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>
17 #include "extent_io.h"
18 #include "extent-io-tree.h"
19 #include "extent_map.h"
21 #include "btrfs_inode.h"
23 #include "check-integrity.h"
25 #include "rcu-string.h"
30 #include "block-group.h"
32 static struct kmem_cache *extent_state_cache;
33 static struct kmem_cache *extent_buffer_cache;
34 static struct bio_set btrfs_bioset;
36 static inline bool extent_state_in_tree(const struct extent_state *state)
38 return !RB_EMPTY_NODE(&state->rb_node);
41 #ifdef CONFIG_BTRFS_DEBUG
42 static LIST_HEAD(states);
43 static DEFINE_SPINLOCK(leak_lock);
45 static inline void btrfs_leak_debug_add(spinlock_t *lock,
46 struct list_head *new,
47 struct list_head *head)
51 spin_lock_irqsave(lock, flags);
53 spin_unlock_irqrestore(lock, flags);
56 static inline void btrfs_leak_debug_del(spinlock_t *lock,
57 struct list_head *entry)
61 spin_lock_irqsave(lock, flags);
63 spin_unlock_irqrestore(lock, flags);
66 void btrfs_extent_buffer_leak_debug_check(struct btrfs_fs_info *fs_info)
68 struct extent_buffer *eb;
72 * If we didn't get into open_ctree our allocated_ebs will not be
73 * initialized, so just skip this.
75 if (!fs_info->allocated_ebs.next)
78 spin_lock_irqsave(&fs_info->eb_leak_lock, flags);
79 while (!list_empty(&fs_info->allocated_ebs)) {
80 eb = list_first_entry(&fs_info->allocated_ebs,
81 struct extent_buffer, leak_list);
83 "BTRFS: buffer leak start %llu len %lu refs %d bflags %lu owner %llu\n",
84 eb->start, eb->len, atomic_read(&eb->refs), eb->bflags,
85 btrfs_header_owner(eb));
86 list_del(&eb->leak_list);
87 kmem_cache_free(extent_buffer_cache, eb);
89 spin_unlock_irqrestore(&fs_info->eb_leak_lock, flags);
92 static inline void btrfs_extent_state_leak_debug_check(void)
94 struct extent_state *state;
96 while (!list_empty(&states)) {
97 state = list_entry(states.next, struct extent_state, leak_list);
98 pr_err("BTRFS: state leak: start %llu end %llu state %u in tree %d refs %d\n",
99 state->start, state->end, state->state,
100 extent_state_in_tree(state),
101 refcount_read(&state->refs));
102 list_del(&state->leak_list);
103 kmem_cache_free(extent_state_cache, state);
107 #define btrfs_debug_check_extent_io_range(tree, start, end) \
108 __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end))
109 static inline void __btrfs_debug_check_extent_io_range(const char *caller,
110 struct extent_io_tree *tree, u64 start, u64 end)
112 struct inode *inode = tree->private_data;
115 if (!inode || !is_data_inode(inode))
118 isize = i_size_read(inode);
119 if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) {
120 btrfs_debug_rl(BTRFS_I(inode)->root->fs_info,
121 "%s: ino %llu isize %llu odd range [%llu,%llu]",
122 caller, btrfs_ino(BTRFS_I(inode)), isize, start, end);
126 #define btrfs_leak_debug_add(lock, new, head) do {} while (0)
127 #define btrfs_leak_debug_del(lock, entry) do {} while (0)
128 #define btrfs_extent_state_leak_debug_check() do {} while (0)
129 #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0)
135 struct rb_node rb_node;
138 struct extent_page_data {
139 struct btrfs_bio_ctrl bio_ctrl;
140 /* tells writepage not to lock the state bits for this range
141 * it still does the unlocking
143 unsigned int extent_locked:1;
145 /* tells the submit_bio code to use REQ_SYNC */
146 unsigned int sync_io:1;
149 static int add_extent_changeset(struct extent_state *state, u32 bits,
150 struct extent_changeset *changeset,
157 if (set && (state->state & bits) == bits)
159 if (!set && (state->state & bits) == 0)
161 changeset->bytes_changed += state->end - state->start + 1;
162 ret = ulist_add(&changeset->range_changed, state->start, state->end,
167 int __must_check submit_one_bio(struct bio *bio, int mirror_num,
168 unsigned long bio_flags)
170 blk_status_t ret = 0;
171 struct extent_io_tree *tree = bio->bi_private;
173 bio->bi_private = NULL;
175 if (is_data_inode(tree->private_data))
176 ret = btrfs_submit_data_bio(tree->private_data, bio, mirror_num,
179 ret = btrfs_submit_metadata_bio(tree->private_data, bio,
180 mirror_num, bio_flags);
182 return blk_status_to_errno(ret);
185 /* Cleanup unsubmitted bios */
186 static void end_write_bio(struct extent_page_data *epd, int ret)
188 struct bio *bio = epd->bio_ctrl.bio;
191 bio->bi_status = errno_to_blk_status(ret);
193 epd->bio_ctrl.bio = NULL;
198 * Submit bio from extent page data via submit_one_bio
200 * Return 0 if everything is OK.
201 * Return <0 for error.
203 static int __must_check flush_write_bio(struct extent_page_data *epd)
206 struct bio *bio = epd->bio_ctrl.bio;
209 ret = submit_one_bio(bio, 0, 0);
211 * Clean up of epd->bio is handled by its endio function.
212 * And endio is either triggered by successful bio execution
213 * or the error handler of submit bio hook.
214 * So at this point, no matter what happened, we don't need
215 * to clean up epd->bio.
217 epd->bio_ctrl.bio = NULL;
222 int __init extent_state_cache_init(void)
224 extent_state_cache = kmem_cache_create("btrfs_extent_state",
225 sizeof(struct extent_state), 0,
226 SLAB_MEM_SPREAD, NULL);
227 if (!extent_state_cache)
232 int __init extent_io_init(void)
234 extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer",
235 sizeof(struct extent_buffer), 0,
236 SLAB_MEM_SPREAD, NULL);
237 if (!extent_buffer_cache)
240 if (bioset_init(&btrfs_bioset, BIO_POOL_SIZE,
241 offsetof(struct btrfs_io_bio, bio),
243 goto free_buffer_cache;
245 if (bioset_integrity_create(&btrfs_bioset, BIO_POOL_SIZE))
251 bioset_exit(&btrfs_bioset);
254 kmem_cache_destroy(extent_buffer_cache);
255 extent_buffer_cache = NULL;
259 void __cold extent_state_cache_exit(void)
261 btrfs_extent_state_leak_debug_check();
262 kmem_cache_destroy(extent_state_cache);
265 void __cold extent_io_exit(void)
268 * Make sure all delayed rcu free are flushed before we
272 kmem_cache_destroy(extent_buffer_cache);
273 bioset_exit(&btrfs_bioset);
277 * For the file_extent_tree, we want to hold the inode lock when we lookup and
278 * update the disk_i_size, but lockdep will complain because our io_tree we hold
279 * the tree lock and get the inode lock when setting delalloc. These two things
280 * are unrelated, so make a class for the file_extent_tree so we don't get the
281 * two locking patterns mixed up.
283 static struct lock_class_key file_extent_tree_class;
285 void extent_io_tree_init(struct btrfs_fs_info *fs_info,
286 struct extent_io_tree *tree, unsigned int owner,
289 tree->fs_info = fs_info;
290 tree->state = RB_ROOT;
291 tree->dirty_bytes = 0;
292 spin_lock_init(&tree->lock);
293 tree->private_data = private_data;
295 if (owner == IO_TREE_INODE_FILE_EXTENT)
296 lockdep_set_class(&tree->lock, &file_extent_tree_class);
299 void extent_io_tree_release(struct extent_io_tree *tree)
301 spin_lock(&tree->lock);
303 * Do a single barrier for the waitqueue_active check here, the state
304 * of the waitqueue should not change once extent_io_tree_release is
308 while (!RB_EMPTY_ROOT(&tree->state)) {
309 struct rb_node *node;
310 struct extent_state *state;
312 node = rb_first(&tree->state);
313 state = rb_entry(node, struct extent_state, rb_node);
314 rb_erase(&state->rb_node, &tree->state);
315 RB_CLEAR_NODE(&state->rb_node);
317 * btree io trees aren't supposed to have tasks waiting for
318 * changes in the flags of extent states ever.
320 ASSERT(!waitqueue_active(&state->wq));
321 free_extent_state(state);
323 cond_resched_lock(&tree->lock);
325 spin_unlock(&tree->lock);
328 static struct extent_state *alloc_extent_state(gfp_t mask)
330 struct extent_state *state;
333 * The given mask might be not appropriate for the slab allocator,
334 * drop the unsupported bits
336 mask &= ~(__GFP_DMA32|__GFP_HIGHMEM);
337 state = kmem_cache_alloc(extent_state_cache, mask);
341 state->failrec = NULL;
342 RB_CLEAR_NODE(&state->rb_node);
343 btrfs_leak_debug_add(&leak_lock, &state->leak_list, &states);
344 refcount_set(&state->refs, 1);
345 init_waitqueue_head(&state->wq);
346 trace_alloc_extent_state(state, mask, _RET_IP_);
350 void free_extent_state(struct extent_state *state)
354 if (refcount_dec_and_test(&state->refs)) {
355 WARN_ON(extent_state_in_tree(state));
356 btrfs_leak_debug_del(&leak_lock, &state->leak_list);
357 trace_free_extent_state(state, _RET_IP_);
358 kmem_cache_free(extent_state_cache, state);
362 static struct rb_node *tree_insert(struct rb_root *root,
363 struct rb_node *search_start,
365 struct rb_node *node,
366 struct rb_node ***p_in,
367 struct rb_node **parent_in)
370 struct rb_node *parent = NULL;
371 struct tree_entry *entry;
373 if (p_in && parent_in) {
379 p = search_start ? &search_start : &root->rb_node;
382 entry = rb_entry(parent, struct tree_entry, rb_node);
384 if (offset < entry->start)
386 else if (offset > entry->end)
393 rb_link_node(node, parent, p);
394 rb_insert_color(node, root);
399 * Search @tree for an entry that contains @offset. Such entry would have
400 * entry->start <= offset && entry->end >= offset.
402 * @tree: the tree to search
403 * @offset: offset that should fall within an entry in @tree
404 * @next_ret: pointer to the first entry whose range ends after @offset
405 * @prev_ret: pointer to the first entry whose range begins before @offset
406 * @p_ret: pointer where new node should be anchored (used when inserting an
408 * @parent_ret: points to entry which would have been the parent of the entry,
411 * This function returns a pointer to the entry that contains @offset byte
412 * address. If no such entry exists, then NULL is returned and the other
413 * pointer arguments to the function are filled, otherwise the found entry is
414 * returned and other pointers are left untouched.
416 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
417 struct rb_node **next_ret,
418 struct rb_node **prev_ret,
419 struct rb_node ***p_ret,
420 struct rb_node **parent_ret)
422 struct rb_root *root = &tree->state;
423 struct rb_node **n = &root->rb_node;
424 struct rb_node *prev = NULL;
425 struct rb_node *orig_prev = NULL;
426 struct tree_entry *entry;
427 struct tree_entry *prev_entry = NULL;
431 entry = rb_entry(prev, struct tree_entry, rb_node);
434 if (offset < entry->start)
436 else if (offset > entry->end)
449 while (prev && offset > prev_entry->end) {
450 prev = rb_next(prev);
451 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
458 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
459 while (prev && offset < prev_entry->start) {
460 prev = rb_prev(prev);
461 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
468 static inline struct rb_node *
469 tree_search_for_insert(struct extent_io_tree *tree,
471 struct rb_node ***p_ret,
472 struct rb_node **parent_ret)
474 struct rb_node *next= NULL;
477 ret = __etree_search(tree, offset, &next, NULL, p_ret, parent_ret);
483 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
486 return tree_search_for_insert(tree, offset, NULL, NULL);
490 * utility function to look for merge candidates inside a given range.
491 * Any extents with matching state are merged together into a single
492 * extent in the tree. Extents with EXTENT_IO in their state field
493 * are not merged because the end_io handlers need to be able to do
494 * operations on them without sleeping (or doing allocations/splits).
496 * This should be called with the tree lock held.
498 static void merge_state(struct extent_io_tree *tree,
499 struct extent_state *state)
501 struct extent_state *other;
502 struct rb_node *other_node;
504 if (state->state & (EXTENT_LOCKED | EXTENT_BOUNDARY))
507 other_node = rb_prev(&state->rb_node);
509 other = rb_entry(other_node, struct extent_state, rb_node);
510 if (other->end == state->start - 1 &&
511 other->state == state->state) {
512 if (tree->private_data &&
513 is_data_inode(tree->private_data))
514 btrfs_merge_delalloc_extent(tree->private_data,
516 state->start = other->start;
517 rb_erase(&other->rb_node, &tree->state);
518 RB_CLEAR_NODE(&other->rb_node);
519 free_extent_state(other);
522 other_node = rb_next(&state->rb_node);
524 other = rb_entry(other_node, struct extent_state, rb_node);
525 if (other->start == state->end + 1 &&
526 other->state == state->state) {
527 if (tree->private_data &&
528 is_data_inode(tree->private_data))
529 btrfs_merge_delalloc_extent(tree->private_data,
531 state->end = other->end;
532 rb_erase(&other->rb_node, &tree->state);
533 RB_CLEAR_NODE(&other->rb_node);
534 free_extent_state(other);
539 static void set_state_bits(struct extent_io_tree *tree,
540 struct extent_state *state, u32 *bits,
541 struct extent_changeset *changeset);
544 * insert an extent_state struct into the tree. 'bits' are set on the
545 * struct before it is inserted.
547 * This may return -EEXIST if the extent is already there, in which case the
548 * state struct is freed.
550 * The tree lock is not taken internally. This is a utility function and
551 * probably isn't what you want to call (see set/clear_extent_bit).
553 static int insert_state(struct extent_io_tree *tree,
554 struct extent_state *state, u64 start, u64 end,
556 struct rb_node **parent,
557 u32 *bits, struct extent_changeset *changeset)
559 struct rb_node *node;
562 btrfs_err(tree->fs_info,
563 "insert state: end < start %llu %llu", end, start);
566 state->start = start;
569 set_state_bits(tree, state, bits, changeset);
571 node = tree_insert(&tree->state, NULL, end, &state->rb_node, p, parent);
573 struct extent_state *found;
574 found = rb_entry(node, struct extent_state, rb_node);
575 btrfs_err(tree->fs_info,
576 "found node %llu %llu on insert of %llu %llu",
577 found->start, found->end, start, end);
580 merge_state(tree, state);
585 * split a given extent state struct in two, inserting the preallocated
586 * struct 'prealloc' as the newly created second half. 'split' indicates an
587 * offset inside 'orig' where it should be split.
590 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
591 * are two extent state structs in the tree:
592 * prealloc: [orig->start, split - 1]
593 * orig: [ split, orig->end ]
595 * The tree locks are not taken by this function. They need to be held
598 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
599 struct extent_state *prealloc, u64 split)
601 struct rb_node *node;
603 if (tree->private_data && is_data_inode(tree->private_data))
604 btrfs_split_delalloc_extent(tree->private_data, orig, split);
606 prealloc->start = orig->start;
607 prealloc->end = split - 1;
608 prealloc->state = orig->state;
611 node = tree_insert(&tree->state, &orig->rb_node, prealloc->end,
612 &prealloc->rb_node, NULL, NULL);
614 free_extent_state(prealloc);
620 static struct extent_state *next_state(struct extent_state *state)
622 struct rb_node *next = rb_next(&state->rb_node);
624 return rb_entry(next, struct extent_state, rb_node);
630 * utility function to clear some bits in an extent state struct.
631 * it will optionally wake up anyone waiting on this state (wake == 1).
633 * If no bits are set on the state struct after clearing things, the
634 * struct is freed and removed from the tree
636 static struct extent_state *clear_state_bit(struct extent_io_tree *tree,
637 struct extent_state *state,
639 struct extent_changeset *changeset)
641 struct extent_state *next;
642 u32 bits_to_clear = *bits & ~EXTENT_CTLBITS;
645 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
646 u64 range = state->end - state->start + 1;
647 WARN_ON(range > tree->dirty_bytes);
648 tree->dirty_bytes -= range;
651 if (tree->private_data && is_data_inode(tree->private_data))
652 btrfs_clear_delalloc_extent(tree->private_data, state, bits);
654 ret = add_extent_changeset(state, bits_to_clear, changeset, 0);
656 state->state &= ~bits_to_clear;
659 if (state->state == 0) {
660 next = next_state(state);
661 if (extent_state_in_tree(state)) {
662 rb_erase(&state->rb_node, &tree->state);
663 RB_CLEAR_NODE(&state->rb_node);
664 free_extent_state(state);
669 merge_state(tree, state);
670 next = next_state(state);
675 static struct extent_state *
676 alloc_extent_state_atomic(struct extent_state *prealloc)
679 prealloc = alloc_extent_state(GFP_ATOMIC);
684 static void extent_io_tree_panic(struct extent_io_tree *tree, int err)
686 btrfs_panic(tree->fs_info, err,
687 "locking error: extent tree was modified by another thread while locked");
691 * clear some bits on a range in the tree. This may require splitting
692 * or inserting elements in the tree, so the gfp mask is used to
693 * indicate which allocations or sleeping are allowed.
695 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
696 * the given range from the tree regardless of state (ie for truncate).
698 * the range [start, end] is inclusive.
700 * This takes the tree lock, and returns 0 on success and < 0 on error.
702 int __clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
703 u32 bits, int wake, int delete,
704 struct extent_state **cached_state,
705 gfp_t mask, struct extent_changeset *changeset)
707 struct extent_state *state;
708 struct extent_state *cached;
709 struct extent_state *prealloc = NULL;
710 struct rb_node *node;
715 btrfs_debug_check_extent_io_range(tree, start, end);
716 trace_btrfs_clear_extent_bit(tree, start, end - start + 1, bits);
718 if (bits & EXTENT_DELALLOC)
719 bits |= EXTENT_NORESERVE;
722 bits |= ~EXTENT_CTLBITS;
724 if (bits & (EXTENT_LOCKED | EXTENT_BOUNDARY))
727 if (!prealloc && gfpflags_allow_blocking(mask)) {
729 * Don't care for allocation failure here because we might end
730 * up not needing the pre-allocated extent state at all, which
731 * is the case if we only have in the tree extent states that
732 * cover our input range and don't cover too any other range.
733 * If we end up needing a new extent state we allocate it later.
735 prealloc = alloc_extent_state(mask);
738 spin_lock(&tree->lock);
740 cached = *cached_state;
743 *cached_state = NULL;
747 if (cached && extent_state_in_tree(cached) &&
748 cached->start <= start && cached->end > start) {
750 refcount_dec(&cached->refs);
755 free_extent_state(cached);
758 * this search will find the extents that end after
761 node = tree_search(tree, start);
764 state = rb_entry(node, struct extent_state, rb_node);
766 if (state->start > end)
768 WARN_ON(state->end < start);
769 last_end = state->end;
771 /* the state doesn't have the wanted bits, go ahead */
772 if (!(state->state & bits)) {
773 state = next_state(state);
778 * | ---- desired range ---- |
780 * | ------------- state -------------- |
782 * We need to split the extent we found, and may flip
783 * bits on second half.
785 * If the extent we found extends past our range, we
786 * just split and search again. It'll get split again
787 * the next time though.
789 * If the extent we found is inside our range, we clear
790 * the desired bit on it.
793 if (state->start < start) {
794 prealloc = alloc_extent_state_atomic(prealloc);
796 err = split_state(tree, state, prealloc, start);
798 extent_io_tree_panic(tree, err);
803 if (state->end <= end) {
804 state = clear_state_bit(tree, state, &bits, wake,
811 * | ---- desired range ---- |
813 * We need to split the extent, and clear the bit
816 if (state->start <= end && state->end > end) {
817 prealloc = alloc_extent_state_atomic(prealloc);
819 err = split_state(tree, state, prealloc, end + 1);
821 extent_io_tree_panic(tree, err);
826 clear_state_bit(tree, prealloc, &bits, wake, changeset);
832 state = clear_state_bit(tree, state, &bits, wake, changeset);
834 if (last_end == (u64)-1)
836 start = last_end + 1;
837 if (start <= end && state && !need_resched())
843 spin_unlock(&tree->lock);
844 if (gfpflags_allow_blocking(mask))
849 spin_unlock(&tree->lock);
851 free_extent_state(prealloc);
857 static void wait_on_state(struct extent_io_tree *tree,
858 struct extent_state *state)
859 __releases(tree->lock)
860 __acquires(tree->lock)
863 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
864 spin_unlock(&tree->lock);
866 spin_lock(&tree->lock);
867 finish_wait(&state->wq, &wait);
871 * waits for one or more bits to clear on a range in the state tree.
872 * The range [start, end] is inclusive.
873 * The tree lock is taken by this function
875 static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
878 struct extent_state *state;
879 struct rb_node *node;
881 btrfs_debug_check_extent_io_range(tree, start, end);
883 spin_lock(&tree->lock);
887 * this search will find all the extents that end after
890 node = tree_search(tree, start);
895 state = rb_entry(node, struct extent_state, rb_node);
897 if (state->start > end)
900 if (state->state & bits) {
901 start = state->start;
902 refcount_inc(&state->refs);
903 wait_on_state(tree, state);
904 free_extent_state(state);
907 start = state->end + 1;
912 if (!cond_resched_lock(&tree->lock)) {
913 node = rb_next(node);
918 spin_unlock(&tree->lock);
921 static void set_state_bits(struct extent_io_tree *tree,
922 struct extent_state *state,
923 u32 *bits, struct extent_changeset *changeset)
925 u32 bits_to_set = *bits & ~EXTENT_CTLBITS;
928 if (tree->private_data && is_data_inode(tree->private_data))
929 btrfs_set_delalloc_extent(tree->private_data, state, bits);
931 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
932 u64 range = state->end - state->start + 1;
933 tree->dirty_bytes += range;
935 ret = add_extent_changeset(state, bits_to_set, changeset, 1);
937 state->state |= bits_to_set;
940 static void cache_state_if_flags(struct extent_state *state,
941 struct extent_state **cached_ptr,
944 if (cached_ptr && !(*cached_ptr)) {
945 if (!flags || (state->state & flags)) {
947 refcount_inc(&state->refs);
952 static void cache_state(struct extent_state *state,
953 struct extent_state **cached_ptr)
955 return cache_state_if_flags(state, cached_ptr,
956 EXTENT_LOCKED | EXTENT_BOUNDARY);
960 * set some bits on a range in the tree. This may require allocations or
961 * sleeping, so the gfp mask is used to indicate what is allowed.
963 * If any of the exclusive bits are set, this will fail with -EEXIST if some
964 * part of the range already has the desired bits set. The start of the
965 * existing range is returned in failed_start in this case.
967 * [start, end] is inclusive This takes the tree lock.
