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 {
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)
189 epd->bio->bi_status = errno_to_blk_status(ret);
196 * Submit bio from extent page data via submit_one_bio
198 * Return 0 if everything is OK.
199 * Return <0 for error.
201 static int __must_check flush_write_bio(struct extent_page_data *epd)
206 ret = submit_one_bio(epd->bio, 0, 0);
208 * Clean up of epd->bio is handled by its endio function.
209 * And endio is either triggered by successful bio execution
210 * or the error handler of submit bio hook.
211 * So at this point, no matter what happened, we don't need
212 * to clean up epd->bio.
219 int __init extent_state_cache_init(void)
221 extent_state_cache = kmem_cache_create("btrfs_extent_state",
222 sizeof(struct extent_state), 0,
223 SLAB_MEM_SPREAD, NULL);
224 if (!extent_state_cache)
229 int __init extent_io_init(void)
231 extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer",
232 sizeof(struct extent_buffer), 0,
233 SLAB_MEM_SPREAD, NULL);
234 if (!extent_buffer_cache)
237 if (bioset_init(&btrfs_bioset, BIO_POOL_SIZE,
238 offsetof(struct btrfs_io_bio, bio),
240 goto free_buffer_cache;
242 if (bioset_integrity_create(&btrfs_bioset, BIO_POOL_SIZE))
248 bioset_exit(&btrfs_bioset);
251 kmem_cache_destroy(extent_buffer_cache);
252 extent_buffer_cache = NULL;
256 void __cold extent_state_cache_exit(void)
258 btrfs_extent_state_leak_debug_check();
259 kmem_cache_destroy(extent_state_cache);
262 void __cold extent_io_exit(void)
265 * Make sure all delayed rcu free are flushed before we
269 kmem_cache_destroy(extent_buffer_cache);
270 bioset_exit(&btrfs_bioset);
274 * For the file_extent_tree, we want to hold the inode lock when we lookup and
275 * update the disk_i_size, but lockdep will complain because our io_tree we hold
276 * the tree lock and get the inode lock when setting delalloc. These two things
277 * are unrelated, so make a class for the file_extent_tree so we don't get the
278 * two locking patterns mixed up.
280 static struct lock_class_key file_extent_tree_class;
282 void extent_io_tree_init(struct btrfs_fs_info *fs_info,
283 struct extent_io_tree *tree, unsigned int owner,
286 tree->fs_info = fs_info;
287 tree->state = RB_ROOT;
288 tree->dirty_bytes = 0;
289 spin_lock_init(&tree->lock);
290 tree->private_data = private_data;
292 if (owner == IO_TREE_INODE_FILE_EXTENT)
293 lockdep_set_class(&tree->lock, &file_extent_tree_class);
296 void extent_io_tree_release(struct extent_io_tree *tree)
298 spin_lock(&tree->lock);
300 * Do a single barrier for the waitqueue_active check here, the state
301 * of the waitqueue should not change once extent_io_tree_release is
305 while (!RB_EMPTY_ROOT(&tree->state)) {
306 struct rb_node *node;
307 struct extent_state *state;
309 node = rb_first(&tree->state);
310 state = rb_entry(node, struct extent_state, rb_node);
311 rb_erase(&state->rb_node, &tree->state);
312 RB_CLEAR_NODE(&state->rb_node);
314 * btree io trees aren't supposed to have tasks waiting for
315 * changes in the flags of extent states ever.
317 ASSERT(!waitqueue_active(&state->wq));
318 free_extent_state(state);
320 cond_resched_lock(&tree->lock);
322 spin_unlock(&tree->lock);
325 static struct extent_state *alloc_extent_state(gfp_t mask)
327 struct extent_state *state;
330 * The given mask might be not appropriate for the slab allocator,
331 * drop the unsupported bits
333 mask &= ~(__GFP_DMA32|__GFP_HIGHMEM);
334 state = kmem_cache_alloc(extent_state_cache, mask);
338 state->failrec = NULL;
339 RB_CLEAR_NODE(&state->rb_node);
340 btrfs_leak_debug_add(&leak_lock, &state->leak_list, &states);
341 refcount_set(&state->refs, 1);
342 init_waitqueue_head(&state->wq);
343 trace_alloc_extent_state(state, mask, _RET_IP_);
347 void free_extent_state(struct extent_state *state)
351 if (refcount_dec_and_test(&state->refs)) {
352 WARN_ON(extent_state_in_tree(state));
353 btrfs_leak_debug_del(&leak_lock, &state->leak_list);
354 trace_free_extent_state(state, _RET_IP_);
355 kmem_cache_free(extent_state_cache, state);
359 static struct rb_node *tree_insert(struct rb_root *root,
360 struct rb_node *search_start,
362 struct rb_node *node,
363 struct rb_node ***p_in,
364 struct rb_node **parent_in)
367 struct rb_node *parent = NULL;
368 struct tree_entry *entry;
370 if (p_in && parent_in) {
376 p = search_start ? &search_start : &root->rb_node;
379 entry = rb_entry(parent, struct tree_entry, rb_node);
381 if (offset < entry->start)
383 else if (offset > entry->end)
390 rb_link_node(node, parent, p);
391 rb_insert_color(node, root);
396 * Search @tree for an entry that contains @offset. Such entry would have
397 * entry->start <= offset && entry->end >= offset.
399 * @tree: the tree to search
400 * @offset: offset that should fall within an entry in @tree
401 * @next_ret: pointer to the first entry whose range ends after @offset
402 * @prev_ret: pointer to the first entry whose range begins before @offset
403 * @p_ret: pointer where new node should be anchored (used when inserting an
405 * @parent_ret: points to entry which would have been the parent of the entry,
408 * This function returns a pointer to the entry that contains @offset byte
409 * address. If no such entry exists, then NULL is returned and the other
410 * pointer arguments to the function are filled, otherwise the found entry is
411 * returned and other pointers are left untouched.
413 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
414 struct rb_node **next_ret,
415 struct rb_node **prev_ret,
416 struct rb_node ***p_ret,
417 struct rb_node **parent_ret)
419 struct rb_root *root = &tree->state;
420 struct rb_node **n = &root->rb_node;
421 struct rb_node *prev = NULL;
422 struct rb_node *orig_prev = NULL;
423 struct tree_entry *entry;
424 struct tree_entry *prev_entry = NULL;
428 entry = rb_entry(prev, struct tree_entry, rb_node);
431 if (offset < entry->start)
433 else if (offset > entry->end)
446 while (prev && offset > prev_entry->end) {
447 prev = rb_next(prev);
448 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
455 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
456 while (prev && offset < prev_entry->start) {
457 prev = rb_prev(prev);
458 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
465 static inline struct rb_node *
466 tree_search_for_insert(struct extent_io_tree *tree,
468 struct rb_node ***p_ret,
469 struct rb_node **parent_ret)
471 struct rb_node *next= NULL;
474 ret = __etree_search(tree, offset, &next, NULL, p_ret, parent_ret);
480 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
483 return tree_search_for_insert(tree, offset, NULL, NULL);
487 * utility function to look for merge candidates inside a given range.
488 * Any extents with matching state are merged together into a single
489 * extent in the tree. Extents with EXTENT_IO in their state field
490 * are not merged because the end_io handlers need to be able to do
491 * operations on them without sleeping (or doing allocations/splits).
493 * This should be called with the tree lock held.
495 static void merge_state(struct extent_io_tree *tree,
496 struct extent_state *state)
498 struct extent_state *other;
499 struct rb_node *other_node;
501 if (state->state & (EXTENT_LOCKED | EXTENT_BOUNDARY))
504 other_node = rb_prev(&state->rb_node);
506 other = rb_entry(other_node, struct extent_state, rb_node);
507 if (other->end == state->start - 1 &&
508 other->state == state->state) {
509 if (tree->private_data &&
510 is_data_inode(tree->private_data))
511 btrfs_merge_delalloc_extent(tree->private_data,
513 state->start = other->start;
514 rb_erase(&other->rb_node, &tree->state);
515 RB_CLEAR_NODE(&other->rb_node);
516 free_extent_state(other);
519 other_node = rb_next(&state->rb_node);
521 other = rb_entry(other_node, struct extent_state, rb_node);
522 if (other->start == state->end + 1 &&
523 other->state == state->state) {
524 if (tree->private_data &&
525 is_data_inode(tree->private_data))
526 btrfs_merge_delalloc_extent(tree->private_data,
528 state->end = other->end;
529 rb_erase(&other->rb_node, &tree->state);
530 RB_CLEAR_NODE(&other->rb_node);
531 free_extent_state(other);
536 static void set_state_bits(struct extent_io_tree *tree,
537 struct extent_state *state, u32 *bits,
538 struct extent_changeset *changeset);
541 * insert an extent_state struct into the tree. 'bits' are set on the
542 * struct before it is inserted.
544 * This may return -EEXIST if the extent is already there, in which case the
545 * state struct is freed.
547 * The tree lock is not taken internally. This is a utility function and
548 * probably isn't what you want to call (see set/clear_extent_bit).
550 static int insert_state(struct extent_io_tree *tree,
551 struct extent_state *state, u64 start, u64 end,
553 struct rb_node **parent,
554 u32 *bits, struct extent_changeset *changeset)
556 struct rb_node *node;
559 btrfs_err(tree->fs_info,
560 "insert state: end < start %llu %llu", end, start);
563 state->start = start;
566 set_state_bits(tree, state, bits, changeset);
568 node = tree_insert(&tree->state, NULL, end, &state->rb_node, p, parent);
570 struct extent_state *found;
571 found = rb_entry(node, struct extent_state, rb_node);
572 btrfs_err(tree->fs_info,
573 "found node %llu %llu on insert of %llu %llu",
574 found->start, found->end, start, end);
577 merge_state(tree, state);
582 * split a given extent state struct in two, inserting the preallocated
583 * struct 'prealloc' as the newly created second half. 'split' indicates an
584 * offset inside 'orig' where it should be split.
587 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
588 * are two extent state structs in the tree:
589 * prealloc: [orig->start, split - 1]
590 * orig: [ split, orig->end ]
592 * The tree locks are not taken by this function. They need to be held
595 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
596 struct extent_state *prealloc, u64 split)
598 struct rb_node *node;
600 if (tree->private_data && is_data_inode(tree->private_data))
601 btrfs_split_delalloc_extent(tree->private_data, orig, split);
603 prealloc->start = orig->start;
604 prealloc->end = split - 1;
605 prealloc->state = orig->state;
608 node = tree_insert(&tree->state, &orig->rb_node, prealloc->end,
609 &prealloc->rb_node, NULL, NULL);
611 free_extent_state(prealloc);
617 static struct extent_state *next_state(struct extent_state *state)
619 struct rb_node *next = rb_next(&state->rb_node);
621 return rb_entry(next, struct extent_state, rb_node);
627 * utility function to clear some bits in an extent state struct.
628 * it will optionally wake up anyone waiting on this state (wake == 1).
630 * If no bits are set on the state struct after clearing things, the
631 * struct is freed and removed from the tree
633 static struct extent_state *clear_state_bit(struct extent_io_tree *tree,
634 struct extent_state *state,
636 struct extent_changeset *changeset)
638 struct extent_state *next;
639 u32 bits_to_clear = *bits & ~EXTENT_CTLBITS;
642 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
643 u64 range = state->end - state->start + 1;
644 WARN_ON(range > tree->dirty_bytes);
645 tree->dirty_bytes -= range;
648 if (tree->private_data && is_data_inode(tree->private_data))
649 btrfs_clear_delalloc_extent(tree->private_data, state, bits);
651 ret = add_extent_changeset(state, bits_to_clear, changeset, 0);
653 state->state &= ~bits_to_clear;
656 if (state->state == 0) {
657 next = next_state(state);
658 if (extent_state_in_tree(state)) {
659 rb_erase(&state->rb_node, &tree->state);
660 RB_CLEAR_NODE(&state->rb_node);
661 free_extent_state(state);
666 merge_state(tree, state);
667 next = next_state(state);
672 static struct extent_state *
673 alloc_extent_state_atomic(struct extent_state *prealloc)
676 prealloc = alloc_extent_state(GFP_ATOMIC);
681 static void extent_io_tree_panic(struct extent_io_tree *tree, int err)
683 btrfs_panic(tree->fs_info, err,
684 "locking error: extent tree was modified by another thread while locked");
688 * clear some bits on a range in the tree. This may require splitting
689 * or inserting elements in the tree, so the gfp mask is used to
690 * indicate which allocations or sleeping are allowed.
692 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
693 * the given range from the tree regardless of state (ie for truncate).
695 * the range [start, end] is inclusive.
697 * This takes the tree lock, and returns 0 on success and < 0 on error.
699 int __clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
700 u32 bits, int wake, int delete,
701 struct extent_state **cached_state,
702 gfp_t mask, struct extent_changeset *changeset)
704 struct extent_state *state;
705 struct extent_state *cached;
706 struct extent_state *prealloc = NULL;
707 struct rb_node *node;
712 btrfs_debug_check_extent_io_range(tree, start, end);
713 trace_btrfs_clear_extent_bit(tree, start, end - start + 1, bits);
715 if (bits & EXTENT_DELALLOC)
716 bits |= EXTENT_NORESERVE;
719 bits |= ~EXTENT_CTLBITS;
721 if (bits & (EXTENT_LOCKED | EXTENT_BOUNDARY))
724 if (!prealloc && gfpflags_allow_blocking(mask)) {
726 * Don't care for allocation failure here because we might end
727 * up not needing the pre-allocated extent state at all, which
728 * is the case if we only have in the tree extent states that
729 * cover our input range and don't cover too any other range.
730 * If we end up needing a new extent state we allocate it later.
732 prealloc = alloc_extent_state(mask);
735 spin_lock(&tree->lock);
737 cached = *cached_state;
740 *cached_state = NULL;
744 if (cached && extent_state_in_tree(cached) &&
745 cached->start <= start && cached->end > start) {
747 refcount_dec(&cached->refs);
752 free_extent_state(cached);
755 * this search will find the extents that end after
758 node = tree_search(tree, start);
761 state = rb_entry(node, struct extent_state, rb_node);
763 if (state->start > end)
765 WARN_ON(state->end < start);
766 last_end = state->end;
768 /* the state doesn't have the wanted bits, go ahead */
769 if (!(state->state & bits)) {
770 state = next_state(state);
775 * | ---- desired range ---- |
777 * | ------------- state -------------- |
779 * We need to split the extent we found, and may flip
780 * bits on second half.
782 * If the extent we found extends past our range, we
783 * just split and search again. It'll get split again
784 * the next time though.
786 * If the extent we found is inside our range, we clear
787 * the desired bit on it.
790 if (state->start < start) {
791 prealloc = alloc_extent_state_atomic(prealloc);
793 err = split_state(tree, state, prealloc, start);
795 extent_io_tree_panic(tree, err);
800 if (state->end <= end) {
801 state = clear_state_bit(tree, state, &bits, wake,
808 * | ---- desired range ---- |
810 * We need to split the extent, and clear the bit
813 if (state->start <= end && state->end > end) {
814 prealloc = alloc_extent_state_atomic(prealloc);
816 err = split_state(tree, state, prealloc, end + 1);
818 extent_io_tree_panic(tree, err);
823 clear_state_bit(tree, prealloc, &bits, wake, changeset);
829 state = clear_state_bit(tree, state, &bits, wake, changeset);
831 if (last_end == (u64)-1)
833 start = last_end + 1;
834 if (start <= end && state && !need_resched())
840 spin_unlock(&tree->lock);
841 if (gfpflags_allow_blocking(mask))
846 spin_unlock(&tree->lock);
848 free_extent_state(prealloc);
854 static void wait_on_state(struct extent_io_tree *tree,
855 struct extent_state *state)
856 __releases(tree->lock)
857 __acquires(tree->lock)
860 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
861 spin_unlock(&tree->lock);
863 spin_lock(&tree->lock);
864 finish_wait(&state->wq, &wait);
868 * waits for one or more bits to clear on a range in the state tree.
869 * The range [start, end] is inclusive.
870 * The tree lock is taken by this function
872 static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
875 struct extent_state *state;
876 struct rb_node *node;
878 btrfs_debug_check_extent_io_range(tree, start, end);
880 spin_lock(&tree->lock);
884 * this search will find all the extents that end after
887 node = tree_search(tree, start);
892 state = rb_entry(node, struct extent_state, rb_node);
894 if (state->start > end)
897 if (state->state & bits) {
898 start = state->start;
899 refcount_inc(&state->refs);
900 wait_on_state(tree, state);
901 free_extent_state(state);
904 start = state->end + 1;
909 if (!cond_resched_lock(&tree->lock)) {
910 node = rb_next(node);
915 spin_unlock(&tree->lock);
918 static void set_state_bits(struct extent_io_tree *tree,
919 struct extent_state *state,
920 u32 *bits, struct extent_changeset *changeset)
922 u32 bits_to_set = *bits & ~EXTENT_CTLBITS;
925 if (tree->private_data && is_data_inode(tree->private_data))
926 btrfs_set_delalloc_extent(tree->private_data, state, bits);
928 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
929 u64 range = state->end - state->start + 1;
930 tree->dirty_bytes += range;
932 ret = add_extent_changeset(state, bits_to_set, changeset, 1);
934 state->state |= bits_to_set;
937 static void cache_state_if_flags(struct extent_state *state,
938 struct extent_state **cached_ptr,
941 if (cached_ptr && !(*cached_ptr)) {
942 if (!flags || (state->state & flags)) {
944 refcount_inc(&state->refs);
949 static void cache_state(struct extent_state *state,
950 struct extent_state **cached_ptr)
952 return cache_state_if_flags(state, cached_ptr,
953 EXTENT_LOCKED | EXTENT_BOUNDARY);
957 * set some bits on a range in the tree. This may require allocations or
958 * sleeping, so the gfp mask is used to indicate what is allowed.
960 * If any of the exclusive bits are set, this will fail with -EEXIST if some
961 * part of the range already has the desired bits set. The start of the
962 * existing range is returned in failed_start in this case.
964 * [start, end] is inclusive This takes the tree lock.
966 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, u32 bits,
967 u32 exclusive_bits, u64 *failed_start,
968 struct extent_state **cached_state, gfp_t mask,
969 struct extent_changeset *changeset)
971 struct extent_state *state;
972 struct extent_state *prealloc = NULL;
973 struct rb_node *node;
975 struct rb_node *parent;
980 btrfs_debug_check_extent_io_range(tree, start, end);
981 trace_btrfs_set_extent_bit(tree, start, end - start + 1, bits);
984 ASSERT(failed_start);
986 ASSERT(failed_start == NULL);
988 if (!prealloc && gfpflags_allow_blocking(mask)) {
990 * Don't care for allocation failure here because we might end
991 * up not needing the pre-allocated extent state at all, which
992 * is the case if we only have in the tree extent states that
993 * cover our input range and don't cover too any other range.
994 * If we end up needing a new extent state we allocate it later.
996 prealloc = alloc_extent_state(mask);
999 spin_lock(&tree->lock);
1000 if (cached_state && *cached_state) {
1001 state = *cached_state;
1002 if (state->start <= start && state->end > start &&
1003 extent_state_in_tree(state)) {
1004 node = &state->rb_node;
1009 * this search will find all the extents that end after
1012 node = tree_search_for_insert(tree, start, &p, &parent);
1014 prealloc = alloc_extent_state_atomic(prealloc);
1016 err = insert_state(tree, prealloc, start, end,
1017 &p, &parent, &bits, changeset);
1019 extent_io_tree_panic(tree, err);
1021 cache_state(prealloc, cached_state);
1025 state = rb_entry(node, struct extent_state, rb_node);
1027 last_start = state->start;
1028 last_end = state->end;
1031 * | ---- desired range ---- |
1034 * Just lock what we found and keep going
1036 if (state->start == start && state->end <= end) {
1037 if (state->state & exclusive_bits) {
1038 *failed_start = state->start;
1043 set_state_bits(tree, state, &bits, changeset);
1044 cache_state(state, cached_state);
1045 merge_state(tree, state);
1046 if (last_end == (u64)-1)
1048 start = last_end + 1;
1049 state = next_state(state);
1050 if (start < end && state && state->start == start &&
1057 * | ---- desired range ---- |
1060 * | ------------- state -------------- |
1062 * We need to split the extent we found, and may flip bits on
1065 * If the extent we found extends past our
1066 * range, we just split and search again. It'll get split
1067 * again the next time though.
1069 * If the extent we found is inside our range, we set the
1070 * desired bit on it.
1072 if (state->start < start) {
1073 if (state->state & exclusive_bits) {
1074 *failed_start = start;
1080 * If this extent already has all the bits we want set, then
1081 * skip it, not necessary to split it or do anything with it.
1083 if ((state->state & bits) == bits) {
1084 start = state->end + 1;
1085 cache_state(state, cached_state);
1089 prealloc = alloc_extent_state_atomic(prealloc);
1091 err = split_state(tree, state, prealloc, start);
1093 extent_io_tree_panic(tree, err);
1098 if (state->end <= end) {
1099 set_state_bits(tree, state, &bits, changeset);
1100 cache_state(state, cached_state);
1101 merge_state(tree, state);
1102 if (last_end == (u64)-1)
1104 start = last_end + 1;
1105 state = next_state(state);
1106 if (start < end && state && state->start == start &&
1113 * | ---- desired range ---- |
1114 * | state | or | state |
1116 * There's a hole, we need to insert something in it and
1117 * ignore the extent we found.
