changeset);
902 extent_io_tree_panic(tree, err);
904 cache_state(prealloc, cached_state);
908 state = rb_entry(node, struct extent_state, rb_node);
910 last_start = state->start;
911 last_end = state->end;
914 * | ---- desired range ---- |
917 * Just lock what we found and keep going
919 if (state->start == start && state->end <= end) {
920 if (state->state & exclusive_bits) {
921 *failed_start = state->start;
926 set_state_bits(tree, state, &bits, changeset);
927 cache_state(state, cached_state);
928 merge_state(tree, state);
929 if (last_end == (u64)-1)
931 start = last_end + 1;
932 state = next_state(state);
933 if (start < end && state && state->start == start &&
940 * | ---- desired range ---- |
943 * | ------------- state -------------- |
945 * We need to split the extent we found, and may flip bits on
948 * If the extent we found extends past our
949 * range, we just split and search again. It'll get split
950 * again the next time though.
952 * If the extent we found is inside our range, we set the
955 if (state->start < start) {
956 if (state->state & exclusive_bits) {
957 *failed_start = start;
962 prealloc = alloc_extent_state_atomic(prealloc);
964 err = split_state(tree, state, prealloc, start);
966 extent_io_tree_panic(tree, err);
971 if (state->end <= end) {
972 set_state_bits(tree, state, &bits, changeset);
973 cache_state(state, cached_state);
974 merge_state(tree, state);
975 if (last_end == (u64)-1)
977 start = last_end + 1;
978 state = next_state(state);
979 if (start < end && state && state->start == start &&
986 * | ---- desired range ---- |
987 * | state | or | state |
989 * There's a hole, we need to insert something in it and
990 * ignore the extent we found.
992 if (state->start > start) {
994 if (end < last_start)
997 this_end = last_start - 1;
999 prealloc = alloc_extent_state_atomic(prealloc);
1003 * Avoid to free 'prealloc' if it can be merged with
1006 err = insert_state(tree, prealloc, start, this_end,
1007 NULL, NULL, &bits, changeset);
1009 extent_io_tree_panic(tree, err);
1011 cache_state(prealloc, cached_state);
1013 start = this_end + 1;
1017 * | ---- desired range ---- |
1019 * We need to split the extent, and set the bit
1022 if (state->start <= end && state->end > end) {
1023 if (state->state & exclusive_bits) {
1024 *failed_start = start;
1029 prealloc = alloc_extent_state_atomic(prealloc);
1031 err = split_state(tree, state, prealloc, end + 1);
1033 extent_io_tree_panic(tree, err);
1035 set_state_bits(tree, prealloc, &bits, changeset);
1036 cache_state(prealloc, cached_state);
1037 merge_state(tree, prealloc);
1045 spin_unlock(&tree->lock);
1047 free_extent_state(prealloc);
1054 spin_unlock(&tree->lock);
1055 if (gfpflags_allow_blocking(mask))
1060 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1061 unsigned bits, u64 * failed_start,
1062 struct extent_state **cached_state, gfp_t mask)
1064 return __set_extent_bit(tree, start, end, bits, 0, failed_start,
1065 cached_state, mask, NULL);
1070 * convert_extent_bit - convert all bits in a given range from one bit to
1072 * @tree: the io tree to search
1073 * @start: the start offset in bytes
1074 * @end: the end offset in bytes (inclusive)
1075 * @bits: the bits to set in this range
1076 * @clear_bits: the bits to clear in this range
1077 * @cached_state: state that we're going to cache
1078 * @mask: the allocation mask
1080 * This will go through and set bits for the given range. If any states exist
1081 * already in this range they are set with the given bit and cleared of the
1082 * clear_bits. This is only meant to be used by things that are mergeable, ie
1083 * converting from say DELALLOC to DIRTY. This is not meant to be used with
1084 * boundary bits like LOCK.
1086 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1087 unsigned bits, unsigned clear_bits,
1088 struct extent_state **cached_state, gfp_t mask)
1090 struct extent_state *state;
1091 struct extent_state *prealloc = NULL;
1092 struct rb_node *node;
1094 struct rb_node *parent;
1098 bool first_iteration = true;
1100 btrfs_debug_check_extent_io_range(tree, start, end);
1103 if (!prealloc && gfpflags_allow_blocking(mask)) {
1105 * Best effort, don't worry if extent state allocation fails
1106 * here for the first iteration. We might have a cached state
1107 * that matches exactly the target range, in which case no
1108 * extent state allocations are needed. We'll only know this
1109 * after locking the tree.
1111 prealloc = alloc_extent_state(mask);
1112 if (!prealloc && !first_iteration)
1116 spin_lock(&tree->lock);
1117 if (cached_state && *cached_state) {
1118 state = *cached_state;
1119 if (state->start <= start && state->end > start &&
1120 extent_state_in_tree(state)) {
1121 node = &state->rb_node;
1127 * this search will find all the extents that end after
1130 node = tree_search_for_insert(tree, start, &p, &parent);
1132 prealloc = alloc_extent_state_atomic(prealloc);
1137 err = insert_state(tree, prealloc, start, end,
1138 &p, &parent, &bits, NULL);
1140 extent_io_tree_panic(tree, err);
1141 cache_state(prealloc, cached_state);
1145 state = rb_entry(node, struct extent_state, rb_node);
1147 last_start = state->start;
1148 last_end = state->end;
1151 * | ---- desired range ---- |
1154 * Just lock what we found and keep going
1156 if (state->start == start && state->end <= end) {
1157 set_state_bits(tree, state, &bits, NULL);
1158 cache_state(state, cached_state);
1159 state = clear_state_bit(tree, state, &clear_bits, 0, NULL);
1160 if (last_end == (u64)-1)
1162 start = last_end + 1;
1163 if (start < end && state && state->start == start &&
1170 * | ---- desired range ---- |
1173 * | ------------- state -------------- |
1175 * We need to split the extent we found, and may flip bits on
1178 * If the extent we found extends past our
1179 * range, we just split and search again. It'll get split
1180 * again the next time though.
1182 * If the extent we found is inside our range, we set the
1183 * desired bit on it.
1185 if (state->start < start) {
1186 prealloc = alloc_extent_state_atomic(prealloc);
1191 err = split_state(tree, state, prealloc, start);
1193 extent_io_tree_panic(tree, err);
1197 if (state->end <= end) {
1198 set_state_bits(tree, state, &bits, NULL);
1199 cache_state(state, cached_state);
1200 state = clear_state_bit(tree, state, &clear_bits, 0,
1202 if (last_end == (u64)-1)
1204 start = last_end + 1;
1205 if (start < end && state && state->start == start &&
1212 * | ---- desired range ---- |
1213 * | state | or | state |
1215 * There's a hole, we need to insert something in it and
1216 * ignore the extent we found.
1218 if (state->start > start) {
1220 if (end < last_start)
1223 this_end = last_start - 1;
1225 prealloc = alloc_extent_state_atomic(prealloc);
1232 * Avoid to free 'prealloc' if it can be merged with
1235 err = insert_state(tree, prealloc, start, this_end,
1236 NULL, NULL, &bits, NULL);
1238 extent_io_tree_panic(tree, err);
1239 cache_state(prealloc, cached_state);
1241 start = this_end + 1;
1245 * | ---- desired range ---- |
1247 * We need to split the extent, and set the bit
1250 if (state->start <= end && state->end > end) {
1251 prealloc = alloc_extent_state_atomic(prealloc);
1257 err = split_state(tree, state, prealloc, end + 1);
1259 extent_io_tree_panic(tree, err);
1261 set_state_bits(tree, prealloc, &bits, NULL);
1262 cache_state(prealloc, cached_state);
1263 clear_state_bit(tree, prealloc, &clear_bits, 0, NULL);
1271 spin_unlock(&tree->lock);
1273 free_extent_state(prealloc);
1280 spin_unlock(&tree->lock);
1281 if (gfpflags_allow_blocking(mask))
1283 first_iteration = false;
1287 /* wrappers around set/clear extent bit */
1288 int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1289 unsigned bits, gfp_t mask,
1290 struct extent_changeset *changeset)
1293 * We don't support EXTENT_LOCKED yet, as current changeset will
1294 * record any bits changed, so for EXTENT_LOCKED case, it will
1295 * either fail with -EEXIST or changeset will record the whole
1298 BUG_ON(bits & EXTENT_LOCKED);
1300 return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL, mask,
1304 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1305 unsigned bits, int wake, int delete,
1306 struct extent_state **cached, gfp_t mask)
1308 return __clear_extent_bit(tree, start, end, bits, wake, delete,
1309 cached, mask, NULL);
1312 int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1313 unsigned bits, gfp_t mask,
1314 struct extent_changeset *changeset)
1317 * Don't support EXTENT_LOCKED case, same reason as
1318 * set_record_extent_bits().
1320 BUG_ON(bits & EXTENT_LOCKED);
1322 return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask,
1327 * either insert or lock state struct between start and end use mask to tell
1328 * us if waiting is desired.
1330 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1331 struct extent_state **cached_state)
1337 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED,
1338 EXTENT_LOCKED, &failed_start,
1339 cached_state, GFP_NOFS, NULL);
1340 if (err == -EEXIST) {
1341 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1342 start = failed_start;
1345 WARN_ON(start > end);
1350 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1355 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1356 &failed_start, NULL, GFP_NOFS, NULL);
1357 if (err == -EEXIST) {
1358 if (failed_start > start)
1359 clear_extent_bit(tree, start, failed_start - 1,
1360 EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS);
1366 void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
1368 unsigned long index = start >> PAGE_CACHE_SHIFT;
1369 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1372 while (index <= end_index) {
1373 page = find_get_page(inode->i_mapping, index);
1374 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1375 clear_page_dirty_for_io(page);
1376 page_cache_release(page);
1381 void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
1383 unsigned long index = start >> PAGE_CACHE_SHIFT;
1384 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1387 while (index <= end_index) {
1388 page = find_get_page(inode->i_mapping, index);
1389 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1390 __set_page_dirty_nobuffers(page);
1391 account_page_redirty(page);
1392 page_cache_release(page);
1398 * helper function to set both pages and extents in the tree writeback
1400 static void set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
1402 unsigned long index = start >> PAGE_CACHE_SHIFT;
1403 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1406 while (index <= end_index) {
1407 page = find_get_page(tree->mapping, index);
1408 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1409 set_page_writeback(page);
1410 page_cache_release(page);
1415 /* find the first state struct with 'bits' set after 'start', and
1416 * return it. tree->lock must be held. NULL will returned if
1417 * nothing was found after 'start'
1419 static struct extent_state *
1420 find_first_extent_bit_state(struct extent_io_tree *tree,
1421 u64 start, unsigned bits)
1423 struct rb_node *node;
1424 struct extent_state *state;
1427 * this search will find all the extents that end after
1430 node = tree_search(tree, start);
1435 state = rb_entry(node, struct extent_state, rb_node);
1436 if (state->end >= start && (state->state & bits))
1439 node = rb_next(node);
1448 * find the first offset in the io tree with 'bits' set. zero is
1449 * returned if we find something, and *start_ret and *end_ret are
1450 * set to reflect the state struct that was found.
1452 * If nothing was found, 1 is returned. If found something, return 0.
