1 // SPDX-License-Identifier: GPL-2.0
3 #include <linux/bitops.h>
4 #include <linux/slab.h>
7 #include <linux/pagemap.h>
8 #include <linux/page-flags.h>
9 #include <linux/spinlock.h>
10 #include <linux/blkdev.h>
11 #include <linux/swap.h>
12 #include <linux/writeback.h>
13 #include <linux/pagevec.h>
14 #include <linux/prefetch.h>
15 #include <linux/cleancache.h>
16 #include "extent_io.h"
17 #include "extent-io-tree.h"
18 #include "extent_map.h"
20 #include "btrfs_inode.h"
22 #include "check-integrity.h"
24 #include "rcu-string.h"
28 static struct kmem_cache *extent_state_cache;
29 static struct kmem_cache *extent_buffer_cache;
30 static struct bio_set btrfs_bioset;
32 static inline bool extent_state_in_tree(const struct extent_state *state)
34 return !RB_EMPTY_NODE(&state->rb_node);
37 #ifdef CONFIG_BTRFS_DEBUG
38 static LIST_HEAD(states);
39 static DEFINE_SPINLOCK(leak_lock);
41 static inline void btrfs_leak_debug_add(spinlock_t *lock,
42 struct list_head *new,
43 struct list_head *head)
47 spin_lock_irqsave(lock, flags);
49 spin_unlock_irqrestore(lock, flags);
52 static inline void btrfs_leak_debug_del(spinlock_t *lock,
53 struct list_head *entry)
57 spin_lock_irqsave(lock, flags);
59 spin_unlock_irqrestore(lock, flags);
62 void btrfs_extent_buffer_leak_debug_check(struct btrfs_fs_info *fs_info)
64 struct extent_buffer *eb;
68 * If we didn't get into open_ctree our allocated_ebs will not be
69 * initialized, so just skip this.
71 if (!fs_info->allocated_ebs.next)
74 spin_lock_irqsave(&fs_info->eb_leak_lock, flags);
75 while (!list_empty(&fs_info->allocated_ebs)) {
76 eb = list_first_entry(&fs_info->allocated_ebs,
77 struct extent_buffer, leak_list);
79 "BTRFS: buffer leak start %llu len %lu refs %d bflags %lu owner %llu\n",
80 eb->start, eb->len, atomic_read(&eb->refs), eb->bflags,
81 btrfs_header_owner(eb));
82 list_del(&eb->leak_list);
83 kmem_cache_free(extent_buffer_cache, eb);
85 spin_unlock_irqrestore(&fs_info->eb_leak_lock, flags);
88 static inline void btrfs_extent_state_leak_debug_check(void)
90 struct extent_state *state;
92 while (!list_empty(&states)) {
93 state = list_entry(states.next, struct extent_state, leak_list);
94 pr_err("BTRFS: state leak: start %llu end %llu state %u in tree %d refs %d\n",
95 state->start, state->end, state->state,
96 extent_state_in_tree(state),
97 refcount_read(&state->refs));
98 list_del(&state->leak_list);
99 kmem_cache_free(extent_state_cache, state);
103 #define btrfs_debug_check_extent_io_range(tree, start, end) \
104 __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end))
105 static inline void __btrfs_debug_check_extent_io_range(const char *caller,
106 struct extent_io_tree *tree, u64 start, u64 end)
108 struct inode *inode = tree->private_data;
111 if (!inode || !is_data_inode(inode))
114 isize = i_size_read(inode);
115 if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) {
116 btrfs_debug_rl(BTRFS_I(inode)->root->fs_info,
117 "%s: ino %llu isize %llu odd range [%llu,%llu]",
118 caller, btrfs_ino(BTRFS_I(inode)), isize, start, end);
122 #define btrfs_leak_debug_add(lock, new, head) do {} while (0)
123 #define btrfs_leak_debug_del(lock, entry) do {} while (0)
124 #define btrfs_extent_state_leak_debug_check() do {} while (0)
125 #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0)
131 struct rb_node rb_node;
134 struct extent_page_data {
136 /* tells writepage not to lock the state bits for this range
137 * it still does the unlocking
139 unsigned int extent_locked:1;
141 /* tells the submit_bio code to use REQ_SYNC */
142 unsigned int sync_io:1;
145 static int add_extent_changeset(struct extent_state *state, unsigned bits,
146 struct extent_changeset *changeset,
153 if (set && (state->state & bits) == bits)
155 if (!set && (state->state & bits) == 0)
157 changeset->bytes_changed += state->end - state->start + 1;
158 ret = ulist_add(&changeset->range_changed, state->start, state->end,
163 int __must_check submit_one_bio(struct bio *bio, int mirror_num,
164 unsigned long bio_flags)
166 blk_status_t ret = 0;
167 struct extent_io_tree *tree = bio->bi_private;
169 bio->bi_private = NULL;
171 if (is_data_inode(tree->private_data))
172 ret = btrfs_submit_data_bio(tree->private_data, bio, mirror_num,
175 ret = btrfs_submit_metadata_bio(tree->private_data, bio,
176 mirror_num, bio_flags);
178 return blk_status_to_errno(ret);
181 /* Cleanup unsubmitted bios */
182 static void end_write_bio(struct extent_page_data *epd, int ret)
185 epd->bio->bi_status = errno_to_blk_status(ret);
192 * Submit bio from extent page data via submit_one_bio
194 * Return 0 if everything is OK.