1119 if (state->start > start) {
1121 if (end < last_start)
1124 this_end = last_start - 1;
1126 prealloc = alloc_extent_state_atomic(prealloc);
1130 * Avoid to free 'prealloc' if it can be merged with
1133 err = insert_state(tree, prealloc, start, this_end,
1134 NULL, NULL, &bits, changeset);
1136 extent_io_tree_panic(tree, err);
1138 cache_state(prealloc, cached_state);
1140 start = this_end + 1;
1144 * | ---- desired range ---- |
1146 * We need to split the extent, and set the bit
1149 if (state->start <= end && state->end > end) {
1150 if (state->state & exclusive_bits) {
1151 *failed_start = start;
1156 prealloc = alloc_extent_state_atomic(prealloc);
1158 err = split_state(tree, state, prealloc, end + 1);
1160 extent_io_tree_panic(tree, err);
1162 set_state_bits(tree, prealloc, &bits, changeset);
1163 cache_state(prealloc, cached_state);
1164 merge_state(tree, prealloc);
1172 spin_unlock(&tree->lock);
1173 if (gfpflags_allow_blocking(mask))
1178 spin_unlock(&tree->lock);
1180 free_extent_state(prealloc);
1187 * convert_extent_bit - convert all bits in a given range from one bit to
1189 * @tree: the io tree to search
1190 * @start: the start offset in bytes
1191 * @end: the end offset in bytes (inclusive)
1192 * @bits: the bits to set in this range
1193 * @clear_bits: the bits to clear in this range
1194 * @cached_state: state that we're going to cache
1196 * This will go through and set bits for the given range. If any states exist
1197 * already in this range they are set with the given bit and cleared of the
1198 * clear_bits. This is only meant to be used by things that are mergeable, ie
1199 * converting from say DELALLOC to DIRTY. This is not meant to be used with
1200 * boundary bits like LOCK.
1202 * All allocations are done with GFP_NOFS.
1204 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1205 u32 bits, u32 clear_bits,
1206 struct extent_state **cached_state)
1208 struct extent_state *state;
1209 struct extent_state *prealloc = NULL;
1210 struct rb_node *node;
1212 struct rb_node *parent;
1216 bool first_iteration = true;
1218 btrfs_debug_check_extent_io_range(tree, start, end);
1219 trace_btrfs_convert_extent_bit(tree, start, end - start + 1, bits,
1225 * Best effort, don't worry if extent state allocation fails
1226 * here for the first iteration. We might have a cached state
1227 * that matches exactly the target range, in which case no
1228 * extent state allocations are needed. We'll only know this
1229 * after locking the tree.
1231 prealloc = alloc_extent_state(GFP_NOFS);
1232 if (!prealloc && !first_iteration)
1236 spin_lock(&tree->lock);
1237 if (cached_state && *cached_state) {
1238 state = *cached_state;
1239 if (state->start <= start && state->end > start &&
1240 extent_state_in_tree(state)) {
1241 node = &state->rb_node;
1247 * this search will find all the extents that end after
1250 node = tree_search_for_insert(tree, start, &p, &parent);
1252 prealloc = alloc_extent_state_atomic(prealloc);
1257 err = insert_state(tree, prealloc, start, end,
1258 &p, &parent, &bits, NULL);
1260 extent_io_tree_panic(tree, err);
1261 cache_state(prealloc, cached_state);
1265 state = rb_entry(node, struct extent_state, rb_node);
1267 last_start = state->start;
1268 last_end = state->end;
1271 * | ---- desired range ---- |
1274 * Just lock what we found and keep going
1276 if (state->start == start && state->end <= end) {
1277 set_state_bits(tree, state, &bits, NULL);
1278 cache_state(state, cached_state);
1279 state = clear_state_bit(tree, state, &clear_bits, 0, NULL);
1280 if (last_end == (u64)-1)
1282 start = last_end + 1;
1283 if (start < end && state && state->start == start &&
1290 * | ---- desired range ---- |
1293 * | ------------- state -------------- |
1295 * We need to split the extent we found, and may flip bits on
1298 * If the extent we found extends past our
1299 * range, we just split and search again. It'll get split
1300 * again the next time though.
1302 * If the extent we found is inside our range, we set the
1303 * desired bit on it.
1305 if (state->start < start) {
1306 prealloc = alloc_extent_state_atomic(prealloc);
1311 err = split_state(tree, state, prealloc, start);
1313 extent_io_tree_panic(tree, err);
1317 if (state->end <= end) {
1318 set_state_bits(tree, state, &bits, NULL);
1319 cache_state(state, cached_state);
1320 state = clear_state_bit(tree, state, &clear_bits, 0,
1322 if (last_end == (u64)-1)
1324 start = last_end + 1;
1325 if (start < end && state && state->start == start &&
1332 * | ---- desired range ---- |
1333 * | state | or | state |
1335 * There's a hole, we need to insert something in it and
1336 * ignore the extent we found.
1338 if (state->start > start) {
1340 if (end < last_start)
1343 this_end = last_start - 1;
1345 prealloc = alloc_extent_state_atomic(prealloc);
1352 * Avoid to free 'prealloc' if it can be merged with
1355 err = insert_state(tree, prealloc, start, this_end,
1356 NULL, NULL, &bits, NULL);
1358 extent_io_tree_panic(tree, err);
1359 cache_state(prealloc, cached_state);
1361 start = this_end + 1;
1365 * | ---- desired range ---- |
1367 * We need to split the extent, and set the bit
1370 if (state->start <= end && state->end > end) {
1371 prealloc = alloc_extent_state_atomic(prealloc);
1377 err = split_state(tree, state, prealloc, end + 1);
1379 extent_io_tree_panic(tree, err);
1381 set_state_bits(tree, prealloc, &bits, NULL);
1382 cache_state(prealloc, cached_state);
1383 clear_state_bit(tree, prealloc, &clear_bits, 0, NULL);
1391 spin_unlock(&tree->lock);
1393 first_iteration = false;
1397 spin_unlock(&tree->lock);
1399 free_extent_state(prealloc);
1404 /* wrappers around set/clear extent bit */
1405 int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1406 u32 bits, struct extent_changeset *changeset)
1409 * We don't support EXTENT_LOCKED yet, as current changeset will
1410 * record any bits changed, so for EXTENT_LOCKED case, it will
1411 * either fail with -EEXIST or changeset will record the whole
1414 BUG_ON(bits & EXTENT_LOCKED);
1416 return set_extent_bit(tree, start, end, bits, 0, NULL, NULL, GFP_NOFS,
1420 int set_extent_bits_nowait(struct extent_io_tree *tree, u64 start, u64 end,
1423 return set_extent_bit(tree, start, end, bits, 0, NULL, NULL,
1427 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1428 u32 bits, int wake, int delete,
1429 struct extent_state **cached)
1431 return __clear_extent_bit(tree, start, end, bits, wake, delete,
1432 cached, GFP_NOFS, NULL);
1435 int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1436 u32 bits, struct extent_changeset *changeset)
1439 * Don't support EXTENT_LOCKED case, same reason as
1440 * set_record_extent_bits().
1442 BUG_ON(bits & EXTENT_LOCKED);
1444 return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, GFP_NOFS,
1449 * either insert or lock state struct between start and end use mask to tell
1450 * us if waiting is desired.
1452 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1453 struct extent_state **cached_state)
1459 err = set_extent_bit(tree, start, end, EXTENT_LOCKED,
1460 EXTENT_LOCKED, &failed_start,
1461 cached_state, GFP_NOFS, NULL);
1462 if (err == -EEXIST) {
1463 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1464 start = failed_start;
1467 WARN_ON(start > end);
1472 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1477 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1478 &failed_start, NULL, GFP_NOFS, NULL);
1479 if (err == -EEXIST) {
1480 if (failed_start > start)
1481 clear_extent_bit(tree, start, failed_start - 1,
1482 EXTENT_LOCKED, 1, 0, NULL);
1488 void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
1490 unsigned long index = start >> PAGE_SHIFT;
1491 unsigned long end_index = end >> PAGE_SHIFT;
1494 while (index <= end_index) {
1495 page = find_get_page(inode->i_mapping, index);
1496 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1497 clear_page_dirty_for_io(page);
1503 void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
1505 unsigned long index = start >> PAGE_SHIFT;
1506 unsigned long end_index = end >> PAGE_SHIFT;
1509 while (index <= end_index) {
1510 page = find_get_page(inode->i_mapping, index);
1511 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1512 __set_page_dirty_nobuffers(page);
1513 account_page_redirty(page);
1519 /* find the first state struct with 'bits' set after 'start', and
1520 * return it. tree->lock must be held. NULL will returned if
1521 * nothing was found after 'start'
1523 static struct extent_state *
1524 find_first_extent_bit_state(struct extent_io_tree *tree, u64 start, u32 bits)
1526 struct rb_node *node;
1527 struct extent_state *state;
1530 * this search will find all the extents that end after
1533 node = tree_search(tree, start);
1538 state = rb_entry(node, struct extent_state, rb_node);
1539 if (state->end >= start && (state->state & bits))
1542 node = rb_next(node);
1551 * Find the first offset in the io tree with one or more @bits set.
1553 * Note: If there are multiple bits set in @bits, any of them will match.
1555 * Return 0 if we find something, and update @start_ret and @end_ret.
1556 * Return 1 if we found nothing.
1558 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1559 u64 *start_ret, u64 *end_ret, u32 bits,
1560 struct extent_state **cached_state)
1562 struct extent_state *state;
1565 spin_lock(&tree->lock);
1566 if (cached_state && *cached_state) {
1567 state = *cached_state;
1568 if (state->end == start - 1 && extent_state_in_tree(state)) {
1569 while ((state = next_state(state)) != NULL) {
1570 if (state->state & bits)
1573 free_extent_state(*cached_state);
1574 *cached_state = NULL;
1577 free_extent_state(*cached_state);
1578 *cached_state = NULL;
1581 state = find_first_extent_bit_state(tree, start, bits);
1584 cache_state_if_flags(state, cached_state, 0);
1585 *start_ret = state->start;
1586 *end_ret = state->end;
1590 spin_unlock(&tree->lock);
1595 * Find a contiguous area of bits
1597 * @tree: io tree to check
1598 * @start: offset to start the search from
1599 * @start_ret: the first offset we found with the bits set
1600 * @end_ret: the final contiguous range of the bits that were set
1601 * @bits: bits to look for
1603 * set_extent_bit and clear_extent_bit can temporarily split contiguous ranges
1604 * to set bits appropriately, and then merge them again. During this time it
1605 * will drop the tree->lock, so use this helper if you want to find the actual
1606 * contiguous area for given bits. We will search to the first bit we find, and
1607 * then walk down the tree until we find a non-contiguous area. The area
1608 * returned will be the full contiguous area with the bits set.
1610 int find_contiguous_extent_bit(struct extent_io_tree *tree, u64 start,
1611 u64 *start_ret, u64 *end_ret, u32 bits)
1613 struct extent_state *state;
1616 spin_lock(&tree->lock);
1617 state = find_first_extent_bit_state(tree, start, bits);
1619 *start_ret = state->start;
1620 *end_ret = state->end;
1621 while ((state = next_state(state)) != NULL) {
1622 if (state->start > (*end_ret + 1))
1624 *end_ret = state->end;
1628 spin_unlock(&tree->lock);
1633 * Find the first range that has @bits not set. This range could start before
1636 * @tree: the tree to search
1637 * @start: offset at/after which the found extent should start
1638 * @start_ret: records the beginning of the range
1639 * @end_ret: records the end of the range (inclusive)
1640 * @bits: the set of bits which must be unset
1642 * Since unallocated range is also considered one which doesn't have the bits
1643 * set it's possible that @end_ret contains -1, this happens in case the range
1644 * spans (last_range_end, end of device]. In this case it's up to the caller to
1645 * trim @end_ret to the appropriate size.
1647 void find_first_clear_extent_bit(struct extent_io_tree *tree, u64 start,
1648 u64 *start_ret, u64 *end_ret, u32 bits)
1650 struct extent_state *state;
1651 struct rb_node *node, *prev = NULL, *next;
1653 spin_lock(&tree->lock);
1655 /* Find first extent with bits cleared */
1657 node = __etree_search(tree, start, &next, &prev, NULL, NULL);
1658 if (!node && !next && !prev) {
1660 * Tree is completely empty, send full range and let
1661 * caller deal with it
1666 } else if (!node && !next) {
1668 * We are past the last allocated chunk, set start at
1669 * the end of the last extent.
1671 state = rb_entry(prev, struct extent_state, rb_node);
1672 *start_ret = state->end + 1;
1679 * At this point 'node' either contains 'start' or start is
1682 state = rb_entry(node, struct extent_state, rb_node);
1684 if (in_range(start, state->start, state->end - state->start + 1)) {
1685 if (state->state & bits) {
1687 * |--range with bits sets--|
1691 start = state->end + 1;
1694 * 'start' falls within a range that doesn't
1695 * have the bits set, so take its start as
1696 * the beginning of the desired range
1698 * |--range with bits cleared----|
1702 *start_ret = state->start;
1707 * |---prev range---|---hole/unset---|---node range---|
1713 * |---hole/unset--||--first node--|
1718 state = rb_entry(prev, struct extent_state,
1720 *start_ret = state->end + 1;
1729 * Find the longest stretch from start until an entry which has the
1733 state = rb_entry(node, struct extent_state, rb_node);
1734 if (state->end >= start && !(state->state & bits)) {
1735 *end_ret = state->end;
1737 *end_ret = state->start - 1;
1741 node = rb_next(node);
1746 spin_unlock(&tree->lock);
1750 * find a contiguous range of bytes in the file marked as delalloc, not
1751 * more than 'max_bytes'. start and end are used to return the range,
1753 * true is returned if we find something, false if nothing was in the tree
1755 bool btrfs_find_delalloc_range(struct extent_io_tree *tree, u64 *start,
1756 u64 *end, u64 max_bytes,
1757 struct extent_state **cached_state)
1759 struct rb_node *node;
1760 struct extent_state *state;
1761 u64 cur_start = *start;
1763 u64 total_bytes = 0;
1765 spin_lock(&tree->lock);
1768 * this search will find all the extents that end after
1771 node = tree_search(tree, cur_start);
1778 state = rb_entry(node, struct extent_state, rb_node);
1779 if (found && (state->start != cur_start ||
1780 (state->state & EXTENT_BOUNDARY))) {
1783 if (!(state->state & EXTENT_DELALLOC)) {
1789 *start = state->start;
1790 *cached_state = state;
1791 refcount_inc(&state->refs);
1795 cur_start = state->end + 1;
1796 node = rb_next(node);
1797 total_bytes += state->end - state->start + 1;
1798 if (total_bytes >= max_bytes)
1804 spin_unlock(&tree->lock);
1808 static int __process_pages_contig(struct address_space *mapping,
1809 struct page *locked_page,
1810 pgoff_t start_index, pgoff_t end_index,
1811 unsigned long page_ops, pgoff_t *index_ret);
1813 static noinline void __unlock_for_delalloc(struct inode *inode,
1814 struct page *locked_page,
1817 unsigned long index = start >> PAGE_SHIFT;
1818 unsigned long end_index = end >> PAGE_SHIFT;
1820 ASSERT(locked_page);
1821 if (index == locked_page->index && end_index == index)
1824 __process_pages_contig(inode->i_mapping, locked_page, index, end_index,
1828 static noinline int lock_delalloc_pages(struct inode *inode,
1829 struct page *locked_page,
1833 unsigned long index = delalloc_start >> PAGE_SHIFT;
1834 unsigned long index_ret = index;
1835 unsigned long end_index = delalloc_end >> PAGE_SHIFT;
1838 ASSERT(locked_page);
1839 if (index == locked_page->index && index == end_index)
1842 ret = __process_pages_contig(inode->i_mapping, locked_page, index,
1843 end_index, PAGE_LOCK, &index_ret);
1845 __unlock_for_delalloc(inode, locked_page, delalloc_start,
1846 (u64)index_ret << PAGE_SHIFT);
1851 * Find and lock a contiguous range of bytes in the file marked as delalloc, no
1852 * more than @max_bytes. @Start and @end are used to return the range,
1854 * Return: true if we find something
1855 * false if nothing was in the tree
1858 noinline_for_stack bool find_lock_delalloc_range(struct inode *inode,
1859 struct page *locked_page, u64 *start,
1862 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
1863 u64 max_bytes = BTRFS_MAX_EXTENT_SIZE;
1867 struct extent_state *cached_state = NULL;
1872 /* step one, find a bunch of delalloc bytes starting at start */
1873 delalloc_start = *start;
1875 found = btrfs_find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1876 max_bytes, &cached_state);
1877 if (!found || delalloc_end <= *start) {
1878 *start = delalloc_start;
1879 *end = delalloc_end;
1880 free_extent_state(cached_state);
1885 * start comes from the offset of locked_page. We have to lock
1886 * pages in order, so we can't process delalloc bytes before
1889 if (delalloc_start < *start)
1890 delalloc_start = *start;
1893 * make sure to limit the number of pages we try to lock down
1895 if (delalloc_end + 1 - delalloc_start > max_bytes)
1896 delalloc_end = delalloc_start + max_bytes - 1;
1898 /* step two, lock all the pages after the page that has start */
1899 ret = lock_delalloc_pages(inode, locked_page,
1900 delalloc_start, delalloc_end);
1901 ASSERT(!ret || ret == -EAGAIN);
1902 if (ret == -EAGAIN) {
1903 /* some of the pages are gone, lets avoid looping by
1904 * shortening the size of the delalloc range we're searching
1906 free_extent_state(cached_state);
1907 cached_state = NULL;
1909 max_bytes = PAGE_SIZE;
1918 /* step three, lock the state bits for the whole range */
1919 lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state);
1921 /* then test to make sure it is all still delalloc */
1922 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1923 EXTENT_DELALLOC, 1, cached_state);
1925 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1927 __unlock_for_delalloc(inode, locked_page,
1928 delalloc_start, delalloc_end);
1932 free_extent_state(cached_state);
1933 *start = delalloc_start;
1934 *end = delalloc_end;
1939 static int __process_pages_contig(struct address_space *mapping,
1940 struct page *locked_page,
1941 pgoff_t start_index, pgoff_t end_index,
1942 unsigned long page_ops, pgoff_t *index_ret)
1944 unsigned long nr_pages = end_index - start_index + 1;
1945 unsigned long pages_processed = 0;
1946 pgoff_t index = start_index;
1947 struct page *pages[16];
1952 if (page_ops & PAGE_LOCK) {
1953 ASSERT(page_ops == PAGE_LOCK);
1954 ASSERT(index_ret && *index_ret == start_index);
1957 if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0)
1958 mapping_set_error(mapping, -EIO);
1960 while (nr_pages > 0) {
1961 ret = find_get_pages_contig(mapping, index,
1962 min_t(unsigned long,
1963 nr_pages, ARRAY_SIZE(pages)), pages);
1966 * Only if we're going to lock these pages,
1967 * can we find nothing at @index.
1969 ASSERT(page_ops & PAGE_LOCK);
1974 for (i = 0; i < ret; i++) {
1975 if (page_ops & PAGE_SET_PRIVATE2)
1976 SetPagePrivate2(pages[i]);
1978 if (locked_page && pages[i] == locked_page) {
1983 if (page_ops & PAGE_START_WRITEBACK) {
1984 clear_page_dirty_for_io(pages[i]);
1985 set_page_writeback(pages[i]);
1987 if (page_ops & PAGE_SET_ERROR)
1988 SetPageError(pages[i]);
1989 if (page_ops & PAGE_END_WRITEBACK)
1990 end_page_writeback(pages[i]);
1991 if (page_ops & PAGE_UNLOCK)
1992 unlock_page(pages[i]);
1993 if (page_ops & PAGE_LOCK) {
1994 lock_page(pages[i]);
1995 if (!PageDirty(pages[i]) ||
1996 pages[i]->mapping != mapping) {
1997 unlock_page(pages[i]);
1998 for (; i < ret; i++)
2012 if (err && index_ret)
2013 *index_ret = start_index + pages_processed - 1;
2017 void extent_clear_unlock_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2018 struct page *locked_page,
2019 u32 clear_bits, unsigned long page_ops)
2021 clear_extent_bit(&inode->io_tree, start, end, clear_bits, 1, 0, NULL);
2023 __process_pages_contig(inode->vfs_inode.i_mapping, locked_page,
2024 start >> PAGE_SHIFT, end >> PAGE_SHIFT,
2029 * count the number of bytes in the tree that have a given bit(s)
2030 * set. This can be fairly slow, except for EXTENT_DIRTY which is
2031 * cached. The total number found is returned.
2033 u64 count_range_bits(struct extent_io_tree *tree,
2034 u64 *start, u64 search_end, u64 max_bytes,
2035 u32 bits, int contig)
2037 struct rb_node *node;
2038 struct extent_state *state;
2039 u64 cur_start = *start;
2040 u64 total_bytes = 0;
2044 if (WARN_ON(search_end <= cur_start))
2047 spin_lock(&tree->lock);
2048 if (cur_start == 0 && bits == EXTENT_DIRTY) {
2049 total_bytes = tree->dirty_bytes;
2053 * this search will find all the extents that end after
2056 node = tree_search(tree, cur_start);
2061 state = rb_entry(node, struct extent_state, rb_node);
2062 if (state->start > search_end)
2064 if (contig && found && state->start > last + 1)
2066 if (state->end >= cur_start && (state->state & bits) == bits) {
2067 total_bytes += min(search_end, state->end) + 1 -
2068 max(cur_start, state->start);
2069 if (total_bytes >= max_bytes)
2072 *start = max(cur_start, state->start);
2076 } else if (contig && found) {
2079 node = rb_next(node);
2084 spin_unlock(&tree->lock);
2089 * set the private field for a given byte offset in the tree. If there isn't
2090 * an extent_state there already, this does nothing.