1454 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1455 u64 *start_ret, u64 *end_ret, unsigned bits,
1456 struct extent_state **cached_state)
1458 struct extent_state *state;
1462 spin_lock(&tree->lock);
1463 if (cached_state && *cached_state) {
1464 state = *cached_state;
1465 if (state->end == start - 1 && extent_state_in_tree(state)) {
1466 n = rb_next(&state->rb_node);
1468 state = rb_entry(n, struct extent_state,
1470 if (state->state & bits)
1474 free_extent_state(*cached_state);
1475 *cached_state = NULL;
1478 free_extent_state(*cached_state);
1479 *cached_state = NULL;
1482 state = find_first_extent_bit_state(tree, start, bits);
1485 cache_state_if_flags(state, cached_state, 0);
1486 *start_ret = state->start;
1487 *end_ret = state->end;
1491 spin_unlock(&tree->lock);
1496 * find a contiguous range of bytes in the file marked as delalloc, not
1497 * more than 'max_bytes'. start and end are used to return the range,
1499 * 1 is returned if we find something, 0 if nothing was in the tree
1501 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1502 u64 *start, u64 *end, u64 max_bytes,
1503 struct extent_state **cached_state)
1505 struct rb_node *node;
1506 struct extent_state *state;
1507 u64 cur_start = *start;
1509 u64 total_bytes = 0;
1511 spin_lock(&tree->lock);
1514 * this search will find all the extents that end after
1517 node = tree_search(tree, cur_start);
1525 state = rb_entry(node, struct extent_state, rb_node);
1526 if (found && (state->start != cur_start ||
1527 (state->state & EXTENT_BOUNDARY))) {
1530 if (!(state->state & EXTENT_DELALLOC)) {
1536 *start = state->start;
1537 *cached_state = state;
1538 atomic_inc(&state->refs);
1542 cur_start = state->end + 1;
1543 node = rb_next(node);
1544 total_bytes += state->end - state->start + 1;
1545 if (total_bytes >= max_bytes)
1551 spin_unlock(&tree->lock);
1555 static noinline void __unlock_for_delalloc(struct inode *inode,
1556 struct page *locked_page,
1560 struct page *pages[16];
1561 unsigned long index = start >> PAGE_CACHE_SHIFT;
1562 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1563 unsigned long nr_pages = end_index - index + 1;
1566 if (index == locked_page->index && end_index == index)
1569 while (nr_pages > 0) {
1570 ret = find_get_pages_contig(inode->i_mapping, index,
1571 min_t(unsigned long, nr_pages,
1572 ARRAY_SIZE(pages)), pages);
1573 for (i = 0; i < ret; i++) {
1574 if (pages[i] != locked_page)
1575 unlock_page(pages[i]);
1576 page_cache_release(pages[i]);
1584 static noinline int lock_delalloc_pages(struct inode *inode,
1585 struct page *locked_page,
1589 unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT;
1590 unsigned long start_index = index;
1591 unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT;
1592 unsigned long pages_locked = 0;
1593 struct page *pages[16];
1594 unsigned long nrpages;
1598 /* the caller is responsible for locking the start index */
1599 if (index == locked_page->index && index == end_index)
1602 /* skip the page at the start index */
1603 nrpages = end_index - index + 1;
1604 while (nrpages > 0) {
1605 ret = find_get_pages_contig(inode->i_mapping, index,
1606 min_t(unsigned long,
1607 nrpages, ARRAY_SIZE(pages)), pages);
1612 /* now we have an array of pages, lock them all */
1613 for (i = 0; i < ret; i++) {
1615 * the caller is taking responsibility for
1618 if (pages[i] != locked_page) {
1619 lock_page(pages[i]);
1620 if (!PageDirty(pages[i]) ||
1621 pages[i]->mapping != inode->i_mapping) {
1623 unlock_page(pages[i]);
1624 page_cache_release(pages[i]);
1628 page_cache_release(pages[i]);
1637 if (ret && pages_locked) {
1638 __unlock_for_delalloc(inode, locked_page,
1640 ((u64)(start_index + pages_locked - 1)) <<
1647 * find a contiguous range of bytes in the file marked as delalloc, not
1648 * more than 'max_bytes'. start and end are used to return the range,
1650 * 1 is returned if we find something, 0 if nothing was in the tree
1652 STATIC u64 find_lock_delalloc_range(struct inode *inode,
1653 struct extent_io_tree *tree,
1654 struct page *locked_page, u64 *start,
1655 u64 *end, u64 max_bytes)
1660 struct extent_state *cached_state = NULL;
1665 /* step one, find a bunch of delalloc bytes starting at start */
1666 delalloc_start = *start;
1668 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1669 max_bytes, &cached_state);
1670 if (!found || delalloc_end <= *start) {
1671 *start = delalloc_start;
1672 *end = delalloc_end;
1673 free_extent_state(cached_state);
1678 * start comes from the offset of locked_page. We have to lock
1679 * pages in order, so we can't process delalloc bytes before
1682 if (delalloc_start < *start)
1683 delalloc_start = *start;
1686 * make sure to limit the number of pages we try to lock down
1688 if (delalloc_end + 1 - delalloc_start > max_bytes)
1689 delalloc_end = delalloc_start + max_bytes - 1;
1691 /* step two, lock all the pages after the page that has start */
1692 ret = lock_delalloc_pages(inode, locked_page,
1693 delalloc_start, delalloc_end);
1694 if (ret == -EAGAIN) {
1695 /* some of the pages are gone, lets avoid looping by
1696 * shortening the size of the delalloc range we're searching
1698 free_extent_state(cached_state);
1699 cached_state = NULL;
1701 max_bytes = PAGE_CACHE_SIZE;
1709 BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */
1711 /* step three, lock the state bits for the whole range */
1712 lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state);
1714 /* then test to make sure it is all still delalloc */
1715 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1716 EXTENT_DELALLOC, 1, cached_state);
1718 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1719 &cached_state, GFP_NOFS);
1720 __unlock_for_delalloc(inode, locked_page,
1721 delalloc_start, delalloc_end);
1725 free_extent_state(cached_state);
1726 *start = delalloc_start;
1727 *end = delalloc_end;
1732 void extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end,
1733 struct page *locked_page,
1734 unsigned clear_bits,
1735 unsigned long page_ops)
1737 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
1739 struct page *pages[16];
1740 unsigned long index = start >> PAGE_CACHE_SHIFT;
1741 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1742 unsigned long nr_pages = end_index - index + 1;
1745 clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS);
1749 if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0)
1750 mapping_set_error(inode->i_mapping, -EIO);
1752 while (nr_pages > 0) {
1753 ret = find_get_pages_contig(inode->i_mapping, index,
1754 min_t(unsigned long,
1755 nr_pages, ARRAY_SIZE(pages)), pages);
1756 for (i = 0; i < ret; i++) {
1758 if (page_ops & PAGE_SET_PRIVATE2)
1759 SetPagePrivate2(pages[i]);
1761 if (pages[i] == locked_page) {
1762 page_cache_release(pages[i]);
1765 if (page_ops & PAGE_CLEAR_DIRTY)
1766 clear_page_dirty_for_io(pages[i]);
1767 if (page_ops & PAGE_SET_WRITEBACK)
1768 set_page_writeback(pages[i]);
1769 if (page_ops & PAGE_SET_ERROR)
1770 SetPageError(pages[i]);
1771 if (page_ops & PAGE_END_WRITEBACK)
1772 end_page_writeback(pages[i]);
1773 if (page_ops & PAGE_UNLOCK)
1774 unlock_page(pages[i]);
1775 page_cache_release(pages[i]);
1784 * count the number of bytes in the tree that have a given bit(s)
1785 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1786 * cached. The total number found is returned.
1788 u64 count_range_bits(struct extent_io_tree *tree,
1789 u64 *start, u64 search_end, u64 max_bytes,
1790 unsigned bits, int contig)
1792 struct rb_node *node;
1793 struct extent_state *state;
1794 u64 cur_start = *start;
1795 u64 total_bytes = 0;
1799 if (WARN_ON(search_end <= cur_start))
1802 spin_lock(&tree->lock);
1803 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1804 total_bytes = tree->dirty_bytes;
1808 * this search will find all the extents that end after
1811 node = tree_search(tree, cur_start);
1816 state = rb_entry(node, struct extent_state, rb_node);
1817 if (state->start > search_end)
1819 if (contig && found && state->start > last + 1)
1821 if (state->end >= cur_start && (state->state & bits) == bits) {
1822 total_bytes += min(search_end, state->end) + 1 -
1823 max(cur_start, state->start);
1824 if (total_bytes >= max_bytes)
1827 *start = max(cur_start, state->start);
1831 } else if (contig && found) {
1834 node = rb_next(node);
1839 spin_unlock(&tree->lock);
1844 * set the private field for a given byte offset in the tree. If there isn't
1845 * an extent_state there already, this does nothing.
1847 static int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1849 struct rb_node *node;
1850 struct extent_state *state;
1853 spin_lock(&tree->lock);
1855 * this search will find all the extents that end after
1858 node = tree_search(tree, start);
1863 state = rb_entry(node, struct extent_state, rb_node);
1864 if (state->start != start) {
1868 state->private = private;
1870 spin_unlock(&tree->lock);
1874 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1876 struct rb_node *node;
1877 struct extent_state *state;
1880 spin_lock(&tree->lock);
1882 * this search will find all the extents that end after
1885 node = tree_search(tree, start);
1890 state = rb_entry(node, struct extent_state, rb_node);
1891 if (state->start != start) {
1895 *private = state->private;
1897 spin_unlock(&tree->lock);
1902 * searches a range in the state tree for a given mask.
1903 * If 'filled' == 1, this returns 1 only if every extent in the tree
1904 * has the bits set. Otherwise, 1 is returned if any bit in the
1905 * range is found set.
1907 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1908 unsigned bits, int filled, struct extent_state *cached)
1910 struct extent_state *state = NULL;
1911 struct rb_node *node;
1914 spin_lock(&tree->lock);
1915 if (cached && extent_state_in_tree(cached) && cached->start <= start &&
1916 cached->end > start)
1917 node = &cached->rb_node;
1919 node = tree_search(tree, start);
1920 while (node && start <= end) {
1921 state = rb_entry(node, struct extent_state, rb_node);
1923 if (filled && state->start > start) {
1928 if (state->start > end)
1931 if (state->state & bits) {
1935 } else if (filled) {
1940 if (state->end == (u64)-1)
1943 start = state->end + 1;
1946 node = rb_next(node);
1953 spin_unlock(&tree->lock);
1958 * helper function to set a given page up to date if all the
1959 * extents in the tree for that page are up to date
1961 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
1963 u64 start = page_offset(page);
1964 u64 end = start + PAGE_CACHE_SIZE - 1;
1965 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1966 SetPageUptodate(page);
1969 int free_io_failure(struct inode *inode, struct io_failure_record *rec)
1973 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1975 set_state_private(failure_tree, rec->start, 0);
1976 ret = clear_extent_bits(failure_tree, rec->start,
1977 rec->start + rec->len - 1,
1978 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1982 ret = clear_extent_bits(&BTRFS_I(inode)->io_tree, rec->start,
1983 rec->start + rec->len - 1,
1984 EXTENT_DAMAGED, GFP_NOFS);
1993 * this bypasses the standard btrfs submit functions deliberately, as
1994 * the standard behavior is to write all copies in a raid setup. here we only
1995 * want to write the one bad copy. so we do the mapping for ourselves and issue
1996 * submit_bio directly.
1997 * to avoid any synchronization issues, wait for the data after writing, which
1998 * actually prevents the read that triggered the error from finishing.
1999 * currently, there can be no more than two copies of every data bit. thus,
2000 * exactly one rewrite is required.
2002 int repair_io_failure(struct inode *inode, u64 start, u64 length, u64 logical,
2003 struct page *page, unsigned int pg_offset, int mirror_num)
2005 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2007 struct btrfs_device *dev;
2010 struct btrfs_bio *bbio = NULL;
2011 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
2014 ASSERT(!(fs_info->sb->s_flags & MS_RDONLY));
2015 BUG_ON(!mirror_num);
2017 /* we can't repair anything in raid56 yet */
2018 if (btrfs_is_parity_mirror(map_tree, logical, length, mirror_num))
2021 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
2024 bio->bi_iter.bi_size = 0;
2025 map_length = length;
2027 ret = btrfs_map_block(fs_info, WRITE, logical,
2028 &map_length, &bbio, mirror_num);
2033 BUG_ON(mirror_num != bbio->mirror_num);
2034 sector = bbio->stripes[mirror_num-1].physical >> 9;
2035 bio->bi_iter.bi_sector = sector;
2036 dev = bbio->stripes[mirror_num-1].dev;
2037 btrfs_put_bbio(bbio);
2038 if (!dev || !dev->bdev || !dev->writeable) {
2042 bio->bi_bdev = dev->bdev;
2043 bio_add_page(bio, page, length, pg_offset);
2045 if (btrfsic_submit_bio_wait(WRITE_SYNC, bio)) {
2046 /* try to remap that extent elsewhere? */
2048 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
2052 btrfs_info_rl_in_rcu(fs_info,
2053 "read error corrected: ino %llu off %llu (dev %s sector %llu)",
2054 btrfs_ino(inode), start,
2055 rcu_str_deref(dev->name), sector);
2060 int repair_eb_io_failure(struct btrfs_root *root, struct extent_buffer *eb,
2063 u64 start = eb->start;
2064 unsigned long i, num_pages = num_extent_pages(eb->start, eb->len);
2067 if (root->fs_info->sb->s_flags & MS_RDONLY)
2070 for (i = 0; i < num_pages; i++) {
2071 struct page *p = eb->pages[i];
2073 ret = repair_io_failure(root->fs_info->btree_inode, start,
2074 PAGE_CACHE_SIZE, start, p,
2075 start - page_offset(p), mirror_num);
2078 start += PAGE_CACHE_SIZE;
2085 * each time an IO finishes, we do a fast check in the IO failure tree
2086 * to see if we need to process or clean up an io_failure_record
2088 int clean_io_failure(struct inode *inode, u64 start, struct page *page,
2089 unsigned int pg_offset)
2092 u64 private_failure;
2093 struct io_failure_record *failrec;
2094 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2095 struct extent_state *state;
2100 ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
2101 (u64)-1, 1, EXTENT_DIRTY, 0);
2105 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, start,
2110 failrec = (struct io_failure_record *)(unsigned long) private_failure;
2111 BUG_ON(!failrec->this_mirror);
2113 if (failrec->in_validation) {
2114 /* there was no real error, just free the record */
2115 pr_debug("clean_io_failure: freeing dummy error at %llu\n",
2119 if (fs_info->sb->s_flags & MS_RDONLY)
2122 spin_lock(&BTRFS_I(inode)->io_tree.lock);
2123 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
2126 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
2128 if (state && state->start <= failrec->start &&
2129 state->end >= failrec->start + failrec->len - 1) {
2130 num_copies = btrfs_num_copies(fs_info, failrec->logical,
2132 if (num_copies > 1) {
2133 repair_io_failure(inode, start, failrec->len,
2134 failrec->logical, page,
2135 pg_offset, failrec->failed_mirror);
2140 free_io_failure(inode, failrec);
2146 * Can be called when
2147 * - hold extent lock
2148 * - under ordered extent
2149 * - the inode is freeing
2151 void btrfs_free_io_failure_record(struct inode *inode, u64 start, u64 end)
2153 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2154 struct io_failure_record *failrec;
2155 struct extent_state *state, *next;
2157 if (RB_EMPTY_ROOT(&failure_tree->state))
2160 spin_lock(&failure_tree->lock);
2161 state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY);
2163 if (state->start > end)
2166 ASSERT(state->end <= end);
2168 next = next_state(state);
2170 failrec = (struct io_failure_record *)(unsigned long)state->private;
2171 free_extent_state(state);
2176 spin_unlock(&failure_tree->lock);
2179 int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end,
2180 struct io_failure_record **failrec_ret)
2182 struct io_failure_record *failrec;
2184 struct extent_map *em;
2185 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2186 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2187 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2191 ret = get_state_private(failure_tree, start, &private);
2193 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2197 failrec->start = start;
2198 failrec->len = end - start + 1;
2199 failrec->this_mirror = 0;
2200 failrec->bio_flags = 0;
2201 failrec->in_validation = 0;
2203 read_lock(&em_tree->lock);
2204 em = lookup_extent_mapping(em_tree, start, failrec->len);
2206 read_unlock(&em_tree->lock);
2211 if (em->start > start || em->start + em->len <= start) {
2212 free_extent_map(em);
2215 read_unlock(&em_tree->lock);
2221 logical = start - em->start;
2222 logical = em->block_start + logical;
2223 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2224 logical = em->block_start;
2225 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2226 extent_set_compress_type(&failrec->bio_flags,
2230 pr_debug("Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu\n",
2231 logical, start, failrec->len);
2233 failrec->logical = logical;
2234 free_extent_map(em);
2236 /* set the bits in the private failure tree */
2237 ret = set_extent_bits(failure_tree, start, end,
2238 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
2240 ret = set_state_private(failure_tree, start,
2241 (u64)(unsigned long)failrec);
2242 /* set the bits in the inode's tree */
2244 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED,
2251 failrec = (struct io_failure_record *)(unsigned long)private;
2252 pr_debug("Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d\n",
2253 failrec->logical, failrec->start, failrec->len,
2254 failrec->in_validation);
2256 * when data can be on disk more than twice, add to failrec here
2257 * (e.g. with a list for failed_mirror) to make
2258 * clean_io_failure() clean all those errors at once.
