1 // SPDX-License-Identifier: GPL-2.0
3 #include <linux/bitops.h>
4 #include <linux/slab.h>
7 #include <linux/pagemap.h>
8 #include <linux/page-flags.h>
9 #include <linux/spinlock.h>
10 #include <linux/blkdev.h>
11 #include <linux/swap.h>
12 #include <linux/writeback.h>
13 #include <linux/pagevec.h>
14 #include <linux/prefetch.h>
15 #include <linux/cleancache.h>
16 #include "extent_io.h"
17 #include "extent_map.h"
19 #include "btrfs_inode.h"
21 #include "check-integrity.h"
23 #include "rcu-string.h"
27 static struct kmem_cache *extent_state_cache;
28 static struct kmem_cache *extent_buffer_cache;
29 static struct bio_set btrfs_bioset;
31 static inline bool extent_state_in_tree(const struct extent_state *state)
33 return !RB_EMPTY_NODE(&state->rb_node);
36 #ifdef CONFIG_BTRFS_DEBUG
37 static LIST_HEAD(buffers);
38 static LIST_HEAD(states);
40 static DEFINE_SPINLOCK(leak_lock);
43 void btrfs_leak_debug_add(struct list_head *new, struct list_head *head)
47 spin_lock_irqsave(&leak_lock, flags);
49 spin_unlock_irqrestore(&leak_lock, flags);
53 void btrfs_leak_debug_del(struct list_head *entry)
57 spin_lock_irqsave(&leak_lock, flags);
59 spin_unlock_irqrestore(&leak_lock, flags);
63 void btrfs_leak_debug_check(void)
65 struct extent_state *state;
66 struct extent_buffer *eb;
68 while (!list_empty(&states)) {
69 state = list_entry(states.next, struct extent_state, leak_list);
70 pr_err("BTRFS: state leak: start %llu end %llu state %u in tree %d refs %d\n",
71 state->start, state->end, state->state,
72 extent_state_in_tree(state),
73 refcount_read(&state->refs));
74 list_del(&state->leak_list);
75 kmem_cache_free(extent_state_cache, state);
78 while (!list_empty(&buffers)) {
79 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
80 pr_err("BTRFS: buffer leak start %llu len %lu refs %d bflags %lu\n",
81 eb->start, eb->len, atomic_read(&eb->refs), eb->bflags);
82 list_del(&eb->leak_list);
83 kmem_cache_free(extent_buffer_cache, eb);
87 #define btrfs_debug_check_extent_io_range(tree, start, end) \
88 __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end))
89 static inline void __btrfs_debug_check_extent_io_range(const char *caller,
90 struct extent_io_tree *tree, u64 start, u64 end)
92 struct inode *inode = tree->private_data;
95 if (!inode || !is_data_inode(inode))
98 isize = i_size_read(inode);
99 if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) {
100 btrfs_debug_rl(BTRFS_I(inode)->root->fs_info,
101 "%s: ino %llu isize %llu odd range [%llu,%llu]",
102 caller, btrfs_ino(BTRFS_I(inode)), isize, start, end);
106 #define btrfs_leak_debug_add(new, head) do {} while (0)
107 #define btrfs_leak_debug_del(entry) do {} while (0)
108 #define btrfs_leak_debug_check() do {} while (0)
109 #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0)
112 #define BUFFER_LRU_MAX 64
117 struct rb_node rb_node;
120 struct extent_page_data {
122 struct extent_io_tree *tree;
123 /* tells writepage not to lock the state bits for this range
124 * it still does the unlocking
126 unsigned int extent_locked:1;
128 /* tells the submit_bio code to use REQ_SYNC */
129 unsigned int sync_io:1;
132 static int add_extent_changeset(struct extent_state *state, unsigned bits,
133 struct extent_changeset *changeset,
140 if (set && (state->state & bits) == bits)
142 if (!set && (state->state & bits) == 0)
144 changeset->bytes_changed += state->end - state->start + 1;
145 ret = ulist_add(&changeset->range_changed, state->start, state->end,
150 static int __must_check submit_one_bio(struct bio *bio, int mirror_num,
151 unsigned long bio_flags)
153 blk_status_t ret = 0;
154 struct bio_vec *bvec = bio_last_bvec_all(bio);
156 struct extent_io_tree *tree = bio->bi_private;
159 mp_bvec_last_segment(bvec, &bv);
160 start = page_offset(bv.bv_page) + bv.bv_offset;
162 bio->bi_private = NULL;
165 ret = tree->ops->submit_bio_hook(tree->private_data, bio,
166 mirror_num, bio_flags, start);
168 btrfsic_submit_bio(bio);
170 return blk_status_to_errno(ret);
173 static void flush_write_bio(struct extent_page_data *epd)
178 ret = submit_one_bio(epd->bio, 0, 0);
179 BUG_ON(ret < 0); /* -ENOMEM */
184 int __init extent_io_init(void)
186 extent_state_cache = kmem_cache_create("btrfs_extent_state",
187 sizeof(struct extent_state), 0,
188 SLAB_MEM_SPREAD, NULL);
189 if (!extent_state_cache)
192 extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer",
193 sizeof(struct extent_buffer), 0,
194 SLAB_MEM_SPREAD, NULL);
195 if (!extent_buffer_cache)
196 goto free_state_cache;
198 if (bioset_init(&btrfs_bioset, BIO_POOL_SIZE,
199 offsetof(struct btrfs_io_bio, bio),
201 goto free_buffer_cache;
203 if (bioset_integrity_create(&btrfs_bioset, BIO_POOL_SIZE))
209 bioset_exit(&btrfs_bioset);
212 kmem_cache_destroy(extent_buffer_cache);
213 extent_buffer_cache = NULL;
216 kmem_cache_destroy(extent_state_cache);
217 extent_state_cache = NULL;
221 void __cold extent_io_exit(void)
223 btrfs_leak_debug_check();
226 * Make sure all delayed rcu free are flushed before we
230 kmem_cache_destroy(extent_state_cache);
231 kmem_cache_destroy(extent_buffer_cache);
232 bioset_exit(&btrfs_bioset);
235 void extent_io_tree_init(struct btrfs_fs_info *fs_info,
236 struct extent_io_tree *tree, unsigned int owner,
239 tree->fs_info = fs_info;
240 tree->state = RB_ROOT;
242 tree->dirty_bytes = 0;
243 spin_lock_init(&tree->lock);
244 tree->private_data = private_data;
248 static struct extent_state *alloc_extent_state(gfp_t mask)
250 struct extent_state *state;
253 * The given mask might be not appropriate for the slab allocator,
254 * drop the unsupported bits
256 mask &= ~(__GFP_DMA32|__GFP_HIGHMEM);
257 state = kmem_cache_alloc(extent_state_cache, mask);
261 state->failrec = NULL;
262 RB_CLEAR_NODE(&state->rb_node);
263 btrfs_leak_debug_add(&state->leak_list, &states);
264 refcount_set(&state->refs, 1);
265 init_waitqueue_head(&state->wq);
266 trace_alloc_extent_state(state, mask, _RET_IP_);
270 void free_extent_state(struct extent_state *state)
274 if (refcount_dec_and_test(&state->refs)) {
275 WARN_ON(extent_state_in_tree(state));
276 btrfs_leak_debug_del(&state->leak_list);
277 trace_free_extent_state(state, _RET_IP_);
278 kmem_cache_free(extent_state_cache, state);
282 static struct rb_node *tree_insert(struct rb_root *root,
283 struct rb_node *search_start,
285 struct rb_node *node,
286 struct rb_node ***p_in,
287 struct rb_node **parent_in)
290 struct rb_node *parent = NULL;
291 struct tree_entry *entry;
293 if (p_in && parent_in) {
299 p = search_start ? &search_start : &root->rb_node;
302 entry = rb_entry(parent, struct tree_entry, rb_node);
304 if (offset < entry->start)
306 else if (offset > entry->end)
313 rb_link_node(node, parent, p);
314 rb_insert_color(node, root);
318 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
319 struct rb_node **next_ret,
320 struct rb_node **prev_ret,
321 struct rb_node ***p_ret,
322 struct rb_node **parent_ret)
324 struct rb_root *root = &tree->state;
325 struct rb_node **n = &root->rb_node;
326 struct rb_node *prev = NULL;
327 struct rb_node *orig_prev = NULL;
328 struct tree_entry *entry;
329 struct tree_entry *prev_entry = NULL;
333 entry = rb_entry(prev, struct tree_entry, rb_node);
336 if (offset < entry->start)
338 else if (offset > entry->end)
351 while (prev && offset > prev_entry->end) {
352 prev = rb_next(prev);
353 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
360 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
361 while (prev && offset < prev_entry->start) {
362 prev = rb_prev(prev);
363 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
370 static inline struct rb_node *
371 tree_search_for_insert(struct extent_io_tree *tree,
373 struct rb_node ***p_ret,
374 struct rb_node **parent_ret)
376 struct rb_node *next= NULL;
379 ret = __etree_search(tree, offset, &next, NULL, p_ret, parent_ret);
385 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
388 return tree_search_for_insert(tree, offset, NULL, NULL);
392 * utility function to look for merge candidates inside a given range.
393 * Any extents with matching state are merged together into a single
394 * extent in the tree. Extents with EXTENT_IO in their state field
395 * are not merged because the end_io handlers need to be able to do
396 * operations on them without sleeping (or doing allocations/splits).
398 * This should be called with the tree lock held.
400 static void merge_state(struct extent_io_tree *tree,
401 struct extent_state *state)
403 struct extent_state *other;
404 struct rb_node *other_node;
406 if (state->state & (EXTENT_LOCKED | EXTENT_BOUNDARY))
409 other_node = rb_prev(&state->rb_node);
411 other = rb_entry(other_node, struct extent_state, rb_node);
412 if (other->end == state->start - 1 &&
413 other->state == state->state) {
414 if (tree->private_data &&
415 is_data_inode(tree->private_data))
416 btrfs_merge_delalloc_extent(tree->private_data,
418 state->start = other->start;
419 rb_erase(&other->rb_node, &tree->state);
420 RB_CLEAR_NODE(&other->rb_node);
421 free_extent_state(other);
424 other_node = rb_next(&state->rb_node);
426 other = rb_entry(other_node, struct extent_state, rb_node);
427 if (other->start == state->end + 1 &&
428 other->state == state->state) {
429 if (tree->private_data &&
430 is_data_inode(tree->private_data))
431 btrfs_merge_delalloc_extent(tree->private_data,
433 state->end = other->end;
434 rb_erase(&other->rb_node, &tree->state);
435 RB_CLEAR_NODE(&other->rb_node);
436 free_extent_state(other);
441 static void set_state_bits(struct extent_io_tree *tree,
442 struct extent_state *state, unsigned *bits,
443 struct extent_changeset *changeset);
446 * insert an extent_state struct into the tree. 'bits' are set on the
447 * struct before it is inserted.
449 * This may return -EEXIST if the extent is already there, in which case the
450 * state struct is freed.
452 * The tree lock is not taken internally. This is a utility function and
453 * probably isn't what you want to call (see set/clear_extent_bit).
455 static int insert_state(struct extent_io_tree *tree,
456 struct extent_state *state, u64 start, u64 end,
458 struct rb_node **parent,
459 unsigned *bits, struct extent_changeset *changeset)
461 struct rb_node *node;
464 WARN(1, KERN_ERR "BTRFS: end < start %llu %llu\n",
466 state->start = start;
469 set_state_bits(tree, state, bits, changeset);
471 node = tree_insert(&tree->state, NULL, end, &state->rb_node, p, parent);
473 struct extent_state *found;
474 found = rb_entry(node, struct extent_state, rb_node);
475 pr_err("BTRFS: found node %llu %llu on insert of %llu %llu\n",
476 found->start, found->end, start, end);
479 merge_state(tree, state);
484 * split a given extent state struct in two, inserting the preallocated
485 * struct 'prealloc' as the newly created second half. 'split' indicates an
486 * offset inside 'orig' where it should be split.
489 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
490 * are two extent state structs in the tree:
491 * prealloc: [orig->start, split - 1]
492 * orig: [ split, orig->end ]
494 * The tree locks are not taken by this function. They need to be held
497 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
498 struct extent_state *prealloc, u64 split)
500 struct rb_node *node;
502 if (tree->private_data && is_data_inode(tree->private_data))
503 btrfs_split_delalloc_extent(tree->private_data, orig, split);
505 prealloc->start = orig->start;
506 prealloc->end = split - 1;
507 prealloc->state = orig->state;
510 node = tree_insert(&tree->state, &orig->rb_node, prealloc->end,
511 &prealloc->rb_node, NULL, NULL);
513 free_extent_state(prealloc);
519 static struct extent_state *next_state(struct extent_state *state)
521 struct rb_node *next = rb_next(&state->rb_node);
523 return rb_entry(next, struct extent_state, rb_node);
529 * utility function to clear some bits in an extent state struct.
530 * it will optionally wake up anyone waiting on this state (wake == 1).
532 * If no bits are set on the state struct after clearing things, the
533 * struct is freed and removed from the tree
535 static struct extent_state *clear_state_bit(struct extent_io_tree *tree,
536 struct extent_state *state,
537 unsigned *bits, int wake,
538 struct extent_changeset *changeset)
540 struct extent_state *next;
541 unsigned bits_to_clear = *bits & ~EXTENT_CTLBITS;
544 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
545 u64 range = state->end - state->start + 1;
546 WARN_ON(range > tree->dirty_bytes);
547 tree->dirty_bytes -= range;
550 if (tree->private_data && is_data_inode(tree->private_data))
551 btrfs_clear_delalloc_extent(tree->private_data, state, bits);
553 ret = add_extent_changeset(state, bits_to_clear, changeset, 0);
555 state->state &= ~bits_to_clear;
558 if (state->state == 0) {
559 next = next_state(state);
560 if (extent_state_in_tree(state)) {
561 rb_erase(&state->rb_node, &tree->state);
562 RB_CLEAR_NODE(&state->rb_node);
563 free_extent_state(state);
568 merge_state(tree, state);
569 next = next_state(state);
574 static struct extent_state *
575 alloc_extent_state_atomic(struct extent_state *prealloc)
578 prealloc = alloc_extent_state(GFP_ATOMIC);
583 static void extent_io_tree_panic(struct extent_io_tree *tree, int err)
585 struct inode *inode = tree->private_data;
587 btrfs_panic(btrfs_sb(inode->i_sb), err,
588 "locking error: extent tree was modified by another thread while locked");
592 * clear some bits on a range in the tree. This may require splitting
593 * or inserting elements in the tree, so the gfp mask is used to
594 * indicate which allocations or sleeping are allowed.
596 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
597 * the given range from the tree regardless of state (ie for truncate).
599 * the range [start, end] is inclusive.
601 * This takes the tree lock, and returns 0 on success and < 0 on error.
603 int __clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
604 unsigned bits, int wake, int delete,
605 struct extent_state **cached_state,
606 gfp_t mask, struct extent_changeset *changeset)
608 struct extent_state *state;
609 struct extent_state *cached;
610 struct extent_state *prealloc = NULL;
611 struct rb_node *node;
616 btrfs_debug_check_extent_io_range(tree, start, end);
617 trace_btrfs_clear_extent_bit(tree, start, end - start + 1, bits);
619 if (bits & EXTENT_DELALLOC)
620 bits |= EXTENT_NORESERVE;
623 bits |= ~EXTENT_CTLBITS;
625 if (bits & (EXTENT_LOCKED | EXTENT_BOUNDARY))
628 if (!prealloc && gfpflags_allow_blocking(mask)) {
630 * Don't care for allocation failure here because we might end
631 * up not needing the pre-allocated extent state at all, which
632 * is the case if we only have in the tree extent states that
633 * cover our input range and don't cover too any other range.
634 * If we end up needing a new extent state we allocate it later.
636 prealloc = alloc_extent_state(mask);
639 spin_lock(&tree->lock);
641 cached = *cached_state;
644 *cached_state = NULL;
648 if (cached && extent_state_in_tree(cached) &&
649 cached->start <= start && cached->end > start) {
651 refcount_dec(&cached->refs);
656 free_extent_state(cached);
659 * this search will find the extents that end after
662 node = tree_search(tree, start);
665 state = rb_entry(node, struct extent_state, rb_node);
667 if (state->start > end)
669 WARN_ON(state->end < start);
670 last_end = state->end;
672 /* the state doesn't have the wanted bits, go ahead */
673 if (!(state->state & bits)) {
674 state = next_state(state);
679 * | ---- desired range ---- |
681 * | ------------- state -------------- |
683 * We need to split the extent we found, and may flip
684 * bits on second half.
686 * If the extent we found extends past our range, we
687 * just split and search again. It'll get split again
688 * the next time though.
690 * If the extent we found is inside our range, we clear
691 * the desired bit on it.
