1 // SPDX-License-Identifier: GPL-2.0
3 #include <linux/bitops.h>
4 #include <linux/slab.h>
7 #include <linux/pagemap.h>
8 #include <linux/page-flags.h>
9 #include <linux/spinlock.h>
10 #include <linux/blkdev.h>
11 #include <linux/swap.h>
12 #include <linux/writeback.h>
13 #include <linux/pagevec.h>
14 #include <linux/prefetch.h>
15 #include <linux/cleancache.h>
17 #include "extent_io.h"
18 #include "extent-io-tree.h"
19 #include "extent_map.h"
21 #include "btrfs_inode.h"
23 #include "check-integrity.h"
25 #include "rcu-string.h"
30 #include "block-group.h"
32 static struct kmem_cache *extent_state_cache;
33 static struct kmem_cache *extent_buffer_cache;
34 static struct bio_set btrfs_bioset;
36 static inline bool extent_state_in_tree(const struct extent_state *state)
38 return !RB_EMPTY_NODE(&state->rb_node);
41 #ifdef CONFIG_BTRFS_DEBUG
42 static LIST_HEAD(states);
43 static DEFINE_SPINLOCK(leak_lock);
45 static inline void btrfs_leak_debug_add(spinlock_t *lock,
46 struct list_head *new,
47 struct list_head *head)
51 spin_lock_irqsave(lock, flags);
53 spin_unlock_irqrestore(lock, flags);
56 static inline void btrfs_leak_debug_del(spinlock_t *lock,
57 struct list_head *entry)
61 spin_lock_irqsave(lock, flags);
63 spin_unlock_irqrestore(lock, flags);
66 void btrfs_extent_buffer_leak_debug_check(struct btrfs_fs_info *fs_info)
68 struct extent_buffer *eb;
72 * If we didn't get into open_ctree our allocated_ebs will not be
73 * initialized, so just skip this.
75 if (!fs_info->allocated_ebs.next)
78 spin_lock_irqsave(&fs_info->eb_leak_lock, flags);
79 while (!list_empty(&fs_info->allocated_ebs)) {
80 eb = list_first_entry(&fs_info->allocated_ebs,
81 struct extent_buffer, leak_list);
83 "BTRFS: buffer leak start %llu len %lu refs %d bflags %lu owner %llu\n",
84 eb->start, eb->len, atomic_read(&eb->refs), eb->bflags,
85 btrfs_header_owner(eb));
86 list_del(&eb->leak_list);
87 kmem_cache_free(extent_buffer_cache, eb);
89 spin_unlock_irqrestore(&fs_info->eb_leak_lock, flags);
92 static inline void btrfs_extent_state_leak_debug_check(void)
94 struct extent_state *state;
96 while (!list_empty(&states)) {
97 state = list_entry(states.next, struct extent_state, leak_list);
98 pr_err("BTRFS: state leak: start %llu end %llu state %u in tree %d refs %d\n",
99 state->start, state->end, state->state,
100 extent_state_in_tree(state),
101 refcount_read(&state->refs));
102 list_del(&state->leak_list);
103 kmem_cache_free(extent_state_cache, state);
107 #define btrfs_debug_check_extent_io_range(tree, start, end) \
108 __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end))
109 static inline void __btrfs_debug_check_extent_io_range(const char *caller,
110 struct extent_io_tree *tree, u64 start, u64 end)
112 struct inode *inode = tree->private_data;
115 if (!inode || !is_data_inode(inode))
118 isize = i_size_read(inode);
119 if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) {
120 btrfs_debug_rl(BTRFS_I(inode)->root->fs_info,
121 "%s: ino %llu isize %llu odd range [%llu,%llu]",
122 caller, btrfs_ino(BTRFS_I(inode)), isize, start, end);
126 #define btrfs_leak_debug_add(lock, new, head) do {} while (0)
127 #define btrfs_leak_debug_del(lock, entry) do {} while (0)
128 #define btrfs_extent_state_leak_debug_check() do {} while (0)
129 #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0)
135 struct rb_node rb_node;
138 struct extent_page_data {
139 struct btrfs_bio_ctrl bio_ctrl;
140 /* tells writepage not to lock the state bits for this range
141 * it still does the unlocking
143 unsigned int extent_locked:1;
145 /* tells the submit_bio code to use REQ_SYNC */
146 unsigned int sync_io:1;
149 static int add_extent_changeset(struct extent_state *state, u32 bits,
150 struct extent_changeset *changeset,
157 if (set && (state->state & bits) == bits)
159 if (!set && (state->state & bits) == 0)
161 changeset->bytes_changed += state->end - state->start + 1;
162 ret = ulist_add(&changeset->range_changed, state->start, state->end,
167 int __must_check submit_one_bio(struct bio *bio, int mirror_num,
168 unsigned long bio_flags)
170 blk_status_t ret = 0;
171 struct extent_io_tree *tree = bio->bi_private;
173 bio->bi_private = NULL;
175 if (is_data_inode(tree->private_data))
176 ret = btrfs_submit_data_bio(tree->private_data, bio, mirror_num,
179 ret = btrfs_submit_metadata_bio(tree->private_data, bio,
180 mirror_num, bio_flags);
182 return blk_status_to_errno(ret);
185 /* Cleanup unsubmitted bios */
186 static void end_write_bio(struct extent_page_data *epd, int ret)
188 struct bio *bio = epd->bio_ctrl.bio;
191 bio->bi_status = errno_to_blk_status(ret);
193 epd->bio_ctrl.bio = NULL;
198 * Submit bio from extent page data via submit_one_bio
200 * Return 0 if everything is OK.
201 * Return <0 for error.
203 static int __must_check flush_write_bio(struct extent_page_data *epd)
206 struct bio *bio = epd->bio_ctrl.bio;
209 ret = submit_one_bio(bio, 0, 0);
211 * Clean up of epd->bio is handled by its endio function.
212 * And endio is either triggered by successful bio execution
213 * or the error handler of submit bio hook.
214 * So at this point, no matter what happened, we don't need
215 * to clean up epd->bio.
217 epd->bio_ctrl.bio = NULL;
222 int __init extent_state_cache_init(void)
224 extent_state_cache = kmem_cache_create("btrfs_extent_state",
225 sizeof(struct extent_state), 0,
226 SLAB_MEM_SPREAD, NULL);
227 if (!extent_state_cache)
232 int __init extent_io_init(void)
234 extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer",
235 sizeof(struct extent_buffer), 0,
236 SLAB_MEM_SPREAD, NULL);
237 if (!extent_buffer_cache)
240 if (bioset_init(&btrfs_bioset, BIO_POOL_SIZE,
241 offsetof(struct btrfs_io_bio, bio),
243 goto free_buffer_cache;
245 if (bioset_integrity_create(&btrfs_bioset, BIO_POOL_SIZE))
251 bioset_exit(&btrfs_bioset);
254 kmem_cache_destroy(extent_buffer_cache);
255 extent_buffer_cache = NULL;
259 void __cold extent_state_cache_exit(void)
261 btrfs_extent_state_leak_debug_check();
262 kmem_cache_destroy(extent_state_cache);
265 void __cold extent_io_exit(void)
268 * Make sure all delayed rcu free are flushed before we
272 kmem_cache_destroy(extent_buffer_cache);
273 bioset_exit(&btrfs_bioset);
277 * For the file_extent_tree, we want to hold the inode lock when we lookup and
278 * update the disk_i_size, but lockdep will complain because our io_tree we hold
279 * the tree lock and get the inode lock when setting delalloc. These two things
280 * are unrelated, so make a class for the file_extent_tree so we don't get the
281 * two locking patterns mixed up.
283 static struct lock_class_key file_extent_tree_class;
285 void extent_io_tree_init(struct btrfs_fs_info *fs_info,
286 struct extent_io_tree *tree, unsigned int owner,
289 tree->fs_info = fs_info;
290 tree->state = RB_ROOT;
291 tree->dirty_bytes = 0;
292 spin_lock_init(&tree->lock);
293 tree->private_data = private_data;
295 if (owner == IO_TREE_INODE_FILE_EXTENT)
296 lockdep_set_class(&tree->lock, &file_extent_tree_class);
299 void extent_io_tree_release(struct extent_io_tree *tree)
301 spin_lock(&tree->lock);
303 * Do a single barrier for the waitqueue_active check here, the state
304 * of the waitqueue should not change once extent_io_tree_release is
308 while (!RB_EMPTY_ROOT(&tree->state)) {
309 struct rb_node *node;
310 struct extent_state *state;
312 node = rb_first(&tree->state);
313 state = rb_entry(node, struct extent_state, rb_node);
314 rb_erase(&state->rb_node, &tree->state);
315 RB_CLEAR_NODE(&state->rb_node);
317 * btree io trees aren't supposed to have tasks waiting for
318 * changes in the flags of extent states ever.
320 ASSERT(!waitqueue_active(&state->wq));
321 free_extent_state(state);
323 cond_resched_lock(&tree->lock);
325 spin_unlock(&tree->lock);
328 static struct extent_state *alloc_extent_state(gfp_t mask)
330 struct extent_state *state;
333 * The given mask might be not appropriate for the slab allocator,
334 * drop the unsupported bits
336 mask &= ~(__GFP_DMA32|__GFP_HIGHMEM);
337 state = kmem_cache_alloc(extent_state_cache, mask);
341 state->failrec = NULL;
342 RB_CLEAR_NODE(&state->rb_node);
343 btrfs_leak_debug_add(&leak_lock, &state->leak_list, &states);
344 refcount_set(&state->refs, 1);
345 init_waitqueue_head(&state->wq);
346 trace_alloc_extent_state(state, mask, _RET_IP_);
350 void free_extent_state(struct extent_state *state)
354 if (refcount_dec_and_test(&state->refs)) {
355 WARN_ON(extent_state_in_tree(state));
356 btrfs_leak_debug_del(&leak_lock, &state->leak_list);
357 trace_free_extent_state(state, _RET_IP_);
358 kmem_cache_free(extent_state_cache, state);
362 static struct rb_node *tree_insert(struct rb_root *root,
363 struct rb_node *search_start,
365 struct rb_node *node,
366 struct rb_node ***p_in,
367 struct rb_node **parent_in)
370 struct rb_node *parent = NULL;
371 struct tree_entry *entry;
373 if (p_in && parent_in) {
379 p = search_start ? &search_start : &root->rb_node;
382 entry = rb_entry(parent, struct tree_entry, rb_node);
384 if (offset < entry->start)
386 else if (offset > entry->end)
393 rb_link_node(node, parent, p);
394 rb_insert_color(node, root);
399 * Search @tree for an entry that contains @offset. Such entry would have
400 * entry->start <= offset && entry->end >= offset.
402 * @tree: the tree to search
403 * @offset: offset that should fall within an entry in @tree
404 * @next_ret: pointer to the first entry whose range ends after @offset
405 * @prev_ret: pointer to the first entry whose range begins before @offset
406 * @p_ret: pointer where new node should be anchored (used when inserting an
408 * @parent_ret: points to entry which would have been the parent of the entry,
411 * This function returns a pointer to the entry that contains @offset byte
412 * address. If no such entry exists, then NULL is returned and the other
413 * pointer arguments to the function are filled, otherwise the found entry is
414 * returned and other pointers are left untouched.
416 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
417 struct rb_node **next_ret,
418 struct rb_node **prev_ret,
419 struct rb_node ***p_ret,
420 struct rb_node **parent_ret)
422 struct rb_root *root = &tree->state;
423 struct rb_node **n = &root->rb_node;
424 struct rb_node *prev = NULL;
425 struct rb_node *orig_prev = NULL;
426 struct tree_entry *entry;
427 struct tree_entry *prev_entry = NULL;
431 entry = rb_entry(prev, struct tree_entry, rb_node);
434 if (offset < entry->start)
436 else if (offset > entry->end)
449 while (prev && offset > prev_entry->end) {
450 prev = rb_next(prev);
451 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
458 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
459 while (prev && offset < prev_entry->start) {
460 prev = rb_prev(prev);
461 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
468 static inline struct rb_node *
469 tree_search_for_insert(struct extent_io_tree *tree,
471 struct rb_node ***p_ret,
472 struct rb_node **parent_ret)
474 struct rb_node *next= NULL;
477 ret = __etree_search(tree, offset, &next, NULL, p_ret, parent_ret);
483 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
486 return tree_search_for_insert(tree, offset, NULL, NULL);
490 * utility function to look for merge candidates inside a given range.
491 * Any extents with matching state are merged together into a single
492 * extent in the tree. Extents with EXTENT_IO in their state field
493 * are not merged because the end_io handlers need to be able to do
494 * operations on them without sleeping (or doing allocations/splits).
496 * This should be called with the tree lock held.
498 static void merge_state(struct extent_io_tree *tree,
499 struct extent_state *state)
501 struct extent_state *other;
502 struct rb_node *other_node;
504 if (state->state & (EXTENT_LOCKED | EXTENT_BOUNDARY))
507 other_node = rb_prev(&state->rb_node);
509 other = rb_entry(other_node, struct extent_state, rb_node);
510 if (other->end == state->start - 1 &&
511 other->state == state->state) {
512 if (tree->private_data &&
513 is_data_inode(tree->private_data))
514 btrfs_merge_delalloc_extent(tree->private_data,
516 state->start = other->start;
517 rb_erase(&other->rb_node, &tree->state);
518 RB_CLEAR_NODE(&other->rb_node);
519 free_extent_state(other);
522 other_node = rb_next(&state->rb_node);
524 other = rb_entry(other_node, struct extent_state, rb_node);
525 if (other->start == state->end + 1 &&
526 other->state == state->state) {
527 if (tree->private_data &&
528 is_data_inode(tree->private_data))
529 btrfs_merge_delalloc_extent(tree->private_data,
531 state->end = other->end;
532 rb_erase(&other->rb_node, &tree->state);
533 RB_CLEAR_NODE(&other->rb_node);
534 free_extent_state(other);
539 static void set_state_bits(struct extent_io_tree *tree,
540 struct extent_state *state, u32 *bits,
541 struct extent_changeset *changeset);
544 * insert an extent_state struct into the tree. 'bits' are set on the
545 * struct before it is inserted.
547 * This may return -EEXIST if the extent is already there, in which case the
548 * state struct is freed.
550 * The tree lock is not taken internally. This is a utility function and
551 * probably isn't what you want to call (see set/clear_extent_bit).
553 static int insert_state(struct extent_io_tree *tree,
554 struct extent_state *state, u64 start, u64 end,
556 struct rb_node **parent,
557 u32 *bits, struct extent_changeset *changeset)
559 struct rb_node *node;
562 btrfs_err(tree->fs_info,
563 "insert state: end < start %llu %llu", end, start);
566 state->start = start;
569 set_state_bits(tree, state, bits, changeset);
571 node = tree_insert(&tree->state, NULL, end, &state->rb_node, p, parent);
573 struct extent_state *found;
574 found = rb_entry(node, struct extent_state, rb_node);
575 btrfs_err(tree->fs_info,
576 "found node %llu %llu on insert of %llu %llu",
577 found->start, found->end, start, end);
580 merge_state(tree, state);
585 * split a given extent state struct in two, inserting the preallocated
586 * struct 'prealloc' as the newly created second half. 'split' indicates an
587 * offset inside 'orig' where it should be split.
590 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
591 * are two extent state structs in the tree:
592 * prealloc: [orig->start, split - 1]
593 * orig: [ split, orig->end ]
595 * The tree locks are not taken by this function. They need to be held
598 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
599 struct extent_state *prealloc, u64 split)
601 struct rb_node *node;
603 if (tree->private_data && is_data_inode(tree->private_data))
604 btrfs_split_delalloc_extent(tree->private_data, orig, split);
606 prealloc->start = orig->start;
607 prealloc->end = split - 1;
608 prealloc->state = orig->state;
611 node = tree_insert(&tree->state, &orig->rb_node, prealloc->end,
612 &prealloc->rb_node, NULL, NULL);
614 free_extent_state(prealloc);
620 static struct extent_state *next_state(struct extent_state *state)
622 struct rb_node *next = rb_next(&state->rb_node);
624 return rb_entry(next, struct extent_state, rb_node);
630 * utility function to clear some bits in an extent state struct.
631 * it will optionally wake up anyone waiting on this state (wake == 1).
633 * If no bits are set on the state struct after clearing things, the
634 * struct is freed and removed from the tree
636 static struct extent_state *clear_state_bit(struct extent_io_tree *tree,
637 struct extent_state *state,
639 struct extent_changeset *changeset)
641 struct extent_state *next;
642 u32 bits_to_clear = *bits & ~EXTENT_CTLBITS;
645 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
646 u64 range = state->end - state->start + 1;
647 WARN_ON(range > tree->dirty_bytes);
648 tree->dirty_bytes -= range;
651 if (tree->private_data && is_data_inode(tree->private_data))
652 btrfs_clear_delalloc_extent(tree->private_data, state, bits);
654 ret = add_extent_changeset(state, bits_to_clear, changeset, 0);
656 state->state &= ~bits_to_clear;
659 if (state->state == 0) {
660 next = next_state(state);
661 if (extent_state_in_tree(state)) {
662 rb_erase(&state->rb_node, &tree->state);
663 RB_CLEAR_NODE(&state->rb_node);
664 free_extent_state(state);
669 merge_state(tree, state);
670 next = next_state(state);
675 static struct extent_state *
676 alloc_extent_state_atomic(struct extent_state *prealloc)
679 prealloc = alloc_extent_state(GFP_ATOMIC);
684 static void extent_io_tree_panic(struct extent_io_tree *tree, int err)
686 btrfs_panic(tree->fs_info, err,
687 "locking error: extent tree was modified by another thread while locked");
691 * clear some bits on a range in the tree. This may require splitting
692 * or inserting elements in the tree, so the gfp mask is used to
693 * indicate which allocations or sleeping are allowed.
695 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
696 * the given range from the tree regardless of state (ie for truncate).
698 * the range [start, end] is inclusive.
700 * This takes the tree lock, and returns 0 on success and < 0 on error.
702 int __clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
703 u32 bits, int wake, int delete,
704 struct extent_state **cached_state,
705 gfp_t mask, struct extent_changeset *changeset)
707 struct extent_state *state;
708 struct extent_state *cached;
709 struct extent_state *prealloc = NULL;
710 struct rb_node *node;
715 btrfs_debug_check_extent_io_range(tree, start, end);
716 trace_btrfs_clear_extent_bit(tree, start, end - start + 1, bits);
718 if (bits & EXTENT_DELALLOC)
719 bits |= EXTENT_NORESERVE;
722 bits |= ~EXTENT_CTLBITS;
724 if (bits & (EXTENT_LOCKED | EXTENT_BOUNDARY))
727 if (!prealloc && gfpflags_allow_blocking(mask)) {
729 * Don't care for allocation failure here because we might end
730 * up not needing the pre-allocated extent state at all, which
731 * is the case if we only have in the tree extent states that
732 * cover our input range and don't cover too any other range.
733 * If we end up needing a new extent state we allocate it later.
735 prealloc = alloc_extent_state(mask);
738 spin_lock(&tree->lock);
740 cached = *cached_state;
743 *cached_state = NULL;
747 if (cached && extent_state_in_tree(cached) &&
748 cached->start <= start && cached->end > start) {
750 refcount_dec(&cached->refs);
755 free_extent_state(cached);
758 * this search will find the extents that end after
761 node = tree_search(tree, start);
764 state = rb_entry(node, struct extent_state, rb_node);
766 if (state->start > end)
768 WARN_ON(state->end < start);
769 last_end = state->end;
771 /* the state doesn't have the wanted bits, go ahead */
772 if (!(state->state & bits)) {
773 state = next_state(state);
778 * | ---- desired range ---- |
780 * | ------------- state -------------- |
782 * We need to split the extent we found, and may flip
783 * bits on second half.
785 * If the extent we found extends past our range, we
786 * just split and search again. It'll get split again
787 * the next time though.
789 * If the extent we found is inside our range, we clear
790 * the desired bit on it.
793 if (state->start < start) {
794 prealloc = alloc_extent_state_atomic(prealloc);
796 err = split_state(tree, state, prealloc, start);
798 extent_io_tree_panic(tree, err);
803 if (state->end <= end) {
804 state = clear_state_bit(tree, state, &bits, wake,
811 * | ---- desired range ---- |
813 * We need to split the extent, and clear the bit
816 if (state->start <= end && state->end > end) {
817 prealloc = alloc_extent_state_atomic(prealloc);
819 err = split_state(tree, state, prealloc, end + 1);
821 extent_io_tree_panic(tree, err);
826 clear_state_bit(tree, prealloc, &bits, wake, changeset);
832 state = clear_state_bit(tree, state, &bits, wake, changeset);
834 if (last_end == (u64)-1)
836 start = last_end + 1;
837 if (start <= end && state && !need_resched())
843 spin_unlock(&tree->lock);
844 if (gfpflags_allow_blocking(mask))
849 spin_unlock(&tree->lock);
851 free_extent_state(prealloc);
857 static void wait_on_state(struct extent_io_tree *tree,
858 struct extent_state *state)
859 __releases(tree->lock)
860 __acquires(tree->lock)
863 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
864 spin_unlock(&tree->lock);
866 spin_lock(&tree->lock);
867 finish_wait(&state->wq, &wait);
871 * waits for one or more bits to clear on a range in the state tree.
872 * The range [start, end] is inclusive.
873 * The tree lock is taken by this function
875 static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
878 struct extent_state *state;
879 struct rb_node *node;
881 btrfs_debug_check_extent_io_range(tree, start, end);
883 spin_lock(&tree->lock);
887 * this search will find all the extents that end after
890 node = tree_search(tree, start);
895 state = rb_entry(node, struct extent_state, rb_node);
897 if (state->start > end)
900 if (state->state & bits) {
901 start = state->start;
902 refcount_inc(&state->refs);
903 wait_on_state(tree, state);
904 free_extent_state(state);
907 start = state->end + 1;
912 if (!cond_resched_lock(&tree->lock)) {
913 node = rb_next(node);
918 spin_unlock(&tree->lock);
921 static void set_state_bits(struct extent_io_tree *tree,
922 struct extent_state *state,
923 u32 *bits, struct extent_changeset *changeset)
925 u32 bits_to_set = *bits & ~EXTENT_CTLBITS;
928 if (tree->private_data && is_data_inode(tree->private_data))
929 btrfs_set_delalloc_extent(tree->private_data, state, bits);
931 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
932 u64 range = state->end - state->start + 1;
933 tree->dirty_bytes += range;
935 ret = add_extent_changeset(state, bits_to_set, changeset, 1);
937 state->state |= bits_to_set;
940 static void cache_state_if_flags(struct extent_state *state,
941 struct extent_state **cached_ptr,
944 if (cached_ptr && !(*cached_ptr)) {
945 if (!flags || (state->state & flags)) {
947 refcount_inc(&state->refs);
952 static void cache_state(struct extent_state *state,
953 struct extent_state **cached_ptr)
955 return cache_state_if_flags(state, cached_ptr,
956 EXTENT_LOCKED | EXTENT_BOUNDARY);
960 * set some bits on a range in the tree. This may require allocations or
961 * sleeping, so the gfp mask is used to indicate what is allowed.