969 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, u32 bits,
970 u32 exclusive_bits, u64 *failed_start,
971 struct extent_state **cached_state, gfp_t mask,
972 struct extent_changeset *changeset)
974 struct extent_state *state;
975 struct extent_state *prealloc = NULL;
976 struct rb_node *node;
978 struct rb_node *parent;
983 btrfs_debug_check_extent_io_range(tree, start, end);
984 trace_btrfs_set_extent_bit(tree, start, end - start + 1, bits);
987 ASSERT(failed_start);
989 ASSERT(failed_start == NULL);
991 if (!prealloc && gfpflags_allow_blocking(mask)) {
993 * Don't care for allocation failure here because we might end
994 * up not needing the pre-allocated extent state at all, which
995 * is the case if we only have in the tree extent states that
996 * cover our input range and don't cover too any other range.
997 * If we end up needing a new extent state we allocate it later.
999 prealloc = alloc_extent_state(mask);
1002 spin_lock(&tree->lock);
1003 if (cached_state && *cached_state) {
1004 state = *cached_state;
1005 if (state->start <= start && state->end > start &&
1006 extent_state_in_tree(state)) {
1007 node = &state->rb_node;
1012 * this search will find all the extents that end after
1015 node = tree_search_for_insert(tree, start, &p, &parent);
1017 prealloc = alloc_extent_state_atomic(prealloc);
1019 err = insert_state(tree, prealloc, start, end,
1020 &p, &parent, &bits, changeset);
1022 extent_io_tree_panic(tree, err);
1024 cache_state(prealloc, cached_state);
1028 state = rb_entry(node, struct extent_state, rb_node);
1030 last_start = state->start;
1031 last_end = state->end;
1034 * | ---- desired range ---- |
1037 * Just lock what we found and keep going
1039 if (state->start == start && state->end <= end) {
1040 if (state->state & exclusive_bits) {
1041 *failed_start = state->start;
1046 set_state_bits(tree, state, &bits, changeset);
1047 cache_state(state, cached_state);
1048 merge_state(tree, state);
1049 if (last_end == (u64)-1)
1051 start = last_end + 1;
1052 state = next_state(state);
1053 if (start < end && state && state->start == start &&
1060 * | ---- desired range ---- |
1063 * | ------------- state -------------- |
1065 * We need to split the extent we found, and may flip bits on
1068 * If the extent we found extends past our
1069 * range, we just split and search again. It'll get split
1070 * again the next time though.
1072 * If the extent we found is inside our range, we set the
1073 * desired bit on it.
1075 if (state->start < start) {
1076 if (state->state & exclusive_bits) {
1077 *failed_start = start;
1083 * If this extent already has all the bits we want set, then
1084 * skip it, not necessary to split it or do anything with it.
1086 if ((state->state & bits) == bits) {
1087 start = state->end + 1;
1088 cache_state(state, cached_state);
1092 prealloc = alloc_extent_state_atomic(prealloc);
1094 err = split_state(tree, state, prealloc, start);
1096 extent_io_tree_panic(tree, err);
1101 if (state->end <= end) {
1102 set_state_bits(tree, state, &bits, changeset);
1103 cache_state(state, cached_state);
1104 merge_state(tree, state);
1105 if (last_end == (u64)-1)
1107 start = last_end + 1;
1108 state = next_state(state);
1109 if (start < end && state && state->start == start &&
1116 * | ---- desired range ---- |
1117 * | state | or | state |
1119 * There's a hole, we need to insert something in it and
1120 * ignore the extent we found.
1122 if (state->start > start) {
1124 if (end < last_start)
1127 this_end = last_start - 1;
1129 prealloc = alloc_extent_state_atomic(prealloc);
1133 * Avoid to free 'prealloc' if it can be merged with
1136 err = insert_state(tree, prealloc, start, this_end,
1137 NULL, NULL, &bits, changeset);
1139 extent_io_tree_panic(tree, err);
1141 cache_state(prealloc, cached_state);
1143 start = this_end + 1;
1147 * | ---- desired range ---- |
1149 * We need to split the extent, and set the bit
1152 if (state->start <= end && state->end > end) {
1153 if (state->state & exclusive_bits) {
1154 *failed_start = start;
1159 prealloc = alloc_extent_state_atomic(prealloc);
1161 err = split_state(tree, state, prealloc, end + 1);
1163 extent_io_tree_panic(tree, err);
1165 set_state_bits(tree, prealloc, &bits, changeset);
1166 cache_state(prealloc, cached_state);
1167 merge_state(tree, prealloc);
1175 spin_unlock(&tree->lock);
1176 if (gfpflags_allow_blocking(mask))
1181 spin_unlock(&tree->lock);
1183 free_extent_state(prealloc);
1190 * convert_extent_bit - convert all bits in a given range from one bit to
1192 * @tree: the io tree to search
1193 * @start: the start offset in bytes
1194 * @end: the end offset in bytes (inclusive)
1195 * @bits: the bits to set in this range
1196 * @clear_bits: the bits to clear in this range
1197 * @cached_state: state that we're going to cache
1199 * This will go through and set bits for the given range. If any states exist
1200 * already in this range they are set with the given bit and cleared of the
1201 * clear_bits. This is only meant to be used by things that are mergeable, ie
1202 * converting from say DELALLOC to DIRTY. This is not meant to be used with
1203 * boundary bits like LOCK.
1205 * All allocations are done with GFP_NOFS.
1207 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1208 u32 bits, u32 clear_bits,
1209 struct extent_state **cached_state)
1211 struct extent_state *state;
1212 struct extent_state *prealloc = NULL;
1213 struct rb_node *node;
1215 struct rb_node *parent;
1219 bool first_iteration = true;
1221 btrfs_debug_check_extent_io_range(tree, start, end);
1222 trace_btrfs_convert_extent_bit(tree, start, end - start + 1, bits,
1228 * Best effort, don't worry if extent state allocation fails
1229 * here for the first iteration. We might have a cached state
1230 * that matches exactly the target range, in which case no
1231 * extent state allocations are needed. We'll only know this
1232 * after locking the tree.
1234 prealloc = alloc_extent_state(GFP_NOFS);
1235 if (!prealloc && !first_iteration)
1239 spin_lock(&tree->lock);
1240 if (cached_state && *cached_state) {
1241 state = *cached_state;
1242 if (state->start <= start && state->end > start &&
1243 extent_state_in_tree(state)) {
1244 node = &state->rb_node;
1250 * this search will find all the extents that end after
1253 node = tree_search_for_insert(tree, start, &p, &parent);
1255 prealloc = alloc_extent_state_atomic(prealloc);
1260 err = insert_state(tree, prealloc, start, end,
1261 &p, &parent, &bits, NULL);
1263 extent_io_tree_panic(tree, err);
1264 cache_state(prealloc, cached_state);
1268 state = rb_entry(node, struct extent_state, rb_node);
1270 last_start = state->start;
1271 last_end = state->end;
1274 * | ---- desired range ---- |
1277 * Just lock what we found and keep going
1279 if (state->start == start && state->end <= end) {
1280 set_state_bits(tree, state, &bits, NULL);
1281 cache_state(state, cached_state);
1282 state = clear_state_bit(tree, state, &clear_bits, 0, NULL);
1283 if (last_end == (u64)-1)
1285 start = last_end + 1;
1286 if (start < end && state && state->start == start &&
1293 * | ---- desired range ---- |
1296 * | ------------- state -------------- |
1298 * We need to split the extent we found, and may flip bits on
1301 * If the extent we found extends past our
1302 * range, we just split and search again. It'll get split
1303 * again the next time though.
1305 * If the extent we found is inside our range, we set the
1306 * desired bit on it.
1308 if (state->start < start) {
1309 prealloc = alloc_extent_state_atomic(prealloc);
1314 err = split_state(tree, state, prealloc, start);
1316 extent_io_tree_panic(tree, err);
1320 if (state->end <= end) {
1321 set_state_bits(tree, state, &bits, NULL);
1322 cache_state(state, cached_state);
1323 state = clear_state_bit(tree, state, &clear_bits, 0,
1325 if (last_end == (u64)-1)
1327 start = last_end + 1;
1328 if (start < end && state && state->start == start &&
1335 * | ---- desired range ---- |
1336 * | state | or | state |
1338 * There's a hole, we need to insert something in it and
1339 * ignore the extent we found.
1341 if (state->start > start) {
1343 if (end < last_start)
1346 this_end = last_start - 1;
1348 prealloc = alloc_extent_state_atomic(prealloc);
1355 * Avoid to free 'prealloc' if it can be merged with
1358 err = insert_state(tree, prealloc, start, this_end,
1359 NULL, NULL, &bits, NULL);
1361 extent_io_tree_panic(tree, err);
1362 cache_state(prealloc, cached_state);
1364 start = this_end + 1;
1368 * | ---- desired range ---- |
1370 * We need to split the extent, and set the bit
1373 if (state->start <= end && state->end > end) {
1374 prealloc = alloc_extent_state_atomic(prealloc);
1380 err = split_state(tree, state, prealloc, end + 1);
1382 extent_io_tree_panic(tree, err);
1384 set_state_bits(tree, prealloc, &bits, NULL);
1385 cache_state(prealloc, cached_state);
1386 clear_state_bit(tree, prealloc, &clear_bits, 0, NULL);
1394 spin_unlock(&tree->lock);
1396 first_iteration = false;
1400 spin_unlock(&tree->lock);
1402 free_extent_state(prealloc);
1407 /* wrappers around set/clear extent bit */
1408 int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1409 u32 bits, struct extent_changeset *changeset)
1412 * We don't support EXTENT_LOCKED yet, as current changeset will
1413 * record any bits changed, so for EXTENT_LOCKED case, it will
1414 * either fail with -EEXIST or changeset will record the whole
1417 BUG_ON(bits & EXTENT_LOCKED);
1419 return set_extent_bit(tree, start, end, bits, 0, NULL, NULL, GFP_NOFS,
1423 int set_extent_bits_nowait(struct extent_io_tree *tree, u64 start, u64 end,
1426 return set_extent_bit(tree, start, end, bits, 0, NULL, NULL,
1430 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1431 u32 bits, int wake, int delete,
1432 struct extent_state **cached)
1434 return __clear_extent_bit(tree, start, end, bits, wake, delete,
1435 cached, GFP_NOFS, NULL);
1438 int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1439 u32 bits, struct extent_changeset *changeset)
1442 * Don't support EXTENT_LOCKED case, same reason as
1443 * set_record_extent_bits().
1445 BUG_ON(bits & EXTENT_LOCKED);
1447 return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, GFP_NOFS,
1452 * either insert or lock state struct between start and end use mask to tell
1453 * us if waiting is desired.
1455 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1456 struct extent_state **cached_state)
1462 err = set_extent_bit(tree, start, end, EXTENT_LOCKED,
1463 EXTENT_LOCKED, &failed_start,
1464 cached_state, GFP_NOFS, NULL);
1465 if (err == -EEXIST) {
1466 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1467 start = failed_start;
1470 WARN_ON(start > end);
1475 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1480 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1481 &failed_start, NULL, GFP_NOFS, NULL);
1482 if (err == -EEXIST) {
1483 if (failed_start > start)
1484 clear_extent_bit(tree, start, failed_start - 1,
1485 EXTENT_LOCKED, 1, 0, NULL);
1491 void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
1493 unsigned long index = start >> PAGE_SHIFT;
1494 unsigned long end_index = end >> PAGE_SHIFT;
1497 while (index <= end_index) {
1498 page = find_get_page(inode->i_mapping, index);
1499 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1500 clear_page_dirty_for_io(page);
1506 void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
1508 unsigned long index = start >> PAGE_SHIFT;
1509 unsigned long end_index = end >> PAGE_SHIFT;
1512 while (index <= end_index) {
1513 page = find_get_page(inode->i_mapping, index);
1514 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1515 __set_page_dirty_nobuffers(page);
1516 account_page_redirty(page);
1522 /* find the first state struct with 'bits' set after 'start', and
1523 * return it. tree->lock must be held. NULL will returned if
1524 * nothing was found after 'start'
1526 static struct extent_state *
1527 find_first_extent_bit_state(struct extent_io_tree *tree, u64 start, u32 bits)
1529 struct rb_node *node;
1530 struct extent_state *state;
1533 * this search will find all the extents that end after
1536 node = tree_search(tree, start);
1541 state = rb_entry(node, struct extent_state, rb_node);
1542 if (state->end >= start && (state->state & bits))
1545 node = rb_next(node);
1554 * Find the first offset in the io tree with one or more @bits set.
1556 * Note: If there are multiple bits set in @bits, any of them will match.
1558 * Return 0 if we find something, and update @start_ret and @end_ret.
1559 * Return 1 if we found nothing.
1561 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1562 u64 *start_ret, u64 *end_ret, u32 bits,
1563 struct extent_state **cached_state)
1565 struct extent_state *state;
1568 spin_lock(&tree->lock);
1569 if (cached_state && *cached_state) {
1570 state = *cached_state;
1571 if (state->end == start - 1 && extent_state_in_tree(state)) {
1572 while ((state = next_state(state)) != NULL) {
1573 if (state->state & bits)
1576 free_extent_state(*cached_state);
1577 *cached_state = NULL;
1580 free_extent_state(*cached_state);
1581 *cached_state = NULL;
1584 state = find_first_extent_bit_state(tree, start, bits);
1587 cache_state_if_flags(state, cached_state, 0);
1588 *start_ret = state->start;
1589 *end_ret = state->end;
1593 spin_unlock(&tree->lock);
1598 * Find a contiguous area of bits
1600 * @tree: io tree to check
1601 * @start: offset to start the search from
1602 * @start_ret: the first offset we found with the bits set
1603 * @end_ret: the final contiguous range of the bits that were set
1604 * @bits: bits to look for
1606 * set_extent_bit and clear_extent_bit can temporarily split contiguous ranges
1607 * to set bits appropriately, and then merge them again. During this time it
1608 * will drop the tree->lock, so use this helper if you want to find the actual
1609 * contiguous area for given bits. We will search to the first bit we find, and
1610 * then walk down the tree until we find a non-contiguous area. The area
1611 * returned will be the full contiguous area with the bits set.
1613 int find_contiguous_extent_bit(struct extent_io_tree *tree, u64 start,
1614 u64 *start_ret, u64 *end_ret, u32 bits)
1616 struct extent_state *state;
1619 spin_lock(&tree->lock);
1620 state = find_first_extent_bit_state(tree, start, bits);
1622 *start_ret = state->start;
1623 *end_ret = state->end;
1624 while ((state = next_state(state)) != NULL) {
1625 if (state->start > (*end_ret + 1))
1627 *end_ret = state->end;
1631 spin_unlock(&tree->lock);
1636 * Find the first range that has @bits not set. This range could start before
1639 * @tree: the tree to search
1640 * @start: offset at/after which the found extent should start
1641 * @start_ret: records the beginning of the range
1642 * @end_ret: records the end of the range (inclusive)
1643 * @bits: the set of bits which must be unset
1645 * Since unallocated range is also considered one which doesn't have the bits
1646 * set it's possible that @end_ret contains -1, this happens in case the range
1647 * spans (last_range_end, end of device]. In this case it's up to the caller to
1648 * trim @end_ret to the appropriate size.
1650 void find_first_clear_extent_bit(struct extent_io_tree *tree, u64 start,
1651 u64 *start_ret, u64 *end_ret, u32 bits)
1653 struct extent_state *state;
1654 struct rb_node *node, *prev = NULL, *next;
1656 spin_lock(&tree->lock);
1658 /* Find first extent with bits cleared */
1660 node = __etree_search(tree, start, &next, &prev, NULL, NULL);
1661 if (!node && !next && !prev) {
1663 * Tree is completely empty, send full range and let
1664 * caller deal with it
1669 } else if (!node && !next) {
1671 * We are past the last allocated chunk, set start at
1672 * the end of the last extent.
1674 state = rb_entry(prev, struct extent_state, rb_node);
1675 *start_ret = state->end + 1;
1682 * At this point 'node' either contains 'start' or start is
1685 state = rb_entry(node, struct extent_state, rb_node);
1687 if (in_range(start, state->start, state->end - state->start + 1)) {
1688 if (state->state & bits) {
1690 * |--range with bits sets--|
1694 start = state->end + 1;
1697 * 'start' falls within a range that doesn't
1698 * have the bits set, so take its start as
1699 * the beginning of the desired range
1701 * |--range with bits cleared----|
1705 *start_ret = state->start;
1710 * |---prev range---|---hole/unset---|---node range---|
1716 * |---hole/unset--||--first node--|
1721 state = rb_entry(prev, struct extent_state,
1723 *start_ret = state->end + 1;
1732 * Find the longest stretch from start until an entry which has the
1736 state = rb_entry(node, struct extent_state, rb_node);
1737 if (state->end >= start && !(state->state & bits)) {
1738 *end_ret = state->end;
1740 *end_ret = state->start - 1;
1744 node = rb_next(node);
1749 spin_unlock(&tree->lock);
1753 * find a contiguous range of bytes in the file marked as delalloc, not
1754 * more than 'max_bytes'. start and end are used to return the range,
1756 * true is returned if we find something, false if nothing was in the tree
1758 bool btrfs_find_delalloc_range(struct extent_io_tree *tree, u64 *start,
1759 u64 *end, u64 max_bytes,
1760 struct extent_state **cached_state)
1762 struct rb_node *node;
1763 struct extent_state *state;
1764 u64 cur_start = *start;
1766 u64 total_bytes = 0;
1768 spin_lock(&tree->lock);
1771 * this search will find all the extents that end after
1774 node = tree_search(tree, cur_start);
1781 state = rb_entry(node, struct extent_state, rb_node);
1782 if (found && (state->start != cur_start ||
1783 (state->state & EXTENT_BOUNDARY))) {
1786 if (!(state->state & EXTENT_DELALLOC)) {
1792 *start = state->start;
1793 *cached_state = state;
1794 refcount_inc(&state->refs);
1798 cur_start = state->end + 1;
1799 node = rb_next(node);
1800 total_bytes += state->end - state->start + 1;
1801 if (total_bytes >= max_bytes)
1807 spin_unlock(&tree->lock);
1812 * Process one page for __process_pages_contig().
1814 * Return >0 if we hit @page == @locked_page.
1815 * Return 0 if we updated the page status.
1816 * Return -EGAIN if the we need to try again.
1817 * (For PAGE_LOCK case but got dirty page or page not belong to mapping)
1819 static int process_one_page(struct btrfs_fs_info *fs_info,
1820 struct address_space *mapping,
1821 struct page *page, struct page *locked_page,
1822 unsigned long page_ops, u64 start, u64 end)
1826 ASSERT(end + 1 - start != 0 && end + 1 - start < U32_MAX);
1827 len = end + 1 - start;
1829 if (page_ops & PAGE_SET_ORDERED)
1830 btrfs_page_clamp_set_ordered(fs_info, page, start, len);
1831 if (page_ops & PAGE_SET_ERROR)
1832 btrfs_page_clamp_set_error(fs_info, page, start, len);
1833 if (page_ops & PAGE_START_WRITEBACK) {
1834 btrfs_page_clamp_clear_dirty(fs_info, page, start, len);
1835 btrfs_page_clamp_set_writeback(fs_info, page, start, len);
1837 if (page_ops & PAGE_END_WRITEBACK)
1838 btrfs_page_clamp_clear_writeback(fs_info, page, start, len);
1840 if (page == locked_page)
1843 if (page_ops & PAGE_LOCK) {
1846 ret = btrfs_page_start_writer_lock(fs_info, page, start, len);
1849 if (!PageDirty(page) || page->mapping != mapping) {
1850 btrfs_page_end_writer_lock(fs_info, page, start, len);
1854 if (page_ops & PAGE_UNLOCK)
1855 btrfs_page_end_writer_lock(fs_info, page, start, len);
1859 static int __process_pages_contig(struct address_space *mapping,
1860 struct page *locked_page,
1861 u64 start, u64 end, unsigned long page_ops,
1864 struct btrfs_fs_info *fs_info = btrfs_sb(mapping->host->i_sb);
1865 pgoff_t start_index = start >> PAGE_SHIFT;
1866 pgoff_t end_index = end >> PAGE_SHIFT;
1867 pgoff_t index = start_index;
1868 unsigned long nr_pages = end_index - start_index + 1;
1869 unsigned long pages_processed = 0;
1870 struct page *pages[16];
1874 if (page_ops & PAGE_LOCK) {
1875 ASSERT(page_ops == PAGE_LOCK);
1876 ASSERT(processed_end && *processed_end == start);
1879 if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0)
1880 mapping_set_error(mapping, -EIO);
1882 while (nr_pages > 0) {
1885 found_pages = find_get_pages_contig(mapping, index,
1886 min_t(unsigned long,
1887 nr_pages, ARRAY_SIZE(pages)), pages);
1888 if (found_pages == 0) {
1890 * Only if we're going to lock these pages, we can find
1891 * nothing at @index.
1893 ASSERT(page_ops & PAGE_LOCK);
1898 for (i = 0; i < found_pages; i++) {
1901 process_ret = process_one_page(fs_info, mapping,
1902 pages[i], locked_page, page_ops,
1904 if (process_ret < 0) {
1905 for (; i < found_pages; i++)
1913 nr_pages -= found_pages;
1914 index += found_pages;
1918 if (err && processed_end) {
1920 * Update @processed_end. I know this is awful since it has
1921 * two different return value patterns (inclusive vs exclusive).