2262 *failrec_ret = failrec;
2267 int btrfs_check_repairable(struct inode *inode, struct bio *failed_bio,
2268 struct io_failure_record *failrec, int failed_mirror)
2272 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
2273 failrec->logical, failrec->len);
2274 if (num_copies == 1) {
2276 * we only have a single copy of the data, so don't bother with
2277 * all the retry and error correction code that follows. no
2278 * matter what the error is, it is very likely to persist.
2280 pr_debug("Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
2281 num_copies, failrec->this_mirror, failed_mirror);
2286 * there are two premises:
2287 * a) deliver good data to the caller
2288 * b) correct the bad sectors on disk
2290 if (failed_bio->bi_vcnt > 1) {
2292 * to fulfill b), we need to know the exact failing sectors, as
2293 * we don't want to rewrite any more than the failed ones. thus,
2294 * we need separate read requests for the failed bio
2296 * if the following BUG_ON triggers, our validation request got
2297 * merged. we need separate requests for our algorithm to work.
2299 BUG_ON(failrec->in_validation);
2300 failrec->in_validation = 1;
2301 failrec->this_mirror = failed_mirror;
2304 * we're ready to fulfill a) and b) alongside. get a good copy
2305 * of the failed sector and if we succeed, we have setup
2306 * everything for repair_io_failure to do the rest for us.
2308 if (failrec->in_validation) {
2309 BUG_ON(failrec->this_mirror != failed_mirror);
2310 failrec->in_validation = 0;
2311 failrec->this_mirror = 0;
2313 failrec->failed_mirror = failed_mirror;
2314 failrec->this_mirror++;
2315 if (failrec->this_mirror == failed_mirror)
2316 failrec->this_mirror++;
2319 if (failrec->this_mirror > num_copies) {
2320 pr_debug("Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
2321 num_copies, failrec->this_mirror, failed_mirror);
2329 struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio,
2330 struct io_failure_record *failrec,
2331 struct page *page, int pg_offset, int icsum,
2332 bio_end_io_t *endio_func, void *data)
2335 struct btrfs_io_bio *btrfs_failed_bio;
2336 struct btrfs_io_bio *btrfs_bio;
2338 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
2342 bio->bi_end_io = endio_func;
2343 bio->bi_iter.bi_sector = failrec->logical >> 9;
2344 bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
2345 bio->bi_iter.bi_size = 0;
2346 bio->bi_private = data;
2348 btrfs_failed_bio = btrfs_io_bio(failed_bio);
2349 if (btrfs_failed_bio->csum) {
2350 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2351 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
2353 btrfs_bio = btrfs_io_bio(bio);
2354 btrfs_bio->csum = btrfs_bio->csum_inline;
2356 memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + icsum,
2360 bio_add_page(bio, page, failrec->len, pg_offset);
2366 * this is a generic handler for readpage errors (default
2367 * readpage_io_failed_hook). if other copies exist, read those and write back
2368 * good data to the failed position. does not investigate in remapping the
2369 * failed extent elsewhere, hoping the device will be smart enough to do this as
2373 static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset,
2374 struct page *page, u64 start, u64 end,
2377 struct io_failure_record *failrec;
2378 struct inode *inode = page->mapping->host;
2379 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2384 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
2386 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
2390 ret = btrfs_check_repairable(inode, failed_bio, failrec, failed_mirror);
2392 free_io_failure(inode, failrec);
2396 if (failed_bio->bi_vcnt > 1)
2397 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
2399 read_mode = READ_SYNC;
2401 phy_offset >>= inode->i_sb->s_blocksize_bits;
2402 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
2403 start - page_offset(page),
2404 (int)phy_offset, failed_bio->bi_end_io,
2407 free_io_failure(inode, failrec);
2411 pr_debug("Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d\n",
2412 read_mode, failrec->this_mirror, failrec->in_validation);
2414 ret = tree->ops->submit_bio_hook(inode, read_mode, bio,
2415 failrec->this_mirror,
2416 failrec->bio_flags, 0);
2418 free_io_failure(inode, failrec);
2425 /* lots and lots of room for performance fixes in the end_bio funcs */
2427 void end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2429 int uptodate = (err == 0);
2430 struct extent_io_tree *tree;
2433 tree = &BTRFS_I(page->mapping->host)->io_tree;
2435 if (tree->ops && tree->ops->writepage_end_io_hook) {
2436 ret = tree->ops->writepage_end_io_hook(page, start,
2437 end, NULL, uptodate);
2443 ClearPageUptodate(page);
2445 ret = ret < 0 ? ret : -EIO;
2446 mapping_set_error(page->mapping, ret);
2451 * after a writepage IO is done, we need to:
2452 * clear the uptodate bits on error
2453 * clear the writeback bits in the extent tree for this IO
2454 * end_page_writeback if the page has no more pending IO
2456 * Scheduling is not allowed, so the extent state tree is expected
2457 * to have one and only one object corresponding to this IO.
2459 static void end_bio_extent_writepage(struct bio *bio)
2461 struct bio_vec *bvec;
2466 bio_for_each_segment_all(bvec, bio, i) {
2467 struct page *page = bvec->bv_page;
2469 /* We always issue full-page reads, but if some block
2470 * in a page fails to read, blk_update_request() will
2471 * advance bv_offset and adjust bv_len to compensate.
2472 * Print a warning for nonzero offsets, and an error
2473 * if they don't add up to a full page. */
2474 if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE) {
2475 if (bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE)
2476 btrfs_err(BTRFS_I(page->mapping->host)->root->fs_info,
2477 "partial page write in btrfs with offset %u and length %u",
2478 bvec->bv_offset, bvec->bv_len);
2480 btrfs_info(BTRFS_I(page->mapping->host)->root->fs_info,
2481 "incomplete page write in btrfs with offset %u and "
2483 bvec->bv_offset, bvec->bv_len);
2486 start = page_offset(page);
2487 end = start + bvec->bv_offset + bvec->bv_len - 1;
2489 end_extent_writepage(page, bio->bi_error, start, end);
2490 end_page_writeback(page);
2497 endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len,
2500 struct extent_state *cached = NULL;
2501 u64 end = start + len - 1;
2503 if (uptodate && tree->track_uptodate)
2504 set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC);
2505 unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC);
2509 * after a readpage IO is done, we need to:
2510 * clear the uptodate bits on error
2511 * set the uptodate bits if things worked
2512 * set the page up to date if all extents in the tree are uptodate
2513 * clear the lock bit in the extent tree
2514 * unlock the page if there are no other extents locked for it
2516 * Scheduling is not allowed, so the extent state tree is expected
2517 * to have one and only one object corresponding to this IO.
2519 static void end_bio_extent_readpage(struct bio *bio)
2521 struct bio_vec *bvec;
2522 int uptodate = !bio->bi_error;
2523 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
2524 struct extent_io_tree *tree;
2529 u64 extent_start = 0;
2535 bio_for_each_segment_all(bvec, bio, i) {
2536 struct page *page = bvec->bv_page;
2537 struct inode *inode = page->mapping->host;
2539 pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, "
2540 "mirror=%u\n", (u64)bio->bi_iter.bi_sector,
2541 bio->bi_error, io_bio->mirror_num);
2542 tree = &BTRFS_I(inode)->io_tree;
2544 /* We always issue full-page reads, but if some block
2545 * in a page fails to read, blk_update_request() will
2546 * advance bv_offset and adjust bv_len to compensate.
2547 * Print a warning for nonzero offsets, and an error
2548 * if they don't add up to a full page. */
2549 if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE) {
2550 if (bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE)
2551 btrfs_err(BTRFS_I(page->mapping->host)->root->fs_info,
2552 "partial page read in btrfs with offset %u and length %u",
2553 bvec->bv_offset, bvec->bv_len);
2555 btrfs_info(BTRFS_I(page->mapping->host)->root->fs_info,
2556 "incomplete page read in btrfs with offset %u and "
2558 bvec->bv_offset, bvec->bv_len);
2561 start = page_offset(page);
2562 end = start + bvec->bv_offset + bvec->bv_len - 1;
2565 mirror = io_bio->mirror_num;
2566 if (likely(uptodate && tree->ops &&
2567 tree->ops->readpage_end_io_hook)) {
2568 ret = tree->ops->readpage_end_io_hook(io_bio, offset,
2574 clean_io_failure(inode, start, page, 0);
2577 if (likely(uptodate))
2580 if (tree->ops && tree->ops->readpage_io_failed_hook) {
2581 ret = tree->ops->readpage_io_failed_hook(page, mirror);
2582 if (!ret && !bio->bi_error)
2586 * The generic bio_readpage_error handles errors the
2587 * following way: If possible, new read requests are
2588 * created and submitted and will end up in
2589 * end_bio_extent_readpage as well (if we're lucky, not
2590 * in the !uptodate case). In that case it returns 0 and
2591 * we just go on with the next page in our bio. If it
2592 * can't handle the error it will return -EIO and we
2593 * remain responsible for that page.
2595 ret = bio_readpage_error(bio, offset, page, start, end,
2598 uptodate = !bio->bi_error;
2604 if (likely(uptodate)) {
2605 loff_t i_size = i_size_read(inode);
2606 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
2609 /* Zero out the end if this page straddles i_size */
2610 off = i_size & (PAGE_CACHE_SIZE-1);
2611 if (page->index == end_index && off)
2612 zero_user_segment(page, off, PAGE_CACHE_SIZE);
2613 SetPageUptodate(page);
2615 ClearPageUptodate(page);
2621 if (unlikely(!uptodate)) {
2623 endio_readpage_release_extent(tree,
2629 endio_readpage_release_extent(tree, start,
2630 end - start + 1, 0);
2631 } else if (!extent_len) {
2632 extent_start = start;
2633 extent_len = end + 1 - start;
2634 } else if (extent_start + extent_len == start) {
2635 extent_len += end + 1 - start;
2637 endio_readpage_release_extent(tree, extent_start,
2638 extent_len, uptodate);
2639 extent_start = start;
2640 extent_len = end + 1 - start;
2645 endio_readpage_release_extent(tree, extent_start, extent_len,
2648 io_bio->end_io(io_bio, bio->bi_error);
2653 * this allocates from the btrfs_bioset. We're returning a bio right now
2654 * but you can call btrfs_io_bio for the appropriate container_of magic
2657 btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
2660 struct btrfs_io_bio *btrfs_bio;
2663 bio = bio_alloc_bioset(gfp_flags, nr_vecs, btrfs_bioset);
2665 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
2666 while (!bio && (nr_vecs /= 2)) {
2667 bio = bio_alloc_bioset(gfp_flags,
2668 nr_vecs, btrfs_bioset);
2673 bio->bi_bdev = bdev;
2674 bio->bi_iter.bi_sector = first_sector;
2675 btrfs_bio = btrfs_io_bio(bio);
2676 btrfs_bio->csum = NULL;
2677 btrfs_bio->csum_allocated = NULL;
2678 btrfs_bio->end_io = NULL;
2683 struct bio *btrfs_bio_clone(struct bio *bio, gfp_t gfp_mask)
2685 struct btrfs_io_bio *btrfs_bio;
2688 new = bio_clone_bioset(bio, gfp_mask, btrfs_bioset);
2690 btrfs_bio = btrfs_io_bio(new);
2691 btrfs_bio->csum = NULL;
2692 btrfs_bio->csum_allocated = NULL;
2693 btrfs_bio->end_io = NULL;
2695 #ifdef CONFIG_BLK_CGROUP
2696 /* FIXME, put this into bio_clone_bioset */
2698 bio_associate_blkcg(new, bio->bi_css);
2704 /* this also allocates from the btrfs_bioset */
2705 struct bio *btrfs_io_bio_alloc(gfp_t gfp_mask, unsigned int nr_iovecs)
2707 struct btrfs_io_bio *btrfs_bio;
2710 bio = bio_alloc_bioset(gfp_mask, nr_iovecs, btrfs_bioset);
2712 btrfs_bio = btrfs_io_bio(bio);
2713 btrfs_bio->csum = NULL;
2714 btrfs_bio->csum_allocated = NULL;
2715 btrfs_bio->end_io = NULL;
2721 static int __must_check submit_one_bio(int rw, struct bio *bio,
2722 int mirror_num, unsigned long bio_flags)
2725 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2726 struct page *page = bvec->bv_page;
2727 struct extent_io_tree *tree = bio->bi_private;
2730 start = page_offset(page) + bvec->bv_offset;
2732 bio->bi_private = NULL;
2736 if (tree->ops && tree->ops->submit_bio_hook)
2737 ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
2738 mirror_num, bio_flags, start);
2740 btrfsic_submit_bio(rw, bio);
2746 static int merge_bio(int rw, struct extent_io_tree *tree, struct page *page,
2747 unsigned long offset, size_t size, struct bio *bio,
2748 unsigned long bio_flags)
2751 if (tree->ops && tree->ops->merge_bio_hook)
2752 ret = tree->ops->merge_bio_hook(rw, page, offset, size, bio,
2759 static int submit_extent_page(int rw, struct extent_io_tree *tree,
2760 struct writeback_control *wbc,
2761 struct page *page, sector_t sector,
2762 size_t size, unsigned long offset,
2763 struct block_device *bdev,
2764 struct bio **bio_ret,
2765 unsigned long max_pages,
2766 bio_end_io_t end_io_func,
2768 unsigned long prev_bio_flags,
2769 unsigned long bio_flags,
2770 bool force_bio_submit)
2775 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
2776 size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE);