963 * If any of the exclusive bits are set, this will fail with -EEXIST if some
964 * part of the range already has the desired bits set. The start of the
965 * existing range is returned in failed_start in this case.
967 * [start, end] is inclusive This takes the tree lock.
969 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, u32 bits,
970 u32 exclusive_bits, u64 *failed_start,
971 struct extent_state **cached_state, gfp_t mask,
972 struct extent_changeset *changeset)
974 struct extent_state *state;
975 struct extent_state *prealloc = NULL;
976 struct rb_node *node;
978 struct rb_node *parent;
983 btrfs_debug_check_extent_io_range(tree, start, end);
984 trace_btrfs_set_extent_bit(tree, start, end - start + 1, bits);
987 ASSERT(failed_start);
989 ASSERT(failed_start == NULL);
991 if (!prealloc && gfpflags_allow_blocking(mask)) {
993 * Don't care for allocation failure here because we might end
994 * up not needing the pre-allocated extent state at all, which
995 * is the case if we only have in the tree extent states that
996 * cover our input range and don't cover too any other range.
997 * If we end up needing a new extent state we allocate it later.
999 prealloc = alloc_extent_state(mask);
1002 spin_lock(&tree->lock);
1003 if (cached_state && *cached_state) {
1004 state = *cached_state;
1005 if (state->start <= start && state->end > start &&
1006 extent_state_in_tree(state)) {
1007 node = &state->rb_node;
1012 * this search will find all the extents that end after
1015 node = tree_search_for_insert(tree, start, &p, &parent);
1017 prealloc = alloc_extent_state_atomic(prealloc);
1019 err = insert_state(tree, prealloc, start, end,
1020 &p, &parent, &bits, changeset);
1022 extent_io_tree_panic(tree, err);
1024 cache_state(prealloc, cached_state);
1028 state = rb_entry(node, struct extent_state, rb_node);
1030 last_start = state->start;
1031 last_end = state->end;
1034 * | ---- desired range ---- |
1037 * Just lock what we found and keep going
1039 if (state->start == start && state->end <= end) {
1040 if (state->state & exclusive_bits) {
1041 *failed_start = state->start;
1046 set_state_bits(tree, state, &bits, changeset);
1047 cache_state(state, cached_state);
1048 merge_state(tree, state);
1049 if (last_end == (u64)-1)
1051 start = last_end + 1;
1052 state = next_state(state);
1053 if (start < end && state && state->start == start &&
1060 * | ---- desired range ---- |
1063 * | ------------- state -------------- |
1065 * We need to split the extent we found, and may flip bits on
1068 * If the extent we found extends past our
1069 * range, we just split and search again. It'll get split
1070 * again the next time though.
1072 * If the extent we found is inside our range, we set the
1073 * desired bit on it.
1075 if (state->start < start) {
1076 if (state->state & exclusive_bits) {
1077 *failed_start = start;
1083 * If this extent already has all the bits we want set, then
1084 * skip it, not necessary to split it or do anything with it.
1086 if ((state->state & bits) == bits) {
1087 start = state->end + 1;
1088 cache_state(state, cached_state);
1092 prealloc = alloc_extent_state_atomic(prealloc);
1094 err = split_state(tree, state, prealloc, start);
1096 extent_io_tree_panic(tree, err);
1101 if (state->end <= end) {
1102 set_state_bits(tree, state, &bits, changeset);
1103 cache_state(state, cached_state);
1104 merge_state(tree, state);
1105 if (last_end == (u64)-1)
1107 start = last_end + 1;
1108 state = next_state(state);
1109 if (start < end && state && state->start == start &&
1116 * | ---- desired range ---- |
1117 * | state | or | state |
1119 * There's a hole, we need to insert something in it and
1120 * ignore the extent we found.
1122 if (state->start > start) {
1124 if (end < last_start)
1127 this_end = last_start - 1;
1129 prealloc = alloc_extent_state_atomic(prealloc);
1133 * Avoid to free 'prealloc' if it can be merged with
1136 err = insert_state(tree, prealloc, start, this_end,
1137 NULL, NULL, &bits, changeset);
1139 extent_io_tree_panic(tree, err);
1141 cache_state(prealloc, cached_state);
1143 start = this_end + 1;
1147 * | ---- desired range ---- |
1149 * We need to split the extent, and set the bit
1152 if (state->start <= end && state->end > end) {
1153 if (state->state & exclusive_bits) {
1154 *failed_start = start;
1159 prealloc = alloc_extent_state_atomic(prealloc);
1161 err = split_state(tree, state, prealloc, end + 1);
1163 extent_io_tree_panic(tree, err);
1165 set_state_bits(tree, prealloc, &bits, changeset);
1166 cache_state(prealloc, cached_state);
1167 merge_state(tree, prealloc);
1175 spin_unlock(&tree->lock);
1176 if (gfpflags_allow_blocking(mask))
1181 spin_unlock(&tree->lock);
1183 free_extent_state(prealloc);
1190 * convert_extent_bit - convert all bits in a given range from one bit to
1192 * @tree: the io tree to search
1193 * @start: the start offset in bytes
1194 * @end: the end offset in bytes (inclusive)
1195 * @bits: the bits to set in this range
1196 * @clear_bits: the bits to clear in this range
1197 * @cached_state: state that we're going to cache
1199 * This will go through and set bits for the given range. If any states exist
1200 * already in this range they are set with the given bit and cleared of the
1201 * clear_bits. This is only meant to be used by things that are mergeable, ie
1202 * converting from say DELALLOC to DIRTY. This is not meant to be used with
1203 * boundary bits like LOCK.
1205 * All allocations are done with GFP_NOFS.
1207 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1208 u32 bits, u32 clear_bits,
1209 struct extent_state **cached_state)
1211 struct extent_state *state;
1212 struct extent_state *prealloc = NULL;
1213 struct rb_node *node;
1215 struct rb_node *parent;
1219 bool first_iteration = true;
1221 btrfs_debug_check_extent_io_range(tree, start, end);
1222 trace_btrfs_convert_extent_bit(tree, start, end - start + 1, bits,
1228 * Best effort, don't worry if extent state allocation fails
1229 * here for the first iteration. We might have a cached state
1230 * that matches exactly the target range, in which case no
1231 * extent state allocations are needed. We'll only know this
1232 * after locking the tree.
1234 prealloc = alloc_extent_state(GFP_NOFS);
1235 if (!prealloc && !first_iteration)
1239 spin_lock(&tree->lock);
1240 if (cached_state && *cached_state) {
1241 state = *cached_state;
1242 if (state->start <= start && state->end > start &&
1243 extent_state_in_tree(state)) {
1244 node = &state->rb_node;
1250 * this search will find all the extents that end after
1253 node = tree_search_for_insert(tree, start, &p, &parent);
1255 prealloc = alloc_extent_state_atomic(prealloc);
1260 err = insert_state(tree, prealloc, start, end,
1261 &p, &parent, &bits, NULL);
1263 extent_io_tree_panic(tree, err);
1264 cache_state(prealloc, cached_state);
1268 state = rb_entry(node, struct extent_state, rb_node);
1270 last_start = state->start;
1271 last_end = state->end;
1274 * | ---- desired range ---- |
1277 * Just lock what we found and keep going
1279 if (state->start == start && state->end <= end) {
1280 set_state_bits(tree, state, &bits, NULL);
1281 cache_state(state, cached_state);
1282 state = clear_state_bit(tree, state, &clear_bits, 0, NULL);
1283 if (last_end == (u64)-1)
1285 start = last_end + 1;
1286 if (start < end && state && state->start == start &&
1293 * | ---- desired range ---- |
1296 * | ------------- state -------------- |
1298 * We need to split the extent we found, and may flip bits on
1301 * If the extent we found extends past our
1302 * range, we just split and search again. It'll get split
1303 * again the next time though.
1305 * If the extent we found is inside our range, we set the
1306 * desired bit on it.
1308 if (state->start < start) {
1309 prealloc = alloc_extent_state_atomic(prealloc);
1314 err = split_state(tree, state, prealloc, start);
1316 extent_io_tree_panic(tree, err);
1320 if (state->end <= end) {
1321 set_state_bits(tree, state, &bits, NULL);
1322 cache_state(state, cached_state);
1323 state = clear_state_bit(tree, state, &clear_bits, 0,
1325 if (last_end == (u64)-1)
1327 start = last_end + 1;
1328 if (start < end && state && state->start == start &&
1335 * | ---- desired range ---- |
1336 * | state | or | state |
1338 * There's a hole, we need to insert something in it and
1339 * ignore the extent we found.
1341 if (state->start > start) {
1343 if (end < last_start)
1346 this_end = last_start - 1;
1348 prealloc = alloc_extent_state_atomic(prealloc);
1355 * Avoid to free 'prealloc' if it can be merged with
1358 err = insert_state(tree, prealloc, start, this_end,
1359 NULL, NULL, &bits, NULL);
1361 extent_io_tree_panic(tree, err);
1362 cache_state(prealloc, cached_state);
1364 start = this_end + 1;
1368 * | ---- desired range ---- |
1370 * We need to split the extent, and set the bit
1373 if (state->start <= end && state->end > end) {
1374 prealloc = alloc_extent_state_atomic(prealloc);
1380 err = split_state(tree, state, prealloc, end + 1);
1382 extent_io_tree_panic(tree, err);
1384 set_state_bits(tree, prealloc, &bits, NULL);
1385 cache_state(prealloc, cached_state);
1386 clear_state_bit(tree, prealloc, &clear_bits, 0, NULL);
1394 spin_unlock(&tree->lock);
1396 first_iteration = false;
1400 spin_unlock(&tree->lock);
1402 free_extent_state(prealloc);
1407 /* wrappers around set/clear extent bit */
1408 int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1409 u32 bits, struct extent_changeset *changeset)
1412 * We don't support EXTENT_LOCKED yet, as current changeset will
1413 * record any bits changed, so for EXTENT_LOCKED case, it will
1414 * either fail with -EEXIST or changeset will record the whole
1417 BUG_ON(bits & EXTENT_LOCKED);
1419 return set_extent_bit(tree, start, end, bits, 0, NULL, NULL, GFP_NOFS,
1423 int set_extent_bits_nowait(struct extent_io_tree *tree, u64 start, u64 end,
1426 return set_extent_bit(tree, start, end, bits, 0, NULL, NULL,
1430 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1431 u32 bits, int wake, int delete,
1432 struct extent_state **cached)
1434 return __clear_extent_bit(tree, start, end, bits, wake, delete,
1435 cached, GFP_NOFS, NULL);
1438 int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1439 u32 bits, struct extent_changeset *changeset)
1442 * Don't support EXTENT_LOCKED case, same reason as
1443 * set_record_extent_bits().
1445 BUG_ON(bits & EXTENT_LOCKED);
1447 return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, GFP_NOFS,
1452 * either insert or lock state struct between start and end use mask to tell
1453 * us if waiting is desired.
1455 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1456 struct extent_state **cached_state)
1462 err = set_extent_bit(tree, start, end, EXTENT_LOCKED,
1463 EXTENT_LOCKED, &failed_start,
1464 cached_state, GFP_NOFS, NULL);
1465 if (err == -EEXIST) {
1466 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1467 start = failed_start;
1470 WARN_ON(start > end);
1475 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1480 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1481 &failed_start, NULL, GFP_NOFS, NULL);
1482 if (err == -EEXIST) {
1483 if (failed_start > start)
1484 clear_extent_bit(tree, start, failed_start - 1,
1485 EXTENT_LOCKED, 1, 0, NULL);
1491 void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
1493 unsigned long index = start >> PAGE_SHIFT;
1494 unsigned long end_index = end >> PAGE_SHIFT;
1497 while (index <= end_index) {
1498 page = find_get_page(inode->i_mapping, index);
1499 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1500 clear_page_dirty_for_io(page);
1506 void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
1508 unsigned long index = start >> PAGE_SHIFT;
1509 unsigned long end_index = end >> PAGE_SHIFT;
1512 while (index <= end_index) {
1513 page = find_get_page(inode->i_mapping, index);
1514 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1515 __set_page_dirty_nobuffers(page);
1516 account_page_redirty(page);
1522 /* find the first state struct with 'bits' set after 'start', and
1523 * return it. tree->lock must be held. NULL will returned if
1524 * nothing was found after 'start'
1526 static struct extent_state *
1527 find_first_extent_bit_state(struct extent_io_tree *tree, u64 start, u32 bits)
1529 struct rb_node *node;
1530 struct extent_state *state;
1533 * this search will find all the extents that end after
1536 node = tree_search(tree, start);
1541 state = rb_entry(node, struct extent_state, rb_node);
1542 if (state->end >= start && (state->state & bits))
1545 node = rb_next(node);
1554 * Find the first offset in the io tree with one or more @bits set.
1556 * Note: If there are multiple bits set in @bits, any of them will match.
1558 * Return 0 if we find something, and update @start_ret and @end_ret.
1559 * Return 1 if we found nothing.
1561 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1562 u64 *start_ret, u64 *end_ret, u32 bits,
1563 struct extent_state **cached_state)
1565 struct extent_state *state;
1568 spin_lock(&tree->lock);
1569 if (cached_state && *cached_state) {
1570 state = *cached_state;
1571 if (state->end == start - 1 && extent_state_in_tree(state)) {
1572 while ((state = next_state(state)) != NULL) {
1573 if (state->state & bits)
1576 free_extent_state(*cached_state);
1577 *cached_state = NULL;
1580 free_extent_state(*cached_state);
1581 *cached_state = NULL;
1584 state = find_first_extent_bit_state(tree, start, bits);
1587 cache_state_if_flags(state, cached_state, 0);
1588 *start_ret = state->start;
1589 *end_ret = state->end;
1593 spin_unlock(&tree->lock);
1598 * Find a contiguous area of bits
1600 * @tree: io tree to check
1601 * @start: offset to start the search from
1602 * @start_ret: the first offset we found with the bits set
1603 * @end_ret: the final contiguous range of the bits that were set
1604 * @bits: bits to look for
1606 * set_extent_bit and clear_extent_bit can temporarily split contiguous ranges
1607 * to set bits appropriately, and then merge them again. During this time it
1608 * will drop the tree->lock, so use this helper if you want to find the actual
1609 * contiguous area for given bits. We will search to the first bit we find, and
1610 * then walk down the tree until we find a non-contiguous area. The area
1611 * returned will be the full contiguous area with the bits set.
1613 int find_contiguous_extent_bit(struct extent_io_tree *tree, u64 start,
1614 u64 *start_ret, u64 *end_ret, u32 bits)
1616 struct extent_state *state;
1619 spin_lock(&tree->lock);
1620 state = find_first_extent_bit_state(tree, start, bits);
1622 *start_ret = state->start;
1623 *end_ret = state->end;
1624 while ((state = next_state(state)) != NULL) {
1625 if (state->start > (*end_ret + 1))
1627 *end_ret = state->end;
1631 spin_unlock(&tree->lock);
1636 * Find the first range that has @bits not set. This range could start before
1639 * @tree: the tree to search
1640 * @start: offset at/after which the found extent should start
1641 * @start_ret: records the beginning of the range
1642 * @end_ret: records the end of the range (inclusive)
1643 * @bits: the set of bits which must be unset
1645 * Since unallocated range is also considered one which doesn't have the bits
1646 * set it's possible that @end_ret contains -1, this happens in case the range
1647 * spans (last_range_end, end of device]. In this case it's up to the caller to
1648 * trim @end_ret to the appropriate size.
1650 void find_first_clear_extent_bit(struct extent_io_tree *tree, u64 start,
1651 u64 *start_ret, u64 *end_ret, u32 bits)
1653 struct extent_state *state;
1654 struct rb_node *node, *prev = NULL, *next;
1656 spin_lock(&tree->lock);
1658 /* Find first extent with bits cleared */
1660 node = __etree_search(tree, start, &next, &prev, NULL, NULL);
1661 if (!node && !next && !prev) {
1663 * Tree is completely empty, send full range and let
1664 * caller deal with it
1669 } else if (!node && !next) {
1671 * We are past the last allocated chunk, set start at
1672 * the end of the last extent.
1674 state = rb_entry(prev, struct extent_state, rb_node);
1675 *start_ret = state->end + 1;
1682 * At this point 'node' either contains 'start' or start is
1685 state = rb_entry(node, struct extent_state, rb_node);
1687 if (in_range(start, state->start, state->end - state->start + 1)) {
1688 if (state->state & bits) {
1690 * |--range with bits sets--|
1694 start = state->end + 1;
1697 * 'start' falls within a range that doesn't
1698 * have the bits set, so take its start as
1699 * the beginning of the desired range
1701 * |--range with bits cleared----|
1705 *start_ret = state->start;
1710 * |---prev range---|---hole/unset---|---node range---|
1716 * |---hole/unset--||--first node--|
1721 state = rb_entry(prev, struct extent_state,
1723 *start_ret = state->end + 1;
1732 * Find the longest stretch from start until an entry which has the
1736 state = rb_entry(node, struct extent_state, rb_node);
1737 if (state->end >= start && !(state->state & bits)) {
1738 *end_ret = state->end;
1740 *end_ret = state->start - 1;
1744 node = rb_next(node);
1749 spin_unlock(&tree->lock);
1753 * find a contiguous range of bytes in the file marked as delalloc, not
1754 * more than 'max_bytes'. start and end are used to return the range,
1756 * true is returned if we find something, false if nothing was in the tree
1758 bool btrfs_find_delalloc_range(struct extent_io_tree *tree, u64 *start,
1759 u64 *end, u64 max_bytes,
1760 struct extent_state **cached_state)
1762 struct rb_node *node;
1763 struct extent_state *state;
1764 u64 cur_start = *start;
1766 u64 total_bytes = 0;
1768 spin_lock(&tree->lock);
1771 * this search will find all the extents that end after
1774 node = tree_search(tree, cur_start);
1781 state = rb_entry(node, struct extent_state, rb_node);
1782 if (found && (state->start != cur_start ||
1783 (state->state & EXTENT_BOUNDARY))) {
1786 if (!(state->state & EXTENT_DELALLOC)) {
1792 *start = state->start;
1793 *cached_state = state;
1794 refcount_inc(&state->refs);
1798 cur_start = state->end + 1;
1799 node = rb_next(node);
1800 total_bytes += state->end - state->start + 1;
1801 if (total_bytes >= max_bytes)
1807 spin_unlock(&tree->lock);
1811 static int __process_pages_contig(struct address_space *mapping,
1812 struct page *locked_page,
1813 pgoff_t start_index, pgoff_t end_index,
1814 unsigned long page_ops, pgoff_t *index_ret);
1816 static noinline void __unlock_for_delalloc(struct inode *inode,
1817 struct page *locked_page,
1820 unsigned long index = start >> PAGE_SHIFT;
1821 unsigned long end_index = end >> PAGE_SHIFT;
1823 ASSERT(locked_page);
1824 if (index == locked_page->index && end_index == index)
1827 __process_pages_contig(inode->i_mapping, locked_page, index, end_index,
1831 static noinline int lock_delalloc_pages(struct inode *inode,
1832 struct page *locked_page,
1836 unsigned long index = delalloc_start >> PAGE_SHIFT;
1837 unsigned long index_ret = index;
1838 unsigned long end_index = delalloc_end >> PAGE_SHIFT;
1841 ASSERT(locked_page);
1842 if (index == locked_page->index && index == end_index)
1845 ret = __process_pages_contig(inode->i_mapping, locked_page, index,
1846 end_index, PAGE_LOCK, &index_ret);
1848 __unlock_for_delalloc(inode, locked_page, delalloc_start,
1849 (u64)index_ret << PAGE_SHIFT);
1854 * Find and lock a contiguous range of bytes in the file marked as delalloc, no
1855 * more than @max_bytes. @Start and @end are used to return the range,
1857 * Return: true if we find something
1858 * false if nothing was in the tree
1861 noinline_for_stack bool find_lock_delalloc_range(struct inode *inode,
1862 struct page *locked_page, u64 *start,
1865 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
1866 u64 max_bytes = BTRFS_MAX_EXTENT_SIZE;
1870 struct extent_state *cached_state = NULL;
1875 /* step one, find a bunch of delalloc bytes starting at start */
1876 delalloc_start = *start;
1878 found = btrfs_find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1879 max_bytes, &cached_state);
1880 if (!found || delalloc_end <= *start) {
1881 *start = delalloc_start;
1882 *end = delalloc_end;
1883 free_extent_state(cached_state);
1888 * start comes from the offset of locked_page. We have to lock
1889 * pages in order, so we can't process delalloc bytes before
1892 if (delalloc_start < *start)
1893 delalloc_start = *start;
1896 * make sure to limit the number of pages we try to lock down
1898 if (delalloc_end + 1 - delalloc_start > max_bytes)
1899 delalloc_end = delalloc_start + max_bytes - 1;
1901 /* step two, lock all the pages after the page that has start */
1902 ret = lock_delalloc_pages(inode, locked_page,
1903 delalloc_start, delalloc_end);
1904 ASSERT(!ret || ret == -EAGAIN);
1905 if (ret == -EAGAIN) {
1906 /* some of the pages are gone, lets avoid looping by
1907 * shortening the size of the delalloc range we're searching
1909 free_extent_state(cached_state);
1910 cached_state = NULL;
1912 max_bytes = PAGE_SIZE;
1921 /* step three, lock the state bits for the whole range */
1922 lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state);
1924 /* then test to make sure it is all still delalloc */
1925 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1926 EXTENT_DELALLOC, 1, cached_state);
1928 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1930 __unlock_for_delalloc(inode, locked_page,
1931 delalloc_start, delalloc_end);
1935 free_extent_state(cached_state);
1936 *start = delalloc_start;
1937 *end = delalloc_end;
1942 static int __process_pages_contig(struct address_space *mapping,
1943 struct page *locked_page,
1944 pgoff_t start_index, pgoff_t end_index,
1945 unsigned long page_ops, pgoff_t *index_ret)
1947 unsigned long nr_pages = end_index - start_index + 1;
1948 unsigned long pages_processed = 0;
1949 pgoff_t index = start_index;
1950 struct page *pages[16];
1955 if (page_ops & PAGE_LOCK) {
1956 ASSERT(page_ops == PAGE_LOCK);
1957 ASSERT(index_ret && *index_ret == start_index);
1960 if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0)
1961 mapping_set_error(mapping, -EIO);
1963 while (nr_pages > 0) {
1964 ret = find_get_pages_contig(mapping, index,
1965 min_t(unsigned long,
1966 nr_pages, ARRAY_SIZE(pages)), pages);
1969 * Only if we're going to lock these pages,
1970 * can we find nothing at @index.