694 if (state->start < start) {
695 prealloc = alloc_extent_state_atomic(prealloc);
697 err = split_state(tree, state, prealloc, start);
699 extent_io_tree_panic(tree, err);
704 if (state->end <= end) {
705 state = clear_state_bit(tree, state, &bits, wake,
712 * | ---- desired range ---- |
714 * We need to split the extent, and clear the bit
717 if (state->start <= end && state->end > end) {
718 prealloc = alloc_extent_state_atomic(prealloc);
720 err = split_state(tree, state, prealloc, end + 1);
722 extent_io_tree_panic(tree, err);
727 clear_state_bit(tree, prealloc, &bits, wake, changeset);
733 state = clear_state_bit(tree, state, &bits, wake, changeset);
735 if (last_end == (u64)-1)
737 start = last_end + 1;
738 if (start <= end && state && !need_resched())
744 spin_unlock(&tree->lock);
745 if (gfpflags_allow_blocking(mask))
750 spin_unlock(&tree->lock);
752 free_extent_state(prealloc);
758 static void wait_on_state(struct extent_io_tree *tree,
759 struct extent_state *state)
760 __releases(tree->lock)
761 __acquires(tree->lock)
764 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
765 spin_unlock(&tree->lock);
767 spin_lock(&tree->lock);
768 finish_wait(&state->wq, &wait);
772 * waits for one or more bits to clear on a range in the state tree.
773 * The range [start, end] is inclusive.
774 * The tree lock is taken by this function
776 static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
779 struct extent_state *state;
780 struct rb_node *node;
782 btrfs_debug_check_extent_io_range(tree, start, end);
784 spin_lock(&tree->lock);
788 * this search will find all the extents that end after
791 node = tree_search(tree, start);
796 state = rb_entry(node, struct extent_state, rb_node);
798 if (state->start > end)
801 if (state->state & bits) {
802 start = state->start;
803 refcount_inc(&state->refs);
804 wait_on_state(tree, state);
805 free_extent_state(state);
808 start = state->end + 1;
813 if (!cond_resched_lock(&tree->lock)) {
814 node = rb_next(node);
819 spin_unlock(&tree->lock);
822 static void set_state_bits(struct extent_io_tree *tree,
823 struct extent_state *state,
824 unsigned *bits, struct extent_changeset *changeset)
826 unsigned bits_to_set = *bits & ~EXTENT_CTLBITS;
829 if (tree->private_data && is_data_inode(tree->private_data))
830 btrfs_set_delalloc_extent(tree->private_data, state, bits);
832 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
833 u64 range = state->end - state->start + 1;
834 tree->dirty_bytes += range;
836 ret = add_extent_changeset(state, bits_to_set, changeset, 1);
838 state->state |= bits_to_set;
841 static void cache_state_if_flags(struct extent_state *state,
842 struct extent_state **cached_ptr,
845 if (cached_ptr && !(*cached_ptr)) {
846 if (!flags || (state->state & flags)) {
848 refcount_inc(&state->refs);
853 static void cache_state(struct extent_state *state,
854 struct extent_state **cached_ptr)
856 return cache_state_if_flags(state, cached_ptr,
857 EXTENT_LOCKED | EXTENT_BOUNDARY);
861 * set some bits on a range in the tree. This may require allocations or
862 * sleeping, so the gfp mask is used to indicate what is allowed.
864 * If any of the exclusive bits are set, this will fail with -EEXIST if some
865 * part of the range already has the desired bits set. The start of the
866 * existing range is returned in failed_start in this case.
868 * [start, end] is inclusive This takes the tree lock.
871 static int __must_check
872 __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
873 unsigned bits, unsigned exclusive_bits,
874 u64 *failed_start, struct extent_state **cached_state,
875 gfp_t mask, struct extent_changeset *changeset)
877 struct extent_state *state;
878 struct extent_state *prealloc = NULL;
879 struct rb_node *node;
881 struct rb_node *parent;
886 btrfs_debug_check_extent_io_range(tree, start, end);
887 trace_btrfs_set_extent_bit(tree, start, end - start + 1, bits);
890 if (!prealloc && gfpflags_allow_blocking(mask)) {
892 * Don't care for allocation failure here because we might end
893 * up not needing the pre-allocated extent state at all, which
894 * is the case if we only have in the tree extent states that
895 * cover our input range and don't cover too any other range.
896 * If we end up needing a new extent state we allocate it later.
898 prealloc = alloc_extent_state(mask);
901 spin_lock(&tree->lock);
902 if (cached_state && *cached_state) {
903 state = *cached_state;
904 if (state->start <= start && state->end > start &&
905 extent_state_in_tree(state)) {
906 node = &state->rb_node;
911 * this search will find all the extents that end after
914 node = tree_search_for_insert(tree, start, &p, &parent);
916 prealloc = alloc_extent_state_atomic(prealloc);
918 err = insert_state(tree, prealloc, start, end,
919 &p, &parent, &bits, changeset);
921 extent_io_tree_panic(tree, err);
923 cache_state(prealloc, cached_state);
927 state = rb_entry(node, struct extent_state, rb_node);
929 last_start = state->start;
930 last_end = state->end;
933 * | ---- desired range ---- |
936 * Just lock what we found and keep going
938 if (state->start == start && state->end <= end) {
939 if (state->state & exclusive_bits) {
940 *failed_start = state->start;
945 set_state_bits(tree, state, &bits, changeset);
946 cache_state(state, cached_state);
947 merge_state(tree, state);
948 if (last_end == (u64)-1)
950 start = last_end + 1;
951 state = next_state(state);
952 if (start < end && state && state->start == start &&
959 * | ---- desired range ---- |
962 * | ------------- state -------------- |
964 * We need to split the extent we found, and may flip bits on
967 * If the extent we found extends past our
968 * range, we just split and search again. It'll get split
969 * again the next time though.
971 * If the extent we found is inside our range, we set the
974 if (state->start < start) {
975 if (state->state & exclusive_bits) {
976 *failed_start = start;
981 prealloc = alloc_extent_state_atomic(prealloc);
983 err = split_state(tree, state, prealloc, start);
985 extent_io_tree_panic(tree, err);
990 if (state->end <= end) {
991 set_state_bits(tree, state, &bits, changeset);
992 cache_state(state, cached_state);
993 merge_state(tree, state);
994 if (last_end == (u64)-1)
996 start = last_end + 1;
997 state = next_state(state);
998 if (start < end && state && state->start == start &&
1005 * | ---- desired range ---- |
1006 * | state | or | state |
1008 * There's a hole, we need to insert something in it and
1009 * ignore the extent we found.
1011 if (state->start > start) {
1013 if (end < last_start)
1016 this_end = last_start - 1;
1018 prealloc = alloc_extent_state_atomic(prealloc);
1022 * Avoid to free 'prealloc' if it can be merged with
1025 err = insert_state(tree, prealloc, start, this_end,
1026 NULL, NULL, &bits, changeset);
1028 extent_io_tree_panic(tree, err);
1030 cache_state(prealloc, cached_state);
1032 start = this_end + 1;
1036 * | ---- desired range ---- |
1038 * We need to split the extent, and set the bit
1041 if (state->start <= end && state->end > end) {
1042 if (state->state & exclusive_bits) {
1043 *failed_start = start;
1048 prealloc = alloc_extent_state_atomic(prealloc);
1050 err = split_state(tree, state, prealloc, end + 1);
1052 extent_io_tree_panic(tree, err);
1054 set_state_bits(tree, prealloc, &bits, changeset);
1055 cache_state(prealloc, cached_state);
1056 merge_state(tree, prealloc);
1064 spin_unlock(&tree->lock);
1065 if (gfpflags_allow_blocking(mask))
1070 spin_unlock(&tree->lock);
1072 free_extent_state(prealloc);
1078 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1079 unsigned bits, u64 * failed_start,
1080 struct extent_state **cached_state, gfp_t mask)
1082 return __set_extent_bit(tree, start, end, bits, 0, failed_start,
1083 cached_state, mask, NULL);
1088 * convert_extent_bit - convert all bits in a given range from one bit to
1090 * @tree: the io tree to search
1091 * @start: the start offset in bytes
1092 * @end: the end offset in bytes (inclusive)
1093 * @bits: the bits to set in this range
1094 * @clear_bits: the bits to clear in this range
1095 * @cached_state: state that we're going to cache
1097 * This will go through and set bits for the given range. If any states exist
1098 * already in this range they are set with the given bit and cleared of the
1099 * clear_bits. This is only meant to be used by things that are mergeable, ie
1100 * converting from say DELALLOC to DIRTY. This is not meant to be used with
1101 * boundary bits like LOCK.
1103 * All allocations are done with GFP_NOFS.
1105 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1106 unsigned bits, unsigned clear_bits,
1107 struct extent_state **cached_state)
1109 struct extent_state *state;
1110 struct extent_state *prealloc = NULL;
1111 struct rb_node *node;
1113 struct rb_node *parent;
1117 bool first_iteration = true;
1119 btrfs_debug_check_extent_io_range(tree, start, end);
1120 trace_btrfs_convert_extent_bit(tree, start, end - start + 1, bits,
1126 * Best effort, don't worry if extent state allocation fails
1127 * here for the first iteration. We might have a cached state
1128 * that matches exactly the target range, in which case no
1129 * extent state allocations are needed. We'll only know this
1130 * after locking the tree.
1132 prealloc = alloc_extent_state(GFP_NOFS);
1133 if (!prealloc && !first_iteration)
1137 spin_lock(&tree->lock);
1138 if (cached_state && *cached_state) {
1139 state = *cached_state;
1140 if (state->start <= start && state->end > start &&
1141 extent_state_in_tree(state)) {
1142 node = &state->rb_node;
1148 * this search will find all the extents that end after
1151 node = tree_search_for_insert(tree, start, &p, &parent);
1153 prealloc = alloc_extent_state_atomic(prealloc);
1158 err = insert_state(tree, prealloc, start, end,
1159 &p, &parent, &bits, NULL);
1161 extent_io_tree_panic(tree, err);
1162 cache_state(prealloc, cached_state);
1166 state = rb_entry(node, struct extent_state, rb_node);
1168 last_start = state->start;
1169 last_end = state->end;
1172 * | ---- desired range ---- |
1175 * Just lock what we found and keep going
1177 if (state->start == start && state->end <= end) {
1178 set_state_bits(tree, state, &bits, NULL);
1179 cache_state(state, cached_state);
1180 state = clear_state_bit(tree, state, &clear_bits, 0, NULL);
1181 if (last_end == (u64)-1)
1183 start = last_end + 1;
1184 if (start < end && state && state->start == start &&
1191 * | ---- desired range ---- |
1194 * | ------------- state -------------- |
1196 * We need to split the extent we found, and may flip bits on
1199 * If the extent we found extends past our
1200 * range, we just split and search again. It'll get split
1201 * again the next time though.
1203 * If the extent we found is inside our range, we set the
1204 * desired bit on it.
1206 if (state->start < start) {
1207 prealloc = alloc_extent_state_atomic(prealloc);
1212 err = split_state(tree, state, prealloc, start);
1214 extent_io_tree_panic(tree, err);
1218 if (state->end <= end) {
1219 set_state_bits(tree, state, &bits, NULL);
1220 cache_state(state, cached_state);
1221 state = clear_state_bit(tree, state, &clear_bits, 0,
1223 if (last_end == (u64)-1)
1225 start = last_end + 1;
1226 if (start < end && state && state->start == start &&
1233 * | ---- desired range ---- |
1234 * | state | or | state |
1236 * There's a hole, we need to insert something in it and
1237 * ignore the extent we found.
1239 if (state->start > start) {
1241 if (end < last_start)
1244 this_end = last_start - 1;
1246 prealloc = alloc_extent_state_atomic(prealloc);
1253 * Avoid to free 'prealloc' if it can be merged with
1256 err = insert_state(tree, prealloc, start, this_end,
1257 NULL, NULL, &bits, NULL);
1259 extent_io_tree_panic(tree, err);
1260 cache_state(prealloc, cached_state);
1262 start = this_end + 1;
1266 * | ---- desired range ---- |
1268 * We need to split the extent, and set the bit
1271 if (state->start <= end && state->end > end) {
1272 prealloc = alloc_extent_state_atomic(prealloc);
1278 err = split_state(tree, state, prealloc, end + 1);
1280 extent_io_tree_panic(tree, err);
1282 set_state_bits(tree, prealloc, &bits, NULL);
1283 cache_state(prealloc, cached_state);
1284 clear_state_bit(tree, prealloc, &clear_bits, 0, NULL);
1292 spin_unlock(&tree->lock);
1294 first_iteration = false;
1298 spin_unlock(&tree->lock);
1300 free_extent_state(prealloc);
1305 /* wrappers around set/clear extent bit */
1306 int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1307 unsigned bits, struct extent_changeset *changeset)
1310 * We don't support EXTENT_LOCKED yet, as current changeset will
1311 * record any bits changed, so for EXTENT_LOCKED case, it will
1312 * either fail with -EEXIST or changeset will record the whole
1315 BUG_ON(bits & EXTENT_LOCKED);
1317 return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL, GFP_NOFS,
1321 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1322 unsigned bits, int wake, int delete,
1323 struct extent_state **cached)
1325 return __clear_extent_bit(tree, start, end, bits, wake, delete,
1326 cached, GFP_NOFS, NULL);
1329 int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1330 unsigned bits, struct extent_changeset *changeset)
1333 * Don't support EXTENT_LOCKED case, same reason as
1334 * set_record_extent_bits().
1336 BUG_ON(bits & EXTENT_LOCKED);
1338 return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, GFP_NOFS,
1343 * either insert or lock state struct between start and end use mask to tell
1344 * us if waiting is desired.
1346 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1347 struct extent_state **cached_state)
1353 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED,
1354 EXTENT_LOCKED, &failed_start,
1355 cached_state, GFP_NOFS, NULL);
1356 if (err == -EEXIST) {
1357 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1358 start = failed_start;
1361 WARN_ON(start > end);
1366 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1371 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1372 &failed_start, NULL, GFP_NOFS, NULL);
1373 if (err == -EEXIST) {
1374 if (failed_start > start)
1375 clear_extent_bit(tree, start, failed_start - 1,
1376 EXTENT_LOCKED, 1, 0, NULL);
1382 void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
1384 unsigned long index = start >> PAGE_SHIFT;
1385 unsigned long end_index = end >> PAGE_SHIFT;
1388 while (index <= end_index) {
1389 page = find_get_page(inode->i_mapping, index);
1390 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1391 clear_page_dirty_for_io(page);
1397 void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
1399 unsigned long index = start >> PAGE_SHIFT;
1400 unsigned long end_index = end >> PAGE_SHIFT;
1403 while (index <= end_index) {
1404 page = find_get_page(inode->i_mapping, index);
1405 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1406 __set_page_dirty_nobuffers(page);
1407 account_page_redirty(page);
1413 /* find the first state struct with 'bits' set after 'start', and
1414 * return it. tree->lock must be held. NULL will returned if
1415 * nothing was found after 'start'
1417 static struct extent_state *
1418 find_first_extent_bit_state(struct extent_io_tree *tree,
1419 u64 start, unsigned bits)
1421 struct rb_node *node;
1422 struct extent_state *state;
1425 * this search will find all the extents that end after
1428 node = tree_search(tree, start);
1433 state = rb_entry(node, struct extent_state, rb_node);
1434 if (state->end >= start && (state->state & bits))
1437 node = rb_next(node);
1446 * find the first offset in the io tree with 'bits' set. zero is
1447 * returned if we find something, and *start_ret and *end_ret are
1448 * set to reflect the state struct that was found.
1450 * If nothing was found, 1 is returned. If found something, return 0.
