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)
115 struct rb_node rb_node;
118 struct extent_page_data {
120 struct extent_io_tree *tree;
121 /* tells writepage not to lock the state bits for this range
122 * it still does the unlocking
124 unsigned int extent_locked:1;
126 /* tells the submit_bio code to use REQ_SYNC */
127 unsigned int sync_io:1;
130 static int add_extent_changeset(struct extent_state *state, unsigned bits,
131 struct extent_changeset *changeset,
138 if (set && (state->state & bits) == bits)
140 if (!set && (state->state & bits) == 0)
142 changeset->bytes_changed += state->end - state->start + 1;
143 ret = ulist_add(&changeset->range_changed, state->start, state->end,
148 static int __must_check submit_one_bio(struct bio *bio, int mirror_num,
149 unsigned long bio_flags)
151 blk_status_t ret = 0;
152 struct bio_vec *bvec = bio_last_bvec_all(bio);
154 struct extent_io_tree *tree = bio->bi_private;
157 mp_bvec_last_segment(bvec, &bv);
158 start = page_offset(bv.bv_page) + bv.bv_offset;
160 bio->bi_private = NULL;
163 ret = tree->ops->submit_bio_hook(tree->private_data, bio,
164 mirror_num, bio_flags, start);
166 btrfsic_submit_bio(bio);
168 return blk_status_to_errno(ret);
171 /* Cleanup unsubmitted bios */
172 static void end_write_bio(struct extent_page_data *epd, int ret)
175 epd->bio->bi_status = errno_to_blk_status(ret);
182 * Submit bio from extent page data via submit_one_bio
184 * Return 0 if everything is OK.
185 * Return <0 for error.
187 static int __must_check flush_write_bio(struct extent_page_data *epd)
192 ret = submit_one_bio(epd->bio, 0, 0);
194 * Clean up of epd->bio is handled by its endio function.
195 * And endio is either triggered by successful bio execution
196 * or the error handler of submit bio hook.
197 * So at this point, no matter what happened, we don't need
198 * to clean up epd->bio.
205 int __init extent_io_init(void)
207 extent_state_cache = kmem_cache_create("btrfs_extent_state",
208 sizeof(struct extent_state), 0,
209 SLAB_MEM_SPREAD, NULL);
210 if (!extent_state_cache)
213 extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer",
214 sizeof(struct extent_buffer), 0,
215 SLAB_MEM_SPREAD, NULL);
216 if (!extent_buffer_cache)
217 goto free_state_cache;
219 if (bioset_init(&btrfs_bioset, BIO_POOL_SIZE,
220 offsetof(struct btrfs_io_bio, bio),
222 goto free_buffer_cache;
224 if (bioset_integrity_create(&btrfs_bioset, BIO_POOL_SIZE))
230 bioset_exit(&btrfs_bioset);
233 kmem_cache_destroy(extent_buffer_cache);
234 extent_buffer_cache = NULL;
237 kmem_cache_destroy(extent_state_cache);
238 extent_state_cache = NULL;
242 void __cold extent_io_exit(void)
244 btrfs_leak_debug_check();
247 * Make sure all delayed rcu free are flushed before we
251 kmem_cache_destroy(extent_state_cache);
252 kmem_cache_destroy(extent_buffer_cache);
253 bioset_exit(&btrfs_bioset);
256 void extent_io_tree_init(struct btrfs_fs_info *fs_info,
257 struct extent_io_tree *tree, unsigned int owner,
260 tree->fs_info = fs_info;
261 tree->state = RB_ROOT;
263 tree->dirty_bytes = 0;
264 spin_lock_init(&tree->lock);
265 tree->private_data = private_data;
269 static struct extent_state *alloc_extent_state(gfp_t mask)
271 struct extent_state *state;
274 * The given mask might be not appropriate for the slab allocator,
275 * drop the unsupported bits
277 mask &= ~(__GFP_DMA32|__GFP_HIGHMEM);
278 state = kmem_cache_alloc(extent_state_cache, mask);
282 state->failrec = NULL;
283 RB_CLEAR_NODE(&state->rb_node);
284 btrfs_leak_debug_add(&state->leak_list, &states);
285 refcount_set(&state->refs, 1);
286 init_waitqueue_head(&state->wq);
287 trace_alloc_extent_state(state, mask, _RET_IP_);
291 void free_extent_state(struct extent_state *state)
295 if (refcount_dec_and_test(&state->refs)) {
296 WARN_ON(extent_state_in_tree(state));
297 btrfs_leak_debug_del(&state->leak_list);
298 trace_free_extent_state(state, _RET_IP_);
299 kmem_cache_free(extent_state_cache, state);
303 static struct rb_node *tree_insert(struct rb_root *root,
304 struct rb_node *search_start,
306 struct rb_node *node,
307 struct rb_node ***p_in,
308 struct rb_node **parent_in)
311 struct rb_node *parent = NULL;
312 struct tree_entry *entry;
314 if (p_in && parent_in) {
320 p = search_start ? &search_start : &root->rb_node;
323 entry = rb_entry(parent, struct tree_entry, rb_node);
325 if (offset < entry->start)
327 else if (offset > entry->end)
334 rb_link_node(node, parent, p);
335 rb_insert_color(node, root);
339 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
340 struct rb_node **next_ret,
341 struct rb_node **prev_ret,
342 struct rb_node ***p_ret,
343 struct rb_node **parent_ret)
345 struct rb_root *root = &tree->state;
346 struct rb_node **n = &root->rb_node;
347 struct rb_node *prev = NULL;
348 struct rb_node *orig_prev = NULL;
349 struct tree_entry *entry;
350 struct tree_entry *prev_entry = NULL;
354 entry = rb_entry(prev, struct tree_entry, rb_node);
357 if (offset < entry->start)
359 else if (offset > entry->end)
372 while (prev && offset > prev_entry->end) {
373 prev = rb_next(prev);
374 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
381 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
382 while (prev && offset < prev_entry->start) {
383 prev = rb_prev(prev);
384 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
391 static inline struct rb_node *
392 tree_search_for_insert(struct extent_io_tree *tree,
394 struct rb_node ***p_ret,
395 struct rb_node **parent_ret)
397 struct rb_node *next= NULL;
400 ret = __etree_search(tree, offset, &next, NULL, p_ret, parent_ret);
406 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
409 return tree_search_for_insert(tree, offset, NULL, NULL);
413 * utility function to look for merge candidates inside a given range.
414 * Any extents with matching state are merged together into a single
415 * extent in the tree. Extents with EXTENT_IO in their state field
416 * are not merged because the end_io handlers need to be able to do
417 * operations on them without sleeping (or doing allocations/splits).
419 * This should be called with the tree lock held.
421 static void merge_state(struct extent_io_tree *tree,
422 struct extent_state *state)
424 struct extent_state *other;
425 struct rb_node *other_node;
427 if (state->state & (EXTENT_LOCKED | EXTENT_BOUNDARY))
430 other_node = rb_prev(&state->rb_node);
432 other = rb_entry(other_node, struct extent_state, rb_node);
433 if (other->end == state->start - 1 &&
434 other->state == state->state) {
435 if (tree->private_data &&
436 is_data_inode(tree->private_data))
437 btrfs_merge_delalloc_extent(tree->private_data,
439 state->start = other->start;
440 rb_erase(&other->rb_node, &tree->state);
441 RB_CLEAR_NODE(&other->rb_node);
442 free_extent_state(other);
445 other_node = rb_next(&state->rb_node);
447 other = rb_entry(other_node, struct extent_state, rb_node);
448 if (other->start == state->end + 1 &&
449 other->state == state->state) {
450 if (tree->private_data &&
451 is_data_inode(tree->private_data))
452 btrfs_merge_delalloc_extent(tree->private_data,
454 state->end = other->end;
455 rb_erase(&other->rb_node, &tree->state);
456 RB_CLEAR_NODE(&other->rb_node);
457 free_extent_state(other);
462 static void set_state_bits(struct extent_io_tree *tree,
463 struct extent_state *state, unsigned *bits,
464 struct extent_changeset *changeset);
467 * insert an extent_state struct into the tree. 'bits' are set on the
468 * struct before it is inserted.
470 * This may return -EEXIST if the extent is already there, in which case the
471 * state struct is freed.
473 * The tree lock is not taken internally. This is a utility function and
474 * probably isn't what you want to call (see set/clear_extent_bit).
476 static int insert_state(struct extent_io_tree *tree,
477 struct extent_state *state, u64 start, u64 end,
479 struct rb_node **parent,
480 unsigned *bits, struct extent_changeset *changeset)
482 struct rb_node *node;
485 WARN(1, KERN_ERR "BTRFS: end < start %llu %llu\n",
487 state->start = start;
490 set_state_bits(tree, state, bits, changeset);
492 node = tree_insert(&tree->state, NULL, end, &state->rb_node, p, parent);
494 struct extent_state *found;
495 found = rb_entry(node, struct extent_state, rb_node);
496 pr_err("BTRFS: found node %llu %llu on insert of %llu %llu\n",
497 found->start, found->end, start, end);
500 merge_state(tree, state);
505 * split a given extent state struct in two, inserting the preallocated
506 * struct 'prealloc' as the newly created second half. 'split' indicates an
507 * offset inside 'orig' where it should be split.
510 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
511 * are two extent state structs in the tree:
512 * prealloc: [orig->start, split - 1]
513 * orig: [ split, orig->end ]
515 * The tree locks are not taken by this function. They need to be held
518 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
519 struct extent_state *prealloc, u64 split)
521 struct rb_node *node;
523 if (tree->private_data && is_data_inode(tree->private_data))
524 btrfs_split_delalloc_extent(tree->private_data, orig, split);
526 prealloc->start = orig->start;
527 prealloc->end = split - 1;
528 prealloc->state = orig->state;
531 node = tree_insert(&tree->state, &orig->rb_node, prealloc->end,
532 &prealloc->rb_node, NULL, NULL);
534 free_extent_state(prealloc);
540 static struct extent_state *next_state(struct extent_state *state)
542 struct rb_node *next = rb_next(&state->rb_node);
544 return rb_entry(next, struct extent_state, rb_node);
550 * utility function to clear some bits in an extent state struct.
551 * it will optionally wake up anyone waiting on this state (wake == 1).
553 * If no bits are set on the state struct after clearing things, the
554 * struct is freed and removed from the tree
556 static struct extent_state *clear_state_bit(struct extent_io_tree *tree,
557 struct extent_state *state,
558 unsigned *bits, int wake,
559 struct extent_changeset *changeset)
561 struct extent_state *next;
562 unsigned bits_to_clear = *bits & ~EXTENT_CTLBITS;
565 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
566 u64 range = state->end - state->start + 1;
567 WARN_ON(range > tree->dirty_bytes);
568 tree->dirty_bytes -= range;
571 if (tree->private_data && is_data_inode(tree->private_data))
572 btrfs_clear_delalloc_extent(tree->private_data, state, bits);
574 ret = add_extent_changeset(state, bits_to_clear, changeset, 0);
576 state->state &= ~bits_to_clear;
579 if (state->state == 0) {
580 next = next_state(state);
581 if (extent_state_in_tree(state)) {
582 rb_erase(&state->rb_node, &tree->state);
583 RB_CLEAR_NODE(&state->rb_node);
584 free_extent_state(state);
589 merge_state(tree, state);
590 next = next_state(state);
595 static struct extent_state *
596 alloc_extent_state_atomic(struct extent_state *prealloc)
599 prealloc = alloc_extent_state(GFP_ATOMIC);
604 static void extent_io_tree_panic(struct extent_io_tree *tree, int err)
606 struct inode *inode = tree->private_data;
608 btrfs_panic(btrfs_sb(inode->i_sb), err,
609 "locking error: extent tree was modified by another thread while locked");
613 * clear some bits on a range in the tree. This may require splitting
614 * or inserting elements in the tree, so the gfp mask is used to
615 * indicate which allocations or sleeping are allowed.
617 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
618 * the given range from the tree regardless of state (ie for truncate).
620 * the range [start, end] is inclusive.
622 * This takes the tree lock, and returns 0 on success and < 0 on error.
624 int __clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
625 unsigned bits, int wake, int delete,
626 struct extent_state **cached_state,
627 gfp_t mask, struct extent_changeset *changeset)
629 struct extent_state *state;
630 struct extent_state *cached;
631 struct extent_state *prealloc = NULL;
632 struct rb_node *node;
637 btrfs_debug_check_extent_io_range(tree, start, end);
638 trace_btrfs_clear_extent_bit(tree, start, end - start + 1, bits);
640 if (bits & EXTENT_DELALLOC)
641 bits |= EXTENT_NORESERVE;
644 bits |= ~EXTENT_CTLBITS;
646 if (bits & (EXTENT_LOCKED | EXTENT_BOUNDARY))
649 if (!prealloc && gfpflags_allow_blocking(mask)) {
651 * Don't care for allocation failure here because we might end
652 * up not needing the pre-allocated extent state at all, which
653 * is the case if we only have in the tree extent states that
654 * cover our input range and don't cover too any other range.
655 * If we end up needing a new extent state we allocate it later.
657 prealloc = alloc_extent_state(mask);
660 spin_lock(&tree->lock);
662 cached = *cached_state;
665 *cached_state = NULL;
669 if (cached && extent_state_in_tree(cached) &&
670 cached->start <= start && cached->end > start) {
672 refcount_dec(&cached->refs);
677 free_extent_state(cached);
680 * this search will find the extents that end after
683 node = tree_search(tree, start);
686 state = rb_entry(node, struct extent_state, rb_node);
688 if (state->start > end)
690 WARN_ON(state->end < start);
691 last_end = state->end;
693 /* the state doesn't have the wanted bits, go ahead */
694 if (!(state->state & bits)) {
695 state = next_state(state);
700 * | ---- desired range ---- |
702 * | ------------- state -------------- |
704 * We need to split the extent we found, and may flip
705 * bits on second half.
707 * If the extent we found extends past our range, we
708 * just split and search again. It'll get split again
709 * the next time though.
711 * If the extent we found is inside our range, we clear
712 * the desired bit on it.
715 if (state->start < start) {
716 prealloc = alloc_extent_state_atomic(prealloc);
718 err = split_state(tree, state, prealloc, start);
720 extent_io_tree_panic(tree, err);
725 if (state->end <= end) {
726 state = clear_state_bit(tree, state, &bits, wake,
733 * | ---- desired range ---- |
735 * We need to split the extent, and clear the bit
738 if (state->start <= end && state->end > end) {
739 prealloc = alloc_extent_state_atomic(prealloc);
741 err = split_state(tree, state, prealloc, end + 1);
743 extent_io_tree_panic(tree, err);
748 clear_state_bit(tree, prealloc, &bits, wake, changeset);
754 state = clear_state_bit(tree, state, &bits, wake, changeset);
756 if (last_end == (u64)-1)
758 start = last_end + 1;
759 if (start <= end && state && !need_resched())
765 spin_unlock(&tree->lock);
766 if (gfpflags_allow_blocking(mask))
771 spin_unlock(&tree->lock);
773 free_extent_state(prealloc);
779 static void wait_on_state(struct extent_io_tree *tree,
780 struct extent_state *state)
781 __releases(tree->lock)
782 __acquires(tree->lock)
785 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
786 spin_unlock(&tree->lock);
788 spin_lock(&tree->lock);
789 finish_wait(&state->wq, &wait);
793 * waits for one or more bits to clear on a range in the state tree.
794 * The range [start, end] is inclusive.
795 * The tree lock is taken by this function
797 static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
800 struct extent_state *state;
801 struct rb_node *node;
803 btrfs_debug_check_extent_io_range(tree, start, end);
805 spin_lock(&tree->lock);
809 * this search will find all the extents that end after
812 node = tree_search(tree, start);
817 state = rb_entry(node, struct extent_state, rb_node);
819 if (state->start > end)
822 if (state->state & bits) {
823 start = state->start;
824 refcount_inc(&state->refs);
825 wait_on_state(tree, state);
826 free_extent_state(state);
829 start = state->end + 1;
834 if (!cond_resched_lock(&tree->lock)) {
835 node = rb_next(node);
840 spin_unlock(&tree->lock);
843 static void set_state_bits(struct extent_io_tree *tree,
844 struct extent_state *state,
845 unsigned *bits, struct extent_changeset *changeset)
847 unsigned bits_to_set = *bits & ~EXTENT_CTLBITS;
850 if (tree->private_data && is_data_inode(tree->private_data))
851 btrfs_set_delalloc_extent(tree->private_data, state, bits);
853 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
854 u64 range = state->end - state->start + 1;
855 tree->dirty_bytes += range;
857 ret = add_extent_changeset(state, bits_to_set, changeset, 1);
859 state->state |= bits_to_set;
862 static void cache_state_if_flags(struct extent_state *state,
863 struct extent_state **cached_ptr,
866 if (cached_ptr && !(*cached_ptr)) {
867 if (!flags || (state->state & flags)) {
869 refcount_inc(&state->refs);
874 static void cache_state(struct extent_state *state,
875 struct extent_state **cached_ptr)
877 return cache_state_if_flags(state, cached_ptr,
878 EXTENT_LOCKED | EXTENT_BOUNDARY);
882 * set some bits on a range in the tree. This may require allocations or
883 * sleeping, so the gfp mask is used to indicate what is allowed.