195 * Return <0 for error.
197 static int __must_check flush_write_bio(struct extent_page_data *epd)
202 ret = submit_one_bio(epd->bio, 0, 0);
204 * Clean up of epd->bio is handled by its endio function.
205 * And endio is either triggered by successful bio execution
206 * or the error handler of submit bio hook.
207 * So at this point, no matter what happened, we don't need
208 * to clean up epd->bio.
215 int __init extent_state_cache_init(void)
217 extent_state_cache = kmem_cache_create("btrfs_extent_state",
218 sizeof(struct extent_state), 0,
219 SLAB_MEM_SPREAD, NULL);
220 if (!extent_state_cache)
225 int __init extent_io_init(void)
227 extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer",
228 sizeof(struct extent_buffer), 0,
229 SLAB_MEM_SPREAD, NULL);
230 if (!extent_buffer_cache)
233 if (bioset_init(&btrfs_bioset, BIO_POOL_SIZE,
234 offsetof(struct btrfs_io_bio, bio),
236 goto free_buffer_cache;
238 if (bioset_integrity_create(&btrfs_bioset, BIO_POOL_SIZE))
244 bioset_exit(&btrfs_bioset);
247 kmem_cache_destroy(extent_buffer_cache);
248 extent_buffer_cache = NULL;
252 void __cold extent_state_cache_exit(void)
254 btrfs_extent_state_leak_debug_check();
255 kmem_cache_destroy(extent_state_cache);
258 void __cold extent_io_exit(void)
261 * Make sure all delayed rcu free are flushed before we
265 kmem_cache_destroy(extent_buffer_cache);
266 bioset_exit(&btrfs_bioset);
270 * For the file_extent_tree, we want to hold the inode lock when we lookup and
271 * update the disk_i_size, but lockdep will complain because our io_tree we hold
272 * the tree lock and get the inode lock when setting delalloc. These two things
273 * are unrelated, so make a class for the file_extent_tree so we don't get the
274 * two locking patterns mixed up.
276 static struct lock_class_key file_extent_tree_class;
278 void extent_io_tree_init(struct btrfs_fs_info *fs_info,
279 struct extent_io_tree *tree, unsigned int owner,
282 tree->fs_info = fs_info;
283 tree->state = RB_ROOT;
284 tree->dirty_bytes = 0;
285 spin_lock_init(&tree->lock);
286 tree->private_data = private_data;
288 if (owner == IO_TREE_INODE_FILE_EXTENT)
289 lockdep_set_class(&tree->lock, &file_extent_tree_class);
292 void extent_io_tree_release(struct extent_io_tree *tree)
294 spin_lock(&tree->lock);
296 * Do a single barrier for the waitqueue_active check here, the state
297 * of the waitqueue should not change once extent_io_tree_release is
301 while (!RB_EMPTY_ROOT(&tree->state)) {
302 struct rb_node *node;
303 struct extent_state *state;
305 node = rb_first(&tree->state);
306 state = rb_entry(node, struct extent_state, rb_node);
307 rb_erase(&state->rb_node, &tree->state);
308 RB_CLEAR_NODE(&state->rb_node);
310 * btree io trees aren't supposed to have tasks waiting for
311 * changes in the flags of extent states ever.