2092 int set_state_failrec(struct extent_io_tree *tree, u64 start,
2093 struct io_failure_record *failrec)
2095 struct rb_node *node;
2096 struct extent_state *state;
2099 spin_lock(&tree->lock);
2101 * this search will find all the extents that end after
2104 node = tree_search(tree, start);
2109 state = rb_entry(node, struct extent_state, rb_node);
2110 if (state->start != start) {
2114 state->failrec = failrec;
2116 spin_unlock(&tree->lock);
2120 struct io_failure_record *get_state_failrec(struct extent_io_tree *tree, u64 start)
2122 struct rb_node *node;
2123 struct extent_state *state;
2124 struct io_failure_record *failrec;
2126 spin_lock(&tree->lock);
2128 * this search will find all the extents that end after
2131 node = tree_search(tree, start);
2133 failrec = ERR_PTR(-ENOENT);
2136 state = rb_entry(node, struct extent_state, rb_node);
2137 if (state->start != start) {
2138 failrec = ERR_PTR(-ENOENT);
2142 failrec = state->failrec;
2144 spin_unlock(&tree->lock);
2149 * searches a range in the state tree for a given mask.
2150 * If 'filled' == 1, this returns 1 only if every extent in the tree
2151 * has the bits set. Otherwise, 1 is returned if any bit in the
2152 * range is found set.
2154 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
2155 u32 bits, int filled, struct extent_state *cached)
2157 struct extent_state *state = NULL;
2158 struct rb_node *node;
2161 spin_lock(&tree->lock);
2162 if (cached && extent_state_in_tree(cached) && cached->start <= start &&
2163 cached->end > start)
2164 node = &cached->rb_node;
2166 node = tree_search(tree, start);
2167 while (node && start <= end) {
2168 state = rb_entry(node, struct extent_state, rb_node);
2170 if (filled && state->start > start) {
2175 if (state->start > end)
2178 if (state->state & bits) {
2182 } else if (filled) {
2187 if (state->end == (u64)-1)
2190 start = state->end + 1;
2193 node = rb_next(node);
2200 spin_unlock(&tree->lock);
2205 * helper function to set a given page up to date if all the
2206 * extents in the tree for that page are up to date
2208 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
2210 u64 start = page_offset(page);
2211 u64 end = start + PAGE_SIZE - 1;
2212 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
2213 SetPageUptodate(page);
2216 int free_io_failure(struct extent_io_tree *failure_tree,
2217 struct extent_io_tree *io_tree,
2218 struct io_failure_record *rec)
2223 set_state_failrec(failure_tree, rec->start, NULL);
2224 ret = clear_extent_bits(failure_tree, rec->start,
2225 rec->start + rec->len - 1,
2226 EXTENT_LOCKED | EXTENT_DIRTY);
2230 ret = clear_extent_bits(io_tree, rec->start,
2231 rec->start + rec->len - 1,
2241 * this bypasses the standard btrfs submit functions deliberately, as
2242 * the standard behavior is to write all copies in a raid setup. here we only
2243 * want to write the one bad copy. so we do the mapping for ourselves and issue
2244 * submit_bio directly.
2245 * to avoid any synchronization issues, wait for the data after writing, which
2246 * actually prevents the read that triggered the error from finishing.
2247 * currently, there can be no more than two copies of every data bit. thus,
2248 * exactly one rewrite is required.
2250 int repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start,
2251 u64 length, u64 logical, struct page *page,
2252 unsigned int pg_offset, int mirror_num)
2255 struct btrfs_device *dev;
2258 struct btrfs_bio *bbio = NULL;
2261 ASSERT(!(fs_info->sb->s_flags & SB_RDONLY));
2262 BUG_ON(!mirror_num);
2264 if (btrfs_is_zoned(fs_info))
2265 return btrfs_repair_one_zone(fs_info, logical);
2267 bio = btrfs_io_bio_alloc(1);
2268 bio->bi_iter.bi_size = 0;
2269 map_length = length;
2272 * Avoid races with device replace and make sure our bbio has devices
2273 * associated to its stripes that don't go away while we are doing the
2274 * read repair operation.
2276 btrfs_bio_counter_inc_blocked(fs_info);
2277 if (btrfs_is_parity_mirror(fs_info, logical, length)) {
2279 * Note that we don't use BTRFS_MAP_WRITE because it's supposed
2280 * to update all raid stripes, but here we just want to correct
2281 * bad stripe, thus BTRFS_MAP_READ is abused to only get the bad
2282 * stripe's dev and sector.
2284 ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, logical,
2285 &map_length, &bbio, 0);
2287 btrfs_bio_counter_dec(fs_info);
2291 ASSERT(bbio->mirror_num == 1);
2293 ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical,
2294 &map_length, &bbio, mirror_num);
2296 btrfs_bio_counter_dec(fs_info);
2300 BUG_ON(mirror_num != bbio->mirror_num);
2303 sector = bbio->stripes[bbio->mirror_num - 1].physical >> 9;
2304 bio->bi_iter.bi_sector = sector;
2305 dev = bbio->stripes[bbio->mirror_num - 1].dev;
2306 btrfs_put_bbio(bbio);
2307 if (!dev || !dev->bdev ||
2308 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) {
2309 btrfs_bio_counter_dec(fs_info);
2313 bio_set_dev(bio, dev->bdev);
2314 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC;
2315 bio_add_page(bio, page, length, pg_offset);
2317 if (btrfsic_submit_bio_wait(bio)) {
2318 /* try to remap that extent elsewhere? */
2319 btrfs_bio_counter_dec(fs_info);
2321 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
2325 btrfs_info_rl_in_rcu(fs_info,
2326 "read error corrected: ino %llu off %llu (dev %s sector %llu)",
2328 rcu_str_deref(dev->name), sector);
2329 btrfs_bio_counter_dec(fs_info);
2334 int btrfs_repair_eb_io_failure(const struct extent_buffer *eb, int mirror_num)
2336 struct btrfs_fs_info *fs_info = eb->fs_info;
2337 u64 start = eb->start;
2338 int i, num_pages = num_extent_pages(eb);
2341 if (sb_rdonly(fs_info->sb))
2344 for (i = 0; i < num_pages; i++) {
2345 struct page *p = eb->pages[i];
2347 ret = repair_io_failure(fs_info, 0, start, PAGE_SIZE, start, p,
2348 start - page_offset(p), mirror_num);
2358 * each time an IO finishes, we do a fast check in the IO failure tree
2359 * to see if we need to process or clean up an io_failure_record
2361 int clean_io_failure(struct btrfs_fs_info *fs_info,
2362 struct extent_io_tree *failure_tree,
2363 struct extent_io_tree *io_tree, u64 start,
2364 struct page *page, u64 ino, unsigned int pg_offset)
2367 struct io_failure_record *failrec;
2368 struct extent_state *state;
2373 ret = count_range_bits(failure_tree, &private, (u64)-1, 1,
2378 failrec = get_state_failrec(failure_tree, start);
2379 if (IS_ERR(failrec))
2382 BUG_ON(!failrec->this_mirror);
2384 if (failrec->in_validation) {
2385 /* there was no real error, just free the record */
2386 btrfs_debug(fs_info,
2387 "clean_io_failure: freeing dummy error at %llu",
2391 if (sb_rdonly(fs_info->sb))
2394 spin_lock(&io_tree->lock);
2395 state = find_first_extent_bit_state(io_tree,
2398 spin_unlock(&io_tree->lock);
2400 if (state && state->start <= failrec->start &&
2401 state->end >= failrec->start + failrec->len - 1) {
2402 num_copies = btrfs_num_copies(fs_info, failrec->logical,
2404 if (num_copies > 1) {
2405 repair_io_failure(fs_info, ino, start, failrec->len,
2406 failrec->logical, page, pg_offset,
2407 failrec->failed_mirror);
2412 free_io_failure(failure_tree, io_tree, failrec);
2418 * Can be called when
2419 * - hold extent lock
2420 * - under ordered extent
2421 * - the inode is freeing
2423 void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start, u64 end)
2425 struct extent_io_tree *failure_tree = &inode->io_failure_tree;
2426 struct io_failure_record *failrec;
2427 struct extent_state *state, *next;
2429 if (RB_EMPTY_ROOT(&failure_tree->state))
2432 spin_lock(&failure_tree->lock);
2433 state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY);
2435 if (state->start > end)
2438 ASSERT(state->end <= end);
2440 next = next_state(state);
2442 failrec = state->failrec;
2443 free_extent_state(state);
2448 spin_unlock(&failure_tree->lock);
2451 static struct io_failure_record *btrfs_get_io_failure_record(struct inode *inode,
2454 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2455 struct io_failure_record *failrec;
2456 struct extent_map *em;
2457 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2458 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2459 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2463 failrec = get_state_failrec(failure_tree, start);
2464 if (!IS_ERR(failrec)) {
2465 btrfs_debug(fs_info,
2466 "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d",
2467 failrec->logical, failrec->start, failrec->len,
2468 failrec->in_validation);
2470 * when data can be on disk more than twice, add to failrec here
2471 * (e.g. with a list for failed_mirror) to make
2472 * clean_io_failure() clean all those errors at once.
2478 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2480 return ERR_PTR(-ENOMEM);
2482 failrec->start = start;
2483 failrec->len = end - start + 1;
2484 failrec->this_mirror = 0;
2485 failrec->bio_flags = 0;
2486 failrec->in_validation = 0;
2488 read_lock(&em_tree->lock);
2489 em = lookup_extent_mapping(em_tree, start, failrec->len);
2491 read_unlock(&em_tree->lock);
2493 return ERR_PTR(-EIO);
2496 if (em->start > start || em->start + em->len <= start) {
2497 free_extent_map(em);
2500 read_unlock(&em_tree->lock);
2503 return ERR_PTR(-EIO);
2506 logical = start - em->start;
2507 logical = em->block_start + logical;
2508 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2509 logical = em->block_start;
2510 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2511 extent_set_compress_type(&failrec->bio_flags, em->compress_type);
2514 btrfs_debug(fs_info,
2515 "Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu",
2516 logical, start, failrec->len);
2518 failrec->logical = logical;
2519 free_extent_map(em);
2521 /* Set the bits in the private failure tree */
2522 ret = set_extent_bits(failure_tree, start, end,
2523 EXTENT_LOCKED | EXTENT_DIRTY);
2525 ret = set_state_failrec(failure_tree, start, failrec);
2526 /* Set the bits in the inode's tree */
2527 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED);
2528 } else if (ret < 0) {
2530 return ERR_PTR(ret);
2536 static bool btrfs_check_repairable(struct inode *inode, bool needs_validation,
2537 struct io_failure_record *failrec,
2540 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2543 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
2544 if (num_copies == 1) {
2546 * we only have a single copy of the data, so don't bother with
2547 * all the retry and error correction code that follows. no
2548 * matter what the error is, it is very likely to persist.
2550 btrfs_debug(fs_info,
2551 "Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
2552 num_copies, failrec->this_mirror, failed_mirror);
2557 * there are two premises:
2558 * a) deliver good data to the caller
2559 * b) correct the bad sectors on disk
2561 if (needs_validation) {
2563 * to fulfill b), we need to know the exact failing sectors, as
2564 * we don't want to rewrite any more than the failed ones. thus,
2565 * we need separate read requests for the failed bio
2567 * if the following BUG_ON triggers, our validation request got
2568 * merged. we need separate requests for our algorithm to work.
2570 BUG_ON(failrec->in_validation);
2571 failrec->in_validation = 1;
2572 failrec->this_mirror = failed_mirror;
2575 * we're ready to fulfill a) and b) alongside. get a good copy
2576 * of the failed sector and if we succeed, we have setup
2577 * everything for repair_io_failure to do the rest for us.
2579 if (failrec->in_validation) {
2580 BUG_ON(failrec->this_mirror != failed_mirror);
2581 failrec->in_validation = 0;
2582 failrec->this_mirror = 0;
2584 failrec->failed_mirror = failed_mirror;
2585 failrec->this_mirror++;
2586 if (failrec->this_mirror == failed_mirror)
2587 failrec->this_mirror++;
2590 if (failrec->this_mirror > num_copies) {
2591 btrfs_debug(fs_info,
2592 "Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
2593 num_copies, failrec->this_mirror, failed_mirror);
2600 static bool btrfs_io_needs_validation(struct inode *inode, struct bio *bio)
2603 const u32 blocksize = inode->i_sb->s_blocksize;
2606 * If bi_status is BLK_STS_OK, then this was a checksum error, not an
2607 * I/O error. In this case, we already know exactly which sector was
2608 * bad, so we don't need to validate.
2610 if (bio->bi_status == BLK_STS_OK)
2614 * We need to validate each sector individually if the failed I/O was
2615 * for multiple sectors.
2617 * There are a few possible bios that can end up here:
2618 * 1. A buffered read bio, which is not cloned.
2619 * 2. A direct I/O read bio, which is cloned.
2620 * 3. A (buffered or direct) repair bio, which is not cloned.
2622 * For cloned bios (case 2), we can get the size from
2623 * btrfs_io_bio->iter; for non-cloned bios (cases 1 and 3), we can get
2624 * it from the bvecs.
2626 if (bio_flagged(bio, BIO_CLONED)) {
2627 if (btrfs_io_bio(bio)->iter.bi_size > blocksize)
2630 struct bio_vec *bvec;
2633 bio_for_each_bvec_all(bvec, bio, i) {
2634 len += bvec->bv_len;
2635 if (len > blocksize)
2642 blk_status_t btrfs_submit_read_repair(struct inode *inode,
2643 struct bio *failed_bio, u32 bio_offset,
2644 struct page *page, unsigned int pgoff,
2645 u64 start, u64 end, int failed_mirror,
2646 submit_bio_hook_t *submit_bio_hook)
2648 struct io_failure_record *failrec;
2649 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2650 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2651 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2652 struct btrfs_io_bio *failed_io_bio = btrfs_io_bio(failed_bio);
2653 const int icsum = bio_offset >> fs_info->sectorsize_bits;
2654 bool need_validation;
2655 struct bio *repair_bio;
2656 struct btrfs_io_bio *repair_io_bio;
2657 blk_status_t status;
2659 btrfs_debug(fs_info,
2660 "repair read error: read error at %llu", start);
2662 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
2664 failrec = btrfs_get_io_failure_record(inode, start, end);
2665 if (IS_ERR(failrec))
2666 return errno_to_blk_status(PTR_ERR(failrec));
2668 need_validation = btrfs_io_needs_validation(inode, failed_bio);
2670 if (!btrfs_check_repairable(inode, need_validation, failrec,
2672 free_io_failure(failure_tree, tree, failrec);
2673 return BLK_STS_IOERR;
2676 repair_bio = btrfs_io_bio_alloc(1);
2677 repair_io_bio = btrfs_io_bio(repair_bio);
2678 repair_bio->bi_opf = REQ_OP_READ;
2679 if (need_validation)
2680 repair_bio->bi_opf |= REQ_FAILFAST_DEV;
2681 repair_bio->bi_end_io = failed_bio->bi_end_io;
2682 repair_bio->bi_iter.bi_sector = failrec->logical >> 9;
2683 repair_bio->bi_private = failed_bio->bi_private;
2685 if (failed_io_bio->csum) {
2686 const u32 csum_size = fs_info->csum_size;
2688 repair_io_bio->csum = repair_io_bio->csum_inline;
2689 memcpy(repair_io_bio->csum,
2690 failed_io_bio->csum + csum_size * icsum, csum_size);
2693 bio_add_page(repair_bio, page, failrec->len, pgoff);
2694 repair_io_bio->logical = failrec->start;
2695 repair_io_bio->iter = repair_bio->bi_iter;
2697 btrfs_debug(btrfs_sb(inode->i_sb),
2698 "repair read error: submitting new read to mirror %d, in_validation=%d",
2699 failrec->this_mirror, failrec->in_validation);
2701 status = submit_bio_hook(inode, repair_bio, failrec->this_mirror,
2702 failrec->bio_flags);
2704 free_io_failure(failure_tree, tree, failrec);
2705 bio_put(repair_bio);
2710 /* lots and lots of room for performance fixes in the end_bio funcs */
2712 void end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2714 int uptodate = (err == 0);
2717 btrfs_writepage_endio_finish_ordered(page, start, end, uptodate);
2720 ClearPageUptodate(page);
2722 ret = err < 0 ? err : -EIO;
2723 mapping_set_error(page->mapping, ret);
2728 * after a writepage IO is done, we need to:
2729 * clear the uptodate bits on error
2730 * clear the writeback bits in the extent tree for this IO
2731 * end_page_writeback if the page has no more pending IO
2733 * Scheduling is not allowed, so the extent state tree is expected
2734 * to have one and only one object corresponding to this IO.
2736 static void end_bio_extent_writepage(struct bio *bio)
2738 int error = blk_status_to_errno(bio->bi_status);
2739 struct bio_vec *bvec;
2742 struct bvec_iter_all iter_all;
2743 bool first_bvec = true;
2745 ASSERT(!bio_flagged(bio, BIO_CLONED));
2746 bio_for_each_segment_all(bvec, bio, iter_all) {
2747 struct page *page = bvec->bv_page;
2748 struct inode *inode = page->mapping->host;
2749 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2751 /* We always issue full-page reads, but if some block
2752 * in a page fails to read, blk_update_request() will
2753 * advance bv_offset and adjust bv_len to compensate.
2754 * Print a warning for nonzero offsets, and an error
2755 * if they don't add up to a full page. */
2756 if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) {
2757 if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE)
2759 "partial page write in btrfs with offset %u and length %u",
2760 bvec->bv_offset, bvec->bv_len);
2763 "incomplete page write in btrfs with offset %u and length %u",
2764 bvec->bv_offset, bvec->bv_len);
2767 start = page_offset(page);
2768 end = start + bvec->bv_offset + bvec->bv_len - 1;
2771 btrfs_record_physical_zoned(inode, start, bio);
2775 end_extent_writepage(page, error, start, end);
2776 end_page_writeback(page);
2783 * Record previously processed extent range
2785 * For endio_readpage_release_extent() to handle a full extent range, reducing
2786 * the extent io operations.
2788 struct processed_extent {
2789 struct btrfs_inode *inode;
2790 /* Start of the range in @inode */
2792 /* End of the range in @inode */
2798 * Try to release processed extent range
2800 * May not release the extent range right now if the current range is
2801 * contiguous to processed extent.
2803 * Will release processed extent when any of @inode, @uptodate, the range is
2804 * no longer contiguous to the processed range.
2806 * Passing @inode == NULL will force processed extent to be released.
2808 static void endio_readpage_release_extent(struct processed_extent *processed,
2809 struct btrfs_inode *inode, u64 start, u64 end,
2812 struct extent_state *cached = NULL;
2813 struct extent_io_tree *tree;
2815 /* The first extent, initialize @processed */
2816 if (!processed->inode)
2820 * Contiguous to processed extent, just uptodate the end.
2822 * Several things to notice:
2824 * - bio can be merged as long as on-disk bytenr is contiguous
2825 * This means we can have page belonging to other inodes, thus need to
2826 * check if the inode still matches.
2827 * - bvec can contain range beyond current page for multi-page bvec
2828 * Thus we need to do processed->end + 1 >= start check
2830 if (processed->inode == inode && processed->uptodate == uptodate &&
2831 processed->end + 1 >= start && end >= processed->end) {
2832 processed->end = end;
2836 tree = &processed->inode->io_tree;
2838 * Now we don't have range contiguous to the processed range, release
2839 * the processed range now.
2841 if (processed->uptodate && tree->track_uptodate)
2842 set_extent_uptodate(tree, processed->start, processed->end,
2843 &cached, GFP_ATOMIC);
2844 unlock_extent_cached_atomic(tree, processed->start, processed->end,
2848 /* Update processed to current range */
2849 processed->inode = inode;
2850 processed->start = start;
2851 processed->end = end;
2852 processed->uptodate = uptodate;
2855 static void begin_page_read(struct btrfs_fs_info *fs_info, struct page *page)
2857 ASSERT(PageLocked(page));
2858 if (fs_info->sectorsize == PAGE_SIZE)
2861 ASSERT(PagePrivate(page));
2862 btrfs_subpage_start_reader(fs_info, page, page_offset(page), PAGE_SIZE);
2865 static void end_page_read(struct page *page, bool uptodate, u64 start, u32 len)
2867 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
2869 ASSERT(page_offset(page) <= start &&
2870 start + len <= page_offset(page) + PAGE_SIZE);
2873 btrfs_page_set_uptodate(fs_info, page, start, len);
2875 btrfs_page_clear_uptodate(fs_info, page, start, len);
2876 btrfs_page_set_error(fs_info, page, start, len);
2879 if (fs_info->sectorsize == PAGE_SIZE)
2881 else if (is_data_inode(page->mapping->host))
2883 * For subpage data, unlock the page if we're the last reader.
2884 * For subpage metadata, page lock is not utilized for read.
2886 btrfs_subpage_end_reader(fs_info, page, start, len);
2890 * Find extent buffer for a givne bytenr.