1923 * But the exclusive pattern is necessary if @start is 0, or we
1924 * underflow and check against processed_end won't work as
1927 if (pages_processed)
1928 *processed_end = min(end,
1929 ((u64)(start_index + pages_processed) << PAGE_SHIFT) - 1);
1931 *processed_end = start;
1936 static noinline void __unlock_for_delalloc(struct inode *inode,
1937 struct page *locked_page,
1940 unsigned long index = start >> PAGE_SHIFT;
1941 unsigned long end_index = end >> PAGE_SHIFT;
1943 ASSERT(locked_page);
1944 if (index == locked_page->index && end_index == index)
1947 __process_pages_contig(inode->i_mapping, locked_page, start, end,
1951 static noinline int lock_delalloc_pages(struct inode *inode,
1952 struct page *locked_page,
1956 unsigned long index = delalloc_start >> PAGE_SHIFT;
1957 unsigned long end_index = delalloc_end >> PAGE_SHIFT;
1958 u64 processed_end = delalloc_start;
1961 ASSERT(locked_page);
1962 if (index == locked_page->index && index == end_index)
1965 ret = __process_pages_contig(inode->i_mapping, locked_page, delalloc_start,
1966 delalloc_end, PAGE_LOCK, &processed_end);
1967 if (ret == -EAGAIN && processed_end > delalloc_start)
1968 __unlock_for_delalloc(inode, locked_page, delalloc_start,
1974 * Find and lock a contiguous range of bytes in the file marked as delalloc, no
1975 * more than @max_bytes. @Start and @end are used to return the range,
1977 * Return: true if we find something
1978 * false if nothing was in the tree
1981 noinline_for_stack bool find_lock_delalloc_range(struct inode *inode,
1982 struct page *locked_page, u64 *start,
1985 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
1986 u64 max_bytes = BTRFS_MAX_EXTENT_SIZE;
1990 struct extent_state *cached_state = NULL;
1995 /* step one, find a bunch of delalloc bytes starting at start */
1996 delalloc_start = *start;
1998 found = btrfs_find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1999 max_bytes, &cached_state);
2000 if (!found || delalloc_end <= *start) {
2001 *start = delalloc_start;
2002 *end = delalloc_end;
2003 free_extent_state(cached_state);
2008 * start comes from the offset of locked_page. We have to lock
2009 * pages in order, so we can't process delalloc bytes before
2012 if (delalloc_start < *start)
2013 delalloc_start = *start;
2016 * make sure to limit the number of pages we try to lock down
2018 if (delalloc_end + 1 - delalloc_start > max_bytes)
2019 delalloc_end = delalloc_start + max_bytes - 1;
2021 /* step two, lock all the pages after the page that has start */
2022 ret = lock_delalloc_pages(inode, locked_page,
2023 delalloc_start, delalloc_end);
2024 ASSERT(!ret || ret == -EAGAIN);
2025 if (ret == -EAGAIN) {
2026 /* some of the pages are gone, lets avoid looping by
2027 * shortening the size of the delalloc range we're searching
2029 free_extent_state(cached_state);
2030 cached_state = NULL;
2032 max_bytes = PAGE_SIZE;
2041 /* step three, lock the state bits for the whole range */
2042 lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state);
2044 /* then test to make sure it is all still delalloc */
2045 ret = test_range_bit(tree, delalloc_start, delalloc_end,
2046 EXTENT_DELALLOC, 1, cached_state);
2048 unlock_extent_cached(tree, delalloc_start, delalloc_end,
2050 __unlock_for_delalloc(inode, locked_page,
2051 delalloc_start, delalloc_end);
2055 free_extent_state(cached_state);
2056 *start = delalloc_start;
2057 *end = delalloc_end;
2062 void extent_clear_unlock_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2063 struct page *locked_page,
2064 u32 clear_bits, unsigned long page_ops)
2066 clear_extent_bit(&inode->io_tree, start, end, clear_bits, 1, 0, NULL);
2068 __process_pages_contig(inode->vfs_inode.i_mapping, locked_page,
2069 start, end, page_ops, NULL);
2073 * count the number of bytes in the tree that have a given bit(s)
2074 * set. This can be fairly slow, except for EXTENT_DIRTY which is
2075 * cached. The total number found is returned.
2077 u64 count_range_bits(struct extent_io_tree *tree,
2078 u64 *start, u64 search_end, u64 max_bytes,
2079 u32 bits, int contig)
2081 struct rb_node *node;
2082 struct extent_state *state;
2083 u64 cur_start = *start;
2084 u64 total_bytes = 0;
2088 if (WARN_ON(search_end <= cur_start))
2091 spin_lock(&tree->lock);
2092 if (cur_start == 0 && bits == EXTENT_DIRTY) {
2093 total_bytes = tree->dirty_bytes;
2097 * this search will find all the extents that end after
2100 node = tree_search(tree, cur_start);
2105 state = rb_entry(node, struct extent_state, rb_node);
2106 if (state->start > search_end)
2108 if (contig && found && state->start > last + 1)
2110 if (state->end >= cur_start && (state->state & bits) == bits) {
2111 total_bytes += min(search_end, state->end) + 1 -
2112 max(cur_start, state->start);
2113 if (total_bytes >= max_bytes)
2116 *start = max(cur_start, state->start);
2120 } else if (contig && found) {
2123 node = rb_next(node);
2128 spin_unlock(&tree->lock);
2133 * set the private field for a given byte offset in the tree. If there isn't
2134 * an extent_state there already, this does nothing.
2136 int set_state_failrec(struct extent_io_tree *tree, u64 start,
2137 struct io_failure_record *failrec)
2139 struct rb_node *node;
2140 struct extent_state *state;
2143 spin_lock(&tree->lock);
2145 * this search will find all the extents that end after
2148 node = tree_search(tree, start);
2153 state = rb_entry(node, struct extent_state, rb_node);
2154 if (state->start != start) {
2158 state->failrec = failrec;
2160 spin_unlock(&tree->lock);
2164 struct io_failure_record *get_state_failrec(struct extent_io_tree *tree, u64 start)
2166 struct rb_node *node;
2167 struct extent_state *state;
2168 struct io_failure_record *failrec;
2170 spin_lock(&tree->lock);
2172 * this search will find all the extents that end after
2175 node = tree_search(tree, start);
2177 failrec = ERR_PTR(-ENOENT);
2180 state = rb_entry(node, struct extent_state, rb_node);
2181 if (state->start != start) {
2182 failrec = ERR_PTR(-ENOENT);
2186 failrec = state->failrec;
2188 spin_unlock(&tree->lock);
2193 * searches a range in the state tree for a given mask.
2194 * If 'filled' == 1, this returns 1 only if every extent in the tree
2195 * has the bits set. Otherwise, 1 is returned if any bit in the
2196 * range is found set.
2198 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
2199 u32 bits, int filled, struct extent_state *cached)
2201 struct extent_state *state = NULL;
2202 struct rb_node *node;
2205 spin_lock(&tree->lock);
2206 if (cached && extent_state_in_tree(cached) && cached->start <= start &&
2207 cached->end > start)
2208 node = &cached->rb_node;
2210 node = tree_search(tree, start);
2211 while (node && start <= end) {
2212 state = rb_entry(node, struct extent_state, rb_node);
2214 if (filled && state->start > start) {
2219 if (state->start > end)
2222 if (state->state & bits) {
2226 } else if (filled) {
2231 if (state->end == (u64)-1)
2234 start = state->end + 1;
2237 node = rb_next(node);
2244 spin_unlock(&tree->lock);
2249 * helper function to set a given page up to date if all the
2250 * extents in the tree for that page are up to date
2252 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
2254 u64 start = page_offset(page);
2255 u64 end = start + PAGE_SIZE - 1;
2256 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
2257 SetPageUptodate(page);
2260 int free_io_failure(struct extent_io_tree *failure_tree,
2261 struct extent_io_tree *io_tree,
2262 struct io_failure_record *rec)
2267 set_state_failrec(failure_tree, rec->start, NULL);
2268 ret = clear_extent_bits(failure_tree, rec->start,
2269 rec->start + rec->len - 1,
2270 EXTENT_LOCKED | EXTENT_DIRTY);
2274 ret = clear_extent_bits(io_tree, rec->start,
2275 rec->start + rec->len - 1,
2285 * this bypasses the standard btrfs submit functions deliberately, as
2286 * the standard behavior is to write all copies in a raid setup. here we only
2287 * want to write the one bad copy. so we do the mapping for ourselves and issue
2288 * submit_bio directly.
2289 * to avoid any synchronization issues, wait for the data after writing, which
2290 * actually prevents the read that triggered the error from finishing.
2291 * currently, there can be no more than two copies of every data bit. thus,
2292 * exactly one rewrite is required.
2294 int repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start,
2295 u64 length, u64 logical, struct page *page,
2296 unsigned int pg_offset, int mirror_num)
2299 struct btrfs_device *dev;
2302 struct btrfs_bio *bbio = NULL;
2305 ASSERT(!(fs_info->sb->s_flags & SB_RDONLY));
2306 BUG_ON(!mirror_num);
2308 if (btrfs_is_zoned(fs_info))
2309 return btrfs_repair_one_zone(fs_info, logical);
2311 bio = btrfs_io_bio_alloc(1);
2312 bio->bi_iter.bi_size = 0;
2313 map_length = length;
2316 * Avoid races with device replace and make sure our bbio has devices
2317 * associated to its stripes that don't go away while we are doing the
2318 * read repair operation.
2320 btrfs_bio_counter_inc_blocked(fs_info);
2321 if (btrfs_is_parity_mirror(fs_info, logical, length)) {
2323 * Note that we don't use BTRFS_MAP_WRITE because it's supposed
2324 * to update all raid stripes, but here we just want to correct
2325 * bad stripe, thus BTRFS_MAP_READ is abused to only get the bad
2326 * stripe's dev and sector.
2328 ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, logical,
2329 &map_length, &bbio, 0);
2331 btrfs_bio_counter_dec(fs_info);
2335 ASSERT(bbio->mirror_num == 1);
2337 ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical,
2338 &map_length, &bbio, mirror_num);
2340 btrfs_bio_counter_dec(fs_info);
2344 BUG_ON(mirror_num != bbio->mirror_num);
2347 sector = bbio->stripes[bbio->mirror_num - 1].physical >> 9;
2348 bio->bi_iter.bi_sector = sector;
2349 dev = bbio->stripes[bbio->mirror_num - 1].dev;
2350 btrfs_put_bbio(bbio);
2351 if (!dev || !dev->bdev ||
2352 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) {
2353 btrfs_bio_counter_dec(fs_info);
2357 bio_set_dev(bio, dev->bdev);
2358 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC;
2359 bio_add_page(bio, page, length, pg_offset);
2361 if (btrfsic_submit_bio_wait(bio)) {
2362 /* try to remap that extent elsewhere? */
2363 btrfs_bio_counter_dec(fs_info);
2365 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
2369 btrfs_info_rl_in_rcu(fs_info,
2370 "read error corrected: ino %llu off %llu (dev %s sector %llu)",
2372 rcu_str_deref(dev->name), sector);
2373 btrfs_bio_counter_dec(fs_info);
2378 int btrfs_repair_eb_io_failure(const struct extent_buffer *eb, int mirror_num)
2380 struct btrfs_fs_info *fs_info = eb->fs_info;
2381 u64 start = eb->start;
2382 int i, num_pages = num_extent_pages(eb);
2385 if (sb_rdonly(fs_info->sb))
2388 for (i = 0; i < num_pages; i++) {
2389 struct page *p = eb->pages[i];
2391 ret = repair_io_failure(fs_info, 0, start, PAGE_SIZE, start, p,
2392 start - page_offset(p), mirror_num);
2402 * each time an IO finishes, we do a fast check in the IO failure tree
2403 * to see if we need to process or clean up an io_failure_record
2405 int clean_io_failure(struct btrfs_fs_info *fs_info,
2406 struct extent_io_tree *failure_tree,
2407 struct extent_io_tree *io_tree, u64 start,
2408 struct page *page, u64 ino, unsigned int pg_offset)
2411 struct io_failure_record *failrec;
2412 struct extent_state *state;
2417 ret = count_range_bits(failure_tree, &private, (u64)-1, 1,
2422 failrec = get_state_failrec(failure_tree, start);
2423 if (IS_ERR(failrec))
2426 BUG_ON(!failrec->this_mirror);
2428 if (sb_rdonly(fs_info->sb))
2431 spin_lock(&io_tree->lock);
2432 state = find_first_extent_bit_state(io_tree,
2435 spin_unlock(&io_tree->lock);
2437 if (state && state->start <= failrec->start &&
2438 state->end >= failrec->start + failrec->len - 1) {
2439 num_copies = btrfs_num_copies(fs_info, failrec->logical,
2441 if (num_copies > 1) {
2442 repair_io_failure(fs_info, ino, start, failrec->len,
2443 failrec->logical, page, pg_offset,
2444 failrec->failed_mirror);
2449 free_io_failure(failure_tree, io_tree, failrec);
2455 * Can be called when
2456 * - hold extent lock
2457 * - under ordered extent
2458 * - the inode is freeing
2460 void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start, u64 end)
2462 struct extent_io_tree *failure_tree = &inode->io_failure_tree;
2463 struct io_failure_record *failrec;
2464 struct extent_state *state, *next;
2466 if (RB_EMPTY_ROOT(&failure_tree->state))
2469 spin_lock(&failure_tree->lock);
2470 state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY);
2472 if (state->start > end)
2475 ASSERT(state->end <= end);
2477 next = next_state(state);
2479 failrec = state->failrec;
2480 free_extent_state(state);
2485 spin_unlock(&failure_tree->lock);
2488 static struct io_failure_record *btrfs_get_io_failure_record(struct inode *inode,
2491 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2492 struct io_failure_record *failrec;
2493 struct extent_map *em;
2494 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2495 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2496 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2497 const u32 sectorsize = fs_info->sectorsize;
2501 failrec = get_state_failrec(failure_tree, start);
2502 if (!IS_ERR(failrec)) {
2503 btrfs_debug(fs_info,
2504 "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu",
2505 failrec->logical, failrec->start, failrec->len);
2507 * when data can be on disk more than twice, add to failrec here
2508 * (e.g. with a list for failed_mirror) to make
2509 * clean_io_failure() clean all those errors at once.
2515 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2517 return ERR_PTR(-ENOMEM);
2519 failrec->start = start;
2520 failrec->len = sectorsize;
2521 failrec->this_mirror = 0;
2522 failrec->bio_flags = 0;
2524 read_lock(&em_tree->lock);
2525 em = lookup_extent_mapping(em_tree, start, failrec->len);
2527 read_unlock(&em_tree->lock);
2529 return ERR_PTR(-EIO);
2532 if (em->start > start || em->start + em->len <= start) {
2533 free_extent_map(em);
2536 read_unlock(&em_tree->lock);
2539 return ERR_PTR(-EIO);
2542 logical = start - em->start;
2543 logical = em->block_start + logical;
2544 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2545 logical = em->block_start;
2546 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2547 extent_set_compress_type(&failrec->bio_flags, em->compress_type);
2550 btrfs_debug(fs_info,
2551 "Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu",
2552 logical, start, failrec->len);
2554 failrec->logical = logical;
2555 free_extent_map(em);
2557 /* Set the bits in the private failure tree */
2558 ret = set_extent_bits(failure_tree, start, start + sectorsize - 1,
2559 EXTENT_LOCKED | EXTENT_DIRTY);
2561 ret = set_state_failrec(failure_tree, start, failrec);
2562 /* Set the bits in the inode's tree */
2563 ret = set_extent_bits(tree, start, start + sectorsize - 1,
2565 } else if (ret < 0) {
2567 return ERR_PTR(ret);
2573 static bool btrfs_check_repairable(struct inode *inode,
2574 struct io_failure_record *failrec,
2577 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2580 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
2581 if (num_copies == 1) {
2583 * we only have a single copy of the data, so don't bother with
2584 * all the retry and error correction code that follows. no
2585 * matter what the error is, it is very likely to persist.
2587 btrfs_debug(fs_info,
2588 "Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
2589 num_copies, failrec->this_mirror, failed_mirror);
2593 /* The failure record should only contain one sector */
2594 ASSERT(failrec->len == fs_info->sectorsize);
2597 * There are two premises:
2598 * a) deliver good data to the caller
2599 * b) correct the bad sectors on disk
2601 * Since we're only doing repair for one sector, we only need to get
2602 * a good copy of the failed sector and if we succeed, we have setup
2603 * everything for repair_io_failure to do the rest for us.
2605 failrec->failed_mirror = failed_mirror;
2606 failrec->this_mirror++;
2607 if (failrec->this_mirror == failed_mirror)
2608 failrec->this_mirror++;
2610 if (failrec->this_mirror > num_copies) {
2611 btrfs_debug(fs_info,
2612 "Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
2613 num_copies, failrec->this_mirror, failed_mirror);
2620 int btrfs_repair_one_sector(struct inode *inode,
2621 struct bio *failed_bio, u32 bio_offset,
2622 struct page *page, unsigned int pgoff,
2623 u64 start, int failed_mirror,
2624 submit_bio_hook_t *submit_bio_hook)
2626 struct io_failure_record *failrec;
2627 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2628 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2629 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2630 struct btrfs_io_bio *failed_io_bio = btrfs_io_bio(failed_bio);
2631 const int icsum = bio_offset >> fs_info->sectorsize_bits;
2632 struct bio *repair_bio;
2633 struct btrfs_io_bio *repair_io_bio;
2634 blk_status_t status;
2636 btrfs_debug(fs_info,
2637 "repair read error: read error at %llu", start);
2639 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
2641 failrec = btrfs_get_io_failure_record(inode, start);
2642 if (IS_ERR(failrec))
2643 return PTR_ERR(failrec);
2646 if (!btrfs_check_repairable(inode, failrec, failed_mirror)) {
2647 free_io_failure(failure_tree, tree, failrec);
2651 repair_bio = btrfs_io_bio_alloc(1);
2652 repair_io_bio = btrfs_io_bio(repair_bio);
2653 repair_bio->bi_opf = REQ_OP_READ;
2654 repair_bio->bi_end_io = failed_bio->bi_end_io;
2655 repair_bio->bi_iter.bi_sector = failrec->logical >> 9;
2656 repair_bio->bi_private = failed_bio->bi_private;
2658 if (failed_io_bio->csum) {
2659 const u32 csum_size = fs_info->csum_size;
2661 repair_io_bio->csum = repair_io_bio->csum_inline;
2662 memcpy(repair_io_bio->csum,
2663 failed_io_bio->csum + csum_size * icsum, csum_size);
2666 bio_add_page(repair_bio, page, failrec->len, pgoff);
2667 repair_io_bio->logical = failrec->start;
2668 repair_io_bio->iter = repair_bio->bi_iter;
2670 btrfs_debug(btrfs_sb(inode->i_sb),
2671 "repair read error: submitting new read to mirror %d",
2672 failrec->this_mirror);
2674 status = submit_bio_hook(inode, repair_bio, failrec->this_mirror,
2675 failrec->bio_flags);
2677 free_io_failure(failure_tree, tree, failrec);
2678 bio_put(repair_bio);
2680 return blk_status_to_errno(status);
2683 static void end_page_read(struct page *page, bool uptodate, u64 start, u32 len)
2685 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
2687 ASSERT(page_offset(page) <= start &&
2688 start + len <= page_offset(page) + PAGE_SIZE);
2691 btrfs_page_set_uptodate(fs_info, page, start, len);
2693 btrfs_page_clear_uptodate(fs_info, page, start, len);
2694 btrfs_page_set_error(fs_info, page, start, len);
2697 if (fs_info->sectorsize == PAGE_SIZE)
2700 btrfs_subpage_end_reader(fs_info, page, start, len);
2703 static blk_status_t submit_read_repair(struct inode *inode,
2704 struct bio *failed_bio, u32 bio_offset,
2705 struct page *page, unsigned int pgoff,
2706 u64 start, u64 end, int failed_mirror,
2707 unsigned int error_bitmap,
2708 submit_bio_hook_t *submit_bio_hook)
2710 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2711 const u32 sectorsize = fs_info->sectorsize;
2712 const int nr_bits = (end + 1 - start) >> fs_info->sectorsize_bits;
2716 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
2718 /* We're here because we had some read errors or csum mismatch */
2719 ASSERT(error_bitmap);
2722 * We only get called on buffered IO, thus page must be mapped and bio
2723 * must not be cloned.
2725 ASSERT(page->mapping && !bio_flagged(failed_bio, BIO_CLONED));
2727 /* Iterate through all the sectors in the range */
2728 for (i = 0; i < nr_bits; i++) {
2729 const unsigned int offset = i * sectorsize;
2730 struct extent_state *cached = NULL;
2731 bool uptodate = false;
2734 if (!(error_bitmap & (1U << i))) {
2736 * This sector has no error, just end the page read
2737 * and unlock the range.
2743 ret = btrfs_repair_one_sector(inode, failed_bio,
2744 bio_offset + offset,
2745 page, pgoff + offset, start + offset,
2746 failed_mirror, submit_bio_hook);
2749 * We have submitted the read repair, the page release
2750 * will be handled by the endio function of the
2751 * submitted repair bio.
2752 * Thus we don't need to do any thing here.
2757 * Repair failed, just record the error but still continue.
2758 * Or the remaining sectors will not be properly unlocked.
2763 end_page_read(page, uptodate, start + offset, sectorsize);
2765 set_extent_uptodate(&BTRFS_I(inode)->io_tree,
2767 start + offset + sectorsize - 1,
2768 &cached, GFP_ATOMIC);
2769 unlock_extent_cached_atomic(&BTRFS_I(inode)->io_tree,
2771 start + offset + sectorsize - 1,
2774 return errno_to_blk_status(error);
2777 /* lots and lots of room for performance fixes in the end_bio funcs */
2779 void end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2781 struct btrfs_inode *inode;
2782 const bool uptodate = (err == 0);
2785 ASSERT(page && page->mapping);
2786 inode = BTRFS_I(page->mapping->host);
2787 btrfs_writepage_endio_finish_ordered(inode, page, start, end, uptodate);
2790 ClearPageUptodate(page);
2792 ret = err < 0 ? err : -EIO;
2793 mapping_set_error(page->mapping, ret);
2798 * after a writepage IO is done, we need to:
2799 * clear the uptodate bits on error
2800 * clear the writeback bits in the extent tree for this IO
2801 * end_page_writeback if the page has no more pending IO
2803 * Scheduling is not allowed, so the extent state tree is expected
2804 * to have one and only one object corresponding to this IO.
2806 static void end_bio_extent_writepage(struct bio *bio)
2808 int error = blk_status_to_errno(bio->bi_status);
2809 struct bio_vec *bvec;
2812 struct bvec_iter_all iter_all;
2813 bool first_bvec = true;
2815 ASSERT(!bio_flagged(bio, BIO_CLONED));
2816 bio_for_each_segment_all(bvec, bio, iter_all) {
2817 struct page *page = bvec->bv_page;
2818 struct inode *inode = page->mapping->host;
2819 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2820 const u32 sectorsize = fs_info->sectorsize;
2822 /* Our read/write should always be sector aligned. */
2823 if (!IS_ALIGNED(bvec->bv_offset, sectorsize))
2825 "partial page write in btrfs with offset %u and length %u",
2826 bvec->bv_offset, bvec->bv_len);
2827 else if (!IS_ALIGNED(bvec->bv_len, sectorsize))
2829 "incomplete page write with offset %u and length %u",
2830 bvec->bv_offset, bvec->bv_len);
2832 start = page_offset(page) + bvec->bv_offset;
2833 end = start + bvec->bv_len - 1;
2836 btrfs_record_physical_zoned(inode, start, bio);
2840 end_extent_writepage(page, error, start, end);
2842 btrfs_page_clear_writeback(fs_info, page, start, bvec->bv_len);
2849 * Record previously processed extent range
2851 * For endio_readpage_release_extent() to handle a full extent range, reducing
2852 * the extent io operations.
2854 struct processed_extent {
2855 struct btrfs_inode *inode;
2856 /* Start of the range in @inode */
2858 /* End of the range in @inode */
2864 * Try to release processed extent range
2866 * May not release the extent range right now if the current range is
2867 * contiguous to processed extent.
2869 * Will release processed extent when any of @inode, @uptodate, the range is
2870 * no longer contiguous to the processed range.
2872 * Passing @inode == NULL will force processed extent to be released.
2874 static void endio_readpage_release_extent(struct processed_extent *processed,
2875 struct btrfs_inode *inode, u64 start, u64 end,
2878 struct extent_state *cached = NULL;
2879 struct extent_io_tree *tree;
2881 /* The first extent, initialize @processed */
2882 if (!processed->inode)
2886 * Contiguous to processed extent, just uptodate the end.