1972 ASSERT(page_ops & PAGE_LOCK);
1977 for (i = 0; i < ret; i++) {
1978 if (page_ops & PAGE_SET_PRIVATE2)
1979 SetPagePrivate2(pages[i]);
1981 if (locked_page && pages[i] == locked_page) {
1986 if (page_ops & PAGE_START_WRITEBACK) {
1987 clear_page_dirty_for_io(pages[i]);
1988 set_page_writeback(pages[i]);
1990 if (page_ops & PAGE_SET_ERROR)
1991 SetPageError(pages[i]);
1992 if (page_ops & PAGE_END_WRITEBACK)
1993 end_page_writeback(pages[i]);
1994 if (page_ops & PAGE_UNLOCK)
1995 unlock_page(pages[i]);
1996 if (page_ops & PAGE_LOCK) {
1997 lock_page(pages[i]);
1998 if (!PageDirty(pages[i]) ||
1999 pages[i]->mapping != mapping) {
2000 unlock_page(pages[i]);
2001 for (; i < ret; i++)
2015 if (err && index_ret)
2016 *index_ret = start_index + pages_processed - 1;
2020 void extent_clear_unlock_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2021 struct page *locked_page,
2022 u32 clear_bits, unsigned long page_ops)
2024 clear_extent_bit(&inode->io_tree, start, end, clear_bits, 1, 0, NULL);
2026 __process_pages_contig(inode->vfs_inode.i_mapping, locked_page,
2027 start >> PAGE_SHIFT, end >> PAGE_SHIFT,
2032 * count the number of bytes in the tree that have a given bit(s)
2033 * set. This can be fairly slow, except for EXTENT_DIRTY which is
2034 * cached. The total number found is returned.
2036 u64 count_range_bits(struct extent_io_tree *tree,
2037 u64 *start, u64 search_end, u64 max_bytes,
2038 u32 bits, int contig)
2040 struct rb_node *node;
2041 struct extent_state *state;
2042 u64 cur_start = *start;
2043 u64 total_bytes = 0;
2047 if (WARN_ON(search_end <= cur_start))
2050 spin_lock(&tree->lock);
2051 if (cur_start == 0 && bits == EXTENT_DIRTY) {
2052 total_bytes = tree->dirty_bytes;
2056 * this search will find all the extents that end after
2059 node = tree_search(tree, cur_start);
2064 state = rb_entry(node, struct extent_state, rb_node);
2065 if (state->start > search_end)
2067 if (contig && found && state->start > last + 1)
2069 if (state->end >= cur_start && (state->state & bits) == bits) {
2070 total_bytes += min(search_end, state->end) + 1 -
2071 max(cur_start, state->start);
2072 if (total_bytes >= max_bytes)
2075 *start = max(cur_start, state->start);
2079 } else if (contig && found) {
2082 node = rb_next(node);
2087 spin_unlock(&tree->lock);
2092 * set the private field for a given byte offset in the tree. If there isn't
2093 * an extent_state there already, this does nothing.
2095 int set_state_failrec(struct extent_io_tree *tree, u64 start,
2096 struct io_failure_record *failrec)
2098 struct rb_node *node;
2099 struct extent_state *state;
2102 spin_lock(&tree->lock);
2104 * this search will find all the extents that end after
2107 node = tree_search(tree, start);
2112 state = rb_entry(node, struct extent_state, rb_node);
2113 if (state->start != start) {
2117 state->failrec = failrec;
2119 spin_unlock(&tree->lock);
2123 struct io_failure_record *get_state_failrec(struct extent_io_tree *tree, u64 start)
2125 struct rb_node *node;
2126 struct extent_state *state;
2127 struct io_failure_record *failrec;
2129 spin_lock(&tree->lock);
2131 * this search will find all the extents that end after
2134 node = tree_search(tree, start);
2136 failrec = ERR_PTR(-ENOENT);
2139 state = rb_entry(node, struct extent_state, rb_node);
2140 if (state->start != start) {
2141 failrec = ERR_PTR(-ENOENT);
2145 failrec = state->failrec;
2147 spin_unlock(&tree->lock);
2152 * searches a range in the state tree for a given mask.
2153 * If 'filled' == 1, this returns 1 only if every extent in the tree
2154 * has the bits set. Otherwise, 1 is returned if any bit in the
2155 * range is found set.
2157 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
2158 u32 bits, int filled, struct extent_state *cached)
2160 struct extent_state *state = NULL;
2161 struct rb_node *node;
2164 spin_lock(&tree->lock);
2165 if (cached && extent_state_in_tree(cached) && cached->start <= start &&
2166 cached->end > start)
2167 node = &cached->rb_node;
2169 node = tree_search(tree, start);
2170 while (node && start <= end) {
2171 state = rb_entry(node, struct extent_state, rb_node);
2173 if (filled && state->start > start) {
2178 if (state->start > end)
2181 if (state->state & bits) {
2185 } else if (filled) {
2190 if (state->end == (u64)-1)
2193 start = state->end + 1;
2196 node = rb_next(node);
2203 spin_unlock(&tree->lock);
2208 * helper function to set a given page up to date if all the
2209 * extents in the tree for that page are up to date
2211 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
2213 u64 start = page_offset(page);
2214 u64 end = start + PAGE_SIZE - 1;
2215 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
2216 SetPageUptodate(page);
2219 int free_io_failure(struct extent_io_tree *failure_tree,
2220 struct extent_io_tree *io_tree,
2221 struct io_failure_record *rec)
2226 set_state_failrec(failure_tree, rec->start, NULL);
2227 ret = clear_extent_bits(failure_tree, rec->start,
2228 rec->start + rec->len - 1,
2229 EXTENT_LOCKED | EXTENT_DIRTY);
2233 ret = clear_extent_bits(io_tree, rec->start,
2234 rec->start + rec->len - 1,
2244 * this bypasses the standard btrfs submit functions deliberately, as
2245 * the standard behavior is to write all copies in a raid setup. here we only
2246 * want to write the one bad copy. so we do the mapping for ourselves and issue
2247 * submit_bio directly.
2248 * to avoid any synchronization issues, wait for the data after writing, which
2249 * actually prevents the read that triggered the error from finishing.
2250 * currently, there can be no more than two copies of every data bit. thus,
2251 * exactly one rewrite is required.
2253 int repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start,
2254 u64 length, u64 logical, struct page *page,
2255 unsigned int pg_offset, int mirror_num)
2258 struct btrfs_device *dev;
2261 struct btrfs_bio *bbio = NULL;
2264 ASSERT(!(fs_info->sb->s_flags & SB_RDONLY));
2265 BUG_ON(!mirror_num);
2267 if (btrfs_is_zoned(fs_info))
2268 return btrfs_repair_one_zone(fs_info, logical);
2270 bio = btrfs_io_bio_alloc(1);
2271 bio->bi_iter.bi_size = 0;
2272 map_length = length;
2275 * Avoid races with device replace and make sure our bbio has devices
2276 * associated to its stripes that don't go away while we are doing the
2277 * read repair operation.
2279 btrfs_bio_counter_inc_blocked(fs_info);
2280 if (btrfs_is_parity_mirror(fs_info, logical, length)) {
2282 * Note that we don't use BTRFS_MAP_WRITE because it's supposed
2283 * to update all raid stripes, but here we just want to correct
2284 * bad stripe, thus BTRFS_MAP_READ is abused to only get the bad
2285 * stripe's dev and sector.
2287 ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, logical,
2288 &map_length, &bbio, 0);
2290 btrfs_bio_counter_dec(fs_info);
2294 ASSERT(bbio->mirror_num == 1);
2296 ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical,
2297 &map_length, &bbio, mirror_num);
2299 btrfs_bio_counter_dec(fs_info);
2303 BUG_ON(mirror_num != bbio->mirror_num);
2306 sector = bbio->stripes[bbio->mirror_num - 1].physical >> 9;
2307 bio->bi_iter.bi_sector = sector;
2308 dev = bbio->stripes[bbio->mirror_num - 1].dev;
2309 btrfs_put_bbio(bbio);
2310 if (!dev || !dev->bdev ||
2311 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) {
2312 btrfs_bio_counter_dec(fs_info);
2316 bio_set_dev(bio, dev->bdev);
2317 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC;
2318 bio_add_page(bio, page, length, pg_offset);
2320 if (btrfsic_submit_bio_wait(bio)) {
2321 /* try to remap that extent elsewhere? */
2322 btrfs_bio_counter_dec(fs_info);
2324 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
2328 btrfs_info_rl_in_rcu(fs_info,
2329 "read error corrected: ino %llu off %llu (dev %s sector %llu)",
2331 rcu_str_deref(dev->name), sector);
2332 btrfs_bio_counter_dec(fs_info);
2337 int btrfs_repair_eb_io_failure(const struct extent_buffer *eb, int mirror_num)
2339 struct btrfs_fs_info *fs_info = eb->fs_info;
2340 u64 start = eb->start;
2341 int i, num_pages = num_extent_pages(eb);
2344 if (sb_rdonly(fs_info->sb))
2347 for (i = 0; i < num_pages; i++) {
2348 struct page *p = eb->pages[i];
2350 ret = repair_io_failure(fs_info, 0, start, PAGE_SIZE, start, p,
2351 start - page_offset(p), mirror_num);
2361 * each time an IO finishes, we do a fast check in the IO failure tree
2362 * to see if we need to process or clean up an io_failure_record
2364 int clean_io_failure(struct btrfs_fs_info *fs_info,
2365 struct extent_io_tree *failure_tree,
2366 struct extent_io_tree *io_tree, u64 start,
2367 struct page *page, u64 ino, unsigned int pg_offset)
2370 struct io_failure_record *failrec;
2371 struct extent_state *state;
2376 ret = count_range_bits(failure_tree, &private, (u64)-1, 1,
2381 failrec = get_state_failrec(failure_tree, start);
2382 if (IS_ERR(failrec))
2385 BUG_ON(!failrec->this_mirror);
2387 if (sb_rdonly(fs_info->sb))
2390 spin_lock(&io_tree->lock);
2391 state = find_first_extent_bit_state(io_tree,
2394 spin_unlock(&io_tree->lock);
2396 if (state && state->start <= failrec->start &&
2397 state->end >= failrec->start + failrec->len - 1) {
2398 num_copies = btrfs_num_copies(fs_info, failrec->logical,
2400 if (num_copies > 1) {
2401 repair_io_failure(fs_info, ino, start, failrec->len,
2402 failrec->logical, page, pg_offset,
2403 failrec->failed_mirror);
2408 free_io_failure(failure_tree, io_tree, failrec);
2414 * Can be called when
2415 * - hold extent lock
2416 * - under ordered extent
2417 * - the inode is freeing
2419 void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start, u64 end)
2421 struct extent_io_tree *failure_tree = &inode->io_failure_tree;
2422 struct io_failure_record *failrec;
2423 struct extent_state *state, *next;
2425 if (RB_EMPTY_ROOT(&failure_tree->state))
2428 spin_lock(&failure_tree->lock);
2429 state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY);
2431 if (state->start > end)
2434 ASSERT(state->end <= end);
2436 next = next_state(state);
2438 failrec = state->failrec;
2439 free_extent_state(state);
2444 spin_unlock(&failure_tree->lock);
2447 static struct io_failure_record *btrfs_get_io_failure_record(struct inode *inode,
2450 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2451 struct io_failure_record *failrec;
2452 struct extent_map *em;
2453 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2454 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2455 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2456 const u32 sectorsize = fs_info->sectorsize;
2460 failrec = get_state_failrec(failure_tree, start);
2461 if (!IS_ERR(failrec)) {
2462 btrfs_debug(fs_info,
2463 "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu",
2464 failrec->logical, failrec->start, failrec->len);
2466 * when data can be on disk more than twice, add to failrec here
2467 * (e.g. with a list for failed_mirror) to make
2468 * clean_io_failure() clean all those errors at once.
2474 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2476 return ERR_PTR(-ENOMEM);
2478 failrec->start = start;
2479 failrec->len = sectorsize;
2480 failrec->this_mirror = 0;
2481 failrec->bio_flags = 0;
2483 read_lock(&em_tree->lock);
2484 em = lookup_extent_mapping(em_tree, start, failrec->len);
2486 read_unlock(&em_tree->lock);
2488 return ERR_PTR(-EIO);
2491 if (em->start > start || em->start + em->len <= start) {
2492 free_extent_map(em);
2495 read_unlock(&em_tree->lock);
2498 return ERR_PTR(-EIO);
2501 logical = start - em->start;
2502 logical = em->block_start + logical;
2503 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2504 logical = em->block_start;
2505 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2506 extent_set_compress_type(&failrec->bio_flags, em->compress_type);
2509 btrfs_debug(fs_info,
2510 "Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu",
2511 logical, start, failrec->len);
2513 failrec->logical = logical;
2514 free_extent_map(em);
2516 /* Set the bits in the private failure tree */
2517 ret = set_extent_bits(failure_tree, start, start + sectorsize - 1,
2518 EXTENT_LOCKED | EXTENT_DIRTY);
2520 ret = set_state_failrec(failure_tree, start, failrec);
2521 /* Set the bits in the inode's tree */
2522 ret = set_extent_bits(tree, start, start + sectorsize - 1,
2524 } else if (ret < 0) {
2526 return ERR_PTR(ret);
2532 static bool btrfs_check_repairable(struct inode *inode,
2533 struct io_failure_record *failrec,
2536 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2539 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
2540 if (num_copies == 1) {
2542 * we only have a single copy of the data, so don't bother with
2543 * all the retry and error correction code that follows. no
2544 * matter what the error is, it is very likely to persist.
2546 btrfs_debug(fs_info,
2547 "Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
2548 num_copies, failrec->this_mirror, failed_mirror);
2552 /* The failure record should only contain one sector */
2553 ASSERT(failrec->len == fs_info->sectorsize);
2556 * There are two premises:
2557 * a) deliver good data to the caller
2558 * b) correct the bad sectors on disk
2560 * Since we're only doing repair for one sector, we only need to get
2561 * a good copy of the failed sector and if we succeed, we have setup
2562 * everything for repair_io_failure to do the rest for us.
2564 failrec->failed_mirror = failed_mirror;
2565 failrec->this_mirror++;
2566 if (failrec->this_mirror == failed_mirror)
2567 failrec->this_mirror++;
2569 if (failrec->this_mirror > num_copies) {
2570 btrfs_debug(fs_info,
2571 "Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
2572 num_copies, failrec->this_mirror, failed_mirror);
2579 int btrfs_repair_one_sector(struct inode *inode,
2580 struct bio *failed_bio, u32 bio_offset,
2581 struct page *page, unsigned int pgoff,
2582 u64 start, int failed_mirror,
2583 submit_bio_hook_t *submit_bio_hook)
2585 struct io_failure_record *failrec;
2586 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2587 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2588 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2589 struct btrfs_io_bio *failed_io_bio = btrfs_io_bio(failed_bio);
2590 const int icsum = bio_offset >> fs_info->sectorsize_bits;
2591 struct bio *repair_bio;
2592 struct btrfs_io_bio *repair_io_bio;
2593 blk_status_t status;
2595 btrfs_debug(fs_info,
2596 "repair read error: read error at %llu", start);
2598 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
2600 failrec = btrfs_get_io_failure_record(inode, start);
2601 if (IS_ERR(failrec))
2602 return PTR_ERR(failrec);
2605 if (!btrfs_check_repairable(inode, failrec, failed_mirror)) {
2606 free_io_failure(failure_tree, tree, failrec);
2610 repair_bio = btrfs_io_bio_alloc(1);
2611 repair_io_bio = btrfs_io_bio(repair_bio);
2612 repair_bio->bi_opf = REQ_OP_READ;
2613 repair_bio->bi_end_io = failed_bio->bi_end_io;
2614 repair_bio->bi_iter.bi_sector = failrec->logical >> 9;
2615 repair_bio->bi_private = failed_bio->bi_private;
2617 if (failed_io_bio->csum) {
2618 const u32 csum_size = fs_info->csum_size;
2620 repair_io_bio->csum = repair_io_bio->csum_inline;
2621 memcpy(repair_io_bio->csum,
2622 failed_io_bio->csum + csum_size * icsum, csum_size);
2625 bio_add_page(repair_bio, page, failrec->len, pgoff);
2626 repair_io_bio->logical = failrec->start;
2627 repair_io_bio->iter = repair_bio->bi_iter;
2629 btrfs_debug(btrfs_sb(inode->i_sb),
2630 "repair read error: submitting new read to mirror %d",
2631 failrec->this_mirror);
2633 status = submit_bio_hook(inode, repair_bio, failrec->this_mirror,
2634 failrec->bio_flags);
2636 free_io_failure(failure_tree, tree, failrec);
2637 bio_put(repair_bio);
2639 return blk_status_to_errno(status);
2642 static void end_page_read(struct page *page, bool uptodate, u64 start, u32 len)
2644 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
2646 ASSERT(page_offset(page) <= start &&
2647 start + len <= page_offset(page) + PAGE_SIZE);
2650 * For subapge metadata case, all btrfs_page_* helpers need page to
2651 * have page::private populated.
2652 * But we can have rare case where the last eb in the page is only
2653 * referred by the IO, and it gets released immedately after it's
2654 * read and verified.
2656 * This can detach the page private completely.
2657 * In that case, we can just skip the page status update completely,
2658 * as the page has no eb anymore.
2660 if (fs_info->sectorsize < PAGE_SIZE && unlikely(!PagePrivate(page))) {
2661 ASSERT(!is_data_inode(page->mapping->host));
2665 btrfs_page_set_uptodate(fs_info, page, start, len);
2667 btrfs_page_clear_uptodate(fs_info, page, start, len);
2668 btrfs_page_set_error(fs_info, page, start, len);
2671 if (fs_info->sectorsize == PAGE_SIZE)
2673 else if (is_data_inode(page->mapping->host))
2675 * For subpage data, unlock the page if we're the last reader.
2676 * For subpage metadata, page lock is not utilized for read.
2678 btrfs_subpage_end_reader(fs_info, page, start, len);
2681 static blk_status_t submit_read_repair(struct inode *inode,
2682 struct bio *failed_bio, u32 bio_offset,
2683 struct page *page, unsigned int pgoff,
2684 u64 start, u64 end, int failed_mirror,
2685 unsigned int error_bitmap,
2686 submit_bio_hook_t *submit_bio_hook)
2688 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2689 const u32 sectorsize = fs_info->sectorsize;
2690 const int nr_bits = (end + 1 - start) >> fs_info->sectorsize_bits;
2694 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
2696 /* We're here because we had some read errors or csum mismatch */
2697 ASSERT(error_bitmap);
2700 * We only get called on buffered IO, thus page must be mapped and bio
2701 * must not be cloned.
2703 ASSERT(page->mapping && !bio_flagged(failed_bio, BIO_CLONED));
2705 /* Iterate through all the sectors in the range */
2706 for (i = 0; i < nr_bits; i++) {
2707 const unsigned int offset = i * sectorsize;
2708 struct extent_state *cached = NULL;
2709 bool uptodate = false;
2712 if (!(error_bitmap & (1U << i))) {
2714 * This sector has no error, just end the page read
2715 * and unlock the range.
2721 ret = btrfs_repair_one_sector(inode, failed_bio,
2722 bio_offset + offset,
2723 page, pgoff + offset, start + offset,
2724 failed_mirror, submit_bio_hook);
2727 * We have submitted the read repair, the page release
2728 * will be handled by the endio function of the
2729 * submitted repair bio.
2730 * Thus we don't need to do any thing here.
2735 * Repair failed, just record the error but still continue.
2736 * Or the remaining sectors will not be properly unlocked.
2741 end_page_read(page, uptodate, start + offset, sectorsize);
2743 set_extent_uptodate(&BTRFS_I(inode)->io_tree,
2745 start + offset + sectorsize - 1,
2746 &cached, GFP_ATOMIC);
2747 unlock_extent_cached_atomic(&BTRFS_I(inode)->io_tree,
2749 start + offset + sectorsize - 1,
2752 return errno_to_blk_status(error);
2755 /* lots and lots of room for performance fixes in the end_bio funcs */
2757 void end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2759 int uptodate = (err == 0);
2762 btrfs_writepage_endio_finish_ordered(page, start, end, uptodate);
2765 ClearPageUptodate(page);
2767 ret = err < 0 ? err : -EIO;
2768 mapping_set_error(page->mapping, ret);
2773 * after a writepage IO is done, we need to:
2774 * clear the uptodate bits on error
2775 * clear the writeback bits in the extent tree for this IO
2776 * end_page_writeback if the page has no more pending IO
2778 * Scheduling is not allowed, so the extent state tree is expected
2779 * to have one and only one object corresponding to this IO.
2781 static void end_bio_extent_writepage(struct bio *bio)
2783 int error = blk_status_to_errno(bio->bi_status);
2784 struct bio_vec *bvec;
2787 struct bvec_iter_all iter_all;
2788 bool first_bvec = true;
2790 ASSERT(!bio_flagged(bio, BIO_CLONED));
2791 bio_for_each_segment_all(bvec, bio, iter_all) {
2792 struct page *page = bvec->bv_page;
2793 struct inode *inode = page->mapping->host;
2794 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2796 /* We always issue full-page reads, but if some block
2797 * in a page fails to read, blk_update_request() will
2798 * advance bv_offset and adjust bv_len to compensate.
2799 * Print a warning for nonzero offsets, and an error
2800 * if they don't add up to a full page. */
2801 if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) {
2802 if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE)
2804 "partial page write in btrfs with offset %u and length %u",
2805 bvec->bv_offset, bvec->bv_len);
2808 "incomplete page write in btrfs with offset %u and length %u",
2809 bvec->bv_offset, bvec->bv_len);
2812 start = page_offset(page);
2813 end = start + bvec->bv_offset + bvec->bv_len - 1;
2816 btrfs_record_physical_zoned(inode, start, bio);
2820 end_extent_writepage(page, error, start, end);
2821 end_page_writeback(page);
2828 * Record previously processed extent range
2830 * For endio_readpage_release_extent() to handle a full extent range, reducing
2831 * the extent io operations.