1452 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1453 u64 *start_ret, u64 *end_ret, unsigned bits,
1454 struct extent_state **cached_state)
1456 struct extent_state *state;
1459 spin_lock(&tree->lock);
1460 if (cached_state && *cached_state) {
1461 state = *cached_state;
1462 if (state->end == start - 1 && extent_state_in_tree(state)) {
1463 while ((state = next_state(state)) != NULL) {
1464 if (state->state & bits)
1467 free_extent_state(*cached_state);
1468 *cached_state = NULL;
1471 free_extent_state(*cached_state);
1472 *cached_state = NULL;
1475 state = find_first_extent_bit_state(tree, start, bits);
1478 cache_state_if_flags(state, cached_state, 0);
1479 *start_ret = state->start;
1480 *end_ret = state->end;
1484 spin_unlock(&tree->lock);
1489 * find a contiguous range of bytes in the file marked as delalloc, not
1490 * more than 'max_bytes'. start and end are used to return the range,
1492 * true is returned if we find something, false if nothing was in the tree
1494 static noinline bool find_delalloc_range(struct extent_io_tree *tree,
1495 u64 *start, u64 *end, u64 max_bytes,
1496 struct extent_state **cached_state)
1498 struct rb_node *node;
1499 struct extent_state *state;
1500 u64 cur_start = *start;
1502 u64 total_bytes = 0;
1504 spin_lock(&tree->lock);
1507 * this search will find all the extents that end after
1510 node = tree_search(tree, cur_start);
1517 state = rb_entry(node, struct extent_state, rb_node);
1518 if (found && (state->start != cur_start ||
1519 (state->state & EXTENT_BOUNDARY))) {
1522 if (!(state->state & EXTENT_DELALLOC)) {
1528 *start = state->start;
1529 *cached_state = state;
1530 refcount_inc(&state->refs);
1534 cur_start = state->end + 1;
1535 node = rb_next(node);
1536 total_bytes += state->end - state->start + 1;
1537 if (total_bytes >= max_bytes)
1543 spin_unlock(&tree->lock);
1547 static int __process_pages_contig(struct address_space *mapping,
1548 struct page *locked_page,
1549 pgoff_t start_index, pgoff_t end_index,
1550 unsigned long page_ops, pgoff_t *index_ret);
1552 static noinline void __unlock_for_delalloc(struct inode *inode,
1553 struct page *locked_page,
1556 unsigned long index = start >> PAGE_SHIFT;
1557 unsigned long end_index = end >> PAGE_SHIFT;
1559 ASSERT(locked_page);
1560 if (index == locked_page->index && end_index == index)
1563 __process_pages_contig(inode->i_mapping, locked_page, index, end_index,
1567 static noinline int lock_delalloc_pages(struct inode *inode,
1568 struct page *locked_page,
1572 unsigned long index = delalloc_start >> PAGE_SHIFT;
1573 unsigned long index_ret = index;
1574 unsigned long end_index = delalloc_end >> PAGE_SHIFT;
1577 ASSERT(locked_page);
1578 if (index == locked_page->index && index == end_index)
1581 ret = __process_pages_contig(inode->i_mapping, locked_page, index,
1582 end_index, PAGE_LOCK, &index_ret);
1584 __unlock_for_delalloc(inode, locked_page, delalloc_start,
1585 (u64)index_ret << PAGE_SHIFT);
1590 * Find and lock a contiguous range of bytes in the file marked as delalloc, no
1591 * more than @max_bytes. @Start and @end are used to return the range,
1593 * Return: true if we find something
1594 * false if nothing was in the tree
1597 noinline_for_stack bool find_lock_delalloc_range(struct inode *inode,
1598 struct extent_io_tree *tree,
1599 struct page *locked_page, u64 *start,
1602 u64 max_bytes = BTRFS_MAX_EXTENT_SIZE;
1606 struct extent_state *cached_state = NULL;
1611 /* step one, find a bunch of delalloc bytes starting at start */
1612 delalloc_start = *start;
1614 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1615 max_bytes, &cached_state);
1616 if (!found || delalloc_end <= *start) {
1617 *start = delalloc_start;
1618 *end = delalloc_end;
1619 free_extent_state(cached_state);
1624 * start comes from the offset of locked_page. We have to lock
1625 * pages in order, so we can't process delalloc bytes before
1628 if (delalloc_start < *start)
1629 delalloc_start = *start;
1632 * make sure to limit the number of pages we try to lock down
1634 if (delalloc_end + 1 - delalloc_start > max_bytes)
1635 delalloc_end = delalloc_start + max_bytes - 1;
1637 /* step two, lock all the pages after the page that has start */
1638 ret = lock_delalloc_pages(inode, locked_page,
1639 delalloc_start, delalloc_end);
1640 ASSERT(!ret || ret == -EAGAIN);
1641 if (ret == -EAGAIN) {
1642 /* some of the pages are gone, lets avoid looping by
1643 * shortening the size of the delalloc range we're searching
1645 free_extent_state(cached_state);
1646 cached_state = NULL;
1648 max_bytes = PAGE_SIZE;
1657 /* step three, lock the state bits for the whole range */
1658 lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state);
1660 /* then test to make sure it is all still delalloc */
1661 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1662 EXTENT_DELALLOC, 1, cached_state);
1664 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1666 __unlock_for_delalloc(inode, locked_page,
1667 delalloc_start, delalloc_end);
1671 free_extent_state(cached_state);
1672 *start = delalloc_start;
1673 *end = delalloc_end;
1678 static int __process_pages_contig(struct address_space *mapping,
1679 struct page *locked_page,
1680 pgoff_t start_index, pgoff_t end_index,
1681 unsigned long page_ops, pgoff_t *index_ret)
1683 unsigned long nr_pages = end_index - start_index + 1;
1684 unsigned long pages_locked = 0;
1685 pgoff_t index = start_index;
1686 struct page *pages[16];
1691 if (page_ops & PAGE_LOCK) {
1692 ASSERT(page_ops == PAGE_LOCK);
1693 ASSERT(index_ret && *index_ret == start_index);
1696 if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0)
1697 mapping_set_error(mapping, -EIO);
1699 while (nr_pages > 0) {
1700 ret = find_get_pages_contig(mapping, index,
1701 min_t(unsigned long,
1702 nr_pages, ARRAY_SIZE(pages)), pages);
1705 * Only if we're going to lock these pages,
1706 * can we find nothing at @index.
1708 ASSERT(page_ops & PAGE_LOCK);
1713 for (i = 0; i < ret; i++) {
1714 if (page_ops & PAGE_SET_PRIVATE2)
1715 SetPagePrivate2(pages[i]);
1717 if (pages[i] == locked_page) {
1722 if (page_ops & PAGE_CLEAR_DIRTY)
1723 clear_page_dirty_for_io(pages[i]);
1724 if (page_ops & PAGE_SET_WRITEBACK)
1725 set_page_writeback(pages[i]);
1726 if (page_ops & PAGE_SET_ERROR)
1727 SetPageError(pages[i]);
1728 if (page_ops & PAGE_END_WRITEBACK)
1729 end_page_writeback(pages[i]);
1730 if (page_ops & PAGE_UNLOCK)
1731 unlock_page(pages[i]);
1732 if (page_ops & PAGE_LOCK) {
1733 lock_page(pages[i]);
1734 if (!PageDirty(pages[i]) ||
1735 pages[i]->mapping != mapping) {
1736 unlock_page(pages[i]);
1750 if (err && index_ret)
1751 *index_ret = start_index + pages_locked - 1;
1755 void extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end,
1756 u64 delalloc_end, struct page *locked_page,
1757 unsigned clear_bits,
1758 unsigned long page_ops)
1760 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, clear_bits, 1, 0,
1763 __process_pages_contig(inode->i_mapping, locked_page,
1764 start >> PAGE_SHIFT, end >> PAGE_SHIFT,
1769 * count the number of bytes in the tree that have a given bit(s)
1770 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1771 * cached. The total number found is returned.
1773 u64 count_range_bits(struct extent_io_tree *tree,
1774 u64 *start, u64 search_end, u64 max_bytes,
1775 unsigned bits, int contig)
1777 struct rb_node *node;
1778 struct extent_state *state;
1779 u64 cur_start = *start;
1780 u64 total_bytes = 0;
1784 if (WARN_ON(search_end <= cur_start))
1787 spin_lock(&tree->lock);
1788 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1789 total_bytes = tree->dirty_bytes;
1793 * this search will find all the extents that end after
1796 node = tree_search(tree, cur_start);
1801 state = rb_entry(node, struct extent_state, rb_node);
1802 if (state->start > search_end)
1804 if (contig && found && state->start > last + 1)
1806 if (state->end >= cur_start && (state->state & bits) == bits) {
1807 total_bytes += min(search_end, state->end) + 1 -
1808 max(cur_start, state->start);
1809 if (total_bytes >= max_bytes)
1812 *start = max(cur_start, state->start);
1816 } else if (contig && found) {
1819 node = rb_next(node);
1824 spin_unlock(&tree->lock);
1829 * set the private field for a given byte offset in the tree. If there isn't
1830 * an extent_state there already, this does nothing.
1832 static noinline int set_state_failrec(struct extent_io_tree *tree, u64 start,
1833 struct io_failure_record *failrec)
1835 struct rb_node *node;
1836 struct extent_state *state;
1839 spin_lock(&tree->lock);
1841 * this search will find all the extents that end after
1844 node = tree_search(tree, start);
1849 state = rb_entry(node, struct extent_state, rb_node);
1850 if (state->start != start) {
1854 state->failrec = failrec;
1856 spin_unlock(&tree->lock);
1860 static noinline int get_state_failrec(struct extent_io_tree *tree, u64 start,
1861 struct io_failure_record **failrec)
1863 struct rb_node *node;
1864 struct extent_state *state;
1867 spin_lock(&tree->lock);
1869 * this search will find all the extents that end after
1872 node = tree_search(tree, start);
1877 state = rb_entry(node, struct extent_state, rb_node);
1878 if (state->start != start) {
1882 *failrec = state->failrec;
1884 spin_unlock(&tree->lock);
1889 * searches a range in the state tree for a given mask.
1890 * If 'filled' == 1, this returns 1 only if every extent in the tree
1891 * has the bits set. Otherwise, 1 is returned if any bit in the
1892 * range is found set.
1894 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1895 unsigned bits, int filled, struct extent_state *cached)
1897 struct extent_state *state = NULL;
1898 struct rb_node *node;
1901 spin_lock(&tree->lock);
1902 if (cached && extent_state_in_tree(cached) && cached->start <= start &&
1903 cached->end > start)
1904 node = &cached->rb_node;
1906 node = tree_search(tree, start);
1907 while (node && start <= end) {
1908 state = rb_entry(node, struct extent_state, rb_node);
1910 if (filled && state->start > start) {
1915 if (state->start > end)
1918 if (state->state & bits) {
1922 } else if (filled) {
1927 if (state->end == (u64)-1)
1930 start = state->end + 1;
1933 node = rb_next(node);
1940 spin_unlock(&tree->lock);
1945 * helper function to set a given page up to date if all the
1946 * extents in the tree for that page are up to date
1948 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
1950 u64 start = page_offset(page);
1951 u64 end = start + PAGE_SIZE - 1;
1952 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1953 SetPageUptodate(page);
1956 int free_io_failure(struct extent_io_tree *failure_tree,
1957 struct extent_io_tree *io_tree,
1958 struct io_failure_record *rec)
1963 set_state_failrec(failure_tree, rec->start, NULL);
1964 ret = clear_extent_bits(failure_tree, rec->start,
1965 rec->start + rec->len - 1,
1966 EXTENT_LOCKED | EXTENT_DIRTY);
1970 ret = clear_extent_bits(io_tree, rec->start,
1971 rec->start + rec->len - 1,
1981 * this bypasses the standard btrfs submit functions deliberately, as
1982 * the standard behavior is to write all copies in a raid setup. here we only
1983 * want to write the one bad copy. so we do the mapping for ourselves and issue
1984 * submit_bio directly.
1985 * to avoid any synchronization issues, wait for the data after writing, which
1986 * actually prevents the read that triggered the error from finishing.
1987 * currently, there can be no more than two copies of every data bit. thus,
1988 * exactly one rewrite is required.
1990 int repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start,
1991 u64 length, u64 logical, struct page *page,
1992 unsigned int pg_offset, int mirror_num)
1995 struct btrfs_device *dev;
1998 struct btrfs_bio *bbio = NULL;
2001 ASSERT(!(fs_info->sb->s_flags & SB_RDONLY));
2002 BUG_ON(!mirror_num);
2004 bio = btrfs_io_bio_alloc(1);
2005 bio->bi_iter.bi_size = 0;
2006 map_length = length;
2009 * Avoid races with device replace and make sure our bbio has devices
2010 * associated to its stripes that don't go away while we are doing the
2011 * read repair operation.
2013 btrfs_bio_counter_inc_blocked(fs_info);
2014 if (btrfs_is_parity_mirror(fs_info, logical, length)) {
2016 * Note that we don't use BTRFS_MAP_WRITE because it's supposed
2017 * to update all raid stripes, but here we just want to correct
2018 * bad stripe, thus BTRFS_MAP_READ is abused to only get the bad
2019 * stripe's dev and sector.
2021 ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, logical,
2022 &map_length, &bbio, 0);
2024 btrfs_bio_counter_dec(fs_info);
2028 ASSERT(bbio->mirror_num == 1);
2030 ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical,
2031 &map_length, &bbio, mirror_num);
2033 btrfs_bio_counter_dec(fs_info);
2037 BUG_ON(mirror_num != bbio->mirror_num);
2040 sector = bbio->stripes[bbio->mirror_num - 1].physical >> 9;
2041 bio->bi_iter.bi_sector = sector;
2042 dev = bbio->stripes[bbio->mirror_num - 1].dev;
2043 btrfs_put_bbio(bbio);
2044 if (!dev || !dev->bdev ||
2045 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) {
2046 btrfs_bio_counter_dec(fs_info);
2050 bio_set_dev(bio, dev->bdev);
2051 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC;
2052 bio_add_page(bio, page, length, pg_offset);
2054 if (btrfsic_submit_bio_wait(bio)) {
2055 /* try to remap that extent elsewhere? */
2056 btrfs_bio_counter_dec(fs_info);
2058 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
2062 btrfs_info_rl_in_rcu(fs_info,
2063 "read error corrected: ino %llu off %llu (dev %s sector %llu)",
2065 rcu_str_deref(dev->name), sector);
2066 btrfs_bio_counter_dec(fs_info);
2071 int repair_eb_io_failure(struct btrfs_fs_info *fs_info,
2072 struct extent_buffer *eb, int mirror_num)
2074 u64 start = eb->start;
2075 int i, num_pages = num_extent_pages(eb);
2078 if (sb_rdonly(fs_info->sb))
2081 for (i = 0; i < num_pages; i++) {
2082 struct page *p = eb->pages[i];
2084 ret = repair_io_failure(fs_info, 0, start, PAGE_SIZE, start, p,
2085 start - page_offset(p), mirror_num);
2095 * each time an IO finishes, we do a fast check in the IO failure tree
2096 * to see if we need to process or clean up an io_failure_record
2098 int clean_io_failure(struct btrfs_fs_info *fs_info,
2099 struct extent_io_tree *failure_tree,
2100 struct extent_io_tree *io_tree, u64 start,
2101 struct page *page, u64 ino, unsigned int pg_offset)
2104 struct io_failure_record *failrec;
2105 struct extent_state *state;
2110 ret = count_range_bits(failure_tree, &private, (u64)-1, 1,
2115 ret = get_state_failrec(failure_tree, start, &failrec);
2119 BUG_ON(!failrec->this_mirror);
2121 if (failrec->in_validation) {
2122 /* there was no real error, just free the record */
2123 btrfs_debug(fs_info,
2124 "clean_io_failure: freeing dummy error at %llu",
2128 if (sb_rdonly(fs_info->sb))
2131 spin_lock(&io_tree->lock);
2132 state = find_first_extent_bit_state(io_tree,
2135 spin_unlock(&io_tree->lock);
2137 if (state && state->start <= failrec->start &&
2138 state->end >= failrec->start + failrec->len - 1) {
2139 num_copies = btrfs_num_copies(fs_info, failrec->logical,
2141 if (num_copies > 1) {
2142 repair_io_failure(fs_info, ino, start, failrec->len,
2143 failrec->logical, page, pg_offset,
2144 failrec->failed_mirror);
2149 free_io_failure(failure_tree, io_tree, failrec);
2155 * Can be called when
2156 * - hold extent lock
2157 * - under ordered extent
2158 * - the inode is freeing
2160 void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start, u64 end)
2162 struct extent_io_tree *failure_tree = &inode->io_failure_tree;
2163 struct io_failure_record *failrec;
2164 struct extent_state *state, *next;
2166 if (RB_EMPTY_ROOT(&failure_tree->state))
2169 spin_lock(&failure_tree->lock);
2170 state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY);
2172 if (state->start > end)
2175 ASSERT(state->end <= end);
2177 next = next_state(state);
2179 failrec = state->failrec;
2180 free_extent_state(state);
2185 spin_unlock(&failure_tree->lock);
2188 int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end,
2189 struct io_failure_record **failrec_ret)
2191 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2192 struct io_failure_record *failrec;
2193 struct extent_map *em;
2194 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2195 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2196 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2200 ret = get_state_failrec(failure_tree, start, &failrec);
2202 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2206 failrec->start = start;
2207 failrec->len = end - start + 1;
2208 failrec->this_mirror = 0;
2209 failrec->bio_flags = 0;
2210 failrec->in_validation = 0;
2212 read_lock(&em_tree->lock);
2213 em = lookup_extent_mapping(em_tree, start, failrec->len);
2215 read_unlock(&em_tree->lock);
2220 if (em->start > start || em->start + em->len <= start) {
2221 free_extent_map(em);
2224 read_unlock(&em_tree->lock);
2230 logical = start - em->start;
2231 logical = em->block_start + logical;
2232 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2233 logical = em->block_start;
2234 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2235 extent_set_compress_type(&failrec->bio_flags,
2239 btrfs_debug(fs_info,
2240 "Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu",
2241 logical, start, failrec->len);
2243 failrec->logical = logical;
2244 free_extent_map(em);
2246 /* set the bits in the private failure tree */
2247 ret = set_extent_bits(failure_tree, start, end,
2248 EXTENT_LOCKED | EXTENT_DIRTY);
2250 ret = set_state_failrec(failure_tree, start, failrec);
2251 /* set the bits in the inode's tree */
2253 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED);
2259 btrfs_debug(fs_info,
2260 "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d",
2261 failrec->logical, failrec->start, failrec->len,
2262 failrec->in_validation);
2264 * when data can be on disk more than twice, add to failrec here
2265 * (e.g. with a list for failed_mirror) to make
2266 * clean_io_failure() clean all those errors at once.