2892 * This is for end_bio_extent_readpage(), thus we can't do any unsafe locking
2895 static struct extent_buffer *find_extent_buffer_readpage(
2896 struct btrfs_fs_info *fs_info, struct page *page, u64 bytenr)
2898 struct extent_buffer *eb;
2901 * For regular sectorsize, we can use page->private to grab extent
2904 if (fs_info->sectorsize == PAGE_SIZE) {
2905 ASSERT(PagePrivate(page) && page->private);
2906 return (struct extent_buffer *)page->private;
2909 /* For subpage case, we need to lookup buffer radix tree */
2911 eb = radix_tree_lookup(&fs_info->buffer_radix,
2912 bytenr >> fs_info->sectorsize_bits);
2919 * after a readpage IO is done, we need to:
2920 * clear the uptodate bits on error
2921 * set the uptodate bits if things worked
2922 * set the page up to date if all extents in the tree are uptodate
2923 * clear the lock bit in the extent tree
2924 * unlock the page if there are no other extents locked for it
2926 * Scheduling is not allowed, so the extent state tree is expected
2927 * to have one and only one object corresponding to this IO.
2929 static void end_bio_extent_readpage(struct bio *bio)
2931 struct bio_vec *bvec;
2932 int uptodate = !bio->bi_status;
2933 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
2934 struct extent_io_tree *tree, *failure_tree;
2935 struct processed_extent processed = { 0 };
2937 * The offset to the beginning of a bio, since one bio can never be
2938 * larger than UINT_MAX, u32 here is enough.
2943 struct bvec_iter_all iter_all;
2945 ASSERT(!bio_flagged(bio, BIO_CLONED));
2946 bio_for_each_segment_all(bvec, bio, iter_all) {
2947 struct page *page = bvec->bv_page;
2948 struct inode *inode = page->mapping->host;
2949 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2950 const u32 sectorsize = fs_info->sectorsize;
2955 btrfs_debug(fs_info,
2956 "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u",
2957 bio->bi_iter.bi_sector, bio->bi_status,
2958 io_bio->mirror_num);
2959 tree = &BTRFS_I(inode)->io_tree;
2960 failure_tree = &BTRFS_I(inode)->io_failure_tree;
2963 * We always issue full-sector reads, but if some block in a
2964 * page fails to read, blk_update_request() will advance
2965 * bv_offset and adjust bv_len to compensate. Print a warning
2966 * for unaligned offsets, and an error if they don't add up to
2969 if (!IS_ALIGNED(bvec->bv_offset, sectorsize))
2971 "partial page read in btrfs with offset %u and length %u",
2972 bvec->bv_offset, bvec->bv_len);
2973 else if (!IS_ALIGNED(bvec->bv_offset + bvec->bv_len,
2976 "incomplete page read with offset %u and length %u",
2977 bvec->bv_offset, bvec->bv_len);
2979 start = page_offset(page) + bvec->bv_offset;
2980 end = start + bvec->bv_len - 1;
2983 mirror = io_bio->mirror_num;
2984 if (likely(uptodate)) {
2985 if (is_data_inode(inode))
2986 ret = btrfs_verify_data_csum(io_bio,
2987 bio_offset, page, start, end);
2989 ret = btrfs_validate_metadata_buffer(io_bio,
2990 page, start, end, mirror);
2994 clean_io_failure(BTRFS_I(inode)->root->fs_info,
2995 failure_tree, tree, start,
2997 btrfs_ino(BTRFS_I(inode)), 0);
3000 if (likely(uptodate))
3003 if (is_data_inode(inode)) {
3006 * The generic bio_readpage_error handles errors the
3007 * following way: If possible, new read requests are
3008 * created and submitted and will end up in
3009 * end_bio_extent_readpage as well (if we're lucky,
3010 * not in the !uptodate case). In that case it returns
3011 * 0 and we just go on with the next page in our bio.
3012 * If it can't handle the error it will return -EIO and
3013 * we remain responsible for that page.
3015 if (!btrfs_submit_read_repair(inode, bio, bio_offset,
3017 start - page_offset(page),
3019 btrfs_submit_data_bio)) {
3020 uptodate = !bio->bi_status;
3021 ASSERT(bio_offset + len > bio_offset);
3026 struct extent_buffer *eb;
3028 eb = find_extent_buffer_readpage(fs_info, page, start);
3029 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
3030 eb->read_mirror = mirror;
3031 atomic_dec(&eb->io_pages);
3032 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD,
3034 btree_readahead_hook(eb, -EIO);
3037 if (likely(uptodate)) {
3038 loff_t i_size = i_size_read(inode);
3039 pgoff_t end_index = i_size >> PAGE_SHIFT;
3042 * Zero out the remaining part if this range straddles
3045 * Here we should only zero the range inside the bvec,
3046 * not touch anything else.
3048 * NOTE: i_size is exclusive while end is inclusive.
3050 if (page->index == end_index && i_size <= end) {
3051 u32 zero_start = max(offset_in_page(i_size),
3052 offset_in_page(start));
3054 zero_user_segment(page, zero_start,
3055 offset_in_page(end) + 1);
3058 ASSERT(bio_offset + len > bio_offset);
3061 /* Update page status and unlock */
3062 end_page_read(page, uptodate, start, len);
3063 endio_readpage_release_extent(&processed, BTRFS_I(inode),
3064 start, end, uptodate);
3066 /* Release the last extent */
3067 endio_readpage_release_extent(&processed, NULL, 0, 0, false);
3068 btrfs_io_bio_free_csum(io_bio);
3073 * Initialize the members up to but not including 'bio'. Use after allocating a
3074 * new bio by bio_alloc_bioset as it does not initialize the bytes outside of
3075 * 'bio' because use of __GFP_ZERO is not supported.
3077 static inline void btrfs_io_bio_init(struct btrfs_io_bio *btrfs_bio)
3079 memset(btrfs_bio, 0, offsetof(struct btrfs_io_bio, bio));
3083 * The following helpers allocate a bio. As it's backed by a bioset, it'll
3084 * never fail. We're returning a bio right now but you can call btrfs_io_bio
3085 * for the appropriate container_of magic
3087 struct bio *btrfs_bio_alloc(u64 first_byte)
3091 bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_VECS, &btrfs_bioset);
3092 bio->bi_iter.bi_sector = first_byte >> 9;
3093 btrfs_io_bio_init(btrfs_io_bio(bio));
3097 struct bio *btrfs_bio_clone(struct bio *bio)
3099 struct btrfs_io_bio *btrfs_bio;
3102 /* Bio allocation backed by a bioset does not fail */
3103 new = bio_clone_fast(bio, GFP_NOFS, &btrfs_bioset);
3104 btrfs_bio = btrfs_io_bio(new);
3105 btrfs_io_bio_init(btrfs_bio);
3106 btrfs_bio->iter = bio->bi_iter;
3110 struct bio *btrfs_io_bio_alloc(unsigned int nr_iovecs)
3114 /* Bio allocation backed by a bioset does not fail */
3115 bio = bio_alloc_bioset(GFP_NOFS, nr_iovecs, &btrfs_bioset);
3116 btrfs_io_bio_init(btrfs_io_bio(bio));
3120 struct bio *btrfs_bio_clone_partial(struct bio *orig, int offset, int size)
3123 struct btrfs_io_bio *btrfs_bio;
3125 /* this will never fail when it's backed by a bioset */
3126 bio = bio_clone_fast(orig, GFP_NOFS, &btrfs_bioset);
3129 btrfs_bio = btrfs_io_bio(bio);
3130 btrfs_io_bio_init(btrfs_bio);
3132 bio_trim(bio, offset >> 9, size >> 9);
3133 btrfs_bio->iter = bio->bi_iter;
3138 * Attempt to add a page to bio
3140 * @bio: destination bio
3141 * @page: page to add to the bio
3142 * @disk_bytenr: offset of the new bio or to check whether we are adding
3143 * a contiguous page to the previous one
3144 * @pg_offset: starting offset in the page
3145 * @size: portion of page that we want to write
3146 * @prev_bio_flags: flags of previous bio to see if we can merge the current one
3147 * @bio_flags: flags of the current bio to see if we can merge them
3148 * @return: true if page was added, false otherwise
3150 * Attempt to add a page to bio considering stripe alignment etc.
3152 * Return true if successfully page added. Otherwise, return false.
3154 static bool btrfs_bio_add_page(struct bio *bio, struct page *page,
3155 u64 disk_bytenr, unsigned int size,
3156 unsigned int pg_offset,
3157 unsigned long prev_bio_flags,
3158 unsigned long bio_flags)
3160 const sector_t sector = disk_bytenr >> SECTOR_SHIFT;
3164 if (prev_bio_flags != bio_flags)
3167 if (prev_bio_flags & EXTENT_BIO_COMPRESSED)
3168 contig = bio->bi_iter.bi_sector == sector;
3170 contig = bio_end_sector(bio) == sector;
3174 if (btrfs_bio_fits_in_stripe(page, size, bio, bio_flags))
3177 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
3178 struct page *first_page = bio_first_bvec_all(bio)->bv_page;
3180 if (!btrfs_bio_fits_in_ordered_extent(first_page, bio, size))
3182 ret = bio_add_zone_append_page(bio, page, size, pg_offset);
3184 ret = bio_add_page(bio, page, size, pg_offset);
3191 * @opf: bio REQ_OP_* and REQ_* flags as one value
3192 * @wbc: optional writeback control for io accounting
3193 * @page: page to add to the bio
3194 * @disk_bytenr: logical bytenr where the write will be
3195 * @size: portion of page that we want to write to
3196 * @pg_offset: offset of the new bio or to check whether we are adding
3197 * a contiguous page to the previous one
3198 * @bio_ret: must be valid pointer, newly allocated bio will be stored there
3199 * @end_io_func: end_io callback for new bio
3200 * @mirror_num: desired mirror to read/write
3201 * @prev_bio_flags: flags of previous bio to see if we can merge the current one
3202 * @bio_flags: flags of the current bio to see if we can merge them
3204 static int submit_extent_page(unsigned int opf,
3205 struct writeback_control *wbc,
3206 struct page *page, u64 disk_bytenr,
3207 size_t size, unsigned long pg_offset,
3208 struct bio **bio_ret,
3209 bio_end_io_t end_io_func,
3211 unsigned long prev_bio_flags,
3212 unsigned long bio_flags,
3213 bool force_bio_submit)
3217 size_t io_size = min_t(size_t, size, PAGE_SIZE);
3218 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
3219 struct extent_io_tree *tree = &inode->io_tree;
3220 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3226 if (force_bio_submit ||
3227 !btrfs_bio_add_page(bio, page, disk_bytenr, io_size,
3228 pg_offset, prev_bio_flags, bio_flags)) {
3229 ret = submit_one_bio(bio, mirror_num, prev_bio_flags);
3237 wbc_account_cgroup_owner(wbc, page, io_size);
3242 bio = btrfs_bio_alloc(disk_bytenr);
3243 bio_add_page(bio, page, io_size, pg_offset);
3244 bio->bi_end_io = end_io_func;
3245 bio->bi_private = tree;
3246 bio->bi_write_hint = page->mapping->host->i_write_hint;
3249 struct block_device *bdev;
3251 bdev = fs_info->fs_devices->latest_bdev;
3252 bio_set_dev(bio, bdev);
3253 wbc_init_bio(wbc, bio);
3254 wbc_account_cgroup_owner(wbc, page, io_size);
3256 if (btrfs_is_zoned(fs_info) && bio_op(bio) == REQ_OP_ZONE_APPEND) {
3257 struct extent_map *em;
3258 struct map_lookup *map;
3260 em = btrfs_get_chunk_map(fs_info, disk_bytenr, io_size);
3264 map = em->map_lookup;
3265 /* We only support single profile for now */
3266 ASSERT(map->num_stripes == 1);
3267 btrfs_io_bio(bio)->device = map->stripes[0].dev;
3269 free_extent_map(em);
3277 static int attach_extent_buffer_page(struct extent_buffer *eb,
3279 struct btrfs_subpage *prealloc)
3281 struct btrfs_fs_info *fs_info = eb->fs_info;
3285 * If the page is mapped to btree inode, we should hold the private
3286 * lock to prevent race.
3287 * For cloned or dummy extent buffers, their pages are not mapped and
3288 * will not race with any other ebs.
3291 lockdep_assert_held(&page->mapping->private_lock);
3293 if (fs_info->sectorsize == PAGE_SIZE) {
3294 if (!PagePrivate(page))
3295 attach_page_private(page, eb);
3297 WARN_ON(page->private != (unsigned long)eb);
3301 /* Already mapped, just free prealloc */
3302 if (PagePrivate(page)) {
3303 btrfs_free_subpage(prealloc);
3308 /* Has preallocated memory for subpage */
3309 attach_page_private(page, prealloc);
3311 /* Do new allocation to attach subpage */
3312 ret = btrfs_attach_subpage(fs_info, page,
3313 BTRFS_SUBPAGE_METADATA);
3317 int set_page_extent_mapped(struct page *page)
3319 struct btrfs_fs_info *fs_info;
3321 ASSERT(page->mapping);
3323 if (PagePrivate(page))
3326 fs_info = btrfs_sb(page->mapping->host->i_sb);
3328 if (fs_info->sectorsize < PAGE_SIZE)
3329 return btrfs_attach_subpage(fs_info, page, BTRFS_SUBPAGE_DATA);
3331 attach_page_private(page, (void *)EXTENT_PAGE_PRIVATE);
3335 void clear_page_extent_mapped(struct page *page)
3337 struct btrfs_fs_info *fs_info;
3339 ASSERT(page->mapping);
3341 if (!PagePrivate(page))
3344 fs_info = btrfs_sb(page->mapping->host->i_sb);
3345 if (fs_info->sectorsize < PAGE_SIZE)
3346 return btrfs_detach_subpage(fs_info, page);
3348 detach_page_private(page);
3351 static struct extent_map *
3352 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
3353 u64 start, u64 len, struct extent_map **em_cached)
3355 struct extent_map *em;
3357 if (em_cached && *em_cached) {
3359 if (extent_map_in_tree(em) && start >= em->start &&
3360 start < extent_map_end(em)) {
3361 refcount_inc(&em->refs);
3365 free_extent_map(em);
3369 em = btrfs_get_extent(BTRFS_I(inode), page, pg_offset, start, len);
3370 if (em_cached && !IS_ERR_OR_NULL(em)) {
3372 refcount_inc(&em->refs);
3378 * basic readpage implementation. Locked extent state structs are inserted
3379 * into the tree that are removed when the IO is done (by the end_io
3381 * XXX JDM: This needs looking at to ensure proper page locking
3382 * return 0 on success, otherwise return error
3384 int btrfs_do_readpage(struct page *page, struct extent_map **em_cached,
3385 struct bio **bio, unsigned long *bio_flags,
3386 unsigned int read_flags, u64 *prev_em_start)
3388 struct inode *inode = page->mapping->host;
3389 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3390 u64 start = page_offset(page);
3391 const u64 end = start + PAGE_SIZE - 1;
3394 u64 last_byte = i_size_read(inode);
3397 struct extent_map *em;
3400 size_t pg_offset = 0;
3402 size_t blocksize = inode->i_sb->s_blocksize;
3403 unsigned long this_bio_flag = 0;
3404 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
3406 ret = set_page_extent_mapped(page);
3408 unlock_extent(tree, start, end);
3409 btrfs_page_set_error(fs_info, page, start, PAGE_SIZE);
3414 if (!PageUptodate(page)) {
3415 if (cleancache_get_page(page) == 0) {
3416 BUG_ON(blocksize != PAGE_SIZE);
3417 unlock_extent(tree, start, end);
3423 if (page->index == last_byte >> PAGE_SHIFT) {
3424 size_t zero_offset = offset_in_page(last_byte);
3427 iosize = PAGE_SIZE - zero_offset;
3428 memzero_page(page, zero_offset, iosize);
3429 flush_dcache_page(page);
3432 begin_page_read(fs_info, page);
3433 while (cur <= end) {
3434 bool force_bio_submit = false;
3437 if (cur >= last_byte) {
3438 struct extent_state *cached = NULL;
3440 iosize = PAGE_SIZE - pg_offset;
3441 memzero_page(page, pg_offset, iosize);
3442 flush_dcache_page(page);
3443 set_extent_uptodate(tree, cur, cur + iosize - 1,
3445 unlock_extent_cached(tree, cur,
3446 cur + iosize - 1, &cached);
3447 end_page_read(page, true, cur, iosize);
3450 em = __get_extent_map(inode, page, pg_offset, cur,
3451 end - cur + 1, em_cached);
3452 if (IS_ERR_OR_NULL(em)) {
3453 unlock_extent(tree, cur, end);
3454 end_page_read(page, false, cur, end + 1 - cur);
3457 extent_offset = cur - em->start;
3458 BUG_ON(extent_map_end(em) <= cur);
3461 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
3462 this_bio_flag |= EXTENT_BIO_COMPRESSED;
3463 extent_set_compress_type(&this_bio_flag,
3467 iosize = min(extent_map_end(em) - cur, end - cur + 1);
3468 cur_end = min(extent_map_end(em) - 1, end);
3469 iosize = ALIGN(iosize, blocksize);
3470 if (this_bio_flag & EXTENT_BIO_COMPRESSED)
3471 disk_bytenr = em->block_start;
3473 disk_bytenr = em->block_start + extent_offset;
3474 block_start = em->block_start;
3475 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
3476 block_start = EXTENT_MAP_HOLE;
3479 * If we have a file range that points to a compressed extent
3480 * and it's followed by a consecutive file range that points
3481 * to the same compressed extent (possibly with a different
3482 * offset and/or length, so it either points to the whole extent
3483 * or only part of it), we must make sure we do not submit a
3484 * single bio to populate the pages for the 2 ranges because
3485 * this makes the compressed extent read zero out the pages
3486 * belonging to the 2nd range. Imagine the following scenario:
3489 * [0 - 8K] [8K - 24K]
3492 * points to extent X, points to extent X,
3493 * offset 4K, length of 8K offset 0, length 16K
3495 * [extent X, compressed length = 4K uncompressed length = 16K]
3497 * If the bio to read the compressed extent covers both ranges,
3498 * it will decompress extent X into the pages belonging to the
3499 * first range and then it will stop, zeroing out the remaining
3500 * pages that belong to the other range that points to extent X.
3501 * So here we make sure we submit 2 bios, one for the first
3502 * range and another one for the third range. Both will target
3503 * the same physical extent from disk, but we can't currently
3504 * make the compressed bio endio callback populate the pages
3505 * for both ranges because each compressed bio is tightly
3506 * coupled with a single extent map, and each range can have
3507 * an extent map with a different offset value relative to the
3508 * uncompressed data of our extent and different lengths. This
3509 * is a corner case so we prioritize correctness over
3510 * non-optimal behavior (submitting 2 bios for the same extent).
3512 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) &&
3513 prev_em_start && *prev_em_start != (u64)-1 &&
3514 *prev_em_start != em->start)
3515 force_bio_submit = true;
3518 *prev_em_start = em->start;
3520 free_extent_map(em);
3523 /* we've found a hole, just zero and go on */
3524 if (block_start == EXTENT_MAP_HOLE) {
3525 struct extent_state *cached = NULL;
3527 memzero_page(page, pg_offset, iosize);
3528 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);
3536 pg_offset += iosize;
3539 /* the get_extent function already copied into the page */
3540 if (test_range_bit(tree, cur, cur_end,
3541 EXTENT_UPTODATE, 1, NULL)) {
3542 check_page_uptodate(tree, page);
3543 unlock_extent(tree, cur, cur + iosize - 1);
3544 end_page_read(page, true, cur, iosize);
3546 pg_offset += iosize;
3549 /* we have an inline extent but it didn't get marked up
3550 * to date. Error out
3552 if (block_start == EXTENT_MAP_INLINE) {
3553 unlock_extent(tree, cur, cur + iosize - 1);
3554 end_page_read(page, false, cur, iosize);
3556 pg_offset += iosize;
3560 ret = submit_extent_page(REQ_OP_READ | read_flags, NULL,
3561 page, disk_bytenr, iosize,
3563 end_bio_extent_readpage, 0,
3569 *bio_flags = this_bio_flag;
3571 unlock_extent(tree, cur, cur + iosize - 1);
3572 end_page_read(page, false, cur, iosize);
3576 pg_offset += iosize;
3582 static inline void contiguous_readpages(struct page *pages[], int nr_pages,
3584 struct extent_map **em_cached,
3586 unsigned long *bio_flags,
3589 struct btrfs_inode *inode = BTRFS_I(pages[0]->mapping->host);
3592 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
3594 for (index = 0; index < nr_pages; index++) {
3595 btrfs_do_readpage(pages[index], em_cached, bio, bio_flags,
3596 REQ_RAHEAD, prev_em_start);
3597 put_page(pages[index]);
3601 static void update_nr_written(struct writeback_control *wbc,
3602 unsigned long nr_written)
3604 wbc->nr_to_write -= nr_written;
3608 * helper for __extent_writepage, doing all of the delayed allocation setup.
3610 * This returns 1 if btrfs_run_delalloc_range function did all the work required
3611 * to write the page (copy into inline extent). In this case the IO has
3612 * been started and the page is already unlocked.
3614 * This returns 0 if all went well (page still locked)
3615 * This returns < 0 if there were errors (page still locked)
3617 static noinline_for_stack int writepage_delalloc(struct btrfs_inode *inode,
3618 struct page *page, struct writeback_control *wbc,
3619 u64 delalloc_start, unsigned long *nr_written)
3621 u64 page_end = delalloc_start + PAGE_SIZE - 1;
3623 u64 delalloc_to_write = 0;
3624 u64 delalloc_end = 0;
3626 int page_started = 0;
3629 while (delalloc_end < page_end) {
3630 found = find_lock_delalloc_range(&inode->vfs_inode, page,
3634 delalloc_start = delalloc_end + 1;
3637 ret = btrfs_run_delalloc_range(inode, page, delalloc_start,
3638 delalloc_end, &page_started, nr_written, wbc);
3642 * btrfs_run_delalloc_range should return < 0 for error
3643 * but just in case, we use > 0 here meaning the IO is
3644 * started, so we don't want to return > 0 unless
3645 * things are going well.