885 * If any of the exclusive bits are set, this will fail with -EEXIST if some
886 * part of the range already has the desired bits set. The start of the
887 * existing range is returned in failed_start in this case.
889 * [start, end] is inclusive This takes the tree lock.
892 static int __must_check
893 __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
894 unsigned bits, unsigned exclusive_bits,
895 u64 *failed_start, struct extent_state **cached_state,
896 gfp_t mask, struct extent_changeset *changeset)
898 struct extent_state *state;
899 struct extent_state *prealloc = NULL;
900 struct rb_node *node;
902 struct rb_node *parent;
907 btrfs_debug_check_extent_io_range(tree, start, end);
908 trace_btrfs_set_extent_bit(tree, start, end - start + 1, bits);
911 if (!prealloc && gfpflags_allow_blocking(mask)) {
913 * Don't care for allocation failure here because we might end
914 * up not needing the pre-allocated extent state at all, which
915 * is the case if we only have in the tree extent states that
916 * cover our input range and don't cover too any other range.
917 * If we end up needing a new extent state we allocate it later.
919 prealloc = alloc_extent_state(mask);
922 spin_lock(&tree->lock);
923 if (cached_state && *cached_state) {
924 state = *cached_state;
925 if (state->start <= start && state->end > start &&
926 extent_state_in_tree(state)) {
927 node = &state->rb_node;
932 * this search will find all the extents that end after
935 node = tree_search_for_insert(tree, start, &p, &parent);
937 prealloc = alloc_extent_state_atomic(prealloc);
939 err = insert_state(tree, prealloc, start, end,
940 &p, &parent, &bits, changeset);
942 extent_io_tree_panic(tree, err);
944 cache_state(prealloc, cached_state);
948 state = rb_entry(node, struct extent_state, rb_node);
950 last_start = state->start;
951 last_end = state->end;
954 * | ---- desired range ---- |
957 * Just lock what we found and keep going
959 if (state->start == start && state->end <= end) {
960 if (state->state & exclusive_bits) {
961 *failed_start = state->start;
966 set_state_bits(tree, state, &bits, changeset);
967 cache_state(state, cached_state);
968 merge_state(tree, state);
969 if (last_end == (u64)-1)
971 start = last_end + 1;
972 state = next_state(state);
973 if (start < end && state && state->start == start &&
980 * | ---- desired range ---- |
983 * | ------------- state -------------- |
985 * We need to split the extent we found, and may flip bits on
988 * If the extent we found extends past our
989 * range, we just split and search again. It'll get split
990 * again the next time though.
992 * If the extent we found is inside our range, we set the
995 if (state->start < start) {
996 if (state->state & exclusive_bits) {
997 *failed_start = start;
1002 prealloc = alloc_extent_state_atomic(prealloc);
1004 err = split_state(tree, state, prealloc, start);
1006 extent_io_tree_panic(tree, err);
1011 if (state->end <= end) {
1012 set_state_bits(tree, state, &bits, changeset);
1013 cache_state(state, cached_state);
1014 merge_state(tree, state);
1015 if (last_end == (u64)-1)
1017 start = last_end + 1;
1018 state = next_state(state);
1019 if (start < end && state && state->start == start &&
1026 * | ---- desired range ---- |
1027 * | state | or | state |
1029 * There's a hole, we need to insert something in it and
1030 * ignore the extent we found.
1032 if (state->start > start) {
1034 if (end < last_start)
1037 this_end = last_start - 1;
1039 prealloc = alloc_extent_state_atomic(prealloc);
1043 * Avoid to free 'prealloc' if it can be merged with
1046 err = insert_state(tree, prealloc, start, this_end,
1047 NULL, NULL, &bits, changeset);
1049 extent_io_tree_panic(tree, err);
1051 cache_state(prealloc, cached_state);
1053 start = this_end + 1;
1057 * | ---- desired range ---- |
1059 * We need to split the extent, and set the bit
1062 if (state->start <= end && state->end > end) {
1063 if (state->state & exclusive_bits) {
1064 *failed_start = start;
1069 prealloc = alloc_extent_state_atomic(prealloc);
1071 err = split_state(tree, state, prealloc, end + 1);
1073 extent_io_tree_panic(tree, err);
1075 set_state_bits(tree, prealloc, &bits, changeset);
1076 cache_state(prealloc, cached_state);
1077 merge_state(tree, prealloc);
1085 spin_unlock(&tree->lock);
1086 if (gfpflags_allow_blocking(mask))
1091 spin_unlock(&tree->lock);
1093 free_extent_state(prealloc);
1099 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1100 unsigned bits, u64 * failed_start,
1101 struct extent_state **cached_state, gfp_t mask)
1103 return __set_extent_bit(tree, start, end, bits, 0, failed_start,
1104 cached_state, mask, NULL);
1109 * convert_extent_bit - convert all bits in a given range from one bit to
1111 * @tree: the io tree to search
1112 * @start: the start offset in bytes
1113 * @end: the end offset in bytes (inclusive)
1114 * @bits: the bits to set in this range
1115 * @clear_bits: the bits to clear in this range
1116 * @cached_state: state that we're going to cache
1118 * This will go through and set bits for the given range. If any states exist
1119 * already in this range they are set with the given bit and cleared of the
1120 * clear_bits. This is only meant to be used by things that are mergeable, ie
1121 * converting from say DELALLOC to DIRTY. This is not meant to be used with
1122 * boundary bits like LOCK.
1124 * All allocations are done with GFP_NOFS.
1126 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1127 unsigned bits, unsigned clear_bits,
1128 struct extent_state **cached_state)
1130 struct extent_state *state;
1131 struct extent_state *prealloc = NULL;
1132 struct rb_node *node;
1134 struct rb_node *parent;
1138 bool first_iteration = true;
1140 btrfs_debug_check_extent_io_range(tree, start, end);
1141 trace_btrfs_convert_extent_bit(tree, start, end - start + 1, bits,
1147 * Best effort, don't worry if extent state allocation fails
1148 * here for the first iteration. We might have a cached state
1149 * that matches exactly the target range, in which case no
1150 * extent state allocations are needed. We'll only know this
1151 * after locking the tree.
1153 prealloc = alloc_extent_state(GFP_NOFS);
1154 if (!prealloc && !first_iteration)
1158 spin_lock(&tree->lock);
1159 if (cached_state && *cached_state) {
1160 state = *cached_state;
1161 if (state->start <= start && state->end > start &&
1162 extent_state_in_tree(state)) {
1163 node = &state->rb_node;
1169 * this search will find all the extents that end after
1172 node = tree_search_for_insert(tree, start, &p, &parent);
1174 prealloc = alloc_extent_state_atomic(prealloc);
1179 err = insert_state(tree, prealloc, start, end,
1180 &p, &parent, &bits, NULL);
1182 extent_io_tree_panic(tree, err);
1183 cache_state(prealloc, cached_state);
1187 state = rb_entry(node, struct extent_state, rb_node);
1189 last_start = state->start;
1190 last_end = state->end;
1193 * | ---- desired range ---- |
1196 * Just lock what we found and keep going
1198 if (state->start == start && state->end <= end) {
1199 set_state_bits(tree, state, &bits, NULL);
1200 cache_state(state, cached_state);
1201 state = clear_state_bit(tree, state, &clear_bits, 0, NULL);
1202 if (last_end == (u64)-1)
1204 start = last_end + 1;
1205 if (start < end && state && state->start == start &&
1212 * | ---- desired range ---- |
1215 * | ------------- state -------------- |
1217 * We need to split the extent we found, and may flip bits on
1220 * If the extent we found extends past our
1221 * range, we just split and search again. It'll get split
1222 * again the next time though.
1224 * If the extent we found is inside our range, we set the
1225 * desired bit on it.
1227 if (state->start < start) {
1228 prealloc = alloc_extent_state_atomic(prealloc);
1233 err = split_state(tree, state, prealloc, start);
1235 extent_io_tree_panic(tree, err);
1239 if (state->end <= end) {
1240 set_state_bits(tree, state, &bits, NULL);
1241 cache_state(state, cached_state);
1242 state = clear_state_bit(tree, state, &clear_bits, 0,
1244 if (last_end == (u64)-1)
1246 start = last_end + 1;
1247 if (start < end && state && state->start == start &&
1254 * | ---- desired range ---- |
1255 * | state | or | state |
1257 * There's a hole, we need to insert something in it and
1258 * ignore the extent we found.
1260 if (state->start > start) {
1262 if (end < last_start)
1265 this_end = last_start - 1;
1267 prealloc = alloc_extent_state_atomic(prealloc);
1274 * Avoid to free 'prealloc' if it can be merged with
1277 err = insert_state(tree, prealloc, start, this_end,
1278 NULL, NULL, &bits, NULL);
1280 extent_io_tree_panic(tree, err);
1281 cache_state(prealloc, cached_state);
1283 start = this_end + 1;
1287 * | ---- desired range ---- |
1289 * We need to split the extent, and set the bit
1292 if (state->start <= end && state->end > end) {
1293 prealloc = alloc_extent_state_atomic(prealloc);
1299 err = split_state(tree, state, prealloc, end + 1);
1301 extent_io_tree_panic(tree, err);
1303 set_state_bits(tree, prealloc, &bits, NULL);
1304 cache_state(prealloc, cached_state);
1305 clear_state_bit(tree, prealloc, &clear_bits, 0, NULL);
1313 spin_unlock(&tree->lock);
1315 first_iteration = false;
1319 spin_unlock(&tree->lock);
1321 free_extent_state(prealloc);
1326 /* wrappers around set/clear extent bit */
1327 int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1328 unsigned bits, struct extent_changeset *changeset)
1331 * We don't support EXTENT_LOCKED yet, as current changeset will
1332 * record any bits changed, so for EXTENT_LOCKED case, it will
1333 * either fail with -EEXIST or changeset will record the whole
1336 BUG_ON(bits & EXTENT_LOCKED);
1338 return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL, GFP_NOFS,
1342 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1343 unsigned bits, int wake, int delete,
1344 struct extent_state **cached)
1346 return __clear_extent_bit(tree, start, end, bits, wake, delete,
1347 cached, GFP_NOFS, NULL);
1350 int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1351 unsigned bits, struct extent_changeset *changeset)
1354 * Don't support EXTENT_LOCKED case, same reason as
1355 * set_record_extent_bits().
1357 BUG_ON(bits & EXTENT_LOCKED);
1359 return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, GFP_NOFS,
1364 * either insert or lock state struct between start and end use mask to tell
1365 * us if waiting is desired.
1367 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1368 struct extent_state **cached_state)
1374 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED,
1375 EXTENT_LOCKED, &failed_start,
1376 cached_state, GFP_NOFS, NULL);
1377 if (err == -EEXIST) {
1378 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1379 start = failed_start;
1382 WARN_ON(start > end);
1387 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1392 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1393 &failed_start, NULL, GFP_NOFS, NULL);
1394 if (err == -EEXIST) {
1395 if (failed_start > start)
1396 clear_extent_bit(tree, start, failed_start - 1,
1397 EXTENT_LOCKED, 1, 0, NULL);
1403 void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
1405 unsigned long index = start >> PAGE_SHIFT;
1406 unsigned long end_index = end >> PAGE_SHIFT;
1409 while (index <= end_index) {
1410 page = find_get_page(inode->i_mapping, index);
1411 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1412 clear_page_dirty_for_io(page);
1418 void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
1420 unsigned long index = start >> PAGE_SHIFT;
1421 unsigned long end_index = end >> PAGE_SHIFT;
1424 while (index <= end_index) {
1425 page = find_get_page(inode->i_mapping, index);
1426 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1427 __set_page_dirty_nobuffers(page);
1428 account_page_redirty(page);
1434 /* find the first state struct with 'bits' set after 'start', and
1435 * return it. tree->lock must be held. NULL will returned if
1436 * nothing was found after 'start'
1438 static struct extent_state *
1439 find_first_extent_bit_state(struct extent_io_tree *tree,
1440 u64 start, unsigned bits)
1442 struct rb_node *node;
1443 struct extent_state *state;
1446 * this search will find all the extents that end after
1449 node = tree_search(tree, start);
1454 state = rb_entry(node, struct extent_state, rb_node);
1455 if (state->end >= start && (state->state & bits))
1458 node = rb_next(node);
1467 * find the first offset in the io tree with 'bits' set. zero is
1468 * returned if we find something, and *start_ret and *end_ret are
1469 * set to reflect the state struct that was found.
1471 * If nothing was found, 1 is returned. If found something, return 0.
1473 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1474 u64 *start_ret, u64 *end_ret, unsigned bits,
1475 struct extent_state **cached_state)
1477 struct extent_state *state;
1480 spin_lock(&tree->lock);
1481 if (cached_state && *cached_state) {
1482 state = *cached_state;
1483 if (state->end == start - 1 && extent_state_in_tree(state)) {
1484 while ((state = next_state(state)) != NULL) {
1485 if (state->state & bits)
1488 free_extent_state(*cached_state);
1489 *cached_state = NULL;
1492 free_extent_state(*cached_state);
1493 *cached_state = NULL;
1496 state = find_first_extent_bit_state(tree, start, bits);
1499 cache_state_if_flags(state, cached_state, 0);
1500 *start_ret = state->start;
1501 *end_ret = state->end;
1505 spin_unlock(&tree->lock);
1510 * find a contiguous range of bytes in the file marked as delalloc, not
1511 * more than 'max_bytes'. start and end are used to return the range,
1513 * true is returned if we find something, false if nothing was in the tree
1515 static noinline bool find_delalloc_range(struct extent_io_tree *tree,
1516 u64 *start, u64 *end, u64 max_bytes,
1517 struct extent_state **cached_state)
1519 struct rb_node *node;
1520 struct extent_state *state;
1521 u64 cur_start = *start;
1523 u64 total_bytes = 0;
1525 spin_lock(&tree->lock);
1528 * this search will find all the extents that end after
1531 node = tree_search(tree, cur_start);
1538 state = rb_entry(node, struct extent_state, rb_node);
1539 if (found && (state->start != cur_start ||
1540 (state->state & EXTENT_BOUNDARY))) {
1543 if (!(state->state & EXTENT_DELALLOC)) {
1549 *start = state->start;
1550 *cached_state = state;
1551 refcount_inc(&state->refs);
1555 cur_start = state->end + 1;
1556 node = rb_next(node);
1557 total_bytes += state->end - state->start + 1;
1558 if (total_bytes >= max_bytes)
1564 spin_unlock(&tree->lock);
1568 static int __process_pages_contig(struct address_space *mapping,
1569 struct page *locked_page,
1570 pgoff_t start_index, pgoff_t end_index,
1571 unsigned long page_ops, pgoff_t *index_ret);
1573 static noinline void __unlock_for_delalloc(struct inode *inode,
1574 struct page *locked_page,
1577 unsigned long index = start >> PAGE_SHIFT;
1578 unsigned long end_index = end >> PAGE_SHIFT;
1580 ASSERT(locked_page);
1581 if (index == locked_page->index && end_index == index)
1584 __process_pages_contig(inode->i_mapping, locked_page, index, end_index,
1588 static noinline int lock_delalloc_pages(struct inode *inode,
1589 struct page *locked_page,
1593 unsigned long index = delalloc_start >> PAGE_SHIFT;
1594 unsigned long index_ret = index;
1595 unsigned long end_index = delalloc_end >> PAGE_SHIFT;
1598 ASSERT(locked_page);
1599 if (index == locked_page->index && index == end_index)
1602 ret = __process_pages_contig(inode->i_mapping, locked_page, index,
1603 end_index, PAGE_LOCK, &index_ret);
1605 __unlock_for_delalloc(inode, locked_page, delalloc_start,
1606 (u64)index_ret << PAGE_SHIFT);
1611 * Find and lock a contiguous range of bytes in the file marked as delalloc, no
1612 * more than @max_bytes. @Start and @end are used to return the range,
1614 * Return: true if we find something
1615 * false if nothing was in the tree
1618 noinline_for_stack bool find_lock_delalloc_range(struct inode *inode,
1619 struct extent_io_tree *tree,
1620 struct page *locked_page, u64 *start,
1623 u64 max_bytes = BTRFS_MAX_EXTENT_SIZE;
1627 struct extent_state *cached_state = NULL;
1632 /* step one, find a bunch of delalloc bytes starting at start */
1633 delalloc_start = *start;
1635 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1636 max_bytes, &cached_state);
1637 if (!found || delalloc_end <= *start) {
1638 *start = delalloc_start;
1639 *end = delalloc_end;
1640 free_extent_state(cached_state);
1645 * start comes from the offset of locked_page. We have to lock
1646 * pages in order, so we can't process delalloc bytes before
1649 if (delalloc_start < *start)
1650 delalloc_start = *start;
1653 * make sure to limit the number of pages we try to lock down
1655 if (delalloc_end + 1 - delalloc_start > max_bytes)
1656 delalloc_end = delalloc_start + max_bytes - 1;
1658 /* step two, lock all the pages after the page that has start */
1659 ret = lock_delalloc_pages(inode, locked_page,
1660 delalloc_start, delalloc_end);
1661 ASSERT(!ret || ret == -EAGAIN);
1662 if (ret == -EAGAIN) {
1663 /* some of the pages are gone, lets avoid looping by
1664 * shortening the size of the delalloc range we're searching
1666 free_extent_state(cached_state);
1667 cached_state = NULL;
1669 max_bytes = PAGE_SIZE;
1678 /* step three, lock the state bits for the whole range */
1679 lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state);
1681 /* then test to make sure it is all still delalloc */
1682 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1683 EXTENT_DELALLOC, 1, cached_state);
1685 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1687 __unlock_for_delalloc(inode, locked_page,
1688 delalloc_start, delalloc_end);
1692 free_extent_state(cached_state);
1693 *start = delalloc_start;
1694 *end = delalloc_end;
1699 static int __process_pages_contig(struct address_space *mapping,
1700 struct page *locked_page,
1701 pgoff_t start_index, pgoff_t end_index,
1702 unsigned long page_ops, pgoff_t *index_ret)
1704 unsigned long nr_pages = end_index - start_index + 1;
1705 unsigned long pages_locked = 0;
1706 pgoff_t index = start_index;
1707 struct page *pages[16];
1712 if (page_ops & PAGE_LOCK) {
1713 ASSERT(page_ops == PAGE_LOCK);
1714 ASSERT(index_ret && *index_ret == start_index);
1717 if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0)
1718 mapping_set_error(mapping, -EIO);
1720 while (nr_pages > 0) {
1721 ret = find_get_pages_contig(mapping, index,
1722 min_t(unsigned long,
1723 nr_pages, ARRAY_SIZE(pages)), pages);
1726 * Only if we're going to lock these pages,
1727 * can we find nothing at @index.