2833 struct processed_extent {
2834 struct btrfs_inode *inode;
2835 /* Start of the range in @inode */
2837 /* End of the range in @inode */
2843 * Try to release processed extent range
2845 * May not release the extent range right now if the current range is
2846 * contiguous to processed extent.
2848 * Will release processed extent when any of @inode, @uptodate, the range is
2849 * no longer contiguous to the processed range.
2851 * Passing @inode == NULL will force processed extent to be released.
2853 static void endio_readpage_release_extent(struct processed_extent *processed,
2854 struct btrfs_inode *inode, u64 start, u64 end,
2857 struct extent_state *cached = NULL;
2858 struct extent_io_tree *tree;
2860 /* The first extent, initialize @processed */
2861 if (!processed->inode)
2865 * Contiguous to processed extent, just uptodate the end.
2867 * Several things to notice:
2869 * - bio can be merged as long as on-disk bytenr is contiguous
2870 * This means we can have page belonging to other inodes, thus need to
2871 * check if the inode still matches.
2872 * - bvec can contain range beyond current page for multi-page bvec
2873 * Thus we need to do processed->end + 1 >= start check
2875 if (processed->inode == inode && processed->uptodate == uptodate &&
2876 processed->end + 1 >= start && end >= processed->end) {
2877 processed->end = end;
2881 tree = &processed->inode->io_tree;
2883 * Now we don't have range contiguous to the processed range, release
2884 * the processed range now.
2886 if (processed->uptodate && tree->track_uptodate)
2887 set_extent_uptodate(tree, processed->start, processed->end,
2888 &cached, GFP_ATOMIC);
2889 unlock_extent_cached_atomic(tree, processed->start, processed->end,
2893 /* Update processed to current range */
2894 processed->inode = inode;
2895 processed->start = start;
2896 processed->end = end;
2897 processed->uptodate = uptodate;
2900 static void begin_page_read(struct btrfs_fs_info *fs_info, struct page *page)
2902 ASSERT(PageLocked(page));
2903 if (fs_info->sectorsize == PAGE_SIZE)
2906 ASSERT(PagePrivate(page));
2907 btrfs_subpage_start_reader(fs_info, page, page_offset(page), PAGE_SIZE);
2911 * Find extent buffer for a givne bytenr.
2913 * This is for end_bio_extent_readpage(), thus we can't do any unsafe locking
2916 static struct extent_buffer *find_extent_buffer_readpage(
2917 struct btrfs_fs_info *fs_info, struct page *page, u64 bytenr)
2919 struct extent_buffer *eb;
2922 * For regular sectorsize, we can use page->private to grab extent
2925 if (fs_info->sectorsize == PAGE_SIZE) {
2926 ASSERT(PagePrivate(page) && page->private);
2927 return (struct extent_buffer *)page->private;
2930 /* For subpage case, we need to lookup buffer radix tree */
2932 eb = radix_tree_lookup(&fs_info->buffer_radix,
2933 bytenr >> fs_info->sectorsize_bits);
2940 * after a readpage IO is done, we need to:
2941 * clear the uptodate bits on error
2942 * set the uptodate bits if things worked
2943 * set the page up to date if all extents in the tree are uptodate
2944 * clear the lock bit in the extent tree
2945 * unlock the page if there are no other extents locked for it
2947 * Scheduling is not allowed, so the extent state tree is expected
2948 * to have one and only one object corresponding to this IO.
2950 static void end_bio_extent_readpage(struct bio *bio)
2952 struct bio_vec *bvec;
2953 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
2954 struct extent_io_tree *tree, *failure_tree;
2955 struct processed_extent processed = { 0 };
2957 * The offset to the beginning of a bio, since one bio can never be
2958 * larger than UINT_MAX, u32 here is enough.
2963 struct bvec_iter_all iter_all;
2965 ASSERT(!bio_flagged(bio, BIO_CLONED));
2966 bio_for_each_segment_all(bvec, bio, iter_all) {
2967 bool uptodate = !bio->bi_status;
2968 struct page *page = bvec->bv_page;
2969 struct inode *inode = page->mapping->host;
2970 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2971 const u32 sectorsize = fs_info->sectorsize;
2972 unsigned int error_bitmap = (unsigned int)-1;
2977 btrfs_debug(fs_info,
2978 "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u",
2979 bio->bi_iter.bi_sector, bio->bi_status,
2980 io_bio->mirror_num);
2981 tree = &BTRFS_I(inode)->io_tree;
2982 failure_tree = &BTRFS_I(inode)->io_failure_tree;
2985 * We always issue full-sector reads, but if some block in a
2986 * page fails to read, blk_update_request() will advance
2987 * bv_offset and adjust bv_len to compensate. Print a warning
2988 * for unaligned offsets, and an error if they don't add up to
2991 if (!IS_ALIGNED(bvec->bv_offset, sectorsize))
2993 "partial page read in btrfs with offset %u and length %u",
2994 bvec->bv_offset, bvec->bv_len);
2995 else if (!IS_ALIGNED(bvec->bv_offset + bvec->bv_len,
2998 "incomplete page read with offset %u and length %u",
2999 bvec->bv_offset, bvec->bv_len);
3001 start = page_offset(page) + bvec->bv_offset;
3002 end = start + bvec->bv_len - 1;
3005 mirror = io_bio->mirror_num;
3006 if (likely(uptodate)) {
3007 if (is_data_inode(inode)) {
3008 error_bitmap = btrfs_verify_data_csum(io_bio,
3009 bio_offset, page, start, end);
3012 ret = btrfs_validate_metadata_buffer(io_bio,
3013 page, start, end, mirror);
3018 clean_io_failure(BTRFS_I(inode)->root->fs_info,
3019 failure_tree, tree, start,
3021 btrfs_ino(BTRFS_I(inode)), 0);
3024 if (likely(uptodate))
3027 if (is_data_inode(inode)) {
3029 * btrfs_submit_read_repair() will handle all the good
3030 * and bad sectors, we just continue to the next bvec.
3032 submit_read_repair(inode, bio, bio_offset, page,
3033 start - page_offset(page), start,
3034 end, mirror, error_bitmap,
3035 btrfs_submit_data_bio);
3037 ASSERT(bio_offset + len > bio_offset);
3041 struct extent_buffer *eb;
3043 eb = find_extent_buffer_readpage(fs_info, page, start);
3044 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
3045 eb->read_mirror = mirror;
3046 atomic_dec(&eb->io_pages);
3047 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD,
3049 btree_readahead_hook(eb, -EIO);
3052 if (likely(uptodate)) {
3053 loff_t i_size = i_size_read(inode);
3054 pgoff_t end_index = i_size >> PAGE_SHIFT;
3057 * Zero out the remaining part if this range straddles
3060 * Here we should only zero the range inside the bvec,
3061 * not touch anything else.
3063 * NOTE: i_size is exclusive while end is inclusive.
3065 if (page->index == end_index && i_size <= end) {
3066 u32 zero_start = max(offset_in_page(i_size),
3067 offset_in_page(start));
3069 zero_user_segment(page, zero_start,
3070 offset_in_page(end) + 1);
3073 ASSERT(bio_offset + len > bio_offset);
3076 /* Update page status and unlock */
3077 end_page_read(page, uptodate, start, len);
3078 endio_readpage_release_extent(&processed, BTRFS_I(inode),
3079 start, end, uptodate);
3081 /* Release the last extent */
3082 endio_readpage_release_extent(&processed, NULL, 0, 0, false);
3083 btrfs_io_bio_free_csum(io_bio);
3088 * Initialize the members up to but not including 'bio'. Use after allocating a
3089 * new bio by bio_alloc_bioset as it does not initialize the bytes outside of
3090 * 'bio' because use of __GFP_ZERO is not supported.
3092 static inline void btrfs_io_bio_init(struct btrfs_io_bio *btrfs_bio)
3094 memset(btrfs_bio, 0, offsetof(struct btrfs_io_bio, bio));
3098 * The following helpers allocate a bio. As it's backed by a bioset, it'll
3099 * never fail. We're returning a bio right now but you can call btrfs_io_bio
3100 * for the appropriate container_of magic
3102 struct bio *btrfs_bio_alloc(u64 first_byte)
3106 bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_VECS, &btrfs_bioset);
3107 bio->bi_iter.bi_sector = first_byte >> 9;
3108 btrfs_io_bio_init(btrfs_io_bio(bio));
3112 struct bio *btrfs_bio_clone(struct bio *bio)
3114 struct btrfs_io_bio *btrfs_bio;
3117 /* Bio allocation backed by a bioset does not fail */
3118 new = bio_clone_fast(bio, GFP_NOFS, &btrfs_bioset);
3119 btrfs_bio = btrfs_io_bio(new);
3120 btrfs_io_bio_init(btrfs_bio);
3121 btrfs_bio->iter = bio->bi_iter;
3125 struct bio *btrfs_io_bio_alloc(unsigned int nr_iovecs)
3129 /* Bio allocation backed by a bioset does not fail */
3130 bio = bio_alloc_bioset(GFP_NOFS, nr_iovecs, &btrfs_bioset);
3131 btrfs_io_bio_init(btrfs_io_bio(bio));
3135 struct bio *btrfs_bio_clone_partial(struct bio *orig, int offset, int size)
3138 struct btrfs_io_bio *btrfs_bio;
3140 /* this will never fail when it's backed by a bioset */
3141 bio = bio_clone_fast(orig, GFP_NOFS, &btrfs_bioset);
3144 btrfs_bio = btrfs_io_bio(bio);
3145 btrfs_io_bio_init(btrfs_bio);
3147 bio_trim(bio, offset >> 9, size >> 9);
3148 btrfs_bio->iter = bio->bi_iter;
3153 * Attempt to add a page to bio
3155 * @bio: destination bio
3156 * @page: page to add to the bio
3157 * @disk_bytenr: offset of the new bio or to check whether we are adding
3158 * a contiguous page to the previous one
3159 * @pg_offset: starting offset in the page
3160 * @size: portion of page that we want to write
3161 * @prev_bio_flags: flags of previous bio to see if we can merge the current one
3162 * @bio_flags: flags of the current bio to see if we can merge them
3163 * @return: true if page was added, false otherwise
3165 * Attempt to add a page to bio considering stripe alignment etc.
3167 * Return true if successfully page added. Otherwise, return false.
3169 static bool btrfs_bio_add_page(struct btrfs_bio_ctrl *bio_ctrl,
3171 u64 disk_bytenr, unsigned int size,
3172 unsigned int pg_offset,
3173 unsigned long bio_flags)
3175 struct bio *bio = bio_ctrl->bio;
3176 u32 bio_size = bio->bi_iter.bi_size;
3177 const sector_t sector = disk_bytenr >> SECTOR_SHIFT;
3182 /* The limit should be calculated when bio_ctrl->bio is allocated */
3183 ASSERT(bio_ctrl->len_to_oe_boundary && bio_ctrl->len_to_stripe_boundary);
3184 if (bio_ctrl->bio_flags != bio_flags)
3187 if (bio_ctrl->bio_flags & EXTENT_BIO_COMPRESSED)
3188 contig = bio->bi_iter.bi_sector == sector;
3190 contig = bio_end_sector(bio) == sector;
3194 if (bio_size + size > bio_ctrl->len_to_oe_boundary ||
3195 bio_size + size > bio_ctrl->len_to_stripe_boundary)
3198 if (bio_op(bio) == REQ_OP_ZONE_APPEND)
3199 ret = bio_add_zone_append_page(bio, page, size, pg_offset);
3201 ret = bio_add_page(bio, page, size, pg_offset);
3206 static int calc_bio_boundaries(struct btrfs_bio_ctrl *bio_ctrl,
3207 struct btrfs_inode *inode)
3209 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3210 struct btrfs_io_geometry geom;
3211 struct btrfs_ordered_extent *ordered;
3212 struct extent_map *em;
3213 u64 logical = (bio_ctrl->bio->bi_iter.bi_sector << SECTOR_SHIFT);
3217 * Pages for compressed extent are never submitted to disk directly,
3218 * thus it has no real boundary, just set them to U32_MAX.
3220 * The split happens for real compressed bio, which happens in
3221 * btrfs_submit_compressed_read/write().
3223 if (bio_ctrl->bio_flags & EXTENT_BIO_COMPRESSED) {
3224 bio_ctrl->len_to_oe_boundary = U32_MAX;
3225 bio_ctrl->len_to_stripe_boundary = U32_MAX;
3228 em = btrfs_get_chunk_map(fs_info, logical, fs_info->sectorsize);
3231 ret = btrfs_get_io_geometry(fs_info, em, btrfs_op(bio_ctrl->bio),
3233 free_extent_map(em);
3237 if (geom.len > U32_MAX)
3238 bio_ctrl->len_to_stripe_boundary = U32_MAX;
3240 bio_ctrl->len_to_stripe_boundary = (u32)geom.len;
3242 if (!btrfs_is_zoned(fs_info) ||
3243 bio_op(bio_ctrl->bio) != REQ_OP_ZONE_APPEND) {
3244 bio_ctrl->len_to_oe_boundary = U32_MAX;
3248 ASSERT(fs_info->max_zone_append_size > 0);
3249 /* Ordered extent not yet created, so we're good */
3250 ordered = btrfs_lookup_ordered_extent(inode, logical);
3252 bio_ctrl->len_to_oe_boundary = U32_MAX;
3256 bio_ctrl->len_to_oe_boundary = min_t(u32, U32_MAX,
3257 ordered->disk_bytenr + ordered->disk_num_bytes - logical);
3258 btrfs_put_ordered_extent(ordered);
3263 * @opf: bio REQ_OP_* and REQ_* flags as one value
3264 * @wbc: optional writeback control for io accounting
3265 * @page: page to add to the bio
3266 * @disk_bytenr: logical bytenr where the write will be
3267 * @size: portion of page that we want to write to
3268 * @pg_offset: offset of the new bio or to check whether we are adding
3269 * a contiguous page to the previous one
3270 * @bio_ret: must be valid pointer, newly allocated bio will be stored there
3271 * @end_io_func: end_io callback for new bio
3272 * @mirror_num: desired mirror to read/write
3273 * @prev_bio_flags: flags of previous bio to see if we can merge the current one
3274 * @bio_flags: flags of the current bio to see if we can merge them
3276 static int submit_extent_page(unsigned int opf,
3277 struct writeback_control *wbc,
3278 struct btrfs_bio_ctrl *bio_ctrl,
3279 struct page *page, u64 disk_bytenr,
3280 size_t size, unsigned long pg_offset,
3281 bio_end_io_t end_io_func,
3283 unsigned long bio_flags,
3284 bool force_bio_submit)
3288 size_t io_size = min_t(size_t, size, PAGE_SIZE);
3289 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
3290 struct extent_io_tree *tree = &inode->io_tree;
3291 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3295 ASSERT(pg_offset < PAGE_SIZE && size <= PAGE_SIZE &&
3296 pg_offset + size <= PAGE_SIZE);
3297 if (bio_ctrl->bio) {
3298 bio = bio_ctrl->bio;
3299 if (force_bio_submit ||
3300 !btrfs_bio_add_page(bio_ctrl, page, disk_bytenr, io_size,
3301 pg_offset, bio_flags)) {
3302 ret = submit_one_bio(bio, mirror_num, bio_ctrl->bio_flags);
3303 bio_ctrl->bio = NULL;
3308 wbc_account_cgroup_owner(wbc, page, io_size);
3313 bio = btrfs_bio_alloc(disk_bytenr);
3314 bio_add_page(bio, page, io_size, pg_offset);
3315 bio->bi_end_io = end_io_func;
3316 bio->bi_private = tree;
3317 bio->bi_write_hint = page->mapping->host->i_write_hint;
3320 struct block_device *bdev;
3322 bdev = fs_info->fs_devices->latest_bdev;
3323 bio_set_dev(bio, bdev);
3324 wbc_init_bio(wbc, bio);
3325 wbc_account_cgroup_owner(wbc, page, io_size);
3327 if (btrfs_is_zoned(fs_info) && bio_op(bio) == REQ_OP_ZONE_APPEND) {
3328 struct btrfs_device *device;
3330 device = btrfs_zoned_get_device(fs_info, disk_bytenr, io_size);
3332 return PTR_ERR(device);
3334 btrfs_io_bio(bio)->device = device;
3337 bio_ctrl->bio = bio;
3338 bio_ctrl->bio_flags = bio_flags;
3339 ret = calc_bio_boundaries(bio_ctrl, inode);
3344 static int attach_extent_buffer_page(struct extent_buffer *eb,
3346 struct btrfs_subpage *prealloc)
3348 struct btrfs_fs_info *fs_info = eb->fs_info;
3352 * If the page is mapped to btree inode, we should hold the private
3353 * lock to prevent race.
3354 * For cloned or dummy extent buffers, their pages are not mapped and
3355 * will not race with any other ebs.
3358 lockdep_assert_held(&page->mapping->private_lock);
3360 if (fs_info->sectorsize == PAGE_SIZE) {
3361 if (!PagePrivate(page))
3362 attach_page_private(page, eb);
3364 WARN_ON(page->private != (unsigned long)eb);
3368 /* Already mapped, just free prealloc */
3369 if (PagePrivate(page)) {
3370 btrfs_free_subpage(prealloc);
3375 /* Has preallocated memory for subpage */
3376 attach_page_private(page, prealloc);
3378 /* Do new allocation to attach subpage */
3379 ret = btrfs_attach_subpage(fs_info, page,
3380 BTRFS_SUBPAGE_METADATA);
3384 int set_page_extent_mapped(struct page *page)
3386 struct btrfs_fs_info *fs_info;
3388 ASSERT(page->mapping);
3390 if (PagePrivate(page))
3393 fs_info = btrfs_sb(page->mapping->host->i_sb);
3395 if (fs_info->sectorsize < PAGE_SIZE)
3396 return btrfs_attach_subpage(fs_info, page, BTRFS_SUBPAGE_DATA);
3398 attach_page_private(page, (void *)EXTENT_PAGE_PRIVATE);
3402 void clear_page_extent_mapped(struct page *page)
3404 struct btrfs_fs_info *fs_info;
3406 ASSERT(page->mapping);
3408 if (!PagePrivate(page))
3411 fs_info = btrfs_sb(page->mapping->host->i_sb);
3412 if (fs_info->sectorsize < PAGE_SIZE)
3413 return btrfs_detach_subpage(fs_info, page);
3415 detach_page_private(page);
3418 static struct extent_map *
3419 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
3420 u64 start, u64 len, struct extent_map **em_cached)
3422 struct extent_map *em;
3424 if (em_cached && *em_cached) {
3426 if (extent_map_in_tree(em) && start >= em->start &&
3427 start < extent_map_end(em)) {
3428 refcount_inc(&em->refs);
3432 free_extent_map(em);
3436 em = btrfs_get_extent(BTRFS_I(inode), page, pg_offset, start, len);
3437 if (em_cached && !IS_ERR_OR_NULL(em)) {
3439 refcount_inc(&em->refs);
3445 * basic readpage implementation. Locked extent state structs are inserted
3446 * into the tree that are removed when the IO is done (by the end_io
3448 * XXX JDM: This needs looking at to ensure proper page locking
3449 * return 0 on success, otherwise return error
3451 int btrfs_do_readpage(struct page *page, struct extent_map **em_cached,
3452 struct btrfs_bio_ctrl *bio_ctrl,
3453 unsigned int read_flags, u64 *prev_em_start)
3455 struct inode *inode = page->mapping->host;
3456 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3457 u64 start = page_offset(page);
3458 const u64 end = start + PAGE_SIZE - 1;
3461 u64 last_byte = i_size_read(inode);
3464 struct extent_map *em;
3467 size_t pg_offset = 0;
3469 size_t blocksize = inode->i_sb->s_blocksize;
3470 unsigned long this_bio_flag = 0;
3471 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
3473 ret = set_page_extent_mapped(page);
3475 unlock_extent(tree, start, end);
3476 btrfs_page_set_error(fs_info, page, start, PAGE_SIZE);
3481 if (!PageUptodate(page)) {
3482 if (cleancache_get_page(page) == 0) {
3483 BUG_ON(blocksize != PAGE_SIZE);
3484 unlock_extent(tree, start, end);
3490 if (page->index == last_byte >> PAGE_SHIFT) {
3491 size_t zero_offset = offset_in_page(last_byte);
3494 iosize = PAGE_SIZE - zero_offset;
3495 memzero_page(page, zero_offset, iosize);
3496 flush_dcache_page(page);
3499 begin_page_read(fs_info, page);
3500 while (cur <= end) {
3501 bool force_bio_submit = false;
3504 if (cur >= last_byte) {
3505 struct extent_state *cached = NULL;
3507 iosize = PAGE_SIZE - pg_offset;
3508 memzero_page(page, pg_offset, iosize);
3509 flush_dcache_page(page);
3510 set_extent_uptodate(tree, cur, cur + iosize - 1,
3512 unlock_extent_cached(tree, cur,
3513 cur + iosize - 1, &cached);
3514 end_page_read(page, true, cur, iosize);
3517 em = __get_extent_map(inode, page, pg_offset, cur,
3518 end - cur + 1, em_cached);
3519 if (IS_ERR_OR_NULL(em)) {
3520 unlock_extent(tree, cur, end);
3521 end_page_read(page, false, cur, end + 1 - cur);
3524 extent_offset = cur - em->start;
3525 BUG_ON(extent_map_end(em) <= cur);
3528 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
3529 this_bio_flag |= EXTENT_BIO_COMPRESSED;
3530 extent_set_compress_type(&this_bio_flag,
3534 iosize = min(extent_map_end(em) - cur, end - cur + 1);
3535 cur_end = min(extent_map_end(em) - 1, end);
3536 iosize = ALIGN(iosize, blocksize);
3537 if (this_bio_flag & EXTENT_BIO_COMPRESSED)
3538 disk_bytenr = em->block_start;
3540 disk_bytenr = em->block_start + extent_offset;
3541 block_start = em->block_start;
3542 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
3543 block_start = EXTENT_MAP_HOLE;
3546 * If we have a file range that points to a compressed extent
3547 * and it's followed by a consecutive file range that points
3548 * to the same compressed extent (possibly with a different
3549 * offset and/or length, so it either points to the whole extent
3550 * or only part of it), we must make sure we do not submit a
3551 * single bio to populate the pages for the 2 ranges because
3552 * this makes the compressed extent read zero out the pages
3553 * belonging to the 2nd range. Imagine the following scenario:
3556 * [0 - 8K] [8K - 24K]
3559 * points to extent X, points to extent X,
3560 * offset 4K, length of 8K offset 0, length 16K
3562 * [extent X, compressed length = 4K uncompressed length = 16K]
3564 * If the bio to read the compressed extent covers both ranges,
3565 * it will decompress extent X into the pages belonging to the
3566 * first range and then it will stop, zeroing out the remaining
3567 * pages that belong to the other range that points to extent X.