3647 return ret < 0 ? ret : -EIO;
3650 * delalloc_end is already one less than the total length, so
3651 * we don't subtract one from PAGE_SIZE
3653 delalloc_to_write += (delalloc_end - delalloc_start +
3654 PAGE_SIZE) >> PAGE_SHIFT;
3655 delalloc_start = delalloc_end + 1;
3657 if (wbc->nr_to_write < delalloc_to_write) {
3660 if (delalloc_to_write < thresh * 2)
3661 thresh = delalloc_to_write;
3662 wbc->nr_to_write = min_t(u64, delalloc_to_write,
3666 /* did the fill delalloc function already unlock and start
3671 * we've unlocked the page, so we can't update
3672 * the mapping's writeback index, just update
3675 wbc->nr_to_write -= *nr_written;
3683 * helper for __extent_writepage. This calls the writepage start hooks,
3684 * and does the loop to map the page into extents and bios.
3686 * We return 1 if the IO is started and the page is unlocked,
3687 * 0 if all went well (page still locked)
3688 * < 0 if there were errors (page still locked)
3690 static noinline_for_stack int __extent_writepage_io(struct btrfs_inode *inode,
3692 struct writeback_control *wbc,
3693 struct extent_page_data *epd,
3695 unsigned long nr_written,
3698 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3699 struct extent_io_tree *tree = &inode->io_tree;
3700 u64 start = page_offset(page);
3701 u64 end = start + PAGE_SIZE - 1;
3705 struct extent_map *em;
3708 u32 opf = REQ_OP_WRITE;
3709 const unsigned int write_flags = wbc_to_write_flags(wbc);
3712 ret = btrfs_writepage_cow_fixup(page, start, end);
3714 /* Fixup worker will requeue */
3715 redirty_page_for_writepage(wbc, page);
3716 update_nr_written(wbc, nr_written);
3722 * we don't want to touch the inode after unlocking the page,
3723 * so we update the mapping writeback index now
3725 update_nr_written(wbc, nr_written + 1);
3727 while (cur <= end) {
3732 if (cur >= i_size) {
3733 btrfs_writepage_endio_finish_ordered(page, cur, end, 1);
3736 em = btrfs_get_extent(inode, NULL, 0, cur, end - cur + 1);
3737 if (IS_ERR_OR_NULL(em)) {
3739 ret = PTR_ERR_OR_ZERO(em);
3743 extent_offset = cur - em->start;
3744 em_end = extent_map_end(em);
3745 ASSERT(cur <= em_end);
3747 ASSERT(IS_ALIGNED(em->start, fs_info->sectorsize));
3748 ASSERT(IS_ALIGNED(em->len, fs_info->sectorsize));
3749 block_start = em->block_start;
3750 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3751 disk_bytenr = em->block_start + extent_offset;
3753 /* Note that em_end from extent_map_end() is exclusive */
3754 iosize = min(em_end, end + 1) - cur;
3756 if (btrfs_use_zone_append(inode, em->block_start))
3757 opf = REQ_OP_ZONE_APPEND;
3759 free_extent_map(em);
3763 * compressed and inline extents are written through other
3766 if (compressed || block_start == EXTENT_MAP_HOLE ||
3767 block_start == EXTENT_MAP_INLINE) {
3771 btrfs_writepage_endio_finish_ordered(page, cur,
3772 cur + iosize - 1, 1);
3777 btrfs_set_range_writeback(tree, cur, cur + iosize - 1);
3778 if (!PageWriteback(page)) {
3779 btrfs_err(inode->root->fs_info,
3780 "page %lu not writeback, cur %llu end %llu",
3781 page->index, cur, end);
3784 ret = submit_extent_page(opf | write_flags, wbc, page,
3785 disk_bytenr, iosize,
3786 cur - page_offset(page), &epd->bio,
3787 end_bio_extent_writepage,
3791 if (PageWriteback(page))
3792 end_page_writeback(page);
3803 * the writepage semantics are similar to regular writepage. extent
3804 * records are inserted to lock ranges in the tree, and as dirty areas
3805 * are found, they are marked writeback. Then the lock bits are removed
3806 * and the end_io handler clears the writeback ranges
3808 * Return 0 if everything goes well.
3809 * Return <0 for error.
3811 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
3812 struct extent_page_data *epd)
3814 struct inode *inode = page->mapping->host;
3815 u64 start = page_offset(page);
3816 u64 page_end = start + PAGE_SIZE - 1;
3820 loff_t i_size = i_size_read(inode);
3821 unsigned long end_index = i_size >> PAGE_SHIFT;
3822 unsigned long nr_written = 0;
3824 trace___extent_writepage(page, inode, wbc);
3826 WARN_ON(!PageLocked(page));
3828 ClearPageError(page);
3830 pg_offset = offset_in_page(i_size);
3831 if (page->index > end_index ||
3832 (page->index == end_index && !pg_offset)) {
3833 page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
3838 if (page->index == end_index) {
3839 memzero_page(page, pg_offset, PAGE_SIZE - pg_offset);
3840 flush_dcache_page(page);
3843 ret = set_page_extent_mapped(page);
3849 if (!epd->extent_locked) {
3850 ret = writepage_delalloc(BTRFS_I(inode), page, wbc, start,
3858 ret = __extent_writepage_io(BTRFS_I(inode), page, wbc, epd, i_size,
3865 /* make sure the mapping tag for page dirty gets cleared */
3866 set_page_writeback(page);
3867 end_page_writeback(page);
3869 if (PageError(page)) {
3870 ret = ret < 0 ? ret : -EIO;
3871 end_extent_writepage(page, ret, start, page_end);
3878 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3880 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
3881 TASK_UNINTERRUPTIBLE);
3884 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3886 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3887 smp_mb__after_atomic();
3888 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3892 * Lock extent buffer status and pages for writeback.
3894 * May try to flush write bio if we can't get the lock.
3896 * Return 0 if the extent buffer doesn't need to be submitted.
3897 * (E.g. the extent buffer is not dirty)
3898 * Return >0 is the extent buffer is submitted to bio.
3899 * Return <0 if something went wrong, no page is locked.
3901 static noinline_for_stack int lock_extent_buffer_for_io(struct extent_buffer *eb,
3902 struct extent_page_data *epd)
3904 struct btrfs_fs_info *fs_info = eb->fs_info;
3905 int i, num_pages, failed_page_nr;
3909 if (!btrfs_try_tree_write_lock(eb)) {
3910 ret = flush_write_bio(epd);
3914 btrfs_tree_lock(eb);
3917 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3918 btrfs_tree_unlock(eb);
3922 ret = flush_write_bio(epd);
3928 wait_on_extent_buffer_writeback(eb);
3929 btrfs_tree_lock(eb);
3930 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3932 btrfs_tree_unlock(eb);
3937 * We need to do this to prevent races in people who check if the eb is
3938 * under IO since we can end up having no IO bits set for a short period
3941 spin_lock(&eb->refs_lock);
3942 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3943 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3944 spin_unlock(&eb->refs_lock);
3945 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3946 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
3948 fs_info->dirty_metadata_batch);
3951 spin_unlock(&eb->refs_lock);
3954 btrfs_tree_unlock(eb);
3957 * Either we don't need to submit any tree block, or we're submitting
3959 * Subpage metadata doesn't use page locking at all, so we can skip
3962 if (!ret || fs_info->sectorsize < PAGE_SIZE)
3965 num_pages = num_extent_pages(eb);
3966 for (i = 0; i < num_pages; i++) {
3967 struct page *p = eb->pages[i];
3969 if (!trylock_page(p)) {
3973 err = flush_write_bio(epd);
3987 /* Unlock already locked pages */
3988 for (i = 0; i < failed_page_nr; i++)
3989 unlock_page(eb->pages[i]);
3991 * Clear EXTENT_BUFFER_WRITEBACK and wake up anyone waiting on it.
3992 * Also set back EXTENT_BUFFER_DIRTY so future attempts to this eb can
3993 * be made and undo everything done before.
3995 btrfs_tree_lock(eb);
3996 spin_lock(&eb->refs_lock);
3997 set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
3998 end_extent_buffer_writeback(eb);
3999 spin_unlock(&eb->refs_lock);
4000 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, eb->len,
4001 fs_info->dirty_metadata_batch);
4002 btrfs_clear_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
4003 btrfs_tree_unlock(eb);
4007 static void set_btree_ioerr(struct page *page, struct extent_buffer *eb)
4009 struct btrfs_fs_info *fs_info = eb->fs_info;
4011 btrfs_page_set_error(fs_info, page, eb->start, eb->len);
4012 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))
4016 * If we error out, we should add back the dirty_metadata_bytes
4017 * to make it consistent.
4019 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4020 eb->len, fs_info->dirty_metadata_batch);
4023 * If writeback for a btree extent that doesn't belong to a log tree
4024 * failed, increment the counter transaction->eb_write_errors.
4025 * We do this because while the transaction is running and before it's
4026 * committing (when we call filemap_fdata[write|wait]_range against
4027 * the btree inode), we might have
4028 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
4029 * returns an error or an error happens during writeback, when we're
4030 * committing the transaction we wouldn't know about it, since the pages
4031 * can be no longer dirty nor marked anymore for writeback (if a
4032 * subsequent modification to the extent buffer didn't happen before the
4033 * transaction commit), which makes filemap_fdata[write|wait]_range not
4034 * able to find the pages tagged with SetPageError at transaction
4035 * commit time. So if this happens we must abort the transaction,
4036 * otherwise we commit a super block with btree roots that point to
4037 * btree nodes/leafs whose content on disk is invalid - either garbage
4038 * or the content of some node/leaf from a past generation that got
4039 * cowed or deleted and is no longer valid.
4041 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
4042 * not be enough - we need to distinguish between log tree extents vs
4043 * non-log tree extents, and the next filemap_fdatawait_range() call
4044 * will catch and clear such errors in the mapping - and that call might
4045 * be from a log sync and not from a transaction commit. Also, checking
4046 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
4047 * not done and would not be reliable - the eb might have been released
4048 * from memory and reading it back again means that flag would not be
4049 * set (since it's a runtime flag, not persisted on disk).
4051 * Using the flags below in the btree inode also makes us achieve the
4052 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
4053 * writeback for all dirty pages and before filemap_fdatawait_range()
4054 * is called, the writeback for all dirty pages had already finished
4055 * with errors - because we were not using AS_EIO/AS_ENOSPC,
4056 * filemap_fdatawait_range() would return success, as it could not know
4057 * that writeback errors happened (the pages were no longer tagged for
4060 switch (eb->log_index) {
4062 set_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags);
4065 set_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags);
4068 set_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags);
4071 BUG(); /* unexpected, logic error */
4076 * The endio specific version which won't touch any unsafe spinlock in endio
4079 static struct extent_buffer *find_extent_buffer_nolock(
4080 struct btrfs_fs_info *fs_info, u64 start)
4082 struct extent_buffer *eb;
4085 eb = radix_tree_lookup(&fs_info->buffer_radix,
4086 start >> fs_info->sectorsize_bits);
4087 if (eb && atomic_inc_not_zero(&eb->refs)) {
4096 * The endio function for subpage extent buffer write.
4098 * Unlike end_bio_extent_buffer_writepage(), we only call end_page_writeback()
4099 * after all extent buffers in the page has finished their writeback.
4101 static void end_bio_subpage_eb_writepage(struct btrfs_fs_info *fs_info,
4104 struct bio_vec *bvec;
4105 struct bvec_iter_all iter_all;
4107 ASSERT(!bio_flagged(bio, BIO_CLONED));
4108 bio_for_each_segment_all(bvec, bio, iter_all) {
4109 struct page *page = bvec->bv_page;
4110 u64 bvec_start = page_offset(page) + bvec->bv_offset;
4111 u64 bvec_end = bvec_start + bvec->bv_len - 1;
4112 u64 cur_bytenr = bvec_start;
4114 ASSERT(IS_ALIGNED(bvec->bv_len, fs_info->nodesize));
4116 /* Iterate through all extent buffers in the range */
4117 while (cur_bytenr <= bvec_end) {
4118 struct extent_buffer *eb;
4122 * Here we can't use find_extent_buffer(), as it may
4123 * try to lock eb->refs_lock, which is not safe in endio
4126 eb = find_extent_buffer_nolock(fs_info, cur_bytenr);
4129 cur_bytenr = eb->start + eb->len;
4131 ASSERT(test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags));
4132 done = atomic_dec_and_test(&eb->io_pages);
4135 if (bio->bi_status ||
4136 test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) {
4137 ClearPageUptodate(page);
4138 set_btree_ioerr(page, eb);
4141 btrfs_subpage_clear_writeback(fs_info, page, eb->start,
4143 end_extent_buffer_writeback(eb);
4145 * free_extent_buffer() will grab spinlock which is not
4146 * safe in endio context. Thus here we manually dec
4149 atomic_dec(&eb->refs);
4155 static void end_bio_extent_buffer_writepage(struct bio *bio)
4157 struct btrfs_fs_info *fs_info;
4158 struct bio_vec *bvec;
4159 struct extent_buffer *eb;
4161 struct bvec_iter_all iter_all;
4163 fs_info = btrfs_sb(bio_first_page_all(bio)->mapping->host->i_sb);
4164 if (fs_info->sectorsize < PAGE_SIZE)
4165 return end_bio_subpage_eb_writepage(fs_info, bio);
4167 ASSERT(!bio_flagged(bio, BIO_CLONED));
4168 bio_for_each_segment_all(bvec, bio, iter_all) {
4169 struct page *page = bvec->bv_page;
4171 eb = (struct extent_buffer *)page->private;
4173 done = atomic_dec_and_test(&eb->io_pages);
4175 if (bio->bi_status ||
4176 test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) {
4177 ClearPageUptodate(page);
4178 set_btree_ioerr(page, eb);
4181 end_page_writeback(page);
4186 end_extent_buffer_writeback(eb);
4193 * Unlike the work in write_one_eb(), we rely completely on extent locking.
4194 * Page locking is only utilized at minimum to keep the VMM code happy.
4196 * Caller should still call write_one_eb() other than this function directly.
4197 * As write_one_eb() has extra preparation before submitting the extent buffer.
4199 static int write_one_subpage_eb(struct extent_buffer *eb,
4200 struct writeback_control *wbc,
4201 struct extent_page_data *epd)
4203 struct btrfs_fs_info *fs_info = eb->fs_info;
4204 struct page *page = eb->pages[0];
4205 unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META;
4206 bool no_dirty_ebs = false;
4209 /* clear_page_dirty_for_io() in subpage helper needs page locked */
4211 btrfs_subpage_set_writeback(fs_info, page, eb->start, eb->len);
4213 /* Check if this is the last dirty bit to update nr_written */
4214 no_dirty_ebs = btrfs_subpage_clear_and_test_dirty(fs_info, page,
4215 eb->start, eb->len);
4217 clear_page_dirty_for_io(page);
4219 ret = submit_extent_page(REQ_OP_WRITE | write_flags, wbc, page,
4220 eb->start, eb->len, eb->start - page_offset(page),
4221 &epd->bio, end_bio_extent_buffer_writepage, 0, 0, 0,
4224 btrfs_subpage_clear_writeback(fs_info, page, eb->start, eb->len);
4225 set_btree_ioerr(page, eb);
4228 if (atomic_dec_and_test(&eb->io_pages))
4229 end_extent_buffer_writeback(eb);
4234 * Submission finished without problem, if no range of the page is
4235 * dirty anymore, we have submitted a page. Update nr_written in wbc.
4238 update_nr_written(wbc, 1);
4242 static noinline_for_stack int write_one_eb(struct extent_buffer *eb,
4243 struct writeback_control *wbc,
4244 struct extent_page_data *epd)
4246 u64 disk_bytenr = eb->start;
4249 unsigned long start, end;
4250 unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META;
4253 clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
4254 num_pages = num_extent_pages(eb);
4255 atomic_set(&eb->io_pages, num_pages);
4257 /* set btree blocks beyond nritems with 0 to avoid stale content. */
4258 nritems = btrfs_header_nritems(eb);
4259 if (btrfs_header_level(eb) > 0) {
4260 end = btrfs_node_key_ptr_offset(nritems);
4262 memzero_extent_buffer(eb, end, eb->len - end);
4266 * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0
4268 start = btrfs_item_nr_offset(nritems);
4269 end = BTRFS_LEAF_DATA_OFFSET + leaf_data_end(eb);
4270 memzero_extent_buffer(eb, start, end - start);
4273 if (eb->fs_info->sectorsize < PAGE_SIZE)
4274 return write_one_subpage_eb(eb, wbc, epd);
4276 for (i = 0; i < num_pages; i++) {
4277 struct page *p = eb->pages[i];
4279 clear_page_dirty_for_io(p);
4280 set_page_writeback(p);
4281 ret = submit_extent_page(REQ_OP_WRITE | write_flags, wbc,
4282 p, disk_bytenr, PAGE_SIZE, 0,
4284 end_bio_extent_buffer_writepage,
4287 set_btree_ioerr(p, eb);
4288 if (PageWriteback(p))
4289 end_page_writeback(p);
4290 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
4291 end_extent_buffer_writeback(eb);
4295 disk_bytenr += PAGE_SIZE;
4296 update_nr_written(wbc, 1);
4300 if (unlikely(ret)) {
4301 for (; i < num_pages; i++) {
4302 struct page *p = eb->pages[i];
4303 clear_page_dirty_for_io(p);
4312 * Submit one subpage btree page.
4314 * The main difference to submit_eb_page() is:
4316 * For subpage, we don't rely on page locking at all.
4319 * We only flush bio if we may be unable to fit current extent buffers into
4322 * Return >=0 for the number of submitted extent buffers.
4323 * Return <0 for fatal error.
4325 static int submit_eb_subpage(struct page *page,
4326 struct writeback_control *wbc,
4327 struct extent_page_data *epd)
4329 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
4331 u64 page_start = page_offset(page);
4333 const int nbits = BTRFS_SUBPAGE_BITMAP_SIZE;
4334 int sectors_per_node = fs_info->nodesize >> fs_info->sectorsize_bits;
4337 /* Lock and write each dirty extent buffers in the range */
4338 while (bit_start < nbits) {
4339 struct btrfs_subpage *subpage = (struct btrfs_subpage *)page->private;
4340 struct extent_buffer *eb;
4341 unsigned long flags;
4345 * Take private lock to ensure the subpage won't be detached
4348 spin_lock(&page->mapping->private_lock);
4349 if (!PagePrivate(page)) {
4350 spin_unlock(&page->mapping->private_lock);
4353 spin_lock_irqsave(&subpage->lock, flags);
4354 if (!((1 << bit_start) & subpage->dirty_bitmap)) {
4355 spin_unlock_irqrestore(&subpage->lock, flags);
4356 spin_unlock(&page->mapping->private_lock);
4361 start = page_start + bit_start * fs_info->sectorsize;
4362 bit_start += sectors_per_node;
4365 * Here we just want to grab the eb without touching extra
4366 * spin locks, so call find_extent_buffer_nolock().
4368 eb = find_extent_buffer_nolock(fs_info, start);
4369 spin_unlock_irqrestore(&subpage->lock, flags);
4370 spin_unlock(&page->mapping->private_lock);
4373 * The eb has already reached 0 refs thus find_extent_buffer()
4374 * doesn't return it. We don't need to write back such eb
4380 ret = lock_extent_buffer_for_io(eb, epd);
4382 free_extent_buffer(eb);
4386 free_extent_buffer(eb);
4389 ret = write_one_eb(eb, wbc, epd);
4390 free_extent_buffer(eb);
4398 /* We hit error, end bio for the submitted extent buffers */
4399 end_write_bio(epd, ret);
4404 * Submit all page(s) of one extent buffer.
4406 * @page: the page of one extent buffer
4407 * @eb_context: to determine if we need to submit this page, if current page
4408 * belongs to this eb, we don't need to submit
4410 * The caller should pass each page in their bytenr order, and here we use
4411 * @eb_context to determine if we have submitted pages of one extent buffer.
4413 * If we have, we just skip until we hit a new page that doesn't belong to
4414 * current @eb_context.
4416 * If not, we submit all the page(s) of the extent buffer.
4418 * Return >0 if we have submitted the extent buffer successfully.
4419 * Return 0 if we don't need to submit the page, as it's already submitted by
4421 * Return <0 for fatal error.
4423 static int submit_eb_page(struct page *page, struct writeback_control *wbc,
4424 struct extent_page_data *epd,
4425 struct extent_buffer **eb_context)
4427 struct address_space *mapping = page->mapping;
4428 struct btrfs_block_group *cache = NULL;
4429 struct extent_buffer *eb;
4432 if (!PagePrivate(page))
4435 if (btrfs_sb(page->mapping->host->i_sb)->sectorsize < PAGE_SIZE)
4436 return submit_eb_subpage(page, wbc, epd);
4438 spin_lock(&mapping->private_lock);
4439 if (!PagePrivate(page)) {
4440 spin_unlock(&mapping->private_lock);
4444 eb = (struct extent_buffer *)page->private;
4447 * Shouldn't happen and normally this would be a BUG_ON but no point
4448 * crashing the machine for something we can survive anyway.