2270 *failrec_ret = failrec;
2275 bool btrfs_check_repairable(struct inode *inode, unsigned failed_bio_pages,
2276 struct io_failure_record *failrec, int failed_mirror)
2278 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2281 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
2282 if (num_copies == 1) {
2284 * we only have a single copy of the data, so don't bother with
2285 * all the retry and error correction code that follows. no
2286 * matter what the error is, it is very likely to persist.
2288 btrfs_debug(fs_info,
2289 "Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
2290 num_copies, failrec->this_mirror, failed_mirror);
2295 * there are two premises:
2296 * a) deliver good data to the caller
2297 * b) correct the bad sectors on disk
2299 if (failed_bio_pages > 1) {
2301 * to fulfill b), we need to know the exact failing sectors, as
2302 * we don't want to rewrite any more than the failed ones. thus,
2303 * we need separate read requests for the failed bio
2305 * if the following BUG_ON triggers, our validation request got
2306 * merged. we need separate requests for our algorithm to work.
2308 BUG_ON(failrec->in_validation);
2309 failrec->in_validation = 1;
2310 failrec->this_mirror = failed_mirror;
2313 * we're ready to fulfill a) and b) alongside. get a good copy
2314 * of the failed sector and if we succeed, we have setup
2315 * everything for repair_io_failure to do the rest for us.
2317 if (failrec->in_validation) {
2318 BUG_ON(failrec->this_mirror != failed_mirror);
2319 failrec->in_validation = 0;
2320 failrec->this_mirror = 0;
2322 failrec->failed_mirror = failed_mirror;
2323 failrec->this_mirror++;
2324 if (failrec->this_mirror == failed_mirror)
2325 failrec->this_mirror++;
2328 if (failrec->this_mirror > num_copies) {
2329 btrfs_debug(fs_info,
2330 "Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
2331 num_copies, failrec->this_mirror, failed_mirror);
2339 struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio,
2340 struct io_failure_record *failrec,
2341 struct page *page, int pg_offset, int icsum,
2342 bio_end_io_t *endio_func, void *data)
2344 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2346 struct btrfs_io_bio *btrfs_failed_bio;
2347 struct btrfs_io_bio *btrfs_bio;
2349 bio = btrfs_io_bio_alloc(1);
2350 bio->bi_end_io = endio_func;
2351 bio->bi_iter.bi_sector = failrec->logical >> 9;
2352 bio_set_dev(bio, fs_info->fs_devices->latest_bdev);
2353 bio->bi_iter.bi_size = 0;
2354 bio->bi_private = data;
2356 btrfs_failed_bio = btrfs_io_bio(failed_bio);
2357 if (btrfs_failed_bio->csum) {
2358 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
2360 btrfs_bio = btrfs_io_bio(bio);
2361 btrfs_bio->csum = btrfs_bio->csum_inline;
2363 memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + icsum,
2367 bio_add_page(bio, page, failrec->len, pg_offset);
2373 * This is a generic handler for readpage errors. If other copies exist, read
2374 * those and write back good data to the failed position. Does not investigate
2375 * in remapping the failed extent elsewhere, hoping the device will be smart
2376 * enough to do this as needed
2378 static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset,
2379 struct page *page, u64 start, u64 end,
2382 struct io_failure_record *failrec;
2383 struct inode *inode = page->mapping->host;
2384 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2385 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2388 blk_status_t status;
2390 unsigned failed_bio_pages = failed_bio->bi_iter.bi_size >> PAGE_SHIFT;
2392 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
2394 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
2398 if (!btrfs_check_repairable(inode, failed_bio_pages, failrec,
2400 free_io_failure(failure_tree, tree, failrec);
2404 if (failed_bio_pages > 1)
2405 read_mode |= REQ_FAILFAST_DEV;
2407 phy_offset >>= inode->i_sb->s_blocksize_bits;
2408 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
2409 start - page_offset(page),
2410 (int)phy_offset, failed_bio->bi_end_io,
2412 bio->bi_opf = REQ_OP_READ | read_mode;
2414 btrfs_debug(btrfs_sb(inode->i_sb),
2415 "Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d",
2416 read_mode, failrec->this_mirror, failrec->in_validation);
2418 status = tree->ops->submit_bio_hook(tree->private_data, bio, failrec->this_mirror,
2419 failrec->bio_flags, 0);
2421 free_io_failure(failure_tree, tree, failrec);
2423 ret = blk_status_to_errno(status);
2429 /* lots and lots of room for performance fixes in the end_bio funcs */
2431 void end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2433 int uptodate = (err == 0);
2436 btrfs_writepage_endio_finish_ordered(page, start, end, uptodate);
2439 ClearPageUptodate(page);
2441 ret = err < 0 ? err : -EIO;
2442 mapping_set_error(page->mapping, ret);
2447 * after a writepage IO is done, we need to:
2448 * clear the uptodate bits on error
2449 * clear the writeback bits in the extent tree for this IO
2450 * end_page_writeback if the page has no more pending IO
2452 * Scheduling is not allowed, so the extent state tree is expected
2453 * to have one and only one object corresponding to this IO.
2455 static void end_bio_extent_writepage(struct bio *bio)
2457 int error = blk_status_to_errno(bio->bi_status);
2458 struct bio_vec *bvec;
2462 struct bvec_iter_all iter_all;
2464 ASSERT(!bio_flagged(bio, BIO_CLONED));
2465 bio_for_each_segment_all(bvec, bio, i, iter_all) {
2466 struct page *page = bvec->bv_page;
2467 struct inode *inode = page->mapping->host;
2468 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2470 /* We always issue full-page reads, but if some block
2471 * in a page fails to read, blk_update_request() will
2472 * advance bv_offset and adjust bv_len to compensate.
2473 * Print a warning for nonzero offsets, and an error
2474 * if they don't add up to a full page. */
2475 if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) {
2476 if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE)
2478 "partial page write in btrfs with offset %u and length %u",
2479 bvec->bv_offset, bvec->bv_len);
2482 "incomplete page write in btrfs with offset %u and length %u",
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, 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_atomic(tree, start, end, &cached);
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_status;
2523 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
2524 struct extent_io_tree *tree, *failure_tree;
2529 u64 extent_start = 0;
2534 struct bvec_iter_all iter_all;
2536 ASSERT(!bio_flagged(bio, BIO_CLONED));
2537 bio_for_each_segment_all(bvec, bio, i, iter_all) {
2538 struct page *page = bvec->bv_page;
2539 struct inode *inode = page->mapping->host;
2540 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2541 bool data_inode = btrfs_ino(BTRFS_I(inode))
2542 != BTRFS_BTREE_INODE_OBJECTID;
2544 btrfs_debug(fs_info,
2545 "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u",
2546 (u64)bio->bi_iter.bi_sector, bio->bi_status,
2547 io_bio->mirror_num);
2548 tree = &BTRFS_I(inode)->io_tree;
2549 failure_tree = &BTRFS_I(inode)->io_failure_tree;
2551 /* We always issue full-page reads, but if some block
2552 * in a page fails to read, blk_update_request() will
2553 * advance bv_offset and adjust bv_len to compensate.
2554 * Print a warning for nonzero offsets, and an error
2555 * if they don't add up to a full page. */
2556 if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) {
2557 if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE)
2559 "partial page read in btrfs with offset %u and length %u",
2560 bvec->bv_offset, bvec->bv_len);
2563 "incomplete page read in btrfs with offset %u and length %u",
2564 bvec->bv_offset, bvec->bv_len);
2567 start = page_offset(page);
2568 end = start + bvec->bv_offset + bvec->bv_len - 1;
2571 mirror = io_bio->mirror_num;
2572 if (likely(uptodate)) {
2573 ret = tree->ops->readpage_end_io_hook(io_bio, offset,
2579 clean_io_failure(BTRFS_I(inode)->root->fs_info,
2580 failure_tree, tree, start,
2582 btrfs_ino(BTRFS_I(inode)), 0);
2585 if (likely(uptodate))
2591 * The generic bio_readpage_error handles errors the
2592 * following way: If possible, new read requests are
2593 * created and submitted and will end up in
2594 * end_bio_extent_readpage as well (if we're lucky,
2595 * not in the !uptodate case). In that case it returns
2596 * 0 and we just go on with the next page in our bio.
2597 * If it can't handle the error it will return -EIO and
2598 * we remain responsible for that page.
2600 ret = bio_readpage_error(bio, offset, page, start, end,
2603 uptodate = !bio->bi_status;
2608 struct extent_buffer *eb;
2610 eb = (struct extent_buffer *)page->private;
2611 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
2612 eb->read_mirror = mirror;
2613 atomic_dec(&eb->io_pages);
2614 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD,
2616 btree_readahead_hook(eb, -EIO);
2621 if (likely(uptodate)) {
2622 loff_t i_size = i_size_read(inode);
2623 pgoff_t end_index = i_size >> PAGE_SHIFT;
2626 /* Zero out the end if this page straddles i_size */
2627 off = offset_in_page(i_size);
2628 if (page->index == end_index && off)
2629 zero_user_segment(page, off, PAGE_SIZE);
2630 SetPageUptodate(page);
2632 ClearPageUptodate(page);
2638 if (unlikely(!uptodate)) {
2640 endio_readpage_release_extent(tree,
2646 endio_readpage_release_extent(tree, start,
2647 end - start + 1, 0);
2648 } else if (!extent_len) {
2649 extent_start = start;
2650 extent_len = end + 1 - start;
2651 } else if (extent_start + extent_len == start) {
2652 extent_len += end + 1 - start;
2654 endio_readpage_release_extent(tree, extent_start,
2655 extent_len, uptodate);
2656 extent_start = start;
2657 extent_len = end + 1 - start;
2662 endio_readpage_release_extent(tree, extent_start, extent_len,
2664 btrfs_io_bio_free_csum(io_bio);
2669 * Initialize the members up to but not including 'bio'. Use after allocating a
2670 * new bio by bio_alloc_bioset as it does not initialize the bytes outside of
2671 * 'bio' because use of __GFP_ZERO is not supported.
2673 static inline void btrfs_io_bio_init(struct btrfs_io_bio *btrfs_bio)
2675 memset(btrfs_bio, 0, offsetof(struct btrfs_io_bio, bio));
2679 * The following helpers allocate a bio. As it's backed by a bioset, it'll
2680 * never fail. We're returning a bio right now but you can call btrfs_io_bio
2681 * for the appropriate container_of magic
2683 struct bio *btrfs_bio_alloc(struct block_device *bdev, u64 first_byte)
2687 bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, &btrfs_bioset);
2688 bio_set_dev(bio, bdev);
2689 bio->bi_iter.bi_sector = first_byte >> 9;
2690 btrfs_io_bio_init(btrfs_io_bio(bio));
2694 struct bio *btrfs_bio_clone(struct bio *bio)
2696 struct btrfs_io_bio *btrfs_bio;
2699 /* Bio allocation backed by a bioset does not fail */
2700 new = bio_clone_fast(bio, GFP_NOFS, &btrfs_bioset);
2701 btrfs_bio = btrfs_io_bio(new);
2702 btrfs_io_bio_init(btrfs_bio);
2703 btrfs_bio->iter = bio->bi_iter;
2707 struct bio *btrfs_io_bio_alloc(unsigned int nr_iovecs)
2711 /* Bio allocation backed by a bioset does not fail */
2712 bio = bio_alloc_bioset(GFP_NOFS, nr_iovecs, &btrfs_bioset);
2713 btrfs_io_bio_init(btrfs_io_bio(bio));
2717 struct bio *btrfs_bio_clone_partial(struct bio *orig, int offset, int size)
2720 struct btrfs_io_bio *btrfs_bio;
2722 /* this will never fail when it's backed by a bioset */
2723 bio = bio_clone_fast(orig, GFP_NOFS, &btrfs_bioset);
2726 btrfs_bio = btrfs_io_bio(bio);
2727 btrfs_io_bio_init(btrfs_bio);
2729 bio_trim(bio, offset >> 9, size >> 9);
2730 btrfs_bio->iter = bio->bi_iter;
2735 * @opf: bio REQ_OP_* and REQ_* flags as one value
2736 * @tree: tree so we can call our merge_bio hook
2737 * @wbc: optional writeback control for io accounting
2738 * @page: page to add to the bio
2739 * @pg_offset: offset of the new bio or to check whether we are adding
2740 * a contiguous page to the previous one
2741 * @size: portion of page that we want to write
2742 * @offset: starting offset in the page
2743 * @bdev: attach newly created bios to this bdev
2744 * @bio_ret: must be valid pointer, newly allocated bio will be stored there
2745 * @end_io_func: end_io callback for new bio
2746 * @mirror_num: desired mirror to read/write
2747 * @prev_bio_flags: flags of previous bio to see if we can merge the current one
2748 * @bio_flags: flags of the current bio to see if we can merge them
2750 static int submit_extent_page(unsigned int opf, struct extent_io_tree *tree,
2751 struct writeback_control *wbc,
2752 struct page *page, u64 offset,
2753 size_t size, unsigned long pg_offset,
2754 struct block_device *bdev,
2755 struct bio **bio_ret,
2756 bio_end_io_t end_io_func,
2758 unsigned long prev_bio_flags,
2759 unsigned long bio_flags,
2760 bool force_bio_submit)
2764 size_t page_size = min_t(size_t, size, PAGE_SIZE);
2765 sector_t sector = offset >> 9;
2771 bool can_merge = true;
2774 if (prev_bio_flags & EXTENT_BIO_COMPRESSED)
2775 contig = bio->bi_iter.bi_sector == sector;
2777 contig = bio_end_sector(bio) == sector;
2780 if (btrfs_bio_fits_in_stripe(page, page_size, bio, bio_flags))
2783 if (prev_bio_flags != bio_flags || !contig || !can_merge ||
2785 bio_add_page(bio, page, page_size, pg_offset) < page_size) {
2786 ret = submit_one_bio(bio, mirror_num, prev_bio_flags);
2794 wbc_account_io(wbc, page, page_size);
2799 bio = btrfs_bio_alloc(bdev, offset);
2800 bio_add_page(bio, page, page_size, pg_offset);
2801 bio->bi_end_io = end_io_func;
2802 bio->bi_private = tree;
2803 bio->bi_write_hint = page->mapping->host->i_write_hint;
2806 wbc_init_bio(wbc, bio);
2807 wbc_account_io(wbc, page, page_size);
2815 static void attach_extent_buffer_page(struct extent_buffer *eb,
2818 if (!PagePrivate(page)) {
2819 SetPagePrivate(page);
2821 set_page_private(page, (unsigned long)eb);
2823 WARN_ON(page->private != (unsigned long)eb);
2827 void set_page_extent_mapped(struct page *page)
2829 if (!PagePrivate(page)) {
2830 SetPagePrivate(page);
2832 set_page_private(page, EXTENT_PAGE_PRIVATE);
2836 static struct extent_map *
2837 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
2838 u64 start, u64 len, get_extent_t *get_extent,
2839 struct extent_map **em_cached)
2841 struct extent_map *em;
2843 if (em_cached && *em_cached) {
2845 if (extent_map_in_tree(em) && start >= em->start &&
2846 start < extent_map_end(em)) {
2847 refcount_inc(&em->refs);
2851 free_extent_map(em);
2855 em = get_extent(BTRFS_I(inode), page, pg_offset, start, len, 0);
2856 if (em_cached && !IS_ERR_OR_NULL(em)) {
2858 refcount_inc(&em->refs);
2864 * basic readpage implementation. Locked extent state structs are inserted
2865 * into the tree that are removed when the IO is done (by the end_io
2867 * XXX JDM: This needs looking at to ensure proper page locking
2868 * return 0 on success, otherwise return error
2870 static int __do_readpage(struct extent_io_tree *tree,
2872 get_extent_t *get_extent,
2873 struct extent_map **em_cached,
2874 struct bio **bio, int mirror_num,
2875 unsigned long *bio_flags, unsigned int read_flags,
2878 struct inode *inode = page->mapping->host;
2879 u64 start = page_offset(page);
2880 const u64 end = start + PAGE_SIZE - 1;
2883 u64 last_byte = i_size_read(inode);
2886 struct extent_map *em;
2887 struct block_device *bdev;
2890 size_t pg_offset = 0;
2892 size_t disk_io_size;
2893 size_t blocksize = inode->i_sb->s_blocksize;
2894 unsigned long this_bio_flag = 0;
2896 set_page_extent_mapped(page);
2898 if (!PageUptodate(page)) {
2899 if (cleancache_get_page(page) == 0) {
2900 BUG_ON(blocksize != PAGE_SIZE);
2901 unlock_extent(tree, start, end);
2906 if (page->index == last_byte >> PAGE_SHIFT) {
2908 size_t zero_offset = offset_in_page(last_byte);
2911 iosize = PAGE_SIZE - zero_offset;
2912 userpage = kmap_atomic(page);
2913 memset(userpage + zero_offset, 0, iosize);
2914 flush_dcache_page(page);
2915 kunmap_atomic(userpage);
2918 while (cur <= end) {
2919 bool force_bio_submit = false;
2922 if (cur >= last_byte) {
2924 struct extent_state *cached = NULL;
2926 iosize = PAGE_SIZE - pg_offset;
2927 userpage = kmap_atomic(page);
2928 memset(userpage + pg_offset, 0, iosize);
2929 flush_dcache_page(page);
2930 kunmap_atomic(userpage);
2931 set_extent_uptodate(tree, cur, cur + iosize - 1,
2933 unlock_extent_cached(tree, cur,
2934 cur + iosize - 1, &cached);
2937 em = __get_extent_map(inode, page, pg_offset, cur,
2938 end - cur + 1, get_extent, em_cached);
2939 if (IS_ERR_OR_NULL(em)) {
2941 unlock_extent(tree, cur, end);
2944 extent_offset = cur - em->start;
2945 BUG_ON(extent_map_end(em) <= cur);
2948 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2949 this_bio_flag |= EXTENT_BIO_COMPRESSED;
2950 extent_set_compress_type(&this_bio_flag,
2954 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2955 cur_end = min(extent_map_end(em) - 1, end);
2956 iosize = ALIGN(iosize, blocksize);
2957 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2958 disk_io_size = em->block_len;
2959 offset = em->block_start;
2961 offset = em->block_start + extent_offset;
2962 disk_io_size = iosize;
2965 block_start = em->block_start;
2966 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2967 block_start = EXTENT_MAP_HOLE;
2970 * If we have a file range that points to a compressed extent
2971 * and it's followed by a consecutive file range that points to
2972 * to the same compressed extent (possibly with a different
2973 * offset and/or length, so it either points to the whole extent
2974 * or only part of it), we must make sure we do not submit a
2975 * single bio to populate the pages for the 2 ranges because
2976 * this makes the compressed extent read zero out the pages
2977 * belonging to the 2nd range. Imagine the following scenario:
2980 * [0 - 8K] [8K - 24K]
2983 * points to extent X, points to extent X,
2984 * offset 4K, length of 8K offset 0, length 16K
2986 * [extent X, compressed length = 4K uncompressed length = 16K]
2988 * If the bio to read the compressed extent covers both ranges,
2989 * it will decompress extent X into the pages belonging to the
2990 * first range and then it will stop, zeroing out the remaining
2991 * pages that belong to the other range that points to extent X.