2888 * Several things to notice:
2890 * - bio can be merged as long as on-disk bytenr is contiguous
2891 * This means we can have page belonging to other inodes, thus need to
2892 * check if the inode still matches.
2893 * - bvec can contain range beyond current page for multi-page bvec
2894 * Thus we need to do processed->end + 1 >= start check
2896 if (processed->inode == inode && processed->uptodate == uptodate &&
2897 processed->end + 1 >= start && end >= processed->end) {
2898 processed->end = end;
2902 tree = &processed->inode->io_tree;
2904 * Now we don't have range contiguous to the processed range, release
2905 * the processed range now.
2907 if (processed->uptodate && tree->track_uptodate)
2908 set_extent_uptodate(tree, processed->start, processed->end,
2909 &cached, GFP_ATOMIC);
2910 unlock_extent_cached_atomic(tree, processed->start, processed->end,
2914 /* Update processed to current range */
2915 processed->inode = inode;
2916 processed->start = start;
2917 processed->end = end;
2918 processed->uptodate = uptodate;
2921 static void begin_page_read(struct btrfs_fs_info *fs_info, struct page *page)
2923 ASSERT(PageLocked(page));
2924 if (fs_info->sectorsize == PAGE_SIZE)
2927 ASSERT(PagePrivate(page));
2928 btrfs_subpage_start_reader(fs_info, page, page_offset(page), PAGE_SIZE);
2932 * Find extent buffer for a givne bytenr.
2934 * This is for end_bio_extent_readpage(), thus we can't do any unsafe locking
2937 static struct extent_buffer *find_extent_buffer_readpage(
2938 struct btrfs_fs_info *fs_info, struct page *page, u64 bytenr)
2940 struct extent_buffer *eb;
2943 * For regular sectorsize, we can use page->private to grab extent
2946 if (fs_info->sectorsize == PAGE_SIZE) {
2947 ASSERT(PagePrivate(page) && page->private);
2948 return (struct extent_buffer *)page->private;
2951 /* For subpage case, we need to lookup buffer radix tree */
2953 eb = radix_tree_lookup(&fs_info->buffer_radix,
2954 bytenr >> fs_info->sectorsize_bits);
2961 * after a readpage IO is done, we need to:
2962 * clear the uptodate bits on error
2963 * set the uptodate bits if things worked
2964 * set the page up to date if all extents in the tree are uptodate
2965 * clear the lock bit in the extent tree
2966 * unlock the page if there are no other extents locked for it
2968 * Scheduling is not allowed, so the extent state tree is expected
2969 * to have one and only one object corresponding to this IO.
2971 static void end_bio_extent_readpage(struct bio *bio)
2973 struct bio_vec *bvec;
2974 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
2975 struct extent_io_tree *tree, *failure_tree;
2976 struct processed_extent processed = { 0 };
2978 * The offset to the beginning of a bio, since one bio can never be
2979 * larger than UINT_MAX, u32 here is enough.
2984 struct bvec_iter_all iter_all;
2986 ASSERT(!bio_flagged(bio, BIO_CLONED));
2987 bio_for_each_segment_all(bvec, bio, iter_all) {
2988 bool uptodate = !bio->bi_status;
2989 struct page *page = bvec->bv_page;
2990 struct inode *inode = page->mapping->host;
2991 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2992 const u32 sectorsize = fs_info->sectorsize;
2993 unsigned int error_bitmap = (unsigned int)-1;
2998 btrfs_debug(fs_info,
2999 "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u",
3000 bio->bi_iter.bi_sector, bio->bi_status,
3001 io_bio->mirror_num);
3002 tree = &BTRFS_I(inode)->io_tree;
3003 failure_tree = &BTRFS_I(inode)->io_failure_tree;
3006 * We always issue full-sector reads, but if some block in a
3007 * page fails to read, blk_update_request() will advance
3008 * bv_offset and adjust bv_len to compensate. Print a warning
3009 * for unaligned offsets, and an error if they don't add up to
3012 if (!IS_ALIGNED(bvec->bv_offset, sectorsize))
3014 "partial page read in btrfs with offset %u and length %u",
3015 bvec->bv_offset, bvec->bv_len);
3016 else if (!IS_ALIGNED(bvec->bv_offset + bvec->bv_len,
3019 "incomplete page read with offset %u and length %u",
3020 bvec->bv_offset, bvec->bv_len);
3022 start = page_offset(page) + bvec->bv_offset;
3023 end = start + bvec->bv_len - 1;
3026 mirror = io_bio->mirror_num;
3027 if (likely(uptodate)) {
3028 if (is_data_inode(inode)) {
3029 error_bitmap = btrfs_verify_data_csum(io_bio,
3030 bio_offset, page, start, end);
3033 ret = btrfs_validate_metadata_buffer(io_bio,
3034 page, start, end, mirror);
3039 clean_io_failure(BTRFS_I(inode)->root->fs_info,
3040 failure_tree, tree, start,
3042 btrfs_ino(BTRFS_I(inode)), 0);
3045 if (likely(uptodate))
3048 if (is_data_inode(inode)) {
3050 * btrfs_submit_read_repair() will handle all the good
3051 * and bad sectors, we just continue to the next bvec.
3053 submit_read_repair(inode, bio, bio_offset, page,
3054 start - page_offset(page), start,
3055 end, mirror, error_bitmap,
3056 btrfs_submit_data_bio);
3058 ASSERT(bio_offset + len > bio_offset);
3062 struct extent_buffer *eb;
3064 eb = find_extent_buffer_readpage(fs_info, page, start);
3065 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
3066 eb->read_mirror = mirror;
3067 atomic_dec(&eb->io_pages);
3068 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD,
3070 btree_readahead_hook(eb, -EIO);
3073 if (likely(uptodate)) {
3074 loff_t i_size = i_size_read(inode);
3075 pgoff_t end_index = i_size >> PAGE_SHIFT;
3078 * Zero out the remaining part if this range straddles
3081 * Here we should only zero the range inside the bvec,
3082 * not touch anything else.
3084 * NOTE: i_size is exclusive while end is inclusive.
3086 if (page->index == end_index && i_size <= end) {
3087 u32 zero_start = max(offset_in_page(i_size),
3088 offset_in_page(start));
3090 zero_user_segment(page, zero_start,
3091 offset_in_page(end) + 1);
3094 ASSERT(bio_offset + len > bio_offset);
3097 /* Update page status and unlock */
3098 end_page_read(page, uptodate, start, len);
3099 endio_readpage_release_extent(&processed, BTRFS_I(inode),
3100 start, end, uptodate);
3102 /* Release the last extent */
3103 endio_readpage_release_extent(&processed, NULL, 0, 0, false);
3104 btrfs_io_bio_free_csum(io_bio);
3109 * Initialize the members up to but not including 'bio'. Use after allocating a
3110 * new bio by bio_alloc_bioset as it does not initialize the bytes outside of
3111 * 'bio' because use of __GFP_ZERO is not supported.
3113 static inline void btrfs_io_bio_init(struct btrfs_io_bio *btrfs_bio)
3115 memset(btrfs_bio, 0, offsetof(struct btrfs_io_bio, bio));
3119 * The following helpers allocate a bio. As it's backed by a bioset, it'll
3120 * never fail. We're returning a bio right now but you can call btrfs_io_bio
3121 * for the appropriate container_of magic
3123 struct bio *btrfs_bio_alloc(u64 first_byte)
3127 bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_VECS, &btrfs_bioset);
3128 bio->bi_iter.bi_sector = first_byte >> 9;
3129 btrfs_io_bio_init(btrfs_io_bio(bio));
3133 struct bio *btrfs_bio_clone(struct bio *bio)
3135 struct btrfs_io_bio *btrfs_bio;
3138 /* Bio allocation backed by a bioset does not fail */
3139 new = bio_clone_fast(bio, GFP_NOFS, &btrfs_bioset);
3140 btrfs_bio = btrfs_io_bio(new);
3141 btrfs_io_bio_init(btrfs_bio);
3142 btrfs_bio->iter = bio->bi_iter;
3146 struct bio *btrfs_io_bio_alloc(unsigned int nr_iovecs)
3150 /* Bio allocation backed by a bioset does not fail */
3151 bio = bio_alloc_bioset(GFP_NOFS, nr_iovecs, &btrfs_bioset);
3152 btrfs_io_bio_init(btrfs_io_bio(bio));
3156 struct bio *btrfs_bio_clone_partial(struct bio *orig, int offset, int size)
3159 struct btrfs_io_bio *btrfs_bio;
3161 /* this will never fail when it's backed by a bioset */
3162 bio = bio_clone_fast(orig, GFP_NOFS, &btrfs_bioset);
3165 btrfs_bio = btrfs_io_bio(bio);
3166 btrfs_io_bio_init(btrfs_bio);
3168 bio_trim(bio, offset >> 9, size >> 9);
3169 btrfs_bio->iter = bio->bi_iter;
3174 * Attempt to add a page to bio
3176 * @bio: destination bio
3177 * @page: page to add to the bio
3178 * @disk_bytenr: offset of the new bio or to check whether we are adding
3179 * a contiguous page to the previous one
3180 * @pg_offset: starting offset in the page
3181 * @size: portion of page that we want to write
3182 * @prev_bio_flags: flags of previous bio to see if we can merge the current one
3183 * @bio_flags: flags of the current bio to see if we can merge them
3184 * @return: true if page was added, false otherwise
3186 * Attempt to add a page to bio considering stripe alignment etc.
3188 * Return true if successfully page added. Otherwise, return false.
3190 static bool btrfs_bio_add_page(struct btrfs_bio_ctrl *bio_ctrl,
3192 u64 disk_bytenr, unsigned int size,
3193 unsigned int pg_offset,
3194 unsigned long bio_flags)
3196 struct bio *bio = bio_ctrl->bio;
3197 u32 bio_size = bio->bi_iter.bi_size;
3198 const sector_t sector = disk_bytenr >> SECTOR_SHIFT;
3203 /* The limit should be calculated when bio_ctrl->bio is allocated */
3204 ASSERT(bio_ctrl->len_to_oe_boundary && bio_ctrl->len_to_stripe_boundary);
3205 if (bio_ctrl->bio_flags != bio_flags)
3208 if (bio_ctrl->bio_flags & EXTENT_BIO_COMPRESSED)
3209 contig = bio->bi_iter.bi_sector == sector;
3211 contig = bio_end_sector(bio) == sector;
3215 if (bio_size + size > bio_ctrl->len_to_oe_boundary ||
3216 bio_size + size > bio_ctrl->len_to_stripe_boundary)
3219 if (bio_op(bio) == REQ_OP_ZONE_APPEND)
3220 ret = bio_add_zone_append_page(bio, page, size, pg_offset);
3222 ret = bio_add_page(bio, page, size, pg_offset);
3227 static int calc_bio_boundaries(struct btrfs_bio_ctrl *bio_ctrl,
3228 struct btrfs_inode *inode)
3230 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3231 struct btrfs_io_geometry geom;
3232 struct btrfs_ordered_extent *ordered;
3233 struct extent_map *em;
3234 u64 logical = (bio_ctrl->bio->bi_iter.bi_sector << SECTOR_SHIFT);
3238 * Pages for compressed extent are never submitted to disk directly,
3239 * thus it has no real boundary, just set them to U32_MAX.
3241 * The split happens for real compressed bio, which happens in
3242 * btrfs_submit_compressed_read/write().
3244 if (bio_ctrl->bio_flags & EXTENT_BIO_COMPRESSED) {
3245 bio_ctrl->len_to_oe_boundary = U32_MAX;
3246 bio_ctrl->len_to_stripe_boundary = U32_MAX;
3249 em = btrfs_get_chunk_map(fs_info, logical, fs_info->sectorsize);
3252 ret = btrfs_get_io_geometry(fs_info, em, btrfs_op(bio_ctrl->bio),
3254 free_extent_map(em);
3258 if (geom.len > U32_MAX)
3259 bio_ctrl->len_to_stripe_boundary = U32_MAX;
3261 bio_ctrl->len_to_stripe_boundary = (u32)geom.len;
3263 if (!btrfs_is_zoned(fs_info) ||
3264 bio_op(bio_ctrl->bio) != REQ_OP_ZONE_APPEND) {
3265 bio_ctrl->len_to_oe_boundary = U32_MAX;
3269 /* Ordered extent not yet created, so we're good */
3270 ordered = btrfs_lookup_ordered_extent(inode, logical);
3272 bio_ctrl->len_to_oe_boundary = U32_MAX;
3276 bio_ctrl->len_to_oe_boundary = min_t(u32, U32_MAX,
3277 ordered->disk_bytenr + ordered->disk_num_bytes - logical);
3278 btrfs_put_ordered_extent(ordered);
3283 * @opf: bio REQ_OP_* and REQ_* flags as one value
3284 * @wbc: optional writeback control for io accounting
3285 * @page: page to add to the bio
3286 * @disk_bytenr: logical bytenr where the write will be
3287 * @size: portion of page that we want to write to
3288 * @pg_offset: offset of the new bio or to check whether we are adding
3289 * a contiguous page to the previous one
3290 * @bio_ret: must be valid pointer, newly allocated bio will be stored there
3291 * @end_io_func: end_io callback for new bio
3292 * @mirror_num: desired mirror to read/write
3293 * @prev_bio_flags: flags of previous bio to see if we can merge the current one
3294 * @bio_flags: flags of the current bio to see if we can merge them
3296 static int submit_extent_page(unsigned int opf,
3297 struct writeback_control *wbc,
3298 struct btrfs_bio_ctrl *bio_ctrl,
3299 struct page *page, u64 disk_bytenr,
3300 size_t size, unsigned long pg_offset,
3301 bio_end_io_t end_io_func,
3303 unsigned long bio_flags,
3304 bool force_bio_submit)
3308 size_t io_size = min_t(size_t, size, PAGE_SIZE);
3309 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
3310 struct extent_io_tree *tree = &inode->io_tree;
3311 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3315 ASSERT(pg_offset < PAGE_SIZE && size <= PAGE_SIZE &&
3316 pg_offset + size <= PAGE_SIZE);
3317 if (bio_ctrl->bio) {
3318 bio = bio_ctrl->bio;
3319 if (force_bio_submit ||
3320 !btrfs_bio_add_page(bio_ctrl, page, disk_bytenr, io_size,
3321 pg_offset, bio_flags)) {
3322 ret = submit_one_bio(bio, mirror_num, bio_ctrl->bio_flags);
3323 bio_ctrl->bio = NULL;
3328 wbc_account_cgroup_owner(wbc, page, io_size);
3333 bio = btrfs_bio_alloc(disk_bytenr);
3334 bio_add_page(bio, page, io_size, pg_offset);
3335 bio->bi_end_io = end_io_func;
3336 bio->bi_private = tree;
3337 bio->bi_write_hint = page->mapping->host->i_write_hint;
3340 struct block_device *bdev;
3342 bdev = fs_info->fs_devices->latest_bdev;
3343 bio_set_dev(bio, bdev);
3344 wbc_init_bio(wbc, bio);
3345 wbc_account_cgroup_owner(wbc, page, io_size);
3347 if (btrfs_is_zoned(fs_info) && bio_op(bio) == REQ_OP_ZONE_APPEND) {
3348 struct btrfs_device *device;
3350 device = btrfs_zoned_get_device(fs_info, disk_bytenr, io_size);
3352 return PTR_ERR(device);
3354 btrfs_io_bio(bio)->device = device;
3357 bio_ctrl->bio = bio;
3358 bio_ctrl->bio_flags = bio_flags;
3359 ret = calc_bio_boundaries(bio_ctrl, inode);
3364 static int attach_extent_buffer_page(struct extent_buffer *eb,
3366 struct btrfs_subpage *prealloc)
3368 struct btrfs_fs_info *fs_info = eb->fs_info;
3372 * If the page is mapped to btree inode, we should hold the private
3373 * lock to prevent race.
3374 * For cloned or dummy extent buffers, their pages are not mapped and
3375 * will not race with any other ebs.
3378 lockdep_assert_held(&page->mapping->private_lock);
3380 if (fs_info->sectorsize == PAGE_SIZE) {
3381 if (!PagePrivate(page))
3382 attach_page_private(page, eb);
3384 WARN_ON(page->private != (unsigned long)eb);
3388 /* Already mapped, just free prealloc */
3389 if (PagePrivate(page)) {
3390 btrfs_free_subpage(prealloc);
3395 /* Has preallocated memory for subpage */
3396 attach_page_private(page, prealloc);
3398 /* Do new allocation to attach subpage */
3399 ret = btrfs_attach_subpage(fs_info, page,
3400 BTRFS_SUBPAGE_METADATA);
3404 int set_page_extent_mapped(struct page *page)
3406 struct btrfs_fs_info *fs_info;
3408 ASSERT(page->mapping);
3410 if (PagePrivate(page))
3413 fs_info = btrfs_sb(page->mapping->host->i_sb);
3415 if (fs_info->sectorsize < PAGE_SIZE)
3416 return btrfs_attach_subpage(fs_info, page, BTRFS_SUBPAGE_DATA);
3418 attach_page_private(page, (void *)EXTENT_PAGE_PRIVATE);
3422 void clear_page_extent_mapped(struct page *page)
3424 struct btrfs_fs_info *fs_info;
3426 ASSERT(page->mapping);
3428 if (!PagePrivate(page))
3431 fs_info = btrfs_sb(page->mapping->host->i_sb);
3432 if (fs_info->sectorsize < PAGE_SIZE)
3433 return btrfs_detach_subpage(fs_info, page);
3435 detach_page_private(page);
3438 static struct extent_map *
3439 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
3440 u64 start, u64 len, struct extent_map **em_cached)
3442 struct extent_map *em;
3444 if (em_cached && *em_cached) {
3446 if (extent_map_in_tree(em) && start >= em->start &&
3447 start < extent_map_end(em)) {
3448 refcount_inc(&em->refs);
3452 free_extent_map(em);
3456 em = btrfs_get_extent(BTRFS_I(inode), page, pg_offset, start, len);
3457 if (em_cached && !IS_ERR_OR_NULL(em)) {
3459 refcount_inc(&em->refs);
3465 * basic readpage implementation. Locked extent state structs are inserted
3466 * into the tree that are removed when the IO is done (by the end_io
3468 * XXX JDM: This needs looking at to ensure proper page locking
3469 * return 0 on success, otherwise return error
3471 int btrfs_do_readpage(struct page *page, struct extent_map **em_cached,
3472 struct btrfs_bio_ctrl *bio_ctrl,
3473 unsigned int read_flags, u64 *prev_em_start)
3475 struct inode *inode = page->mapping->host;
3476 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3477 u64 start = page_offset(page);
3478 const u64 end = start + PAGE_SIZE - 1;
3481 u64 last_byte = i_size_read(inode);
3484 struct extent_map *em;
3487 size_t pg_offset = 0;
3489 size_t blocksize = inode->i_sb->s_blocksize;
3490 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
3492 ret = set_page_extent_mapped(page);
3494 unlock_extent(tree, start, end);
3495 btrfs_page_set_error(fs_info, page, start, PAGE_SIZE);
3500 if (!PageUptodate(page)) {
3501 if (cleancache_get_page(page) == 0) {
3502 BUG_ON(blocksize != PAGE_SIZE);
3503 unlock_extent(tree, start, end);
3509 if (page->index == last_byte >> PAGE_SHIFT) {
3510 size_t zero_offset = offset_in_page(last_byte);
3513 iosize = PAGE_SIZE - zero_offset;
3514 memzero_page(page, zero_offset, iosize);
3515 flush_dcache_page(page);
3518 begin_page_read(fs_info, page);
3519 while (cur <= end) {
3520 unsigned long this_bio_flag = 0;
3521 bool force_bio_submit = false;
3524 if (cur >= last_byte) {
3525 struct extent_state *cached = NULL;
3527 iosize = PAGE_SIZE - pg_offset;
3528 memzero_page(page, pg_offset, iosize);
3529 flush_dcache_page(page);
3530 set_extent_uptodate(tree, cur, cur + iosize - 1,
3532 unlock_extent_cached(tree, cur,
3533 cur + iosize - 1, &cached);
3534 end_page_read(page, true, cur, iosize);
3537 em = __get_extent_map(inode, page, pg_offset, cur,
3538 end - cur + 1, em_cached);
3539 if (IS_ERR_OR_NULL(em)) {
3540 unlock_extent(tree, cur, end);
3541 end_page_read(page, false, cur, end + 1 - cur);
3544 extent_offset = cur - em->start;
3545 BUG_ON(extent_map_end(em) <= cur);
3548 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
3549 this_bio_flag |= EXTENT_BIO_COMPRESSED;
3550 extent_set_compress_type(&this_bio_flag,
3554 iosize = min(extent_map_end(em) - cur, end - cur + 1);
3555 cur_end = min(extent_map_end(em) - 1, end);
3556 iosize = ALIGN(iosize, blocksize);
3557 if (this_bio_flag & EXTENT_BIO_COMPRESSED)
3558 disk_bytenr = em->block_start;
3560 disk_bytenr = em->block_start + extent_offset;
3561 block_start = em->block_start;
3562 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
3563 block_start = EXTENT_MAP_HOLE;
3566 * If we have a file range that points to a compressed extent
3567 * and it's followed by a consecutive file range that points
3568 * to the same compressed extent (possibly with a different
3569 * offset and/or length, so it either points to the whole extent
3570 * or only part of it), we must make sure we do not submit a
3571 * single bio to populate the pages for the 2 ranges because
3572 * this makes the compressed extent read zero out the pages
3573 * belonging to the 2nd range. Imagine the following scenario:
3576 * [0 - 8K] [8K - 24K]
3579 * points to extent X, points to extent X,
3580 * offset 4K, length of 8K offset 0, length 16K
3582 * [extent X, compressed length = 4K uncompressed length = 16K]
3584 * If the bio to read the compressed extent covers both ranges,
3585 * it will decompress extent X into the pages belonging to the
3586 * first range and then it will stop, zeroing out the remaining
3587 * pages that belong to the other range that points to extent X.
3588 * So here we make sure we submit 2 bios, one for the first
3589 * range and another one for the third range. Both will target
3590 * the same physical extent from disk, but we can't currently
3591 * make the compressed bio endio callback populate the pages
3592 * for both ranges because each compressed bio is tightly
3593 * coupled with a single extent map, and each range can have
3594 * an extent map with a different offset value relative to the
3595 * uncompressed data of our extent and different lengths. This
3596 * is a corner case so we prioritize correctness over
3597 * non-optimal behavior (submitting 2 bios for the same extent).