2778 if (bio_ret && *bio_ret) {
2781 contig = bio->bi_iter.bi_sector == sector;
2783 contig = bio_end_sector(bio) == sector;
2785 if (prev_bio_flags != bio_flags || !contig ||
2787 merge_bio(rw, tree, page, offset, page_size, bio, bio_flags) ||
2788 bio_add_page(bio, page, page_size, offset) < page_size) {
2789 ret = submit_one_bio(rw, bio, mirror_num,
2798 wbc_account_io(wbc, page, page_size);
2803 bio = btrfs_bio_alloc(bdev, sector, BIO_MAX_PAGES,
2804 GFP_NOFS | __GFP_HIGH);
2808 bio_add_page(bio, page, page_size, offset);
2809 bio->bi_end_io = end_io_func;
2810 bio->bi_private = tree;
2812 wbc_init_bio(wbc, bio);
2813 wbc_account_io(wbc, page, page_size);
2819 ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
2824 static void attach_extent_buffer_page(struct extent_buffer *eb,
2827 if (!PagePrivate(page)) {
2828 SetPagePrivate(page);
2829 page_cache_get(page);
2830 set_page_private(page, (unsigned long)eb);
2832 WARN_ON(page->private != (unsigned long)eb);
2836 void set_page_extent_mapped(struct page *page)
2838 if (!PagePrivate(page)) {
2839 SetPagePrivate(page);
2840 page_cache_get(page);
2841 set_page_private(page, EXTENT_PAGE_PRIVATE);
2845 static struct extent_map *
2846 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
2847 u64 start, u64 len, get_extent_t *get_extent,
2848 struct extent_map **em_cached)
2850 struct extent_map *em;
2852 if (em_cached && *em_cached) {
2854 if (extent_map_in_tree(em) && start >= em->start &&
2855 start < extent_map_end(em)) {
2856 atomic_inc(&em->refs);
2860 free_extent_map(em);
2864 em = get_extent(inode, page, pg_offset, start, len, 0);
2865 if (em_cached && !IS_ERR_OR_NULL(em)) {
2867 atomic_inc(&em->refs);
2873 * basic readpage implementation. Locked extent state structs are inserted
2874 * into the tree that are removed when the IO is done (by the end_io
2876 * XXX JDM: This needs looking at to ensure proper page locking
2878 static int __do_readpage(struct extent_io_tree *tree,
2880 get_extent_t *get_extent,
2881 struct extent_map **em_cached,
2882 struct bio **bio, int mirror_num,
2883 unsigned long *bio_flags, int rw,
2886 struct inode *inode = page->mapping->host;
2887 u64 start = page_offset(page);
2888 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2892 u64 last_byte = i_size_read(inode);
2896 struct extent_map *em;
2897 struct block_device *bdev;
2900 int parent_locked = *bio_flags & EXTENT_BIO_PARENT_LOCKED;
2901 size_t pg_offset = 0;
2903 size_t disk_io_size;
2904 size_t blocksize = inode->i_sb->s_blocksize;
2905 unsigned long this_bio_flag = *bio_flags & EXTENT_BIO_PARENT_LOCKED;
2907 set_page_extent_mapped(page);
2910 if (!PageUptodate(page)) {
2911 if (cleancache_get_page(page) == 0) {
2912 BUG_ON(blocksize != PAGE_SIZE);
2913 unlock_extent(tree, start, end);
2918 if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
2920 size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
2923 iosize = PAGE_CACHE_SIZE - zero_offset;
2924 userpage = kmap_atomic(page);
2925 memset(userpage + zero_offset, 0, iosize);
2926 flush_dcache_page(page);
2927 kunmap_atomic(userpage);
2930 while (cur <= end) {
2931 unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
2932 bool force_bio_submit = false;
2934 if (cur >= last_byte) {
2936 struct extent_state *cached = NULL;
2938 iosize = PAGE_CACHE_SIZE - pg_offset;
2939 userpage = kmap_atomic(page);
2940 memset(userpage + pg_offset, 0, iosize);
2941 flush_dcache_page(page);
2942 kunmap_atomic(userpage);
2943 set_extent_uptodate(tree, cur, cur + iosize - 1,
2946 unlock_extent_cached(tree, cur,
2951 em = __get_extent_map(inode, page, pg_offset, cur,
2952 end - cur + 1, get_extent, em_cached);
2953 if (IS_ERR_OR_NULL(em)) {
2956 unlock_extent(tree, cur, end);
2959 extent_offset = cur - em->start;
2960 BUG_ON(extent_map_end(em) <= cur);
2963 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2964 this_bio_flag |= EXTENT_BIO_COMPRESSED;
2965 extent_set_compress_type(&this_bio_flag,
2969 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2970 cur_end = min(extent_map_end(em) - 1, end);
2971 iosize = ALIGN(iosize, blocksize);
2972 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2973 disk_io_size = em->block_len;
2974 sector = em->block_start >> 9;
2976 sector = (em->block_start + extent_offset) >> 9;
2977 disk_io_size = iosize;
2980 block_start = em->block_start;
2981 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2982 block_start = EXTENT_MAP_HOLE;
2985 * If we have a file range that points to a compressed extent
2986 * and it's followed by a consecutive file range that points to
2987 * to the same compressed extent (possibly with a different
2988 * offset and/or length, so it either points to the whole extent
2989 * or only part of it), we must make sure we do not submit a
2990 * single bio to populate the pages for the 2 ranges because
2991 * this makes the compressed extent read zero out the pages
2992 * belonging to the 2nd range. Imagine the following scenario:
2995 * [0 - 8K] [8K - 24K]
2998 * points to extent X, points to extent X,
2999 * offset 4K, length of 8K offset 0, length 16K
3001 * [extent X, compressed length = 4K uncompressed length = 16K]
3003 * If the bio to read the compressed extent covers both ranges,
3004 * it will decompress extent X into the pages belonging to the
3005 * first range and then it will stop, zeroing out the remaining
3006 * pages that belong to the other range that points to extent X.
3007 * So here we make sure we submit 2 bios, one for the first
3008 * range and another one for the third range. Both will target
3009 * the same physical extent from disk, but we can't currently
3010 * make the compressed bio endio callback populate the pages
3011 * for both ranges because each compressed bio is tightly
3012 * coupled with a single extent map, and each range can have
3013 * an extent map with a different offset value relative to the
3014 * uncompressed data of our extent and different lengths. This
3015 * is a corner case so we prioritize correctness over
3016 * non-optimal behavior (submitting 2 bios for the same extent).
3018 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) &&
3019 prev_em_start && *prev_em_start != (u64)-1 &&
3020 *prev_em_start != em->orig_start)
3021 force_bio_submit = true;
3024 *prev_em_start = em->orig_start;
3026 free_extent_map(em);
3029 /* we've found a hole, just zero and go on */
3030 if (block_start == EXTENT_MAP_HOLE) {
3032 struct extent_state *cached = NULL;
3034 userpage = kmap_atomic(page);
3035 memset(userpage + pg_offset, 0, iosize);
3036 flush_dcache_page(page);
3037 kunmap_atomic(userpage);
3039 set_extent_uptodate(tree, cur, cur + iosize - 1,
3042 free_extent_state(cached);
3044 unlock_extent_cached(tree, cur,
3048 pg_offset += iosize;
3051 /* the get_extent function already copied into the page */
3052 if (test_range_bit(tree, cur, cur_end,
3053 EXTENT_UPTODATE, 1, NULL)) {
3054 check_page_uptodate(tree, page);
3056 unlock_extent(tree, cur, cur + iosize - 1);
3058 pg_offset += iosize;
3061 /* we have an inline extent but it didn't get marked up
3062 * to date. Error out
3064 if (block_start == EXTENT_MAP_INLINE) {
3067 unlock_extent(tree, cur, cur + iosize - 1);
3069 pg_offset += iosize;
3074 ret = submit_extent_page(rw, tree, NULL, page,
3075 sector, disk_io_size, pg_offset,
3077 end_bio_extent_readpage, mirror_num,
3083 *bio_flags = this_bio_flag;
3087 unlock_extent(tree, cur, cur + iosize - 1);
3090 pg_offset += iosize;
3094 if (!PageError(page))
3095 SetPageUptodate(page);
3101 static inline void __do_contiguous_readpages(struct extent_io_tree *tree,
3102 struct page *pages[], int nr_pages,
3104 get_extent_t *get_extent,
3105 struct extent_map **em_cached,
3106 struct bio **bio, int mirror_num,
3107 unsigned long *bio_flags, int rw,
3110 struct inode *inode;
3111 struct btrfs_ordered_extent *ordered;
3114 inode = pages[0]->mapping->host;
3116 lock_extent(tree, start, end);
3117 ordered = btrfs_lookup_ordered_range(inode, start,
3121 unlock_extent(tree, start, end);
3122 btrfs_start_ordered_extent(inode, ordered, 1);
3123 btrfs_put_ordered_extent(ordered);
3126 for (index = 0; index < nr_pages; index++) {
3127 __do_readpage(tree, pages[index], get_extent, em_cached, bio,
3128 mirror_num, bio_flags, rw, prev_em_start);
3129 page_cache_release(pages[index]);
3133 static void __extent_readpages(struct extent_io_tree *tree,
3134 struct page *pages[],
3135 int nr_pages, get_extent_t *get_extent,
3136 struct extent_map **em_cached,
3137 struct bio **bio, int mirror_num,
3138 unsigned long *bio_flags, int rw,
3145 int first_index = 0;
3147 for (index = 0; index < nr_pages; index++) {
3148 page_start = page_offset(pages[index]);
3151 end = start + PAGE_CACHE_SIZE - 1;
3152 first_index = index;
3153 } else if (end + 1 == page_start) {
3154 end += PAGE_CACHE_SIZE;
3156 __do_contiguous_readpages(tree, &pages[first_index],
3157 index - first_index, start,
3158 end, get_extent, em_cached,
3159 bio, mirror_num, bio_flags,
3162 end = start + PAGE_CACHE_SIZE - 1;
3163 first_index = index;
3168 __do_contiguous_readpages(tree, &pages[first_index],
3169 index - first_index, start,
3170 end, get_extent, em_cached, bio,
3171 mirror_num, bio_flags, rw,
3175 static int __extent_read_full_page(struct extent_io_tree *tree,
3177 get_extent_t *get_extent,
3178 struct bio **bio, int mirror_num,
3179 unsigned long *bio_flags, int rw)
3181 struct inode *inode = page->mapping->host;
3182 struct btrfs_ordered_extent *ordered;
3183 u64 start = page_offset(page);
3184 u64 end = start + PAGE_CACHE_SIZE - 1;
3188 lock_extent(tree, start, end);
3189 ordered = btrfs_lookup_ordered_range(inode, start,
3193 unlock_extent(tree, start, end);
3194 btrfs_start_ordered_extent(inode, ordered, 1);
3195 btrfs_put_ordered_extent(ordered);
3198 ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num,
3199 bio_flags, rw, NULL);
3203 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
3204 get_extent_t *get_extent, int mirror_num)
3206 struct bio *bio = NULL;
3207 unsigned long bio_flags = 0;
3210 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
3213 ret = submit_one_bio(READ, bio, mirror_num, bio_flags);
3217 int extent_read_full_page_nolock(struct extent_io_tree *tree, struct page *page,
3218 get_extent_t *get_extent, int mirror_num)
3220 struct bio *bio = NULL;
3221 unsigned long bio_flags = EXTENT_BIO_PARENT_LOCKED;
3224 ret = __do_readpage(tree, page, get_extent, NULL, &bio, mirror_num,
3225 &bio_flags, READ, NULL);
3227 ret = submit_one_bio(READ, bio, mirror_num, bio_flags);
3231 static noinline void update_nr_written(struct page *page,
3232 struct writeback_control *wbc,
3233 unsigned long nr_written)
3235 wbc->nr_to_write -= nr_written;
3236 if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
3237 wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
3238 page->mapping->writeback_index = page->index + nr_written;
3242 * helper for __extent_writepage, doing all of the delayed allocation setup.
3244 * This returns 1 if our fill_delalloc function did all the work required
3245 * to write the page (copy into inline extent). In this case the IO has
3246 * been started and the page is already unlocked.
3248 * This returns 0 if all went well (page still locked)
3249 * This returns < 0 if there were errors (page still locked)
3251 static noinline_for_stack int writepage_delalloc(struct inode *inode,
3252 struct page *page, struct writeback_control *wbc,
3253 struct extent_page_data *epd,
3255 unsigned long *nr_written)
3257 struct extent_io_tree *tree = epd->tree;
3258 u64 page_end = delalloc_start + PAGE_CACHE_SIZE - 1;
3260 u64 delalloc_to_write = 0;
3261 u64 delalloc_end = 0;
3263 int page_started = 0;
3265 if (epd->extent_locked || !tree->ops || !tree->ops->fill_delalloc)
3268 while (delalloc_end < page_end) {
3269 nr_delalloc = find_lock_delalloc_range(inode, tree,
3273 BTRFS_MAX_EXTENT_SIZE);
3274 if (nr_delalloc == 0) {
3275 delalloc_start = delalloc_end + 1;
3278 ret = tree->ops->fill_delalloc(inode, page,
3283 /* File system has been set read-only */
3286 /* fill_delalloc should be return < 0 for error
3287 * but just in case, we use > 0 here meaning the
3288 * IO is started, so we don't want to return > 0
3289 * unless things are going well.
3291 ret = ret < 0 ? ret : -EIO;
3295 * delalloc_end is already one less than the total
3296 * length, so we don't subtract one from
3299 delalloc_to_write += (delalloc_end - delalloc_start +
3302 delalloc_start = delalloc_end + 1;
3304 if (wbc->nr_to_write < delalloc_to_write) {
3307 if (delalloc_to_write < thresh * 2)
3308 thresh = delalloc_to_write;
3309 wbc->nr_to_write = min_t(u64, delalloc_to_write,
3313 /* did the fill delalloc function already unlock and start
3318 * we've unlocked the page, so we can't update
3319 * the mapping's writeback index, just update
3322 wbc->nr_to_write -= *nr_written;
3333 * helper for __extent_writepage. This calls the writepage start hooks,
3334 * and does the loop to map the page into extents and bios.