3568 * So here we make sure we submit 2 bios, one for the first
3569 * range and another one for the third range. Both will target
3570 * the same physical extent from disk, but we can't currently
3571 * make the compressed bio endio callback populate the pages
3572 * for both ranges because each compressed bio is tightly
3573 * coupled with a single extent map, and each range can have
3574 * an extent map with a different offset value relative to the
3575 * uncompressed data of our extent and different lengths. This
3576 * is a corner case so we prioritize correctness over
3577 * non-optimal behavior (submitting 2 bios for the same extent).
3579 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) &&
3580 prev_em_start && *prev_em_start != (u64)-1 &&
3581 *prev_em_start != em->start)
3582 force_bio_submit = true;
3585 *prev_em_start = em->start;
3587 free_extent_map(em);
3590 /* we've found a hole, just zero and go on */
3591 if (block_start == EXTENT_MAP_HOLE) {
3592 struct extent_state *cached = NULL;
3594 memzero_page(page, pg_offset, iosize);
3595 flush_dcache_page(page);
3597 set_extent_uptodate(tree, cur, cur + iosize - 1,
3599 unlock_extent_cached(tree, cur,
3600 cur + iosize - 1, &cached);
3601 end_page_read(page, true, cur, iosize);
3603 pg_offset += iosize;
3606 /* the get_extent function already copied into the page */
3607 if (test_range_bit(tree, cur, cur_end,
3608 EXTENT_UPTODATE, 1, NULL)) {
3609 check_page_uptodate(tree, page);
3610 unlock_extent(tree, cur, cur + iosize - 1);
3611 end_page_read(page, true, cur, iosize);
3613 pg_offset += iosize;
3616 /* we have an inline extent but it didn't get marked up
3617 * to date. Error out
3619 if (block_start == EXTENT_MAP_INLINE) {
3620 unlock_extent(tree, cur, cur + iosize - 1);
3621 end_page_read(page, false, cur, iosize);
3623 pg_offset += iosize;
3627 ret = submit_extent_page(REQ_OP_READ | read_flags, NULL,
3628 bio_ctrl, page, disk_bytenr, iosize,
3630 end_bio_extent_readpage, 0,
3636 unlock_extent(tree, cur, cur + iosize - 1);
3637 end_page_read(page, false, cur, iosize);
3641 pg_offset += iosize;
3647 static inline void contiguous_readpages(struct page *pages[], int nr_pages,
3649 struct extent_map **em_cached,
3650 struct btrfs_bio_ctrl *bio_ctrl,
3653 struct btrfs_inode *inode = BTRFS_I(pages[0]->mapping->host);
3656 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
3658 for (index = 0; index < nr_pages; index++) {
3659 btrfs_do_readpage(pages[index], em_cached, bio_ctrl,
3660 REQ_RAHEAD, prev_em_start);
3661 put_page(pages[index]);
3665 static void update_nr_written(struct writeback_control *wbc,
3666 unsigned long nr_written)
3668 wbc->nr_to_write -= nr_written;
3672 * helper for __extent_writepage, doing all of the delayed allocation setup.
3674 * This returns 1 if btrfs_run_delalloc_range function did all the work required
3675 * to write the page (copy into inline extent). In this case the IO has
3676 * been started and the page is already unlocked.
3678 * This returns 0 if all went well (page still locked)
3679 * This returns < 0 if there were errors (page still locked)
3681 static noinline_for_stack int writepage_delalloc(struct btrfs_inode *inode,
3682 struct page *page, struct writeback_control *wbc,
3683 u64 delalloc_start, unsigned long *nr_written)
3685 u64 page_end = delalloc_start + PAGE_SIZE - 1;
3687 u64 delalloc_to_write = 0;
3688 u64 delalloc_end = 0;
3690 int page_started = 0;
3693 while (delalloc_end < page_end) {
3694 found = find_lock_delalloc_range(&inode->vfs_inode, page,
3698 delalloc_start = delalloc_end + 1;
3701 ret = btrfs_run_delalloc_range(inode, page, delalloc_start,
3702 delalloc_end, &page_started, nr_written, wbc);
3706 * btrfs_run_delalloc_range should return < 0 for error
3707 * but just in case, we use > 0 here meaning the IO is
3708 * started, so we don't want to return > 0 unless
3709 * things are going well.
3711 return ret < 0 ? ret : -EIO;
3714 * delalloc_end is already one less than the total length, so
3715 * we don't subtract one from PAGE_SIZE
3717 delalloc_to_write += (delalloc_end - delalloc_start +
3718 PAGE_SIZE) >> PAGE_SHIFT;
3719 delalloc_start = delalloc_end + 1;
3721 if (wbc->nr_to_write < delalloc_to_write) {
3724 if (delalloc_to_write < thresh * 2)
3725 thresh = delalloc_to_write;
3726 wbc->nr_to_write = min_t(u64, delalloc_to_write,
3730 /* did the fill delalloc function already unlock and start
3735 * we've unlocked the page, so we can't update
3736 * the mapping's writeback index, just update
3739 wbc->nr_to_write -= *nr_written;
3747 * helper for __extent_writepage. This calls the writepage start hooks,
3748 * and does the loop to map the page into extents and bios.
3750 * We return 1 if the IO is started and the page is unlocked,
3751 * 0 if all went well (page still locked)
3752 * < 0 if there were errors (page still locked)
3754 static noinline_for_stack int __extent_writepage_io(struct btrfs_inode *inode,
3756 struct writeback_control *wbc,
3757 struct extent_page_data *epd,
3759 unsigned long nr_written,
3762 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3763 struct extent_io_tree *tree = &inode->io_tree;
3764 u64 start = page_offset(page);
3765 u64 end = start + PAGE_SIZE - 1;
3769 struct extent_map *em;
3772 u32 opf = REQ_OP_WRITE;
3773 const unsigned int write_flags = wbc_to_write_flags(wbc);
3776 ret = btrfs_writepage_cow_fixup(page, start, end);
3778 /* Fixup worker will requeue */
3779 redirty_page_for_writepage(wbc, page);
3780 update_nr_written(wbc, nr_written);
3786 * we don't want to touch the inode after unlocking the page,
3787 * so we update the mapping writeback index now
3789 update_nr_written(wbc, nr_written + 1);
3791 while (cur <= end) {
3796 if (cur >= i_size) {
3797 btrfs_writepage_endio_finish_ordered(page, cur, end, 1);
3800 em = btrfs_get_extent(inode, NULL, 0, cur, end - cur + 1);
3801 if (IS_ERR_OR_NULL(em)) {
3803 ret = PTR_ERR_OR_ZERO(em);
3807 extent_offset = cur - em->start;
3808 em_end = extent_map_end(em);
3809 ASSERT(cur <= em_end);
3811 ASSERT(IS_ALIGNED(em->start, fs_info->sectorsize));
3812 ASSERT(IS_ALIGNED(em->len, fs_info->sectorsize));
3813 block_start = em->block_start;
3814 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3815 disk_bytenr = em->block_start + extent_offset;
3817 /* Note that em_end from extent_map_end() is exclusive */
3818 iosize = min(em_end, end + 1) - cur;
3820 if (btrfs_use_zone_append(inode, em->block_start))
3821 opf = REQ_OP_ZONE_APPEND;
3823 free_extent_map(em);
3827 * compressed and inline extents are written through other
3830 if (compressed || block_start == EXTENT_MAP_HOLE ||
3831 block_start == EXTENT_MAP_INLINE) {
3835 btrfs_writepage_endio_finish_ordered(page, cur,
3836 cur + iosize - 1, 1);
3841 btrfs_set_range_writeback(tree, cur, cur + iosize - 1);
3842 if (!PageWriteback(page)) {
3843 btrfs_err(inode->root->fs_info,
3844 "page %lu not writeback, cur %llu end %llu",
3845 page->index, cur, end);
3848 ret = submit_extent_page(opf | write_flags, wbc,
3849 &epd->bio_ctrl, page,
3850 disk_bytenr, iosize,
3851 cur - page_offset(page),
3852 end_bio_extent_writepage,
3856 if (PageWriteback(page))
3857 end_page_writeback(page);
3868 * the writepage semantics are similar to regular writepage. extent
3869 * records are inserted to lock ranges in the tree, and as dirty areas
3870 * are found, they are marked writeback. Then the lock bits are removed
3871 * and the end_io handler clears the writeback ranges
3873 * Return 0 if everything goes well.
3874 * Return <0 for error.
3876 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
3877 struct extent_page_data *epd)
3879 struct inode *inode = page->mapping->host;
3880 u64 start = page_offset(page);
3881 u64 page_end = start + PAGE_SIZE - 1;
3885 loff_t i_size = i_size_read(inode);
3886 unsigned long end_index = i_size >> PAGE_SHIFT;
3887 unsigned long nr_written = 0;
3889 trace___extent_writepage(page, inode, wbc);
3891 WARN_ON(!PageLocked(page));
3893 ClearPageError(page);
3895 pg_offset = offset_in_page(i_size);
3896 if (page->index > end_index ||
3897 (page->index == end_index && !pg_offset)) {
3898 page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
3903 if (page->index == end_index) {
3904 memzero_page(page, pg_offset, PAGE_SIZE - pg_offset);
3905 flush_dcache_page(page);
3908 ret = set_page_extent_mapped(page);
3914 if (!epd->extent_locked) {
3915 ret = writepage_delalloc(BTRFS_I(inode), page, wbc, start,
3923 ret = __extent_writepage_io(BTRFS_I(inode), page, wbc, epd, i_size,
3930 /* make sure the mapping tag for page dirty gets cleared */
3931 set_page_writeback(page);
3932 end_page_writeback(page);
3934 if (PageError(page)) {
3935 ret = ret < 0 ? ret : -EIO;
3936 end_extent_writepage(page, ret, start, page_end);
3943 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3945 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
3946 TASK_UNINTERRUPTIBLE);
3949 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3951 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3952 smp_mb__after_atomic();
3953 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3957 * Lock extent buffer status and pages for writeback.
3959 * May try to flush write bio if we can't get the lock.
3961 * Return 0 if the extent buffer doesn't need to be submitted.
3962 * (E.g. the extent buffer is not dirty)
3963 * Return >0 is the extent buffer is submitted to bio.
3964 * Return <0 if something went wrong, no page is locked.
3966 static noinline_for_stack int lock_extent_buffer_for_io(struct extent_buffer *eb,
3967 struct extent_page_data *epd)
3969 struct btrfs_fs_info *fs_info = eb->fs_info;
3970 int i, num_pages, failed_page_nr;
3974 if (!btrfs_try_tree_write_lock(eb)) {
3975 ret = flush_write_bio(epd);
3979 btrfs_tree_lock(eb);
3982 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3983 btrfs_tree_unlock(eb);
3987 ret = flush_write_bio(epd);
3993 wait_on_extent_buffer_writeback(eb);
3994 btrfs_tree_lock(eb);
3995 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3997 btrfs_tree_unlock(eb);
4002 * We need to do this to prevent races in people who check if the eb is
4003 * under IO since we can end up having no IO bits set for a short period
4006 spin_lock(&eb->refs_lock);
4007 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
4008 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
4009 spin_unlock(&eb->refs_lock);
4010 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
4011 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4013 fs_info->dirty_metadata_batch);
4016 spin_unlock(&eb->refs_lock);
4019 btrfs_tree_unlock(eb);
4022 * Either we don't need to submit any tree block, or we're submitting
4024 * Subpage metadata doesn't use page locking at all, so we can skip
4027 if (!ret || fs_info->sectorsize < PAGE_SIZE)
4030 num_pages = num_extent_pages(eb);
4031 for (i = 0; i < num_pages; i++) {
4032 struct page *p = eb->pages[i];
4034 if (!trylock_page(p)) {
4038 err = flush_write_bio(epd);
4052 /* Unlock already locked pages */
4053 for (i = 0; i < failed_page_nr; i++)
4054 unlock_page(eb->pages[i]);
4056 * Clear EXTENT_BUFFER_WRITEBACK and wake up anyone waiting on it.
4057 * Also set back EXTENT_BUFFER_DIRTY so future attempts to this eb can
4058 * be made and undo everything done before.
4060 btrfs_tree_lock(eb);
4061 spin_lock(&eb->refs_lock);
4062 set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
4063 end_extent_buffer_writeback(eb);
4064 spin_unlock(&eb->refs_lock);
4065 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, eb->len,
4066 fs_info->dirty_metadata_batch);
4067 btrfs_clear_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
4068 btrfs_tree_unlock(eb);
4072 static void set_btree_ioerr(struct page *page, struct extent_buffer *eb)
4074 struct btrfs_fs_info *fs_info = eb->fs_info;
4076 btrfs_page_set_error(fs_info, page, eb->start, eb->len);
4077 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))
4081 * If we error out, we should add back the dirty_metadata_bytes
4082 * to make it consistent.
4084 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4085 eb->len, fs_info->dirty_metadata_batch);
4088 * If writeback for a btree extent that doesn't belong to a log tree
4089 * failed, increment the counter transaction->eb_write_errors.
4090 * We do this because while the transaction is running and before it's
4091 * committing (when we call filemap_fdata[write|wait]_range against
4092 * the btree inode), we might have
4093 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
4094 * returns an error or an error happens during writeback, when we're
4095 * committing the transaction we wouldn't know about it, since the pages
4096 * can be no longer dirty nor marked anymore for writeback (if a
4097 * subsequent modification to the extent buffer didn't happen before the
4098 * transaction commit), which makes filemap_fdata[write|wait]_range not
4099 * able to find the pages tagged with SetPageError at transaction
4100 * commit time. So if this happens we must abort the transaction,
4101 * otherwise we commit a super block with btree roots that point to
4102 * btree nodes/leafs whose content on disk is invalid - either garbage
4103 * or the content of some node/leaf from a past generation that got
4104 * cowed or deleted and is no longer valid.
4106 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
4107 * not be enough - we need to distinguish between log tree extents vs
4108 * non-log tree extents, and the next filemap_fdatawait_range() call
4109 * will catch and clear such errors in the mapping - and that call might
4110 * be from a log sync and not from a transaction commit. Also, checking
4111 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
4112 * not done and would not be reliable - the eb might have been released
4113 * from memory and reading it back again means that flag would not be
4114 * set (since it's a runtime flag, not persisted on disk).
4116 * Using the flags below in the btree inode also makes us achieve the
4117 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
4118 * writeback for all dirty pages and before filemap_fdatawait_range()
4119 * is called, the writeback for all dirty pages had already finished
4120 * with errors - because we were not using AS_EIO/AS_ENOSPC,
4121 * filemap_fdatawait_range() would return success, as it could not know
4122 * that writeback errors happened (the pages were no longer tagged for
4125 switch (eb->log_index) {
4127 set_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags);
4130 set_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags);
4133 set_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags);
4136 BUG(); /* unexpected, logic error */
4141 * The endio specific version which won't touch any unsafe spinlock in endio
4144 static struct extent_buffer *find_extent_buffer_nolock(
4145 struct btrfs_fs_info *fs_info, u64 start)
4147 struct extent_buffer *eb;
4150 eb = radix_tree_lookup(&fs_info->buffer_radix,
4151 start >> fs_info->sectorsize_bits);
4152 if (eb && atomic_inc_not_zero(&eb->refs)) {
4161 * The endio function for subpage extent buffer write.
4163 * Unlike end_bio_extent_buffer_writepage(), we only call end_page_writeback()
4164 * after all extent buffers in the page has finished their writeback.
4166 static void end_bio_subpage_eb_writepage(struct btrfs_fs_info *fs_info,
4169 struct bio_vec *bvec;
4170 struct bvec_iter_all iter_all;
4172 ASSERT(!bio_flagged(bio, BIO_CLONED));
4173 bio_for_each_segment_all(bvec, bio, iter_all) {
4174 struct page *page = bvec->bv_page;
4175 u64 bvec_start = page_offset(page) + bvec->bv_offset;
4176 u64 bvec_end = bvec_start + bvec->bv_len - 1;
4177 u64 cur_bytenr = bvec_start;
4179 ASSERT(IS_ALIGNED(bvec->bv_len, fs_info->nodesize));
4181 /* Iterate through all extent buffers in the range */
4182 while (cur_bytenr <= bvec_end) {
4183 struct extent_buffer *eb;
4187 * Here we can't use find_extent_buffer(), as it may
4188 * try to lock eb->refs_lock, which is not safe in endio
4191 eb = find_extent_buffer_nolock(fs_info, cur_bytenr);
4194 cur_bytenr = eb->start + eb->len;
4196 ASSERT(test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags));
4197 done = atomic_dec_and_test(&eb->io_pages);
4200 if (bio->bi_status ||
4201 test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) {
4202 ClearPageUptodate(page);
4203 set_btree_ioerr(page, eb);
4206 btrfs_subpage_clear_writeback(fs_info, page, eb->start,
4208 end_extent_buffer_writeback(eb);
4210 * free_extent_buffer() will grab spinlock which is not
4211 * safe in endio context. Thus here we manually dec
4214 atomic_dec(&eb->refs);
4220 static void end_bio_extent_buffer_writepage(struct bio *bio)
4222 struct btrfs_fs_info *fs_info;
4223 struct bio_vec *bvec;
4224 struct extent_buffer *eb;
4226 struct bvec_iter_all iter_all;
4228 fs_info = btrfs_sb(bio_first_page_all(bio)->mapping->host->i_sb);
4229 if (fs_info->sectorsize < PAGE_SIZE)
4230 return end_bio_subpage_eb_writepage(fs_info, bio);
4232 ASSERT(!bio_flagged(bio, BIO_CLONED));
4233 bio_for_each_segment_all(bvec, bio, iter_all) {
4234 struct page *page = bvec->bv_page;
4236 eb = (struct extent_buffer *)page->private;
4238 done = atomic_dec_and_test(&eb->io_pages);
4240 if (bio->bi_status ||
4241 test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) {
4242 ClearPageUptodate(page);
4243 set_btree_ioerr(page, eb);
4246 end_page_writeback(page);
4251 end_extent_buffer_writeback(eb);
4258 * Unlike the work in write_one_eb(), we rely completely on extent locking.
4259 * Page locking is only utilized at minimum to keep the VMM code happy.
4261 * Caller should still call write_one_eb() other than this function directly.
4262 * As write_one_eb() has extra preparation before submitting the extent buffer.
4264 static int write_one_subpage_eb(struct extent_buffer *eb,
4265 struct writeback_control *wbc,
4266 struct extent_page_data *epd)
4268 struct btrfs_fs_info *fs_info = eb->fs_info;
4269 struct page *page = eb->pages[0];
4270 unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META;
4271 bool no_dirty_ebs = false;
4274 /* clear_page_dirty_for_io() in subpage helper needs page locked */
4276 btrfs_subpage_set_writeback(fs_info, page, eb->start, eb->len);
4278 /* Check if this is the last dirty bit to update nr_written */
4279 no_dirty_ebs = btrfs_subpage_clear_and_test_dirty(fs_info, page,
4280 eb->start, eb->len);
4282 clear_page_dirty_for_io(page);
4284 ret = submit_extent_page(REQ_OP_WRITE | write_flags, wbc,
4285 &epd->bio_ctrl, page, eb->start, eb->len,
4286 eb->start - page_offset(page),
4287 end_bio_extent_buffer_writepage, 0, 0, false);
4289 btrfs_subpage_clear_writeback(fs_info, page, eb->start, eb->len);
4290 set_btree_ioerr(page, eb);
4293 if (atomic_dec_and_test(&eb->io_pages))
4294 end_extent_buffer_writeback(eb);
4299 * Submission finished without problem, if no range of the page is
4300 * dirty anymore, we have submitted a page. Update nr_written in wbc.
4303 update_nr_written(wbc, 1);
4307 static noinline_for_stack int write_one_eb(struct extent_buffer *eb,
4308 struct writeback_control *wbc,
4309 struct extent_page_data *epd)
4311 u64 disk_bytenr = eb->start;
4314 unsigned long start, end;
4315 unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META;
4318 clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
4319 num_pages = num_extent_pages(eb);
4320 atomic_set(&eb->io_pages, num_pages);
4322 /* set btree blocks beyond nritems with 0 to avoid stale content. */
4323 nritems = btrfs_header_nritems(eb);
4324 if (btrfs_header_level(eb) > 0) {
4325 end = btrfs_node_key_ptr_offset(nritems);
4327 memzero_extent_buffer(eb, end, eb->len - end);
4331 * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0
4333 start = btrfs_item_nr_offset(nritems);
4334 end = BTRFS_LEAF_DATA_OFFSET + leaf_data_end(eb);
4335 memzero_extent_buffer(eb, start, end - start);
4338 if (eb->fs_info->sectorsize < PAGE_SIZE)
4339 return write_one_subpage_eb(eb, wbc, epd);
4341 for (i = 0; i < num_pages; i++) {
4342 struct page *p = eb->pages[i];
4344 clear_page_dirty_for_io(p);
4345 set_page_writeback(p);
4346 ret = submit_extent_page(REQ_OP_WRITE | write_flags, wbc,
4347 &epd->bio_ctrl, p, disk_bytenr,
4349 end_bio_extent_buffer_writepage,
4352 set_btree_ioerr(p, eb);
4353 if (PageWriteback(p))
4354 end_page_writeback(p);
4355 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
4356 end_extent_buffer_writeback(eb);
4360 disk_bytenr += PAGE_SIZE;
4361 update_nr_written(wbc, 1);
4365 if (unlikely(ret)) {
4366 for (; i < num_pages; i++) {
4367 struct page *p = eb->pages[i];
4368 clear_page_dirty_for_io(p);
4377 * Submit one subpage btree page.
4379 * The main difference to submit_eb_page() is:
4381 * For subpage, we don't rely on page locking at all.
4384 * We only flush bio if we may be unable to fit current extent buffers into
4387 * Return >=0 for the number of submitted extent buffers.
4388 * Return <0 for fatal error.