1729 ASSERT(page_ops & PAGE_LOCK);
1734 for (i = 0; i < ret; i++) {
1735 if (page_ops & PAGE_SET_PRIVATE2)
1736 SetPagePrivate2(pages[i]);
1738 if (pages[i] == locked_page) {
1743 if (page_ops & PAGE_CLEAR_DIRTY)
1744 clear_page_dirty_for_io(pages[i]);
1745 if (page_ops & PAGE_SET_WRITEBACK)
1746 set_page_writeback(pages[i]);
1747 if (page_ops & PAGE_SET_ERROR)
1748 SetPageError(pages[i]);
1749 if (page_ops & PAGE_END_WRITEBACK)
1750 end_page_writeback(pages[i]);
1751 if (page_ops & PAGE_UNLOCK)
1752 unlock_page(pages[i]);
1753 if (page_ops & PAGE_LOCK) {
1754 lock_page(pages[i]);
1755 if (!PageDirty(pages[i]) ||
1756 pages[i]->mapping != mapping) {
1757 unlock_page(pages[i]);
1771 if (err && index_ret)
1772 *index_ret = start_index + pages_locked - 1;
1776 void extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end,
1777 u64 delalloc_end, struct page *locked_page,
1778 unsigned clear_bits,
1779 unsigned long page_ops)
1781 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, clear_bits, 1, 0,
1784 __process_pages_contig(inode->i_mapping, locked_page,
1785 start >> PAGE_SHIFT, end >> PAGE_SHIFT,
1790 * count the number of bytes in the tree that have a given bit(s)
1791 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1792 * cached. The total number found is returned.
1794 u64 count_range_bits(struct extent_io_tree *tree,
1795 u64 *start, u64 search_end, u64 max_bytes,
1796 unsigned bits, int contig)
1798 struct rb_node *node;
1799 struct extent_state *state;
1800 u64 cur_start = *start;
1801 u64 total_bytes = 0;
1805 if (WARN_ON(search_end <= cur_start))
1808 spin_lock(&tree->lock);
1809 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1810 total_bytes = tree->dirty_bytes;
1814 * this search will find all the extents that end after
1817 node = tree_search(tree, cur_start);
1822 state = rb_entry(node, struct extent_state, rb_node);
1823 if (state->start > search_end)
1825 if (contig && found && state->start > last + 1)
1827 if (state->end >= cur_start && (state->state & bits) == bits) {
1828 total_bytes += min(search_end, state->end) + 1 -
1829 max(cur_start, state->start);
1830 if (total_bytes >= max_bytes)
1833 *start = max(cur_start, state->start);
1837 } else if (contig && found) {
1840 node = rb_next(node);
1845 spin_unlock(&tree->lock);
1850 * set the private field for a given byte offset in the tree. If there isn't
1851 * an extent_state there already, this does nothing.
1853 static noinline int set_state_failrec(struct extent_io_tree *tree, u64 start,
1854 struct io_failure_record *failrec)
1856 struct rb_node *node;
1857 struct extent_state *state;
1860 spin_lock(&tree->lock);
1862 * this search will find all the extents that end after
1865 node = tree_search(tree, start);
1870 state = rb_entry(node, struct extent_state, rb_node);
1871 if (state->start != start) {
1875 state->failrec = failrec;
1877 spin_unlock(&tree->lock);
1881 static noinline int get_state_failrec(struct extent_io_tree *tree, u64 start,
1882 struct io_failure_record **failrec)
1884 struct rb_node *node;
1885 struct extent_state *state;
1888 spin_lock(&tree->lock);
1890 * this search will find all the extents that end after
1893 node = tree_search(tree, start);
1898 state = rb_entry(node, struct extent_state, rb_node);
1899 if (state->start != start) {
1903 *failrec = state->failrec;
1905 spin_unlock(&tree->lock);
1910 * searches a range in the state tree for a given mask.
1911 * If 'filled' == 1, this returns 1 only if every extent in the tree
1912 * has the bits set. Otherwise, 1 is returned if any bit in the
1913 * range is found set.
1915 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1916 unsigned bits, int filled, struct extent_state *cached)
1918 struct extent_state *state = NULL;
1919 struct rb_node *node;
1922 spin_lock(&tree->lock);
1923 if (cached && extent_state_in_tree(cached) && cached->start <= start &&
1924 cached->end > start)
1925 node = &cached->rb_node;
1927 node = tree_search(tree, start);
1928 while (node && start <= end) {
1929 state = rb_entry(node, struct extent_state, rb_node);
1931 if (filled && state->start > start) {
1936 if (state->start > end)
1939 if (state->state & bits) {
1943 } else if (filled) {
1948 if (state->end == (u64)-1)
1951 start = state->end + 1;
1954 node = rb_next(node);
1961 spin_unlock(&tree->lock);
1966 * helper function to set a given page up to date if all the
1967 * extents in the tree for that page are up to date
1969 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
1971 u64 start = page_offset(page);
1972 u64 end = start + PAGE_SIZE - 1;
1973 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1974 SetPageUptodate(page);
1977 int free_io_failure(struct extent_io_tree *failure_tree,
1978 struct extent_io_tree *io_tree,
1979 struct io_failure_record *rec)
1984 set_state_failrec(failure_tree, rec->start, NULL);
1985 ret = clear_extent_bits(failure_tree, rec->start,
1986 rec->start + rec->len - 1,
1987 EXTENT_LOCKED | EXTENT_DIRTY);
1991 ret = clear_extent_bits(io_tree, rec->start,
1992 rec->start + rec->len - 1,
2002 * this bypasses the standard btrfs submit functions deliberately, as
2003 * the standard behavior is to write all copies in a raid setup. here we only
2004 * want to write the one bad copy. so we do the mapping for ourselves and issue
2005 * submit_bio directly.
2006 * to avoid any synchronization issues, wait for the data after writing, which
2007 * actually prevents the read that triggered the error from finishing.
2008 * currently, there can be no more than two copies of every data bit. thus,
2009 * exactly one rewrite is required.
2011 int repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start,
2012 u64 length, u64 logical, struct page *page,
2013 unsigned int pg_offset, int mirror_num)
2016 struct btrfs_device *dev;
2019 struct btrfs_bio *bbio = NULL;
2022 ASSERT(!(fs_info->sb->s_flags & SB_RDONLY));
2023 BUG_ON(!mirror_num);
2025 bio = btrfs_io_bio_alloc(1);
2026 bio->bi_iter.bi_size = 0;
2027 map_length = length;
2030 * Avoid races with device replace and make sure our bbio has devices
2031 * associated to its stripes that don't go away while we are doing the
2032 * read repair operation.
2034 btrfs_bio_counter_inc_blocked(fs_info);
2035 if (btrfs_is_parity_mirror(fs_info, logical, length)) {
2037 * Note that we don't use BTRFS_MAP_WRITE because it's supposed
2038 * to update all raid stripes, but here we just want to correct
2039 * bad stripe, thus BTRFS_MAP_READ is abused to only get the bad
2040 * stripe's dev and sector.
2042 ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, logical,
2043 &map_length, &bbio, 0);
2045 btrfs_bio_counter_dec(fs_info);
2049 ASSERT(bbio->mirror_num == 1);
2051 ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical,
2052 &map_length, &bbio, mirror_num);
2054 btrfs_bio_counter_dec(fs_info);
2058 BUG_ON(mirror_num != bbio->mirror_num);
2061 sector = bbio->stripes[bbio->mirror_num - 1].physical >> 9;
2062 bio->bi_iter.bi_sector = sector;
2063 dev = bbio->stripes[bbio->mirror_num - 1].dev;
2064 btrfs_put_bbio(bbio);
2065 if (!dev || !dev->bdev ||
2066 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) {
2067 btrfs_bio_counter_dec(fs_info);
2071 bio_set_dev(bio, dev->bdev);
2072 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC;
2073 bio_add_page(bio, page, length, pg_offset);
2075 if (btrfsic_submit_bio_wait(bio)) {
2076 /* try to remap that extent elsewhere? */
2077 btrfs_bio_counter_dec(fs_info);
2079 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
2083 btrfs_info_rl_in_rcu(fs_info,
2084 "read error corrected: ino %llu off %llu (dev %s sector %llu)",
2086 rcu_str_deref(dev->name), sector);
2087 btrfs_bio_counter_dec(fs_info);
2092 int repair_eb_io_failure(struct btrfs_fs_info *fs_info,
2093 struct extent_buffer *eb, int mirror_num)
2095 u64 start = eb->start;
2096 int i, num_pages = num_extent_pages(eb);
2099 if (sb_rdonly(fs_info->sb))
2102 for (i = 0; i < num_pages; i++) {
2103 struct page *p = eb->pages[i];
2105 ret = repair_io_failure(fs_info, 0, start, PAGE_SIZE, start, p,
2106 start - page_offset(p), mirror_num);
2116 * each time an IO finishes, we do a fast check in the IO failure tree
2117 * to see if we need to process or clean up an io_failure_record
2119 int clean_io_failure(struct btrfs_fs_info *fs_info,
2120 struct extent_io_tree *failure_tree,
2121 struct extent_io_tree *io_tree, u64 start,
2122 struct page *page, u64 ino, unsigned int pg_offset)
2125 struct io_failure_record *failrec;
2126 struct extent_state *state;
2131 ret = count_range_bits(failure_tree, &private, (u64)-1, 1,
2136 ret = get_state_failrec(failure_tree, start, &failrec);
2140 BUG_ON(!failrec->this_mirror);
2142 if (failrec->in_validation) {
2143 /* there was no real error, just free the record */
2144 btrfs_debug(fs_info,
2145 "clean_io_failure: freeing dummy error at %llu",
2149 if (sb_rdonly(fs_info->sb))
2152 spin_lock(&io_tree->lock);
2153 state = find_first_extent_bit_state(io_tree,
2156 spin_unlock(&io_tree->lock);
2158 if (state && state->start <= failrec->start &&
2159 state->end >= failrec->start + failrec->len - 1) {
2160 num_copies = btrfs_num_copies(fs_info, failrec->logical,
2162 if (num_copies > 1) {
2163 repair_io_failure(fs_info, ino, start, failrec->len,
2164 failrec->logical, page, pg_offset,
2165 failrec->failed_mirror);
2170 free_io_failure(failure_tree, io_tree, failrec);
2176 * Can be called when
2177 * - hold extent lock
2178 * - under ordered extent
2179 * - the inode is freeing
2181 void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start, u64 end)
2183 struct extent_io_tree *failure_tree = &inode->io_failure_tree;
2184 struct io_failure_record *failrec;
2185 struct extent_state *state, *next;
2187 if (RB_EMPTY_ROOT(&failure_tree->state))
2190 spin_lock(&failure_tree->lock);
2191 state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY);
2193 if (state->start > end)
2196 ASSERT(state->end <= end);
2198 next = next_state(state);
2200 failrec = state->failrec;
2201 free_extent_state(state);
2206 spin_unlock(&failure_tree->lock);
2209 int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end,
2210 struct io_failure_record **failrec_ret)
2212 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2213 struct io_failure_record *failrec;
2214 struct extent_map *em;
2215 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2216 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2217 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2221 ret = get_state_failrec(failure_tree, start, &failrec);
2223 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2227 failrec->start = start;
2228 failrec->len = end - start + 1;
2229 failrec->this_mirror = 0;
2230 failrec->bio_flags = 0;
2231 failrec->in_validation = 0;
2233 read_lock(&em_tree->lock);
2234 em = lookup_extent_mapping(em_tree, start, failrec->len);
2236 read_unlock(&em_tree->lock);
2241 if (em->start > start || em->start + em->len <= start) {
2242 free_extent_map(em);
2245 read_unlock(&em_tree->lock);
2251 logical = start - em->start;
2252 logical = em->block_start + logical;
2253 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2254 logical = em->block_start;
2255 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2256 extent_set_compress_type(&failrec->bio_flags,
2260 btrfs_debug(fs_info,
2261 "Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu",
2262 logical, start, failrec->len);
2264 failrec->logical = logical;
2265 free_extent_map(em);
2267 /* set the bits in the private failure tree */
2268 ret = set_extent_bits(failure_tree, start, end,
2269 EXTENT_LOCKED | EXTENT_DIRTY);
2271 ret = set_state_failrec(failure_tree, start, failrec);
2272 /* set the bits in the inode's tree */
2274 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED);
2280 btrfs_debug(fs_info,
2281 "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d",
2282 failrec->logical, failrec->start, failrec->len,
2283 failrec->in_validation);
2285 * when data can be on disk more than twice, add to failrec here
2286 * (e.g. with a list for failed_mirror) to make
2287 * clean_io_failure() clean all those errors at once.
2291 *failrec_ret = failrec;
2296 bool btrfs_check_repairable(struct inode *inode, unsigned failed_bio_pages,
2297 struct io_failure_record *failrec, int failed_mirror)
2299 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2302 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
2303 if (num_copies == 1) {
2305 * we only have a single copy of the data, so don't bother with
2306 * all the retry and error correction code that follows. no
2307 * matter what the error is, it is very likely to persist.
2309 btrfs_debug(fs_info,
2310 "Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
2311 num_copies, failrec->this_mirror, failed_mirror);
2316 * there are two premises:
2317 * a) deliver good data to the caller
2318 * b) correct the bad sectors on disk
2320 if (failed_bio_pages > 1) {
2322 * to fulfill b), we need to know the exact failing sectors, as
2323 * we don't want to rewrite any more than the failed ones. thus,
2324 * we need separate read requests for the failed bio
2326 * if the following BUG_ON triggers, our validation request got
2327 * merged. we need separate requests for our algorithm to work.
2329 BUG_ON(failrec->in_validation);
2330 failrec->in_validation = 1;
2331 failrec->this_mirror = failed_mirror;
2334 * we're ready to fulfill a) and b) alongside. get a good copy
2335 * of the failed sector and if we succeed, we have setup
2336 * everything for repair_io_failure to do the rest for us.