3336 * We return 1 if the IO is started and the page is unlocked,
3337 * 0 if all went well (page still locked)
3338 * < 0 if there were errors (page still locked)
3340 static noinline_for_stack int __extent_writepage_io(struct inode *inode,
3342 struct writeback_control *wbc,
3343 struct extent_page_data *epd,
3345 unsigned long nr_written,
3346 int write_flags, int *nr_ret)
3348 struct extent_io_tree *tree = epd->tree;
3349 u64 start = page_offset(page);
3350 u64 page_end = start + PAGE_CACHE_SIZE - 1;
3357 struct extent_state *cached_state = NULL;
3358 struct extent_map *em;
3359 struct block_device *bdev;
3360 size_t pg_offset = 0;
3366 if (tree->ops && tree->ops->writepage_start_hook) {
3367 ret = tree->ops->writepage_start_hook(page, start,
3370 /* Fixup worker will requeue */
3372 wbc->pages_skipped++;
3374 redirty_page_for_writepage(wbc, page);
3376 update_nr_written(page, wbc, nr_written);
3384 * we don't want to touch the inode after unlocking the page,
3385 * so we update the mapping writeback index now
3387 update_nr_written(page, wbc, nr_written + 1);
3390 if (i_size <= start) {
3391 if (tree->ops && tree->ops->writepage_end_io_hook)
3392 tree->ops->writepage_end_io_hook(page, start,
3397 blocksize = inode->i_sb->s_blocksize;
3399 while (cur <= end) {
3401 if (cur >= i_size) {
3402 if (tree->ops && tree->ops->writepage_end_io_hook)
3403 tree->ops->writepage_end_io_hook(page, cur,
3407 em = epd->get_extent(inode, page, pg_offset, cur,
3409 if (IS_ERR_OR_NULL(em)) {
3411 ret = PTR_ERR_OR_ZERO(em);
3415 extent_offset = cur - em->start;
3416 em_end = extent_map_end(em);
3417 BUG_ON(em_end <= cur);
3419 iosize = min(em_end - cur, end - cur + 1);
3420 iosize = ALIGN(iosize, blocksize);
3421 sector = (em->block_start + extent_offset) >> 9;
3423 block_start = em->block_start;
3424 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3425 free_extent_map(em);
3429 * compressed and inline extents are written through other
3432 if (compressed || block_start == EXTENT_MAP_HOLE ||
3433 block_start == EXTENT_MAP_INLINE) {
3435 * end_io notification does not happen here for
3436 * compressed extents
3438 if (!compressed && tree->ops &&
3439 tree->ops->writepage_end_io_hook)
3440 tree->ops->writepage_end_io_hook(page, cur,
3443 else if (compressed) {
3444 /* we don't want to end_page_writeback on
3445 * a compressed extent. this happens
3452 pg_offset += iosize;
3456 if (tree->ops && tree->ops->writepage_io_hook) {
3457 ret = tree->ops->writepage_io_hook(page, cur,
3465 unsigned long max_nr = (i_size >> PAGE_CACHE_SHIFT) + 1;
3467 set_range_writeback(tree, cur, cur + iosize - 1);
3468 if (!PageWriteback(page)) {
3469 btrfs_err(BTRFS_I(inode)->root->fs_info,
3470 "page %lu not writeback, cur %llu end %llu",
3471 page->index, cur, end);
3474 ret = submit_extent_page(write_flags, tree, wbc, page,
3475 sector, iosize, pg_offset,
3476 bdev, &epd->bio, max_nr,
3477 end_bio_extent_writepage,
3483 pg_offset += iosize;
3491 /* drop our reference on any cached states */
3492 free_extent_state(cached_state);
3497 * the writepage semantics are similar to regular writepage. extent
3498 * records are inserted to lock ranges in the tree, and as dirty areas
3499 * are found, they are marked writeback. Then the lock bits are removed
3500 * and the end_io handler clears the writeback ranges
3502 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
3505 struct inode *inode = page->mapping->host;
3506 struct extent_page_data *epd = data;
3507 u64 start = page_offset(page);
3508 u64 page_end = start + PAGE_CACHE_SIZE - 1;
3511 size_t pg_offset = 0;
3512 loff_t i_size = i_size_read(inode);
3513 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
3515 unsigned long nr_written = 0;
3517 if (wbc->sync_mode == WB_SYNC_ALL)
3518 write_flags = WRITE_SYNC;
3520 write_flags = WRITE;
3522 trace___extent_writepage(page, inode, wbc);
3524 WARN_ON(!PageLocked(page));
3526 ClearPageError(page);
3528 pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
3529 if (page->index > end_index ||
3530 (page->index == end_index && !pg_offset)) {
3531 page->mapping->a_ops->invalidatepage(page, 0, PAGE_CACHE_SIZE);
3536 if (page->index == end_index) {
3539 userpage = kmap_atomic(page);
3540 memset(userpage + pg_offset, 0,
3541 PAGE_CACHE_SIZE - pg_offset);
3542 kunmap_atomic(userpage);
3543 flush_dcache_page(page);
3548 set_page_extent_mapped(page);
3550 ret = writepage_delalloc(inode, page, wbc, epd, start, &nr_written);
3556 ret = __extent_writepage_io(inode, page, wbc, epd,
3557 i_size, nr_written, write_flags, &nr);
3563 /* make sure the mapping tag for page dirty gets cleared */
3564 set_page_writeback(page);
3565 end_page_writeback(page);
3567 if (PageError(page)) {
3568 ret = ret < 0 ? ret : -EIO;
3569 end_extent_writepage(page, ret, start, page_end);
3578 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3580 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
3581 TASK_UNINTERRUPTIBLE);
3584 static noinline_for_stack int
3585 lock_extent_buffer_for_io(struct extent_buffer *eb,
3586 struct btrfs_fs_info *fs_info,
3587 struct extent_page_data *epd)
3589 unsigned long i, num_pages;
3593 if (!btrfs_try_tree_write_lock(eb)) {
3595 flush_write_bio(epd);
3596 btrfs_tree_lock(eb);
3599 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3600 btrfs_tree_unlock(eb);
3604 flush_write_bio(epd);
3608 wait_on_extent_buffer_writeback(eb);
3609 btrfs_tree_lock(eb);
3610 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3612 btrfs_tree_unlock(eb);
3617 * We need to do this to prevent races in people who check if the eb is
3618 * under IO since we can end up having no IO bits set for a short period
3621 spin_lock(&eb->refs_lock);
3622 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3623 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3624 spin_unlock(&eb->refs_lock);
3625 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3626 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
3628 fs_info->dirty_metadata_batch);
3631 spin_unlock(&eb->refs_lock);
3634 btrfs_tree_unlock(eb);
3639 num_pages = num_extent_pages(eb->start, eb->len);
3640 for (i = 0; i < num_pages; i++) {
3641 struct page *p = eb->pages[i];
3643 if (!trylock_page(p)) {
3645 flush_write_bio(epd);
3655 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3657 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3658 smp_mb__after_atomic();
3659 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3662 static void set_btree_ioerr(struct page *page)
3664 struct extent_buffer *eb = (struct extent_buffer *)page->private;
3665 struct btrfs_inode *btree_ino = BTRFS_I(eb->fs_info->btree_inode);
3668 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))
3672 * If writeback for a btree extent that doesn't belong to a log tree
3673 * failed, increment the counter transaction->eb_write_errors.
3674 * We do this because while the transaction is running and before it's
3675 * committing (when we call filemap_fdata[write|wait]_range against
3676 * the btree inode), we might have
3677 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
3678 * returns an error or an error happens during writeback, when we're
3679 * committing the transaction we wouldn't know about it, since the pages
3680 * can be no longer dirty nor marked anymore for writeback (if a
3681 * subsequent modification to the extent buffer didn't happen before the
3682 * transaction commit), which makes filemap_fdata[write|wait]_range not
3683 * able to find the pages tagged with SetPageError at transaction
3684 * commit time. So if this happens we must abort the transaction,
3685 * otherwise we commit a super block with btree roots that point to
3686 * btree nodes/leafs whose content on disk is invalid - either garbage
3687 * or the content of some node/leaf from a past generation that got
3688 * cowed or deleted and is no longer valid.
3690 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
3691 * not be enough - we need to distinguish between log tree extents vs
3692 * non-log tree extents, and the next filemap_fdatawait_range() call
3693 * will catch and clear such errors in the mapping - and that call might
3694 * be from a log sync and not from a transaction commit. Also, checking
3695 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
3696 * not done and would not be reliable - the eb might have been released
3697 * from memory and reading it back again means that flag would not be
3698 * set (since it's a runtime flag, not persisted on disk).
3700 * Using the flags below in the btree inode also makes us achieve the
3701 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
3702 * writeback for all dirty pages and before filemap_fdatawait_range()
3703 * is called, the writeback for all dirty pages had already finished
3704 * with errors - because we were not using AS_EIO/AS_ENOSPC,
3705 * filemap_fdatawait_range() would return success, as it could not know
3706 * that writeback errors happened (the pages were no longer tagged for
3709 switch (eb->log_index) {
3711 set_bit(BTRFS_INODE_BTREE_ERR, &btree_ino->runtime_flags);
3714 set_bit(BTRFS_INODE_BTREE_LOG1_ERR, &btree_ino->runtime_flags);
3717 set_bit(BTRFS_INODE_BTREE_LOG2_ERR, &btree_ino->runtime_flags);
3720 BUG(); /* unexpected, logic error */
3724 static void end_bio_extent_buffer_writepage(struct bio *bio)
3726 struct bio_vec *bvec;
3727 struct extent_buffer *eb;
3730 bio_for_each_segment_all(bvec, bio, i) {
3731 struct page *page = bvec->bv_page;
3733 eb = (struct extent_buffer *)page->private;
3735 done = atomic_dec_and_test(&eb->io_pages);
3737 if (bio->bi_error ||
3738 test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) {
3739 ClearPageUptodate(page);
3740 set_btree_ioerr(page);
3743 end_page_writeback(page);
3748 end_extent_buffer_writeback(eb);
3754 static noinline_for_stack int write_one_eb(struct extent_buffer *eb,
3755 struct btrfs_fs_info *fs_info,
3756 struct writeback_control *wbc,
3757 struct extent_page_data *epd)
3759 struct block_device *bdev = fs_info->fs_devices->latest_bdev;
3760 struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
3761 u64 offset = eb->start;
3762 unsigned long i, num_pages;
3763 unsigned long bio_flags = 0;
3764 int rw = (epd->sync_io ? WRITE_SYNC : WRITE) | REQ_META;
3767 clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
3768 num_pages = num_extent_pages(eb->start, eb->len);
3769 atomic_set(&eb->io_pages, num_pages);
3770 if (btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID)
3771 bio_flags = EXTENT_BIO_TREE_LOG;
3773 for (i = 0; i < num_pages; i++) {
3774 struct page *p = eb->pages[i];
3776 clear_page_dirty_for_io(p);
3777 set_page_writeback(p);
3778 ret = submit_extent_page(rw, tree, wbc, p, offset >> 9,
3779 PAGE_CACHE_SIZE, 0, bdev, &epd->bio,
3780 -1, end_bio_extent_buffer_writepage,
3781 0, epd->bio_flags, bio_flags, false);
3782 epd->bio_flags = bio_flags;
3785 end_page_writeback(p);
3786 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3787 end_extent_buffer_writeback(eb);
3791 offset += PAGE_CACHE_SIZE;
3792 update_nr_written(p, wbc, 1);
3796 if (unlikely(ret)) {
3797 for (; i < num_pages; i++) {
3798 struct page *p = eb->pages[i];
3799 clear_page_dirty_for_io(p);
3807 int btree_write_cache_pages(struct address_space *mapping,
3808 struct writeback_control *wbc)
3810 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
3811 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
3812 struct extent_buffer *eb, *prev_eb = NULL;
3813 struct extent_page_data epd = {
3817 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3822 int nr_to_write_done = 0;
3823 struct pagevec pvec;
3826 pgoff_t end; /* Inclusive */
3830 pagevec_init(&pvec, 0);
3831 if (wbc->range_cyclic) {
3832 index = mapping->writeback_index; /* Start from prev offset */
3835 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3836 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3839 if (wbc->sync_mode == WB_SYNC_ALL)
3840 tag = PAGECACHE_TAG_TOWRITE;
3842 tag = PAGECACHE_TAG_DIRTY;
3844 if (wbc->sync_mode == WB_SYNC_ALL)
3845 tag_pages_for_writeback(mapping, index, end);
3846 while (!done && !nr_to_write_done && (index <= end) &&
3847 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3848 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3852 for (i = 0; i < nr_pages; i++) {
3853 struct page *page = pvec.pages[i];
3855 if (!PagePrivate(page))
3858 if (!wbc->range_cyclic && page->index > end) {
3863 spin_lock(&mapping->private_lock);
3864 if (!PagePrivate(page)) {
3865 spin_unlock(&mapping->private_lock);
3869 eb = (struct extent_buffer *)page->private;
3872 * Shouldn't happen and normally this would be a BUG_ON
3873 * but no sense in crashing the users box for something
3874 * we can survive anyway.
3877 spin_unlock(&mapping->private_lock);
3881 if (eb == prev_eb) {
3882 spin_unlock(&mapping->private_lock);
3886 ret = atomic_inc_not_zero(&eb->refs);
3887 spin_unlock(&mapping->private_lock);
3892 ret = lock_extent_buffer_for_io(eb, fs_info, &epd);
3894 free_extent_buffer(eb);
3898 ret = write_one_eb(eb, fs_info, wbc, &epd);
3901 free_extent_buffer(eb);
3904 free_extent_buffer(eb);
3907 * the filesystem may choose to bump up nr_to_write.