4390 static int submit_eb_subpage(struct page *page,
4391 struct writeback_control *wbc,
4392 struct extent_page_data *epd)
4394 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
4396 u64 page_start = page_offset(page);
4398 const int nbits = BTRFS_SUBPAGE_BITMAP_SIZE;
4399 int sectors_per_node = fs_info->nodesize >> fs_info->sectorsize_bits;
4402 /* Lock and write each dirty extent buffers in the range */
4403 while (bit_start < nbits) {
4404 struct btrfs_subpage *subpage = (struct btrfs_subpage *)page->private;
4405 struct extent_buffer *eb;
4406 unsigned long flags;
4410 * Take private lock to ensure the subpage won't be detached
4413 spin_lock(&page->mapping->private_lock);
4414 if (!PagePrivate(page)) {
4415 spin_unlock(&page->mapping->private_lock);
4418 spin_lock_irqsave(&subpage->lock, flags);
4419 if (!((1 << bit_start) & subpage->dirty_bitmap)) {
4420 spin_unlock_irqrestore(&subpage->lock, flags);
4421 spin_unlock(&page->mapping->private_lock);
4426 start = page_start + bit_start * fs_info->sectorsize;
4427 bit_start += sectors_per_node;
4430 * Here we just want to grab the eb without touching extra
4431 * spin locks, so call find_extent_buffer_nolock().
4433 eb = find_extent_buffer_nolock(fs_info, start);
4434 spin_unlock_irqrestore(&subpage->lock, flags);
4435 spin_unlock(&page->mapping->private_lock);
4438 * The eb has already reached 0 refs thus find_extent_buffer()
4439 * doesn't return it. We don't need to write back such eb
4445 ret = lock_extent_buffer_for_io(eb, epd);
4447 free_extent_buffer(eb);
4451 free_extent_buffer(eb);
4454 ret = write_one_eb(eb, wbc, epd);
4455 free_extent_buffer(eb);
4463 /* We hit error, end bio for the submitted extent buffers */
4464 end_write_bio(epd, ret);
4469 * Submit all page(s) of one extent buffer.
4471 * @page: the page of one extent buffer
4472 * @eb_context: to determine if we need to submit this page, if current page
4473 * belongs to this eb, we don't need to submit
4475 * The caller should pass each page in their bytenr order, and here we use
4476 * @eb_context to determine if we have submitted pages of one extent buffer.
4478 * If we have, we just skip until we hit a new page that doesn't belong to
4479 * current @eb_context.
4481 * If not, we submit all the page(s) of the extent buffer.
4483 * Return >0 if we have submitted the extent buffer successfully.
4484 * Return 0 if we don't need to submit the page, as it's already submitted by
4486 * Return <0 for fatal error.
4488 static int submit_eb_page(struct page *page, struct writeback_control *wbc,
4489 struct extent_page_data *epd,
4490 struct extent_buffer **eb_context)
4492 struct address_space *mapping = page->mapping;
4493 struct btrfs_block_group *cache = NULL;
4494 struct extent_buffer *eb;
4497 if (!PagePrivate(page))
4500 if (btrfs_sb(page->mapping->host->i_sb)->sectorsize < PAGE_SIZE)
4501 return submit_eb_subpage(page, wbc, epd);
4503 spin_lock(&mapping->private_lock);
4504 if (!PagePrivate(page)) {
4505 spin_unlock(&mapping->private_lock);
4509 eb = (struct extent_buffer *)page->private;
4512 * Shouldn't happen and normally this would be a BUG_ON but no point
4513 * crashing the machine for something we can survive anyway.
4516 spin_unlock(&mapping->private_lock);
4520 if (eb == *eb_context) {
4521 spin_unlock(&mapping->private_lock);
4524 ret = atomic_inc_not_zero(&eb->refs);
4525 spin_unlock(&mapping->private_lock);
4529 if (!btrfs_check_meta_write_pointer(eb->fs_info, eb, &cache)) {
4531 * If for_sync, this hole will be filled with
4532 * trasnsaction commit.
4534 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
4538 free_extent_buffer(eb);
4544 ret = lock_extent_buffer_for_io(eb, epd);
4546 btrfs_revert_meta_write_pointer(cache, eb);
4548 btrfs_put_block_group(cache);
4549 free_extent_buffer(eb);
4553 btrfs_put_block_group(cache);
4554 ret = write_one_eb(eb, wbc, epd);
4555 free_extent_buffer(eb);
4561 int btree_write_cache_pages(struct address_space *mapping,
4562 struct writeback_control *wbc)
4564 struct extent_buffer *eb_context = NULL;
4565 struct extent_page_data epd = {
4568 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4570 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
4573 int nr_to_write_done = 0;
4574 struct pagevec pvec;
4577 pgoff_t end; /* Inclusive */
4581 pagevec_init(&pvec);
4582 if (wbc->range_cyclic) {
4583 index = mapping->writeback_index; /* Start from prev offset */
4586 * Start from the beginning does not need to cycle over the
4587 * range, mark it as scanned.
4589 scanned = (index == 0);
4591 index = wbc->range_start >> PAGE_SHIFT;
4592 end = wbc->range_end >> PAGE_SHIFT;
4595 if (wbc->sync_mode == WB_SYNC_ALL)
4596 tag = PAGECACHE_TAG_TOWRITE;
4598 tag = PAGECACHE_TAG_DIRTY;
4599 btrfs_zoned_meta_io_lock(fs_info);
4601 if (wbc->sync_mode == WB_SYNC_ALL)
4602 tag_pages_for_writeback(mapping, index, end);
4603 while (!done && !nr_to_write_done && (index <= end) &&
4604 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
4608 for (i = 0; i < nr_pages; i++) {
4609 struct page *page = pvec.pages[i];
4611 ret = submit_eb_page(page, wbc, &epd, &eb_context);
4620 * the filesystem may choose to bump up nr_to_write.
4621 * We have to make sure to honor the new nr_to_write
4624 nr_to_write_done = wbc->nr_to_write <= 0;
4626 pagevec_release(&pvec);
4629 if (!scanned && !done) {
4631 * We hit the last page and there is more work to be done: wrap
4632 * back to the start of the file
4639 end_write_bio(&epd, ret);
4643 * If something went wrong, don't allow any metadata write bio to be
4646 * This would prevent use-after-free if we had dirty pages not
4647 * cleaned up, which can still happen by fuzzed images.
4650 * Allowing existing tree block to be allocated for other trees.
4652 * - Log tree operations
4653 * Exiting tree blocks get allocated to log tree, bumps its
4654 * generation, then get cleaned in tree re-balance.
4655 * Such tree block will not be written back, since it's clean,
4656 * thus no WRITTEN flag set.
4657 * And after log writes back, this tree block is not traced by
4658 * any dirty extent_io_tree.
4660 * - Offending tree block gets re-dirtied from its original owner
4661 * Since it has bumped generation, no WRITTEN flag, it can be
4662 * reused without COWing. This tree block will not be traced
4663 * by btrfs_transaction::dirty_pages.
4665 * Now such dirty tree block will not be cleaned by any dirty
4666 * extent io tree. Thus we don't want to submit such wild eb
4667 * if the fs already has error.
4669 if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
4670 ret = flush_write_bio(&epd);
4673 end_write_bio(&epd, ret);
4676 btrfs_zoned_meta_io_unlock(fs_info);
4681 * Walk the list of dirty pages of the given address space and write all of them.
4683 * @mapping: address space structure to write
4684 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
4685 * @epd: holds context for the write, namely the bio
4687 * If a page is already under I/O, write_cache_pages() skips it, even
4688 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
4689 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
4690 * and msync() need to guarantee that all the data which was dirty at the time
4691 * the call was made get new I/O started against them. If wbc->sync_mode is
4692 * WB_SYNC_ALL then we were called for data integrity and we must wait for
4693 * existing IO to complete.
4695 static int extent_write_cache_pages(struct address_space *mapping,
4696 struct writeback_control *wbc,
4697 struct extent_page_data *epd)
4699 struct inode *inode = mapping->host;
4702 int nr_to_write_done = 0;
4703 struct pagevec pvec;
4706 pgoff_t end; /* Inclusive */
4708 int range_whole = 0;
4713 * We have to hold onto the inode so that ordered extents can do their
4714 * work when the IO finishes. The alternative to this is failing to add
4715 * an ordered extent if the igrab() fails there and that is a huge pain
4716 * to deal with, so instead just hold onto the inode throughout the
4717 * writepages operation. If it fails here we are freeing up the inode
4718 * anyway and we'd rather not waste our time writing out stuff that is
4719 * going to be truncated anyway.
4724 pagevec_init(&pvec);
4725 if (wbc->range_cyclic) {
4726 index = mapping->writeback_index; /* Start from prev offset */
4729 * Start from the beginning does not need to cycle over the
4730 * range, mark it as scanned.
4732 scanned = (index == 0);
4734 index = wbc->range_start >> PAGE_SHIFT;
4735 end = wbc->range_end >> PAGE_SHIFT;
4736 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
4742 * We do the tagged writepage as long as the snapshot flush bit is set
4743 * and we are the first one who do the filemap_flush() on this inode.
4745 * The nr_to_write == LONG_MAX is needed to make sure other flushers do
4746 * not race in and drop the bit.
4748 if (range_whole && wbc->nr_to_write == LONG_MAX &&
4749 test_and_clear_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
4750 &BTRFS_I(inode)->runtime_flags))
4751 wbc->tagged_writepages = 1;
4753 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
4754 tag = PAGECACHE_TAG_TOWRITE;
4756 tag = PAGECACHE_TAG_DIRTY;
4758 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
4759 tag_pages_for_writeback(mapping, index, end);
4761 while (!done && !nr_to_write_done && (index <= end) &&
4762 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping,
4763 &index, end, tag))) {
4766 for (i = 0; i < nr_pages; i++) {
4767 struct page *page = pvec.pages[i];
4769 done_index = page->index + 1;
4771 * At this point we hold neither the i_pages lock nor
4772 * the page lock: the page may be truncated or
4773 * invalidated (changing page->mapping to NULL),
4774 * or even swizzled back from swapper_space to
4775 * tmpfs file mapping
4777 if (!trylock_page(page)) {
4778 ret = flush_write_bio(epd);
4783 if (unlikely(page->mapping != mapping)) {
4788 if (wbc->sync_mode != WB_SYNC_NONE) {
4789 if (PageWriteback(page)) {
4790 ret = flush_write_bio(epd);
4793 wait_on_page_writeback(page);
4796 if (PageWriteback(page) ||
4797 !clear_page_dirty_for_io(page)) {
4802 ret = __extent_writepage(page, wbc, epd);
4809 * the filesystem may choose to bump up nr_to_write.
4810 * We have to make sure to honor the new nr_to_write
4813 nr_to_write_done = wbc->nr_to_write <= 0;
4815 pagevec_release(&pvec);
4818 if (!scanned && !done) {
4820 * We hit the last page and there is more work to be done: wrap
4821 * back to the start of the file
4827 * If we're looping we could run into a page that is locked by a
4828 * writer and that writer could be waiting on writeback for a
4829 * page in our current bio, and thus deadlock, so flush the
4832 ret = flush_write_bio(epd);
4837 if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole))
4838 mapping->writeback_index = done_index;
4840 btrfs_add_delayed_iput(inode);
4844 int extent_write_full_page(struct page *page, struct writeback_control *wbc)
4847 struct extent_page_data epd = {
4850 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4853 ret = __extent_writepage(page, wbc, &epd);
4856 end_write_bio(&epd, ret);
4860 ret = flush_write_bio(&epd);
4865 int extent_write_locked_range(struct inode *inode, u64 start, u64 end,
4869 struct address_space *mapping = inode->i_mapping;
4871 unsigned long nr_pages = (end - start + PAGE_SIZE) >>
4874 struct extent_page_data epd = {
4877 .sync_io = mode == WB_SYNC_ALL,
4879 struct writeback_control wbc_writepages = {
4881 .nr_to_write = nr_pages * 2,
4882 .range_start = start,
4883 .range_end = end + 1,
4884 /* We're called from an async helper function */
4885 .punt_to_cgroup = 1,
4886 .no_cgroup_owner = 1,
4889 wbc_attach_fdatawrite_inode(&wbc_writepages, inode);
4890 while (start <= end) {
4891 page = find_get_page(mapping, start >> PAGE_SHIFT);
4892 if (clear_page_dirty_for_io(page))
4893 ret = __extent_writepage(page, &wbc_writepages, &epd);
4895 btrfs_writepage_endio_finish_ordered(page, start,
4896 start + PAGE_SIZE - 1, 1);
4905 ret = flush_write_bio(&epd);
4907 end_write_bio(&epd, ret);
4909 wbc_detach_inode(&wbc_writepages);
4913 int extent_writepages(struct address_space *mapping,
4914 struct writeback_control *wbc)
4917 struct extent_page_data epd = {
4920 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4923 ret = extent_write_cache_pages(mapping, wbc, &epd);
4926 end_write_bio(&epd, ret);
4929 ret = flush_write_bio(&epd);
4933 void extent_readahead(struct readahead_control *rac)
4935 struct btrfs_bio_ctrl bio_ctrl = { 0 };
4936 struct page *pagepool[16];
4937 struct extent_map *em_cached = NULL;
4938 u64 prev_em_start = (u64)-1;
4941 while ((nr = readahead_page_batch(rac, pagepool))) {
4942 u64 contig_start = readahead_pos(rac);
4943 u64 contig_end = contig_start + readahead_batch_length(rac) - 1;
4945 contiguous_readpages(pagepool, nr, contig_start, contig_end,
4946 &em_cached, &bio_ctrl, &prev_em_start);
4950 free_extent_map(em_cached);
4953 if (submit_one_bio(bio_ctrl.bio, 0, bio_ctrl.bio_flags))
4959 * basic invalidatepage code, this waits on any locked or writeback
4960 * ranges corresponding to the page, and then deletes any extent state
4961 * records from the tree
4963 int extent_invalidatepage(struct extent_io_tree *tree,
4964 struct page *page, unsigned long offset)
4966 struct extent_state *cached_state = NULL;
4967 u64 start = page_offset(page);
4968 u64 end = start + PAGE_SIZE - 1;
4969 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
4971 /* This function is only called for the btree inode */
4972 ASSERT(tree->owner == IO_TREE_BTREE_INODE_IO);
4974 start += ALIGN(offset, blocksize);
4978 lock_extent_bits(tree, start, end, &cached_state);
4979 wait_on_page_writeback(page);
4982 * Currently for btree io tree, only EXTENT_LOCKED is utilized,
4983 * so here we only need to unlock the extent range to free any
4984 * existing extent state.
4986 unlock_extent_cached(tree, start, end, &cached_state);
4991 * a helper for releasepage, this tests for areas of the page that
4992 * are locked or under IO and drops the related state bits if it is safe
4995 static int try_release_extent_state(struct extent_io_tree *tree,
4996 struct page *page, gfp_t mask)
4998 u64 start = page_offset(page);
4999 u64 end = start + PAGE_SIZE - 1;
5002 if (test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL)) {
5006 * At this point we can safely clear everything except the
5007 * locked bit, the nodatasum bit and the delalloc new bit.
5008 * The delalloc new bit will be cleared by ordered extent
5011 ret = __clear_extent_bit(tree, start, end,
5012 ~(EXTENT_LOCKED | EXTENT_NODATASUM | EXTENT_DELALLOC_NEW),
5013 0, 0, NULL, mask, NULL);
5015 /* if clear_extent_bit failed for enomem reasons,
5016 * we can't allow the release to continue.
5027 * a helper for releasepage. As long as there are no locked extents
5028 * in the range corresponding to the page, both state records and extent
5029 * map records are removed
5031 int try_release_extent_mapping(struct page *page, gfp_t mask)
5033 struct extent_map *em;
5034 u64 start = page_offset(page);
5035 u64 end = start + PAGE_SIZE - 1;
5036 struct btrfs_inode *btrfs_inode = BTRFS_I(page->mapping->host);
5037 struct extent_io_tree *tree = &btrfs_inode->io_tree;
5038 struct extent_map_tree *map = &btrfs_inode->extent_tree;
5040 if (gfpflags_allow_blocking(mask) &&
5041 page->mapping->host->i_size > SZ_16M) {
5043 while (start <= end) {
5044 struct btrfs_fs_info *fs_info;
5047 len = end - start + 1;
5048 write_lock(&map->lock);
5049 em = lookup_extent_mapping(map, start, len);
5051 write_unlock(&map->lock);
5054 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
5055 em->start != start) {
5056 write_unlock(&map->lock);
5057 free_extent_map(em);
5060 if (test_range_bit(tree, em->start,
5061 extent_map_end(em) - 1,
5062 EXTENT_LOCKED, 0, NULL))
5065 * If it's not in the list of modified extents, used
5066 * by a fast fsync, we can remove it. If it's being
5067 * logged we can safely remove it since fsync took an
5068 * extra reference on the em.
5070 if (list_empty(&em->list) ||
5071 test_bit(EXTENT_FLAG_LOGGING, &em->flags))
5074 * If it's in the list of modified extents, remove it
5075 * only if its generation is older then the current one,
5076 * in which case we don't need it for a fast fsync.
5077 * Otherwise don't remove it, we could be racing with an
5078 * ongoing fast fsync that could miss the new extent.
5080 fs_info = btrfs_inode->root->fs_info;
5081 spin_lock(&fs_info->trans_lock);
5082 cur_gen = fs_info->generation;
5083 spin_unlock(&fs_info->trans_lock);
5084 if (em->generation >= cur_gen)
5088 * We only remove extent maps that are not in the list of
5089 * modified extents or that are in the list but with a
5090 * generation lower then the current generation, so there
5091 * is no need to set the full fsync flag on the inode (it
5092 * hurts the fsync performance for workloads with a data
5093 * size that exceeds or is close to the system's memory).
5095 remove_extent_mapping(map, em);
5096 /* once for the rb tree */
5097 free_extent_map(em);
5099 start = extent_map_end(em);
5100 write_unlock(&map->lock);
5103 free_extent_map(em);
5105 cond_resched(); /* Allow large-extent preemption. */
5108 return try_release_extent_state(tree, page, mask);
5112 * helper function for fiemap, which doesn't want to see any holes.
5113 * This maps until we find something past 'last'
5115 static struct extent_map *get_extent_skip_holes(struct btrfs_inode *inode,
5116 u64 offset, u64 last)
5118 u64 sectorsize = btrfs_inode_sectorsize(inode);
5119 struct extent_map *em;
5126 len = last - offset;
5129 len = ALIGN(len, sectorsize);
5130 em = btrfs_get_extent_fiemap(inode, offset, len);
5131 if (IS_ERR_OR_NULL(em))
5134 /* if this isn't a hole return it */
5135 if (em->block_start != EXTENT_MAP_HOLE)
5138 /* this is a hole, advance to the next extent */
5139 offset = extent_map_end(em);
5140 free_extent_map(em);
5148 * To cache previous fiemap extent
5150 * Will be used for merging fiemap extent
5152 struct fiemap_cache {
5161 * Helper to submit fiemap extent.
5163 * Will try to merge current fiemap extent specified by @offset, @phys,
5164 * @len and @flags with cached one.
5165 * And only when we fails to merge, cached one will be submitted as
5168 * Return value is the same as fiemap_fill_next_extent().
5170 static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
5171 struct fiemap_cache *cache,
5172 u64 offset, u64 phys, u64 len, u32 flags)
5180 * Sanity check, extent_fiemap() should have ensured that new
5181 * fiemap extent won't overlap with cached one.
5184 * NOTE: Physical address can overlap, due to compression
5186 if (cache->offset + cache->len > offset) {
5192 * Only merges fiemap extents if
5193 * 1) Their logical addresses are continuous
5195 * 2) Their physical addresses are continuous
5196 * So truly compressed (physical size smaller than logical size)
5197 * extents won't get merged with each other
5199 * 3) Share same flags except FIEMAP_EXTENT_LAST
5200 * So regular extent won't get merged with prealloc extent
5202 if (cache->offset + cache->len == offset &&
5203 cache->phys + cache->len == phys &&
5204 (cache->flags & ~FIEMAP_EXTENT_LAST) ==
5205 (flags & ~FIEMAP_EXTENT_LAST)) {
5207 cache->flags |= flags;
5208 goto try_submit_last;
5211 /* Not mergeable, need to submit cached one */
5212 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
5213 cache->len, cache->flags);
5214 cache->cached = false;
5218 cache->cached = true;
5219 cache->offset = offset;
5222 cache->flags = flags;
5224 if (cache->flags & FIEMAP_EXTENT_LAST) {
5225 ret = fiemap_fill_next_extent(fieinfo, cache->offset,
5226 cache->phys, cache->len, cache->flags);
5227 cache->cached = false;
5233 * Emit last fiemap cache
5235 * The last fiemap cache may still be cached in the following case:
5237 * |<- Fiemap range ->|
5238 * |<------------ First extent ----------->|
5240 * In this case, the first extent range will be cached but not emitted.
5241 * So we must emit it before ending extent_fiemap().
5243 static int emit_last_fiemap_cache(struct fiemap_extent_info *fieinfo,
5244 struct fiemap_cache *cache)
5251 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
5252 cache->len, cache->flags);
5253 cache->cached = false;
5259 int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,
5264 u64 max = start + len;
5268 u64 last_for_get_extent = 0;
5270 u64 isize = i_size_read(&inode->vfs_inode);
5271 struct btrfs_key found_key;
5272 struct extent_map *em = NULL;
5273 struct extent_state *cached_state = NULL;
5274 struct btrfs_path *path;
5275 struct btrfs_root *root = inode->root;
5276 struct fiemap_cache cache = { 0 };
5277 struct ulist *roots;
5278 struct ulist *tmp_ulist;
5287 path = btrfs_alloc_path();
5291 roots = ulist_alloc(GFP_KERNEL);
5292 tmp_ulist = ulist_alloc(GFP_KERNEL);
5293 if (!roots || !tmp_ulist) {
5295 goto out_free_ulist;
5299 * We can't initialize that to 'start' as this could miss extents due
5300 * to extent item merging
5303 start = round_down(start, btrfs_inode_sectorsize(inode));
5304 len = round_up(max, btrfs_inode_sectorsize(inode)) - start;
5307 * lookup the last file extent. We're not using i_size here
5308 * because there might be preallocation past i_size
5310 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode), -1,
5313 goto out_free_ulist;
5321 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
5322 found_type = found_key.type;
5324 /* No extents, but there might be delalloc bits */
5325 if (found_key.objectid != btrfs_ino(inode) ||
5326 found_type != BTRFS_EXTENT_DATA_KEY) {
5327 /* have to trust i_size as the end */
5329 last_for_get_extent = isize;
5332 * remember the start of the last extent. There are a
5333 * bunch of different factors that go into the length of the
5334 * extent, so its much less complex to remember where it started
5336 last = found_key.offset;
5337 last_for_get_extent = last + 1;
5339 btrfs_release_path(path);
5342 * we might have some extents allocated but more delalloc past those
5343 * extents. so, we trust isize unless the start of the last extent is
5348 last_for_get_extent = isize;
5351 lock_extent_bits(&inode->io_tree, start, start + len - 1,
5354 em = get_extent_skip_holes(inode, start, last_for_get_extent);
5363 u64 offset_in_extent = 0;
5365 /* break if the extent we found is outside the range */
5366 if (em->start >= max || extent_map_end(em) < off)
5370 * get_extent may return an extent that starts before our
5371 * requested range. We have to make sure the ranges
5372 * we return to fiemap always move forward and don't
5373 * overlap, so adjust the offsets here
5375 em_start = max(em->start, off);
5378 * record the offset from the start of the extent
5379 * for adjusting the disk offset below. Only do this if the
5380 * extent isn't compressed since our in ram offset may be past
5381 * what we have actually allocated on disk.