2338 if (failrec->in_validation) {
2339 BUG_ON(failrec->this_mirror != failed_mirror);
2340 failrec->in_validation = 0;
2341 failrec->this_mirror = 0;
2343 failrec->failed_mirror = failed_mirror;
2344 failrec->this_mirror++;
2345 if (failrec->this_mirror == failed_mirror)
2346 failrec->this_mirror++;
2349 if (failrec->this_mirror > num_copies) {
2350 btrfs_debug(fs_info,
2351 "Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
2352 num_copies, failrec->this_mirror, failed_mirror);
2360 struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio,
2361 struct io_failure_record *failrec,
2362 struct page *page, int pg_offset, int icsum,
2363 bio_end_io_t *endio_func, void *data)
2365 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2367 struct btrfs_io_bio *btrfs_failed_bio;
2368 struct btrfs_io_bio *btrfs_bio;
2370 bio = btrfs_io_bio_alloc(1);
2371 bio->bi_end_io = endio_func;
2372 bio->bi_iter.bi_sector = failrec->logical >> 9;
2373 bio_set_dev(bio, fs_info->fs_devices->latest_bdev);
2374 bio->bi_iter.bi_size = 0;
2375 bio->bi_private = data;
2377 btrfs_failed_bio = btrfs_io_bio(failed_bio);
2378 if (btrfs_failed_bio->csum) {
2379 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
2381 btrfs_bio = btrfs_io_bio(bio);
2382 btrfs_bio->csum = btrfs_bio->csum_inline;
2384 memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + icsum,
2388 bio_add_page(bio, page, failrec->len, pg_offset);
2394 * This is a generic handler for readpage errors. If other copies exist, read
2395 * those and write back good data to the failed position. Does not investigate
2396 * in remapping the failed extent elsewhere, hoping the device will be smart
2397 * enough to do this as needed
2399 static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset,
2400 struct page *page, u64 start, u64 end,
2403 struct io_failure_record *failrec;
2404 struct inode *inode = page->mapping->host;
2405 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2406 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2409 blk_status_t status;
2411 unsigned failed_bio_pages = failed_bio->bi_iter.bi_size >> PAGE_SHIFT;
2413 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
2415 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
2419 if (!btrfs_check_repairable(inode, failed_bio_pages, failrec,
2421 free_io_failure(failure_tree, tree, failrec);
2425 if (failed_bio_pages > 1)
2426 read_mode |= REQ_FAILFAST_DEV;
2428 phy_offset >>= inode->i_sb->s_blocksize_bits;
2429 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
2430 start - page_offset(page),
2431 (int)phy_offset, failed_bio->bi_end_io,
2433 bio->bi_opf = REQ_OP_READ | read_mode;
2435 btrfs_debug(btrfs_sb(inode->i_sb),
2436 "Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d",
2437 read_mode, failrec->this_mirror, failrec->in_validation);
2439 status = tree->ops->submit_bio_hook(tree->private_data, bio, failrec->this_mirror,
2440 failrec->bio_flags, 0);
2442 free_io_failure(failure_tree, tree, failrec);
2444 ret = blk_status_to_errno(status);
2450 /* lots and lots of room for performance fixes in the end_bio funcs */
2452 void end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2454 int uptodate = (err == 0);
2457 btrfs_writepage_endio_finish_ordered(page, start, end, uptodate);
2460 ClearPageUptodate(page);
2462 ret = err < 0 ? err : -EIO;
2463 mapping_set_error(page->mapping, ret);
2468 * after a writepage IO is done, we need to:
2469 * clear the uptodate bits on error
2470 * clear the writeback bits in the extent tree for this IO
2471 * end_page_writeback if the page has no more pending IO
2473 * Scheduling is not allowed, so the extent state tree is expected
2474 * to have one and only one object corresponding to this IO.
2476 static void end_bio_extent_writepage(struct bio *bio)
2478 int error = blk_status_to_errno(bio->bi_status);
2479 struct bio_vec *bvec;
2483 struct bvec_iter_all iter_all;
2485 ASSERT(!bio_flagged(bio, BIO_CLONED));
2486 bio_for_each_segment_all(bvec, bio, i, iter_all) {
2487 struct page *page = bvec->bv_page;
2488 struct inode *inode = page->mapping->host;
2489 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2491 /* We always issue full-page reads, but if some block
2492 * in a page fails to read, blk_update_request() will
2493 * advance bv_offset and adjust bv_len to compensate.
2494 * Print a warning for nonzero offsets, and an error
2495 * if they don't add up to a full page. */
2496 if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) {
2497 if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE)
2499 "partial page write in btrfs with offset %u and length %u",
2500 bvec->bv_offset, bvec->bv_len);
2503 "incomplete page write in btrfs with offset %u and length %u",
2504 bvec->bv_offset, bvec->bv_len);
2507 start = page_offset(page);
2508 end = start + bvec->bv_offset + bvec->bv_len - 1;
2510 end_extent_writepage(page, error, start, end);
2511 end_page_writeback(page);
2518 endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len,
2521 struct extent_state *cached = NULL;
2522 u64 end = start + len - 1;
2524 if (uptodate && tree->track_uptodate)
2525 set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC);
2526 unlock_extent_cached_atomic(tree, start, end, &cached);
2530 * after a readpage IO is done, we need to:
2531 * clear the uptodate bits on error
2532 * set the uptodate bits if things worked
2533 * set the page up to date if all extents in the tree are uptodate
2534 * clear the lock bit in the extent tree
2535 * unlock the page if there are no other extents locked for it
2537 * Scheduling is not allowed, so the extent state tree is expected
2538 * to have one and only one object corresponding to this IO.
2540 static void end_bio_extent_readpage(struct bio *bio)
2542 struct bio_vec *bvec;
2543 int uptodate = !bio->bi_status;
2544 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
2545 struct extent_io_tree *tree, *failure_tree;
2550 u64 extent_start = 0;
2555 struct bvec_iter_all iter_all;
2557 ASSERT(!bio_flagged(bio, BIO_CLONED));
2558 bio_for_each_segment_all(bvec, bio, i, iter_all) {
2559 struct page *page = bvec->bv_page;
2560 struct inode *inode = page->mapping->host;
2561 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2562 bool data_inode = btrfs_ino(BTRFS_I(inode))
2563 != BTRFS_BTREE_INODE_OBJECTID;
2565 btrfs_debug(fs_info,
2566 "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u",
2567 (u64)bio->bi_iter.bi_sector, bio->bi_status,
2568 io_bio->mirror_num);
2569 tree = &BTRFS_I(inode)->io_tree;
2570 failure_tree = &BTRFS_I(inode)->io_failure_tree;
2572 /* We always issue full-page reads, but if some block
2573 * in a page fails to read, blk_update_request() will
2574 * advance bv_offset and adjust bv_len to compensate.
2575 * Print a warning for nonzero offsets, and an error
2576 * if they don't add up to a full page. */
2577 if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) {
2578 if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE)
2580 "partial page read in btrfs with offset %u and length %u",
2581 bvec->bv_offset, bvec->bv_len);
2584 "incomplete page read in btrfs with offset %u and length %u",
2585 bvec->bv_offset, bvec->bv_len);
2588 start = page_offset(page);
2589 end = start + bvec->bv_offset + bvec->bv_len - 1;
2592 mirror = io_bio->mirror_num;
2593 if (likely(uptodate)) {
2594 ret = tree->ops->readpage_end_io_hook(io_bio, offset,
2600 clean_io_failure(BTRFS_I(inode)->root->fs_info,
2601 failure_tree, tree, start,
2603 btrfs_ino(BTRFS_I(inode)), 0);
2606 if (likely(uptodate))
2612 * The generic bio_readpage_error handles errors the
2613 * following way: If possible, new read requests are
2614 * created and submitted and will end up in
2615 * end_bio_extent_readpage as well (if we're lucky,
2616 * not in the !uptodate case). In that case it returns
2617 * 0 and we just go on with the next page in our bio.
2618 * If it can't handle the error it will return -EIO and
2619 * we remain responsible for that page.
2621 ret = bio_readpage_error(bio, offset, page, start, end,
2624 uptodate = !bio->bi_status;
2629 struct extent_buffer *eb;
2631 eb = (struct extent_buffer *)page->private;
2632 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
2633 eb->read_mirror = mirror;
2634 atomic_dec(&eb->io_pages);
2635 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD,
2637 btree_readahead_hook(eb, -EIO);
2640 if (likely(uptodate)) {
2641 loff_t i_size = i_size_read(inode);
2642 pgoff_t end_index = i_size >> PAGE_SHIFT;
2645 /* Zero out the end if this page straddles i_size */
2646 off = offset_in_page(i_size);
2647 if (page->index == end_index && off)
2648 zero_user_segment(page, off, PAGE_SIZE);
2649 SetPageUptodate(page);
2651 ClearPageUptodate(page);
2657 if (unlikely(!uptodate)) {
2659 endio_readpage_release_extent(tree,
2665 endio_readpage_release_extent(tree, start,
2666 end - start + 1, 0);
2667 } else if (!extent_len) {
2668 extent_start = start;
2669 extent_len = end + 1 - start;
2670 } else if (extent_start + extent_len == start) {
2671 extent_len += end + 1 - start;
2673 endio_readpage_release_extent(tree, extent_start,
2674 extent_len, uptodate);
2675 extent_start = start;
2676 extent_len = end + 1 - start;
2681 endio_readpage_release_extent(tree, extent_start, extent_len,
2683 btrfs_io_bio_free_csum(io_bio);
2688 * Initialize the members up to but not including 'bio'. Use after allocating a
2689 * new bio by bio_alloc_bioset as it does not initialize the bytes outside of
2690 * 'bio' because use of __GFP_ZERO is not supported.
2692 static inline void btrfs_io_bio_init(struct btrfs_io_bio *btrfs_bio)
2694 memset(btrfs_bio, 0, offsetof(struct btrfs_io_bio, bio));
2698 * The following helpers allocate a bio. As it's backed by a bioset, it'll
2699 * never fail. We're returning a bio right now but you can call btrfs_io_bio
2700 * for the appropriate container_of magic
2702 struct bio *btrfs_bio_alloc(struct block_device *bdev, u64 first_byte)
2706 bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, &btrfs_bioset);
2707 bio_set_dev(bio, bdev);
2708 bio->bi_iter.bi_sector = first_byte >> 9;
2709 btrfs_io_bio_init(btrfs_io_bio(bio));
2713 struct bio *btrfs_bio_clone(struct bio *bio)
2715 struct btrfs_io_bio *btrfs_bio;
2718 /* Bio allocation backed by a bioset does not fail */
2719 new = bio_clone_fast(bio, GFP_NOFS, &btrfs_bioset);
2720 btrfs_bio = btrfs_io_bio(new);
2721 btrfs_io_bio_init(btrfs_bio);
2722 btrfs_bio->iter = bio->bi_iter;
2726 struct bio *btrfs_io_bio_alloc(unsigned int nr_iovecs)
2730 /* Bio allocation backed by a bioset does not fail */
2731 bio = bio_alloc_bioset(GFP_NOFS, nr_iovecs, &btrfs_bioset);
2732 btrfs_io_bio_init(btrfs_io_bio(bio));
2736 struct bio *btrfs_bio_clone_partial(struct bio *orig, int offset, int size)
2739 struct btrfs_io_bio *btrfs_bio;
2741 /* this will never fail when it's backed by a bioset */
2742 bio = bio_clone_fast(orig, GFP_NOFS, &btrfs_bioset);
2745 btrfs_bio = btrfs_io_bio(bio);
2746 btrfs_io_bio_init(btrfs_bio);
2748 bio_trim(bio, offset >> 9, size >> 9);
2749 btrfs_bio->iter = bio->bi_iter;
2754 * @opf: bio REQ_OP_* and REQ_* flags as one value
2755 * @tree: tree so we can call our merge_bio hook
2756 * @wbc: optional writeback control for io accounting
2757 * @page: page to add to the bio
2758 * @pg_offset: offset of the new bio or to check whether we are adding
2759 * a contiguous page to the previous one
2760 * @size: portion of page that we want to write
2761 * @offset: starting offset in the page
2762 * @bdev: attach newly created bios to this bdev
2763 * @bio_ret: must be valid pointer, newly allocated bio will be stored there
2764 * @end_io_func: end_io callback for new bio
2765 * @mirror_num: desired mirror to read/write
2766 * @prev_bio_flags: flags of previous bio to see if we can merge the current one
2767 * @bio_flags: flags of the current bio to see if we can merge them
2769 static int submit_extent_page(unsigned int opf, struct extent_io_tree *tree,
2770 struct writeback_control *wbc,
2771 struct page *page, u64 offset,
2772 size_t size, unsigned long pg_offset,
2773 struct block_device *bdev,
2774 struct bio **bio_ret,
2775 bio_end_io_t end_io_func,
2777 unsigned long prev_bio_flags,
2778 unsigned long bio_flags,
2779 bool force_bio_submit)
2783 size_t page_size = min_t(size_t, size, PAGE_SIZE);
2784 sector_t sector = offset >> 9;
2790 bool can_merge = true;
2793 if (prev_bio_flags & EXTENT_BIO_COMPRESSED)
2794 contig = bio->bi_iter.bi_sector == sector;
2796 contig = bio_end_sector(bio) == sector;
2799 if (btrfs_bio_fits_in_stripe(page, page_size, bio, bio_flags))
2802 if (prev_bio_flags != bio_flags || !contig || !can_merge ||
2804 bio_add_page(bio, page, page_size, pg_offset) < page_size) {
2805 ret = submit_one_bio(bio, mirror_num, prev_bio_flags);
2813 wbc_account_io(wbc, page, page_size);
2818 bio = btrfs_bio_alloc(bdev, offset);
2819 bio_add_page(bio, page, page_size, pg_offset);
2820 bio->bi_end_io = end_io_func;
2821 bio->bi_private = tree;
2822 bio->bi_write_hint = page->mapping->host->i_write_hint;
2825 wbc_init_bio(wbc, bio);
2826 wbc_account_io(wbc, page, page_size);
2834 static void attach_extent_buffer_page(struct extent_buffer *eb,
2837 if (!PagePrivate(page)) {
2838 SetPagePrivate(page);
2840 set_page_private(page, (unsigned long)eb);
2842 WARN_ON(page->private != (unsigned long)eb);
2846 void set_page_extent_mapped(struct page *page)
2848 if (!PagePrivate(page)) {
2849 SetPagePrivate(page);
2851 set_page_private(page, EXTENT_PAGE_PRIVATE);
2855 static struct extent_map *
2856 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
2857 u64 start, u64 len, get_extent_t *get_extent,
2858 struct extent_map **em_cached)
2860 struct extent_map *em;
2862 if (em_cached && *em_cached) {
2864 if (extent_map_in_tree(em) && start >= em->start &&
2865 start < extent_map_end(em)) {
2866 refcount_inc(&em->refs);
2870 free_extent_map(em);
2874 em = get_extent(BTRFS_I(inode), page, pg_offset, start, len, 0);
2875 if (em_cached && !IS_ERR_OR_NULL(em)) {
2877 refcount_inc(&em->refs);
2883 * basic readpage implementation. Locked extent state structs are inserted
2884 * into the tree that are removed when the IO is done (by the end_io
2886 * XXX JDM: This needs looking at to ensure proper page locking
2887 * return 0 on success, otherwise return error
2889 static int __do_readpage(struct extent_io_tree *tree,
2891 get_extent_t *get_extent,
2892 struct extent_map **em_cached,
2893 struct bio **bio, int mirror_num,
2894 unsigned long *bio_flags, unsigned int read_flags,
2897 struct inode *inode = page->mapping->host;
2898 u64 start = page_offset(page);
2899 const u64 end = start + PAGE_SIZE - 1;
2902 u64 last_byte = i_size_read(inode);
2905 struct extent_map *em;
2906 struct block_device *bdev;
2909 size_t pg_offset = 0;
2911 size_t disk_io_size;
2912 size_t blocksize = inode->i_sb->s_blocksize;
2913 unsigned long this_bio_flag = 0;
2915 set_page_extent_mapped(page);
2917 if (!PageUptodate(page)) {
2918 if (cleancache_get_page(page) == 0) {
2919 BUG_ON(blocksize != PAGE_SIZE);
2920 unlock_extent(tree, start, end);
2925 if (page->index == last_byte >> PAGE_SHIFT) {
2927 size_t zero_offset = offset_in_page(last_byte);
2930 iosize = PAGE_SIZE - zero_offset;
2931 userpage = kmap_atomic(page);
2932 memset(userpage + zero_offset, 0, iosize);
2933 flush_dcache_page(page);
2934 kunmap_atomic(userpage);
2937 while (cur <= end) {
2938 bool force_bio_submit = false;
2941 if (cur >= last_byte) {
2943 struct extent_state *cached = NULL;
2945 iosize = PAGE_SIZE - pg_offset;
2946 userpage = kmap_atomic(page);
2947 memset(userpage + pg_offset, 0, iosize);
2948 flush_dcache_page(page);
2949 kunmap_atomic(userpage);
2950 set_extent_uptodate(tree, cur, cur + iosize - 1,
2952 unlock_extent_cached(tree, cur,
2953 cur + iosize - 1, &cached);
2956 em = __get_extent_map(inode, page, pg_offset, cur,
2957 end - cur + 1, get_extent, em_cached);
2958 if (IS_ERR_OR_NULL(em)) {
2960 unlock_extent(tree, cur, end);
2963 extent_offset = cur - em->start;
2964 BUG_ON(extent_map_end(em) <= cur);
2967 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2968 this_bio_flag |= EXTENT_BIO_COMPRESSED;
2969 extent_set_compress_type(&this_bio_flag,
2973 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2974 cur_end = min(extent_map_end(em) - 1, end);
2975 iosize = ALIGN(iosize, blocksize);
2976 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2977 disk_io_size = em->block_len;
2978 offset = em->block_start;
2980 offset = em->block_start + extent_offset;
2981 disk_io_size = iosize;
2984 block_start = em->block_start;
2985 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2986 block_start = EXTENT_MAP_HOLE;
2989 * If we have a file range that points to a compressed extent
2990 * and it's followed by a consecutive file range that points to
2991 * to the same compressed extent (possibly with a different
2992 * offset and/or length, so it either points to the whole extent
2993 * or only part of it), we must make sure we do not submit a
2994 * single bio to populate the pages for the 2 ranges because
2995 * this makes the compressed extent read zero out the pages
2996 * belonging to the 2nd range. Imagine the following scenario:
2999 * [0 - 8K] [8K - 24K]
3002 * points to extent X, points to extent X,
3003 * offset 4K, length of 8K offset 0, length 16K
3005 * [extent X, compressed length = 4K uncompressed length = 16K]
3007 * If the bio to read the compressed extent covers both ranges,
3008 * it will decompress extent X into the pages belonging to the
3009 * first range and then it will stop, zeroing out the remaining
3010 * pages that belong to the other range that points to extent X.