2992 * So here we make sure we submit 2 bios, one for the first
2993 * range and another one for the third range. Both will target
2994 * the same physical extent from disk, but we can't currently
2995 * make the compressed bio endio callback populate the pages
2996 * for both ranges because each compressed bio is tightly
2997 * coupled with a single extent map, and each range can have
2998 * an extent map with a different offset value relative to the
2999 * uncompressed data of our extent and different lengths. This
3000 * is a corner case so we prioritize correctness over
3001 * non-optimal behavior (submitting 2 bios for the same extent).
3003 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) &&
3004 prev_em_start && *prev_em_start != (u64)-1 &&
3005 *prev_em_start != em->start)
3006 force_bio_submit = true;
3009 *prev_em_start = em->start;
3011 free_extent_map(em);
3014 /* we've found a hole, just zero and go on */
3015 if (block_start == EXTENT_MAP_HOLE) {
3017 struct extent_state *cached = NULL;
3019 userpage = kmap_atomic(page);
3020 memset(userpage + pg_offset, 0, iosize);
3021 flush_dcache_page(page);
3022 kunmap_atomic(userpage);
3024 set_extent_uptodate(tree, cur, cur + iosize - 1,
3026 unlock_extent_cached(tree, cur,
3027 cur + iosize - 1, &cached);
3029 pg_offset += iosize;
3032 /* the get_extent function already copied into the page */
3033 if (test_range_bit(tree, cur, cur_end,
3034 EXTENT_UPTODATE, 1, NULL)) {
3035 check_page_uptodate(tree, page);
3036 unlock_extent(tree, cur, cur + iosize - 1);
3038 pg_offset += iosize;
3041 /* we have an inline extent but it didn't get marked up
3042 * to date. Error out
3044 if (block_start == EXTENT_MAP_INLINE) {
3046 unlock_extent(tree, cur, cur + iosize - 1);
3048 pg_offset += iosize;
3052 ret = submit_extent_page(REQ_OP_READ | read_flags, tree, NULL,
3053 page, offset, disk_io_size,
3054 pg_offset, bdev, bio,
3055 end_bio_extent_readpage, mirror_num,
3061 *bio_flags = this_bio_flag;
3064 unlock_extent(tree, cur, cur + iosize - 1);
3068 pg_offset += iosize;
3072 if (!PageError(page))
3073 SetPageUptodate(page);
3079 static inline void __do_contiguous_readpages(struct extent_io_tree *tree,
3080 struct page *pages[], int nr_pages,
3082 struct extent_map **em_cached,
3084 unsigned long *bio_flags,
3087 struct inode *inode;
3088 struct btrfs_ordered_extent *ordered;
3091 inode = pages[0]->mapping->host;
3093 lock_extent(tree, start, end);
3094 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
3098 unlock_extent(tree, start, end);
3099 btrfs_start_ordered_extent(inode, ordered, 1);
3100 btrfs_put_ordered_extent(ordered);
3103 for (index = 0; index < nr_pages; index++) {
3104 __do_readpage(tree, pages[index], btrfs_get_extent, em_cached,
3105 bio, 0, bio_flags, REQ_RAHEAD, prev_em_start);
3106 put_page(pages[index]);
3110 static void __extent_readpages(struct extent_io_tree *tree,
3111 struct page *pages[],
3113 struct extent_map **em_cached,
3114 struct bio **bio, unsigned long *bio_flags,
3121 int first_index = 0;
3123 for (index = 0; index < nr_pages; index++) {
3124 page_start = page_offset(pages[index]);
3127 end = start + PAGE_SIZE - 1;
3128 first_index = index;
3129 } else if (end + 1 == page_start) {
3132 __do_contiguous_readpages(tree, &pages[first_index],
3133 index - first_index, start,
3138 end = start + PAGE_SIZE - 1;
3139 first_index = index;
3144 __do_contiguous_readpages(tree, &pages[first_index],
3145 index - first_index, start,
3146 end, em_cached, bio,
3147 bio_flags, prev_em_start);
3150 static int __extent_read_full_page(struct extent_io_tree *tree,
3152 get_extent_t *get_extent,
3153 struct bio **bio, int mirror_num,
3154 unsigned long *bio_flags,
3155 unsigned int read_flags)
3157 struct inode *inode = page->mapping->host;
3158 struct btrfs_ordered_extent *ordered;
3159 u64 start = page_offset(page);
3160 u64 end = start + PAGE_SIZE - 1;
3164 lock_extent(tree, start, end);
3165 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
3169 unlock_extent(tree, start, end);
3170 btrfs_start_ordered_extent(inode, ordered, 1);
3171 btrfs_put_ordered_extent(ordered);
3174 ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num,
3175 bio_flags, read_flags, NULL);
3179 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
3180 get_extent_t *get_extent, int mirror_num)
3182 struct bio *bio = NULL;
3183 unsigned long bio_flags = 0;
3186 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
3189 ret = submit_one_bio(bio, mirror_num, bio_flags);
3193 static void update_nr_written(struct writeback_control *wbc,
3194 unsigned long nr_written)
3196 wbc->nr_to_write -= nr_written;
3200 * helper for __extent_writepage, doing all of the delayed allocation setup.
3202 * This returns 1 if btrfs_run_delalloc_range function did all the work required
3203 * to write the page (copy into inline extent). In this case the IO has
3204 * been started and the page is already unlocked.
3206 * This returns 0 if all went well (page still locked)
3207 * This returns < 0 if there were errors (page still locked)
3209 static noinline_for_stack int writepage_delalloc(struct inode *inode,
3210 struct page *page, struct writeback_control *wbc,
3211 u64 delalloc_start, unsigned long *nr_written)
3213 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
3214 u64 page_end = delalloc_start + PAGE_SIZE - 1;
3216 u64 delalloc_to_write = 0;
3217 u64 delalloc_end = 0;
3219 int page_started = 0;
3222 while (delalloc_end < page_end) {
3223 found = find_lock_delalloc_range(inode, tree,
3228 delalloc_start = delalloc_end + 1;
3231 ret = btrfs_run_delalloc_range(inode, page, delalloc_start,
3232 delalloc_end, &page_started, nr_written, wbc);
3233 /* File system has been set read-only */
3237 * btrfs_run_delalloc_range should return < 0 for error
3238 * but just in case, we use > 0 here meaning the IO is
3239 * started, so we don't want to return > 0 unless
3240 * things are going well.
3242 ret = ret < 0 ? ret : -EIO;
3246 * delalloc_end is already one less than the total length, so
3247 * we don't subtract one from PAGE_SIZE
3249 delalloc_to_write += (delalloc_end - delalloc_start +
3250 PAGE_SIZE) >> PAGE_SHIFT;
3251 delalloc_start = delalloc_end + 1;
3253 if (wbc->nr_to_write < delalloc_to_write) {
3256 if (delalloc_to_write < thresh * 2)
3257 thresh = delalloc_to_write;
3258 wbc->nr_to_write = min_t(u64, delalloc_to_write,
3262 /* did the fill delalloc function already unlock and start
3267 * we've unlocked the page, so we can't update
3268 * the mapping's writeback index, just update
3271 wbc->nr_to_write -= *nr_written;
3282 * helper for __extent_writepage. This calls the writepage start hooks,
3283 * and does the loop to map the page into extents and bios.
3285 * We return 1 if the IO is started and the page is unlocked,
3286 * 0 if all went well (page still locked)
3287 * < 0 if there were errors (page still locked)
3289 static noinline_for_stack int __extent_writepage_io(struct inode *inode,
3291 struct writeback_control *wbc,
3292 struct extent_page_data *epd,
3294 unsigned long nr_written,
3295 unsigned int write_flags, int *nr_ret)
3297 struct extent_io_tree *tree = epd->tree;
3298 u64 start = page_offset(page);
3299 u64 page_end = start + PAGE_SIZE - 1;
3305 struct extent_map *em;
3306 struct block_device *bdev;
3307 size_t pg_offset = 0;
3313 ret = btrfs_writepage_cow_fixup(page, start, page_end);
3315 /* Fixup worker will requeue */
3317 wbc->pages_skipped++;
3319 redirty_page_for_writepage(wbc, page);
3321 update_nr_written(wbc, nr_written);
3327 * we don't want to touch the inode after unlocking the page,
3328 * so we update the mapping writeback index now
3330 update_nr_written(wbc, nr_written + 1);
3333 if (i_size <= start) {
3334 btrfs_writepage_endio_finish_ordered(page, start, page_end, 1);
3338 blocksize = inode->i_sb->s_blocksize;
3340 while (cur <= end) {
3344 if (cur >= i_size) {
3345 btrfs_writepage_endio_finish_ordered(page, cur,
3349 em = btrfs_get_extent(BTRFS_I(inode), page, pg_offset, cur,
3351 if (IS_ERR_OR_NULL(em)) {
3353 ret = PTR_ERR_OR_ZERO(em);
3357 extent_offset = cur - em->start;
3358 em_end = extent_map_end(em);
3359 BUG_ON(em_end <= cur);
3361 iosize = min(em_end - cur, end - cur + 1);
3362 iosize = ALIGN(iosize, blocksize);
3363 offset = em->block_start + extent_offset;
3365 block_start = em->block_start;
3366 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3367 free_extent_map(em);
3371 * compressed and inline extents are written through other
3374 if (compressed || block_start == EXTENT_MAP_HOLE ||
3375 block_start == EXTENT_MAP_INLINE) {
3377 * end_io notification does not happen here for
3378 * compressed extents
3381 btrfs_writepage_endio_finish_ordered(page, cur,
3384 else if (compressed) {
3385 /* we don't want to end_page_writeback on
3386 * a compressed extent. this happens
3393 pg_offset += iosize;
3397 btrfs_set_range_writeback(tree, cur, cur + iosize - 1);
3398 if (!PageWriteback(page)) {
3399 btrfs_err(BTRFS_I(inode)->root->fs_info,
3400 "page %lu not writeback, cur %llu end %llu",
3401 page->index, cur, end);
3404 ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc,
3405 page, offset, iosize, pg_offset,
3407 end_bio_extent_writepage,
3411 if (PageWriteback(page))
3412 end_page_writeback(page);
3416 pg_offset += iosize;
3425 * the writepage semantics are similar to regular writepage. extent
3426 * records are inserted to lock ranges in the tree, and as dirty areas
3427 * are found, they are marked writeback. Then the lock bits are removed
3428 * and the end_io handler clears the writeback ranges
3430 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
3431 struct extent_page_data *epd)
3433 struct inode *inode = page->mapping->host;
3434 u64 start = page_offset(page);
3435 u64 page_end = start + PAGE_SIZE - 1;
3438 size_t pg_offset = 0;
3439 loff_t i_size = i_size_read(inode);
3440 unsigned long end_index = i_size >> PAGE_SHIFT;
3441 unsigned int write_flags = 0;
3442 unsigned long nr_written = 0;
3444 write_flags = wbc_to_write_flags(wbc);
3446 trace___extent_writepage(page, inode, wbc);
3448 WARN_ON(!PageLocked(page));
3450 ClearPageError(page);
3452 pg_offset = offset_in_page(i_size);
3453 if (page->index > end_index ||
3454 (page->index == end_index && !pg_offset)) {
3455 page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
3460 if (page->index == end_index) {
3463 userpage = kmap_atomic(page);
3464 memset(userpage + pg_offset, 0,
3465 PAGE_SIZE - pg_offset);
3466 kunmap_atomic(userpage);
3467 flush_dcache_page(page);
3472 set_page_extent_mapped(page);
3474 if (!epd->extent_locked) {
3475 ret = writepage_delalloc(inode, page, wbc, start, &nr_written);
3482 ret = __extent_writepage_io(inode, page, wbc, epd,
3483 i_size, nr_written, write_flags, &nr);
3489 /* make sure the mapping tag for page dirty gets cleared */
3490 set_page_writeback(page);
3491 end_page_writeback(page);
3493 if (PageError(page)) {
3494 ret = ret < 0 ? ret : -EIO;
3495 end_extent_writepage(page, ret, start, page_end);
3504 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3506 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
3507 TASK_UNINTERRUPTIBLE);
3510 static noinline_for_stack int
3511 lock_extent_buffer_for_io(struct extent_buffer *eb,
3512 struct btrfs_fs_info *fs_info,
3513 struct extent_page_data *epd)
3519 if (!btrfs_try_tree_write_lock(eb)) {
3521 flush_write_bio(epd);
3522 btrfs_tree_lock(eb);
3525 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3526 btrfs_tree_unlock(eb);
3530 flush_write_bio(epd);
3534 wait_on_extent_buffer_writeback(eb);
3535 btrfs_tree_lock(eb);
3536 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3538 btrfs_tree_unlock(eb);
3543 * We need to do this to prevent races in people who check if the eb is
3544 * under IO since we can end up having no IO bits set for a short period
3547 spin_lock(&eb->refs_lock);
3548 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3549 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3550 spin_unlock(&eb->refs_lock);
3551 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3552 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
3554 fs_info->dirty_metadata_batch);
3557 spin_unlock(&eb->refs_lock);
3560 btrfs_tree_unlock(eb);
3565 num_pages = num_extent_pages(eb);
3566 for (i = 0; i < num_pages; i++) {
3567 struct page *p = eb->pages[i];
3569 if (!trylock_page(p)) {
3571 flush_write_bio(epd);
3581 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3583 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3584 smp_mb__after_atomic();
3585 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3588 static void set_btree_ioerr(struct page *page)
3590 struct extent_buffer *eb = (struct extent_buffer *)page->private;
3593 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))
3597 * If writeback for a btree extent that doesn't belong to a log tree
3598 * failed, increment the counter transaction->eb_write_errors.
3599 * We do this because while the transaction is running and before it's
3600 * committing (when we call filemap_fdata[write|wait]_range against
3601 * the btree inode), we might have
3602 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
3603 * returns an error or an error happens during writeback, when we're
3604 * committing the transaction we wouldn't know about it, since the pages
3605 * can be no longer dirty nor marked anymore for writeback (if a
3606 * subsequent modification to the extent buffer didn't happen before the
3607 * transaction commit), which makes filemap_fdata[write|wait]_range not
3608 * able to find the pages tagged with SetPageError at transaction
3609 * commit time. So if this happens we must abort the transaction,
3610 * otherwise we commit a super block with btree roots that point to
3611 * btree nodes/leafs whose content on disk is invalid - either garbage
3612 * or the content of some node/leaf from a past generation that got
3613 * cowed or deleted and is no longer valid.
3615 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
3616 * not be enough - we need to distinguish between log tree extents vs
3617 * non-log tree extents, and the next filemap_fdatawait_range() call
3618 * will catch and clear such errors in the mapping - and that call might
3619 * be from a log sync and not from a transaction commit. Also, checking
3620 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
3621 * not done and would not be reliable - the eb might have been released
3622 * from memory and reading it back again means that flag would not be
3623 * set (since it's a runtime flag, not persisted on disk).