4451 spin_unlock(&mapping->private_lock);
4455 if (eb == *eb_context) {
4456 spin_unlock(&mapping->private_lock);
4459 ret = atomic_inc_not_zero(&eb->refs);
4460 spin_unlock(&mapping->private_lock);
4464 if (!btrfs_check_meta_write_pointer(eb->fs_info, eb, &cache)) {
4466 * If for_sync, this hole will be filled with
4467 * trasnsaction commit.
4469 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
4473 free_extent_buffer(eb);
4479 ret = lock_extent_buffer_for_io(eb, epd);
4481 btrfs_revert_meta_write_pointer(cache, eb);
4483 btrfs_put_block_group(cache);
4484 free_extent_buffer(eb);
4488 btrfs_put_block_group(cache);
4489 ret = write_one_eb(eb, wbc, epd);
4490 free_extent_buffer(eb);
4496 int btree_write_cache_pages(struct address_space *mapping,
4497 struct writeback_control *wbc)
4499 struct extent_buffer *eb_context = NULL;
4500 struct extent_page_data epd = {
4503 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4505 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
4508 int nr_to_write_done = 0;
4509 struct pagevec pvec;
4512 pgoff_t end; /* Inclusive */
4516 pagevec_init(&pvec);
4517 if (wbc->range_cyclic) {
4518 index = mapping->writeback_index; /* Start from prev offset */
4521 * Start from the beginning does not need to cycle over the
4522 * range, mark it as scanned.
4524 scanned = (index == 0);
4526 index = wbc->range_start >> PAGE_SHIFT;
4527 end = wbc->range_end >> PAGE_SHIFT;
4530 if (wbc->sync_mode == WB_SYNC_ALL)
4531 tag = PAGECACHE_TAG_TOWRITE;
4533 tag = PAGECACHE_TAG_DIRTY;
4534 btrfs_zoned_meta_io_lock(fs_info);
4536 if (wbc->sync_mode == WB_SYNC_ALL)
4537 tag_pages_for_writeback(mapping, index, end);
4538 while (!done && !nr_to_write_done && (index <= end) &&
4539 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
4543 for (i = 0; i < nr_pages; i++) {
4544 struct page *page = pvec.pages[i];
4546 ret = submit_eb_page(page, wbc, &epd, &eb_context);
4555 * the filesystem may choose to bump up nr_to_write.
4556 * We have to make sure to honor the new nr_to_write
4559 nr_to_write_done = wbc->nr_to_write <= 0;
4561 pagevec_release(&pvec);
4564 if (!scanned && !done) {
4566 * We hit the last page and there is more work to be done: wrap
4567 * back to the start of the file
4574 end_write_bio(&epd, ret);
4578 * If something went wrong, don't allow any metadata write bio to be
4581 * This would prevent use-after-free if we had dirty pages not
4582 * cleaned up, which can still happen by fuzzed images.
4585 * Allowing existing tree block to be allocated for other trees.
4587 * - Log tree operations
4588 * Exiting tree blocks get allocated to log tree, bumps its
4589 * generation, then get cleaned in tree re-balance.
4590 * Such tree block will not be written back, since it's clean,
4591 * thus no WRITTEN flag set.
4592 * And after log writes back, this tree block is not traced by
4593 * any dirty extent_io_tree.
4595 * - Offending tree block gets re-dirtied from its original owner
4596 * Since it has bumped generation, no WRITTEN flag, it can be
4597 * reused without COWing. This tree block will not be traced
4598 * by btrfs_transaction::dirty_pages.
4600 * Now such dirty tree block will not be cleaned by any dirty
4601 * extent io tree. Thus we don't want to submit such wild eb
4602 * if the fs already has error.
4604 if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
4605 ret = flush_write_bio(&epd);
4608 end_write_bio(&epd, ret);
4611 btrfs_zoned_meta_io_unlock(fs_info);
4616 * Walk the list of dirty pages of the given address space and write all of them.
4618 * @mapping: address space structure to write
4619 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
4620 * @epd: holds context for the write, namely the bio
4622 * If a page is already under I/O, write_cache_pages() skips it, even
4623 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
4624 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
4625 * and msync() need to guarantee that all the data which was dirty at the time
4626 * the call was made get new I/O started against them. If wbc->sync_mode is
4627 * WB_SYNC_ALL then we were called for data integrity and we must wait for
4628 * existing IO to complete.
4630 static int extent_write_cache_pages(struct address_space *mapping,
4631 struct writeback_control *wbc,
4632 struct extent_page_data *epd)
4634 struct inode *inode = mapping->host;
4637 int nr_to_write_done = 0;
4638 struct pagevec pvec;
4641 pgoff_t end; /* Inclusive */
4643 int range_whole = 0;
4648 * We have to hold onto the inode so that ordered extents can do their
4649 * work when the IO finishes. The alternative to this is failing to add
4650 * an ordered extent if the igrab() fails there and that is a huge pain
4651 * to deal with, so instead just hold onto the inode throughout the
4652 * writepages operation. If it fails here we are freeing up the inode
4653 * anyway and we'd rather not waste our time writing out stuff that is
4654 * going to be truncated anyway.
4659 pagevec_init(&pvec);
4660 if (wbc->range_cyclic) {
4661 index = mapping->writeback_index; /* Start from prev offset */
4664 * Start from the beginning does not need to cycle over the
4665 * range, mark it as scanned.
4667 scanned = (index == 0);
4669 index = wbc->range_start >> PAGE_SHIFT;
4670 end = wbc->range_end >> PAGE_SHIFT;
4671 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
4677 * We do the tagged writepage as long as the snapshot flush bit is set
4678 * and we are the first one who do the filemap_flush() on this inode.
4680 * The nr_to_write == LONG_MAX is needed to make sure other flushers do
4681 * not race in and drop the bit.
4683 if (range_whole && wbc->nr_to_write == LONG_MAX &&
4684 test_and_clear_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
4685 &BTRFS_I(inode)->runtime_flags))
4686 wbc->tagged_writepages = 1;
4688 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
4689 tag = PAGECACHE_TAG_TOWRITE;
4691 tag = PAGECACHE_TAG_DIRTY;
4693 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
4694 tag_pages_for_writeback(mapping, index, end);
4696 while (!done && !nr_to_write_done && (index <= end) &&
4697 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping,
4698 &index, end, tag))) {
4701 for (i = 0; i < nr_pages; i++) {
4702 struct page *page = pvec.pages[i];
4704 done_index = page->index + 1;
4706 * At this point we hold neither the i_pages lock nor
4707 * the page lock: the page may be truncated or
4708 * invalidated (changing page->mapping to NULL),
4709 * or even swizzled back from swapper_space to
4710 * tmpfs file mapping
4712 if (!trylock_page(page)) {
4713 ret = flush_write_bio(epd);
4718 if (unlikely(page->mapping != mapping)) {
4723 if (wbc->sync_mode != WB_SYNC_NONE) {
4724 if (PageWriteback(page)) {
4725 ret = flush_write_bio(epd);
4728 wait_on_page_writeback(page);
4731 if (PageWriteback(page) ||
4732 !clear_page_dirty_for_io(page)) {
4737 ret = __extent_writepage(page, wbc, epd);
4744 * the filesystem may choose to bump up nr_to_write.
4745 * We have to make sure to honor the new nr_to_write
4748 nr_to_write_done = wbc->nr_to_write <= 0;
4750 pagevec_release(&pvec);
4753 if (!scanned && !done) {
4755 * We hit the last page and there is more work to be done: wrap
4756 * back to the start of the file
4762 * If we're looping we could run into a page that is locked by a
4763 * writer and that writer could be waiting on writeback for a
4764 * page in our current bio, and thus deadlock, so flush the
4767 ret = flush_write_bio(epd);
4772 if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole))
4773 mapping->writeback_index = done_index;
4775 btrfs_add_delayed_iput(inode);
4779 int extent_write_full_page(struct page *page, struct writeback_control *wbc)
4782 struct extent_page_data epd = {
4785 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4788 ret = __extent_writepage(page, wbc, &epd);
4791 end_write_bio(&epd, ret);
4795 ret = flush_write_bio(&epd);
4800 int extent_write_locked_range(struct inode *inode, u64 start, u64 end,
4804 struct address_space *mapping = inode->i_mapping;
4806 unsigned long nr_pages = (end - start + PAGE_SIZE) >>
4809 struct extent_page_data epd = {
4812 .sync_io = mode == WB_SYNC_ALL,
4814 struct writeback_control wbc_writepages = {
4816 .nr_to_write = nr_pages * 2,
4817 .range_start = start,
4818 .range_end = end + 1,
4819 /* We're called from an async helper function */
4820 .punt_to_cgroup = 1,
4821 .no_cgroup_owner = 1,
4824 wbc_attach_fdatawrite_inode(&wbc_writepages, inode);
4825 while (start <= end) {
4826 page = find_get_page(mapping, start >> PAGE_SHIFT);
4827 if (clear_page_dirty_for_io(page))
4828 ret = __extent_writepage(page, &wbc_writepages, &epd);
4830 btrfs_writepage_endio_finish_ordered(page, start,
4831 start + PAGE_SIZE - 1, 1);
4840 ret = flush_write_bio(&epd);
4842 end_write_bio(&epd, ret);
4844 wbc_detach_inode(&wbc_writepages);
4848 int extent_writepages(struct address_space *mapping,
4849 struct writeback_control *wbc)
4852 struct extent_page_data epd = {
4855 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4858 ret = extent_write_cache_pages(mapping, wbc, &epd);
4861 end_write_bio(&epd, ret);
4864 ret = flush_write_bio(&epd);
4868 void extent_readahead(struct readahead_control *rac)
4870 struct bio *bio = NULL;
4871 unsigned long bio_flags = 0;
4872 struct page *pagepool[16];
4873 struct extent_map *em_cached = NULL;
4874 u64 prev_em_start = (u64)-1;
4877 while ((nr = readahead_page_batch(rac, pagepool))) {
4878 u64 contig_start = readahead_pos(rac);
4879 u64 contig_end = contig_start + readahead_batch_length(rac) - 1;
4881 contiguous_readpages(pagepool, nr, contig_start, contig_end,
4882 &em_cached, &bio, &bio_flags, &prev_em_start);
4886 free_extent_map(em_cached);
4889 if (submit_one_bio(bio, 0, bio_flags))
4895 * basic invalidatepage code, this waits on any locked or writeback
4896 * ranges corresponding to the page, and then deletes any extent state
4897 * records from the tree
4899 int extent_invalidatepage(struct extent_io_tree *tree,
4900 struct page *page, unsigned long offset)
4902 struct extent_state *cached_state = NULL;
4903 u64 start = page_offset(page);
4904 u64 end = start + PAGE_SIZE - 1;
4905 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
4907 /* This function is only called for the btree inode */
4908 ASSERT(tree->owner == IO_TREE_BTREE_INODE_IO);
4910 start += ALIGN(offset, blocksize);
4914 lock_extent_bits(tree, start, end, &cached_state);
4915 wait_on_page_writeback(page);
4918 * Currently for btree io tree, only EXTENT_LOCKED is utilized,
4919 * so here we only need to unlock the extent range to free any
4920 * existing extent state.
4922 unlock_extent_cached(tree, start, end, &cached_state);
4927 * a helper for releasepage, this tests for areas of the page that
4928 * are locked or under IO and drops the related state bits if it is safe
4931 static int try_release_extent_state(struct extent_io_tree *tree,
4932 struct page *page, gfp_t mask)
4934 u64 start = page_offset(page);
4935 u64 end = start + PAGE_SIZE - 1;
4938 if (test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL)) {
4942 * At this point we can safely clear everything except the
4943 * locked bit, the nodatasum bit and the delalloc new bit.
4944 * The delalloc new bit will be cleared by ordered extent
4947 ret = __clear_extent_bit(tree, start, end,
4948 ~(EXTENT_LOCKED | EXTENT_NODATASUM | EXTENT_DELALLOC_NEW),
4949 0, 0, NULL, mask, NULL);
4951 /* if clear_extent_bit failed for enomem reasons,
4952 * we can't allow the release to continue.
4963 * a helper for releasepage. As long as there are no locked extents
4964 * in the range corresponding to the page, both state records and extent
4965 * map records are removed
4967 int try_release_extent_mapping(struct page *page, gfp_t mask)
4969 struct extent_map *em;
4970 u64 start = page_offset(page);
4971 u64 end = start + PAGE_SIZE - 1;
4972 struct btrfs_inode *btrfs_inode = BTRFS_I(page->mapping->host);
4973 struct extent_io_tree *tree = &btrfs_inode->io_tree;
4974 struct extent_map_tree *map = &btrfs_inode->extent_tree;
4976 if (gfpflags_allow_blocking(mask) &&
4977 page->mapping->host->i_size > SZ_16M) {
4979 while (start <= end) {
4980 struct btrfs_fs_info *fs_info;
4983 len = end - start + 1;
4984 write_lock(&map->lock);
4985 em = lookup_extent_mapping(map, start, len);
4987 write_unlock(&map->lock);
4990 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
4991 em->start != start) {
4992 write_unlock(&map->lock);
4993 free_extent_map(em);
4996 if (test_range_bit(tree, em->start,
4997 extent_map_end(em) - 1,
4998 EXTENT_LOCKED, 0, NULL))
5001 * If it's not in the list of modified extents, used
5002 * by a fast fsync, we can remove it. If it's being
5003 * logged we can safely remove it since fsync took an
5004 * extra reference on the em.
5006 if (list_empty(&em->list) ||
5007 test_bit(EXTENT_FLAG_LOGGING, &em->flags))
5010 * If it's in the list of modified extents, remove it
5011 * only if its generation is older then the current one,
5012 * in which case we don't need it for a fast fsync.
5013 * Otherwise don't remove it, we could be racing with an
5014 * ongoing fast fsync that could miss the new extent.
5016 fs_info = btrfs_inode->root->fs_info;
5017 spin_lock(&fs_info->trans_lock);
5018 cur_gen = fs_info->generation;
5019 spin_unlock(&fs_info->trans_lock);
5020 if (em->generation >= cur_gen)
5024 * We only remove extent maps that are not in the list of
5025 * modified extents or that are in the list but with a
5026 * generation lower then the current generation, so there
5027 * is no need to set the full fsync flag on the inode (it
5028 * hurts the fsync performance for workloads with a data
5029 * size that exceeds or is close to the system's memory).
5031 remove_extent_mapping(map, em);
5032 /* once for the rb tree */
5033 free_extent_map(em);
5035 start = extent_map_end(em);
5036 write_unlock(&map->lock);
5039 free_extent_map(em);
5041 cond_resched(); /* Allow large-extent preemption. */
5044 return try_release_extent_state(tree, page, mask);
5048 * helper function for fiemap, which doesn't want to see any holes.
5049 * This maps until we find something past 'last'
5051 static struct extent_map *get_extent_skip_holes(struct btrfs_inode *inode,
5052 u64 offset, u64 last)
5054 u64 sectorsize = btrfs_inode_sectorsize(inode);
5055 struct extent_map *em;
5062 len = last - offset;
5065 len = ALIGN(len, sectorsize);
5066 em = btrfs_get_extent_fiemap(inode, offset, len);
5067 if (IS_ERR_OR_NULL(em))
5070 /* if this isn't a hole return it */
5071 if (em->block_start != EXTENT_MAP_HOLE)
5074 /* this is a hole, advance to the next extent */
5075 offset = extent_map_end(em);
5076 free_extent_map(em);
5084 * To cache previous fiemap extent
5086 * Will be used for merging fiemap extent
5088 struct fiemap_cache {
5097 * Helper to submit fiemap extent.
5099 * Will try to merge current fiemap extent specified by @offset, @phys,
5100 * @len and @flags with cached one.
5101 * And only when we fails to merge, cached one will be submitted as
5104 * Return value is the same as fiemap_fill_next_extent().
5106 static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
5107 struct fiemap_cache *cache,
5108 u64 offset, u64 phys, u64 len, u32 flags)
5116 * Sanity check, extent_fiemap() should have ensured that new
5117 * fiemap extent won't overlap with cached one.
5120 * NOTE: Physical address can overlap, due to compression
5122 if (cache->offset + cache->len > offset) {
5128 * Only merges fiemap extents if
5129 * 1) Their logical addresses are continuous
5131 * 2) Their physical addresses are continuous
5132 * So truly compressed (physical size smaller than logical size)
5133 * extents won't get merged with each other
5135 * 3) Share same flags except FIEMAP_EXTENT_LAST
5136 * So regular extent won't get merged with prealloc extent
5138 if (cache->offset + cache->len == offset &&
5139 cache->phys + cache->len == phys &&
5140 (cache->flags & ~FIEMAP_EXTENT_LAST) ==
5141 (flags & ~FIEMAP_EXTENT_LAST)) {
5143 cache->flags |= flags;
5144 goto try_submit_last;
5147 /* Not mergeable, need to submit cached one */
5148 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
5149 cache->len, cache->flags);
5150 cache->cached = false;
5154 cache->cached = true;
5155 cache->offset = offset;
5158 cache->flags = flags;
5160 if (cache->flags & FIEMAP_EXTENT_LAST) {
5161 ret = fiemap_fill_next_extent(fieinfo, cache->offset,
5162 cache->phys, cache->len, cache->flags);
5163 cache->cached = false;
5169 * Emit last fiemap cache
5171 * The last fiemap cache may still be cached in the following case:
5173 * |<- Fiemap range ->|
5174 * |<------------ First extent ----------->|
5176 * In this case, the first extent range will be cached but not emitted.
5177 * So we must emit it before ending extent_fiemap().
5179 static int emit_last_fiemap_cache(struct fiemap_extent_info *fieinfo,
5180 struct fiemap_cache *cache)
5187 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
5188 cache->len, cache->flags);
5189 cache->cached = false;
5195 int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,
5200 u64 max = start + len;
5204 u64 last_for_get_extent = 0;
5206 u64 isize = i_size_read(&inode->vfs_inode);
5207 struct btrfs_key found_key;
5208 struct extent_map *em = NULL;
5209 struct extent_state *cached_state = NULL;
5210 struct btrfs_path *path;
5211 struct btrfs_root *root = inode->root;
5212 struct fiemap_cache cache = { 0 };
5213 struct ulist *roots;
5214 struct ulist *tmp_ulist;
5223 path = btrfs_alloc_path();
5227 roots = ulist_alloc(GFP_KERNEL);
5228 tmp_ulist = ulist_alloc(GFP_KERNEL);
5229 if (!roots || !tmp_ulist) {
5231 goto out_free_ulist;
5235 * We can't initialize that to 'start' as this could miss extents due
5236 * to extent item merging
5239 start = round_down(start, btrfs_inode_sectorsize(inode));
5240 len = round_up(max, btrfs_inode_sectorsize(inode)) - start;
5243 * lookup the last file extent. We're not using i_size here
5244 * because there might be preallocation past i_size
5246 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode), -1,
5249 goto out_free_ulist;
5257 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
5258 found_type = found_key.type;
5260 /* No extents, but there might be delalloc bits */
5261 if (found_key.objectid != btrfs_ino(inode) ||
5262 found_type != BTRFS_EXTENT_DATA_KEY) {
5263 /* have to trust i_size as the end */
5265 last_for_get_extent = isize;
5268 * remember the start of the last extent. There are a
5269 * bunch of different factors that go into the length of the
5270 * extent, so its much less complex to remember where it started
5272 last = found_key.offset;
5273 last_for_get_extent = last + 1;
5275 btrfs_release_path(path);
5278 * we might have some extents allocated but more delalloc past those
5279 * extents. so, we trust isize unless the start of the last extent is
5284 last_for_get_extent = isize;
5287 lock_extent_bits(&inode->io_tree, start, start + len - 1,
5290 em = get_extent_skip_holes(inode, start, last_for_get_extent);
5299 u64 offset_in_extent = 0;
5301 /* break if the extent we found is outside the range */
5302 if (em->start >= max || extent_map_end(em) < off)
5306 * get_extent may return an extent that starts before our
5307 * requested range. We have to make sure the ranges
5308 * we return to fiemap always move forward and don't
5309 * overlap, so adjust the offsets here
5311 em_start = max(em->start, off);
5314 * record the offset from the start of the extent
5315 * for adjusting the disk offset below. Only do this if the
5316 * extent isn't compressed since our in ram offset may be past
5317 * what we have actually allocated on disk.