3011 * So here we make sure we submit 2 bios, one for the first
3012 * range and another one for the third range. Both will target
3013 * the same physical extent from disk, but we can't currently
3014 * make the compressed bio endio callback populate the pages
3015 * for both ranges because each compressed bio is tightly
3016 * coupled with a single extent map, and each range can have
3017 * an extent map with a different offset value relative to the
3018 * uncompressed data of our extent and different lengths. This
3019 * is a corner case so we prioritize correctness over
3020 * non-optimal behavior (submitting 2 bios for the same extent).
3022 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) &&
3023 prev_em_start && *prev_em_start != (u64)-1 &&
3024 *prev_em_start != em->start)
3025 force_bio_submit = true;
3028 *prev_em_start = em->start;
3030 free_extent_map(em);
3033 /* we've found a hole, just zero and go on */
3034 if (block_start == EXTENT_MAP_HOLE) {
3036 struct extent_state *cached = NULL;
3038 userpage = kmap_atomic(page);
3039 memset(userpage + pg_offset, 0, iosize);
3040 flush_dcache_page(page);
3041 kunmap_atomic(userpage);
3043 set_extent_uptodate(tree, cur, cur + iosize - 1,
3045 unlock_extent_cached(tree, cur,
3046 cur + iosize - 1, &cached);
3048 pg_offset += iosize;
3051 /* the get_extent function already copied into the page */
3052 if (test_range_bit(tree, cur, cur_end,
3053 EXTENT_UPTODATE, 1, NULL)) {
3054 check_page_uptodate(tree, page);
3055 unlock_extent(tree, cur, cur + iosize - 1);
3057 pg_offset += iosize;
3060 /* we have an inline extent but it didn't get marked up
3061 * to date. Error out
3063 if (block_start == EXTENT_MAP_INLINE) {
3065 unlock_extent(tree, cur, cur + iosize - 1);
3067 pg_offset += iosize;
3071 ret = submit_extent_page(REQ_OP_READ | read_flags, tree, NULL,
3072 page, offset, disk_io_size,
3073 pg_offset, bdev, bio,
3074 end_bio_extent_readpage, mirror_num,
3080 *bio_flags = this_bio_flag;
3083 unlock_extent(tree, cur, cur + iosize - 1);
3087 pg_offset += iosize;
3091 if (!PageError(page))
3092 SetPageUptodate(page);
3098 static inline void contiguous_readpages(struct extent_io_tree *tree,
3099 struct page *pages[], int nr_pages,
3101 struct extent_map **em_cached,
3103 unsigned long *bio_flags,
3106 struct inode *inode;
3107 struct btrfs_ordered_extent *ordered;
3110 inode = pages[0]->mapping->host;
3112 lock_extent(tree, start, end);
3113 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
3117 unlock_extent(tree, start, end);
3118 btrfs_start_ordered_extent(inode, ordered, 1);
3119 btrfs_put_ordered_extent(ordered);
3122 for (index = 0; index < nr_pages; index++) {
3123 __do_readpage(tree, pages[index], btrfs_get_extent, em_cached,
3124 bio, 0, bio_flags, REQ_RAHEAD, prev_em_start);
3125 put_page(pages[index]);
3129 static int __extent_read_full_page(struct extent_io_tree *tree,
3131 get_extent_t *get_extent,
3132 struct bio **bio, int mirror_num,
3133 unsigned long *bio_flags,
3134 unsigned int read_flags)
3136 struct inode *inode = page->mapping->host;
3137 struct btrfs_ordered_extent *ordered;
3138 u64 start = page_offset(page);
3139 u64 end = start + PAGE_SIZE - 1;
3143 lock_extent(tree, start, end);
3144 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
3148 unlock_extent(tree, start, end);
3149 btrfs_start_ordered_extent(inode, ordered, 1);
3150 btrfs_put_ordered_extent(ordered);
3153 ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num,
3154 bio_flags, read_flags, NULL);
3158 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
3159 get_extent_t *get_extent, int mirror_num)
3161 struct bio *bio = NULL;
3162 unsigned long bio_flags = 0;
3165 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
3168 ret = submit_one_bio(bio, mirror_num, bio_flags);
3172 static void update_nr_written(struct writeback_control *wbc,
3173 unsigned long nr_written)
3175 wbc->nr_to_write -= nr_written;
3179 * helper for __extent_writepage, doing all of the delayed allocation setup.
3181 * This returns 1 if btrfs_run_delalloc_range function did all the work required
3182 * to write the page (copy into inline extent). In this case the IO has
3183 * been started and the page is already unlocked.
3185 * This returns 0 if all went well (page still locked)
3186 * This returns < 0 if there were errors (page still locked)
3188 static noinline_for_stack int writepage_delalloc(struct inode *inode,
3189 struct page *page, struct writeback_control *wbc,
3190 u64 delalloc_start, unsigned long *nr_written)
3192 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
3193 u64 page_end = delalloc_start + PAGE_SIZE - 1;
3195 u64 delalloc_to_write = 0;
3196 u64 delalloc_end = 0;
3198 int page_started = 0;
3201 while (delalloc_end < page_end) {
3202 found = find_lock_delalloc_range(inode, tree,
3207 delalloc_start = delalloc_end + 1;
3210 ret = btrfs_run_delalloc_range(inode, page, delalloc_start,
3211 delalloc_end, &page_started, nr_written, wbc);
3212 /* File system has been set read-only */
3216 * btrfs_run_delalloc_range should return < 0 for error
3217 * but just in case, we use > 0 here meaning the IO is
3218 * started, so we don't want to return > 0 unless
3219 * things are going well.
3221 ret = ret < 0 ? ret : -EIO;
3225 * delalloc_end is already one less than the total length, so
3226 * we don't subtract one from PAGE_SIZE
3228 delalloc_to_write += (delalloc_end - delalloc_start +
3229 PAGE_SIZE) >> PAGE_SHIFT;
3230 delalloc_start = delalloc_end + 1;
3232 if (wbc->nr_to_write < delalloc_to_write) {
3235 if (delalloc_to_write < thresh * 2)
3236 thresh = delalloc_to_write;
3237 wbc->nr_to_write = min_t(u64, delalloc_to_write,
3241 /* did the fill delalloc function already unlock and start
3246 * we've unlocked the page, so we can't update
3247 * the mapping's writeback index, just update
3250 wbc->nr_to_write -= *nr_written;
3261 * helper for __extent_writepage. This calls the writepage start hooks,
3262 * and does the loop to map the page into extents and bios.
3264 * We return 1 if the IO is started and the page is unlocked,
3265 * 0 if all went well (page still locked)
3266 * < 0 if there were errors (page still locked)
3268 static noinline_for_stack int __extent_writepage_io(struct inode *inode,
3270 struct writeback_control *wbc,
3271 struct extent_page_data *epd,
3273 unsigned long nr_written,
3274 unsigned int write_flags, int *nr_ret)
3276 struct extent_io_tree *tree = epd->tree;
3277 u64 start = page_offset(page);
3278 u64 page_end = start + PAGE_SIZE - 1;
3284 struct extent_map *em;
3285 struct block_device *bdev;
3286 size_t pg_offset = 0;
3292 ret = btrfs_writepage_cow_fixup(page, start, page_end);
3294 /* Fixup worker will requeue */
3296 wbc->pages_skipped++;
3298 redirty_page_for_writepage(wbc, page);
3300 update_nr_written(wbc, nr_written);
3306 * we don't want to touch the inode after unlocking the page,
3307 * so we update the mapping writeback index now
3309 update_nr_written(wbc, nr_written + 1);
3312 if (i_size <= start) {
3313 btrfs_writepage_endio_finish_ordered(page, start, page_end, 1);
3317 blocksize = inode->i_sb->s_blocksize;
3319 while (cur <= end) {
3323 if (cur >= i_size) {
3324 btrfs_writepage_endio_finish_ordered(page, cur,
3328 em = btrfs_get_extent(BTRFS_I(inode), page, pg_offset, cur,
3330 if (IS_ERR_OR_NULL(em)) {
3332 ret = PTR_ERR_OR_ZERO(em);
3336 extent_offset = cur - em->start;
3337 em_end = extent_map_end(em);
3338 BUG_ON(em_end <= cur);
3340 iosize = min(em_end - cur, end - cur + 1);
3341 iosize = ALIGN(iosize, blocksize);
3342 offset = em->block_start + extent_offset;
3344 block_start = em->block_start;
3345 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3346 free_extent_map(em);
3350 * compressed and inline extents are written through other
3353 if (compressed || block_start == EXTENT_MAP_HOLE ||
3354 block_start == EXTENT_MAP_INLINE) {
3356 * end_io notification does not happen here for
3357 * compressed extents
3360 btrfs_writepage_endio_finish_ordered(page, cur,
3363 else if (compressed) {
3364 /* we don't want to end_page_writeback on
3365 * a compressed extent. this happens
3372 pg_offset += iosize;
3376 btrfs_set_range_writeback(tree, cur, cur + iosize - 1);
3377 if (!PageWriteback(page)) {
3378 btrfs_err(BTRFS_I(inode)->root->fs_info,
3379 "page %lu not writeback, cur %llu end %llu",
3380 page->index, cur, end);
3383 ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc,
3384 page, offset, iosize, pg_offset,
3386 end_bio_extent_writepage,
3390 if (PageWriteback(page))
3391 end_page_writeback(page);
3395 pg_offset += iosize;
3404 * the writepage semantics are similar to regular writepage. extent
3405 * records are inserted to lock ranges in the tree, and as dirty areas
3406 * are found, they are marked writeback. Then the lock bits are removed
3407 * and the end_io handler clears the writeback ranges
3409 * Return 0 if everything goes well.
3410 * Return <0 for error.
3412 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
3413 struct extent_page_data *epd)
3415 struct inode *inode = page->mapping->host;
3416 u64 start = page_offset(page);
3417 u64 page_end = start + PAGE_SIZE - 1;
3420 size_t pg_offset = 0;
3421 loff_t i_size = i_size_read(inode);
3422 unsigned long end_index = i_size >> PAGE_SHIFT;
3423 unsigned int write_flags = 0;
3424 unsigned long nr_written = 0;
3426 write_flags = wbc_to_write_flags(wbc);
3428 trace___extent_writepage(page, inode, wbc);
3430 WARN_ON(!PageLocked(page));
3432 ClearPageError(page);
3434 pg_offset = offset_in_page(i_size);
3435 if (page->index > end_index ||
3436 (page->index == end_index && !pg_offset)) {
3437 page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
3442 if (page->index == end_index) {
3445 userpage = kmap_atomic(page);
3446 memset(userpage + pg_offset, 0,
3447 PAGE_SIZE - pg_offset);
3448 kunmap_atomic(userpage);
3449 flush_dcache_page(page);
3454 set_page_extent_mapped(page);
3456 if (!epd->extent_locked) {
3457 ret = writepage_delalloc(inode, page, wbc, start, &nr_written);
3464 ret = __extent_writepage_io(inode, page, wbc, epd,
3465 i_size, nr_written, write_flags, &nr);
3471 /* make sure the mapping tag for page dirty gets cleared */
3472 set_page_writeback(page);
3473 end_page_writeback(page);
3475 if (PageError(page)) {
3476 ret = ret < 0 ? ret : -EIO;
3477 end_extent_writepage(page, ret, start, page_end);
3487 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3489 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
3490 TASK_UNINTERRUPTIBLE);
3494 * Lock eb pages and flush the bio if we can't the locks
3496 * Return 0 if nothing went wrong
3497 * Return >0 is same as 0, except bio is not submitted
3498 * Return <0 if something went wrong, no page is locked
3500 static noinline_for_stack int
3501 lock_extent_buffer_for_io(struct extent_buffer *eb,
3502 struct btrfs_fs_info *fs_info,
3503 struct extent_page_data *epd)
3505 int i, num_pages, failed_page_nr;
3509 if (!btrfs_try_tree_write_lock(eb)) {
3510 ret = flush_write_bio(epd);
3514 btrfs_tree_lock(eb);
3517 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3518 btrfs_tree_unlock(eb);
3522 ret = flush_write_bio(epd);
3528 wait_on_extent_buffer_writeback(eb);
3529 btrfs_tree_lock(eb);
3530 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3532 btrfs_tree_unlock(eb);
3537 * We need to do this to prevent races in people who check if the eb is
3538 * under IO since we can end up having no IO bits set for a short period
3541 spin_lock(&eb->refs_lock);
3542 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3543 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3544 spin_unlock(&eb->refs_lock);
3545 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3546 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
3548 fs_info->dirty_metadata_batch);
3551 spin_unlock(&eb->refs_lock);
3554 btrfs_tree_unlock(eb);
3559 num_pages = num_extent_pages(eb);
3560 for (i = 0; i < num_pages; i++) {
3561 struct page *p = eb->pages[i];
3563 if (!trylock_page(p)) {
3565 ret = flush_write_bio(epd);
3578 /* Unlock already locked pages */
3579 for (i = 0; i < failed_page_nr; i++)
3580 unlock_page(eb->pages[i]);
3584 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3586 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3587 smp_mb__after_atomic();
3588 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3591 static void set_btree_ioerr(struct page *page)
3593 struct extent_buffer *eb = (struct extent_buffer *)page->private;
3596 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))
3600 * If writeback for a btree extent that doesn't belong to a log tree
3601 * failed, increment the counter transaction->eb_write_errors.
3602 * We do this because while the transaction is running and before it's
3603 * committing (when we call filemap_fdata[write|wait]_range against
3604 * the btree inode), we might have
3605 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
3606 * returns an error or an error happens during writeback, when we're
3607 * committing the transaction we wouldn't know about it, since the pages
3608 * can be no longer dirty nor marked anymore for writeback (if a
3609 * subsequent modification to the extent buffer didn't happen before the
3610 * transaction commit), which makes filemap_fdata[write|wait]_range not
3611 * able to find the pages tagged with SetPageError at transaction
3612 * commit time. So if this happens we must abort the transaction,
3613 * otherwise we commit a super block with btree roots that point to
3614 * btree nodes/leafs whose content on disk is invalid - either garbage
3615 * or the content of some node/leaf from a past generation that got
3616 * cowed or deleted and is no longer valid.
3618 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
3619 * not be enough - we need to distinguish between log tree extents vs
3620 * non-log tree extents, and the next filemap_fdatawait_range() call
3621 * will catch and clear such errors in the mapping - and that call might
3622 * be from a log sync and not from a transaction commit. Also, checking
3623 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
3624 * not done and would not be reliable - the eb might have been released
3625 * from memory and reading it back again means that flag would not be
3626 * set (since it's a runtime flag, not persisted on disk).