5383 if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
5384 offset_in_extent = em_start - em->start;
5385 em_end = extent_map_end(em);
5386 em_len = em_end - em_start;
5388 if (em->block_start < EXTENT_MAP_LAST_BYTE)
5389 disko = em->block_start + offset_in_extent;
5394 * bump off for our next call to get_extent
5396 off = extent_map_end(em);
5400 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
5402 flags |= FIEMAP_EXTENT_LAST;
5403 } else if (em->block_start == EXTENT_MAP_INLINE) {
5404 flags |= (FIEMAP_EXTENT_DATA_INLINE |
5405 FIEMAP_EXTENT_NOT_ALIGNED);
5406 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
5407 flags |= (FIEMAP_EXTENT_DELALLOC |
5408 FIEMAP_EXTENT_UNKNOWN);
5409 } else if (fieinfo->fi_extents_max) {
5410 u64 bytenr = em->block_start -
5411 (em->start - em->orig_start);
5414 * As btrfs supports shared space, this information
5415 * can be exported to userspace tools via
5416 * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0
5417 * then we're just getting a count and we can skip the
5420 ret = btrfs_check_shared(root, btrfs_ino(inode),
5421 bytenr, roots, tmp_ulist);
5425 flags |= FIEMAP_EXTENT_SHARED;
5428 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
5429 flags |= FIEMAP_EXTENT_ENCODED;
5430 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5431 flags |= FIEMAP_EXTENT_UNWRITTEN;
5433 free_extent_map(em);
5435 if ((em_start >= last) || em_len == (u64)-1 ||
5436 (last == (u64)-1 && isize <= em_end)) {
5437 flags |= FIEMAP_EXTENT_LAST;
5441 /* now scan forward to see if this is really the last extent. */
5442 em = get_extent_skip_holes(inode, off, last_for_get_extent);
5448 flags |= FIEMAP_EXTENT_LAST;
5451 ret = emit_fiemap_extent(fieinfo, &cache, em_start, disko,
5461 ret = emit_last_fiemap_cache(fieinfo, &cache);
5462 free_extent_map(em);
5464 unlock_extent_cached(&inode->io_tree, start, start + len - 1,
5468 btrfs_free_path(path);
5470 ulist_free(tmp_ulist);
5474 static void __free_extent_buffer(struct extent_buffer *eb)
5476 kmem_cache_free(extent_buffer_cache, eb);
5479 int extent_buffer_under_io(const struct extent_buffer *eb)
5481 return (atomic_read(&eb->io_pages) ||
5482 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
5483 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
5486 static bool page_range_has_eb(struct btrfs_fs_info *fs_info, struct page *page)
5488 struct btrfs_subpage *subpage;
5490 lockdep_assert_held(&page->mapping->private_lock);
5492 if (PagePrivate(page)) {
5493 subpage = (struct btrfs_subpage *)page->private;
5494 if (atomic_read(&subpage->eb_refs))
5500 static void detach_extent_buffer_page(struct extent_buffer *eb, struct page *page)
5502 struct btrfs_fs_info *fs_info = eb->fs_info;
5503 const bool mapped = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
5506 * For mapped eb, we're going to change the page private, which should
5507 * be done under the private_lock.
5510 spin_lock(&page->mapping->private_lock);
5512 if (!PagePrivate(page)) {
5514 spin_unlock(&page->mapping->private_lock);
5518 if (fs_info->sectorsize == PAGE_SIZE) {
5520 * We do this since we'll remove the pages after we've
5521 * removed the eb from the radix tree, so we could race
5522 * and have this page now attached to the new eb. So
5523 * only clear page_private if it's still connected to
5526 if (PagePrivate(page) &&
5527 page->private == (unsigned long)eb) {
5528 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
5529 BUG_ON(PageDirty(page));
5530 BUG_ON(PageWriteback(page));
5532 * We need to make sure we haven't be attached
5535 detach_page_private(page);
5538 spin_unlock(&page->mapping->private_lock);
5543 * For subpage, we can have dummy eb with page private. In this case,
5544 * we can directly detach the private as such page is only attached to
5545 * one dummy eb, no sharing.
5548 btrfs_detach_subpage(fs_info, page);
5552 btrfs_page_dec_eb_refs(fs_info, page);
5555 * We can only detach the page private if there are no other ebs in the
5558 if (!page_range_has_eb(fs_info, page))
5559 btrfs_detach_subpage(fs_info, page);
5561 spin_unlock(&page->mapping->private_lock);
5564 /* Release all pages attached to the extent buffer */
5565 static void btrfs_release_extent_buffer_pages(struct extent_buffer *eb)
5570 ASSERT(!extent_buffer_under_io(eb));
5572 num_pages = num_extent_pages(eb);
5573 for (i = 0; i < num_pages; i++) {
5574 struct page *page = eb->pages[i];
5579 detach_extent_buffer_page(eb, page);
5581 /* One for when we allocated the page */
5587 * Helper for releasing the extent buffer.
5589 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
5591 btrfs_release_extent_buffer_pages(eb);
5592 btrfs_leak_debug_del(&eb->fs_info->eb_leak_lock, &eb->leak_list);
5593 __free_extent_buffer(eb);
5596 static struct extent_buffer *
5597 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
5600 struct extent_buffer *eb = NULL;
5602 eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL);
5605 eb->fs_info = fs_info;
5607 init_rwsem(&eb->lock);
5609 btrfs_leak_debug_add(&fs_info->eb_leak_lock, &eb->leak_list,
5610 &fs_info->allocated_ebs);
5611 INIT_LIST_HEAD(&eb->release_list);
5613 spin_lock_init(&eb->refs_lock);
5614 atomic_set(&eb->refs, 1);
5615 atomic_set(&eb->io_pages, 0);
5617 ASSERT(len <= BTRFS_MAX_METADATA_BLOCKSIZE);
5622 struct extent_buffer *btrfs_clone_extent_buffer(const struct extent_buffer *src)
5626 struct extent_buffer *new;
5627 int num_pages = num_extent_pages(src);
5629 new = __alloc_extent_buffer(src->fs_info, src->start, src->len);
5634 * Set UNMAPPED before calling btrfs_release_extent_buffer(), as
5635 * btrfs_release_extent_buffer() have different behavior for
5636 * UNMAPPED subpage extent buffer.
5638 set_bit(EXTENT_BUFFER_UNMAPPED, &new->bflags);
5640 for (i = 0; i < num_pages; i++) {
5643 p = alloc_page(GFP_NOFS);
5645 btrfs_release_extent_buffer(new);
5648 ret = attach_extent_buffer_page(new, p, NULL);
5651 btrfs_release_extent_buffer(new);
5654 WARN_ON(PageDirty(p));
5656 copy_page(page_address(p), page_address(src->pages[i]));
5658 set_extent_buffer_uptodate(new);
5663 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
5664 u64 start, unsigned long len)
5666 struct extent_buffer *eb;
5670 eb = __alloc_extent_buffer(fs_info, start, len);
5674 num_pages = num_extent_pages(eb);
5675 for (i = 0; i < num_pages; i++) {
5678 eb->pages[i] = alloc_page(GFP_NOFS);
5681 ret = attach_extent_buffer_page(eb, eb->pages[i], NULL);
5685 set_extent_buffer_uptodate(eb);
5686 btrfs_set_header_nritems(eb, 0);
5687 set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
5691 for (; i > 0; i--) {
5692 detach_extent_buffer_page(eb, eb->pages[i - 1]);
5693 __free_page(eb->pages[i - 1]);
5695 __free_extent_buffer(eb);
5699 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
5702 return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize);
5705 static void check_buffer_tree_ref(struct extent_buffer *eb)
5709 * The TREE_REF bit is first set when the extent_buffer is added
5710 * to the radix tree. It is also reset, if unset, when a new reference
5711 * is created by find_extent_buffer.
5713 * It is only cleared in two cases: freeing the last non-tree
5714 * reference to the extent_buffer when its STALE bit is set or
5715 * calling releasepage when the tree reference is the only reference.
5717 * In both cases, care is taken to ensure that the extent_buffer's
5718 * pages are not under io. However, releasepage can be concurrently
5719 * called with creating new references, which is prone to race
5720 * conditions between the calls to check_buffer_tree_ref in those
5721 * codepaths and clearing TREE_REF in try_release_extent_buffer.
5723 * The actual lifetime of the extent_buffer in the radix tree is
5724 * adequately protected by the refcount, but the TREE_REF bit and
5725 * its corresponding reference are not. To protect against this
5726 * class of races, we call check_buffer_tree_ref from the codepaths
5727 * which trigger io after they set eb->io_pages. Note that once io is
5728 * initiated, TREE_REF can no longer be cleared, so that is the
5729 * moment at which any such race is best fixed.
5731 refs = atomic_read(&eb->refs);
5732 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5735 spin_lock(&eb->refs_lock);
5736 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5737 atomic_inc(&eb->refs);
5738 spin_unlock(&eb->refs_lock);
5741 static void mark_extent_buffer_accessed(struct extent_buffer *eb,
5742 struct page *accessed)
5746 check_buffer_tree_ref(eb);
5748 num_pages = num_extent_pages(eb);
5749 for (i = 0; i < num_pages; i++) {
5750 struct page *p = eb->pages[i];
5753 mark_page_accessed(p);
5757 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
5760 struct extent_buffer *eb;
5762 eb = find_extent_buffer_nolock(fs_info, start);
5766 * Lock our eb's refs_lock to avoid races with free_extent_buffer().
5767 * When we get our eb it might be flagged with EXTENT_BUFFER_STALE and
5768 * another task running free_extent_buffer() might have seen that flag
5769 * set, eb->refs == 2, that the buffer isn't under IO (dirty and
5770 * writeback flags not set) and it's still in the tree (flag
5771 * EXTENT_BUFFER_TREE_REF set), therefore being in the process of
5772 * decrementing the extent buffer's reference count twice. So here we
5773 * could race and increment the eb's reference count, clear its stale
5774 * flag, mark it as dirty and drop our reference before the other task
5775 * finishes executing free_extent_buffer, which would later result in
5776 * an attempt to free an extent buffer that is dirty.
5778 if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) {
5779 spin_lock(&eb->refs_lock);
5780 spin_unlock(&eb->refs_lock);
5782 mark_extent_buffer_accessed(eb, NULL);
5786 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5787 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
5790 struct extent_buffer *eb, *exists = NULL;
5793 eb = find_extent_buffer(fs_info, start);
5796 eb = alloc_dummy_extent_buffer(fs_info, start);
5798 return ERR_PTR(-ENOMEM);
5799 eb->fs_info = fs_info;
5801 ret = radix_tree_preload(GFP_NOFS);
5803 exists = ERR_PTR(ret);
5806 spin_lock(&fs_info->buffer_lock);
5807 ret = radix_tree_insert(&fs_info->buffer_radix,
5808 start >> fs_info->sectorsize_bits, eb);
5809 spin_unlock(&fs_info->buffer_lock);
5810 radix_tree_preload_end();
5811 if (ret == -EEXIST) {
5812 exists = find_extent_buffer(fs_info, start);
5818 check_buffer_tree_ref(eb);
5819 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
5823 btrfs_release_extent_buffer(eb);
5828 static struct extent_buffer *grab_extent_buffer(
5829 struct btrfs_fs_info *fs_info, struct page *page)
5831 struct extent_buffer *exists;
5834 * For subpage case, we completely rely on radix tree to ensure we
5835 * don't try to insert two ebs for the same bytenr. So here we always
5836 * return NULL and just continue.
5838 if (fs_info->sectorsize < PAGE_SIZE)
5841 /* Page not yet attached to an extent buffer */
5842 if (!PagePrivate(page))
5846 * We could have already allocated an eb for this page and attached one
5847 * so lets see if we can get a ref on the existing eb, and if we can we
5848 * know it's good and we can just return that one, else we know we can
5849 * just overwrite page->private.
5851 exists = (struct extent_buffer *)page->private;
5852 if (atomic_inc_not_zero(&exists->refs))
5855 WARN_ON(PageDirty(page));
5856 detach_page_private(page);
5860 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
5861 u64 start, u64 owner_root, int level)
5863 unsigned long len = fs_info->nodesize;
5866 unsigned long index = start >> PAGE_SHIFT;
5867 struct extent_buffer *eb;
5868 struct extent_buffer *exists = NULL;
5870 struct address_space *mapping = fs_info->btree_inode->i_mapping;
5874 if (!IS_ALIGNED(start, fs_info->sectorsize)) {
5875 btrfs_err(fs_info, "bad tree block start %llu", start);
5876 return ERR_PTR(-EINVAL);
5879 #if BITS_PER_LONG == 32
5880 if (start >= MAX_LFS_FILESIZE) {
5881 btrfs_err_rl(fs_info,
5882 "extent buffer %llu is beyond 32bit page cache limit", start);
5883 btrfs_err_32bit_limit(fs_info);
5884 return ERR_PTR(-EOVERFLOW);
5886 if (start >= BTRFS_32BIT_EARLY_WARN_THRESHOLD)
5887 btrfs_warn_32bit_limit(fs_info);
5890 if (fs_info->sectorsize < PAGE_SIZE &&
5891 offset_in_page(start) + len > PAGE_SIZE) {
5893 "tree block crosses page boundary, start %llu nodesize %lu",
5895 return ERR_PTR(-EINVAL);
5898 eb = find_extent_buffer(fs_info, start);
5902 eb = __alloc_extent_buffer(fs_info, start, len);
5904 return ERR_PTR(-ENOMEM);
5905 btrfs_set_buffer_lockdep_class(owner_root, eb, level);
5907 num_pages = num_extent_pages(eb);
5908 for (i = 0; i < num_pages; i++, index++) {
5909 struct btrfs_subpage *prealloc = NULL;
5911 p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL);
5913 exists = ERR_PTR(-ENOMEM);
5918 * Preallocate page->private for subpage case, so that we won't
5919 * allocate memory with private_lock hold. The memory will be
5920 * freed by attach_extent_buffer_page() or freed manually if
5923 * Although we have ensured one subpage eb can only have one
5924 * page, but it may change in the future for 16K page size
5925 * support, so we still preallocate the memory in the loop.
5927 ret = btrfs_alloc_subpage(fs_info, &prealloc,
5928 BTRFS_SUBPAGE_METADATA);
5932 exists = ERR_PTR(ret);
5936 spin_lock(&mapping->private_lock);
5937 exists = grab_extent_buffer(fs_info, p);
5939 spin_unlock(&mapping->private_lock);
5942 mark_extent_buffer_accessed(exists, p);
5943 btrfs_free_subpage(prealloc);
5946 /* Should not fail, as we have preallocated the memory */
5947 ret = attach_extent_buffer_page(eb, p, prealloc);
5950 * To inform we have extra eb under allocation, so that
5951 * detach_extent_buffer_page() won't release the page private
5952 * when the eb hasn't yet been inserted into radix tree.
5954 * The ref will be decreased when the eb released the page, in
5955 * detach_extent_buffer_page().
5956 * Thus needs no special handling in error path.
5958 btrfs_page_inc_eb_refs(fs_info, p);
5959 spin_unlock(&mapping->private_lock);
5961 WARN_ON(btrfs_page_test_dirty(fs_info, p, eb->start, eb->len));
5963 if (!PageUptodate(p))
5967 * We can't unlock the pages just yet since the extent buffer
5968 * hasn't been properly inserted in the radix tree, this
5969 * opens a race with btree_releasepage which can free a page
5970 * while we are still filling in all pages for the buffer and
5975 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5977 ret = radix_tree_preload(GFP_NOFS);
5979 exists = ERR_PTR(ret);
5983 spin_lock(&fs_info->buffer_lock);
5984 ret = radix_tree_insert(&fs_info->buffer_radix,
5985 start >> fs_info->sectorsize_bits, eb);
5986 spin_unlock(&fs_info->buffer_lock);
5987 radix_tree_preload_end();
5988 if (ret == -EEXIST) {
5989 exists = find_extent_buffer(fs_info, start);
5995 /* add one reference for the tree */
5996 check_buffer_tree_ref(eb);
5997 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
6000 * Now it's safe to unlock the pages because any calls to
6001 * btree_releasepage will correctly detect that a page belongs to a
6002 * live buffer and won't free them prematurely.
6004 for (i = 0; i < num_pages; i++)
6005 unlock_page(eb->pages[i]);
6009 WARN_ON(!atomic_dec_and_test(&eb->refs));
6010 for (i = 0; i < num_pages; i++) {
6012 unlock_page(eb->pages[i]);
6015 btrfs_release_extent_buffer(eb);
6019 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
6021 struct extent_buffer *eb =
6022 container_of(head, struct extent_buffer, rcu_head);
6024 __free_extent_buffer(eb);
6027 static int release_extent_buffer(struct extent_buffer *eb)
6028 __releases(&eb->refs_lock)
6030 lockdep_assert_held(&eb->refs_lock);
6032 WARN_ON(atomic_read(&eb->refs) == 0);
6033 if (atomic_dec_and_test(&eb->refs)) {
6034 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
6035 struct btrfs_fs_info *fs_info = eb->fs_info;
6037 spin_unlock(&eb->refs_lock);
6039 spin_lock(&fs_info->buffer_lock);
6040 radix_tree_delete(&fs_info->buffer_radix,
6041 eb->start >> fs_info->sectorsize_bits);
6042 spin_unlock(&fs_info->buffer_lock);
6044 spin_unlock(&eb->refs_lock);
6047 btrfs_leak_debug_del(&eb->fs_info->eb_leak_lock, &eb->leak_list);
6048 /* Should be safe to release our pages at this point */
6049 btrfs_release_extent_buffer_pages(eb);
6050 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
6051 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags))) {
6052 __free_extent_buffer(eb);
6056 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
6059 spin_unlock(&eb->refs_lock);
6064 void free_extent_buffer(struct extent_buffer *eb)
6072 refs = atomic_read(&eb->refs);
6073 if ((!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && refs <= 3)
6074 || (test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) &&
6077 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
6082 spin_lock(&eb->refs_lock);
6083 if (atomic_read(&eb->refs) == 2 &&
6084 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
6085 !extent_buffer_under_io(eb) &&
6086 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
6087 atomic_dec(&eb->refs);
6090 * I know this is terrible, but it's temporary until we stop tracking
6091 * the uptodate bits and such for the extent buffers.
6093 release_extent_buffer(eb);
6096 void free_extent_buffer_stale(struct extent_buffer *eb)
6101 spin_lock(&eb->refs_lock);
6102 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
6104 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
6105 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
6106 atomic_dec(&eb->refs);
6107 release_extent_buffer(eb);
6110 static void btree_clear_page_dirty(struct page *page)
6112 ASSERT(PageDirty(page));
6113 ASSERT(PageLocked(page));
6114 clear_page_dirty_for_io(page);
6115 xa_lock_irq(&page->mapping->i_pages);
6116 if (!PageDirty(page))
6117 __xa_clear_mark(&page->mapping->i_pages,
6118 page_index(page), PAGECACHE_TAG_DIRTY);
6119 xa_unlock_irq(&page->mapping->i_pages);
6122 static void clear_subpage_extent_buffer_dirty(const struct extent_buffer *eb)
6124 struct btrfs_fs_info *fs_info = eb->fs_info;
6125 struct page *page = eb->pages[0];
6128 /* btree_clear_page_dirty() needs page locked */
6130 last = btrfs_subpage_clear_and_test_dirty(fs_info, page, eb->start,
6133 btree_clear_page_dirty(page);
6135 WARN_ON(atomic_read(&eb->refs) == 0);
6138 void clear_extent_buffer_dirty(const struct extent_buffer *eb)
6144 if (eb->fs_info->sectorsize < PAGE_SIZE)
6145 return clear_subpage_extent_buffer_dirty(eb);
6147 num_pages = num_extent_pages(eb);
6149 for (i = 0; i < num_pages; i++) {
6150 page = eb->pages[i];
6151 if (!PageDirty(page))
6154 btree_clear_page_dirty(page);
6155 ClearPageError(page);
6158 WARN_ON(atomic_read(&eb->refs) == 0);
6161 bool set_extent_buffer_dirty(struct extent_buffer *eb)
6167 check_buffer_tree_ref(eb);
6169 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
6171 num_pages = num_extent_pages(eb);
6172 WARN_ON(atomic_read(&eb->refs) == 0);
6173 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
6176 bool subpage = eb->fs_info->sectorsize < PAGE_SIZE;
6179 * For subpage case, we can have other extent buffers in the
6180 * same page, and in clear_subpage_extent_buffer_dirty() we
6181 * have to clear page dirty without subpage lock held.