3599 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) &&
3600 prev_em_start && *prev_em_start != (u64)-1 &&
3601 *prev_em_start != em->start)
3602 force_bio_submit = true;
3605 *prev_em_start = em->start;
3607 free_extent_map(em);
3610 /* we've found a hole, just zero and go on */
3611 if (block_start == EXTENT_MAP_HOLE) {
3612 struct extent_state *cached = NULL;
3614 memzero_page(page, pg_offset, iosize);
3615 flush_dcache_page(page);
3617 set_extent_uptodate(tree, cur, cur + iosize - 1,
3619 unlock_extent_cached(tree, cur,
3620 cur + iosize - 1, &cached);
3621 end_page_read(page, true, cur, iosize);
3623 pg_offset += iosize;
3626 /* the get_extent function already copied into the page */
3627 if (test_range_bit(tree, cur, cur_end,
3628 EXTENT_UPTODATE, 1, NULL)) {
3629 check_page_uptodate(tree, page);
3630 unlock_extent(tree, cur, cur + iosize - 1);
3631 end_page_read(page, true, cur, iosize);
3633 pg_offset += iosize;
3636 /* we have an inline extent but it didn't get marked up
3637 * to date. Error out
3639 if (block_start == EXTENT_MAP_INLINE) {
3640 unlock_extent(tree, cur, cur + iosize - 1);
3641 end_page_read(page, false, cur, iosize);
3643 pg_offset += iosize;
3647 ret = submit_extent_page(REQ_OP_READ | read_flags, NULL,
3648 bio_ctrl, page, disk_bytenr, iosize,
3650 end_bio_extent_readpage, 0,
3656 unlock_extent(tree, cur, cur + iosize - 1);
3657 end_page_read(page, false, cur, iosize);
3661 pg_offset += iosize;
3667 static inline void contiguous_readpages(struct page *pages[], int nr_pages,
3669 struct extent_map **em_cached,
3670 struct btrfs_bio_ctrl *bio_ctrl,
3673 struct btrfs_inode *inode = BTRFS_I(pages[0]->mapping->host);
3676 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
3678 for (index = 0; index < nr_pages; index++) {
3679 btrfs_do_readpage(pages[index], em_cached, bio_ctrl,
3680 REQ_RAHEAD, prev_em_start);
3681 put_page(pages[index]);
3685 static void update_nr_written(struct writeback_control *wbc,
3686 unsigned long nr_written)
3688 wbc->nr_to_write -= nr_written;
3692 * helper for __extent_writepage, doing all of the delayed allocation setup.
3694 * This returns 1 if btrfs_run_delalloc_range function did all the work required
3695 * to write the page (copy into inline extent). In this case the IO has
3696 * been started and the page is already unlocked.
3698 * This returns 0 if all went well (page still locked)
3699 * This returns < 0 if there were errors (page still locked)
3701 static noinline_for_stack int writepage_delalloc(struct btrfs_inode *inode,
3702 struct page *page, struct writeback_control *wbc,
3703 u64 delalloc_start, unsigned long *nr_written)
3705 u64 page_end = delalloc_start + PAGE_SIZE - 1;
3707 u64 delalloc_to_write = 0;
3708 u64 delalloc_end = 0;
3710 int page_started = 0;
3713 while (delalloc_end < page_end) {
3714 found = find_lock_delalloc_range(&inode->vfs_inode, page,
3718 delalloc_start = delalloc_end + 1;
3721 ret = btrfs_run_delalloc_range(inode, page, delalloc_start,
3722 delalloc_end, &page_started, nr_written, wbc);
3726 * btrfs_run_delalloc_range should return < 0 for error
3727 * but just in case, we use > 0 here meaning the IO is
3728 * started, so we don't want to return > 0 unless
3729 * things are going well.
3731 return ret < 0 ? ret : -EIO;
3734 * delalloc_end is already one less than the total length, so
3735 * we don't subtract one from PAGE_SIZE
3737 delalloc_to_write += (delalloc_end - delalloc_start +
3738 PAGE_SIZE) >> PAGE_SHIFT;
3739 delalloc_start = delalloc_end + 1;
3741 if (wbc->nr_to_write < delalloc_to_write) {
3744 if (delalloc_to_write < thresh * 2)
3745 thresh = delalloc_to_write;
3746 wbc->nr_to_write = min_t(u64, delalloc_to_write,
3750 /* did the fill delalloc function already unlock and start
3755 * we've unlocked the page, so we can't update
3756 * the mapping's writeback index, just update
3759 wbc->nr_to_write -= *nr_written;
3767 * Find the first byte we need to write.
3769 * For subpage, one page can contain several sectors, and
3770 * __extent_writepage_io() will just grab all extent maps in the page
3771 * range and try to submit all non-inline/non-compressed extents.
3773 * This is a big problem for subpage, we shouldn't re-submit already written
3775 * This function will lookup subpage dirty bit to find which range we really
3778 * Return the next dirty range in [@start, @end).
3779 * If no dirty range is found, @start will be page_offset(page) + PAGE_SIZE.
3781 static void find_next_dirty_byte(struct btrfs_fs_info *fs_info,
3782 struct page *page, u64 *start, u64 *end)
3784 struct btrfs_subpage *subpage = (struct btrfs_subpage *)page->private;
3785 u64 orig_start = *start;
3786 /* Declare as unsigned long so we can use bitmap ops */
3787 unsigned long dirty_bitmap;
3788 unsigned long flags;
3789 int nbits = (orig_start - page_offset(page)) >> fs_info->sectorsize_bits;
3790 int range_start_bit = nbits;
3794 * For regular sector size == page size case, since one page only
3795 * contains one sector, we return the page offset directly.
3797 if (fs_info->sectorsize == PAGE_SIZE) {
3798 *start = page_offset(page);
3799 *end = page_offset(page) + PAGE_SIZE;
3803 /* We should have the page locked, but just in case */
3804 spin_lock_irqsave(&subpage->lock, flags);
3805 dirty_bitmap = subpage->dirty_bitmap;
3806 spin_unlock_irqrestore(&subpage->lock, flags);
3808 bitmap_next_set_region(&dirty_bitmap, &range_start_bit, &range_end_bit,
3809 BTRFS_SUBPAGE_BITMAP_SIZE);
3810 *start = page_offset(page) + range_start_bit * fs_info->sectorsize;
3811 *end = page_offset(page) + range_end_bit * fs_info->sectorsize;
3815 * helper for __extent_writepage. This calls the writepage start hooks,
3816 * and does the loop to map the page into extents and bios.
3818 * We return 1 if the IO is started and the page is unlocked,
3819 * 0 if all went well (page still locked)
3820 * < 0 if there were errors (page still locked)
3822 static noinline_for_stack int __extent_writepage_io(struct btrfs_inode *inode,
3824 struct writeback_control *wbc,
3825 struct extent_page_data *epd,
3827 unsigned long nr_written,
3830 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3831 u64 cur = page_offset(page);
3832 u64 end = cur + PAGE_SIZE - 1;
3835 struct extent_map *em;
3838 u32 opf = REQ_OP_WRITE;
3839 const unsigned int write_flags = wbc_to_write_flags(wbc);
3842 ret = btrfs_writepage_cow_fixup(page);
3844 /* Fixup worker will requeue */
3845 redirty_page_for_writepage(wbc, page);
3846 update_nr_written(wbc, nr_written);
3852 * we don't want to touch the inode after unlocking the page,
3853 * so we update the mapping writeback index now
3855 update_nr_written(wbc, nr_written + 1);
3857 while (cur <= end) {
3860 u64 dirty_range_start = cur;
3861 u64 dirty_range_end;
3864 if (cur >= i_size) {
3865 btrfs_writepage_endio_finish_ordered(inode, page, cur,
3870 find_next_dirty_byte(fs_info, page, &dirty_range_start,
3872 if (cur < dirty_range_start) {
3873 cur = dirty_range_start;
3877 em = btrfs_get_extent(inode, NULL, 0, cur, end - cur + 1);
3878 if (IS_ERR_OR_NULL(em)) {
3879 btrfs_page_set_error(fs_info, page, cur, end - cur + 1);
3880 ret = PTR_ERR_OR_ZERO(em);
3884 extent_offset = cur - em->start;
3885 em_end = extent_map_end(em);
3886 ASSERT(cur <= em_end);
3888 ASSERT(IS_ALIGNED(em->start, fs_info->sectorsize));
3889 ASSERT(IS_ALIGNED(em->len, fs_info->sectorsize));
3890 block_start = em->block_start;
3891 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3892 disk_bytenr = em->block_start + extent_offset;
3895 * Note that em_end from extent_map_end() and dirty_range_end from
3896 * find_next_dirty_byte() are all exclusive
3898 iosize = min(min(em_end, end + 1), dirty_range_end) - cur;
3900 if (btrfs_use_zone_append(inode, em->block_start))
3901 opf = REQ_OP_ZONE_APPEND;
3903 free_extent_map(em);
3907 * compressed and inline extents are written through other
3910 if (compressed || block_start == EXTENT_MAP_HOLE ||
3911 block_start == EXTENT_MAP_INLINE) {
3915 btrfs_writepage_endio_finish_ordered(inode,
3916 page, cur, cur + iosize - 1, true);
3921 btrfs_set_range_writeback(inode, cur, cur + iosize - 1);
3922 if (!PageWriteback(page)) {
3923 btrfs_err(inode->root->fs_info,
3924 "page %lu not writeback, cur %llu end %llu",
3925 page->index, cur, end);
3929 * Although the PageDirty bit is cleared before entering this
3930 * function, subpage dirty bit is not cleared.
3931 * So clear subpage dirty bit here so next time we won't submit
3932 * page for range already written to disk.
3934 btrfs_page_clear_dirty(fs_info, page, cur, iosize);
3936 ret = submit_extent_page(opf | write_flags, wbc,
3937 &epd->bio_ctrl, page,
3938 disk_bytenr, iosize,
3939 cur - page_offset(page),
3940 end_bio_extent_writepage,
3943 btrfs_page_set_error(fs_info, page, cur, iosize);
3944 if (PageWriteback(page))
3945 btrfs_page_clear_writeback(fs_info, page, cur,
3957 * the writepage semantics are similar to regular writepage. extent
3958 * records are inserted to lock ranges in the tree, and as dirty areas
3959 * are found, they are marked writeback. Then the lock bits are removed
3960 * and the end_io handler clears the writeback ranges
3962 * Return 0 if everything goes well.
3963 * Return <0 for error.
3965 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
3966 struct extent_page_data *epd)
3968 struct inode *inode = page->mapping->host;
3969 u64 start = page_offset(page);
3970 u64 page_end = start + PAGE_SIZE - 1;
3974 loff_t i_size = i_size_read(inode);
3975 unsigned long end_index = i_size >> PAGE_SHIFT;
3976 unsigned long nr_written = 0;
3978 trace___extent_writepage(page, inode, wbc);
3980 WARN_ON(!PageLocked(page));
3982 ClearPageError(page);
3984 pg_offset = offset_in_page(i_size);
3985 if (page->index > end_index ||
3986 (page->index == end_index && !pg_offset)) {
3987 page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
3992 if (page->index == end_index) {
3993 memzero_page(page, pg_offset, PAGE_SIZE - pg_offset);
3994 flush_dcache_page(page);
3997 ret = set_page_extent_mapped(page);
4003 if (!epd->extent_locked) {
4004 ret = writepage_delalloc(BTRFS_I(inode), page, wbc, start,
4012 ret = __extent_writepage_io(BTRFS_I(inode), page, wbc, epd, i_size,
4019 /* make sure the mapping tag for page dirty gets cleared */
4020 set_page_writeback(page);
4021 end_page_writeback(page);
4023 if (PageError(page)) {
4024 ret = ret < 0 ? ret : -EIO;
4025 end_extent_writepage(page, ret, start, page_end);
4032 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
4034 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
4035 TASK_UNINTERRUPTIBLE);
4038 static void end_extent_buffer_writeback(struct extent_buffer *eb)
4040 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
4041 smp_mb__after_atomic();
4042 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
4046 * Lock extent buffer status and pages for writeback.
4048 * May try to flush write bio if we can't get the lock.
4050 * Return 0 if the extent buffer doesn't need to be submitted.
4051 * (E.g. the extent buffer is not dirty)
4052 * Return >0 is the extent buffer is submitted to bio.
4053 * Return <0 if something went wrong, no page is locked.
4055 static noinline_for_stack int lock_extent_buffer_for_io(struct extent_buffer *eb,
4056 struct extent_page_data *epd)
4058 struct btrfs_fs_info *fs_info = eb->fs_info;
4059 int i, num_pages, failed_page_nr;
4063 if (!btrfs_try_tree_write_lock(eb)) {
4064 ret = flush_write_bio(epd);
4068 btrfs_tree_lock(eb);
4071 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
4072 btrfs_tree_unlock(eb);
4076 ret = flush_write_bio(epd);
4082 wait_on_extent_buffer_writeback(eb);
4083 btrfs_tree_lock(eb);
4084 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
4086 btrfs_tree_unlock(eb);
4091 * We need to do this to prevent races in people who check if the eb is
4092 * under IO since we can end up having no IO bits set for a short period
4095 spin_lock(&eb->refs_lock);
4096 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
4097 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
4098 spin_unlock(&eb->refs_lock);
4099 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
4100 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4102 fs_info->dirty_metadata_batch);
4105 spin_unlock(&eb->refs_lock);
4108 btrfs_tree_unlock(eb);
4111 * Either we don't need to submit any tree block, or we're submitting
4113 * Subpage metadata doesn't use page locking at all, so we can skip
4116 if (!ret || fs_info->sectorsize < PAGE_SIZE)
4119 num_pages = num_extent_pages(eb);
4120 for (i = 0; i < num_pages; i++) {
4121 struct page *p = eb->pages[i];
4123 if (!trylock_page(p)) {
4127 err = flush_write_bio(epd);
4141 /* Unlock already locked pages */
4142 for (i = 0; i < failed_page_nr; i++)
4143 unlock_page(eb->pages[i]);
4145 * Clear EXTENT_BUFFER_WRITEBACK and wake up anyone waiting on it.
4146 * Also set back EXTENT_BUFFER_DIRTY so future attempts to this eb can
4147 * be made and undo everything done before.
4149 btrfs_tree_lock(eb);
4150 spin_lock(&eb->refs_lock);
4151 set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
4152 end_extent_buffer_writeback(eb);
4153 spin_unlock(&eb->refs_lock);
4154 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, eb->len,
4155 fs_info->dirty_metadata_batch);
4156 btrfs_clear_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
4157 btrfs_tree_unlock(eb);
4161 static void set_btree_ioerr(struct page *page, struct extent_buffer *eb)
4163 struct btrfs_fs_info *fs_info = eb->fs_info;
4165 btrfs_page_set_error(fs_info, page, eb->start, eb->len);
4166 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))
4170 * If we error out, we should add back the dirty_metadata_bytes
4171 * to make it consistent.
4173 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4174 eb->len, fs_info->dirty_metadata_batch);
4177 * If writeback for a btree extent that doesn't belong to a log tree
4178 * failed, increment the counter transaction->eb_write_errors.
4179 * We do this because while the transaction is running and before it's
4180 * committing (when we call filemap_fdata[write|wait]_range against
4181 * the btree inode), we might have
4182 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
4183 * returns an error or an error happens during writeback, when we're
4184 * committing the transaction we wouldn't know about it, since the pages
4185 * can be no longer dirty nor marked anymore for writeback (if a
4186 * subsequent modification to the extent buffer didn't happen before the
4187 * transaction commit), which makes filemap_fdata[write|wait]_range not
4188 * able to find the pages tagged with SetPageError at transaction
4189 * commit time. So if this happens we must abort the transaction,
4190 * otherwise we commit a super block with btree roots that point to
4191 * btree nodes/leafs whose content on disk is invalid - either garbage
4192 * or the content of some node/leaf from a past generation that got
4193 * cowed or deleted and is no longer valid.
4195 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
4196 * not be enough - we need to distinguish between log tree extents vs
4197 * non-log tree extents, and the next filemap_fdatawait_range() call
4198 * will catch and clear such errors in the mapping - and that call might
4199 * be from a log sync and not from a transaction commit. Also, checking
4200 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
4201 * not done and would not be reliable - the eb might have been released
4202 * from memory and reading it back again means that flag would not be
4203 * set (since it's a runtime flag, not persisted on disk).
4205 * Using the flags below in the btree inode also makes us achieve the
4206 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
4207 * writeback for all dirty pages and before filemap_fdatawait_range()
4208 * is called, the writeback for all dirty pages had already finished
4209 * with errors - because we were not using AS_EIO/AS_ENOSPC,
4210 * filemap_fdatawait_range() would return success, as it could not know
4211 * that writeback errors happened (the pages were no longer tagged for
4214 switch (eb->log_index) {
4216 set_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags);
4219 set_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags);
4222 set_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags);
4225 BUG(); /* unexpected, logic error */
4230 * The endio specific version which won't touch any unsafe spinlock in endio
4233 static struct extent_buffer *find_extent_buffer_nolock(
4234 struct btrfs_fs_info *fs_info, u64 start)
4236 struct extent_buffer *eb;
4239 eb = radix_tree_lookup(&fs_info->buffer_radix,
4240 start >> fs_info->sectorsize_bits);
4241 if (eb && atomic_inc_not_zero(&eb->refs)) {
4250 * The endio function for subpage extent buffer write.
4252 * Unlike end_bio_extent_buffer_writepage(), we only call end_page_writeback()
4253 * after all extent buffers in the page has finished their writeback.
4255 static void end_bio_subpage_eb_writepage(struct bio *bio)
4257 struct btrfs_fs_info *fs_info;
4258 struct bio_vec *bvec;
4259 struct bvec_iter_all iter_all;
4261 fs_info = btrfs_sb(bio_first_page_all(bio)->mapping->host->i_sb);
4262 ASSERT(fs_info->sectorsize < PAGE_SIZE);
4264 ASSERT(!bio_flagged(bio, BIO_CLONED));
4265 bio_for_each_segment_all(bvec, bio, iter_all) {
4266 struct page *page = bvec->bv_page;
4267 u64 bvec_start = page_offset(page) + bvec->bv_offset;
4268 u64 bvec_end = bvec_start + bvec->bv_len - 1;
4269 u64 cur_bytenr = bvec_start;
4271 ASSERT(IS_ALIGNED(bvec->bv_len, fs_info->nodesize));
4273 /* Iterate through all extent buffers in the range */
4274 while (cur_bytenr <= bvec_end) {
4275 struct extent_buffer *eb;
4279 * Here we can't use find_extent_buffer(), as it may
4280 * try to lock eb->refs_lock, which is not safe in endio
4283 eb = find_extent_buffer_nolock(fs_info, cur_bytenr);
4286 cur_bytenr = eb->start + eb->len;
4288 ASSERT(test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags));
4289 done = atomic_dec_and_test(&eb->io_pages);
4292 if (bio->bi_status ||
4293 test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) {
4294 ClearPageUptodate(page);
4295 set_btree_ioerr(page, eb);
4298 btrfs_subpage_clear_writeback(fs_info, page, eb->start,
4300 end_extent_buffer_writeback(eb);
4302 * free_extent_buffer() will grab spinlock which is not
4303 * safe in endio context. Thus here we manually dec
4306 atomic_dec(&eb->refs);
4312 static void end_bio_extent_buffer_writepage(struct bio *bio)
4314 struct bio_vec *bvec;
4315 struct extent_buffer *eb;
4317 struct bvec_iter_all iter_all;
4319 ASSERT(!bio_flagged(bio, BIO_CLONED));
4320 bio_for_each_segment_all(bvec, bio, iter_all) {
4321 struct page *page = bvec->bv_page;
4323 eb = (struct extent_buffer *)page->private;
4325 done = atomic_dec_and_test(&eb->io_pages);
4327 if (bio->bi_status ||
4328 test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) {
4329 ClearPageUptodate(page);
4330 set_btree_ioerr(page, eb);
4333 end_page_writeback(page);
4338 end_extent_buffer_writeback(eb);
4344 static void prepare_eb_write(struct extent_buffer *eb)
4347 unsigned long start;
4350 clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
4351 atomic_set(&eb->io_pages, num_extent_pages(eb));
4353 /* Set btree blocks beyond nritems with 0 to avoid stale content */
4354 nritems = btrfs_header_nritems(eb);
4355 if (btrfs_header_level(eb) > 0) {
4356 end = btrfs_node_key_ptr_offset(nritems);
4357 memzero_extent_buffer(eb, end, eb->len - end);
4361 * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0
4363 start = btrfs_item_nr_offset(nritems);
4364 end = BTRFS_LEAF_DATA_OFFSET + leaf_data_end(eb);
4365 memzero_extent_buffer(eb, start, end - start);
4370 * Unlike the work in write_one_eb(), we rely completely on extent locking.
4371 * Page locking is only utilized at minimum to keep the VMM code happy.
4373 static int write_one_subpage_eb(struct extent_buffer *eb,
4374 struct writeback_control *wbc,
4375 struct extent_page_data *epd)
4377 struct btrfs_fs_info *fs_info = eb->fs_info;
4378 struct page *page = eb->pages[0];
4379 unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META;
4380 bool no_dirty_ebs = false;
4383 prepare_eb_write(eb);
4385 /* clear_page_dirty_for_io() in subpage helper needs page locked */
4387 btrfs_subpage_set_writeback(fs_info, page, eb->start, eb->len);
4389 /* Check if this is the last dirty bit to update nr_written */
4390 no_dirty_ebs = btrfs_subpage_clear_and_test_dirty(fs_info, page,
4391 eb->start, eb->len);
4393 clear_page_dirty_for_io(page);
4395 ret = submit_extent_page(REQ_OP_WRITE | write_flags, wbc,
4396 &epd->bio_ctrl, page, eb->start, eb->len,
4397 eb->start - page_offset(page),
4398 end_bio_subpage_eb_writepage, 0, 0, false);
4400 btrfs_subpage_clear_writeback(fs_info, page, eb->start, eb->len);
4401 set_btree_ioerr(page, eb);
4404 if (atomic_dec_and_test(&eb->io_pages))
4405 end_extent_buffer_writeback(eb);
4410 * Submission finished without problem, if no range of the page is
4411 * dirty anymore, we have submitted a page. Update nr_written in wbc.
4414 update_nr_written(wbc, 1);
4418 static noinline_for_stack int write_one_eb(struct extent_buffer *eb,
4419 struct writeback_control *wbc,
4420 struct extent_page_data *epd)
4422 u64 disk_bytenr = eb->start;
4424 unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META;
4427 prepare_eb_write(eb);
4429 num_pages = num_extent_pages(eb);
4430 for (i = 0; i < num_pages; i++) {
4431 struct page *p = eb->pages[i];
4433 clear_page_dirty_for_io(p);
4434 set_page_writeback(p);
4435 ret = submit_extent_page(REQ_OP_WRITE | write_flags, wbc,
4436 &epd->bio_ctrl, p, disk_bytenr,
4438 end_bio_extent_buffer_writepage,
4441 set_btree_ioerr(p, eb);
4442 if (PageWriteback(p))
4443 end_page_writeback(p);
4444 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
4445 end_extent_buffer_writeback(eb);
4449 disk_bytenr += PAGE_SIZE;
4450 update_nr_written(wbc, 1);
4454 if (unlikely(ret)) {
4455 for (; i < num_pages; i++) {
4456 struct page *p = eb->pages[i];
4457 clear_page_dirty_for_io(p);
4466 * Submit one subpage btree page.