3628 * Using the flags below in the btree inode also makes us achieve the
3629 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
3630 * writeback for all dirty pages and before filemap_fdatawait_range()
3631 * is called, the writeback for all dirty pages had already finished
3632 * with errors - because we were not using AS_EIO/AS_ENOSPC,
3633 * filemap_fdatawait_range() would return success, as it could not know
3634 * that writeback errors happened (the pages were no longer tagged for
3637 switch (eb->log_index) {
3639 set_bit(BTRFS_FS_BTREE_ERR, &eb->fs_info->flags);
3642 set_bit(BTRFS_FS_LOG1_ERR, &eb->fs_info->flags);
3645 set_bit(BTRFS_FS_LOG2_ERR, &eb->fs_info->flags);
3648 BUG(); /* unexpected, logic error */
3652 static void end_bio_extent_buffer_writepage(struct bio *bio)
3654 struct bio_vec *bvec;
3655 struct extent_buffer *eb;
3657 struct bvec_iter_all iter_all;
3659 ASSERT(!bio_flagged(bio, BIO_CLONED));
3660 bio_for_each_segment_all(bvec, bio, i, iter_all) {
3661 struct page *page = bvec->bv_page;
3663 eb = (struct extent_buffer *)page->private;
3665 done = atomic_dec_and_test(&eb->io_pages);
3667 if (bio->bi_status ||
3668 test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) {
3669 ClearPageUptodate(page);
3670 set_btree_ioerr(page);
3673 end_page_writeback(page);
3678 end_extent_buffer_writeback(eb);
3684 static noinline_for_stack int write_one_eb(struct extent_buffer *eb,
3685 struct btrfs_fs_info *fs_info,
3686 struct writeback_control *wbc,
3687 struct extent_page_data *epd)
3689 struct block_device *bdev = fs_info->fs_devices->latest_bdev;
3690 struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
3691 u64 offset = eb->start;
3694 unsigned long start, end;
3695 unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META;
3698 clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
3699 num_pages = num_extent_pages(eb);
3700 atomic_set(&eb->io_pages, num_pages);
3702 /* set btree blocks beyond nritems with 0 to avoid stale content. */
3703 nritems = btrfs_header_nritems(eb);
3704 if (btrfs_header_level(eb) > 0) {
3705 end = btrfs_node_key_ptr_offset(nritems);
3707 memzero_extent_buffer(eb, end, eb->len - end);
3711 * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0
3713 start = btrfs_item_nr_offset(nritems);
3714 end = BTRFS_LEAF_DATA_OFFSET + leaf_data_end(fs_info, eb);
3715 memzero_extent_buffer(eb, start, end - start);
3718 for (i = 0; i < num_pages; i++) {
3719 struct page *p = eb->pages[i];
3721 clear_page_dirty_for_io(p);
3722 set_page_writeback(p);
3723 ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc,
3724 p, offset, PAGE_SIZE, 0, bdev,
3726 end_bio_extent_buffer_writepage,
3730 if (PageWriteback(p))
3731 end_page_writeback(p);
3732 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3733 end_extent_buffer_writeback(eb);
3737 offset += PAGE_SIZE;
3738 update_nr_written(wbc, 1);
3742 if (unlikely(ret)) {
3743 for (; i < num_pages; i++) {
3744 struct page *p = eb->pages[i];
3745 clear_page_dirty_for_io(p);
3753 int btree_write_cache_pages(struct address_space *mapping,
3754 struct writeback_control *wbc)
3756 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
3757 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
3758 struct extent_buffer *eb, *prev_eb = NULL;
3759 struct extent_page_data epd = {
3763 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3767 int nr_to_write_done = 0;
3768 struct pagevec pvec;
3771 pgoff_t end; /* Inclusive */
3775 pagevec_init(&pvec);
3776 if (wbc->range_cyclic) {
3777 index = mapping->writeback_index; /* Start from prev offset */
3780 index = wbc->range_start >> PAGE_SHIFT;
3781 end = wbc->range_end >> PAGE_SHIFT;
3784 if (wbc->sync_mode == WB_SYNC_ALL)
3785 tag = PAGECACHE_TAG_TOWRITE;
3787 tag = PAGECACHE_TAG_DIRTY;
3789 if (wbc->sync_mode == WB_SYNC_ALL)
3790 tag_pages_for_writeback(mapping, index, end);
3791 while (!done && !nr_to_write_done && (index <= end) &&
3792 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
3797 for (i = 0; i < nr_pages; i++) {
3798 struct page *page = pvec.pages[i];
3800 if (!PagePrivate(page))
3803 spin_lock(&mapping->private_lock);
3804 if (!PagePrivate(page)) {
3805 spin_unlock(&mapping->private_lock);
3809 eb = (struct extent_buffer *)page->private;
3812 * Shouldn't happen and normally this would be a BUG_ON
3813 * but no sense in crashing the users box for something
3814 * we can survive anyway.
3817 spin_unlock(&mapping->private_lock);
3821 if (eb == prev_eb) {
3822 spin_unlock(&mapping->private_lock);
3826 ret = atomic_inc_not_zero(&eb->refs);
3827 spin_unlock(&mapping->private_lock);
3832 ret = lock_extent_buffer_for_io(eb, fs_info, &epd);
3834 free_extent_buffer(eb);
3838 ret = write_one_eb(eb, fs_info, wbc, &epd);
3841 free_extent_buffer(eb);
3844 free_extent_buffer(eb);
3847 * the filesystem may choose to bump up nr_to_write.
3848 * We have to make sure to honor the new nr_to_write
3851 nr_to_write_done = wbc->nr_to_write <= 0;
3853 pagevec_release(&pvec);
3856 if (!scanned && !done) {
3858 * We hit the last page and there is more work to be done: wrap
3859 * back to the start of the file
3867 end_write_bio(&epd, ret);
3870 ret = flush_write_bio(&epd);
3875 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3876 * @mapping: address space structure to write
3877 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3878 * @data: data passed to __extent_writepage function
3880 * If a page is already under I/O, write_cache_pages() skips it, even
3881 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3882 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3883 * and msync() need to guarantee that all the data which was dirty at the time
3884 * the call was made get new I/O started against them. If wbc->sync_mode is
3885 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3886 * existing IO to complete.
3888 static int extent_write_cache_pages(struct address_space *mapping,
3889 struct writeback_control *wbc,
3890 struct extent_page_data *epd)
3892 struct inode *inode = mapping->host;
3895 int nr_to_write_done = 0;
3896 struct pagevec pvec;
3899 pgoff_t end; /* Inclusive */
3901 int range_whole = 0;
3906 * We have to hold onto the inode so that ordered extents can do their
3907 * work when the IO finishes. The alternative to this is failing to add
3908 * an ordered extent if the igrab() fails there and that is a huge pain
3909 * to deal with, so instead just hold onto the inode throughout the
3910 * writepages operation. If it fails here we are freeing up the inode
3911 * anyway and we'd rather not waste our time writing out stuff that is
3912 * going to be truncated anyway.
3917 pagevec_init(&pvec);
3918 if (wbc->range_cyclic) {
3919 index = mapping->writeback_index; /* Start from prev offset */
3922 index = wbc->range_start >> PAGE_SHIFT;
3923 end = wbc->range_end >> PAGE_SHIFT;
3924 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
3930 * We do the tagged writepage as long as the snapshot flush bit is set
3931 * and we are the first one who do the filemap_flush() on this inode.
3933 * The nr_to_write == LONG_MAX is needed to make sure other flushers do
3934 * not race in and drop the bit.
3936 if (range_whole && wbc->nr_to_write == LONG_MAX &&
3937 test_and_clear_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
3938 &BTRFS_I(inode)->runtime_flags))
3939 wbc->tagged_writepages = 1;
3941 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
3942 tag = PAGECACHE_TAG_TOWRITE;
3944 tag = PAGECACHE_TAG_DIRTY;
3946 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
3947 tag_pages_for_writeback(mapping, index, end);
3949 while (!done && !nr_to_write_done && (index <= end) &&
3950 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping,
3951 &index, end, tag))) {
3955 for (i = 0; i < nr_pages; i++) {
3956 struct page *page = pvec.pages[i];
3958 done_index = page->index;
3960 * At this point we hold neither the i_pages lock nor
3961 * the page lock: the page may be truncated or
3962 * invalidated (changing page->mapping to NULL),
3963 * or even swizzled back from swapper_space to
3964 * tmpfs file mapping
3966 if (!trylock_page(page)) {
3967 ret = flush_write_bio(epd);
3972 if (unlikely(page->mapping != mapping)) {
3977 if (wbc->sync_mode != WB_SYNC_NONE) {
3978 if (PageWriteback(page)) {
3979 ret = flush_write_bio(epd);
3982 wait_on_page_writeback(page);
3985 if (PageWriteback(page) ||
3986 !clear_page_dirty_for_io(page)) {
3991 ret = __extent_writepage(page, wbc, epd);
3994 * done_index is set past this page,
3995 * so media errors will not choke
3996 * background writeout for the entire
3997 * file. This has consequences for
3998 * range_cyclic semantics (ie. it may
3999 * not be suitable for data integrity
4002 done_index = page->index + 1;
4008 * the filesystem may choose to bump up nr_to_write.
4009 * We have to make sure to honor the new nr_to_write
4012 nr_to_write_done = wbc->nr_to_write <= 0;
4014 pagevec_release(&pvec);
4017 if (!scanned && !done) {
4019 * We hit the last page and there is more work to be done: wrap
4020 * back to the start of the file
4027 if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole))
4028 mapping->writeback_index = done_index;
4030 btrfs_add_delayed_iput(inode);
4034 int extent_write_full_page(struct page *page, struct writeback_control *wbc)
4037 struct extent_page_data epd = {
4039 .tree = &BTRFS_I(page->mapping->host)->io_tree,
4041 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4044 ret = __extent_writepage(page, wbc, &epd);
4047 end_write_bio(&epd, ret);
4051 ret = flush_write_bio(&epd);
4056 int extent_write_locked_range(struct inode *inode, u64 start, u64 end,
4060 struct address_space *mapping = inode->i_mapping;
4061 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
4063 unsigned long nr_pages = (end - start + PAGE_SIZE) >>
4066 struct extent_page_data epd = {
4070 .sync_io = mode == WB_SYNC_ALL,
4072 struct writeback_control wbc_writepages = {
4074 .nr_to_write = nr_pages * 2,
4075 .range_start = start,
4076 .range_end = end + 1,
4079 while (start <= end) {
4080 page = find_get_page(mapping, start >> PAGE_SHIFT);
4081 if (clear_page_dirty_for_io(page))
4082 ret = __extent_writepage(page, &wbc_writepages, &epd);
4084 btrfs_writepage_endio_finish_ordered(page, start,
4085 start + PAGE_SIZE - 1, 1);
4094 end_write_bio(&epd, ret);
4097 ret = flush_write_bio(&epd);
4101 int extent_writepages(struct address_space *mapping,
4102 struct writeback_control *wbc)
4105 struct extent_page_data epd = {
4107 .tree = &BTRFS_I(mapping->host)->io_tree,
4109 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4112 ret = extent_write_cache_pages(mapping, wbc, &epd);
4115 end_write_bio(&epd, ret);
4118 ret = flush_write_bio(&epd);
4122 int extent_readpages(struct address_space *mapping, struct list_head *pages,
4125 struct bio *bio = NULL;
4126 unsigned long bio_flags = 0;
4127 struct page *pagepool[16];
4128 struct extent_map *em_cached = NULL;
4129 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
4131 u64 prev_em_start = (u64)-1;
4133 while (!list_empty(pages)) {
4136 for (nr = 0; nr < ARRAY_SIZE(pagepool) && !list_empty(pages);) {
4137 struct page *page = lru_to_page(pages);
4139 prefetchw(&page->flags);
4140 list_del(&page->lru);
4141 if (add_to_page_cache_lru(page, mapping, page->index,
4142 readahead_gfp_mask(mapping))) {
4147 pagepool[nr++] = page;
4148 contig_end = page_offset(page) + PAGE_SIZE - 1;
4152 u64 contig_start = page_offset(pagepool[0]);
4154 ASSERT(contig_start + nr * PAGE_SIZE - 1 == contig_end);
4156 contiguous_readpages(tree, pagepool, nr, contig_start,
4157 contig_end, &em_cached, &bio, &bio_flags,
4163 free_extent_map(em_cached);
4166 return submit_one_bio(bio, 0, bio_flags);
4171 * basic invalidatepage code, this waits on any locked or writeback
4172 * ranges corresponding to the page, and then deletes any extent state
4173 * records from the tree
4175 int extent_invalidatepage(struct extent_io_tree *tree,
4176 struct page *page, unsigned long offset)
4178 struct extent_state *cached_state = NULL;
4179 u64 start = page_offset(page);
4180 u64 end = start + PAGE_SIZE - 1;
4181 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
4183 start += ALIGN(offset, blocksize);
4187 lock_extent_bits(tree, start, end, &cached_state);
4188 wait_on_page_writeback(page);
4189 clear_extent_bit(tree, start, end,
4190 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4191 EXTENT_DO_ACCOUNTING,
4192 1, 1, &cached_state);
4197 * a helper for releasepage, this tests for areas of the page that
4198 * are locked or under IO and drops the related state bits if it is safe
4201 static int try_release_extent_state(struct extent_io_tree *tree,
4202 struct page *page, gfp_t mask)
4204 u64 start = page_offset(page);
4205 u64 end = start + PAGE_SIZE - 1;
4208 if (test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL)) {
4212 * at this point we can safely clear everything except the
4213 * locked bit and the nodatasum bit
4215 ret = __clear_extent_bit(tree, start, end,
4216 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
4217 0, 0, NULL, mask, NULL);
4219 /* if clear_extent_bit failed for enomem reasons,
4220 * we can't allow the release to continue.
4231 * a helper for releasepage. As long as there are no locked extents
4232 * in the range corresponding to the page, both state records and extent
4233 * map records are removed
4235 int try_release_extent_mapping(struct page *page, gfp_t mask)
4237 struct extent_map *em;
4238 u64 start = page_offset(page);
4239 u64 end = start + PAGE_SIZE - 1;
4240 struct btrfs_inode *btrfs_inode = BTRFS_I(page->mapping->host);
4241 struct extent_io_tree *tree = &btrfs_inode->io_tree;
4242 struct extent_map_tree *map = &btrfs_inode->extent_tree;
4244 if (gfpflags_allow_blocking(mask) &&
4245 page->mapping->host->i_size > SZ_16M) {
4247 while (start <= end) {
4248 len = end - start + 1;
4249 write_lock(&map->lock);
4250 em = lookup_extent_mapping(map, start, len);
4252 write_unlock(&map->lock);
4255 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
4256 em->start != start) {
4257 write_unlock(&map->lock);
4258 free_extent_map(em);
4261 if (!test_range_bit(tree, em->start,
4262 extent_map_end(em) - 1,
4263 EXTENT_LOCKED, 0, NULL)) {
4264 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4265 &btrfs_inode->runtime_flags);
4266 remove_extent_mapping(map, em);
4267 /* once for the rb tree */
4268 free_extent_map(em);
4270 start = extent_map_end(em);
4271 write_unlock(&map->lock);
4274 free_extent_map(em);
4277 return try_release_extent_state(tree, page, mask);
4281 * helper function for fiemap, which doesn't want to see any holes.
4282 * This maps until we find something past 'last'
4284 static struct extent_map *get_extent_skip_holes(struct inode *inode,
4285 u64 offset, u64 last)
4287 u64 sectorsize = btrfs_inode_sectorsize(inode);
4288 struct extent_map *em;
4295 len = last - offset;
4298 len = ALIGN(len, sectorsize);
4299 em = btrfs_get_extent_fiemap(BTRFS_I(inode), offset, len);
4300 if (IS_ERR_OR_NULL(em))
4303 /* if this isn't a hole return it */
4304 if (em->block_start != EXTENT_MAP_HOLE)
4307 /* this is a hole, advance to the next extent */
4308 offset = extent_map_end(em);
4309 free_extent_map(em);
4317 * To cache previous fiemap extent
4319 * Will be used for merging fiemap extent
4321 struct fiemap_cache {
4330 * Helper to submit fiemap extent.
4332 * Will try to merge current fiemap extent specified by @offset, @phys,
4333 * @len and @flags with cached one.
4334 * And only when we fails to merge, cached one will be submitted as
4337 * Return value is the same as fiemap_fill_next_extent().
4339 static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
4340 struct fiemap_cache *cache,
4341 u64 offset, u64 phys, u64 len, u32 flags)
4349 * Sanity check, extent_fiemap() should have ensured that new
4350 * fiemap extent won't overlap with cached one.
4353 * NOTE: Physical address can overlap, due to compression
4355 if (cache->offset + cache->len > offset) {
4361 * Only merges fiemap extents if
4362 * 1) Their logical addresses are continuous
4364 * 2) Their physical addresses are continuous
4365 * So truly compressed (physical size smaller than logical size)
4366 * extents won't get merged with each other
4368 * 3) Share same flags except FIEMAP_EXTENT_LAST
4369 * So regular extent won't get merged with prealloc extent
4371 if (cache->offset + cache->len == offset &&
4372 cache->phys + cache->len == phys &&
4373 (cache->flags & ~FIEMAP_EXTENT_LAST) ==
4374 (flags & ~FIEMAP_EXTENT_LAST)) {
4376 cache->flags |= flags;
4377 goto try_submit_last;
4380 /* Not mergeable, need to submit cached one */
4381 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
4382 cache->len, cache->flags);
4383 cache->cached = false;
4387 cache->cached = true;
4388 cache->offset = offset;
4391 cache->flags = flags;
4393 if (cache->flags & FIEMAP_EXTENT_LAST) {
4394 ret = fiemap_fill_next_extent(fieinfo, cache->offset,
4395 cache->phys, cache->len, cache->flags);
4396 cache->cached = false;
4402 * Emit last fiemap cache
4404 * The last fiemap cache may still be cached in the following case:
4406 * |<- Fiemap range ->|
4407 * |<------------ First extent ----------->|
4409 * In this case, the first extent range will be cached but not emitted.