3908 * We have to make sure to honor the new nr_to_write
3911 nr_to_write_done = wbc->nr_to_write <= 0;
3913 pagevec_release(&pvec);
3916 if (!scanned && !done) {
3918 * We hit the last page and there is more work to be done: wrap
3919 * back to the start of the file
3925 flush_write_bio(&epd);
3930 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3931 * @mapping: address space structure to write
3932 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3933 * @writepage: function called for each page
3934 * @data: data passed to writepage function
3936 * If a page is already under I/O, write_cache_pages() skips it, even
3937 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3938 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3939 * and msync() need to guarantee that all the data which was dirty at the time
3940 * the call was made get new I/O started against them. If wbc->sync_mode is
3941 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3942 * existing IO to complete.
3944 static int extent_write_cache_pages(struct extent_io_tree *tree,
3945 struct address_space *mapping,
3946 struct writeback_control *wbc,
3947 writepage_t writepage, void *data,
3948 void (*flush_fn)(void *))
3950 struct inode *inode = mapping->host;
3954 int nr_to_write_done = 0;
3955 struct pagevec pvec;
3958 pgoff_t end; /* Inclusive */
3963 * We have to hold onto the inode so that ordered extents can do their
3964 * work when the IO finishes. The alternative to this is failing to add
3965 * an ordered extent if the igrab() fails there and that is a huge pain
3966 * to deal with, so instead just hold onto the inode throughout the
3967 * writepages operation. If it fails here we are freeing up the inode
3968 * anyway and we'd rather not waste our time writing out stuff that is
3969 * going to be truncated anyway.
3974 pagevec_init(&pvec, 0);
3975 if (wbc->range_cyclic) {
3976 index = mapping->writeback_index; /* Start from prev offset */
3979 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3980 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3983 if (wbc->sync_mode == WB_SYNC_ALL)
3984 tag = PAGECACHE_TAG_TOWRITE;
3986 tag = PAGECACHE_TAG_DIRTY;
3988 if (wbc->sync_mode == WB_SYNC_ALL)
3989 tag_pages_for_writeback(mapping, index, end);
3990 while (!done && !nr_to_write_done && (index <= end) &&
3991 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3992 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3996 for (i = 0; i < nr_pages; i++) {
3997 struct page *page = pvec.pages[i];
4000 * At this point we hold neither mapping->tree_lock nor
4001 * lock on the page itself: the page may be truncated or
4002 * invalidated (changing page->mapping to NULL), or even
4003 * swizzled back from swapper_space to tmpfs file
4006 if (!trylock_page(page)) {
4011 if (unlikely(page->mapping != mapping)) {
4016 if (!wbc->range_cyclic && page->index > end) {
4022 if (wbc->sync_mode != WB_SYNC_NONE) {
4023 if (PageWriteback(page))
4025 wait_on_page_writeback(page);
4028 if (PageWriteback(page) ||
4029 !clear_page_dirty_for_io(page)) {
4034 ret = (*writepage)(page, wbc, data);
4036 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
4040 if (!err && ret < 0)
4044 * the filesystem may choose to bump up nr_to_write.
4045 * We have to make sure to honor the new nr_to_write
4048 nr_to_write_done = wbc->nr_to_write <= 0;
4050 pagevec_release(&pvec);
4053 if (!scanned && !done && !err) {
4055 * We hit the last page and there is more work to be done: wrap
4056 * back to the start of the file
4062 btrfs_add_delayed_iput(inode);
4066 static void flush_epd_write_bio(struct extent_page_data *epd)
4075 ret = submit_one_bio(rw, epd->bio, 0, epd->bio_flags);
4076 BUG_ON(ret < 0); /* -ENOMEM */
4081 static noinline void flush_write_bio(void *data)
4083 struct extent_page_data *epd = data;
4084 flush_epd_write_bio(epd);
4087 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
4088 get_extent_t *get_extent,
4089 struct writeback_control *wbc)
4092 struct extent_page_data epd = {
4095 .get_extent = get_extent,
4097 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4101 ret = __extent_writepage(page, wbc, &epd);
4103 flush_epd_write_bio(&epd);
4107 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
4108 u64 start, u64 end, get_extent_t *get_extent,
4112 struct address_space *mapping = inode->i_mapping;
4114 unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
4117 struct extent_page_data epd = {
4120 .get_extent = get_extent,
4122 .sync_io = mode == WB_SYNC_ALL,
4125 struct writeback_control wbc_writepages = {
4127 .nr_to_write = nr_pages * 2,
4128 .range_start = start,
4129 .range_end = end + 1,
4132 while (start <= end) {
4133 page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
4134 if (clear_page_dirty_for_io(page))
4135 ret = __extent_writepage(page, &wbc_writepages, &epd);
4137 if (tree->ops && tree->ops->writepage_end_io_hook)
4138 tree->ops->writepage_end_io_hook(page, start,
4139 start + PAGE_CACHE_SIZE - 1,
4143 page_cache_release(page);
4144 start += PAGE_CACHE_SIZE;
4147 flush_epd_write_bio(&epd);
4151 int extent_writepages(struct extent_io_tree *tree,
4152 struct address_space *mapping,
4153 get_extent_t *get_extent,
4154 struct writeback_control *wbc)
4157 struct extent_page_data epd = {
4160 .get_extent = get_extent,
4162 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4166 ret = extent_write_cache_pages(tree, mapping, wbc,
4167 __extent_writepage, &epd,
4169 flush_epd_write_bio(&epd);
4173 int extent_readpages(struct extent_io_tree *tree,
4174 struct address_space *mapping,
4175 struct list_head *pages, unsigned nr_pages,
4176 get_extent_t get_extent)
4178 struct bio *bio = NULL;
4180 unsigned long bio_flags = 0;
4181 struct page *pagepool[16];
4183 struct extent_map *em_cached = NULL;
4185 u64 prev_em_start = (u64)-1;
4187 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
4188 page = list_entry(pages->prev, struct page, lru);
4190 prefetchw(&page->flags);
4191 list_del(&page->lru);
4192 if (add_to_page_cache_lru(page, mapping,
4193 page->index, GFP_NOFS)) {
4194 page_cache_release(page);
4198 pagepool[nr++] = page;
4199 if (nr < ARRAY_SIZE(pagepool))
4201 __extent_readpages(tree, pagepool, nr, get_extent, &em_cached,
4202 &bio, 0, &bio_flags, READ, &prev_em_start);
4206 __extent_readpages(tree, pagepool, nr, get_extent, &em_cached,
4207 &bio, 0, &bio_flags, READ, &prev_em_start);
4210 free_extent_map(em_cached);
4212 BUG_ON(!list_empty(pages));
4214 return submit_one_bio(READ, bio, 0, bio_flags);
4219 * basic invalidatepage code, this waits on any locked or writeback
4220 * ranges corresponding to the page, and then deletes any extent state
4221 * records from the tree
4223 int extent_invalidatepage(struct extent_io_tree *tree,
4224 struct page *page, unsigned long offset)
4226 struct extent_state *cached_state = NULL;
4227 u64 start = page_offset(page);
4228 u64 end = start + PAGE_CACHE_SIZE - 1;
4229 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
4231 start += ALIGN(offset, blocksize);
4235 lock_extent_bits(tree, start, end, &cached_state);
4236 wait_on_page_writeback(page);
4237 clear_extent_bit(tree, start, end,
4238 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4239 EXTENT_DO_ACCOUNTING,
4240 1, 1, &cached_state, GFP_NOFS);
4245 * a helper for releasepage, this tests for areas of the page that
4246 * are locked or under IO and drops the related state bits if it is safe
4249 static int try_release_extent_state(struct extent_map_tree *map,
4250 struct extent_io_tree *tree,
4251 struct page *page, gfp_t mask)
4253 u64 start = page_offset(page);
4254 u64 end = start + PAGE_CACHE_SIZE - 1;
4257 if (test_range_bit(tree, start, end,
4258 EXTENT_IOBITS, 0, NULL))
4261 if ((mask & GFP_NOFS) == GFP_NOFS)
4264 * at this point we can safely clear everything except the
4265 * locked bit and the nodatasum bit
4267 ret = clear_extent_bit(tree, start, end,
4268 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
4271 /* if clear_extent_bit failed for enomem reasons,
4272 * we can't allow the release to continue.
4283 * a helper for releasepage. As long as there are no locked extents
4284 * in the range corresponding to the page, both state records and extent
4285 * map records are removed
4287 int try_release_extent_mapping(struct extent_map_tree *map,
4288 struct extent_io_tree *tree, struct page *page,
4291 struct extent_map *em;
4292 u64 start = page_offset(page);
4293 u64 end = start + PAGE_CACHE_SIZE - 1;
4295 if (gfpflags_allow_blocking(mask) &&
4296 page->mapping->host->i_size > SZ_16M) {
4298 while (start <= end) {
4299 len = end - start + 1;
4300 write_lock(&map->lock);
4301 em = lookup_extent_mapping(map, start, len);
4303 write_unlock(&map->lock);
4306 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
4307 em->start != start) {
4308 write_unlock(&map->lock);
4309 free_extent_map(em);
4312 if (!test_range_bit(tree, em->start,
4313 extent_map_end(em) - 1,
4314 EXTENT_LOCKED | EXTENT_WRITEBACK,
4316 remove_extent_mapping(map, em);
4317 /* once for the rb tree */
4318 free_extent_map(em);
4320 start = extent_map_end(em);
4321 write_unlock(&map->lock);
4324 free_extent_map(em);
4327 return try_release_extent_state(map, tree, page, mask);
4331 * helper function for fiemap, which doesn't want to see any holes.
4332 * This maps until we find something past 'last'
4334 static struct extent_map *get_extent_skip_holes(struct inode *inode,
4337 get_extent_t *get_extent)
4339 u64 sectorsize = BTRFS_I(inode)->root->sectorsize;
4340 struct extent_map *em;
4347 len = last - offset;
4350 len = ALIGN(len, sectorsize);
4351 em = get_extent(inode, NULL, 0, offset, len, 0);
4352 if (IS_ERR_OR_NULL(em))
4355 /* if this isn't a hole return it */
4356 if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) &&
4357 em->block_start != EXTENT_MAP_HOLE) {
4361 /* this is a hole, advance to the next extent */
4362 offset = extent_map_end(em);
4363 free_extent_map(em);
4370 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4371 __u64 start, __u64 len, get_extent_t *get_extent)
4375 u64 max = start + len;
4379 u64 last_for_get_extent = 0;
4381 u64 isize = i_size_read(inode);
4382 struct btrfs_key found_key;
4383 struct extent_map *em = NULL;
4384 struct extent_state *cached_state = NULL;
4385 struct btrfs_path *path;
4386 struct btrfs_root *root = BTRFS_I(inode)->root;
4395 path = btrfs_alloc_path();
4398 path->leave_spinning = 1;
4400 start = round_down(start, BTRFS_I(inode)->root->sectorsize);
4401 len = round_up(max, BTRFS_I(inode)->root->sectorsize) - start;
4404 * lookup the last file extent. We're not using i_size here
4405 * because there might be preallocation past i_size
4407 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode), -1,
4410 btrfs_free_path(path);
4415 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
4416 found_type = found_key.type;
4418 /* No extents, but there might be delalloc bits */
4419 if (found_key.objectid != btrfs_ino(inode) ||
4420 found_type != BTRFS_EXTENT_DATA_KEY) {
4421 /* have to trust i_size as the end */
4423 last_for_get_extent = isize;
4426 * remember the start of the last extent. There are a
4427 * bunch of different factors that go into the length of the
4428 * extent, so its much less complex to remember where it started
4430 last = found_key.offset;
4431 last_for_get_extent = last + 1;
4433 btrfs_release_path(path);
4436 * we might have some extents allocated but more delalloc past those
4437 * extents. so, we trust isize unless the start of the last extent is
4442 last_for_get_extent = isize;
4445 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4448 em = get_extent_skip_holes(inode, start, last_for_get_extent,
4458 u64 offset_in_extent = 0;
4460 /* break if the extent we found is outside the range */
4461 if (em->start >= max || extent_map_end(em) < off)
4465 * get_extent may return an extent that starts before our
4466 * requested range. We have to make sure the ranges
4467 * we return to fiemap always move forward and don't
4468 * overlap, so adjust the offsets here
4470 em_start = max(em->start, off);
4473 * record the offset from the start of the extent
4474 * for adjusting the disk offset below. Only do this if the
4475 * extent isn't compressed since our in ram offset may be past
4476 * what we have actually allocated on disk.
4478 if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4479 offset_in_extent = em_start - em->start;
4480 em_end = extent_map_end(em);
4481 em_len = em_end - em_start;
4486 * bump off for our next call to get_extent
4488 off = extent_map_end(em);
4492 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
4494 flags |= FIEMAP_EXTENT_LAST;
4495 } else if (em->block_start == EXTENT_MAP_INLINE) {
4496 flags |= (FIEMAP_EXTENT_DATA_INLINE |
4497 FIEMAP_EXTENT_NOT_ALIGNED);
4498 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
4499 flags |= (FIEMAP_EXTENT_DELALLOC |
4500 FIEMAP_EXTENT_UNKNOWN);
4501 } else if (fieinfo->fi_extents_max) {
4502 u64 bytenr = em->block_start -
4503 (em->start - em->orig_start);
4505 disko = em->block_start + offset_in_extent;
4508 * As btrfs supports shared space, this information
4509 * can be exported to userspace tools via
4510 * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0
4511 * then we're just getting a count and we can skip the
4514 ret = btrfs_check_shared(NULL, root->fs_info,
4516 btrfs_ino(inode), bytenr);
4520 flags |= FIEMAP_EXTENT_SHARED;
4523 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4524 flags |= FIEMAP_EXTENT_ENCODED;
4525 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4526 flags |= FIEMAP_EXTENT_UNWRITTEN;
4528 free_extent_map(em);
4530 if ((em_start >= last) || em_len == (u64)-1 ||
4531 (last == (u64)-1 && isize <= em_end)) {
4532 flags |= FIEMAP_EXTENT_LAST;
4536 /* now scan forward to see if this is really the last extent. */
4537 em = get_extent_skip_holes(inode, off, last_for_get_extent,
4544 flags |= FIEMAP_EXTENT_LAST;
4547 ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
4556 free_extent_map(em);
4558 btrfs_free_path(path);
4559 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4560 &cached_state, GFP_NOFS);
4564 static void __free_extent_buffer(struct extent_buffer *eb)
4566 btrfs_leak_debug_del(&eb->leak_list);
4567 kmem_cache_free(extent_buffer_cache, eb);
4570 int extent_buffer_under_io(struct extent_buffer *eb)
4572 return (atomic_read(&eb->io_pages) ||
4573 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4574 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4578 * Helper for releasing extent buffer page.