3625 * Using the flags below in the btree inode also makes us achieve the
3626 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
3627 * writeback for all dirty pages and before filemap_fdatawait_range()
3628 * is called, the writeback for all dirty pages had already finished
3629 * with errors - because we were not using AS_EIO/AS_ENOSPC,
3630 * filemap_fdatawait_range() would return success, as it could not know
3631 * that writeback errors happened (the pages were no longer tagged for
3634 switch (eb->log_index) {
3636 set_bit(BTRFS_FS_BTREE_ERR, &eb->fs_info->flags);
3639 set_bit(BTRFS_FS_LOG1_ERR, &eb->fs_info->flags);
3642 set_bit(BTRFS_FS_LOG2_ERR, &eb->fs_info->flags);
3645 BUG(); /* unexpected, logic error */
3649 static void end_bio_extent_buffer_writepage(struct bio *bio)
3651 struct bio_vec *bvec;
3652 struct extent_buffer *eb;
3654 struct bvec_iter_all iter_all;
3656 ASSERT(!bio_flagged(bio, BIO_CLONED));
3657 bio_for_each_segment_all(bvec, bio, i, iter_all) {
3658 struct page *page = bvec->bv_page;
3660 eb = (struct extent_buffer *)page->private;
3662 done = atomic_dec_and_test(&eb->io_pages);
3664 if (bio->bi_status ||
3665 test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) {
3666 ClearPageUptodate(page);
3667 set_btree_ioerr(page);
3670 end_page_writeback(page);
3675 end_extent_buffer_writeback(eb);
3681 static noinline_for_stack int write_one_eb(struct extent_buffer *eb,
3682 struct btrfs_fs_info *fs_info,
3683 struct writeback_control *wbc,
3684 struct extent_page_data *epd)
3686 struct block_device *bdev = fs_info->fs_devices->latest_bdev;
3687 struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
3688 u64 offset = eb->start;
3691 unsigned long start, end;
3692 unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META;
3695 clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
3696 num_pages = num_extent_pages(eb);
3697 atomic_set(&eb->io_pages, num_pages);
3699 /* set btree blocks beyond nritems with 0 to avoid stale content. */
3700 nritems = btrfs_header_nritems(eb);
3701 if (btrfs_header_level(eb) > 0) {
3702 end = btrfs_node_key_ptr_offset(nritems);
3704 memzero_extent_buffer(eb, end, eb->len - end);
3708 * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0
3710 start = btrfs_item_nr_offset(nritems);
3711 end = BTRFS_LEAF_DATA_OFFSET + leaf_data_end(fs_info, eb);
3712 memzero_extent_buffer(eb, start, end - start);
3715 for (i = 0; i < num_pages; i++) {
3716 struct page *p = eb->pages[i];
3718 clear_page_dirty_for_io(p);
3719 set_page_writeback(p);
3720 ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc,
3721 p, offset, PAGE_SIZE, 0, bdev,
3723 end_bio_extent_buffer_writepage,
3727 if (PageWriteback(p))
3728 end_page_writeback(p);
3729 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3730 end_extent_buffer_writeback(eb);
3734 offset += PAGE_SIZE;
3735 update_nr_written(wbc, 1);
3739 if (unlikely(ret)) {
3740 for (; i < num_pages; i++) {
3741 struct page *p = eb->pages[i];
3742 clear_page_dirty_for_io(p);
3750 int btree_write_cache_pages(struct address_space *mapping,
3751 struct writeback_control *wbc)
3753 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
3754 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
3755 struct extent_buffer *eb, *prev_eb = NULL;
3756 struct extent_page_data epd = {
3760 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3764 int nr_to_write_done = 0;
3765 struct pagevec pvec;
3768 pgoff_t end; /* Inclusive */
3772 pagevec_init(&pvec);
3773 if (wbc->range_cyclic) {
3774 index = mapping->writeback_index; /* Start from prev offset */
3777 index = wbc->range_start >> PAGE_SHIFT;
3778 end = wbc->range_end >> PAGE_SHIFT;
3781 if (wbc->sync_mode == WB_SYNC_ALL)
3782 tag = PAGECACHE_TAG_TOWRITE;
3784 tag = PAGECACHE_TAG_DIRTY;
3786 if (wbc->sync_mode == WB_SYNC_ALL)
3787 tag_pages_for_writeback(mapping, index, end);
3788 while (!done && !nr_to_write_done && (index <= end) &&
3789 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
3794 for (i = 0; i < nr_pages; i++) {
3795 struct page *page = pvec.pages[i];
3797 if (!PagePrivate(page))
3800 spin_lock(&mapping->private_lock);
3801 if (!PagePrivate(page)) {
3802 spin_unlock(&mapping->private_lock);
3806 eb = (struct extent_buffer *)page->private;
3809 * Shouldn't happen and normally this would be a BUG_ON
3810 * but no sense in crashing the users box for something
3811 * we can survive anyway.
3814 spin_unlock(&mapping->private_lock);
3818 if (eb == prev_eb) {
3819 spin_unlock(&mapping->private_lock);
3823 ret = atomic_inc_not_zero(&eb->refs);
3824 spin_unlock(&mapping->private_lock);
3829 ret = lock_extent_buffer_for_io(eb, fs_info, &epd);
3831 free_extent_buffer(eb);
3835 ret = write_one_eb(eb, fs_info, wbc, &epd);
3838 free_extent_buffer(eb);
3841 free_extent_buffer(eb);
3844 * the filesystem may choose to bump up nr_to_write.
3845 * We have to make sure to honor the new nr_to_write
3848 nr_to_write_done = wbc->nr_to_write <= 0;
3850 pagevec_release(&pvec);
3853 if (!scanned && !done) {
3855 * We hit the last page and there is more work to be done: wrap
3856 * back to the start of the file
3862 flush_write_bio(&epd);
3867 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3868 * @mapping: address space structure to write
3869 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3870 * @data: data passed to __extent_writepage function
3872 * If a page is already under I/O, write_cache_pages() skips it, even
3873 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3874 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3875 * and msync() need to guarantee that all the data which was dirty at the time
3876 * the call was made get new I/O started against them. If wbc->sync_mode is
3877 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3878 * existing IO to complete.
3880 static int extent_write_cache_pages(struct address_space *mapping,
3881 struct writeback_control *wbc,
3882 struct extent_page_data *epd)
3884 struct inode *inode = mapping->host;
3887 int nr_to_write_done = 0;
3888 struct pagevec pvec;
3891 pgoff_t end; /* Inclusive */
3893 int range_whole = 0;
3898 * We have to hold onto the inode so that ordered extents can do their
3899 * work when the IO finishes. The alternative to this is failing to add
3900 * an ordered extent if the igrab() fails there and that is a huge pain
3901 * to deal with, so instead just hold onto the inode throughout the
3902 * writepages operation. If it fails here we are freeing up the inode
3903 * anyway and we'd rather not waste our time writing out stuff that is
3904 * going to be truncated anyway.
3909 pagevec_init(&pvec);
3910 if (wbc->range_cyclic) {
3911 index = mapping->writeback_index; /* Start from prev offset */
3914 index = wbc->range_start >> PAGE_SHIFT;
3915 end = wbc->range_end >> PAGE_SHIFT;
3916 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
3922 * We do the tagged writepage as long as the snapshot flush bit is set
3923 * and we are the first one who do the filemap_flush() on this inode.
3925 * The nr_to_write == LONG_MAX is needed to make sure other flushers do
3926 * not race in and drop the bit.
3928 if (range_whole && wbc->nr_to_write == LONG_MAX &&
3929 test_and_clear_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
3930 &BTRFS_I(inode)->runtime_flags))
3931 wbc->tagged_writepages = 1;
3933 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
3934 tag = PAGECACHE_TAG_TOWRITE;
3936 tag = PAGECACHE_TAG_DIRTY;
3938 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
3939 tag_pages_for_writeback(mapping, index, end);
3941 while (!done && !nr_to_write_done && (index <= end) &&
3942 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping,
3943 &index, end, tag))) {
3947 for (i = 0; i < nr_pages; i++) {
3948 struct page *page = pvec.pages[i];
3950 done_index = page->index;
3952 * At this point we hold neither the i_pages lock nor
3953 * the page lock: the page may be truncated or
3954 * invalidated (changing page->mapping to NULL),
3955 * or even swizzled back from swapper_space to
3956 * tmpfs file mapping
3958 if (!trylock_page(page)) {
3959 flush_write_bio(epd);
3963 if (unlikely(page->mapping != mapping)) {
3968 if (wbc->sync_mode != WB_SYNC_NONE) {
3969 if (PageWriteback(page))
3970 flush_write_bio(epd);
3971 wait_on_page_writeback(page);
3974 if (PageWriteback(page) ||
3975 !clear_page_dirty_for_io(page)) {
3980 ret = __extent_writepage(page, wbc, epd);
3982 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
3988 * done_index is set past this page,
3989 * so media errors will not choke
3990 * background writeout for the entire
3991 * file. This has consequences for
3992 * range_cyclic semantics (ie. it may
3993 * not be suitable for data integrity
3996 done_index = page->index + 1;
4002 * the filesystem may choose to bump up nr_to_write.
4003 * We have to make sure to honor the new nr_to_write
4006 nr_to_write_done = wbc->nr_to_write <= 0;
4008 pagevec_release(&pvec);
4011 if (!scanned && !done) {
4013 * We hit the last page and there is more work to be done: wrap
4014 * back to the start of the file
4021 if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole))
4022 mapping->writeback_index = done_index;
4024 btrfs_add_delayed_iput(inode);
4028 int extent_write_full_page(struct page *page, struct writeback_control *wbc)
4031 struct extent_page_data epd = {
4033 .tree = &BTRFS_I(page->mapping->host)->io_tree,
4035 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4038 ret = __extent_writepage(page, wbc, &epd);
4040 flush_write_bio(&epd);
4044 int extent_write_locked_range(struct inode *inode, u64 start, u64 end,
4048 struct address_space *mapping = inode->i_mapping;
4049 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
4051 unsigned long nr_pages = (end - start + PAGE_SIZE) >>
4054 struct extent_page_data epd = {
4058 .sync_io = mode == WB_SYNC_ALL,
4060 struct writeback_control wbc_writepages = {
4062 .nr_to_write = nr_pages * 2,
4063 .range_start = start,
4064 .range_end = end + 1,
4067 while (start <= end) {
4068 page = find_get_page(mapping, start >> PAGE_SHIFT);
4069 if (clear_page_dirty_for_io(page))
4070 ret = __extent_writepage(page, &wbc_writepages, &epd);
4072 btrfs_writepage_endio_finish_ordered(page, start,
4073 start + PAGE_SIZE - 1, 1);
4080 flush_write_bio(&epd);
4084 int extent_writepages(struct address_space *mapping,
4085 struct writeback_control *wbc)
4088 struct extent_page_data epd = {
4090 .tree = &BTRFS_I(mapping->host)->io_tree,
4092 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4095 ret = extent_write_cache_pages(mapping, wbc, &epd);
4096 flush_write_bio(&epd);
4100 int extent_readpages(struct address_space *mapping, struct list_head *pages,
4103 struct bio *bio = NULL;
4104 unsigned long bio_flags = 0;
4105 struct page *pagepool[16];
4106 struct extent_map *em_cached = NULL;
4107 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
4109 u64 prev_em_start = (u64)-1;
4111 while (!list_empty(pages)) {
4112 for (nr = 0; nr < ARRAY_SIZE(pagepool) && !list_empty(pages);) {
4113 struct page *page = lru_to_page(pages);
4115 prefetchw(&page->flags);
4116 list_del(&page->lru);
4117 if (add_to_page_cache_lru(page, mapping, page->index,
4118 readahead_gfp_mask(mapping))) {
4123 pagepool[nr++] = page;
4126 __extent_readpages(tree, pagepool, nr, &em_cached, &bio,
4127 &bio_flags, &prev_em_start);
4131 free_extent_map(em_cached);
4134 return submit_one_bio(bio, 0, bio_flags);
4139 * basic invalidatepage code, this waits on any locked or writeback
4140 * ranges corresponding to the page, and then deletes any extent state
4141 * records from the tree
4143 int extent_invalidatepage(struct extent_io_tree *tree,
4144 struct page *page, unsigned long offset)
4146 struct extent_state *cached_state = NULL;
4147 u64 start = page_offset(page);
4148 u64 end = start + PAGE_SIZE - 1;
4149 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
4151 start += ALIGN(offset, blocksize);
4155 lock_extent_bits(tree, start, end, &cached_state);
4156 wait_on_page_writeback(page);
4157 clear_extent_bit(tree, start, end,
4158 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4159 EXTENT_DO_ACCOUNTING,
4160 1, 1, &cached_state);
4165 * a helper for releasepage, this tests for areas of the page that
4166 * are locked or under IO and drops the related state bits if it is safe
4169 static int try_release_extent_state(struct extent_io_tree *tree,
4170 struct page *page, gfp_t mask)
4172 u64 start = page_offset(page);
4173 u64 end = start + PAGE_SIZE - 1;
4176 if (test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL)) {
4180 * at this point we can safely clear everything except the
4181 * locked bit and the nodatasum bit
4183 ret = __clear_extent_bit(tree, start, end,
4184 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
4185 0, 0, NULL, mask, NULL);
4187 /* if clear_extent_bit failed for enomem reasons,
4188 * we can't allow the release to continue.
4199 * a helper for releasepage. As long as there are no locked extents
4200 * in the range corresponding to the page, both state records and extent
4201 * map records are removed
4203 int try_release_extent_mapping(struct page *page, gfp_t mask)
4205 struct extent_map *em;
4206 u64 start = page_offset(page);
4207 u64 end = start + PAGE_SIZE - 1;
4208 struct btrfs_inode *btrfs_inode = BTRFS_I(page->mapping->host);
4209 struct extent_io_tree *tree = &btrfs_inode->io_tree;
4210 struct extent_map_tree *map = &btrfs_inode->extent_tree;
4212 if (gfpflags_allow_blocking(mask) &&
4213 page->mapping->host->i_size > SZ_16M) {
4215 while (start <= end) {
4216 len = end - start + 1;
4217 write_lock(&map->lock);
4218 em = lookup_extent_mapping(map, start, len);
4220 write_unlock(&map->lock);
4223 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
4224 em->start != start) {
4225 write_unlock(&map->lock);
4226 free_extent_map(em);
4229 if (!test_range_bit(tree, em->start,
4230 extent_map_end(em) - 1,
4231 EXTENT_LOCKED, 0, NULL)) {
4232 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4233 &btrfs_inode->runtime_flags);
4234 remove_extent_mapping(map, em);
4235 /* once for the rb tree */
4236 free_extent_map(em);
4238 start = extent_map_end(em);
4239 write_unlock(&map->lock);
4242 free_extent_map(em);
4245 return try_release_extent_state(tree, page, mask);
4249 * helper function for fiemap, which doesn't want to see any holes.
4250 * This maps until we find something past 'last'
4252 static struct extent_map *get_extent_skip_holes(struct inode *inode,
4253 u64 offset, u64 last)
4255 u64 sectorsize = btrfs_inode_sectorsize(inode);
4256 struct extent_map *em;
4263 len = last - offset;
4266 len = ALIGN(len, sectorsize);
4267 em = btrfs_get_extent_fiemap(BTRFS_I(inode), offset, len);
4268 if (IS_ERR_OR_NULL(em))
4271 /* if this isn't a hole return it */
4272 if (em->block_start != EXTENT_MAP_HOLE)
4275 /* this is a hole, advance to the next extent */
4276 offset = extent_map_end(em);
4277 free_extent_map(em);
4285 * To cache previous fiemap extent
4287 * Will be used for merging fiemap extent
4289 struct fiemap_cache {
4298 * Helper to submit fiemap extent.
4300 * Will try to merge current fiemap extent specified by @offset, @phys,
4301 * @len and @flags with cached one.
4302 * And only when we fails to merge, cached one will be submitted as
4305 * Return value is the same as fiemap_fill_next_extent().
4307 static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
4308 struct fiemap_cache *cache,
4309 u64 offset, u64 phys, u64 len, u32 flags)
4317 * Sanity check, extent_fiemap() should have ensured that new
4318 * fiemap extent won't overlap with cached one.
4321 * NOTE: Physical address can overlap, due to compression
4323 if (cache->offset + cache->len > offset) {
4329 * Only merges fiemap extents if
4330 * 1) Their logical addresses are continuous
4332 * 2) Their physical addresses are continuous
4333 * So truly compressed (physical size smaller than logical size)
4334 * extents won't get merged with each other
4336 * 3) Share same flags except FIEMAP_EXTENT_LAST
4337 * So regular extent won't get merged with prealloc extent
4339 if (cache->offset + cache->len == offset &&
4340 cache->phys + cache->len == phys &&
4341 (cache->flags & ~FIEMAP_EXTENT_LAST) ==
4342 (flags & ~FIEMAP_EXTENT_LAST)) {
4344 cache->flags |= flags;
4345 goto try_submit_last;
4348 /* Not mergeable, need to submit cached one */
4349 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
4350 cache->len, cache->flags);
4351 cache->cached = false;
4355 cache->cached = true;
4356 cache->offset = offset;
4359 cache->flags = flags;
4361 if (cache->flags & FIEMAP_EXTENT_LAST) {
4362 ret = fiemap_fill_next_extent(fieinfo, cache->offset,
4363 cache->phys, cache->len, cache->flags);
4364 cache->cached = false;
4370 * Emit last fiemap cache
4372 * The last fiemap cache may still be cached in the following case:
4374 * |<- Fiemap range ->|
4375 * |<------------ First extent ----------->|
4377 * In this case, the first extent range will be cached but not emitted.