4410 * So we must emit it before ending extent_fiemap().
4412 static int emit_last_fiemap_cache(struct btrfs_fs_info *fs_info,
4413 struct fiemap_extent_info *fieinfo,
4414 struct fiemap_cache *cache)
4421 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
4422 cache->len, cache->flags);
4423 cache->cached = false;
4429 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4430 __u64 start, __u64 len)
4434 u64 max = start + len;
4438 u64 last_for_get_extent = 0;
4440 u64 isize = i_size_read(inode);
4441 struct btrfs_key found_key;
4442 struct extent_map *em = NULL;
4443 struct extent_state *cached_state = NULL;
4444 struct btrfs_path *path;
4445 struct btrfs_root *root = BTRFS_I(inode)->root;
4446 struct fiemap_cache cache = { 0 };
4455 path = btrfs_alloc_path();
4458 path->leave_spinning = 1;
4460 start = round_down(start, btrfs_inode_sectorsize(inode));
4461 len = round_up(max, btrfs_inode_sectorsize(inode)) - start;
4464 * lookup the last file extent. We're not using i_size here
4465 * because there might be preallocation past i_size
4467 ret = btrfs_lookup_file_extent(NULL, root, path,
4468 btrfs_ino(BTRFS_I(inode)), -1, 0);
4470 btrfs_free_path(path);
4479 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
4480 found_type = found_key.type;
4482 /* No extents, but there might be delalloc bits */
4483 if (found_key.objectid != btrfs_ino(BTRFS_I(inode)) ||
4484 found_type != BTRFS_EXTENT_DATA_KEY) {
4485 /* have to trust i_size as the end */
4487 last_for_get_extent = isize;
4490 * remember the start of the last extent. There are a
4491 * bunch of different factors that go into the length of the
4492 * extent, so its much less complex to remember where it started
4494 last = found_key.offset;
4495 last_for_get_extent = last + 1;
4497 btrfs_release_path(path);
4500 * we might have some extents allocated but more delalloc past those
4501 * extents. so, we trust isize unless the start of the last extent is
4506 last_for_get_extent = isize;
4509 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4512 em = get_extent_skip_holes(inode, start, last_for_get_extent);
4521 u64 offset_in_extent = 0;
4523 /* break if the extent we found is outside the range */
4524 if (em->start >= max || extent_map_end(em) < off)
4528 * get_extent may return an extent that starts before our
4529 * requested range. We have to make sure the ranges
4530 * we return to fiemap always move forward and don't
4531 * overlap, so adjust the offsets here
4533 em_start = max(em->start, off);
4536 * record the offset from the start of the extent
4537 * for adjusting the disk offset below. Only do this if the
4538 * extent isn't compressed since our in ram offset may be past
4539 * what we have actually allocated on disk.
4541 if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4542 offset_in_extent = em_start - em->start;
4543 em_end = extent_map_end(em);
4544 em_len = em_end - em_start;
4546 if (em->block_start < EXTENT_MAP_LAST_BYTE)
4547 disko = em->block_start + offset_in_extent;
4552 * bump off for our next call to get_extent
4554 off = extent_map_end(em);
4558 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
4560 flags |= FIEMAP_EXTENT_LAST;
4561 } else if (em->block_start == EXTENT_MAP_INLINE) {
4562 flags |= (FIEMAP_EXTENT_DATA_INLINE |
4563 FIEMAP_EXTENT_NOT_ALIGNED);
4564 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
4565 flags |= (FIEMAP_EXTENT_DELALLOC |
4566 FIEMAP_EXTENT_UNKNOWN);
4567 } else if (fieinfo->fi_extents_max) {
4568 u64 bytenr = em->block_start -
4569 (em->start - em->orig_start);
4572 * As btrfs supports shared space, this information
4573 * can be exported to userspace tools via
4574 * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0
4575 * then we're just getting a count and we can skip the
4578 ret = btrfs_check_shared(root,
4579 btrfs_ino(BTRFS_I(inode)),
4584 flags |= FIEMAP_EXTENT_SHARED;
4587 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4588 flags |= FIEMAP_EXTENT_ENCODED;
4589 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4590 flags |= FIEMAP_EXTENT_UNWRITTEN;
4592 free_extent_map(em);
4594 if ((em_start >= last) || em_len == (u64)-1 ||
4595 (last == (u64)-1 && isize <= em_end)) {
4596 flags |= FIEMAP_EXTENT_LAST;
4600 /* now scan forward to see if this is really the last extent. */
4601 em = get_extent_skip_holes(inode, off, last_for_get_extent);
4607 flags |= FIEMAP_EXTENT_LAST;
4610 ret = emit_fiemap_extent(fieinfo, &cache, em_start, disko,
4620 ret = emit_last_fiemap_cache(root->fs_info, fieinfo, &cache);
4621 free_extent_map(em);
4623 btrfs_free_path(path);
4624 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4629 static void __free_extent_buffer(struct extent_buffer *eb)
4631 btrfs_leak_debug_del(&eb->leak_list);
4632 kmem_cache_free(extent_buffer_cache, eb);
4635 int extent_buffer_under_io(struct extent_buffer *eb)
4637 return (atomic_read(&eb->io_pages) ||
4638 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4639 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4643 * Release all pages attached to the extent buffer.
4645 static void btrfs_release_extent_buffer_pages(struct extent_buffer *eb)
4649 int mapped = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
4651 BUG_ON(extent_buffer_under_io(eb));
4653 num_pages = num_extent_pages(eb);
4654 for (i = 0; i < num_pages; i++) {
4655 struct page *page = eb->pages[i];
4660 spin_lock(&page->mapping->private_lock);
4662 * We do this since we'll remove the pages after we've
4663 * removed the eb from the radix tree, so we could race
4664 * and have this page now attached to the new eb. So
4665 * only clear page_private if it's still connected to
4668 if (PagePrivate(page) &&
4669 page->private == (unsigned long)eb) {
4670 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4671 BUG_ON(PageDirty(page));
4672 BUG_ON(PageWriteback(page));
4674 * We need to make sure we haven't be attached
4677 ClearPagePrivate(page);
4678 set_page_private(page, 0);
4679 /* One for the page private */
4684 spin_unlock(&page->mapping->private_lock);
4686 /* One for when we allocated the page */
4692 * Helper for releasing the extent buffer.
4694 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4696 btrfs_release_extent_buffer_pages(eb);
4697 __free_extent_buffer(eb);
4700 static struct extent_buffer *
4701 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
4704 struct extent_buffer *eb = NULL;
4706 eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL);
4709 eb->fs_info = fs_info;
4711 rwlock_init(&eb->lock);
4712 atomic_set(&eb->blocking_readers, 0);
4713 atomic_set(&eb->blocking_writers, 0);
4714 eb->lock_nested = false;
4715 init_waitqueue_head(&eb->write_lock_wq);
4716 init_waitqueue_head(&eb->read_lock_wq);
4718 btrfs_leak_debug_add(&eb->leak_list, &buffers);
4720 spin_lock_init(&eb->refs_lock);
4721 atomic_set(&eb->refs, 1);
4722 atomic_set(&eb->io_pages, 0);
4725 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4727 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4728 > MAX_INLINE_EXTENT_BUFFER_SIZE);
4729 BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE);
4731 #ifdef CONFIG_BTRFS_DEBUG
4732 atomic_set(&eb->spinning_writers, 0);
4733 atomic_set(&eb->spinning_readers, 0);
4734 atomic_set(&eb->read_locks, 0);
4735 atomic_set(&eb->write_locks, 0);
4741 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src)
4745 struct extent_buffer *new;
4746 int num_pages = num_extent_pages(src);
4748 new = __alloc_extent_buffer(src->fs_info, src->start, src->len);
4752 for (i = 0; i < num_pages; i++) {
4753 p = alloc_page(GFP_NOFS);
4755 btrfs_release_extent_buffer(new);
4758 attach_extent_buffer_page(new, p);
4759 WARN_ON(PageDirty(p));
4762 copy_page(page_address(p), page_address(src->pages[i]));
4765 set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
4766 set_bit(EXTENT_BUFFER_UNMAPPED, &new->bflags);
4771 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4772 u64 start, unsigned long len)
4774 struct extent_buffer *eb;
4778 eb = __alloc_extent_buffer(fs_info, start, len);
4782 num_pages = num_extent_pages(eb);
4783 for (i = 0; i < num_pages; i++) {
4784 eb->pages[i] = alloc_page(GFP_NOFS);
4788 set_extent_buffer_uptodate(eb);
4789 btrfs_set_header_nritems(eb, 0);
4790 set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
4795 __free_page(eb->pages[i - 1]);
4796 __free_extent_buffer(eb);
4800 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4803 return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize);
4806 static void check_buffer_tree_ref(struct extent_buffer *eb)
4809 /* the ref bit is tricky. We have to make sure it is set
4810 * if we have the buffer dirty. Otherwise the
4811 * code to free a buffer can end up dropping a dirty
4814 * Once the ref bit is set, it won't go away while the
4815 * buffer is dirty or in writeback, and it also won't
4816 * go away while we have the reference count on the
4819 * We can't just set the ref bit without bumping the
4820 * ref on the eb because free_extent_buffer might
4821 * see the ref bit and try to clear it. If this happens
4822 * free_extent_buffer might end up dropping our original
4823 * ref by mistake and freeing the page before we are able
4824 * to add one more ref.
4826 * So bump the ref count first, then set the bit. If someone
4827 * beat us to it, drop the ref we added.
4829 refs = atomic_read(&eb->refs);
4830 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4833 spin_lock(&eb->refs_lock);
4834 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4835 atomic_inc(&eb->refs);
4836 spin_unlock(&eb->refs_lock);
4839 static void mark_extent_buffer_accessed(struct extent_buffer *eb,
4840 struct page *accessed)
4844 check_buffer_tree_ref(eb);
4846 num_pages = num_extent_pages(eb);
4847 for (i = 0; i < num_pages; i++) {
4848 struct page *p = eb->pages[i];
4851 mark_page_accessed(p);
4855 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
4858 struct extent_buffer *eb;
4861 eb = radix_tree_lookup(&fs_info->buffer_radix,
4862 start >> PAGE_SHIFT);
4863 if (eb && atomic_inc_not_zero(&eb->refs)) {
4866 * Lock our eb's refs_lock to avoid races with
4867 * free_extent_buffer. When we get our eb it might be flagged
4868 * with EXTENT_BUFFER_STALE and another task running
4869 * free_extent_buffer might have seen that flag set,
4870 * eb->refs == 2, that the buffer isn't under IO (dirty and
4871 * writeback flags not set) and it's still in the tree (flag
4872 * EXTENT_BUFFER_TREE_REF set), therefore being in the process
4873 * of decrementing the extent buffer's reference count twice.
4874 * So here we could race and increment the eb's reference count,
4875 * clear its stale flag, mark it as dirty and drop our reference
4876 * before the other task finishes executing free_extent_buffer,
4877 * which would later result in an attempt to free an extent
4878 * buffer that is dirty.
4880 if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) {
4881 spin_lock(&eb->refs_lock);
4882 spin_unlock(&eb->refs_lock);
4884 mark_extent_buffer_accessed(eb, NULL);
4892 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4893 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
4896 struct extent_buffer *eb, *exists = NULL;
4899 eb = find_extent_buffer(fs_info, start);
4902 eb = alloc_dummy_extent_buffer(fs_info, start);
4905 eb->fs_info = fs_info;
4907 ret = radix_tree_preload(GFP_NOFS);
4910 spin_lock(&fs_info->buffer_lock);
4911 ret = radix_tree_insert(&fs_info->buffer_radix,
4912 start >> PAGE_SHIFT, eb);
4913 spin_unlock(&fs_info->buffer_lock);
4914 radix_tree_preload_end();
4915 if (ret == -EEXIST) {
4916 exists = find_extent_buffer(fs_info, start);
4922 check_buffer_tree_ref(eb);
4923 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
4927 btrfs_release_extent_buffer(eb);
4932 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
4935 unsigned long len = fs_info->nodesize;
4938 unsigned long index = start >> PAGE_SHIFT;
4939 struct extent_buffer *eb;
4940 struct extent_buffer *exists = NULL;
4942 struct address_space *mapping = fs_info->btree_inode->i_mapping;
4946 if (!IS_ALIGNED(start, fs_info->sectorsize)) {
4947 btrfs_err(fs_info, "bad tree block start %llu", start);
4948 return ERR_PTR(-EINVAL);
4951 eb = find_extent_buffer(fs_info, start);
4955 eb = __alloc_extent_buffer(fs_info, start, len);
4957 return ERR_PTR(-ENOMEM);
4959 num_pages = num_extent_pages(eb);
4960 for (i = 0; i < num_pages; i++, index++) {
4961 p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL);
4963 exists = ERR_PTR(-ENOMEM);
4967 spin_lock(&mapping->private_lock);
4968 if (PagePrivate(p)) {
4970 * We could have already allocated an eb for this page
4971 * and attached one so lets see if we can get a ref on
4972 * the existing eb, and if we can we know it's good and
4973 * we can just return that one, else we know we can just
4974 * overwrite page->private.
4976 exists = (struct extent_buffer *)p->private;
4977 if (atomic_inc_not_zero(&exists->refs)) {
4978 spin_unlock(&mapping->private_lock);
4981 mark_extent_buffer_accessed(exists, p);
4987 * Do this so attach doesn't complain and we need to
4988 * drop the ref the old guy had.
4990 ClearPagePrivate(p);
4991 WARN_ON(PageDirty(p));
4994 attach_extent_buffer_page(eb, p);
4995 spin_unlock(&mapping->private_lock);
4996 WARN_ON(PageDirty(p));
4998 if (!PageUptodate(p))
5002 * We can't unlock the pages just yet since the extent buffer
5003 * hasn't been properly inserted in the radix tree, this
5004 * opens a race with btree_releasepage which can free a page
5005 * while we are still filling in all pages for the buffer and
5010 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5012 ret = radix_tree_preload(GFP_NOFS);
5014 exists = ERR_PTR(ret);
5018 spin_lock(&fs_info->buffer_lock);
5019 ret = radix_tree_insert(&fs_info->buffer_radix,
5020 start >> PAGE_SHIFT, eb);
5021 spin_unlock(&fs_info->buffer_lock);
5022 radix_tree_preload_end();
5023 if (ret == -EEXIST) {
5024 exists = find_extent_buffer(fs_info, start);
5030 /* add one reference for the tree */
5031 check_buffer_tree_ref(eb);
5032 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
5035 * Now it's safe to unlock the pages because any calls to
5036 * btree_releasepage will correctly detect that a page belongs to a
5037 * live buffer and won't free them prematurely.
5039 for (i = 0; i < num_pages; i++)
5040 unlock_page(eb->pages[i]);
5044 WARN_ON(!atomic_dec_and_test(&eb->refs));
5045 for (i = 0; i < num_pages; i++) {
5047 unlock_page(eb->pages[i]);
5050 btrfs_release_extent_buffer(eb);
5054 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
5056 struct extent_buffer *eb =
5057 container_of(head, struct extent_buffer, rcu_head);
5059 __free_extent_buffer(eb);
5062 static int release_extent_buffer(struct extent_buffer *eb)
5064 lockdep_assert_held(&eb->refs_lock);
5066 WARN_ON(atomic_read(&eb->refs) == 0);
5067 if (atomic_dec_and_test(&eb->refs)) {
5068 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
5069 struct btrfs_fs_info *fs_info = eb->fs_info;
5071 spin_unlock(&eb->refs_lock);
5073 spin_lock(&fs_info->buffer_lock);
5074 radix_tree_delete(&fs_info->buffer_radix,
5075 eb->start >> PAGE_SHIFT);
5076 spin_unlock(&fs_info->buffer_lock);
5078 spin_unlock(&eb->refs_lock);
5081 /* Should be safe to release our pages at this point */
5082 btrfs_release_extent_buffer_pages(eb);
5083 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5084 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags))) {
5085 __free_extent_buffer(eb);
5089 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
5092 spin_unlock(&eb->refs_lock);
5097 void free_extent_buffer(struct extent_buffer *eb)
5105 refs = atomic_read(&eb->refs);
5106 if ((!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && refs <= 3)
5107 || (test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) &&
5110 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
5115 spin_lock(&eb->refs_lock);
5116 if (atomic_read(&eb->refs) == 2 &&
5117 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
5118 !extent_buffer_under_io(eb) &&
5119 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5120 atomic_dec(&eb->refs);
5123 * I know this is terrible, but it's temporary until we stop tracking
5124 * the uptodate bits and such for the extent buffers.