4468 * The main difference to submit_eb_page() is:
4470 * For subpage, we don't rely on page locking at all.
4473 * We only flush bio if we may be unable to fit current extent buffers into
4476 * Return >=0 for the number of submitted extent buffers.
4477 * Return <0 for fatal error.
4479 static int submit_eb_subpage(struct page *page,
4480 struct writeback_control *wbc,
4481 struct extent_page_data *epd)
4483 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
4485 u64 page_start = page_offset(page);
4487 const int nbits = BTRFS_SUBPAGE_BITMAP_SIZE;
4488 int sectors_per_node = fs_info->nodesize >> fs_info->sectorsize_bits;
4491 /* Lock and write each dirty extent buffers in the range */
4492 while (bit_start < nbits) {
4493 struct btrfs_subpage *subpage = (struct btrfs_subpage *)page->private;
4494 struct extent_buffer *eb;
4495 unsigned long flags;
4499 * Take private lock to ensure the subpage won't be detached
4502 spin_lock(&page->mapping->private_lock);
4503 if (!PagePrivate(page)) {
4504 spin_unlock(&page->mapping->private_lock);
4507 spin_lock_irqsave(&subpage->lock, flags);
4508 if (!((1 << bit_start) & subpage->dirty_bitmap)) {
4509 spin_unlock_irqrestore(&subpage->lock, flags);
4510 spin_unlock(&page->mapping->private_lock);
4515 start = page_start + bit_start * fs_info->sectorsize;
4516 bit_start += sectors_per_node;
4519 * Here we just want to grab the eb without touching extra
4520 * spin locks, so call find_extent_buffer_nolock().
4522 eb = find_extent_buffer_nolock(fs_info, start);
4523 spin_unlock_irqrestore(&subpage->lock, flags);
4524 spin_unlock(&page->mapping->private_lock);
4527 * The eb has already reached 0 refs thus find_extent_buffer()
4528 * doesn't return it. We don't need to write back such eb
4534 ret = lock_extent_buffer_for_io(eb, epd);
4536 free_extent_buffer(eb);
4540 free_extent_buffer(eb);
4543 ret = write_one_subpage_eb(eb, wbc, epd);
4544 free_extent_buffer(eb);
4552 /* We hit error, end bio for the submitted extent buffers */
4553 end_write_bio(epd, ret);
4558 * Submit all page(s) of one extent buffer.
4560 * @page: the page of one extent buffer
4561 * @eb_context: to determine if we need to submit this page, if current page
4562 * belongs to this eb, we don't need to submit
4564 * The caller should pass each page in their bytenr order, and here we use
4565 * @eb_context to determine if we have submitted pages of one extent buffer.
4567 * If we have, we just skip until we hit a new page that doesn't belong to
4568 * current @eb_context.
4570 * If not, we submit all the page(s) of the extent buffer.
4572 * Return >0 if we have submitted the extent buffer successfully.
4573 * Return 0 if we don't need to submit the page, as it's already submitted by
4575 * Return <0 for fatal error.
4577 static int submit_eb_page(struct page *page, struct writeback_control *wbc,
4578 struct extent_page_data *epd,
4579 struct extent_buffer **eb_context)
4581 struct address_space *mapping = page->mapping;
4582 struct btrfs_block_group *cache = NULL;
4583 struct extent_buffer *eb;
4586 if (!PagePrivate(page))
4589 if (btrfs_sb(page->mapping->host->i_sb)->sectorsize < PAGE_SIZE)
4590 return submit_eb_subpage(page, wbc, epd);
4592 spin_lock(&mapping->private_lock);
4593 if (!PagePrivate(page)) {
4594 spin_unlock(&mapping->private_lock);
4598 eb = (struct extent_buffer *)page->private;
4601 * Shouldn't happen and normally this would be a BUG_ON but no point
4602 * crashing the machine for something we can survive anyway.
4605 spin_unlock(&mapping->private_lock);
4609 if (eb == *eb_context) {
4610 spin_unlock(&mapping->private_lock);
4613 ret = atomic_inc_not_zero(&eb->refs);
4614 spin_unlock(&mapping->private_lock);
4618 if (!btrfs_check_meta_write_pointer(eb->fs_info, eb, &cache)) {
4620 * If for_sync, this hole will be filled with
4621 * trasnsaction commit.
4623 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
4627 free_extent_buffer(eb);
4633 ret = lock_extent_buffer_for_io(eb, epd);
4635 btrfs_revert_meta_write_pointer(cache, eb);
4637 btrfs_put_block_group(cache);
4638 free_extent_buffer(eb);
4642 btrfs_put_block_group(cache);
4643 ret = write_one_eb(eb, wbc, epd);
4644 free_extent_buffer(eb);
4650 int btree_write_cache_pages(struct address_space *mapping,
4651 struct writeback_control *wbc)
4653 struct extent_buffer *eb_context = NULL;
4654 struct extent_page_data epd = {
4657 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4659 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
4662 int nr_to_write_done = 0;
4663 struct pagevec pvec;
4666 pgoff_t end; /* Inclusive */
4670 pagevec_init(&pvec);
4671 if (wbc->range_cyclic) {
4672 index = mapping->writeback_index; /* Start from prev offset */
4675 * Start from the beginning does not need to cycle over the
4676 * range, mark it as scanned.
4678 scanned = (index == 0);
4680 index = wbc->range_start >> PAGE_SHIFT;
4681 end = wbc->range_end >> PAGE_SHIFT;
4684 if (wbc->sync_mode == WB_SYNC_ALL)
4685 tag = PAGECACHE_TAG_TOWRITE;
4687 tag = PAGECACHE_TAG_DIRTY;
4688 btrfs_zoned_meta_io_lock(fs_info);
4690 if (wbc->sync_mode == WB_SYNC_ALL)
4691 tag_pages_for_writeback(mapping, index, end);
4692 while (!done && !nr_to_write_done && (index <= end) &&
4693 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
4697 for (i = 0; i < nr_pages; i++) {
4698 struct page *page = pvec.pages[i];
4700 ret = submit_eb_page(page, wbc, &epd, &eb_context);
4709 * the filesystem may choose to bump up nr_to_write.
4710 * We have to make sure to honor the new nr_to_write
4713 nr_to_write_done = wbc->nr_to_write <= 0;
4715 pagevec_release(&pvec);
4718 if (!scanned && !done) {
4720 * We hit the last page and there is more work to be done: wrap
4721 * back to the start of the file
4728 end_write_bio(&epd, ret);
4732 * If something went wrong, don't allow any metadata write bio to be
4735 * This would prevent use-after-free if we had dirty pages not
4736 * cleaned up, which can still happen by fuzzed images.
4739 * Allowing existing tree block to be allocated for other trees.
4741 * - Log tree operations
4742 * Exiting tree blocks get allocated to log tree, bumps its
4743 * generation, then get cleaned in tree re-balance.
4744 * Such tree block will not be written back, since it's clean,
4745 * thus no WRITTEN flag set.
4746 * And after log writes back, this tree block is not traced by
4747 * any dirty extent_io_tree.
4749 * - Offending tree block gets re-dirtied from its original owner
4750 * Since it has bumped generation, no WRITTEN flag, it can be
4751 * reused without COWing. This tree block will not be traced
4752 * by btrfs_transaction::dirty_pages.
4754 * Now such dirty tree block will not be cleaned by any dirty
4755 * extent io tree. Thus we don't want to submit such wild eb
4756 * if the fs already has error.
4758 if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
4759 ret = flush_write_bio(&epd);
4762 end_write_bio(&epd, ret);
4765 btrfs_zoned_meta_io_unlock(fs_info);
4770 * Walk the list of dirty pages of the given address space and write all of them.
4772 * @mapping: address space structure to write
4773 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
4774 * @epd: holds context for the write, namely the bio
4776 * If a page is already under I/O, write_cache_pages() skips it, even
4777 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
4778 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
4779 * and msync() need to guarantee that all the data which was dirty at the time
4780 * the call was made get new I/O started against them. If wbc->sync_mode is
4781 * WB_SYNC_ALL then we were called for data integrity and we must wait for
4782 * existing IO to complete.
4784 static int extent_write_cache_pages(struct address_space *mapping,
4785 struct writeback_control *wbc,
4786 struct extent_page_data *epd)
4788 struct inode *inode = mapping->host;
4791 int nr_to_write_done = 0;
4792 struct pagevec pvec;
4795 pgoff_t end; /* Inclusive */
4797 int range_whole = 0;
4802 * We have to hold onto the inode so that ordered extents can do their
4803 * work when the IO finishes. The alternative to this is failing to add
4804 * an ordered extent if the igrab() fails there and that is a huge pain
4805 * to deal with, so instead just hold onto the inode throughout the
4806 * writepages operation. If it fails here we are freeing up the inode
4807 * anyway and we'd rather not waste our time writing out stuff that is
4808 * going to be truncated anyway.
4813 pagevec_init(&pvec);
4814 if (wbc->range_cyclic) {
4815 index = mapping->writeback_index; /* Start from prev offset */
4818 * Start from the beginning does not need to cycle over the
4819 * range, mark it as scanned.
4821 scanned = (index == 0);
4823 index = wbc->range_start >> PAGE_SHIFT;
4824 end = wbc->range_end >> PAGE_SHIFT;
4825 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
4831 * We do the tagged writepage as long as the snapshot flush bit is set
4832 * and we are the first one who do the filemap_flush() on this inode.
4834 * The nr_to_write == LONG_MAX is needed to make sure other flushers do
4835 * not race in and drop the bit.
4837 if (range_whole && wbc->nr_to_write == LONG_MAX &&
4838 test_and_clear_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
4839 &BTRFS_I(inode)->runtime_flags))
4840 wbc->tagged_writepages = 1;
4842 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
4843 tag = PAGECACHE_TAG_TOWRITE;
4845 tag = PAGECACHE_TAG_DIRTY;
4847 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
4848 tag_pages_for_writeback(mapping, index, end);
4850 while (!done && !nr_to_write_done && (index <= end) &&
4851 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping,
4852 &index, end, tag))) {
4855 for (i = 0; i < nr_pages; i++) {
4856 struct page *page = pvec.pages[i];
4858 done_index = page->index + 1;
4860 * At this point we hold neither the i_pages lock nor
4861 * the page lock: the page may be truncated or
4862 * invalidated (changing page->mapping to NULL),
4863 * or even swizzled back from swapper_space to
4864 * tmpfs file mapping
4866 if (!trylock_page(page)) {
4867 ret = flush_write_bio(epd);
4872 if (unlikely(page->mapping != mapping)) {
4877 if (wbc->sync_mode != WB_SYNC_NONE) {
4878 if (PageWriteback(page)) {
4879 ret = flush_write_bio(epd);
4882 wait_on_page_writeback(page);
4885 if (PageWriteback(page) ||
4886 !clear_page_dirty_for_io(page)) {
4891 ret = __extent_writepage(page, wbc, epd);
4898 * the filesystem may choose to bump up nr_to_write.
4899 * We have to make sure to honor the new nr_to_write
4902 nr_to_write_done = wbc->nr_to_write <= 0;
4904 pagevec_release(&pvec);
4907 if (!scanned && !done) {
4909 * We hit the last page and there is more work to be done: wrap
4910 * back to the start of the file
4916 * If we're looping we could run into a page that is locked by a
4917 * writer and that writer could be waiting on writeback for a
4918 * page in our current bio, and thus deadlock, so flush the
4921 ret = flush_write_bio(epd);
4926 if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole))
4927 mapping->writeback_index = done_index;
4929 btrfs_add_delayed_iput(inode);
4933 int extent_write_full_page(struct page *page, struct writeback_control *wbc)
4936 struct extent_page_data epd = {
4939 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4942 ret = __extent_writepage(page, wbc, &epd);
4945 end_write_bio(&epd, ret);
4949 ret = flush_write_bio(&epd);
4954 int extent_write_locked_range(struct inode *inode, u64 start, u64 end,
4958 struct address_space *mapping = inode->i_mapping;
4960 unsigned long nr_pages = (end - start + PAGE_SIZE) >>
4963 struct extent_page_data epd = {
4966 .sync_io = mode == WB_SYNC_ALL,
4968 struct writeback_control wbc_writepages = {
4970 .nr_to_write = nr_pages * 2,
4971 .range_start = start,
4972 .range_end = end + 1,
4973 /* We're called from an async helper function */
4974 .punt_to_cgroup = 1,
4975 .no_cgroup_owner = 1,
4978 wbc_attach_fdatawrite_inode(&wbc_writepages, inode);
4979 while (start <= end) {
4980 page = find_get_page(mapping, start >> PAGE_SHIFT);
4981 if (clear_page_dirty_for_io(page))
4982 ret = __extent_writepage(page, &wbc_writepages, &epd);
4984 btrfs_writepage_endio_finish_ordered(BTRFS_I(inode),
4985 page, start, start + PAGE_SIZE - 1, true);
4994 ret = flush_write_bio(&epd);
4996 end_write_bio(&epd, ret);
4998 wbc_detach_inode(&wbc_writepages);
5002 int extent_writepages(struct address_space *mapping,
5003 struct writeback_control *wbc)
5006 struct extent_page_data epd = {
5009 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
5012 ret = extent_write_cache_pages(mapping, wbc, &epd);
5015 end_write_bio(&epd, ret);
5018 ret = flush_write_bio(&epd);
5022 void extent_readahead(struct readahead_control *rac)
5024 struct btrfs_bio_ctrl bio_ctrl = { 0 };
5025 struct page *pagepool[16];
5026 struct extent_map *em_cached = NULL;
5027 u64 prev_em_start = (u64)-1;
5030 while ((nr = readahead_page_batch(rac, pagepool))) {
5031 u64 contig_start = readahead_pos(rac);
5032 u64 contig_end = contig_start + readahead_batch_length(rac) - 1;
5034 contiguous_readpages(pagepool, nr, contig_start, contig_end,
5035 &em_cached, &bio_ctrl, &prev_em_start);
5039 free_extent_map(em_cached);
5042 if (submit_one_bio(bio_ctrl.bio, 0, bio_ctrl.bio_flags))
5048 * basic invalidatepage code, this waits on any locked or writeback
5049 * ranges corresponding to the page, and then deletes any extent state
5050 * records from the tree
5052 int extent_invalidatepage(struct extent_io_tree *tree,
5053 struct page *page, unsigned long offset)
5055 struct extent_state *cached_state = NULL;
5056 u64 start = page_offset(page);
5057 u64 end = start + PAGE_SIZE - 1;
5058 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
5060 /* This function is only called for the btree inode */
5061 ASSERT(tree->owner == IO_TREE_BTREE_INODE_IO);
5063 start += ALIGN(offset, blocksize);
5067 lock_extent_bits(tree, start, end, &cached_state);
5068 wait_on_page_writeback(page);
5071 * Currently for btree io tree, only EXTENT_LOCKED is utilized,
5072 * so here we only need to unlock the extent range to free any
5073 * existing extent state.
5075 unlock_extent_cached(tree, start, end, &cached_state);
5080 * a helper for releasepage, this tests for areas of the page that
5081 * are locked or under IO and drops the related state bits if it is safe
5084 static int try_release_extent_state(struct extent_io_tree *tree,
5085 struct page *page, gfp_t mask)
5087 u64 start = page_offset(page);
5088 u64 end = start + PAGE_SIZE - 1;
5091 if (test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL)) {
5095 * At this point we can safely clear everything except the
5096 * locked bit, the nodatasum bit and the delalloc new bit.
5097 * The delalloc new bit will be cleared by ordered extent
5100 ret = __clear_extent_bit(tree, start, end,
5101 ~(EXTENT_LOCKED | EXTENT_NODATASUM | EXTENT_DELALLOC_NEW),
5102 0, 0, NULL, mask, NULL);
5104 /* if clear_extent_bit failed for enomem reasons,
5105 * we can't allow the release to continue.
5116 * a helper for releasepage. As long as there are no locked extents
5117 * in the range corresponding to the page, both state records and extent
5118 * map records are removed
5120 int try_release_extent_mapping(struct page *page, gfp_t mask)
5122 struct extent_map *em;
5123 u64 start = page_offset(page);
5124 u64 end = start + PAGE_SIZE - 1;
5125 struct btrfs_inode *btrfs_inode = BTRFS_I(page->mapping->host);
5126 struct extent_io_tree *tree = &btrfs_inode->io_tree;
5127 struct extent_map_tree *map = &btrfs_inode->extent_tree;
5129 if (gfpflags_allow_blocking(mask) &&
5130 page->mapping->host->i_size > SZ_16M) {
5132 while (start <= end) {
5133 struct btrfs_fs_info *fs_info;
5136 len = end - start + 1;
5137 write_lock(&map->lock);
5138 em = lookup_extent_mapping(map, start, len);
5140 write_unlock(&map->lock);
5143 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
5144 em->start != start) {
5145 write_unlock(&map->lock);
5146 free_extent_map(em);
5149 if (test_range_bit(tree, em->start,
5150 extent_map_end(em) - 1,
5151 EXTENT_LOCKED, 0, NULL))
5154 * If it's not in the list of modified extents, used
5155 * by a fast fsync, we can remove it. If it's being
5156 * logged we can safely remove it since fsync took an
5157 * extra reference on the em.
5159 if (list_empty(&em->list) ||
5160 test_bit(EXTENT_FLAG_LOGGING, &em->flags))
5163 * If it's in the list of modified extents, remove it
5164 * only if its generation is older then the current one,
5165 * in which case we don't need it for a fast fsync.
5166 * Otherwise don't remove it, we could be racing with an
5167 * ongoing fast fsync that could miss the new extent.
5169 fs_info = btrfs_inode->root->fs_info;
5170 spin_lock(&fs_info->trans_lock);
5171 cur_gen = fs_info->generation;
5172 spin_unlock(&fs_info->trans_lock);
5173 if (em->generation >= cur_gen)
5177 * We only remove extent maps that are not in the list of
5178 * modified extents or that are in the list but with a
5179 * generation lower then the current generation, so there
5180 * is no need to set the full fsync flag on the inode (it
5181 * hurts the fsync performance for workloads with a data
5182 * size that exceeds or is close to the system's memory).
5184 remove_extent_mapping(map, em);
5185 /* once for the rb tree */
5186 free_extent_map(em);
5188 start = extent_map_end(em);
5189 write_unlock(&map->lock);
5192 free_extent_map(em);
5194 cond_resched(); /* Allow large-extent preemption. */
5197 return try_release_extent_state(tree, page, mask);
5201 * helper function for fiemap, which doesn't want to see any holes.
5202 * This maps until we find something past 'last'
5204 static struct extent_map *get_extent_skip_holes(struct btrfs_inode *inode,
5205 u64 offset, u64 last)
5207 u64 sectorsize = btrfs_inode_sectorsize(inode);
5208 struct extent_map *em;
5215 len = last - offset;
5218 len = ALIGN(len, sectorsize);
5219 em = btrfs_get_extent_fiemap(inode, offset, len);
5220 if (IS_ERR_OR_NULL(em))
5223 /* if this isn't a hole return it */
5224 if (em->block_start != EXTENT_MAP_HOLE)
5227 /* this is a hole, advance to the next extent */
5228 offset = extent_map_end(em);
5229 free_extent_map(em);
5237 * To cache previous fiemap extent
5239 * Will be used for merging fiemap extent
5241 struct fiemap_cache {
5250 * Helper to submit fiemap extent.
5252 * Will try to merge current fiemap extent specified by @offset, @phys,
5253 * @len and @flags with cached one.
5254 * And only when we fails to merge, cached one will be submitted as
5257 * Return value is the same as fiemap_fill_next_extent().
5259 static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
5260 struct fiemap_cache *cache,
5261 u64 offset, u64 phys, u64 len, u32 flags)
5269 * Sanity check, extent_fiemap() should have ensured that new
5270 * fiemap extent won't overlap with cached one.
5273 * NOTE: Physical address can overlap, due to compression
5275 if (cache->offset + cache->len > offset) {
5281 * Only merges fiemap extents if
5282 * 1) Their logical addresses are continuous
5284 * 2) Their physical addresses are continuous
5285 * So truly compressed (physical size smaller than logical size)
5286 * extents won't get merged with each other
5288 * 3) Share same flags except FIEMAP_EXTENT_LAST
5289 * So regular extent won't get merged with prealloc extent
5291 if (cache->offset + cache->len == offset &&
5292 cache->phys + cache->len == phys &&
5293 (cache->flags & ~FIEMAP_EXTENT_LAST) ==
5294 (flags & ~FIEMAP_EXTENT_LAST)) {
5296 cache->flags |= flags;
5297 goto try_submit_last;
5300 /* Not mergeable, need to submit cached one */
5301 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
5302 cache->len, cache->flags);
5303 cache->cached = false;
5307 cache->cached = true;
5308 cache->offset = offset;
5311 cache->flags = flags;
5313 if (cache->flags & FIEMAP_EXTENT_LAST) {
5314 ret = fiemap_fill_next_extent(fieinfo, cache->offset,
5315 cache->phys, cache->len, cache->flags);
5316 cache->cached = false;
5322 * Emit last fiemap cache
5324 * The last fiemap cache may still be cached in the following case:
5326 * |<- Fiemap range ->|
5327 * |<------------ First extent ----------->|
5329 * In this case, the first extent range will be cached but not emitted.
5330 * So we must emit it before ending extent_fiemap().
5332 static int emit_last_fiemap_cache(struct fiemap_extent_info *fieinfo,
5333 struct fiemap_cache *cache)
5340 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
5341 cache->len, cache->flags);
5342 cache->cached = false;
5348 int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,
5353 u64 max = start + len;
5357 u64 last_for_get_extent = 0;
5359 u64 isize = i_size_read(&inode->vfs_inode);
5360 struct btrfs_key found_key;
5361 struct extent_map *em = NULL;
5362 struct extent_state *cached_state = NULL;
5363 struct btrfs_path *path;
5364 struct btrfs_root *root = inode->root;
5365 struct fiemap_cache cache = { 0 };
5366 struct ulist *roots;
5367 struct ulist *tmp_ulist;
5376 path = btrfs_alloc_path();
5380 roots = ulist_alloc(GFP_KERNEL);
5381 tmp_ulist = ulist_alloc(GFP_KERNEL);
5382 if (!roots || !tmp_ulist) {
5384 goto out_free_ulist;
5388 * We can't initialize that to 'start' as this could miss extents due
5389 * to extent item merging
5392 start = round_down(start, btrfs_inode_sectorsize(inode));
5393 len = round_up(max, btrfs_inode_sectorsize(inode)) - start;
5396 * lookup the last file extent. We're not using i_size here
5397 * because there might be preallocation past i_size
5399 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode), -1,
5402 goto out_free_ulist;
5410 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
5411 found_type = found_key.type;
5413 /* No extents, but there might be delalloc bits */
5414 if (found_key.objectid != btrfs_ino(inode) ||
5415 found_type != BTRFS_EXTENT_DATA_KEY) {
5416 /* have to trust i_size as the end */
5418 last_for_get_extent = isize;
5421 * remember the start of the last extent. There are a
5422 * bunch of different factors that go into the length of the
5423 * extent, so its much less complex to remember where it started
5425 last = found_key.offset;
5426 last_for_get_extent = last + 1;
5428 btrfs_release_path(path);
5431 * we might have some extents allocated but more delalloc past those
5432 * extents. so, we trust isize unless the start of the last extent is
5437 last_for_get_extent = isize;
5440 lock_extent_bits(&inode->io_tree, start, start + len - 1,
5443 em = get_extent_skip_holes(inode, start, last_for_get_extent);
5452 u64 offset_in_extent = 0;
5454 /* break if the extent we found is outside the range */
5455 if (em->start >= max || extent_map_end(em) < off)
5459 * get_extent may return an extent that starts before our
5460 * requested range. We have to make sure the ranges
5461 * we return to fiemap always move forward and don't
5462 * overlap, so adjust the offsets here
5464 em_start = max(em->start, off);
5467 * record the offset from the start of the extent
5468 * for adjusting the disk offset below. Only do this if the
5469 * extent isn't compressed since our in ram offset may be past
5470 * what we have actually allocated on disk.