4580 static void btrfs_release_extent_buffer_page(struct extent_buffer *eb)
4582 unsigned long index;
4584 int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4586 BUG_ON(extent_buffer_under_io(eb));
4588 index = num_extent_pages(eb->start, eb->len);
4594 page = eb->pages[index];
4598 spin_lock(&page->mapping->private_lock);
4600 * We do this since we'll remove the pages after we've
4601 * removed the eb from the radix tree, so we could race
4602 * and have this page now attached to the new eb. So
4603 * only clear page_private if it's still connected to
4606 if (PagePrivate(page) &&
4607 page->private == (unsigned long)eb) {
4608 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4609 BUG_ON(PageDirty(page));
4610 BUG_ON(PageWriteback(page));
4612 * We need to make sure we haven't be attached
4615 ClearPagePrivate(page);
4616 set_page_private(page, 0);
4617 /* One for the page private */
4618 page_cache_release(page);
4622 spin_unlock(&page->mapping->private_lock);
4624 /* One for when we alloced the page */
4625 page_cache_release(page);
4626 } while (index != 0);
4630 * Helper for releasing the extent buffer.
4632 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4634 btrfs_release_extent_buffer_page(eb);
4635 __free_extent_buffer(eb);
4638 static struct extent_buffer *
4639 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
4642 struct extent_buffer *eb = NULL;
4644 eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL);
4647 eb->fs_info = fs_info;
4649 rwlock_init(&eb->lock);
4650 atomic_set(&eb->write_locks, 0);
4651 atomic_set(&eb->read_locks, 0);
4652 atomic_set(&eb->blocking_readers, 0);
4653 atomic_set(&eb->blocking_writers, 0);
4654 atomic_set(&eb->spinning_readers, 0);
4655 atomic_set(&eb->spinning_writers, 0);
4656 eb->lock_nested = 0;
4657 init_waitqueue_head(&eb->write_lock_wq);
4658 init_waitqueue_head(&eb->read_lock_wq);
4660 btrfs_leak_debug_add(&eb->leak_list, &buffers);
4662 spin_lock_init(&eb->refs_lock);
4663 atomic_set(&eb->refs, 1);
4664 atomic_set(&eb->io_pages, 0);
4667 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4669 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4670 > MAX_INLINE_EXTENT_BUFFER_SIZE);
4671 BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE);
4676 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src)
4680 struct extent_buffer *new;
4681 unsigned long num_pages = num_extent_pages(src->start, src->len);
4683 new = __alloc_extent_buffer(src->fs_info, src->start, src->len);
4687 for (i = 0; i < num_pages; i++) {
4688 p = alloc_page(GFP_NOFS);
4690 btrfs_release_extent_buffer(new);
4693 attach_extent_buffer_page(new, p);
4694 WARN_ON(PageDirty(p));
4699 copy_extent_buffer(new, src, 0, 0, src->len);
4700 set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
4701 set_bit(EXTENT_BUFFER_DUMMY, &new->bflags);
4706 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4707 u64 start, unsigned long len)
4709 struct extent_buffer *eb;
4710 unsigned long num_pages;
4713 num_pages = num_extent_pages(start, len);
4715 eb = __alloc_extent_buffer(fs_info, start, len);
4719 for (i = 0; i < num_pages; i++) {
4720 eb->pages[i] = alloc_page(GFP_NOFS);
4724 set_extent_buffer_uptodate(eb);
4725 btrfs_set_header_nritems(eb, 0);
4726 set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4731 __free_page(eb->pages[i - 1]);
4732 __free_extent_buffer(eb);
4736 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4743 * Called only from tests that don't always have a fs_info
4744 * available, but we know that nodesize is 4096
4748 len = fs_info->tree_root->nodesize;
4751 return __alloc_dummy_extent_buffer(fs_info, start, len);
4754 static void check_buffer_tree_ref(struct extent_buffer *eb)
4757 /* the ref bit is tricky. We have to make sure it is set
4758 * if we have the buffer dirty. Otherwise the
4759 * code to free a buffer can end up dropping a dirty
4762 * Once the ref bit is set, it won't go away while the
4763 * buffer is dirty or in writeback, and it also won't
4764 * go away while we have the reference count on the
4767 * We can't just set the ref bit without bumping the
4768 * ref on the eb because free_extent_buffer might
4769 * see the ref bit and try to clear it. If this happens
4770 * free_extent_buffer might end up dropping our original
4771 * ref by mistake and freeing the page before we are able
4772 * to add one more ref.
4774 * So bump the ref count first, then set the bit. If someone
4775 * beat us to it, drop the ref we added.
4777 refs = atomic_read(&eb->refs);
4778 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4781 spin_lock(&eb->refs_lock);
4782 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4783 atomic_inc(&eb->refs);
4784 spin_unlock(&eb->refs_lock);
4787 static void mark_extent_buffer_accessed(struct extent_buffer *eb,
4788 struct page *accessed)
4790 unsigned long num_pages, i;
4792 check_buffer_tree_ref(eb);
4794 num_pages = num_extent_pages(eb->start, eb->len);
4795 for (i = 0; i < num_pages; i++) {
4796 struct page *p = eb->pages[i];
4799 mark_page_accessed(p);
4803 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
4806 struct extent_buffer *eb;
4809 eb = radix_tree_lookup(&fs_info->buffer_radix,
4810 start >> PAGE_CACHE_SHIFT);
4811 if (eb && atomic_inc_not_zero(&eb->refs)) {
4814 * Lock our eb's refs_lock to avoid races with
4815 * free_extent_buffer. When we get our eb it might be flagged
4816 * with EXTENT_BUFFER_STALE and another task running
4817 * free_extent_buffer might have seen that flag set,
4818 * eb->refs == 2, that the buffer isn't under IO (dirty and
4819 * writeback flags not set) and it's still in the tree (flag
4820 * EXTENT_BUFFER_TREE_REF set), therefore being in the process
4821 * of decrementing the extent buffer's reference count twice.
4822 * So here we could race and increment the eb's reference count,
4823 * clear its stale flag, mark it as dirty and drop our reference
4824 * before the other task finishes executing free_extent_buffer,
4825 * which would later result in an attempt to free an extent
4826 * buffer that is dirty.
4828 if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) {
4829 spin_lock(&eb->refs_lock);
4830 spin_unlock(&eb->refs_lock);
4832 mark_extent_buffer_accessed(eb, NULL);
4840 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4841 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
4844 struct extent_buffer *eb, *exists = NULL;
4847 eb = find_extent_buffer(fs_info, start);
4850 eb = alloc_dummy_extent_buffer(fs_info, start);
4853 eb->fs_info = fs_info;
4855 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
4858 spin_lock(&fs_info->buffer_lock);
4859 ret = radix_tree_insert(&fs_info->buffer_radix,
4860 start >> PAGE_CACHE_SHIFT, eb);
4861 spin_unlock(&fs_info->buffer_lock);
4862 radix_tree_preload_end();
4863 if (ret == -EEXIST) {
4864 exists = find_extent_buffer(fs_info, start);
4870 check_buffer_tree_ref(eb);
4871 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
4874 * We will free dummy extent buffer's if they come into
4875 * free_extent_buffer with a ref count of 2, but if we are using this we
4876 * want the buffers to stay in memory until we're done with them, so
4877 * bump the ref count again.
4879 atomic_inc(&eb->refs);
4882 btrfs_release_extent_buffer(eb);
4887 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
4890 unsigned long len = fs_info->tree_root->nodesize;
4891 unsigned long num_pages = num_extent_pages(start, len);
4893 unsigned long index = start >> PAGE_CACHE_SHIFT;
4894 struct extent_buffer *eb;
4895 struct extent_buffer *exists = NULL;
4897 struct address_space *mapping = fs_info->btree_inode->i_mapping;
4901 eb = find_extent_buffer(fs_info, start);
4905 eb = __alloc_extent_buffer(fs_info, start, len);
4909 for (i = 0; i < num_pages; i++, index++) {
4910 p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL);
4914 spin_lock(&mapping->private_lock);
4915 if (PagePrivate(p)) {
4917 * We could have already allocated an eb for this page
4918 * and attached one so lets see if we can get a ref on
4919 * the existing eb, and if we can we know it's good and
4920 * we can just return that one, else we know we can just
4921 * overwrite page->private.
4923 exists = (struct extent_buffer *)p->private;
4924 if (atomic_inc_not_zero(&exists->refs)) {
4925 spin_unlock(&mapping->private_lock);
4927 page_cache_release(p);
4928 mark_extent_buffer_accessed(exists, p);
4934 * Do this so attach doesn't complain and we need to
4935 * drop the ref the old guy had.
4937 ClearPagePrivate(p);
4938 WARN_ON(PageDirty(p));
4939 page_cache_release(p);
4941 attach_extent_buffer_page(eb, p);
4942 spin_unlock(&mapping->private_lock);
4943 WARN_ON(PageDirty(p));
4945 if (!PageUptodate(p))
4949 * see below about how we avoid a nasty race with release page
4950 * and why we unlock later
4954 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4956 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
4960 spin_lock(&fs_info->buffer_lock);
4961 ret = radix_tree_insert(&fs_info->buffer_radix,
4962 start >> PAGE_CACHE_SHIFT, eb);
4963 spin_unlock(&fs_info->buffer_lock);
4964 radix_tree_preload_end();
4965 if (ret == -EEXIST) {
4966 exists = find_extent_buffer(fs_info, start);
4972 /* add one reference for the tree */
4973 check_buffer_tree_ref(eb);
4974 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
4977 * there is a race where release page may have
4978 * tried to find this extent buffer in the radix
4979 * but failed. It will tell the VM it is safe to
4980 * reclaim the, and it will clear the page private bit.
4981 * We must make sure to set the page private bit properly
4982 * after the extent buffer is in the radix tree so
4983 * it doesn't get lost
4985 SetPageChecked(eb->pages[0]);
4986 for (i = 1; i < num_pages; i++) {
4988 ClearPageChecked(p);
4991 unlock_page(eb->pages[0]);
4995 WARN_ON(!atomic_dec_and_test(&eb->refs));
4996 for (i = 0; i < num_pages; i++) {
4998 unlock_page(eb->pages[i]);
5001 btrfs_release_extent_buffer(eb);
5005 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
5007 struct extent_buffer *eb =
5008 container_of(head, struct extent_buffer, rcu_head);
5010 __free_extent_buffer(eb);
5013 /* Expects to have eb->eb_lock already held */
5014 static int release_extent_buffer(struct extent_buffer *eb)
5016 WARN_ON(atomic_read(&eb->refs) == 0);
5017 if (atomic_dec_and_test(&eb->refs)) {
5018 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
5019 struct btrfs_fs_info *fs_info = eb->fs_info;
5021 spin_unlock(&eb->refs_lock);
5023 spin_lock(&fs_info->buffer_lock);
5024 radix_tree_delete(&fs_info->buffer_radix,
5025 eb->start >> PAGE_CACHE_SHIFT);
5026 spin_unlock(&fs_info->buffer_lock);
5028 spin_unlock(&eb->refs_lock);
5031 /* Should be safe to release our pages at this point */
5032 btrfs_release_extent_buffer_page(eb);
5033 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5034 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))) {
5035 __free_extent_buffer(eb);
5039 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
5042 spin_unlock(&eb->refs_lock);
5047 void free_extent_buffer(struct extent_buffer *eb)
5055 refs = atomic_read(&eb->refs);
5058 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
5063 spin_lock(&eb->refs_lock);
5064 if (atomic_read(&eb->refs) == 2 &&
5065 test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))
5066 atomic_dec(&eb->refs);
5068 if (atomic_read(&eb->refs) == 2 &&
5069 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
5070 !extent_buffer_under_io(eb) &&
5071 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5072 atomic_dec(&eb->refs);
5075 * I know this is terrible, but it's temporary until we stop tracking
5076 * the uptodate bits and such for the extent buffers.