6182 * This can cause race where our page gets dirty cleared after
6185 * Thankfully, clear_subpage_extent_buffer_dirty() has locked
6186 * its page for other reasons, we can use page lock to prevent
6190 lock_page(eb->pages[0]);
6191 for (i = 0; i < num_pages; i++)
6192 btrfs_page_set_dirty(eb->fs_info, eb->pages[i],
6193 eb->start, eb->len);
6195 unlock_page(eb->pages[0]);
6197 #ifdef CONFIG_BTRFS_DEBUG
6198 for (i = 0; i < num_pages; i++)
6199 ASSERT(PageDirty(eb->pages[i]));
6205 void clear_extent_buffer_uptodate(struct extent_buffer *eb)
6207 struct btrfs_fs_info *fs_info = eb->fs_info;
6212 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
6213 num_pages = num_extent_pages(eb);
6214 for (i = 0; i < num_pages; i++) {
6215 page = eb->pages[i];
6217 btrfs_page_clear_uptodate(fs_info, page,
6218 eb->start, eb->len);
6222 void set_extent_buffer_uptodate(struct extent_buffer *eb)
6224 struct btrfs_fs_info *fs_info = eb->fs_info;
6229 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
6230 num_pages = num_extent_pages(eb);
6231 for (i = 0; i < num_pages; i++) {
6232 page = eb->pages[i];
6233 btrfs_page_set_uptodate(fs_info, page, eb->start, eb->len);
6237 static int read_extent_buffer_subpage(struct extent_buffer *eb, int wait,
6240 struct btrfs_fs_info *fs_info = eb->fs_info;
6241 struct extent_io_tree *io_tree;
6242 struct page *page = eb->pages[0];
6243 struct btrfs_bio_ctrl bio_ctrl = { 0 };
6246 ASSERT(!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags));
6247 ASSERT(PagePrivate(page));
6248 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
6250 if (wait == WAIT_NONE) {
6251 if (!try_lock_extent(io_tree, eb->start, eb->start + eb->len - 1))
6254 ret = lock_extent(io_tree, eb->start, eb->start + eb->len - 1);
6260 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags) ||
6261 PageUptodate(page) ||
6262 btrfs_subpage_test_uptodate(fs_info, page, eb->start, eb->len)) {
6263 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
6264 unlock_extent(io_tree, eb->start, eb->start + eb->len - 1);
6268 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
6269 eb->read_mirror = 0;
6270 atomic_set(&eb->io_pages, 1);
6271 check_buffer_tree_ref(eb);
6272 btrfs_subpage_clear_error(fs_info, page, eb->start, eb->len);
6274 ret = submit_extent_page(REQ_OP_READ | REQ_META, NULL, &bio_ctrl,
6275 page, eb->start, eb->len,
6276 eb->start - page_offset(page),
6277 end_bio_extent_readpage, mirror_num, 0,
6281 * In the endio function, if we hit something wrong we will
6282 * increase the io_pages, so here we need to decrease it for
6285 atomic_dec(&eb->io_pages);
6290 tmp = submit_one_bio(bio_ctrl.bio, mirror_num, 0);
6291 bio_ctrl.bio = NULL;
6295 if (ret || wait != WAIT_COMPLETE)
6298 wait_extent_bit(io_tree, eb->start, eb->start + eb->len - 1, EXTENT_LOCKED);
6299 if (!test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
6304 int read_extent_buffer_pages(struct extent_buffer *eb, int wait, int mirror_num)
6310 int locked_pages = 0;
6311 int all_uptodate = 1;
6313 unsigned long num_reads = 0;
6314 struct btrfs_bio_ctrl bio_ctrl = { 0 };
6316 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
6319 if (eb->fs_info->sectorsize < PAGE_SIZE)
6320 return read_extent_buffer_subpage(eb, wait, mirror_num);
6322 num_pages = num_extent_pages(eb);
6323 for (i = 0; i < num_pages; i++) {
6324 page = eb->pages[i];
6325 if (wait == WAIT_NONE) {
6327 * WAIT_NONE is only utilized by readahead. If we can't
6328 * acquire the lock atomically it means either the eb
6329 * is being read out or under modification.
6330 * Either way the eb will be or has been cached,
6331 * readahead can exit safely.
6333 if (!trylock_page(page))
6341 * We need to firstly lock all pages to make sure that
6342 * the uptodate bit of our pages won't be affected by
6343 * clear_extent_buffer_uptodate().
6345 for (i = 0; i < num_pages; i++) {
6346 page = eb->pages[i];
6347 if (!PageUptodate(page)) {
6354 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
6358 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
6359 eb->read_mirror = 0;
6360 atomic_set(&eb->io_pages, num_reads);
6362 * It is possible for releasepage to clear the TREE_REF bit before we
6363 * set io_pages. See check_buffer_tree_ref for a more detailed comment.
6365 check_buffer_tree_ref(eb);
6366 for (i = 0; i < num_pages; i++) {
6367 page = eb->pages[i];
6369 if (!PageUptodate(page)) {
6371 atomic_dec(&eb->io_pages);
6376 ClearPageError(page);
6377 err = submit_extent_page(REQ_OP_READ | REQ_META, NULL,
6378 &bio_ctrl, page, page_offset(page),
6379 PAGE_SIZE, 0, end_bio_extent_readpage,
6380 mirror_num, 0, false);
6383 * We failed to submit the bio so it's the
6384 * caller's responsibility to perform cleanup
6385 * i.e unlock page/set error bit.
6390 atomic_dec(&eb->io_pages);
6398 err = submit_one_bio(bio_ctrl.bio, mirror_num, bio_ctrl.bio_flags);
6399 bio_ctrl.bio = NULL;
6404 if (ret || wait != WAIT_COMPLETE)
6407 for (i = 0; i < num_pages; i++) {
6408 page = eb->pages[i];
6409 wait_on_page_locked(page);
6410 if (!PageUptodate(page))
6417 while (locked_pages > 0) {
6419 page = eb->pages[locked_pages];
6425 static bool report_eb_range(const struct extent_buffer *eb, unsigned long start,
6428 btrfs_warn(eb->fs_info,
6429 "access to eb bytenr %llu len %lu out of range start %lu len %lu",
6430 eb->start, eb->len, start, len);
6431 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
6437 * Check if the [start, start + len) range is valid before reading/writing
6439 * NOTE: @start and @len are offset inside the eb, not logical address.
6441 * Caller should not touch the dst/src memory if this function returns error.
6443 static inline int check_eb_range(const struct extent_buffer *eb,
6444 unsigned long start, unsigned long len)
6446 unsigned long offset;
6448 /* start, start + len should not go beyond eb->len nor overflow */
6449 if (unlikely(check_add_overflow(start, len, &offset) || offset > eb->len))
6450 return report_eb_range(eb, start, len);
6455 void read_extent_buffer(const struct extent_buffer *eb, void *dstv,
6456 unsigned long start, unsigned long len)
6462 char *dst = (char *)dstv;
6463 unsigned long i = get_eb_page_index(start);
6465 if (check_eb_range(eb, start, len))
6468 offset = get_eb_offset_in_page(eb, start);
6471 page = eb->pages[i];
6473 cur = min(len, (PAGE_SIZE - offset));
6474 kaddr = page_address(page);
6475 memcpy(dst, kaddr + offset, cur);
6484 int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb,
6486 unsigned long start, unsigned long len)
6492 char __user *dst = (char __user *)dstv;
6493 unsigned long i = get_eb_page_index(start);
6496 WARN_ON(start > eb->len);
6497 WARN_ON(start + len > eb->start + eb->len);
6499 offset = get_eb_offset_in_page(eb, start);
6502 page = eb->pages[i];
6504 cur = min(len, (PAGE_SIZE - offset));
6505 kaddr = page_address(page);
6506 if (copy_to_user_nofault(dst, kaddr + offset, cur)) {
6520 int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,
6521 unsigned long start, unsigned long len)
6527 char *ptr = (char *)ptrv;
6528 unsigned long i = get_eb_page_index(start);
6531 if (check_eb_range(eb, start, len))
6534 offset = get_eb_offset_in_page(eb, start);
6537 page = eb->pages[i];
6539 cur = min(len, (PAGE_SIZE - offset));
6541 kaddr = page_address(page);
6542 ret = memcmp(ptr, kaddr + offset, cur);
6555 * Check that the extent buffer is uptodate.
6557 * For regular sector size == PAGE_SIZE case, check if @page is uptodate.
6558 * For subpage case, check if the range covered by the eb has EXTENT_UPTODATE.
6560 static void assert_eb_page_uptodate(const struct extent_buffer *eb,
6563 struct btrfs_fs_info *fs_info = eb->fs_info;
6565 if (fs_info->sectorsize < PAGE_SIZE) {
6568 uptodate = btrfs_subpage_test_uptodate(fs_info, page,
6569 eb->start, eb->len);
6572 WARN_ON(!PageUptodate(page));
6576 void write_extent_buffer_chunk_tree_uuid(const struct extent_buffer *eb,
6581 assert_eb_page_uptodate(eb, eb->pages[0]);
6582 kaddr = page_address(eb->pages[0]) +
6583 get_eb_offset_in_page(eb, offsetof(struct btrfs_header,
6585 memcpy(kaddr, srcv, BTRFS_FSID_SIZE);
6588 void write_extent_buffer_fsid(const struct extent_buffer *eb, const void *srcv)
6592 assert_eb_page_uptodate(eb, eb->pages[0]);
6593 kaddr = page_address(eb->pages[0]) +
6594 get_eb_offset_in_page(eb, offsetof(struct btrfs_header, fsid));
6595 memcpy(kaddr, srcv, BTRFS_FSID_SIZE);
6598 void write_extent_buffer(const struct extent_buffer *eb, const void *srcv,
6599 unsigned long start, unsigned long len)
6605 char *src = (char *)srcv;
6606 unsigned long i = get_eb_page_index(start);
6608 WARN_ON(test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags));
6610 if (check_eb_range(eb, start, len))
6613 offset = get_eb_offset_in_page(eb, start);
6616 page = eb->pages[i];
6617 assert_eb_page_uptodate(eb, page);
6619 cur = min(len, PAGE_SIZE - offset);
6620 kaddr = page_address(page);
6621 memcpy(kaddr + offset, src, cur);
6630 void memzero_extent_buffer(const struct extent_buffer *eb, unsigned long start,
6637 unsigned long i = get_eb_page_index(start);
6639 if (check_eb_range(eb, start, len))
6642 offset = get_eb_offset_in_page(eb, start);
6645 page = eb->pages[i];
6646 assert_eb_page_uptodate(eb, page);
6648 cur = min(len, PAGE_SIZE - offset);
6649 kaddr = page_address(page);
6650 memset(kaddr + offset, 0, cur);
6658 void copy_extent_buffer_full(const struct extent_buffer *dst,
6659 const struct extent_buffer *src)
6664 ASSERT(dst->len == src->len);
6666 if (dst->fs_info->sectorsize == PAGE_SIZE) {
6667 num_pages = num_extent_pages(dst);
6668 for (i = 0; i < num_pages; i++)
6669 copy_page(page_address(dst->pages[i]),
6670 page_address(src->pages[i]));
6672 size_t src_offset = get_eb_offset_in_page(src, 0);
6673 size_t dst_offset = get_eb_offset_in_page(dst, 0);
6675 ASSERT(src->fs_info->sectorsize < PAGE_SIZE);
6676 memcpy(page_address(dst->pages[0]) + dst_offset,
6677 page_address(src->pages[0]) + src_offset,
6682 void copy_extent_buffer(const struct extent_buffer *dst,
6683 const struct extent_buffer *src,
6684 unsigned long dst_offset, unsigned long src_offset,
6687 u64 dst_len = dst->len;
6692 unsigned long i = get_eb_page_index(dst_offset);
6694 if (check_eb_range(dst, dst_offset, len) ||
6695 check_eb_range(src, src_offset, len))
6698 WARN_ON(src->len != dst_len);
6700 offset = get_eb_offset_in_page(dst, dst_offset);
6703 page = dst->pages[i];
6704 assert_eb_page_uptodate(dst, page);
6706 cur = min(len, (unsigned long)(PAGE_SIZE - offset));
6708 kaddr = page_address(page);
6709 read_extent_buffer(src, kaddr + offset, src_offset, cur);
6719 * eb_bitmap_offset() - calculate the page and offset of the byte containing the
6721 * @eb: the extent buffer
6722 * @start: offset of the bitmap item in the extent buffer
6724 * @page_index: return index of the page in the extent buffer that contains the
6726 * @page_offset: return offset into the page given by page_index
6728 * This helper hides the ugliness of finding the byte in an extent buffer which
6729 * contains a given bit.
6731 static inline void eb_bitmap_offset(const struct extent_buffer *eb,
6732 unsigned long start, unsigned long nr,
6733 unsigned long *page_index,
6734 size_t *page_offset)
6736 size_t byte_offset = BIT_BYTE(nr);
6740 * The byte we want is the offset of the extent buffer + the offset of
6741 * the bitmap item in the extent buffer + the offset of the byte in the
6744 offset = start + offset_in_page(eb->start) + byte_offset;
6746 *page_index = offset >> PAGE_SHIFT;
6747 *page_offset = offset_in_page(offset);
6751 * extent_buffer_test_bit - determine whether a bit in a bitmap item is set
6752 * @eb: the extent buffer
6753 * @start: offset of the bitmap item in the extent buffer
6754 * @nr: bit number to test
6756 int extent_buffer_test_bit(const struct extent_buffer *eb, unsigned long start,
6764 eb_bitmap_offset(eb, start, nr, &i, &offset);
6765 page = eb->pages[i];
6766 assert_eb_page_uptodate(eb, page);
6767 kaddr = page_address(page);
6768 return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1)));
6772 * extent_buffer_bitmap_set - set an area of a bitmap
6773 * @eb: the extent buffer
6774 * @start: offset of the bitmap item in the extent buffer
6775 * @pos: bit number of the first bit
6776 * @len: number of bits to set
6778 void extent_buffer_bitmap_set(const struct extent_buffer *eb, unsigned long start,
6779 unsigned long pos, unsigned long len)
6785 const unsigned int size = pos + len;
6786 int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
6787 u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos);
6789 eb_bitmap_offset(eb, start, pos, &i, &offset);
6790 page = eb->pages[i];
6791 assert_eb_page_uptodate(eb, page);
6792 kaddr = page_address(page);
6794 while (len >= bits_to_set) {
6795 kaddr[offset] |= mask_to_set;
6797 bits_to_set = BITS_PER_BYTE;
6799 if (++offset >= PAGE_SIZE && len > 0) {
6801 page = eb->pages[++i];
6802 assert_eb_page_uptodate(eb, page);
6803 kaddr = page_address(page);
6807 mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
6808 kaddr[offset] |= mask_to_set;
6814 * extent_buffer_bitmap_clear - clear an area of a bitmap
6815 * @eb: the extent buffer
6816 * @start: offset of the bitmap item in the extent buffer
6817 * @pos: bit number of the first bit
6818 * @len: number of bits to clear
6820 void extent_buffer_bitmap_clear(const struct extent_buffer *eb,
6821 unsigned long start, unsigned long pos,
6828 const unsigned int size = pos + len;
6829 int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
6830 u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos);
6832 eb_bitmap_offset(eb, start, pos, &i, &offset);
6833 page = eb->pages[i];
6834 assert_eb_page_uptodate(eb, page);
6835 kaddr = page_address(page);
6837 while (len >= bits_to_clear) {
6838 kaddr[offset] &= ~mask_to_clear;
6839 len -= bits_to_clear;
6840 bits_to_clear = BITS_PER_BYTE;
6842 if (++offset >= PAGE_SIZE && len > 0) {
6844 page = eb->pages[++i];
6845 assert_eb_page_uptodate(eb, page);
6846 kaddr = page_address(page);
6850 mask_to_clear &= BITMAP_LAST_BYTE_MASK(size);
6851 kaddr[offset] &= ~mask_to_clear;
6855 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
6857 unsigned long distance = (src > dst) ? src - dst : dst - src;
6858 return distance < len;
6861 static void copy_pages(struct page *dst_page, struct page *src_page,
6862 unsigned long dst_off, unsigned long src_off,
6865 char *dst_kaddr = page_address(dst_page);
6867 int must_memmove = 0;
6869 if (dst_page != src_page) {
6870 src_kaddr = page_address(src_page);
6872 src_kaddr = dst_kaddr;
6873 if (areas_overlap(src_off, dst_off, len))
6878 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
6880 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
6883 void memcpy_extent_buffer(const struct extent_buffer *dst,
6884 unsigned long dst_offset, unsigned long src_offset,
6888 size_t dst_off_in_page;
6889 size_t src_off_in_page;
6890 unsigned long dst_i;
6891 unsigned long src_i;
6893 if (check_eb_range(dst, dst_offset, len) ||
6894 check_eb_range(dst, src_offset, len))
6898 dst_off_in_page = get_eb_offset_in_page(dst, dst_offset);
6899 src_off_in_page = get_eb_offset_in_page(dst, src_offset);
6901 dst_i = get_eb_page_index(dst_offset);
6902 src_i = get_eb_page_index(src_offset);
6904 cur = min(len, (unsigned long)(PAGE_SIZE -
6906 cur = min_t(unsigned long, cur,
6907 (unsigned long)(PAGE_SIZE - dst_off_in_page));
6909 copy_pages(dst->pages[dst_i], dst->pages[src_i],
6910 dst_off_in_page, src_off_in_page, cur);
6918 void memmove_extent_buffer(const struct extent_buffer *dst,
6919 unsigned long dst_offset, unsigned long src_offset,
6923 size_t dst_off_in_page;
6924 size_t src_off_in_page;
6925 unsigned long dst_end = dst_offset + len - 1;
6926 unsigned long src_end = src_offset + len - 1;
6927 unsigned long dst_i;
6928 unsigned long src_i;
6930 if (check_eb_range(dst, dst_offset, len) ||
6931 check_eb_range(dst, src_offset, len))
6933 if (dst_offset < src_offset) {
6934 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
6938 dst_i = get_eb_page_index(dst_end);
6939 src_i = get_eb_page_index(src_end);
6941 dst_off_in_page = get_eb_offset_in_page(dst, dst_end);
6942 src_off_in_page = get_eb_offset_in_page(dst, src_end);
6944 cur = min_t(unsigned long, len, src_off_in_page + 1);
6945 cur = min(cur, dst_off_in_page + 1);
6946 copy_pages(dst->pages[dst_i], dst->pages[src_i],
6947 dst_off_in_page - cur + 1,
6948 src_off_in_page - cur + 1, cur);
6956 static struct extent_buffer *get_next_extent_buffer(
6957 struct btrfs_fs_info *fs_info, struct page *page, u64 bytenr)
6959 struct extent_buffer *gang[BTRFS_SUBPAGE_BITMAP_SIZE];
6960 struct extent_buffer *found = NULL;
6961 u64 page_start = page_offset(page);
6965 ASSERT(in_range(bytenr, page_start, PAGE_SIZE));
6966 ASSERT(PAGE_SIZE / fs_info->nodesize <= BTRFS_SUBPAGE_BITMAP_SIZE);
6967 lockdep_assert_held(&fs_info->buffer_lock);
6969 ret = radix_tree_gang_lookup(&fs_info->buffer_radix, (void **)gang,
6970 bytenr >> fs_info->sectorsize_bits,
6971 PAGE_SIZE / fs_info->nodesize);
6972 for (i = 0; i < ret; i++) {
6973 /* Already beyond page end */
6974 if (gang[i]->start >= page_start + PAGE_SIZE)
6977 if (gang[i]->start >= bytenr) {
6985 static int try_release_subpage_extent_buffer(struct page *page)
6987 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
6988 u64 cur = page_offset(page);
6989 const u64 end = page_offset(page) + PAGE_SIZE;
6993 struct extent_buffer *eb = NULL;
6996 * Unlike try_release_extent_buffer() which uses page->private
6997 * to grab buffer, for subpage case we rely on radix tree, thus
6998 * we need to ensure radix tree consistency.
7000 * We also want an atomic snapshot of the radix tree, thus go
7001 * with spinlock rather than RCU.
7003 spin_lock(&fs_info->buffer_lock);
7004 eb = get_next_extent_buffer(fs_info, page, cur);
7006 /* No more eb in the page range after or at cur */
7007 spin_unlock(&fs_info->buffer_lock);
7010 cur = eb->start + eb->len;
7013 * The same as try_release_extent_buffer(), to ensure the eb
7014 * won't disappear out from under us.
7016 spin_lock(&eb->refs_lock);
7017 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
7018 spin_unlock(&eb->refs_lock);
7019 spin_unlock(&fs_info->buffer_lock);
7022 spin_unlock(&fs_info->buffer_lock);
7025 * If tree ref isn't set then we know the ref on this eb is a
7026 * real ref, so just return, this eb will likely be freed soon
7029 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
7030 spin_unlock(&eb->refs_lock);
7035 * Here we don't care about the return value, we will always
7036 * check the page private at the end. And
7037 * release_extent_buffer() will release the refs_lock.
7039 release_extent_buffer(eb);
7042 * Finally to check if we have cleared page private, as if we have
7043 * released all ebs in the page, the page private should be cleared now.
7045 spin_lock(&page->mapping->private_lock);
7046 if (!PagePrivate(page))
7050 spin_unlock(&page->mapping->private_lock);
7055 int try_release_extent_buffer(struct page *page)
7057 struct extent_buffer *eb;
7059 if (btrfs_sb(page->mapping->host->i_sb)->sectorsize < PAGE_SIZE)
7060 return try_release_subpage_extent_buffer(page);
7063 * We need to make sure nobody is changing page->private, as we rely on
7064 * page->private as the pointer to extent buffer.
7066 spin_lock(&page->mapping->private_lock);
7067 if (!PagePrivate(page)) {
7068 spin_unlock(&page->mapping->private_lock);
7072 eb = (struct extent_buffer *)page->private;
7076 * This is a little awful but should be ok, we need to make sure that
7077 * the eb doesn't disappear out from under us while we're looking at
7080 spin_lock(&eb->refs_lock);
7081 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
7082 spin_unlock(&eb->refs_lock);
7083 spin_unlock(&page->mapping->private_lock);
7086 spin_unlock(&page->mapping->private_lock);
7089 * If tree ref isn't set then we know the ref on this eb is a real ref,
7090 * so just return, this page will likely be freed soon anyway.
7092 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
7093 spin_unlock(&eb->refs_lock);
7097 return release_extent_buffer(eb);
7101 * btrfs_readahead_tree_block - attempt to readahead a child block
7102 * @fs_info: the fs_info
7103 * @bytenr: bytenr to read
7104 * @owner_root: objectid of the root that owns this eb
7105 * @gen: generation for the uptodate check, can be 0
7106 * @level: level for the eb
7108 * Attempt to readahead a tree block at @bytenr. If @gen is 0 then we do a
7109 * normal uptodate check of the eb, without checking the generation. If we have
7110 * to read the block we will not block on anything.
7112 void btrfs_readahead_tree_block(struct btrfs_fs_info *fs_info,
7113 u64 bytenr, u64 owner_root, u64 gen, int level)
7115 struct extent_buffer *eb;
7118 eb = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level);
7122 if (btrfs_buffer_uptodate(eb, gen, 1)) {
7123 free_extent_buffer(eb);
7127 ret = read_extent_buffer_pages(eb, WAIT_NONE, 0);
7129 free_extent_buffer_stale(eb);
7131 free_extent_buffer(eb);
7135 * btrfs_readahead_node_child - readahead a node's child block
7136 * @node: parent node we're reading from
7137 * @slot: slot in the parent node for the child we want to read
7139 * A helper for btrfs_readahead_tree_block, we simply read the bytenr pointed at
7140 * the slot in the node provided.
7142 void btrfs_readahead_node_child(struct extent_buffer *node, int slot)
7144 btrfs_readahead_tree_block(node->fs_info,
7145 btrfs_node_blockptr(node, slot),
7146 btrfs_header_owner(node),
7147 btrfs_node_ptr_generation(node, slot),
7148 btrfs_header_level(node) - 1);