5126 release_extent_buffer(eb);
5129 void free_extent_buffer_stale(struct extent_buffer *eb)
5134 spin_lock(&eb->refs_lock);
5135 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
5137 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
5138 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5139 atomic_dec(&eb->refs);
5140 release_extent_buffer(eb);
5143 void clear_extent_buffer_dirty(struct extent_buffer *eb)
5149 num_pages = num_extent_pages(eb);
5151 for (i = 0; i < num_pages; i++) {
5152 page = eb->pages[i];
5153 if (!PageDirty(page))
5157 WARN_ON(!PagePrivate(page));
5159 clear_page_dirty_for_io(page);
5160 xa_lock_irq(&page->mapping->i_pages);
5161 if (!PageDirty(page))
5162 __xa_clear_mark(&page->mapping->i_pages,
5163 page_index(page), PAGECACHE_TAG_DIRTY);
5164 xa_unlock_irq(&page->mapping->i_pages);
5165 ClearPageError(page);
5168 WARN_ON(atomic_read(&eb->refs) == 0);
5171 bool set_extent_buffer_dirty(struct extent_buffer *eb)
5177 check_buffer_tree_ref(eb);
5179 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
5181 num_pages = num_extent_pages(eb);
5182 WARN_ON(atomic_read(&eb->refs) == 0);
5183 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
5186 for (i = 0; i < num_pages; i++)
5187 set_page_dirty(eb->pages[i]);
5189 #ifdef CONFIG_BTRFS_DEBUG
5190 for (i = 0; i < num_pages; i++)
5191 ASSERT(PageDirty(eb->pages[i]));
5197 void clear_extent_buffer_uptodate(struct extent_buffer *eb)
5203 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5204 num_pages = num_extent_pages(eb);
5205 for (i = 0; i < num_pages; i++) {
5206 page = eb->pages[i];
5208 ClearPageUptodate(page);
5212 void set_extent_buffer_uptodate(struct extent_buffer *eb)
5218 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5219 num_pages = num_extent_pages(eb);
5220 for (i = 0; i < num_pages; i++) {
5221 page = eb->pages[i];
5222 SetPageUptodate(page);
5226 int read_extent_buffer_pages(struct extent_io_tree *tree,
5227 struct extent_buffer *eb, int wait, int mirror_num)
5233 int locked_pages = 0;
5234 int all_uptodate = 1;
5236 unsigned long num_reads = 0;
5237 struct bio *bio = NULL;
5238 unsigned long bio_flags = 0;
5240 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
5243 num_pages = num_extent_pages(eb);
5244 for (i = 0; i < num_pages; i++) {
5245 page = eb->pages[i];
5246 if (wait == WAIT_NONE) {
5247 if (!trylock_page(page))
5255 * We need to firstly lock all pages to make sure that
5256 * the uptodate bit of our pages won't be affected by
5257 * clear_extent_buffer_uptodate().
5259 for (i = 0; i < num_pages; i++) {
5260 page = eb->pages[i];
5261 if (!PageUptodate(page)) {
5268 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5272 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
5273 eb->read_mirror = 0;
5274 atomic_set(&eb->io_pages, num_reads);
5275 for (i = 0; i < num_pages; i++) {
5276 page = eb->pages[i];
5278 if (!PageUptodate(page)) {
5280 atomic_dec(&eb->io_pages);
5285 ClearPageError(page);
5286 err = __extent_read_full_page(tree, page,
5287 btree_get_extent, &bio,
5288 mirror_num, &bio_flags,
5293 * We use &bio in above __extent_read_full_page,
5294 * so we ensure that if it returns error, the
5295 * current page fails to add itself to bio and
5296 * it's been unlocked.
5298 * We must dec io_pages by ourselves.
5300 atomic_dec(&eb->io_pages);
5308 err = submit_one_bio(bio, mirror_num, bio_flags);
5313 if (ret || wait != WAIT_COMPLETE)
5316 for (i = 0; i < num_pages; i++) {
5317 page = eb->pages[i];
5318 wait_on_page_locked(page);
5319 if (!PageUptodate(page))
5326 while (locked_pages > 0) {
5328 page = eb->pages[locked_pages];
5334 void read_extent_buffer(const struct extent_buffer *eb, void *dstv,
5335 unsigned long start, unsigned long len)
5341 char *dst = (char *)dstv;
5342 size_t start_offset = offset_in_page(eb->start);
5343 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5345 if (start + len > eb->len) {
5346 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n",
5347 eb->start, eb->len, start, len);
5348 memset(dst, 0, 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 memcpy(dst, kaddr + offset, cur);
5368 int read_extent_buffer_to_user(const struct extent_buffer *eb,
5370 unsigned long start, unsigned long len)
5376 char __user *dst = (char __user *)dstv;
5377 size_t start_offset = offset_in_page(eb->start);
5378 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5381 WARN_ON(start > eb->len);
5382 WARN_ON(start + len > eb->start + eb->len);
5384 offset = offset_in_page(start_offset + start);
5387 page = eb->pages[i];
5389 cur = min(len, (PAGE_SIZE - offset));
5390 kaddr = page_address(page);
5391 if (copy_to_user(dst, kaddr + offset, cur)) {
5406 * return 0 if the item is found within a page.
5407 * return 1 if the item spans two pages.
5408 * return -EINVAL otherwise.
5410 int map_private_extent_buffer(const struct extent_buffer *eb,
5411 unsigned long start, unsigned long min_len,
5412 char **map, unsigned long *map_start,
5413 unsigned long *map_len)
5418 size_t start_offset = offset_in_page(eb->start);
5419 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5420 unsigned long end_i = (start_offset + start + min_len - 1) >>
5423 if (start + min_len > eb->len) {
5424 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n",
5425 eb->start, eb->len, start, min_len);
5433 offset = start_offset;
5437 *map_start = ((u64)i << PAGE_SHIFT) - start_offset;
5441 kaddr = page_address(p);
5442 *map = kaddr + offset;
5443 *map_len = PAGE_SIZE - offset;
5447 int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,
5448 unsigned long start, unsigned long len)
5454 char *ptr = (char *)ptrv;
5455 size_t start_offset = offset_in_page(eb->start);
5456 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5459 WARN_ON(start > eb->len);
5460 WARN_ON(start + len > eb->start + eb->len);
5462 offset = offset_in_page(start_offset + start);
5465 page = eb->pages[i];
5467 cur = min(len, (PAGE_SIZE - offset));
5469 kaddr = page_address(page);
5470 ret = memcmp(ptr, kaddr + offset, cur);
5482 void write_extent_buffer_chunk_tree_uuid(struct extent_buffer *eb,
5487 WARN_ON(!PageUptodate(eb->pages[0]));
5488 kaddr = page_address(eb->pages[0]);
5489 memcpy(kaddr + offsetof(struct btrfs_header, chunk_tree_uuid), srcv,
5493 void write_extent_buffer_fsid(struct extent_buffer *eb, const void *srcv)
5497 WARN_ON(!PageUptodate(eb->pages[0]));
5498 kaddr = page_address(eb->pages[0]);
5499 memcpy(kaddr + offsetof(struct btrfs_header, fsid), srcv,
5503 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
5504 unsigned long start, unsigned long len)
5510 char *src = (char *)srcv;
5511 size_t start_offset = offset_in_page(eb->start);
5512 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5514 WARN_ON(start > eb->len);
5515 WARN_ON(start + len > eb->start + eb->len);
5517 offset = offset_in_page(start_offset + start);
5520 page = eb->pages[i];
5521 WARN_ON(!PageUptodate(page));
5523 cur = min(len, PAGE_SIZE - offset);
5524 kaddr = page_address(page);
5525 memcpy(kaddr + offset, src, cur);
5534 void memzero_extent_buffer(struct extent_buffer *eb, unsigned long start,
5541 size_t start_offset = offset_in_page(eb->start);
5542 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5544 WARN_ON(start > eb->len);
5545 WARN_ON(start + len > eb->start + eb->len);
5547 offset = offset_in_page(start_offset + start);
5550 page = eb->pages[i];
5551 WARN_ON(!PageUptodate(page));
5553 cur = min(len, PAGE_SIZE - offset);
5554 kaddr = page_address(page);
5555 memset(kaddr + offset, 0, cur);
5563 void copy_extent_buffer_full(struct extent_buffer *dst,
5564 struct extent_buffer *src)
5569 ASSERT(dst->len == src->len);
5571 num_pages = num_extent_pages(dst);
5572 for (i = 0; i < num_pages; i++)
5573 copy_page(page_address(dst->pages[i]),
5574 page_address(src->pages[i]));
5577 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
5578 unsigned long dst_offset, unsigned long src_offset,
5581 u64 dst_len = dst->len;
5586 size_t start_offset = offset_in_page(dst->start);
5587 unsigned long i = (start_offset + dst_offset) >> PAGE_SHIFT;
5589 WARN_ON(src->len != dst_len);
5591 offset = offset_in_page(start_offset + dst_offset);
5594 page = dst->pages[i];
5595 WARN_ON(!PageUptodate(page));
5597 cur = min(len, (unsigned long)(PAGE_SIZE - offset));
5599 kaddr = page_address(page);
5600 read_extent_buffer(src, kaddr + offset, src_offset, cur);
5610 * eb_bitmap_offset() - calculate the page and offset of the byte containing the
5612 * @eb: the extent buffer
5613 * @start: offset of the bitmap item in the extent buffer
5615 * @page_index: return index of the page in the extent buffer that contains the
5617 * @page_offset: return offset into the page given by page_index
5619 * This helper hides the ugliness of finding the byte in an extent buffer which
5620 * contains a given bit.
5622 static inline void eb_bitmap_offset(struct extent_buffer *eb,
5623 unsigned long start, unsigned long nr,
5624 unsigned long *page_index,
5625 size_t *page_offset)
5627 size_t start_offset = offset_in_page(eb->start);
5628 size_t byte_offset = BIT_BYTE(nr);
5632 * The byte we want is the offset of the extent buffer + the offset of
5633 * the bitmap item in the extent buffer + the offset of the byte in the
5636 offset = start_offset + start + byte_offset;
5638 *page_index = offset >> PAGE_SHIFT;
5639 *page_offset = offset_in_page(offset);
5643 * extent_buffer_test_bit - determine whether a bit in a bitmap item is set
5644 * @eb: the extent buffer
5645 * @start: offset of the bitmap item in the extent buffer
5646 * @nr: bit number to test
5648 int extent_buffer_test_bit(struct extent_buffer *eb, unsigned long start,
5656 eb_bitmap_offset(eb, start, nr, &i, &offset);
5657 page = eb->pages[i];
5658 WARN_ON(!PageUptodate(page));
5659 kaddr = page_address(page);
5660 return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1)));
5664 * extent_buffer_bitmap_set - set an area of a bitmap
5665 * @eb: the extent buffer
5666 * @start: offset of the bitmap item in the extent buffer
5667 * @pos: bit number of the first bit
5668 * @len: number of bits to set
5670 void extent_buffer_bitmap_set(struct extent_buffer *eb, unsigned long start,
5671 unsigned long pos, unsigned long len)
5677 const unsigned int size = pos + len;
5678 int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5679 u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos);
5681 eb_bitmap_offset(eb, start, pos, &i, &offset);
5682 page = eb->pages[i];
5683 WARN_ON(!PageUptodate(page));
5684 kaddr = page_address(page);
5686 while (len >= bits_to_set) {
5687 kaddr[offset] |= mask_to_set;
5689 bits_to_set = BITS_PER_BYTE;
5691 if (++offset >= PAGE_SIZE && len > 0) {
5693 page = eb->pages[++i];
5694 WARN_ON(!PageUptodate(page));
5695 kaddr = page_address(page);
5699 mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
5700 kaddr[offset] |= mask_to_set;
5706 * extent_buffer_bitmap_clear - clear an area of a bitmap
5707 * @eb: the extent buffer
5708 * @start: offset of the bitmap item in the extent buffer
5709 * @pos: bit number of the first bit
5710 * @len: number of bits to clear
5712 void extent_buffer_bitmap_clear(struct extent_buffer *eb, unsigned long start,
5713 unsigned long pos, unsigned long len)
5719 const unsigned int size = pos + len;
5720 int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5721 u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos);
5723 eb_bitmap_offset(eb, start, pos, &i, &offset);
5724 page = eb->pages[i];
5725 WARN_ON(!PageUptodate(page));
5726 kaddr = page_address(page);
5728 while (len >= bits_to_clear) {
5729 kaddr[offset] &= ~mask_to_clear;
5730 len -= bits_to_clear;
5731 bits_to_clear = BITS_PER_BYTE;
5733 if (++offset >= PAGE_SIZE && len > 0) {
5735 page = eb->pages[++i];
5736 WARN_ON(!PageUptodate(page));
5737 kaddr = page_address(page);
5741 mask_to_clear &= BITMAP_LAST_BYTE_MASK(size);
5742 kaddr[offset] &= ~mask_to_clear;
5746 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
5748 unsigned long distance = (src > dst) ? src - dst : dst - src;
5749 return distance < len;
5752 static void copy_pages(struct page *dst_page, struct page *src_page,
5753 unsigned long dst_off, unsigned long src_off,
5756 char *dst_kaddr = page_address(dst_page);
5758 int must_memmove = 0;
5760 if (dst_page != src_page) {
5761 src_kaddr = page_address(src_page);
5763 src_kaddr = dst_kaddr;
5764 if (areas_overlap(src_off, dst_off, len))
5769 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
5771 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
5774 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5775 unsigned long src_offset, unsigned long len)
5777 struct btrfs_fs_info *fs_info = dst->fs_info;
5779 size_t dst_off_in_page;
5780 size_t src_off_in_page;
5781 size_t start_offset = offset_in_page(dst->start);
5782 unsigned long dst_i;
5783 unsigned long src_i;
5785 if (src_offset + len > dst->len) {
5787 "memmove bogus src_offset %lu move len %lu dst len %lu",
5788 src_offset, len, dst->len);
5791 if (dst_offset + len > dst->len) {
5793 "memmove bogus dst_offset %lu move len %lu dst len %lu",
5794 dst_offset, len, dst->len);
5799 dst_off_in_page = offset_in_page(start_offset + dst_offset);
5800 src_off_in_page = offset_in_page(start_offset + src_offset);
5802 dst_i = (start_offset + dst_offset) >> PAGE_SHIFT;
5803 src_i = (start_offset + src_offset) >> PAGE_SHIFT;
5805 cur = min(len, (unsigned long)(PAGE_SIZE -
5807 cur = min_t(unsigned long, cur,
5808 (unsigned long)(PAGE_SIZE - dst_off_in_page));
5810 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5811 dst_off_in_page, src_off_in_page, cur);
5819 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5820 unsigned long src_offset, unsigned long len)
5822 struct btrfs_fs_info *fs_info = dst->fs_info;
5824 size_t dst_off_in_page;
5825 size_t src_off_in_page;
5826 unsigned long dst_end = dst_offset + len - 1;
5827 unsigned long src_end = src_offset + len - 1;
5828 size_t start_offset = offset_in_page(dst->start);
5829 unsigned long dst_i;
5830 unsigned long src_i;
5832 if (src_offset + len > dst->len) {
5834 "memmove bogus src_offset %lu move len %lu len %lu",
5835 src_offset, len, dst->len);
5838 if (dst_offset + len > dst->len) {
5840 "memmove bogus dst_offset %lu move len %lu len %lu",
5841 dst_offset, len, dst->len);
5844 if (dst_offset < src_offset) {
5845 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
5849 dst_i = (start_offset + dst_end) >> PAGE_SHIFT;
5850 src_i = (start_offset + src_end) >> PAGE_SHIFT;
5852 dst_off_in_page = offset_in_page(start_offset + dst_end);
5853 src_off_in_page = offset_in_page(start_offset + src_end);
5855 cur = min_t(unsigned long, len, src_off_in_page + 1);
5856 cur = min(cur, dst_off_in_page + 1);
5857 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5858 dst_off_in_page - cur + 1,
5859 src_off_in_page - cur + 1, cur);
5867 int try_release_extent_buffer(struct page *page)
5869 struct extent_buffer *eb;
5872 * We need to make sure nobody is attaching this page to an eb right
5875 spin_lock(&page->mapping->private_lock);
5876 if (!PagePrivate(page)) {
5877 spin_unlock(&page->mapping->private_lock);
5881 eb = (struct extent_buffer *)page->private;
5885 * This is a little awful but should be ok, we need to make sure that
5886 * the eb doesn't disappear out from under us while we're looking at
5889 spin_lock(&eb->refs_lock);
5890 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
5891 spin_unlock(&eb->refs_lock);
5892 spin_unlock(&page->mapping->private_lock);
5895 spin_unlock(&page->mapping->private_lock);
5898 * If tree ref isn't set then we know the ref on this eb is a real ref,
5899 * so just return, this page will likely be freed soon anyway.
5901 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
5902 spin_unlock(&eb->refs_lock);
5906 return release_extent_buffer(eb);