5319 if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
5320 offset_in_extent = em_start - em->start;
5321 em_end = extent_map_end(em);
5322 em_len = em_end - em_start;
5324 if (em->block_start < EXTENT_MAP_LAST_BYTE)
5325 disko = em->block_start + offset_in_extent;
5330 * bump off for our next call to get_extent
5332 off = extent_map_end(em);
5336 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
5338 flags |= FIEMAP_EXTENT_LAST;
5339 } else if (em->block_start == EXTENT_MAP_INLINE) {
5340 flags |= (FIEMAP_EXTENT_DATA_INLINE |
5341 FIEMAP_EXTENT_NOT_ALIGNED);
5342 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
5343 flags |= (FIEMAP_EXTENT_DELALLOC |
5344 FIEMAP_EXTENT_UNKNOWN);
5345 } else if (fieinfo->fi_extents_max) {
5346 u64 bytenr = em->block_start -
5347 (em->start - em->orig_start);
5350 * As btrfs supports shared space, this information
5351 * can be exported to userspace tools via
5352 * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0
5353 * then we're just getting a count and we can skip the
5356 ret = btrfs_check_shared(root, btrfs_ino(inode),
5357 bytenr, roots, tmp_ulist);
5361 flags |= FIEMAP_EXTENT_SHARED;
5364 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
5365 flags |= FIEMAP_EXTENT_ENCODED;
5366 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5367 flags |= FIEMAP_EXTENT_UNWRITTEN;
5369 free_extent_map(em);
5371 if ((em_start >= last) || em_len == (u64)-1 ||
5372 (last == (u64)-1 && isize <= em_end)) {
5373 flags |= FIEMAP_EXTENT_LAST;
5377 /* now scan forward to see if this is really the last extent. */
5378 em = get_extent_skip_holes(inode, off, last_for_get_extent);
5384 flags |= FIEMAP_EXTENT_LAST;
5387 ret = emit_fiemap_extent(fieinfo, &cache, em_start, disko,
5397 ret = emit_last_fiemap_cache(fieinfo, &cache);
5398 free_extent_map(em);
5400 unlock_extent_cached(&inode->io_tree, start, start + len - 1,
5404 btrfs_free_path(path);
5406 ulist_free(tmp_ulist);
5410 static void __free_extent_buffer(struct extent_buffer *eb)
5412 kmem_cache_free(extent_buffer_cache, eb);
5415 int extent_buffer_under_io(const struct extent_buffer *eb)
5417 return (atomic_read(&eb->io_pages) ||
5418 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
5419 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
5422 static bool page_range_has_eb(struct btrfs_fs_info *fs_info, struct page *page)
5424 struct btrfs_subpage *subpage;
5426 lockdep_assert_held(&page->mapping->private_lock);
5428 if (PagePrivate(page)) {
5429 subpage = (struct btrfs_subpage *)page->private;
5430 if (atomic_read(&subpage->eb_refs))
5436 static void detach_extent_buffer_page(struct extent_buffer *eb, struct page *page)
5438 struct btrfs_fs_info *fs_info = eb->fs_info;
5439 const bool mapped = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
5442 * For mapped eb, we're going to change the page private, which should
5443 * be done under the private_lock.
5446 spin_lock(&page->mapping->private_lock);
5448 if (!PagePrivate(page)) {
5450 spin_unlock(&page->mapping->private_lock);
5454 if (fs_info->sectorsize == PAGE_SIZE) {
5456 * We do this since we'll remove the pages after we've
5457 * removed the eb from the radix tree, so we could race
5458 * and have this page now attached to the new eb. So
5459 * only clear page_private if it's still connected to
5462 if (PagePrivate(page) &&
5463 page->private == (unsigned long)eb) {
5464 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
5465 BUG_ON(PageDirty(page));
5466 BUG_ON(PageWriteback(page));
5468 * We need to make sure we haven't be attached
5471 detach_page_private(page);
5474 spin_unlock(&page->mapping->private_lock);
5479 * For subpage, we can have dummy eb with page private. In this case,
5480 * we can directly detach the private as such page is only attached to
5481 * one dummy eb, no sharing.
5484 btrfs_detach_subpage(fs_info, page);
5488 btrfs_page_dec_eb_refs(fs_info, page);
5491 * We can only detach the page private if there are no other ebs in the
5494 if (!page_range_has_eb(fs_info, page))
5495 btrfs_detach_subpage(fs_info, page);
5497 spin_unlock(&page->mapping->private_lock);
5500 /* Release all pages attached to the extent buffer */
5501 static void btrfs_release_extent_buffer_pages(struct extent_buffer *eb)
5506 ASSERT(!extent_buffer_under_io(eb));
5508 num_pages = num_extent_pages(eb);
5509 for (i = 0; i < num_pages; i++) {
5510 struct page *page = eb->pages[i];
5515 detach_extent_buffer_page(eb, page);
5517 /* One for when we allocated the page */
5523 * Helper for releasing the extent buffer.
5525 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
5527 btrfs_release_extent_buffer_pages(eb);
5528 btrfs_leak_debug_del(&eb->fs_info->eb_leak_lock, &eb->leak_list);
5529 __free_extent_buffer(eb);
5532 static struct extent_buffer *
5533 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
5536 struct extent_buffer *eb = NULL;
5538 eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL);
5541 eb->fs_info = fs_info;
5543 init_rwsem(&eb->lock);
5545 btrfs_leak_debug_add(&fs_info->eb_leak_lock, &eb->leak_list,
5546 &fs_info->allocated_ebs);
5547 INIT_LIST_HEAD(&eb->release_list);
5549 spin_lock_init(&eb->refs_lock);
5550 atomic_set(&eb->refs, 1);
5551 atomic_set(&eb->io_pages, 0);
5553 ASSERT(len <= BTRFS_MAX_METADATA_BLOCKSIZE);
5558 struct extent_buffer *btrfs_clone_extent_buffer(const struct extent_buffer *src)
5562 struct extent_buffer *new;
5563 int num_pages = num_extent_pages(src);
5565 new = __alloc_extent_buffer(src->fs_info, src->start, src->len);
5570 * Set UNMAPPED before calling btrfs_release_extent_buffer(), as
5571 * btrfs_release_extent_buffer() have different behavior for
5572 * UNMAPPED subpage extent buffer.
5574 set_bit(EXTENT_BUFFER_UNMAPPED, &new->bflags);
5576 for (i = 0; i < num_pages; i++) {
5579 p = alloc_page(GFP_NOFS);
5581 btrfs_release_extent_buffer(new);
5584 ret = attach_extent_buffer_page(new, p, NULL);
5587 btrfs_release_extent_buffer(new);
5590 WARN_ON(PageDirty(p));
5592 copy_page(page_address(p), page_address(src->pages[i]));
5594 set_extent_buffer_uptodate(new);
5599 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
5600 u64 start, unsigned long len)
5602 struct extent_buffer *eb;
5606 eb = __alloc_extent_buffer(fs_info, start, len);
5610 num_pages = num_extent_pages(eb);
5611 for (i = 0; i < num_pages; i++) {
5614 eb->pages[i] = alloc_page(GFP_NOFS);
5617 ret = attach_extent_buffer_page(eb, eb->pages[i], NULL);
5621 set_extent_buffer_uptodate(eb);
5622 btrfs_set_header_nritems(eb, 0);
5623 set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
5627 for (; i > 0; i--) {
5628 detach_extent_buffer_page(eb, eb->pages[i - 1]);
5629 __free_page(eb->pages[i - 1]);
5631 __free_extent_buffer(eb);
5635 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
5638 return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize);
5641 static void check_buffer_tree_ref(struct extent_buffer *eb)
5645 * The TREE_REF bit is first set when the extent_buffer is added
5646 * to the radix tree. It is also reset, if unset, when a new reference
5647 * is created by find_extent_buffer.
5649 * It is only cleared in two cases: freeing the last non-tree
5650 * reference to the extent_buffer when its STALE bit is set or
5651 * calling releasepage when the tree reference is the only reference.
5653 * In both cases, care is taken to ensure that the extent_buffer's
5654 * pages are not under io. However, releasepage can be concurrently
5655 * called with creating new references, which is prone to race
5656 * conditions between the calls to check_buffer_tree_ref in those
5657 * codepaths and clearing TREE_REF in try_release_extent_buffer.
5659 * The actual lifetime of the extent_buffer in the radix tree is
5660 * adequately protected by the refcount, but the TREE_REF bit and
5661 * its corresponding reference are not. To protect against this
5662 * class of races, we call check_buffer_tree_ref from the codepaths
5663 * which trigger io after they set eb->io_pages. Note that once io is
5664 * initiated, TREE_REF can no longer be cleared, so that is the
5665 * moment at which any such race is best fixed.
5667 refs = atomic_read(&eb->refs);
5668 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5671 spin_lock(&eb->refs_lock);
5672 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5673 atomic_inc(&eb->refs);
5674 spin_unlock(&eb->refs_lock);
5677 static void mark_extent_buffer_accessed(struct extent_buffer *eb,
5678 struct page *accessed)
5682 check_buffer_tree_ref(eb);
5684 num_pages = num_extent_pages(eb);
5685 for (i = 0; i < num_pages; i++) {
5686 struct page *p = eb->pages[i];
5689 mark_page_accessed(p);
5693 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
5696 struct extent_buffer *eb;
5698 eb = find_extent_buffer_nolock(fs_info, start);
5702 * Lock our eb's refs_lock to avoid races with free_extent_buffer().
5703 * When we get our eb it might be flagged with EXTENT_BUFFER_STALE and
5704 * another task running free_extent_buffer() might have seen that flag
5705 * set, eb->refs == 2, that the buffer isn't under IO (dirty and
5706 * writeback flags not set) and it's still in the tree (flag
5707 * EXTENT_BUFFER_TREE_REF set), therefore being in the process of
5708 * decrementing the extent buffer's reference count twice. So here we
5709 * could race and increment the eb's reference count, clear its stale
5710 * flag, mark it as dirty and drop our reference before the other task
5711 * finishes executing free_extent_buffer, which would later result in
5712 * an attempt to free an extent buffer that is dirty.
5714 if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) {
5715 spin_lock(&eb->refs_lock);
5716 spin_unlock(&eb->refs_lock);
5718 mark_extent_buffer_accessed(eb, NULL);
5722 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5723 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
5726 struct extent_buffer *eb, *exists = NULL;
5729 eb = find_extent_buffer(fs_info, start);
5732 eb = alloc_dummy_extent_buffer(fs_info, start);
5734 return ERR_PTR(-ENOMEM);
5735 eb->fs_info = fs_info;
5737 ret = radix_tree_preload(GFP_NOFS);
5739 exists = ERR_PTR(ret);
5742 spin_lock(&fs_info->buffer_lock);
5743 ret = radix_tree_insert(&fs_info->buffer_radix,
5744 start >> fs_info->sectorsize_bits, eb);
5745 spin_unlock(&fs_info->buffer_lock);
5746 radix_tree_preload_end();
5747 if (ret == -EEXIST) {
5748 exists = find_extent_buffer(fs_info, start);
5754 check_buffer_tree_ref(eb);
5755 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
5759 btrfs_release_extent_buffer(eb);
5764 static struct extent_buffer *grab_extent_buffer(
5765 struct btrfs_fs_info *fs_info, struct page *page)
5767 struct extent_buffer *exists;
5770 * For subpage case, we completely rely on radix tree to ensure we
5771 * don't try to insert two ebs for the same bytenr. So here we always
5772 * return NULL and just continue.
5774 if (fs_info->sectorsize < PAGE_SIZE)
5777 /* Page not yet attached to an extent buffer */
5778 if (!PagePrivate(page))
5782 * We could have already allocated an eb for this page and attached one
5783 * so lets see if we can get a ref on the existing eb, and if we can we
5784 * know it's good and we can just return that one, else we know we can
5785 * just overwrite page->private.
5787 exists = (struct extent_buffer *)page->private;
5788 if (atomic_inc_not_zero(&exists->refs))
5791 WARN_ON(PageDirty(page));
5792 detach_page_private(page);
5796 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
5797 u64 start, u64 owner_root, int level)
5799 unsigned long len = fs_info->nodesize;
5802 unsigned long index = start >> PAGE_SHIFT;
5803 struct extent_buffer *eb;
5804 struct extent_buffer *exists = NULL;
5806 struct address_space *mapping = fs_info->btree_inode->i_mapping;
5810 if (!IS_ALIGNED(start, fs_info->sectorsize)) {
5811 btrfs_err(fs_info, "bad tree block start %llu", start);
5812 return ERR_PTR(-EINVAL);
5815 #if BITS_PER_LONG == 32
5816 if (start >= MAX_LFS_FILESIZE) {
5817 btrfs_err_rl(fs_info,
5818 "extent buffer %llu is beyond 32bit page cache limit", start);
5819 btrfs_err_32bit_limit(fs_info);
5820 return ERR_PTR(-EOVERFLOW);
5822 if (start >= BTRFS_32BIT_EARLY_WARN_THRESHOLD)
5823 btrfs_warn_32bit_limit(fs_info);
5826 if (fs_info->sectorsize < PAGE_SIZE &&
5827 offset_in_page(start) + len > PAGE_SIZE) {
5829 "tree block crosses page boundary, start %llu nodesize %lu",
5831 return ERR_PTR(-EINVAL);
5834 eb = find_extent_buffer(fs_info, start);
5838 eb = __alloc_extent_buffer(fs_info, start, len);
5840 return ERR_PTR(-ENOMEM);
5841 btrfs_set_buffer_lockdep_class(owner_root, eb, level);
5843 num_pages = num_extent_pages(eb);
5844 for (i = 0; i < num_pages; i++, index++) {
5845 struct btrfs_subpage *prealloc = NULL;
5847 p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL);
5849 exists = ERR_PTR(-ENOMEM);
5854 * Preallocate page->private for subpage case, so that we won't
5855 * allocate memory with private_lock hold. The memory will be
5856 * freed by attach_extent_buffer_page() or freed manually if
5859 * Although we have ensured one subpage eb can only have one
5860 * page, but it may change in the future for 16K page size
5861 * support, so we still preallocate the memory in the loop.
5863 ret = btrfs_alloc_subpage(fs_info, &prealloc,
5864 BTRFS_SUBPAGE_METADATA);
5868 exists = ERR_PTR(ret);
5872 spin_lock(&mapping->private_lock);
5873 exists = grab_extent_buffer(fs_info, p);
5875 spin_unlock(&mapping->private_lock);
5878 mark_extent_buffer_accessed(exists, p);
5879 btrfs_free_subpage(prealloc);
5882 /* Should not fail, as we have preallocated the memory */
5883 ret = attach_extent_buffer_page(eb, p, prealloc);
5886 * To inform we have extra eb under allocation, so that
5887 * detach_extent_buffer_page() won't release the page private
5888 * when the eb hasn't yet been inserted into radix tree.
5890 * The ref will be decreased when the eb released the page, in
5891 * detach_extent_buffer_page().
5892 * Thus needs no special handling in error path.
5894 btrfs_page_inc_eb_refs(fs_info, p);
5895 spin_unlock(&mapping->private_lock);
5897 WARN_ON(btrfs_page_test_dirty(fs_info, p, eb->start, eb->len));
5899 if (!PageUptodate(p))
5903 * We can't unlock the pages just yet since the extent buffer
5904 * hasn't been properly inserted in the radix tree, this
5905 * opens a race with btree_releasepage which can free a page
5906 * while we are still filling in all pages for the buffer and
5911 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5913 ret = radix_tree_preload(GFP_NOFS);
5915 exists = ERR_PTR(ret);
5919 spin_lock(&fs_info->buffer_lock);
5920 ret = radix_tree_insert(&fs_info->buffer_radix,
5921 start >> fs_info->sectorsize_bits, eb);
5922 spin_unlock(&fs_info->buffer_lock);
5923 radix_tree_preload_end();
5924 if (ret == -EEXIST) {
5925 exists = find_extent_buffer(fs_info, start);
5931 /* add one reference for the tree */
5932 check_buffer_tree_ref(eb);
5933 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
5936 * Now it's safe to unlock the pages because any calls to
5937 * btree_releasepage will correctly detect that a page belongs to a
5938 * live buffer and won't free them prematurely.
5940 for (i = 0; i < num_pages; i++)
5941 unlock_page(eb->pages[i]);
5945 WARN_ON(!atomic_dec_and_test(&eb->refs));
5946 for (i = 0; i < num_pages; i++) {
5948 unlock_page(eb->pages[i]);
5951 btrfs_release_extent_buffer(eb);
5955 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
5957 struct extent_buffer *eb =
5958 container_of(head, struct extent_buffer, rcu_head);
5960 __free_extent_buffer(eb);
5963 static int release_extent_buffer(struct extent_buffer *eb)
5964 __releases(&eb->refs_lock)
5966 lockdep_assert_held(&eb->refs_lock);
5968 WARN_ON(atomic_read(&eb->refs) == 0);
5969 if (atomic_dec_and_test(&eb->refs)) {
5970 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
5971 struct btrfs_fs_info *fs_info = eb->fs_info;
5973 spin_unlock(&eb->refs_lock);
5975 spin_lock(&fs_info->buffer_lock);
5976 radix_tree_delete(&fs_info->buffer_radix,
5977 eb->start >> fs_info->sectorsize_bits);
5978 spin_unlock(&fs_info->buffer_lock);
5980 spin_unlock(&eb->refs_lock);
5983 btrfs_leak_debug_del(&eb->fs_info->eb_leak_lock, &eb->leak_list);
5984 /* Should be safe to release our pages at this point */
5985 btrfs_release_extent_buffer_pages(eb);
5986 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5987 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags))) {
5988 __free_extent_buffer(eb);
5992 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
5995 spin_unlock(&eb->refs_lock);
6000 void free_extent_buffer(struct extent_buffer *eb)
6008 refs = atomic_read(&eb->refs);
6009 if ((!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && refs <= 3)
6010 || (test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) &&
6013 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
6018 spin_lock(&eb->refs_lock);
6019 if (atomic_read(&eb->refs) == 2 &&
6020 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
6021 !extent_buffer_under_io(eb) &&
6022 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
6023 atomic_dec(&eb->refs);
6026 * I know this is terrible, but it's temporary until we stop tracking
6027 * the uptodate bits and such for the extent buffers.
6029 release_extent_buffer(eb);
6032 void free_extent_buffer_stale(struct extent_buffer *eb)
6037 spin_lock(&eb->refs_lock);
6038 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
6040 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
6041 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
6042 atomic_dec(&eb->refs);
6043 release_extent_buffer(eb);
6046 static void btree_clear_page_dirty(struct page *page)
6048 ASSERT(PageDirty(page));
6049 ASSERT(PageLocked(page));
6050 clear_page_dirty_for_io(page);
6051 xa_lock_irq(&page->mapping->i_pages);
6052 if (!PageDirty(page))
6053 __xa_clear_mark(&page->mapping->i_pages,
6054 page_index(page), PAGECACHE_TAG_DIRTY);
6055 xa_unlock_irq(&page->mapping->i_pages);
6058 static void clear_subpage_extent_buffer_dirty(const struct extent_buffer *eb)
6060 struct btrfs_fs_info *fs_info = eb->fs_info;
6061 struct page *page = eb->pages[0];
6064 /* btree_clear_page_dirty() needs page locked */
6066 last = btrfs_subpage_clear_and_test_dirty(fs_info, page, eb->start,
6069 btree_clear_page_dirty(page);
6071 WARN_ON(atomic_read(&eb->refs) == 0);
6074 void clear_extent_buffer_dirty(const struct extent_buffer *eb)
6080 if (eb->fs_info->sectorsize < PAGE_SIZE)
6081 return clear_subpage_extent_buffer_dirty(eb);
6083 num_pages = num_extent_pages(eb);
6085 for (i = 0; i < num_pages; i++) {
6086 page = eb->pages[i];
6087 if (!PageDirty(page))
6090 btree_clear_page_dirty(page);
6091 ClearPageError(page);
6094 WARN_ON(atomic_read(&eb->refs) == 0);
6097 bool set_extent_buffer_dirty(struct extent_buffer *eb)
6103 check_buffer_tree_ref(eb);
6105 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
6107 num_pages = num_extent_pages(eb);
6108 WARN_ON(atomic_read(&eb->refs) == 0);
6109 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
6112 bool subpage = eb->fs_info->sectorsize < PAGE_SIZE;
6115 * For subpage case, we can have other extent buffers in the
6116 * same page, and in clear_subpage_extent_buffer_dirty() we
6117 * have to clear page dirty without subpage lock held.
6118 * This can cause race where our page gets dirty cleared after
6121 * Thankfully, clear_subpage_extent_buffer_dirty() has locked
6122 * its page for other reasons, we can use page lock to prevent
6126 lock_page(eb->pages[0]);
6127 for (i = 0; i < num_pages; i++)
6128 btrfs_page_set_dirty(eb->fs_info, eb->pages[i],
6129 eb->start, eb->len);
6131 unlock_page(eb->pages[0]);
6133 #ifdef CONFIG_BTRFS_DEBUG
6134 for (i = 0; i < num_pages; i++)
6135 ASSERT(PageDirty(eb->pages[i]));
6141 void clear_extent_buffer_uptodate(struct extent_buffer *eb)
6143 struct btrfs_fs_info *fs_info = eb->fs_info;
6148 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
6149 num_pages = num_extent_pages(eb);
6150 for (i = 0; i < num_pages; i++) {
6151 page = eb->pages[i];
6153 btrfs_page_clear_uptodate(fs_info, page,
6154 eb->start, eb->len);
6158 void set_extent_buffer_uptodate(struct extent_buffer *eb)
6160 struct btrfs_fs_info *fs_info = eb->fs_info;
6165 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
6166 num_pages = num_extent_pages(eb);
6167 for (i = 0; i < num_pages; i++) {
6168 page = eb->pages[i];
6169 btrfs_page_set_uptodate(fs_info, page, eb->start, eb->len);
6173 static int read_extent_buffer_subpage(struct extent_buffer *eb, int wait,
6176 struct btrfs_fs_info *fs_info = eb->fs_info;
6177 struct extent_io_tree *io_tree;
6178 struct page *page = eb->pages[0];
6179 struct bio *bio = NULL;
6182 ASSERT(!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags));
6183 ASSERT(PagePrivate(page));
6184 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
6186 if (wait == WAIT_NONE) {
6187 ret = try_lock_extent(io_tree, eb->start,
6188 eb->start + eb->len - 1);
6192 ret = lock_extent(io_tree, eb->start, eb->start + eb->len - 1);
6198 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags) ||
6199 PageUptodate(page) ||
6200 btrfs_subpage_test_uptodate(fs_info, page, eb->start, eb->len)) {
6201 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
6202 unlock_extent(io_tree, eb->start, eb->start + eb->len - 1);
6206 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
6207 eb->read_mirror = 0;
6208 atomic_set(&eb->io_pages, 1);
6209 check_buffer_tree_ref(eb);
6210 btrfs_subpage_clear_error(fs_info, page, eb->start, eb->len);
6212 ret = submit_extent_page(REQ_OP_READ | REQ_META, NULL, page, eb->start,
6213 eb->len, eb->start - page_offset(page), &bio,
6214 end_bio_extent_readpage, mirror_num, 0, 0,
6218 * In the endio function, if we hit something wrong we will
6219 * increase the io_pages, so here we need to decrease it for
6222 atomic_dec(&eb->io_pages);
6227 tmp = submit_one_bio(bio, mirror_num, 0);
6231 if (ret || wait != WAIT_COMPLETE)
6234 wait_extent_bit(io_tree, eb->start, eb->start + eb->len - 1, EXTENT_LOCKED);
6235 if (!test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
6240 int read_extent_buffer_pages(struct extent_buffer *eb, int wait, int mirror_num)
6246 int locked_pages = 0;
6247 int all_uptodate = 1;
6249 unsigned long num_reads = 0;
6250 struct bio *bio = NULL;
6251 unsigned long bio_flags = 0;
6253 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
6256 if (eb->fs_info->sectorsize < PAGE_SIZE)
6257 return read_extent_buffer_subpage(eb, wait, mirror_num);
6259 num_pages = num_extent_pages(eb);
6260 for (i = 0; i < num_pages; i++) {
6261 page = eb->pages[i];
6262 if (wait == WAIT_NONE) {
6264 * WAIT_NONE is only utilized by readahead. If we can't
6265 * acquire the lock atomically it means either the eb
6266 * is being read out or under modification.