4378 * So we must emit it before ending extent_fiemap().
4380 static int emit_last_fiemap_cache(struct btrfs_fs_info *fs_info,
4381 struct fiemap_extent_info *fieinfo,
4382 struct fiemap_cache *cache)
4389 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
4390 cache->len, cache->flags);
4391 cache->cached = false;
4397 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4398 __u64 start, __u64 len)
4402 u64 max = start + len;
4406 u64 last_for_get_extent = 0;
4408 u64 isize = i_size_read(inode);
4409 struct btrfs_key found_key;
4410 struct extent_map *em = NULL;
4411 struct extent_state *cached_state = NULL;
4412 struct btrfs_path *path;
4413 struct btrfs_root *root = BTRFS_I(inode)->root;
4414 struct fiemap_cache cache = { 0 };
4423 path = btrfs_alloc_path();
4426 path->leave_spinning = 1;
4428 start = round_down(start, btrfs_inode_sectorsize(inode));
4429 len = round_up(max, btrfs_inode_sectorsize(inode)) - start;
4432 * lookup the last file extent. We're not using i_size here
4433 * because there might be preallocation past i_size
4435 ret = btrfs_lookup_file_extent(NULL, root, path,
4436 btrfs_ino(BTRFS_I(inode)), -1, 0);
4438 btrfs_free_path(path);
4447 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
4448 found_type = found_key.type;
4450 /* No extents, but there might be delalloc bits */
4451 if (found_key.objectid != btrfs_ino(BTRFS_I(inode)) ||
4452 found_type != BTRFS_EXTENT_DATA_KEY) {
4453 /* have to trust i_size as the end */
4455 last_for_get_extent = isize;
4458 * remember the start of the last extent. There are a
4459 * bunch of different factors that go into the length of the
4460 * extent, so its much less complex to remember where it started
4462 last = found_key.offset;
4463 last_for_get_extent = last + 1;
4465 btrfs_release_path(path);
4468 * we might have some extents allocated but more delalloc past those
4469 * extents. so, we trust isize unless the start of the last extent is
4474 last_for_get_extent = isize;
4477 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4480 em = get_extent_skip_holes(inode, start, last_for_get_extent);
4489 u64 offset_in_extent = 0;
4491 /* break if the extent we found is outside the range */
4492 if (em->start >= max || extent_map_end(em) < off)
4496 * get_extent may return an extent that starts before our
4497 * requested range. We have to make sure the ranges
4498 * we return to fiemap always move forward and don't
4499 * overlap, so adjust the offsets here
4501 em_start = max(em->start, off);
4504 * record the offset from the start of the extent
4505 * for adjusting the disk offset below. Only do this if the
4506 * extent isn't compressed since our in ram offset may be past
4507 * what we have actually allocated on disk.
4509 if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4510 offset_in_extent = em_start - em->start;
4511 em_end = extent_map_end(em);
4512 em_len = em_end - em_start;
4514 if (em->block_start < EXTENT_MAP_LAST_BYTE)
4515 disko = em->block_start + offset_in_extent;
4520 * bump off for our next call to get_extent
4522 off = extent_map_end(em);
4526 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
4528 flags |= FIEMAP_EXTENT_LAST;
4529 } else if (em->block_start == EXTENT_MAP_INLINE) {
4530 flags |= (FIEMAP_EXTENT_DATA_INLINE |
4531 FIEMAP_EXTENT_NOT_ALIGNED);
4532 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
4533 flags |= (FIEMAP_EXTENT_DELALLOC |
4534 FIEMAP_EXTENT_UNKNOWN);
4535 } else if (fieinfo->fi_extents_max) {
4536 u64 bytenr = em->block_start -
4537 (em->start - em->orig_start);
4540 * As btrfs supports shared space, this information
4541 * can be exported to userspace tools via
4542 * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0
4543 * then we're just getting a count and we can skip the
4546 ret = btrfs_check_shared(root,
4547 btrfs_ino(BTRFS_I(inode)),
4552 flags |= FIEMAP_EXTENT_SHARED;
4555 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4556 flags |= FIEMAP_EXTENT_ENCODED;
4557 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4558 flags |= FIEMAP_EXTENT_UNWRITTEN;
4560 free_extent_map(em);
4562 if ((em_start >= last) || em_len == (u64)-1 ||
4563 (last == (u64)-1 && isize <= em_end)) {
4564 flags |= FIEMAP_EXTENT_LAST;
4568 /* now scan forward to see if this is really the last extent. */
4569 em = get_extent_skip_holes(inode, off, last_for_get_extent);
4575 flags |= FIEMAP_EXTENT_LAST;
4578 ret = emit_fiemap_extent(fieinfo, &cache, em_start, disko,
4588 ret = emit_last_fiemap_cache(root->fs_info, fieinfo, &cache);
4589 free_extent_map(em);
4591 btrfs_free_path(path);
4592 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4597 static void __free_extent_buffer(struct extent_buffer *eb)
4599 btrfs_leak_debug_del(&eb->leak_list);
4600 kmem_cache_free(extent_buffer_cache, eb);
4603 int extent_buffer_under_io(struct extent_buffer *eb)
4605 return (atomic_read(&eb->io_pages) ||
4606 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4607 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4611 * Release all pages attached to the extent buffer.
4613 static void btrfs_release_extent_buffer_pages(struct extent_buffer *eb)
4617 int mapped = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
4619 BUG_ON(extent_buffer_under_io(eb));
4621 num_pages = num_extent_pages(eb);
4622 for (i = 0; i < num_pages; i++) {
4623 struct page *page = eb->pages[i];
4628 spin_lock(&page->mapping->private_lock);
4630 * We do this since we'll remove the pages after we've
4631 * removed the eb from the radix tree, so we could race
4632 * and have this page now attached to the new eb. So
4633 * only clear page_private if it's still connected to
4636 if (PagePrivate(page) &&
4637 page->private == (unsigned long)eb) {
4638 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4639 BUG_ON(PageDirty(page));
4640 BUG_ON(PageWriteback(page));
4642 * We need to make sure we haven't be attached
4645 ClearPagePrivate(page);
4646 set_page_private(page, 0);
4647 /* One for the page private */
4652 spin_unlock(&page->mapping->private_lock);
4654 /* One for when we allocated the page */
4660 * Helper for releasing the extent buffer.
4662 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4664 btrfs_release_extent_buffer_pages(eb);
4665 __free_extent_buffer(eb);
4668 static struct extent_buffer *
4669 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
4672 struct extent_buffer *eb = NULL;
4674 eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL);
4677 eb->fs_info = fs_info;
4679 rwlock_init(&eb->lock);
4680 atomic_set(&eb->write_locks, 0);
4681 atomic_set(&eb->read_locks, 0);
4682 atomic_set(&eb->blocking_readers, 0);
4683 atomic_set(&eb->blocking_writers, 0);
4684 atomic_set(&eb->spinning_readers, 0);
4685 eb->lock_nested = 0;
4686 init_waitqueue_head(&eb->write_lock_wq);
4687 init_waitqueue_head(&eb->read_lock_wq);
4689 btrfs_leak_debug_add(&eb->leak_list, &buffers);
4691 spin_lock_init(&eb->refs_lock);
4692 atomic_set(&eb->refs, 1);
4693 atomic_set(&eb->io_pages, 0);
4696 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4698 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4699 > MAX_INLINE_EXTENT_BUFFER_SIZE);
4700 BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE);
4702 #ifdef CONFIG_BTRFS_DEBUG
4703 atomic_set(&eb->spinning_writers, 0);
4709 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src)
4713 struct extent_buffer *new;
4714 int num_pages = num_extent_pages(src);
4716 new = __alloc_extent_buffer(src->fs_info, src->start, src->len);
4720 for (i = 0; i < num_pages; i++) {
4721 p = alloc_page(GFP_NOFS);
4723 btrfs_release_extent_buffer(new);
4726 attach_extent_buffer_page(new, p);
4727 WARN_ON(PageDirty(p));
4730 copy_page(page_address(p), page_address(src->pages[i]));
4733 set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
4734 set_bit(EXTENT_BUFFER_UNMAPPED, &new->bflags);
4739 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4740 u64 start, unsigned long len)
4742 struct extent_buffer *eb;
4746 eb = __alloc_extent_buffer(fs_info, start, len);
4750 num_pages = num_extent_pages(eb);
4751 for (i = 0; i < num_pages; i++) {
4752 eb->pages[i] = alloc_page(GFP_NOFS);
4756 set_extent_buffer_uptodate(eb);
4757 btrfs_set_header_nritems(eb, 0);
4758 set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
4763 __free_page(eb->pages[i - 1]);
4764 __free_extent_buffer(eb);
4768 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4771 return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize);
4774 static void check_buffer_tree_ref(struct extent_buffer *eb)
4777 /* the ref bit is tricky. We have to make sure it is set
4778 * if we have the buffer dirty. Otherwise the
4779 * code to free a buffer can end up dropping a dirty
4782 * Once the ref bit is set, it won't go away while the
4783 * buffer is dirty or in writeback, and it also won't
4784 * go away while we have the reference count on the
4787 * We can't just set the ref bit without bumping the
4788 * ref on the eb because free_extent_buffer might
4789 * see the ref bit and try to clear it. If this happens
4790 * free_extent_buffer might end up dropping our original
4791 * ref by mistake and freeing the page before we are able
4792 * to add one more ref.
4794 * So bump the ref count first, then set the bit. If someone
4795 * beat us to it, drop the ref we added.
4797 refs = atomic_read(&eb->refs);
4798 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4801 spin_lock(&eb->refs_lock);
4802 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4803 atomic_inc(&eb->refs);
4804 spin_unlock(&eb->refs_lock);
4807 static void mark_extent_buffer_accessed(struct extent_buffer *eb,
4808 struct page *accessed)
4812 check_buffer_tree_ref(eb);
4814 num_pages = num_extent_pages(eb);
4815 for (i = 0; i < num_pages; i++) {
4816 struct page *p = eb->pages[i];
4819 mark_page_accessed(p);
4823 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
4826 struct extent_buffer *eb;
4829 eb = radix_tree_lookup(&fs_info->buffer_radix,
4830 start >> PAGE_SHIFT);
4831 if (eb && atomic_inc_not_zero(&eb->refs)) {
4834 * Lock our eb's refs_lock to avoid races with
4835 * free_extent_buffer. When we get our eb it might be flagged
4836 * with EXTENT_BUFFER_STALE and another task running
4837 * free_extent_buffer might have seen that flag set,
4838 * eb->refs == 2, that the buffer isn't under IO (dirty and
4839 * writeback flags not set) and it's still in the tree (flag
4840 * EXTENT_BUFFER_TREE_REF set), therefore being in the process
4841 * of decrementing the extent buffer's reference count twice.
4842 * So here we could race and increment the eb's reference count,
4843 * clear its stale flag, mark it as dirty and drop our reference
4844 * before the other task finishes executing free_extent_buffer,
4845 * which would later result in an attempt to free an extent
4846 * buffer that is dirty.
4848 if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) {
4849 spin_lock(&eb->refs_lock);
4850 spin_unlock(&eb->refs_lock);
4852 mark_extent_buffer_accessed(eb, NULL);
4860 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4861 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
4864 struct extent_buffer *eb, *exists = NULL;
4867 eb = find_extent_buffer(fs_info, start);
4870 eb = alloc_dummy_extent_buffer(fs_info, start);
4873 eb->fs_info = fs_info;
4875 ret = radix_tree_preload(GFP_NOFS);
4878 spin_lock(&fs_info->buffer_lock);
4879 ret = radix_tree_insert(&fs_info->buffer_radix,
4880 start >> PAGE_SHIFT, eb);
4881 spin_unlock(&fs_info->buffer_lock);
4882 radix_tree_preload_end();
4883 if (ret == -EEXIST) {
4884 exists = find_extent_buffer(fs_info, start);
4890 check_buffer_tree_ref(eb);
4891 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
4895 btrfs_release_extent_buffer(eb);
4900 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
4903 unsigned long len = fs_info->nodesize;
4906 unsigned long index = start >> PAGE_SHIFT;
4907 struct extent_buffer *eb;
4908 struct extent_buffer *exists = NULL;
4910 struct address_space *mapping = fs_info->btree_inode->i_mapping;
4914 if (!IS_ALIGNED(start, fs_info->sectorsize)) {
4915 btrfs_err(fs_info, "bad tree block start %llu", start);
4916 return ERR_PTR(-EINVAL);
4919 eb = find_extent_buffer(fs_info, start);
4923 eb = __alloc_extent_buffer(fs_info, start, len);
4925 return ERR_PTR(-ENOMEM);
4927 num_pages = num_extent_pages(eb);
4928 for (i = 0; i < num_pages; i++, index++) {
4929 p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL);
4931 exists = ERR_PTR(-ENOMEM);
4935 spin_lock(&mapping->private_lock);
4936 if (PagePrivate(p)) {
4938 * We could have already allocated an eb for this page
4939 * and attached one so lets see if we can get a ref on
4940 * the existing eb, and if we can we know it's good and
4941 * we can just return that one, else we know we can just
4942 * overwrite page->private.
4944 exists = (struct extent_buffer *)p->private;
4945 if (atomic_inc_not_zero(&exists->refs)) {
4946 spin_unlock(&mapping->private_lock);
4949 mark_extent_buffer_accessed(exists, p);
4955 * Do this so attach doesn't complain and we need to
4956 * drop the ref the old guy had.
4958 ClearPagePrivate(p);
4959 WARN_ON(PageDirty(p));
4962 attach_extent_buffer_page(eb, p);
4963 spin_unlock(&mapping->private_lock);
4964 WARN_ON(PageDirty(p));
4966 if (!PageUptodate(p))
4970 * We can't unlock the pages just yet since the extent buffer
4971 * hasn't been properly inserted in the radix tree, this
4972 * opens a race with btree_releasepage which can free a page
4973 * while we are still filling in all pages for the buffer and
4978 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4980 ret = radix_tree_preload(GFP_NOFS);
4982 exists = ERR_PTR(ret);
4986 spin_lock(&fs_info->buffer_lock);
4987 ret = radix_tree_insert(&fs_info->buffer_radix,
4988 start >> PAGE_SHIFT, eb);
4989 spin_unlock(&fs_info->buffer_lock);
4990 radix_tree_preload_end();
4991 if (ret == -EEXIST) {
4992 exists = find_extent_buffer(fs_info, start);
4998 /* add one reference for the tree */
4999 check_buffer_tree_ref(eb);
5000 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
5003 * Now it's safe to unlock the pages because any calls to
5004 * btree_releasepage will correctly detect that a page belongs to a
5005 * live buffer and won't free them prematurely.
5007 for (i = 0; i < num_pages; i++)
5008 unlock_page(eb->pages[i]);
5012 WARN_ON(!atomic_dec_and_test(&eb->refs));
5013 for (i = 0; i < num_pages; i++) {
5015 unlock_page(eb->pages[i]);
5018 btrfs_release_extent_buffer(eb);
5022 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
5024 struct extent_buffer *eb =
5025 container_of(head, struct extent_buffer, rcu_head);
5027 __free_extent_buffer(eb);
5030 static int release_extent_buffer(struct extent_buffer *eb)
5032 lockdep_assert_held(&eb->refs_lock);
5034 WARN_ON(atomic_read(&eb->refs) == 0);
5035 if (atomic_dec_and_test(&eb->refs)) {
5036 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
5037 struct btrfs_fs_info *fs_info = eb->fs_info;
5039 spin_unlock(&eb->refs_lock);
5041 spin_lock(&fs_info->buffer_lock);
5042 radix_tree_delete(&fs_info->buffer_radix,
5043 eb->start >> PAGE_SHIFT);
5044 spin_unlock(&fs_info->buffer_lock);
5046 spin_unlock(&eb->refs_lock);
5049 /* Should be safe to release our pages at this point */
5050 btrfs_release_extent_buffer_pages(eb);
5051 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5052 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags))) {
5053 __free_extent_buffer(eb);
5057 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
5060 spin_unlock(&eb->refs_lock);
5065 void free_extent_buffer(struct extent_buffer *eb)
5073 refs = atomic_read(&eb->refs);
5074 if ((!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && refs <= 3)
5075 || (test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) &&
5078 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
5083 spin_lock(&eb->refs_lock);
5084 if (atomic_read(&eb->refs) == 2 &&
5085 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
5086 !extent_buffer_under_io(eb) &&
5087 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5088 atomic_dec(&eb->refs);
5091 * I know this is terrible, but it's temporary until we stop tracking
5092 * the uptodate bits and such for the extent buffers.