5472 if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
5473 offset_in_extent = em_start - em->start;
5474 em_end = extent_map_end(em);
5475 em_len = em_end - em_start;
5477 if (em->block_start < EXTENT_MAP_LAST_BYTE)
5478 disko = em->block_start + offset_in_extent;
5483 * bump off for our next call to get_extent
5485 off = extent_map_end(em);
5489 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
5491 flags |= FIEMAP_EXTENT_LAST;
5492 } else if (em->block_start == EXTENT_MAP_INLINE) {
5493 flags |= (FIEMAP_EXTENT_DATA_INLINE |
5494 FIEMAP_EXTENT_NOT_ALIGNED);
5495 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
5496 flags |= (FIEMAP_EXTENT_DELALLOC |
5497 FIEMAP_EXTENT_UNKNOWN);
5498 } else if (fieinfo->fi_extents_max) {
5499 u64 bytenr = em->block_start -
5500 (em->start - em->orig_start);
5503 * As btrfs supports shared space, this information
5504 * can be exported to userspace tools via
5505 * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0
5506 * then we're just getting a count and we can skip the
5509 ret = btrfs_check_shared(root, btrfs_ino(inode),
5510 bytenr, roots, tmp_ulist);
5514 flags |= FIEMAP_EXTENT_SHARED;
5517 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
5518 flags |= FIEMAP_EXTENT_ENCODED;
5519 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5520 flags |= FIEMAP_EXTENT_UNWRITTEN;
5522 free_extent_map(em);
5524 if ((em_start >= last) || em_len == (u64)-1 ||
5525 (last == (u64)-1 && isize <= em_end)) {
5526 flags |= FIEMAP_EXTENT_LAST;
5530 /* now scan forward to see if this is really the last extent. */
5531 em = get_extent_skip_holes(inode, off, last_for_get_extent);
5537 flags |= FIEMAP_EXTENT_LAST;
5540 ret = emit_fiemap_extent(fieinfo, &cache, em_start, disko,
5550 ret = emit_last_fiemap_cache(fieinfo, &cache);
5551 free_extent_map(em);
5553 unlock_extent_cached(&inode->io_tree, start, start + len - 1,
5557 btrfs_free_path(path);
5559 ulist_free(tmp_ulist);
5563 static void __free_extent_buffer(struct extent_buffer *eb)
5565 kmem_cache_free(extent_buffer_cache, eb);
5568 int extent_buffer_under_io(const struct extent_buffer *eb)
5570 return (atomic_read(&eb->io_pages) ||
5571 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
5572 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
5575 static bool page_range_has_eb(struct btrfs_fs_info *fs_info, struct page *page)
5577 struct btrfs_subpage *subpage;
5579 lockdep_assert_held(&page->mapping->private_lock);
5581 if (PagePrivate(page)) {
5582 subpage = (struct btrfs_subpage *)page->private;
5583 if (atomic_read(&subpage->eb_refs))
5586 * Even there is no eb refs here, we may still have
5587 * end_page_read() call relying on page::private.
5589 if (atomic_read(&subpage->readers))
5595 static void detach_extent_buffer_page(struct extent_buffer *eb, struct page *page)
5597 struct btrfs_fs_info *fs_info = eb->fs_info;
5598 const bool mapped = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
5601 * For mapped eb, we're going to change the page private, which should
5602 * be done under the private_lock.
5605 spin_lock(&page->mapping->private_lock);
5607 if (!PagePrivate(page)) {
5609 spin_unlock(&page->mapping->private_lock);
5613 if (fs_info->sectorsize == PAGE_SIZE) {
5615 * We do this since we'll remove the pages after we've
5616 * removed the eb from the radix tree, so we could race
5617 * and have this page now attached to the new eb. So
5618 * only clear page_private if it's still connected to
5621 if (PagePrivate(page) &&
5622 page->private == (unsigned long)eb) {
5623 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
5624 BUG_ON(PageDirty(page));
5625 BUG_ON(PageWriteback(page));
5627 * We need to make sure we haven't be attached
5630 detach_page_private(page);
5633 spin_unlock(&page->mapping->private_lock);
5638 * For subpage, we can have dummy eb with page private. In this case,
5639 * we can directly detach the private as such page is only attached to
5640 * one dummy eb, no sharing.
5643 btrfs_detach_subpage(fs_info, page);
5647 btrfs_page_dec_eb_refs(fs_info, page);
5650 * We can only detach the page private if there are no other ebs in the
5651 * page range and no unfinished IO.
5653 if (!page_range_has_eb(fs_info, page))
5654 btrfs_detach_subpage(fs_info, page);
5656 spin_unlock(&page->mapping->private_lock);
5659 /* Release all pages attached to the extent buffer */
5660 static void btrfs_release_extent_buffer_pages(struct extent_buffer *eb)
5665 ASSERT(!extent_buffer_under_io(eb));
5667 num_pages = num_extent_pages(eb);
5668 for (i = 0; i < num_pages; i++) {
5669 struct page *page = eb->pages[i];
5674 detach_extent_buffer_page(eb, page);
5676 /* One for when we allocated the page */
5682 * Helper for releasing the extent buffer.
5684 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
5686 btrfs_release_extent_buffer_pages(eb);
5687 btrfs_leak_debug_del(&eb->fs_info->eb_leak_lock, &eb->leak_list);
5688 __free_extent_buffer(eb);
5691 static struct extent_buffer *
5692 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
5695 struct extent_buffer *eb = NULL;
5697 eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL);
5700 eb->fs_info = fs_info;
5702 init_rwsem(&eb->lock);
5704 btrfs_leak_debug_add(&fs_info->eb_leak_lock, &eb->leak_list,
5705 &fs_info->allocated_ebs);
5706 INIT_LIST_HEAD(&eb->release_list);
5708 spin_lock_init(&eb->refs_lock);
5709 atomic_set(&eb->refs, 1);
5710 atomic_set(&eb->io_pages, 0);
5712 ASSERT(len <= BTRFS_MAX_METADATA_BLOCKSIZE);
5717 struct extent_buffer *btrfs_clone_extent_buffer(const struct extent_buffer *src)
5721 struct extent_buffer *new;
5722 int num_pages = num_extent_pages(src);
5724 new = __alloc_extent_buffer(src->fs_info, src->start, src->len);
5729 * Set UNMAPPED before calling btrfs_release_extent_buffer(), as
5730 * btrfs_release_extent_buffer() have different behavior for
5731 * UNMAPPED subpage extent buffer.
5733 set_bit(EXTENT_BUFFER_UNMAPPED, &new->bflags);
5735 for (i = 0; i < num_pages; i++) {
5738 p = alloc_page(GFP_NOFS);
5740 btrfs_release_extent_buffer(new);
5743 ret = attach_extent_buffer_page(new, p, NULL);
5746 btrfs_release_extent_buffer(new);
5749 WARN_ON(PageDirty(p));
5751 copy_page(page_address(p), page_address(src->pages[i]));
5753 set_extent_buffer_uptodate(new);
5758 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
5759 u64 start, unsigned long len)
5761 struct extent_buffer *eb;
5765 eb = __alloc_extent_buffer(fs_info, start, len);
5769 num_pages = num_extent_pages(eb);
5770 for (i = 0; i < num_pages; i++) {
5773 eb->pages[i] = alloc_page(GFP_NOFS);
5776 ret = attach_extent_buffer_page(eb, eb->pages[i], NULL);
5780 set_extent_buffer_uptodate(eb);
5781 btrfs_set_header_nritems(eb, 0);
5782 set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
5786 for (; i > 0; i--) {
5787 detach_extent_buffer_page(eb, eb->pages[i - 1]);
5788 __free_page(eb->pages[i - 1]);
5790 __free_extent_buffer(eb);
5794 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
5797 return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize);
5800 static void check_buffer_tree_ref(struct extent_buffer *eb)
5804 * The TREE_REF bit is first set when the extent_buffer is added
5805 * to the radix tree. It is also reset, if unset, when a new reference
5806 * is created by find_extent_buffer.
5808 * It is only cleared in two cases: freeing the last non-tree
5809 * reference to the extent_buffer when its STALE bit is set or
5810 * calling releasepage when the tree reference is the only reference.
5812 * In both cases, care is taken to ensure that the extent_buffer's
5813 * pages are not under io. However, releasepage can be concurrently
5814 * called with creating new references, which is prone to race
5815 * conditions between the calls to check_buffer_tree_ref in those
5816 * codepaths and clearing TREE_REF in try_release_extent_buffer.
5818 * The actual lifetime of the extent_buffer in the radix tree is
5819 * adequately protected by the refcount, but the TREE_REF bit and
5820 * its corresponding reference are not. To protect against this
5821 * class of races, we call check_buffer_tree_ref from the codepaths
5822 * which trigger io after they set eb->io_pages. Note that once io is
5823 * initiated, TREE_REF can no longer be cleared, so that is the
5824 * moment at which any such race is best fixed.
5826 refs = atomic_read(&eb->refs);
5827 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5830 spin_lock(&eb->refs_lock);
5831 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5832 atomic_inc(&eb->refs);
5833 spin_unlock(&eb->refs_lock);
5836 static void mark_extent_buffer_accessed(struct extent_buffer *eb,
5837 struct page *accessed)
5841 check_buffer_tree_ref(eb);
5843 num_pages = num_extent_pages(eb);
5844 for (i = 0; i < num_pages; i++) {
5845 struct page *p = eb->pages[i];
5848 mark_page_accessed(p);
5852 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
5855 struct extent_buffer *eb;
5857 eb = find_extent_buffer_nolock(fs_info, start);
5861 * Lock our eb's refs_lock to avoid races with free_extent_buffer().
5862 * When we get our eb it might be flagged with EXTENT_BUFFER_STALE and
5863 * another task running free_extent_buffer() might have seen that flag
5864 * set, eb->refs == 2, that the buffer isn't under IO (dirty and
5865 * writeback flags not set) and it's still in the tree (flag
5866 * EXTENT_BUFFER_TREE_REF set), therefore being in the process of
5867 * decrementing the extent buffer's reference count twice. So here we
5868 * could race and increment the eb's reference count, clear its stale
5869 * flag, mark it as dirty and drop our reference before the other task
5870 * finishes executing free_extent_buffer, which would later result in
5871 * an attempt to free an extent buffer that is dirty.
5873 if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) {
5874 spin_lock(&eb->refs_lock);
5875 spin_unlock(&eb->refs_lock);
5877 mark_extent_buffer_accessed(eb, NULL);
5881 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5882 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
5885 struct extent_buffer *eb, *exists = NULL;
5888 eb = find_extent_buffer(fs_info, start);
5891 eb = alloc_dummy_extent_buffer(fs_info, start);
5893 return ERR_PTR(-ENOMEM);
5894 eb->fs_info = fs_info;
5896 ret = radix_tree_preload(GFP_NOFS);
5898 exists = ERR_PTR(ret);
5901 spin_lock(&fs_info->buffer_lock);
5902 ret = radix_tree_insert(&fs_info->buffer_radix,
5903 start >> fs_info->sectorsize_bits, eb);
5904 spin_unlock(&fs_info->buffer_lock);
5905 radix_tree_preload_end();
5906 if (ret == -EEXIST) {
5907 exists = find_extent_buffer(fs_info, start);
5913 check_buffer_tree_ref(eb);
5914 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
5918 btrfs_release_extent_buffer(eb);
5923 static struct extent_buffer *grab_extent_buffer(
5924 struct btrfs_fs_info *fs_info, struct page *page)
5926 struct extent_buffer *exists;
5929 * For subpage case, we completely rely on radix tree to ensure we
5930 * don't try to insert two ebs for the same bytenr. So here we always
5931 * return NULL and just continue.
5933 if (fs_info->sectorsize < PAGE_SIZE)
5936 /* Page not yet attached to an extent buffer */
5937 if (!PagePrivate(page))
5941 * We could have already allocated an eb for this page and attached one
5942 * so lets see if we can get a ref on the existing eb, and if we can we
5943 * know it's good and we can just return that one, else we know we can
5944 * just overwrite page->private.
5946 exists = (struct extent_buffer *)page->private;
5947 if (atomic_inc_not_zero(&exists->refs))
5950 WARN_ON(PageDirty(page));
5951 detach_page_private(page);
5955 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
5956 u64 start, u64 owner_root, int level)
5958 unsigned long len = fs_info->nodesize;
5961 unsigned long index = start >> PAGE_SHIFT;
5962 struct extent_buffer *eb;
5963 struct extent_buffer *exists = NULL;
5965 struct address_space *mapping = fs_info->btree_inode->i_mapping;
5969 if (!IS_ALIGNED(start, fs_info->sectorsize)) {
5970 btrfs_err(fs_info, "bad tree block start %llu", start);
5971 return ERR_PTR(-EINVAL);
5974 #if BITS_PER_LONG == 32
5975 if (start >= MAX_LFS_FILESIZE) {
5976 btrfs_err_rl(fs_info,
5977 "extent buffer %llu is beyond 32bit page cache limit", start);
5978 btrfs_err_32bit_limit(fs_info);
5979 return ERR_PTR(-EOVERFLOW);
5981 if (start >= BTRFS_32BIT_EARLY_WARN_THRESHOLD)
5982 btrfs_warn_32bit_limit(fs_info);
5985 if (fs_info->sectorsize < PAGE_SIZE &&
5986 offset_in_page(start) + len > PAGE_SIZE) {
5988 "tree block crosses page boundary, start %llu nodesize %lu",
5990 return ERR_PTR(-EINVAL);
5993 eb = find_extent_buffer(fs_info, start);
5997 eb = __alloc_extent_buffer(fs_info, start, len);
5999 return ERR_PTR(-ENOMEM);
6000 btrfs_set_buffer_lockdep_class(owner_root, eb, level);
6002 num_pages = num_extent_pages(eb);
6003 for (i = 0; i < num_pages; i++, index++) {
6004 struct btrfs_subpage *prealloc = NULL;
6006 p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL);
6008 exists = ERR_PTR(-ENOMEM);
6013 * Preallocate page->private for subpage case, so that we won't
6014 * allocate memory with private_lock hold. The memory will be
6015 * freed by attach_extent_buffer_page() or freed manually if
6018 * Although we have ensured one subpage eb can only have one
6019 * page, but it may change in the future for 16K page size
6020 * support, so we still preallocate the memory in the loop.
6022 ret = btrfs_alloc_subpage(fs_info, &prealloc,
6023 BTRFS_SUBPAGE_METADATA);
6027 exists = ERR_PTR(ret);
6031 spin_lock(&mapping->private_lock);
6032 exists = grab_extent_buffer(fs_info, p);
6034 spin_unlock(&mapping->private_lock);
6037 mark_extent_buffer_accessed(exists, p);
6038 btrfs_free_subpage(prealloc);
6041 /* Should not fail, as we have preallocated the memory */
6042 ret = attach_extent_buffer_page(eb, p, prealloc);
6045 * To inform we have extra eb under allocation, so that
6046 * detach_extent_buffer_page() won't release the page private
6047 * when the eb hasn't yet been inserted into radix tree.
6049 * The ref will be decreased when the eb released the page, in
6050 * detach_extent_buffer_page().
6051 * Thus needs no special handling in error path.
6053 btrfs_page_inc_eb_refs(fs_info, p);
6054 spin_unlock(&mapping->private_lock);
6056 WARN_ON(btrfs_page_test_dirty(fs_info, p, eb->start, eb->len));
6058 if (!PageUptodate(p))
6062 * We can't unlock the pages just yet since the extent buffer
6063 * hasn't been properly inserted in the radix tree, this
6064 * opens a race with btree_releasepage which can free a page
6065 * while we are still filling in all pages for the buffer and
6070 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
6072 ret = radix_tree_preload(GFP_NOFS);
6074 exists = ERR_PTR(ret);
6078 spin_lock(&fs_info->buffer_lock);
6079 ret = radix_tree_insert(&fs_info->buffer_radix,
6080 start >> fs_info->sectorsize_bits, eb);
6081 spin_unlock(&fs_info->buffer_lock);
6082 radix_tree_preload_end();
6083 if (ret == -EEXIST) {
6084 exists = find_extent_buffer(fs_info, start);
6090 /* add one reference for the tree */
6091 check_buffer_tree_ref(eb);
6092 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
6095 * Now it's safe to unlock the pages because any calls to
6096 * btree_releasepage will correctly detect that a page belongs to a
6097 * live buffer and won't free them prematurely.
6099 for (i = 0; i < num_pages; i++)
6100 unlock_page(eb->pages[i]);
6104 WARN_ON(!atomic_dec_and_test(&eb->refs));
6105 for (i = 0; i < num_pages; i++) {
6107 unlock_page(eb->pages[i]);
6110 btrfs_release_extent_buffer(eb);
6114 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
6116 struct extent_buffer *eb =
6117 container_of(head, struct extent_buffer, rcu_head);
6119 __free_extent_buffer(eb);
6122 static int release_extent_buffer(struct extent_buffer *eb)
6123 __releases(&eb->refs_lock)
6125 lockdep_assert_held(&eb->refs_lock);
6127 WARN_ON(atomic_read(&eb->refs) == 0);
6128 if (atomic_dec_and_test(&eb->refs)) {
6129 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
6130 struct btrfs_fs_info *fs_info = eb->fs_info;
6132 spin_unlock(&eb->refs_lock);
6134 spin_lock(&fs_info->buffer_lock);
6135 radix_tree_delete(&fs_info->buffer_radix,
6136 eb->start >> fs_info->sectorsize_bits);
6137 spin_unlock(&fs_info->buffer_lock);
6139 spin_unlock(&eb->refs_lock);
6142 btrfs_leak_debug_del(&eb->fs_info->eb_leak_lock, &eb->leak_list);
6143 /* Should be safe to release our pages at this point */
6144 btrfs_release_extent_buffer_pages(eb);
6145 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
6146 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags))) {
6147 __free_extent_buffer(eb);
6151 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
6154 spin_unlock(&eb->refs_lock);
6159 void free_extent_buffer(struct extent_buffer *eb)
6167 refs = atomic_read(&eb->refs);
6168 if ((!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && refs <= 3)
6169 || (test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) &&
6172 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
6177 spin_lock(&eb->refs_lock);
6178 if (atomic_read(&eb->refs) == 2 &&
6179 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
6180 !extent_buffer_under_io(eb) &&
6181 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
6182 atomic_dec(&eb->refs);
6185 * I know this is terrible, but it's temporary until we stop tracking
6186 * the uptodate bits and such for the extent buffers.
6188 release_extent_buffer(eb);
6191 void free_extent_buffer_stale(struct extent_buffer *eb)
6196 spin_lock(&eb->refs_lock);
6197 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
6199 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
6200 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
6201 atomic_dec(&eb->refs);
6202 release_extent_buffer(eb);
6205 static void btree_clear_page_dirty(struct page *page)
6207 ASSERT(PageDirty(page));
6208 ASSERT(PageLocked(page));
6209 clear_page_dirty_for_io(page);
6210 xa_lock_irq(&page->mapping->i_pages);
6211 if (!PageDirty(page))
6212 __xa_clear_mark(&page->mapping->i_pages,
6213 page_index(page), PAGECACHE_TAG_DIRTY);
6214 xa_unlock_irq(&page->mapping->i_pages);
6217 static void clear_subpage_extent_buffer_dirty(const struct extent_buffer *eb)
6219 struct btrfs_fs_info *fs_info = eb->fs_info;
6220 struct page *page = eb->pages[0];
6223 /* btree_clear_page_dirty() needs page locked */
6225 last = btrfs_subpage_clear_and_test_dirty(fs_info, page, eb->start,
6228 btree_clear_page_dirty(page);
6230 WARN_ON(atomic_read(&eb->refs) == 0);
6233 void clear_extent_buffer_dirty(const struct extent_buffer *eb)
6239 if (eb->fs_info->sectorsize < PAGE_SIZE)
6240 return clear_subpage_extent_buffer_dirty(eb);
6242 num_pages = num_extent_pages(eb);
6244 for (i = 0; i < num_pages; i++) {
6245 page = eb->pages[i];
6246 if (!PageDirty(page))
6249 btree_clear_page_dirty(page);
6250 ClearPageError(page);
6253 WARN_ON(atomic_read(&eb->refs) == 0);
6256 bool set_extent_buffer_dirty(struct extent_buffer *eb)
6262 check_buffer_tree_ref(eb);
6264 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
6266 num_pages = num_extent_pages(eb);
6267 WARN_ON(atomic_read(&eb->refs) == 0);
6268 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
6271 bool subpage = eb->fs_info->sectorsize < PAGE_SIZE;
6274 * For subpage case, we can have other extent buffers in the
6275 * same page, and in clear_subpage_extent_buffer_dirty() we
6276 * have to clear page dirty without subpage lock held.