6267 * Either way the eb will be or has been cached,
6268 * readahead can exit safely.
6270 if (!trylock_page(page))
6278 * We need to firstly lock all pages to make sure that
6279 * the uptodate bit of our pages won't be affected by
6280 * clear_extent_buffer_uptodate().
6282 for (i = 0; i < num_pages; i++) {
6283 page = eb->pages[i];
6284 if (!PageUptodate(page)) {
6291 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
6295 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
6296 eb->read_mirror = 0;
6297 atomic_set(&eb->io_pages, num_reads);
6299 * It is possible for releasepage to clear the TREE_REF bit before we
6300 * set io_pages. See check_buffer_tree_ref for a more detailed comment.
6302 check_buffer_tree_ref(eb);
6303 for (i = 0; i < num_pages; i++) {
6304 page = eb->pages[i];
6306 if (!PageUptodate(page)) {
6308 atomic_dec(&eb->io_pages);
6313 ClearPageError(page);
6314 err = submit_extent_page(REQ_OP_READ | REQ_META, NULL,
6315 page, page_offset(page), PAGE_SIZE, 0,
6316 &bio, end_bio_extent_readpage,
6317 mirror_num, 0, 0, false);
6320 * We failed to submit the bio so it's the
6321 * caller's responsibility to perform cleanup
6322 * i.e unlock page/set error bit.
6327 atomic_dec(&eb->io_pages);
6335 err = submit_one_bio(bio, mirror_num, bio_flags);
6340 if (ret || wait != WAIT_COMPLETE)
6343 for (i = 0; i < num_pages; i++) {
6344 page = eb->pages[i];
6345 wait_on_page_locked(page);
6346 if (!PageUptodate(page))
6353 while (locked_pages > 0) {
6355 page = eb->pages[locked_pages];
6361 static bool report_eb_range(const struct extent_buffer *eb, unsigned long start,
6364 btrfs_warn(eb->fs_info,
6365 "access to eb bytenr %llu len %lu out of range start %lu len %lu",
6366 eb->start, eb->len, start, len);
6367 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
6373 * Check if the [start, start + len) range is valid before reading/writing
6375 * NOTE: @start and @len are offset inside the eb, not logical address.
6377 * Caller should not touch the dst/src memory if this function returns error.
6379 static inline int check_eb_range(const struct extent_buffer *eb,
6380 unsigned long start, unsigned long len)
6382 unsigned long offset;
6384 /* start, start + len should not go beyond eb->len nor overflow */
6385 if (unlikely(check_add_overflow(start, len, &offset) || offset > eb->len))
6386 return report_eb_range(eb, start, len);
6391 void read_extent_buffer(const struct extent_buffer *eb, void *dstv,
6392 unsigned long start, unsigned long len)
6398 char *dst = (char *)dstv;
6399 unsigned long i = get_eb_page_index(start);
6401 if (check_eb_range(eb, start, len))
6404 offset = get_eb_offset_in_page(eb, start);
6407 page = eb->pages[i];
6409 cur = min(len, (PAGE_SIZE - offset));
6410 kaddr = page_address(page);
6411 memcpy(dst, kaddr + offset, cur);
6420 int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb,
6422 unsigned long start, unsigned long len)
6428 char __user *dst = (char __user *)dstv;
6429 unsigned long i = get_eb_page_index(start);
6432 WARN_ON(start > eb->len);
6433 WARN_ON(start + len > eb->start + eb->len);
6435 offset = get_eb_offset_in_page(eb, start);
6438 page = eb->pages[i];
6440 cur = min(len, (PAGE_SIZE - offset));
6441 kaddr = page_address(page);
6442 if (copy_to_user_nofault(dst, kaddr + offset, cur)) {
6456 int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,
6457 unsigned long start, unsigned long len)
6463 char *ptr = (char *)ptrv;
6464 unsigned long i = get_eb_page_index(start);
6467 if (check_eb_range(eb, start, len))
6470 offset = get_eb_offset_in_page(eb, start);
6473 page = eb->pages[i];
6475 cur = min(len, (PAGE_SIZE - offset));
6477 kaddr = page_address(page);
6478 ret = memcmp(ptr, kaddr + offset, cur);
6491 * Check that the extent buffer is uptodate.
6493 * For regular sector size == PAGE_SIZE case, check if @page is uptodate.
6494 * For subpage case, check if the range covered by the eb has EXTENT_UPTODATE.
6496 static void assert_eb_page_uptodate(const struct extent_buffer *eb,
6499 struct btrfs_fs_info *fs_info = eb->fs_info;
6501 if (fs_info->sectorsize < PAGE_SIZE) {
6504 uptodate = btrfs_subpage_test_uptodate(fs_info, page,
6505 eb->start, eb->len);
6508 WARN_ON(!PageUptodate(page));
6512 void write_extent_buffer_chunk_tree_uuid(const struct extent_buffer *eb,
6517 assert_eb_page_uptodate(eb, eb->pages[0]);
6518 kaddr = page_address(eb->pages[0]) + get_eb_offset_in_page(eb, 0);
6519 memcpy(kaddr + offsetof(struct btrfs_header, chunk_tree_uuid), srcv,
6523 void write_extent_buffer_fsid(const struct extent_buffer *eb, const void *srcv)
6527 assert_eb_page_uptodate(eb, eb->pages[0]);
6528 kaddr = page_address(eb->pages[0]) + get_eb_offset_in_page(eb, 0);
6529 memcpy(kaddr + offsetof(struct btrfs_header, fsid), srcv,
6533 void write_extent_buffer(const struct extent_buffer *eb, const void *srcv,
6534 unsigned long start, unsigned long len)
6540 char *src = (char *)srcv;
6541 unsigned long i = get_eb_page_index(start);
6543 WARN_ON(test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags));
6545 if (check_eb_range(eb, start, len))
6548 offset = get_eb_offset_in_page(eb, start);
6551 page = eb->pages[i];
6552 assert_eb_page_uptodate(eb, page);
6554 cur = min(len, PAGE_SIZE - offset);
6555 kaddr = page_address(page);
6556 memcpy(kaddr + offset, src, cur);
6565 void memzero_extent_buffer(const struct extent_buffer *eb, unsigned long start,
6572 unsigned long i = get_eb_page_index(start);
6574 if (check_eb_range(eb, start, len))
6577 offset = get_eb_offset_in_page(eb, start);
6580 page = eb->pages[i];
6581 assert_eb_page_uptodate(eb, page);
6583 cur = min(len, PAGE_SIZE - offset);
6584 kaddr = page_address(page);
6585 memset(kaddr + offset, 0, cur);
6593 void copy_extent_buffer_full(const struct extent_buffer *dst,
6594 const struct extent_buffer *src)
6599 ASSERT(dst->len == src->len);
6601 if (dst->fs_info->sectorsize == PAGE_SIZE) {
6602 num_pages = num_extent_pages(dst);
6603 for (i = 0; i < num_pages; i++)
6604 copy_page(page_address(dst->pages[i]),
6605 page_address(src->pages[i]));
6607 size_t src_offset = get_eb_offset_in_page(src, 0);
6608 size_t dst_offset = get_eb_offset_in_page(dst, 0);
6610 ASSERT(src->fs_info->sectorsize < PAGE_SIZE);
6611 memcpy(page_address(dst->pages[0]) + dst_offset,
6612 page_address(src->pages[0]) + src_offset,
6617 void copy_extent_buffer(const struct extent_buffer *dst,
6618 const struct extent_buffer *src,
6619 unsigned long dst_offset, unsigned long src_offset,
6622 u64 dst_len = dst->len;
6627 unsigned long i = get_eb_page_index(dst_offset);
6629 if (check_eb_range(dst, dst_offset, len) ||
6630 check_eb_range(src, src_offset, len))
6633 WARN_ON(src->len != dst_len);
6635 offset = get_eb_offset_in_page(dst, dst_offset);
6638 page = dst->pages[i];
6639 assert_eb_page_uptodate(dst, page);
6641 cur = min(len, (unsigned long)(PAGE_SIZE - offset));
6643 kaddr = page_address(page);
6644 read_extent_buffer(src, kaddr + offset, src_offset, cur);
6654 * eb_bitmap_offset() - calculate the page and offset of the byte containing the
6656 * @eb: the extent buffer
6657 * @start: offset of the bitmap item in the extent buffer
6659 * @page_index: return index of the page in the extent buffer that contains the
6661 * @page_offset: return offset into the page given by page_index
6663 * This helper hides the ugliness of finding the byte in an extent buffer which
6664 * contains a given bit.
6666 static inline void eb_bitmap_offset(const struct extent_buffer *eb,
6667 unsigned long start, unsigned long nr,
6668 unsigned long *page_index,
6669 size_t *page_offset)
6671 size_t byte_offset = BIT_BYTE(nr);
6675 * The byte we want is the offset of the extent buffer + the offset of
6676 * the bitmap item in the extent buffer + the offset of the byte in the
6679 offset = start + offset_in_page(eb->start) + byte_offset;
6681 *page_index = offset >> PAGE_SHIFT;
6682 *page_offset = offset_in_page(offset);
6686 * extent_buffer_test_bit - determine whether a bit in a bitmap item is set
6687 * @eb: the extent buffer
6688 * @start: offset of the bitmap item in the extent buffer
6689 * @nr: bit number to test
6691 int extent_buffer_test_bit(const struct extent_buffer *eb, unsigned long start,
6699 eb_bitmap_offset(eb, start, nr, &i, &offset);
6700 page = eb->pages[i];
6701 assert_eb_page_uptodate(eb, page);
6702 kaddr = page_address(page);
6703 return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1)));
6707 * extent_buffer_bitmap_set - set an area of a bitmap
6708 * @eb: the extent buffer
6709 * @start: offset of the bitmap item in the extent buffer
6710 * @pos: bit number of the first bit
6711 * @len: number of bits to set
6713 void extent_buffer_bitmap_set(const struct extent_buffer *eb, unsigned long start,
6714 unsigned long pos, unsigned long len)
6720 const unsigned int size = pos + len;
6721 int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
6722 u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos);
6724 eb_bitmap_offset(eb, start, pos, &i, &offset);
6725 page = eb->pages[i];
6726 assert_eb_page_uptodate(eb, page);
6727 kaddr = page_address(page);
6729 while (len >= bits_to_set) {
6730 kaddr[offset] |= mask_to_set;
6732 bits_to_set = BITS_PER_BYTE;
6734 if (++offset >= PAGE_SIZE && len > 0) {
6736 page = eb->pages[++i];
6737 assert_eb_page_uptodate(eb, page);
6738 kaddr = page_address(page);
6742 mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
6743 kaddr[offset] |= mask_to_set;
6749 * extent_buffer_bitmap_clear - clear an area of a bitmap
6750 * @eb: the extent buffer
6751 * @start: offset of the bitmap item in the extent buffer
6752 * @pos: bit number of the first bit
6753 * @len: number of bits to clear
6755 void extent_buffer_bitmap_clear(const struct extent_buffer *eb,
6756 unsigned long start, unsigned long pos,
6763 const unsigned int size = pos + len;
6764 int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
6765 u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos);
6767 eb_bitmap_offset(eb, start, pos, &i, &offset);
6768 page = eb->pages[i];
6769 assert_eb_page_uptodate(eb, page);
6770 kaddr = page_address(page);
6772 while (len >= bits_to_clear) {
6773 kaddr[offset] &= ~mask_to_clear;
6774 len -= bits_to_clear;
6775 bits_to_clear = BITS_PER_BYTE;
6777 if (++offset >= PAGE_SIZE && len > 0) {
6779 page = eb->pages[++i];
6780 assert_eb_page_uptodate(eb, page);
6781 kaddr = page_address(page);
6785 mask_to_clear &= BITMAP_LAST_BYTE_MASK(size);
6786 kaddr[offset] &= ~mask_to_clear;
6790 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
6792 unsigned long distance = (src > dst) ? src - dst : dst - src;
6793 return distance < len;
6796 static void copy_pages(struct page *dst_page, struct page *src_page,
6797 unsigned long dst_off, unsigned long src_off,
6800 char *dst_kaddr = page_address(dst_page);
6802 int must_memmove = 0;
6804 if (dst_page != src_page) {
6805 src_kaddr = page_address(src_page);
6807 src_kaddr = dst_kaddr;
6808 if (areas_overlap(src_off, dst_off, len))
6813 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
6815 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
6818 void memcpy_extent_buffer(const struct extent_buffer *dst,
6819 unsigned long dst_offset, unsigned long src_offset,
6823 size_t dst_off_in_page;
6824 size_t src_off_in_page;
6825 unsigned long dst_i;
6826 unsigned long src_i;
6828 if (check_eb_range(dst, dst_offset, len) ||
6829 check_eb_range(dst, src_offset, len))
6833 dst_off_in_page = get_eb_offset_in_page(dst, dst_offset);
6834 src_off_in_page = get_eb_offset_in_page(dst, src_offset);
6836 dst_i = get_eb_page_index(dst_offset);
6837 src_i = get_eb_page_index(src_offset);
6839 cur = min(len, (unsigned long)(PAGE_SIZE -
6841 cur = min_t(unsigned long, cur,
6842 (unsigned long)(PAGE_SIZE - dst_off_in_page));
6844 copy_pages(dst->pages[dst_i], dst->pages[src_i],
6845 dst_off_in_page, src_off_in_page, cur);
6853 void memmove_extent_buffer(const struct extent_buffer *dst,
6854 unsigned long dst_offset, unsigned long src_offset,
6858 size_t dst_off_in_page;
6859 size_t src_off_in_page;
6860 unsigned long dst_end = dst_offset + len - 1;
6861 unsigned long src_end = src_offset + len - 1;
6862 unsigned long dst_i;
6863 unsigned long src_i;
6865 if (check_eb_range(dst, dst_offset, len) ||
6866 check_eb_range(dst, src_offset, len))
6868 if (dst_offset < src_offset) {
6869 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
6873 dst_i = get_eb_page_index(dst_end);
6874 src_i = get_eb_page_index(src_end);
6876 dst_off_in_page = get_eb_offset_in_page(dst, dst_end);
6877 src_off_in_page = get_eb_offset_in_page(dst, src_end);
6879 cur = min_t(unsigned long, len, src_off_in_page + 1);
6880 cur = min(cur, dst_off_in_page + 1);
6881 copy_pages(dst->pages[dst_i], dst->pages[src_i],
6882 dst_off_in_page - cur + 1,
6883 src_off_in_page - cur + 1, cur);
6891 static struct extent_buffer *get_next_extent_buffer(
6892 struct btrfs_fs_info *fs_info, struct page *page, u64 bytenr)
6894 struct extent_buffer *gang[BTRFS_SUBPAGE_BITMAP_SIZE];
6895 struct extent_buffer *found = NULL;
6896 u64 page_start = page_offset(page);
6900 ASSERT(in_range(bytenr, page_start, PAGE_SIZE));
6901 ASSERT(PAGE_SIZE / fs_info->nodesize <= BTRFS_SUBPAGE_BITMAP_SIZE);
6902 lockdep_assert_held(&fs_info->buffer_lock);
6904 ret = radix_tree_gang_lookup(&fs_info->buffer_radix, (void **)gang,
6905 bytenr >> fs_info->sectorsize_bits,
6906 PAGE_SIZE / fs_info->nodesize);
6907 for (i = 0; i < ret; i++) {
6908 /* Already beyond page end */
6909 if (gang[i]->start >= page_start + PAGE_SIZE)
6912 if (gang[i]->start >= bytenr) {
6920 static int try_release_subpage_extent_buffer(struct page *page)
6922 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
6923 u64 cur = page_offset(page);
6924 const u64 end = page_offset(page) + PAGE_SIZE;
6928 struct extent_buffer *eb = NULL;
6931 * Unlike try_release_extent_buffer() which uses page->private
6932 * to grab buffer, for subpage case we rely on radix tree, thus
6933 * we need to ensure radix tree consistency.
6935 * We also want an atomic snapshot of the radix tree, thus go
6936 * with spinlock rather than RCU.
6938 spin_lock(&fs_info->buffer_lock);
6939 eb = get_next_extent_buffer(fs_info, page, cur);
6941 /* No more eb in the page range after or at cur */
6942 spin_unlock(&fs_info->buffer_lock);
6945 cur = eb->start + eb->len;
6948 * The same as try_release_extent_buffer(), to ensure the eb
6949 * won't disappear out from under us.
6951 spin_lock(&eb->refs_lock);
6952 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
6953 spin_unlock(&eb->refs_lock);
6954 spin_unlock(&fs_info->buffer_lock);
6957 spin_unlock(&fs_info->buffer_lock);
6960 * If tree ref isn't set then we know the ref on this eb is a
6961 * real ref, so just return, this eb will likely be freed soon
6964 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
6965 spin_unlock(&eb->refs_lock);
6970 * Here we don't care about the return value, we will always
6971 * check the page private at the end. And
6972 * release_extent_buffer() will release the refs_lock.
6974 release_extent_buffer(eb);
6977 * Finally to check if we have cleared page private, as if we have
6978 * released all ebs in the page, the page private should be cleared now.
6980 spin_lock(&page->mapping->private_lock);
6981 if (!PagePrivate(page))
6985 spin_unlock(&page->mapping->private_lock);
6990 int try_release_extent_buffer(struct page *page)
6992 struct extent_buffer *eb;
6994 if (btrfs_sb(page->mapping->host->i_sb)->sectorsize < PAGE_SIZE)
6995 return try_release_subpage_extent_buffer(page);
6998 * We need to make sure nobody is changing page->private, as we rely on
6999 * page->private as the pointer to extent buffer.
7001 spin_lock(&page->mapping->private_lock);
7002 if (!PagePrivate(page)) {
7003 spin_unlock(&page->mapping->private_lock);
7007 eb = (struct extent_buffer *)page->private;
7011 * This is a little awful but should be ok, we need to make sure that
7012 * the eb doesn't disappear out from under us while we're looking at
7015 spin_lock(&eb->refs_lock);
7016 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
7017 spin_unlock(&eb->refs_lock);
7018 spin_unlock(&page->mapping->private_lock);
7021 spin_unlock(&page->mapping->private_lock);
7024 * If tree ref isn't set then we know the ref on this eb is a real ref,
7025 * so just return, this page will likely be freed soon anyway.
7027 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
7028 spin_unlock(&eb->refs_lock);
7032 return release_extent_buffer(eb);
7036 * btrfs_readahead_tree_block - attempt to readahead a child block
7037 * @fs_info: the fs_info
7038 * @bytenr: bytenr to read
7039 * @owner_root: objectid of the root that owns this eb
7040 * @gen: generation for the uptodate check, can be 0
7041 * @level: level for the eb
7043 * Attempt to readahead a tree block at @bytenr. If @gen is 0 then we do a
7044 * normal uptodate check of the eb, without checking the generation. If we have
7045 * to read the block we will not block on anything.
7047 void btrfs_readahead_tree_block(struct btrfs_fs_info *fs_info,
7048 u64 bytenr, u64 owner_root, u64 gen, int level)
7050 struct extent_buffer *eb;
7053 eb = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level);
7057 if (btrfs_buffer_uptodate(eb, gen, 1)) {
7058 free_extent_buffer(eb);
7062 ret = read_extent_buffer_pages(eb, WAIT_NONE, 0);
7064 free_extent_buffer_stale(eb);
7066 free_extent_buffer(eb);
7070 * btrfs_readahead_node_child - readahead a node's child block
7071 * @node: parent node we're reading from
7072 * @slot: slot in the parent node for the child we want to read
7074 * A helper for btrfs_readahead_tree_block, we simply read the bytenr pointed at
7075 * the slot in the node provided.
7077 void btrfs_readahead_node_child(struct extent_buffer *node, int slot)
7079 btrfs_readahead_tree_block(node->fs_info,
7080 btrfs_node_blockptr(node, slot),
7081 btrfs_header_owner(node),
7082 btrfs_node_ptr_generation(node, slot),
7083 btrfs_header_level(node) - 1);