5094 release_extent_buffer(eb);
5097 void free_extent_buffer_stale(struct extent_buffer *eb)
5102 spin_lock(&eb->refs_lock);
5103 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
5105 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
5106 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5107 atomic_dec(&eb->refs);
5108 release_extent_buffer(eb);
5111 void clear_extent_buffer_dirty(struct extent_buffer *eb)
5117 num_pages = num_extent_pages(eb);
5119 for (i = 0; i < num_pages; i++) {
5120 page = eb->pages[i];
5121 if (!PageDirty(page))
5125 WARN_ON(!PagePrivate(page));
5127 clear_page_dirty_for_io(page);
5128 xa_lock_irq(&page->mapping->i_pages);
5129 if (!PageDirty(page))
5130 __xa_clear_mark(&page->mapping->i_pages,
5131 page_index(page), PAGECACHE_TAG_DIRTY);
5132 xa_unlock_irq(&page->mapping->i_pages);
5133 ClearPageError(page);
5136 WARN_ON(atomic_read(&eb->refs) == 0);
5139 bool set_extent_buffer_dirty(struct extent_buffer *eb)
5145 check_buffer_tree_ref(eb);
5147 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
5149 num_pages = num_extent_pages(eb);
5150 WARN_ON(atomic_read(&eb->refs) == 0);
5151 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
5154 for (i = 0; i < num_pages; i++)
5155 set_page_dirty(eb->pages[i]);
5157 #ifdef CONFIG_BTRFS_DEBUG
5158 for (i = 0; i < num_pages; i++)
5159 ASSERT(PageDirty(eb->pages[i]));
5165 void clear_extent_buffer_uptodate(struct extent_buffer *eb)
5171 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5172 num_pages = num_extent_pages(eb);
5173 for (i = 0; i < num_pages; i++) {
5174 page = eb->pages[i];
5176 ClearPageUptodate(page);
5180 void set_extent_buffer_uptodate(struct extent_buffer *eb)
5186 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5187 num_pages = num_extent_pages(eb);
5188 for (i = 0; i < num_pages; i++) {
5189 page = eb->pages[i];
5190 SetPageUptodate(page);
5194 int read_extent_buffer_pages(struct extent_io_tree *tree,
5195 struct extent_buffer *eb, int wait, int mirror_num)
5201 int locked_pages = 0;
5202 int all_uptodate = 1;
5204 unsigned long num_reads = 0;
5205 struct bio *bio = NULL;
5206 unsigned long bio_flags = 0;
5208 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
5211 num_pages = num_extent_pages(eb);
5212 for (i = 0; i < num_pages; i++) {
5213 page = eb->pages[i];
5214 if (wait == WAIT_NONE) {
5215 if (!trylock_page(page))
5223 * We need to firstly lock all pages to make sure that
5224 * the uptodate bit of our pages won't be affected by
5225 * clear_extent_buffer_uptodate().
5227 for (i = 0; i < num_pages; i++) {
5228 page = eb->pages[i];
5229 if (!PageUptodate(page)) {
5236 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5240 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
5241 eb->read_mirror = 0;
5242 atomic_set(&eb->io_pages, num_reads);
5243 for (i = 0; i < num_pages; i++) {
5244 page = eb->pages[i];
5246 if (!PageUptodate(page)) {
5248 atomic_dec(&eb->io_pages);
5253 ClearPageError(page);
5254 err = __extent_read_full_page(tree, page,
5255 btree_get_extent, &bio,
5256 mirror_num, &bio_flags,
5261 * We use &bio in above __extent_read_full_page,
5262 * so we ensure that if it returns error, the
5263 * current page fails to add itself to bio and
5264 * it's been unlocked.
5266 * We must dec io_pages by ourselves.
5268 atomic_dec(&eb->io_pages);
5276 err = submit_one_bio(bio, mirror_num, bio_flags);
5281 if (ret || wait != WAIT_COMPLETE)
5284 for (i = 0; i < num_pages; i++) {
5285 page = eb->pages[i];
5286 wait_on_page_locked(page);
5287 if (!PageUptodate(page))
5294 while (locked_pages > 0) {
5296 page = eb->pages[locked_pages];
5302 void read_extent_buffer(const struct extent_buffer *eb, void *dstv,
5303 unsigned long start, unsigned long len)
5309 char *dst = (char *)dstv;
5310 size_t start_offset = offset_in_page(eb->start);
5311 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5313 if (start + len > eb->len) {
5314 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n",
5315 eb->start, eb->len, start, len);
5316 memset(dst, 0, len);
5320 offset = offset_in_page(start_offset + start);
5323 page = eb->pages[i];
5325 cur = min(len, (PAGE_SIZE - offset));
5326 kaddr = page_address(page);
5327 memcpy(dst, kaddr + offset, cur);
5336 int read_extent_buffer_to_user(const struct extent_buffer *eb,
5338 unsigned long start, unsigned long len)
5344 char __user *dst = (char __user *)dstv;
5345 size_t start_offset = offset_in_page(eb->start);
5346 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5349 WARN_ON(start > eb->len);
5350 WARN_ON(start + len > eb->start + eb->len);
5352 offset = offset_in_page(start_offset + start);
5355 page = eb->pages[i];
5357 cur = min(len, (PAGE_SIZE - offset));
5358 kaddr = page_address(page);
5359 if (copy_to_user(dst, kaddr + offset, cur)) {
5374 * return 0 if the item is found within a page.
5375 * return 1 if the item spans two pages.
5376 * return -EINVAL otherwise.
5378 int map_private_extent_buffer(const struct extent_buffer *eb,
5379 unsigned long start, unsigned long min_len,
5380 char **map, unsigned long *map_start,
5381 unsigned long *map_len)
5386 size_t start_offset = offset_in_page(eb->start);
5387 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5388 unsigned long end_i = (start_offset + start + min_len - 1) >>
5391 if (start + min_len > eb->len) {
5392 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n",
5393 eb->start, eb->len, start, min_len);
5401 offset = start_offset;
5405 *map_start = ((u64)i << PAGE_SHIFT) - start_offset;
5409 kaddr = page_address(p);
5410 *map = kaddr + offset;
5411 *map_len = PAGE_SIZE - offset;
5415 int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,
5416 unsigned long start, unsigned long len)
5422 char *ptr = (char *)ptrv;
5423 size_t start_offset = offset_in_page(eb->start);
5424 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5427 WARN_ON(start > eb->len);
5428 WARN_ON(start + len > eb->start + eb->len);
5430 offset = offset_in_page(start_offset + start);
5433 page = eb->pages[i];
5435 cur = min(len, (PAGE_SIZE - offset));
5437 kaddr = page_address(page);
5438 ret = memcmp(ptr, kaddr + offset, cur);
5450 void write_extent_buffer_chunk_tree_uuid(struct extent_buffer *eb,
5455 WARN_ON(!PageUptodate(eb->pages[0]));
5456 kaddr = page_address(eb->pages[0]);
5457 memcpy(kaddr + offsetof(struct btrfs_header, chunk_tree_uuid), srcv,
5461 void write_extent_buffer_fsid(struct extent_buffer *eb, const void *srcv)
5465 WARN_ON(!PageUptodate(eb->pages[0]));
5466 kaddr = page_address(eb->pages[0]);
5467 memcpy(kaddr + offsetof(struct btrfs_header, fsid), srcv,
5471 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
5472 unsigned long start, unsigned long len)
5478 char *src = (char *)srcv;
5479 size_t start_offset = offset_in_page(eb->start);
5480 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5482 WARN_ON(start > eb->len);
5483 WARN_ON(start + len > eb->start + eb->len);
5485 offset = offset_in_page(start_offset + start);
5488 page = eb->pages[i];
5489 WARN_ON(!PageUptodate(page));
5491 cur = min(len, PAGE_SIZE - offset);
5492 kaddr = page_address(page);
5493 memcpy(kaddr + offset, src, cur);
5502 void memzero_extent_buffer(struct extent_buffer *eb, unsigned long start,
5509 size_t start_offset = offset_in_page(eb->start);
5510 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5512 WARN_ON(start > eb->len);
5513 WARN_ON(start + len > eb->start + eb->len);
5515 offset = offset_in_page(start_offset + start);
5518 page = eb->pages[i];
5519 WARN_ON(!PageUptodate(page));
5521 cur = min(len, PAGE_SIZE - offset);
5522 kaddr = page_address(page);
5523 memset(kaddr + offset, 0, cur);
5531 void copy_extent_buffer_full(struct extent_buffer *dst,
5532 struct extent_buffer *src)
5537 ASSERT(dst->len == src->len);
5539 num_pages = num_extent_pages(dst);
5540 for (i = 0; i < num_pages; i++)
5541 copy_page(page_address(dst->pages[i]),
5542 page_address(src->pages[i]));
5545 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
5546 unsigned long dst_offset, unsigned long src_offset,
5549 u64 dst_len = dst->len;
5554 size_t start_offset = offset_in_page(dst->start);
5555 unsigned long i = (start_offset + dst_offset) >> PAGE_SHIFT;
5557 WARN_ON(src->len != dst_len);
5559 offset = offset_in_page(start_offset + dst_offset);
5562 page = dst->pages[i];
5563 WARN_ON(!PageUptodate(page));
5565 cur = min(len, (unsigned long)(PAGE_SIZE - offset));
5567 kaddr = page_address(page);
5568 read_extent_buffer(src, kaddr + offset, src_offset, cur);
5578 * eb_bitmap_offset() - calculate the page and offset of the byte containing the
5580 * @eb: the extent buffer
5581 * @start: offset of the bitmap item in the extent buffer
5583 * @page_index: return index of the page in the extent buffer that contains the
5585 * @page_offset: return offset into the page given by page_index
5587 * This helper hides the ugliness of finding the byte in an extent buffer which
5588 * contains a given bit.
5590 static inline void eb_bitmap_offset(struct extent_buffer *eb,
5591 unsigned long start, unsigned long nr,
5592 unsigned long *page_index,
5593 size_t *page_offset)
5595 size_t start_offset = offset_in_page(eb->start);
5596 size_t byte_offset = BIT_BYTE(nr);
5600 * The byte we want is the offset of the extent buffer + the offset of
5601 * the bitmap item in the extent buffer + the offset of the byte in the
5604 offset = start_offset + start + byte_offset;
5606 *page_index = offset >> PAGE_SHIFT;
5607 *page_offset = offset_in_page(offset);
5611 * extent_buffer_test_bit - determine whether a bit in a bitmap item is set
5612 * @eb: the extent buffer
5613 * @start: offset of the bitmap item in the extent buffer
5614 * @nr: bit number to test
5616 int extent_buffer_test_bit(struct extent_buffer *eb, unsigned long start,
5624 eb_bitmap_offset(eb, start, nr, &i, &offset);
5625 page = eb->pages[i];
5626 WARN_ON(!PageUptodate(page));
5627 kaddr = page_address(page);
5628 return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1)));
5632 * extent_buffer_bitmap_set - set an area of a bitmap
5633 * @eb: the extent buffer
5634 * @start: offset of the bitmap item in the extent buffer
5635 * @pos: bit number of the first bit
5636 * @len: number of bits to set
5638 void extent_buffer_bitmap_set(struct extent_buffer *eb, unsigned long start,
5639 unsigned long pos, unsigned long len)
5645 const unsigned int size = pos + len;
5646 int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5647 u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos);
5649 eb_bitmap_offset(eb, start, pos, &i, &offset);
5650 page = eb->pages[i];
5651 WARN_ON(!PageUptodate(page));
5652 kaddr = page_address(page);
5654 while (len >= bits_to_set) {
5655 kaddr[offset] |= mask_to_set;
5657 bits_to_set = BITS_PER_BYTE;
5659 if (++offset >= PAGE_SIZE && len > 0) {
5661 page = eb->pages[++i];
5662 WARN_ON(!PageUptodate(page));
5663 kaddr = page_address(page);
5667 mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
5668 kaddr[offset] |= mask_to_set;
5674 * extent_buffer_bitmap_clear - clear an area of a bitmap
5675 * @eb: the extent buffer
5676 * @start: offset of the bitmap item in the extent buffer
5677 * @pos: bit number of the first bit
5678 * @len: number of bits to clear
5680 void extent_buffer_bitmap_clear(struct extent_buffer *eb, unsigned long start,
5681 unsigned long pos, unsigned long len)
5687 const unsigned int size = pos + len;
5688 int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5689 u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos);
5691 eb_bitmap_offset(eb, start, pos, &i, &offset);
5692 page = eb->pages[i];
5693 WARN_ON(!PageUptodate(page));
5694 kaddr = page_address(page);
5696 while (len >= bits_to_clear) {
5697 kaddr[offset] &= ~mask_to_clear;
5698 len -= bits_to_clear;
5699 bits_to_clear = BITS_PER_BYTE;
5701 if (++offset >= PAGE_SIZE && len > 0) {
5703 page = eb->pages[++i];
5704 WARN_ON(!PageUptodate(page));
5705 kaddr = page_address(page);
5709 mask_to_clear &= BITMAP_LAST_BYTE_MASK(size);
5710 kaddr[offset] &= ~mask_to_clear;
5714 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
5716 unsigned long distance = (src > dst) ? src - dst : dst - src;
5717 return distance < len;
5720 static void copy_pages(struct page *dst_page, struct page *src_page,
5721 unsigned long dst_off, unsigned long src_off,
5724 char *dst_kaddr = page_address(dst_page);
5726 int must_memmove = 0;
5728 if (dst_page != src_page) {
5729 src_kaddr = page_address(src_page);
5731 src_kaddr = dst_kaddr;
5732 if (areas_overlap(src_off, dst_off, len))
5737 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
5739 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
5742 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5743 unsigned long src_offset, unsigned long len)
5745 struct btrfs_fs_info *fs_info = dst->fs_info;
5747 size_t dst_off_in_page;
5748 size_t src_off_in_page;
5749 size_t start_offset = offset_in_page(dst->start);
5750 unsigned long dst_i;
5751 unsigned long src_i;
5753 if (src_offset + len > dst->len) {
5755 "memmove bogus src_offset %lu move len %lu dst len %lu",
5756 src_offset, len, dst->len);
5759 if (dst_offset + len > dst->len) {
5761 "memmove bogus dst_offset %lu move len %lu dst len %lu",
5762 dst_offset, len, dst->len);
5767 dst_off_in_page = offset_in_page(start_offset + dst_offset);
5768 src_off_in_page = offset_in_page(start_offset + src_offset);
5770 dst_i = (start_offset + dst_offset) >> PAGE_SHIFT;
5771 src_i = (start_offset + src_offset) >> PAGE_SHIFT;
5773 cur = min(len, (unsigned long)(PAGE_SIZE -
5775 cur = min_t(unsigned long, cur,
5776 (unsigned long)(PAGE_SIZE - dst_off_in_page));
5778 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5779 dst_off_in_page, src_off_in_page, cur);
5787 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5788 unsigned long src_offset, unsigned long len)
5790 struct btrfs_fs_info *fs_info = dst->fs_info;
5792 size_t dst_off_in_page;
5793 size_t src_off_in_page;
5794 unsigned long dst_end = dst_offset + len - 1;
5795 unsigned long src_end = src_offset + len - 1;
5796 size_t start_offset = offset_in_page(dst->start);
5797 unsigned long dst_i;
5798 unsigned long src_i;
5800 if (src_offset + len > dst->len) {
5802 "memmove bogus src_offset %lu move len %lu len %lu",
5803 src_offset, len, dst->len);
5806 if (dst_offset + len > dst->len) {
5808 "memmove bogus dst_offset %lu move len %lu len %lu",
5809 dst_offset, len, dst->len);
5812 if (dst_offset < src_offset) {
5813 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
5817 dst_i = (start_offset + dst_end) >> PAGE_SHIFT;
5818 src_i = (start_offset + src_end) >> PAGE_SHIFT;
5820 dst_off_in_page = offset_in_page(start_offset + dst_end);
5821 src_off_in_page = offset_in_page(start_offset + src_end);
5823 cur = min_t(unsigned long, len, src_off_in_page + 1);
5824 cur = min(cur, dst_off_in_page + 1);
5825 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5826 dst_off_in_page - cur + 1,
5827 src_off_in_page - cur + 1, cur);
5835 int try_release_extent_buffer(struct page *page)
5837 struct extent_buffer *eb;
5840 * We need to make sure nobody is attaching this page to an eb right
5843 spin_lock(&page->mapping->private_lock);
5844 if (!PagePrivate(page)) {
5845 spin_unlock(&page->mapping->private_lock);
5849 eb = (struct extent_buffer *)page->private;
5853 * This is a little awful but should be ok, we need to make sure that
5854 * the eb doesn't disappear out from under us while we're looking at
5857 spin_lock(&eb->refs_lock);
5858 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
5859 spin_unlock(&eb->refs_lock);
5860 spin_unlock(&page->mapping->private_lock);
5863 spin_unlock(&page->mapping->private_lock);
5866 * If tree ref isn't set then we know the ref on this eb is a real ref,
5867 * so just return, this page will likely be freed soon anyway.
5869 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
5870 spin_unlock(&eb->refs_lock);
5874 return release_extent_buffer(eb);