6277 * This can cause race where our page gets dirty cleared after
6280 * Thankfully, clear_subpage_extent_buffer_dirty() has locked
6281 * its page for other reasons, we can use page lock to prevent
6285 lock_page(eb->pages[0]);
6286 for (i = 0; i < num_pages; i++)
6287 btrfs_page_set_dirty(eb->fs_info, eb->pages[i],
6288 eb->start, eb->len);
6290 unlock_page(eb->pages[0]);
6292 #ifdef CONFIG_BTRFS_DEBUG
6293 for (i = 0; i < num_pages; i++)
6294 ASSERT(PageDirty(eb->pages[i]));
6300 void clear_extent_buffer_uptodate(struct extent_buffer *eb)
6302 struct btrfs_fs_info *fs_info = eb->fs_info;
6307 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
6308 num_pages = num_extent_pages(eb);
6309 for (i = 0; i < num_pages; i++) {
6310 page = eb->pages[i];
6312 btrfs_page_clear_uptodate(fs_info, page,
6313 eb->start, eb->len);
6317 void set_extent_buffer_uptodate(struct extent_buffer *eb)
6319 struct btrfs_fs_info *fs_info = eb->fs_info;
6324 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
6325 num_pages = num_extent_pages(eb);
6326 for (i = 0; i < num_pages; i++) {
6327 page = eb->pages[i];
6328 btrfs_page_set_uptodate(fs_info, page, eb->start, eb->len);
6332 static int read_extent_buffer_subpage(struct extent_buffer *eb, int wait,
6335 struct btrfs_fs_info *fs_info = eb->fs_info;
6336 struct extent_io_tree *io_tree;
6337 struct page *page = eb->pages[0];
6338 struct btrfs_bio_ctrl bio_ctrl = { 0 };
6341 ASSERT(!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags));
6342 ASSERT(PagePrivate(page));
6343 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
6345 if (wait == WAIT_NONE) {
6346 if (!try_lock_extent(io_tree, eb->start, eb->start + eb->len - 1))
6349 ret = lock_extent(io_tree, eb->start, eb->start + eb->len - 1);
6355 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags) ||
6356 PageUptodate(page) ||
6357 btrfs_subpage_test_uptodate(fs_info, page, eb->start, eb->len)) {
6358 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
6359 unlock_extent(io_tree, eb->start, eb->start + eb->len - 1);
6363 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
6364 eb->read_mirror = 0;
6365 atomic_set(&eb->io_pages, 1);
6366 check_buffer_tree_ref(eb);
6367 btrfs_subpage_clear_error(fs_info, page, eb->start, eb->len);
6369 btrfs_subpage_start_reader(fs_info, page, eb->start, eb->len);
6370 ret = submit_extent_page(REQ_OP_READ | REQ_META, NULL, &bio_ctrl,
6371 page, eb->start, eb->len,
6372 eb->start - page_offset(page),
6373 end_bio_extent_readpage, mirror_num, 0,
6377 * In the endio function, if we hit something wrong we will
6378 * increase the io_pages, so here we need to decrease it for
6381 atomic_dec(&eb->io_pages);
6386 tmp = submit_one_bio(bio_ctrl.bio, mirror_num, 0);
6387 bio_ctrl.bio = NULL;
6391 if (ret || wait != WAIT_COMPLETE)
6394 wait_extent_bit(io_tree, eb->start, eb->start + eb->len - 1, EXTENT_LOCKED);
6395 if (!test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
6400 int read_extent_buffer_pages(struct extent_buffer *eb, int wait, int mirror_num)
6406 int locked_pages = 0;
6407 int all_uptodate = 1;
6409 unsigned long num_reads = 0;
6410 struct btrfs_bio_ctrl bio_ctrl = { 0 };
6412 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
6415 if (eb->fs_info->sectorsize < PAGE_SIZE)
6416 return read_extent_buffer_subpage(eb, wait, mirror_num);
6418 num_pages = num_extent_pages(eb);
6419 for (i = 0; i < num_pages; i++) {
6420 page = eb->pages[i];
6421 if (wait == WAIT_NONE) {
6423 * WAIT_NONE is only utilized by readahead. If we can't
6424 * acquire the lock atomically it means either the eb
6425 * is being read out or under modification.
6426 * Either way the eb will be or has been cached,
6427 * readahead can exit safely.
6429 if (!trylock_page(page))
6437 * We need to firstly lock all pages to make sure that
6438 * the uptodate bit of our pages won't be affected by
6439 * clear_extent_buffer_uptodate().
6441 for (i = 0; i < num_pages; i++) {
6442 page = eb->pages[i];
6443 if (!PageUptodate(page)) {
6450 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
6454 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
6455 eb->read_mirror = 0;
6456 atomic_set(&eb->io_pages, num_reads);
6458 * It is possible for releasepage to clear the TREE_REF bit before we
6459 * set io_pages. See check_buffer_tree_ref for a more detailed comment.
6461 check_buffer_tree_ref(eb);
6462 for (i = 0; i < num_pages; i++) {
6463 page = eb->pages[i];
6465 if (!PageUptodate(page)) {
6467 atomic_dec(&eb->io_pages);
6472 ClearPageError(page);
6473 err = submit_extent_page(REQ_OP_READ | REQ_META, NULL,
6474 &bio_ctrl, page, page_offset(page),
6475 PAGE_SIZE, 0, end_bio_extent_readpage,
6476 mirror_num, 0, false);
6479 * We failed to submit the bio so it's the
6480 * caller's responsibility to perform cleanup
6481 * i.e unlock page/set error bit.
6486 atomic_dec(&eb->io_pages);
6494 err = submit_one_bio(bio_ctrl.bio, mirror_num, bio_ctrl.bio_flags);
6495 bio_ctrl.bio = NULL;
6500 if (ret || wait != WAIT_COMPLETE)
6503 for (i = 0; i < num_pages; i++) {
6504 page = eb->pages[i];
6505 wait_on_page_locked(page);
6506 if (!PageUptodate(page))
6513 while (locked_pages > 0) {
6515 page = eb->pages[locked_pages];
6521 static bool report_eb_range(const struct extent_buffer *eb, unsigned long start,
6524 btrfs_warn(eb->fs_info,
6525 "access to eb bytenr %llu len %lu out of range start %lu len %lu",
6526 eb->start, eb->len, start, len);
6527 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
6533 * Check if the [start, start + len) range is valid before reading/writing
6535 * NOTE: @start and @len are offset inside the eb, not logical address.
6537 * Caller should not touch the dst/src memory if this function returns error.
6539 static inline int check_eb_range(const struct extent_buffer *eb,
6540 unsigned long start, unsigned long len)
6542 unsigned long offset;
6544 /* start, start + len should not go beyond eb->len nor overflow */
6545 if (unlikely(check_add_overflow(start, len, &offset) || offset > eb->len))
6546 return report_eb_range(eb, start, len);
6551 void read_extent_buffer(const struct extent_buffer *eb, void *dstv,
6552 unsigned long start, unsigned long len)
6558 char *dst = (char *)dstv;
6559 unsigned long i = get_eb_page_index(start);
6561 if (check_eb_range(eb, start, len))
6564 offset = get_eb_offset_in_page(eb, start);
6567 page = eb->pages[i];
6569 cur = min(len, (PAGE_SIZE - offset));
6570 kaddr = page_address(page);
6571 memcpy(dst, kaddr + offset, cur);
6580 int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb,
6582 unsigned long start, unsigned long len)
6588 char __user *dst = (char __user *)dstv;
6589 unsigned long i = get_eb_page_index(start);
6592 WARN_ON(start > eb->len);
6593 WARN_ON(start + len > eb->start + eb->len);
6595 offset = get_eb_offset_in_page(eb, start);
6598 page = eb->pages[i];
6600 cur = min(len, (PAGE_SIZE - offset));
6601 kaddr = page_address(page);
6602 if (copy_to_user_nofault(dst, kaddr + offset, cur)) {
6616 int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,
6617 unsigned long start, unsigned long len)
6623 char *ptr = (char *)ptrv;
6624 unsigned long i = get_eb_page_index(start);
6627 if (check_eb_range(eb, start, len))
6630 offset = get_eb_offset_in_page(eb, start);
6633 page = eb->pages[i];
6635 cur = min(len, (PAGE_SIZE - offset));
6637 kaddr = page_address(page);
6638 ret = memcmp(ptr, kaddr + offset, cur);
6651 * Check that the extent buffer is uptodate.
6653 * For regular sector size == PAGE_SIZE case, check if @page is uptodate.
6654 * For subpage case, check if the range covered by the eb has EXTENT_UPTODATE.
6656 static void assert_eb_page_uptodate(const struct extent_buffer *eb,
6659 struct btrfs_fs_info *fs_info = eb->fs_info;
6661 if (fs_info->sectorsize < PAGE_SIZE) {
6664 uptodate = btrfs_subpage_test_uptodate(fs_info, page,
6665 eb->start, eb->len);
6668 WARN_ON(!PageUptodate(page));
6672 void write_extent_buffer_chunk_tree_uuid(const struct extent_buffer *eb,
6677 assert_eb_page_uptodate(eb, eb->pages[0]);
6678 kaddr = page_address(eb->pages[0]) +
6679 get_eb_offset_in_page(eb, offsetof(struct btrfs_header,
6681 memcpy(kaddr, srcv, BTRFS_FSID_SIZE);
6684 void write_extent_buffer_fsid(const struct extent_buffer *eb, const void *srcv)
6688 assert_eb_page_uptodate(eb, eb->pages[0]);
6689 kaddr = page_address(eb->pages[0]) +
6690 get_eb_offset_in_page(eb, offsetof(struct btrfs_header, fsid));
6691 memcpy(kaddr, srcv, BTRFS_FSID_SIZE);
6694 void write_extent_buffer(const struct extent_buffer *eb, const void *srcv,
6695 unsigned long start, unsigned long len)
6701 char *src = (char *)srcv;
6702 unsigned long i = get_eb_page_index(start);
6704 WARN_ON(test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags));
6706 if (check_eb_range(eb, start, len))
6709 offset = get_eb_offset_in_page(eb, start);
6712 page = eb->pages[i];
6713 assert_eb_page_uptodate(eb, page);
6715 cur = min(len, PAGE_SIZE - offset);
6716 kaddr = page_address(page);
6717 memcpy(kaddr + offset, src, cur);
6726 void memzero_extent_buffer(const struct extent_buffer *eb, unsigned long start,
6733 unsigned long i = get_eb_page_index(start);
6735 if (check_eb_range(eb, start, len))
6738 offset = get_eb_offset_in_page(eb, start);
6741 page = eb->pages[i];
6742 assert_eb_page_uptodate(eb, page);
6744 cur = min(len, PAGE_SIZE - offset);
6745 kaddr = page_address(page);
6746 memset(kaddr + offset, 0, cur);
6754 void copy_extent_buffer_full(const struct extent_buffer *dst,
6755 const struct extent_buffer *src)
6760 ASSERT(dst->len == src->len);
6762 if (dst->fs_info->sectorsize == PAGE_SIZE) {
6763 num_pages = num_extent_pages(dst);
6764 for (i = 0; i < num_pages; i++)
6765 copy_page(page_address(dst->pages[i]),
6766 page_address(src->pages[i]));
6768 size_t src_offset = get_eb_offset_in_page(src, 0);
6769 size_t dst_offset = get_eb_offset_in_page(dst, 0);
6771 ASSERT(src->fs_info->sectorsize < PAGE_SIZE);
6772 memcpy(page_address(dst->pages[0]) + dst_offset,
6773 page_address(src->pages[0]) + src_offset,
6778 void copy_extent_buffer(const struct extent_buffer *dst,
6779 const struct extent_buffer *src,
6780 unsigned long dst_offset, unsigned long src_offset,
6783 u64 dst_len = dst->len;
6788 unsigned long i = get_eb_page_index(dst_offset);
6790 if (check_eb_range(dst, dst_offset, len) ||
6791 check_eb_range(src, src_offset, len))
6794 WARN_ON(src->len != dst_len);
6796 offset = get_eb_offset_in_page(dst, dst_offset);
6799 page = dst->pages[i];
6800 assert_eb_page_uptodate(dst, page);
6802 cur = min(len, (unsigned long)(PAGE_SIZE - offset));
6804 kaddr = page_address(page);
6805 read_extent_buffer(src, kaddr + offset, src_offset, cur);
6815 * eb_bitmap_offset() - calculate the page and offset of the byte containing the
6817 * @eb: the extent buffer
6818 * @start: offset of the bitmap item in the extent buffer
6820 * @page_index: return index of the page in the extent buffer that contains the
6822 * @page_offset: return offset into the page given by page_index
6824 * This helper hides the ugliness of finding the byte in an extent buffer which
6825 * contains a given bit.
6827 static inline void eb_bitmap_offset(const struct extent_buffer *eb,
6828 unsigned long start, unsigned long nr,
6829 unsigned long *page_index,
6830 size_t *page_offset)
6832 size_t byte_offset = BIT_BYTE(nr);
6836 * The byte we want is the offset of the extent buffer + the offset of
6837 * the bitmap item in the extent buffer + the offset of the byte in the
6840 offset = start + offset_in_page(eb->start) + byte_offset;
6842 *page_index = offset >> PAGE_SHIFT;
6843 *page_offset = offset_in_page(offset);
6847 * extent_buffer_test_bit - determine whether a bit in a bitmap item is set
6848 * @eb: the extent buffer
6849 * @start: offset of the bitmap item in the extent buffer
6850 * @nr: bit number to test
6852 int extent_buffer_test_bit(const struct extent_buffer *eb, unsigned long start,
6860 eb_bitmap_offset(eb, start, nr, &i, &offset);
6861 page = eb->pages[i];
6862 assert_eb_page_uptodate(eb, page);
6863 kaddr = page_address(page);
6864 return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1)));
6868 * extent_buffer_bitmap_set - set an area of a bitmap
6869 * @eb: the extent buffer
6870 * @start: offset of the bitmap item in the extent buffer
6871 * @pos: bit number of the first bit
6872 * @len: number of bits to set
6874 void extent_buffer_bitmap_set(const struct extent_buffer *eb, unsigned long start,
6875 unsigned long pos, unsigned long len)
6881 const unsigned int size = pos + len;
6882 int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
6883 u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos);
6885 eb_bitmap_offset(eb, start, pos, &i, &offset);
6886 page = eb->pages[i];
6887 assert_eb_page_uptodate(eb, page);
6888 kaddr = page_address(page);
6890 while (len >= bits_to_set) {
6891 kaddr[offset] |= mask_to_set;
6893 bits_to_set = BITS_PER_BYTE;
6895 if (++offset >= PAGE_SIZE && len > 0) {
6897 page = eb->pages[++i];
6898 assert_eb_page_uptodate(eb, page);
6899 kaddr = page_address(page);
6903 mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
6904 kaddr[offset] |= mask_to_set;
6910 * extent_buffer_bitmap_clear - clear an area of a bitmap
6911 * @eb: the extent buffer
6912 * @start: offset of the bitmap item in the extent buffer
6913 * @pos: bit number of the first bit
6914 * @len: number of bits to clear
6916 void extent_buffer_bitmap_clear(const struct extent_buffer *eb,
6917 unsigned long start, unsigned long pos,
6924 const unsigned int size = pos + len;
6925 int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
6926 u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos);
6928 eb_bitmap_offset(eb, start, pos, &i, &offset);
6929 page = eb->pages[i];
6930 assert_eb_page_uptodate(eb, page);
6931 kaddr = page_address(page);
6933 while (len >= bits_to_clear) {
6934 kaddr[offset] &= ~mask_to_clear;
6935 len -= bits_to_clear;
6936 bits_to_clear = BITS_PER_BYTE;
6938 if (++offset >= PAGE_SIZE && len > 0) {
6940 page = eb->pages[++i];
6941 assert_eb_page_uptodate(eb, page);
6942 kaddr = page_address(page);
6946 mask_to_clear &= BITMAP_LAST_BYTE_MASK(size);
6947 kaddr[offset] &= ~mask_to_clear;
6951 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
6953 unsigned long distance = (src > dst) ? src - dst : dst - src;
6954 return distance < len;
6957 static void copy_pages(struct page *dst_page, struct page *src_page,
6958 unsigned long dst_off, unsigned long src_off,
6961 char *dst_kaddr = page_address(dst_page);
6963 int must_memmove = 0;
6965 if (dst_page != src_page) {
6966 src_kaddr = page_address(src_page);
6968 src_kaddr = dst_kaddr;
6969 if (areas_overlap(src_off, dst_off, len))
6974 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
6976 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
6979 void memcpy_extent_buffer(const struct extent_buffer *dst,
6980 unsigned long dst_offset, unsigned long src_offset,
6984 size_t dst_off_in_page;
6985 size_t src_off_in_page;
6986 unsigned long dst_i;
6987 unsigned long src_i;
6989 if (check_eb_range(dst, dst_offset, len) ||
6990 check_eb_range(dst, src_offset, len))
6994 dst_off_in_page = get_eb_offset_in_page(dst, dst_offset);
6995 src_off_in_page = get_eb_offset_in_page(dst, src_offset);
6997 dst_i = get_eb_page_index(dst_offset);
6998 src_i = get_eb_page_index(src_offset);
7000 cur = min(len, (unsigned long)(PAGE_SIZE -
7002 cur = min_t(unsigned long, cur,
7003 (unsigned long)(PAGE_SIZE - dst_off_in_page));
7005 copy_pages(dst->pages[dst_i], dst->pages[src_i],
7006 dst_off_in_page, src_off_in_page, cur);
7014 void memmove_extent_buffer(const struct extent_buffer *dst,
7015 unsigned long dst_offset, unsigned long src_offset,
7019 size_t dst_off_in_page;
7020 size_t src_off_in_page;
7021 unsigned long dst_end = dst_offset + len - 1;
7022 unsigned long src_end = src_offset + len - 1;
7023 unsigned long dst_i;
7024 unsigned long src_i;
7026 if (check_eb_range(dst, dst_offset, len) ||
7027 check_eb_range(dst, src_offset, len))
7029 if (dst_offset < src_offset) {
7030 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
7034 dst_i = get_eb_page_index(dst_end);
7035 src_i = get_eb_page_index(src_end);
7037 dst_off_in_page = get_eb_offset_in_page(dst, dst_end);
7038 src_off_in_page = get_eb_offset_in_page(dst, src_end);
7040 cur = min_t(unsigned long, len, src_off_in_page + 1);
7041 cur = min(cur, dst_off_in_page + 1);
7042 copy_pages(dst->pages[dst_i], dst->pages[src_i],
7043 dst_off_in_page - cur + 1,
7044 src_off_in_page - cur + 1, cur);
7052 static struct extent_buffer *get_next_extent_buffer(
7053 struct btrfs_fs_info *fs_info, struct page *page, u64 bytenr)
7055 struct extent_buffer *gang[BTRFS_SUBPAGE_BITMAP_SIZE];
7056 struct extent_buffer *found = NULL;
7057 u64 page_start = page_offset(page);
7061 ASSERT(in_range(bytenr, page_start, PAGE_SIZE));
7062 ASSERT(PAGE_SIZE / fs_info->nodesize <= BTRFS_SUBPAGE_BITMAP_SIZE);
7063 lockdep_assert_held(&fs_info->buffer_lock);
7065 ret = radix_tree_gang_lookup(&fs_info->buffer_radix, (void **)gang,
7066 bytenr >> fs_info->sectorsize_bits,
7067 PAGE_SIZE / fs_info->nodesize);
7068 for (i = 0; i < ret; i++) {
7069 /* Already beyond page end */
7070 if (gang[i]->start >= page_start + PAGE_SIZE)
7073 if (gang[i]->start >= bytenr) {
7081 static int try_release_subpage_extent_buffer(struct page *page)
7083 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
7084 u64 cur = page_offset(page);
7085 const u64 end = page_offset(page) + PAGE_SIZE;
7089 struct extent_buffer *eb = NULL;
7092 * Unlike try_release_extent_buffer() which uses page->private
7093 * to grab buffer, for subpage case we rely on radix tree, thus
7094 * we need to ensure radix tree consistency.
7096 * We also want an atomic snapshot of the radix tree, thus go
7097 * with spinlock rather than RCU.
7099 spin_lock(&fs_info->buffer_lock);
7100 eb = get_next_extent_buffer(fs_info, page, cur);
7102 /* No more eb in the page range after or at cur */
7103 spin_unlock(&fs_info->buffer_lock);
7106 cur = eb->start + eb->len;
7109 * The same as try_release_extent_buffer(), to ensure the eb
7110 * won't disappear out from under us.
7112 spin_lock(&eb->refs_lock);
7113 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
7114 spin_unlock(&eb->refs_lock);
7115 spin_unlock(&fs_info->buffer_lock);
7118 spin_unlock(&fs_info->buffer_lock);
7121 * If tree ref isn't set then we know the ref on this eb is a
7122 * real ref, so just return, this eb will likely be freed soon
7125 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
7126 spin_unlock(&eb->refs_lock);
7131 * Here we don't care about the return value, we will always
7132 * check the page private at the end. And
7133 * release_extent_buffer() will release the refs_lock.
7135 release_extent_buffer(eb);
7138 * Finally to check if we have cleared page private, as if we have
7139 * released all ebs in the page, the page private should be cleared now.
7141 spin_lock(&page->mapping->private_lock);
7142 if (!PagePrivate(page))
7146 spin_unlock(&page->mapping->private_lock);
7151 int try_release_extent_buffer(struct page *page)
7153 struct extent_buffer *eb;
7155 if (btrfs_sb(page->mapping->host->i_sb)->sectorsize < PAGE_SIZE)
7156 return try_release_subpage_extent_buffer(page);
7159 * We need to make sure nobody is changing page->private, as we rely on
7160 * page->private as the pointer to extent buffer.
7162 spin_lock(&page->mapping->private_lock);
7163 if (!PagePrivate(page)) {
7164 spin_unlock(&page->mapping->private_lock);
7168 eb = (struct extent_buffer *)page->private;
7172 * This is a little awful but should be ok, we need to make sure that
7173 * the eb doesn't disappear out from under us while we're looking at
7176 spin_lock(&eb->refs_lock);
7177 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
7178 spin_unlock(&eb->refs_lock);
7179 spin_unlock(&page->mapping->private_lock);
7182 spin_unlock(&page->mapping->private_lock);
7185 * If tree ref isn't set then we know the ref on this eb is a real ref,
7186 * so just return, this page will likely be freed soon anyway.
7188 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
7189 spin_unlock(&eb->refs_lock);
7193 return release_extent_buffer(eb);
7197 * btrfs_readahead_tree_block - attempt to readahead a child block
7198 * @fs_info: the fs_info
7199 * @bytenr: bytenr to read
7200 * @owner_root: objectid of the root that owns this eb
7201 * @gen: generation for the uptodate check, can be 0
7202 * @level: level for the eb
7204 * Attempt to readahead a tree block at @bytenr. If @gen is 0 then we do a
7205 * normal uptodate check of the eb, without checking the generation. If we have
7206 * to read the block we will not block on anything.
7208 void btrfs_readahead_tree_block(struct btrfs_fs_info *fs_info,
7209 u64 bytenr, u64 owner_root, u64 gen, int level)
7211 struct extent_buffer *eb;
7214 eb = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level);
7218 if (btrfs_buffer_uptodate(eb, gen, 1)) {
7219 free_extent_buffer(eb);
7223 ret = read_extent_buffer_pages(eb, WAIT_NONE, 0);
7225 free_extent_buffer_stale(eb);
7227 free_extent_buffer(eb);
7231 * btrfs_readahead_node_child - readahead a node's child block
7232 * @node: parent node we're reading from
7233 * @slot: slot in the parent node for the child we want to read
7235 * A helper for btrfs_readahead_tree_block, we simply read the bytenr pointed at
7236 * the slot in the node provided.
7238 void btrfs_readahead_node_child(struct extent_buffer *node, int slot)
7240 btrfs_readahead_tree_block(node->fs_info,
7241 btrfs_node_blockptr(node, slot),
7242 btrfs_header_owner(node),
7243 btrfs_node_ptr_generation(node, slot),
7244 btrfs_header_level(node) - 1);