5078 release_extent_buffer(eb);
5081 void free_extent_buffer_stale(struct extent_buffer *eb)
5086 spin_lock(&eb->refs_lock);
5087 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
5089 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
5090 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5091 atomic_dec(&eb->refs);
5092 release_extent_buffer(eb);
5095 void clear_extent_buffer_dirty(struct extent_buffer *eb)
5098 unsigned long num_pages;
5101 num_pages = num_extent_pages(eb->start, eb->len);
5103 for (i = 0; i < num_pages; i++) {
5104 page = eb->pages[i];
5105 if (!PageDirty(page))
5109 WARN_ON(!PagePrivate(page));
5111 clear_page_dirty_for_io(page);
5112 spin_lock_irq(&page->mapping->tree_lock);
5113 if (!PageDirty(page)) {
5114 radix_tree_tag_clear(&page->mapping->page_tree,
5116 PAGECACHE_TAG_DIRTY);
5118 spin_unlock_irq(&page->mapping->tree_lock);
5119 ClearPageError(page);
5122 WARN_ON(atomic_read(&eb->refs) == 0);
5125 int set_extent_buffer_dirty(struct extent_buffer *eb)
5128 unsigned long num_pages;
5131 check_buffer_tree_ref(eb);
5133 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
5135 num_pages = num_extent_pages(eb->start, eb->len);
5136 WARN_ON(atomic_read(&eb->refs) == 0);
5137 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
5139 for (i = 0; i < num_pages; i++)
5140 set_page_dirty(eb->pages[i]);
5144 void clear_extent_buffer_uptodate(struct extent_buffer *eb)
5148 unsigned long num_pages;
5150 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5151 num_pages = num_extent_pages(eb->start, eb->len);
5152 for (i = 0; i < num_pages; i++) {
5153 page = eb->pages[i];
5155 ClearPageUptodate(page);
5159 void set_extent_buffer_uptodate(struct extent_buffer *eb)
5163 unsigned long num_pages;
5165 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5166 num_pages = num_extent_pages(eb->start, eb->len);
5167 for (i = 0; i < num_pages; i++) {
5168 page = eb->pages[i];
5169 SetPageUptodate(page);
5173 int extent_buffer_uptodate(struct extent_buffer *eb)
5175 return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5178 int read_extent_buffer_pages(struct extent_io_tree *tree,
5179 struct extent_buffer *eb, u64 start, int wait,
5180 get_extent_t *get_extent, int mirror_num)
5183 unsigned long start_i;
5187 int locked_pages = 0;
5188 int all_uptodate = 1;
5189 unsigned long num_pages;
5190 unsigned long num_reads = 0;
5191 struct bio *bio = NULL;
5192 unsigned long bio_flags = 0;
5194 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
5198 WARN_ON(start < eb->start);
5199 start_i = (start >> PAGE_CACHE_SHIFT) -
5200 (eb->start >> PAGE_CACHE_SHIFT);
5205 num_pages = num_extent_pages(eb->start, eb->len);
5206 for (i = start_i; i < num_pages; i++) {
5207 page = eb->pages[i];
5208 if (wait == WAIT_NONE) {
5209 if (!trylock_page(page))
5215 if (!PageUptodate(page)) {
5222 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5226 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
5227 eb->read_mirror = 0;
5228 atomic_set(&eb->io_pages, num_reads);
5229 for (i = start_i; i < num_pages; i++) {
5230 page = eb->pages[i];
5231 if (!PageUptodate(page)) {
5232 ClearPageError(page);
5233 err = __extent_read_full_page(tree, page,
5235 mirror_num, &bio_flags,
5245 err = submit_one_bio(READ | REQ_META, bio, mirror_num,
5251 if (ret || wait != WAIT_COMPLETE)
5254 for (i = start_i; i < num_pages; i++) {
5255 page = eb->pages[i];
5256 wait_on_page_locked(page);
5257 if (!PageUptodate(page))
5265 while (locked_pages > 0) {
5266 page = eb->pages[i];
5274 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
5275 unsigned long start,
5282 char *dst = (char *)dstv;
5283 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5284 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5286 WARN_ON(start > eb->len);
5287 WARN_ON(start + len > eb->start + eb->len);
5289 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5292 page = eb->pages[i];
5294 cur = min(len, (PAGE_CACHE_SIZE - offset));
5295 kaddr = page_address(page);
5296 memcpy(dst, kaddr + offset, cur);
5305 int read_extent_buffer_to_user(struct extent_buffer *eb, void __user *dstv,
5306 unsigned long start,
5313 char __user *dst = (char __user *)dstv;
5314 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5315 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5318 WARN_ON(start > eb->len);
5319 WARN_ON(start + len > eb->start + eb->len);
5321 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5324 page = eb->pages[i];
5326 cur = min(len, (PAGE_CACHE_SIZE - offset));
5327 kaddr = page_address(page);
5328 if (copy_to_user(dst, kaddr + offset, cur)) {
5342 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
5343 unsigned long min_len, char **map,
5344 unsigned long *map_start,
5345 unsigned long *map_len)
5347 size_t offset = start & (PAGE_CACHE_SIZE - 1);
5350 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5351 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5352 unsigned long end_i = (start_offset + start + min_len - 1) >>
5359 offset = start_offset;
5363 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
5366 if (start + min_len > eb->len) {
5367 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, "
5369 eb->start, eb->len, start, min_len);
5374 kaddr = page_address(p);
5375 *map = kaddr + offset;
5376 *map_len = PAGE_CACHE_SIZE - offset;
5380 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
5381 unsigned long start,
5388 char *ptr = (char *)ptrv;
5389 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5390 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5393 WARN_ON(start > eb->len);
5394 WARN_ON(start + len > eb->start + eb->len);
5396 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5399 page = eb->pages[i];
5401 cur = min(len, (PAGE_CACHE_SIZE - offset));
5403 kaddr = page_address(page);
5404 ret = memcmp(ptr, kaddr + offset, cur);
5416 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
5417 unsigned long start, unsigned long len)
5423 char *src = (char *)srcv;
5424 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5425 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5427 WARN_ON(start > eb->len);
5428 WARN_ON(start + len > eb->start + eb->len);
5430 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5433 page = eb->pages[i];
5434 WARN_ON(!PageUptodate(page));
5436 cur = min(len, PAGE_CACHE_SIZE - offset);
5437 kaddr = page_address(page);
5438 memcpy(kaddr + offset, src, cur);
5447 void memset_extent_buffer(struct extent_buffer *eb, char c,
5448 unsigned long start, unsigned long len)
5454 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5455 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5457 WARN_ON(start > eb->len);
5458 WARN_ON(start + len > eb->start + eb->len);
5460 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5463 page = eb->pages[i];
5464 WARN_ON(!PageUptodate(page));
5466 cur = min(len, PAGE_CACHE_SIZE - offset);
5467 kaddr = page_address(page);
5468 memset(kaddr + offset, c, cur);
5476 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
5477 unsigned long dst_offset, unsigned long src_offset,
5480 u64 dst_len = dst->len;
5485 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5486 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
5488 WARN_ON(src->len != dst_len);
5490 offset = (start_offset + dst_offset) &
5491 (PAGE_CACHE_SIZE - 1);
5494 page = dst->pages[i];
5495 WARN_ON(!PageUptodate(page));
5497 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
5499 kaddr = page_address(page);
5500 read_extent_buffer(src, kaddr + offset, src_offset, cur);
5510 * The extent buffer bitmap operations are done with byte granularity because
5511 * bitmap items are not guaranteed to be aligned to a word and therefore a
5512 * single word in a bitmap may straddle two pages in the extent buffer.
5514 #define BIT_BYTE(nr) ((nr) / BITS_PER_BYTE)
5515 #define BYTE_MASK ((1 << BITS_PER_BYTE) - 1)
5516 #define BITMAP_FIRST_BYTE_MASK(start) \
5517 ((BYTE_MASK << ((start) & (BITS_PER_BYTE - 1))) & BYTE_MASK)
5518 #define BITMAP_LAST_BYTE_MASK(nbits) \
5519 (BYTE_MASK >> (-(nbits) & (BITS_PER_BYTE - 1)))
5522 * eb_bitmap_offset() - calculate the page and offset of the byte containing the
5524 * @eb: the extent buffer
5525 * @start: offset of the bitmap item in the extent buffer
5527 * @page_index: return index of the page in the extent buffer that contains the
5529 * @page_offset: return offset into the page given by page_index
5531 * This helper hides the ugliness of finding the byte in an extent buffer which
5532 * contains a given bit.
5534 static inline void eb_bitmap_offset(struct extent_buffer *eb,
5535 unsigned long start, unsigned long nr,
5536 unsigned long *page_index,
5537 size_t *page_offset)
5539 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5540 size_t byte_offset = BIT_BYTE(nr);
5544 * The byte we want is the offset of the extent buffer + the offset of
5545 * the bitmap item in the extent buffer + the offset of the byte in the
5548 offset = start_offset + start + byte_offset;
5550 *page_index = offset >> PAGE_CACHE_SHIFT;
5551 *page_offset = offset & (PAGE_CACHE_SIZE - 1);
5555 * extent_buffer_test_bit - determine whether a bit in a bitmap item is set
5556 * @eb: the extent buffer
5557 * @start: offset of the bitmap item in the extent buffer
5558 * @nr: bit number to test
5560 int extent_buffer_test_bit(struct extent_buffer *eb, unsigned long start,
5568 eb_bitmap_offset(eb, start, nr, &i, &offset);
5569 page = eb->pages[i];
5570 WARN_ON(!PageUptodate(page));
5571 kaddr = page_address(page);
5572 return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1)));
5576 * extent_buffer_bitmap_set - set an area of a bitmap
5577 * @eb: the extent buffer
5578 * @start: offset of the bitmap item in the extent buffer
5579 * @pos: bit number of the first bit
5580 * @len: number of bits to set
5582 void extent_buffer_bitmap_set(struct extent_buffer *eb, unsigned long start,
5583 unsigned long pos, unsigned long len)
5589 const unsigned int size = pos + len;
5590 int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5591 unsigned int mask_to_set = BITMAP_FIRST_BYTE_MASK(pos);
5593 eb_bitmap_offset(eb, start, pos, &i, &offset);
5594 page = eb->pages[i];
5595 WARN_ON(!PageUptodate(page));
5596 kaddr = page_address(page);
5598 while (len >= bits_to_set) {
5599 kaddr[offset] |= mask_to_set;
5601 bits_to_set = BITS_PER_BYTE;
5603 if (++offset >= PAGE_CACHE_SIZE && len > 0) {
5605 page = eb->pages[++i];
5606 WARN_ON(!PageUptodate(page));
5607 kaddr = page_address(page);
5611 mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
5612 kaddr[offset] |= mask_to_set;
5618 * extent_buffer_bitmap_clear - clear an area of a bitmap
5619 * @eb: the extent buffer
5620 * @start: offset of the bitmap item in the extent buffer
5621 * @pos: bit number of the first bit
5622 * @len: number of bits to clear
5624 void extent_buffer_bitmap_clear(struct extent_buffer *eb, unsigned long start,
5625 unsigned long pos, unsigned long len)
5631 const unsigned int size = pos + len;
5632 int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5633 unsigned int mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos);
5635 eb_bitmap_offset(eb, start, pos, &i, &offset);
5636 page = eb->pages[i];
5637 WARN_ON(!PageUptodate(page));
5638 kaddr = page_address(page);
5640 while (len >= bits_to_clear) {
5641 kaddr[offset] &= ~mask_to_clear;
5642 len -= bits_to_clear;
5643 bits_to_clear = BITS_PER_BYTE;
5644 mask_to_clear = ~0U;
5645 if (++offset >= PAGE_CACHE_SIZE && len > 0) {
5647 page = eb->pages[++i];
5648 WARN_ON(!PageUptodate(page));
5649 kaddr = page_address(page);
5653 mask_to_clear &= BITMAP_LAST_BYTE_MASK(size);
5654 kaddr[offset] &= ~mask_to_clear;
5658 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
5660 unsigned long distance = (src > dst) ? src - dst : dst - src;
5661 return distance < len;
5664 static void copy_pages(struct page *dst_page, struct page *src_page,
5665 unsigned long dst_off, unsigned long src_off,
5668 char *dst_kaddr = page_address(dst_page);
5670 int must_memmove = 0;
5672 if (dst_page != src_page) {
5673 src_kaddr = page_address(src_page);
5675 src_kaddr = dst_kaddr;
5676 if (areas_overlap(src_off, dst_off, len))
5681 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
5683 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
5686 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5687 unsigned long src_offset, unsigned long len)
5690 size_t dst_off_in_page;
5691 size_t src_off_in_page;
5692 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5693 unsigned long dst_i;
5694 unsigned long src_i;
5696 if (src_offset + len > dst->len) {
5697 btrfs_err(dst->fs_info,
5698 "memmove bogus src_offset %lu move "
5699 "len %lu dst len %lu", src_offset, len, dst->len);
5702 if (dst_offset + len > dst->len) {
5703 btrfs_err(dst->fs_info,
5704 "memmove bogus dst_offset %lu move "
5705 "len %lu dst len %lu", dst_offset, len, dst->len);
5710 dst_off_in_page = (start_offset + dst_offset) &
5711 (PAGE_CACHE_SIZE - 1);
5712 src_off_in_page = (start_offset + src_offset) &
5713 (PAGE_CACHE_SIZE - 1);
5715 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
5716 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
5718 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
5720 cur = min_t(unsigned long, cur,
5721 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
5723 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5724 dst_off_in_page, src_off_in_page, cur);
5732 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5733 unsigned long src_offset, unsigned long len)
5736 size_t dst_off_in_page;
5737 size_t src_off_in_page;
5738 unsigned long dst_end = dst_offset + len - 1;
5739 unsigned long src_end = src_offset + len - 1;
5740 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5741 unsigned long dst_i;
5742 unsigned long src_i;
5744 if (src_offset + len > dst->len) {
5745 btrfs_err(dst->fs_info, "memmove bogus src_offset %lu move "
5746 "len %lu len %lu", src_offset, len, dst->len);
5749 if (dst_offset + len > dst->len) {
5750 btrfs_err(dst->fs_info, "memmove bogus dst_offset %lu move "
5751 "len %lu len %lu", dst_offset, len, dst->len);
5754 if (dst_offset < src_offset) {
5755 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
5759 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
5760 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
5762 dst_off_in_page = (start_offset + dst_end) &
5763 (PAGE_CACHE_SIZE - 1);
5764 src_off_in_page = (start_offset + src_end) &
5765 (PAGE_CACHE_SIZE - 1);
5767 cur = min_t(unsigned long, len, src_off_in_page + 1);
5768 cur = min(cur, dst_off_in_page + 1);
5769 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5770 dst_off_in_page - cur + 1,
5771 src_off_in_page - cur + 1, cur);
5779 int try_release_extent_buffer(struct page *page)
5781 struct extent_buffer *eb;
5784 * We need to make sure noboody is attaching this page to an eb right
5787 spin_lock(&page->mapping->private_lock);
5788 if (!PagePrivate(page)) {
5789 spin_unlock(&page->mapping->private_lock);
5793 eb = (struct extent_buffer *)page->private;
5797 * This is a little awful but should be ok, we need to make sure that
5798 * the eb doesn't disappear out from under us while we're looking at
5801 spin_lock(&eb->refs_lock);
5802 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
5803 spin_unlock(&eb->refs_lock);
5804 spin_unlock(&page->mapping->private_lock);
5807 spin_unlock(&page->mapping->private_lock);
5810 * If tree ref isn't set then we know the ref on this eb is a real ref,
5811 * so just return, this page will likely be freed soon anyway.
5813 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
5814 spin_unlock(&eb->refs_lock);
5818 return release_extent_buffer(eb);