313 ASSERT(!waitqueue_active(&state->wq));
314 free_extent_state(state);
316 cond_resched_lock(&tree->lock);
318 spin_unlock(&tree->lock);
321 static struct extent_state *alloc_extent_state(gfp_t mask)
323 struct extent_state *state;
326 * The given mask might be not appropriate for the slab allocator,
327 * drop the unsupported bits
329 mask &= ~(__GFP_DMA32|__GFP_HIGHMEM);
330 state = kmem_cache_alloc(extent_state_cache, mask);
334 state->failrec = NULL;
335 RB_CLEAR_NODE(&state->rb_node);
336 btrfs_leak_debug_add(&leak_lock, &state->leak_list, &states);
337 refcount_set(&state->refs, 1);
338 init_waitqueue_head(&state->wq);
339 trace_alloc_extent_state(state, mask, _RET_IP_);
343 void free_extent_state(struct extent_state *state)
347 if (refcount_dec_and_test(&state->refs)) {
348 WARN_ON(extent_state_in_tree(state));
349 btrfs_leak_debug_del(&leak_lock, &state->leak_list);
350 trace_free_extent_state(state, _RET_IP_);
351 kmem_cache_free(extent_state_cache, state);
355 static struct rb_node *tree_insert(struct rb_root *root,
356 struct rb_node *search_start,
358 struct rb_node *node,
359 struct rb_node ***p_in,
360 struct rb_node **parent_in)
363 struct rb_node *parent = NULL;
364 struct tree_entry *entry;
366 if (p_in && parent_in) {
372 p = search_start ? &search_start : &root->rb_node;
375 entry = rb_entry(parent, struct tree_entry, rb_node);
377 if (offset < entry->start)
379 else if (offset > entry->end)
386 rb_link_node(node, parent, p);
387 rb_insert_color(node, root);
392 * __etree_search - searche @tree for an entry that contains @offset. Such
393 * entry would have entry->start <= offset && entry->end >= offset.
395 * @tree - the tree to search
396 * @offset - offset that should fall within an entry in @tree
397 * @next_ret - pointer to the first entry whose range ends after @offset
398 * @prev - pointer to the first entry whose range begins before @offset
399 * @p_ret - pointer where new node should be anchored (used when inserting an
401 * @parent_ret - points to entry which would have been the parent of the entry,
404 * This function returns a pointer to the entry that contains @offset byte
405 * address. If no such entry exists, then NULL is returned and the other
406 * pointer arguments to the function are filled, otherwise the found entry is
407 * returned and other pointers are left untouched.
409 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
410 struct rb_node **next_ret,
411 struct rb_node **prev_ret,
412 struct rb_node ***p_ret,
413 struct rb_node **parent_ret)
415 struct rb_root *root = &tree->state;
416 struct rb_node **n = &root->rb_node;
417 struct rb_node *prev = NULL;
418 struct rb_node *orig_prev = NULL;
419 struct tree_entry *entry;
420 struct tree_entry *prev_entry = NULL;
424 entry = rb_entry(prev, struct tree_entry, rb_node);
427 if (offset < entry->start)
429 else if (offset > entry->end)
442 while (prev && offset > prev_entry->end) {
443 prev = rb_next(prev);
444 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
451 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
452 while (prev && offset < prev_entry->start) {
453 prev = rb_prev(prev);
454 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
461 static inline struct rb_node *
462 tree_search_for_insert(struct extent_io_tree *tree,
464 struct rb_node ***p_ret,
465 struct rb_node **parent_ret)
467 struct rb_node *next= NULL;
470 ret = __etree_search(tree, offset, &next, NULL, p_ret, parent_ret);
476 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
479 return tree_search_for_insert(tree, offset, NULL, NULL);
483 * utility function to look for merge candidates inside a given range.
484 * Any extents with matching state are merged together into a single
485 * extent in the tree. Extents with EXTENT_IO in their state field
486 * are not merged because the end_io handlers need to be able to do
487 * operations on them without sleeping (or doing allocations/splits).
489 * This should be called with the tree lock held.
491 static void merge_state(struct extent_io_tree *tree,
492 struct extent_state *state)
494 struct extent_state *other;
495 struct rb_node *other_node;
497 if (state->state & (EXTENT_LOCKED | EXTENT_BOUNDARY))
500 other_node = rb_prev(&state->rb_node);
502 other = rb_entry(other_node, struct extent_state, rb_node);
503 if (other->end == state->start - 1 &&
504 other->state == state->state) {
505 if (tree->private_data &&
506 is_data_inode(tree->private_data))
507 btrfs_merge_delalloc_extent(tree->private_data,
509 state->start = other->start;
510 rb_erase(&other->rb_node, &tree->state);
511 RB_CLEAR_NODE(&other->rb_node);
512 free_extent_state(other);
515 other_node = rb_next(&state->rb_node);
517 other = rb_entry(other_node, struct extent_state, rb_node);
518 if (other->start == state->end + 1 &&
519 other->state == state->state) {
520 if (tree->private_data &&
521 is_data_inode(tree->private_data))
522 btrfs_merge_delalloc_extent(tree->private_data,
524 state->end = other->end;
525 rb_erase(&other->rb_node, &tree->state);
526 RB_CLEAR_NODE(&other->rb_node);
527 free_extent_state(other);
532 static void set_state_bits(struct extent_io_tree *tree,
533 struct extent_state *state, unsigned *bits,
534 struct extent_changeset *changeset);
537 * insert an extent_state struct into the tree. 'bits' are set on the
538 * struct before it is inserted.
540 * This may return -EEXIST if the extent is already there, in which case the
541 * state struct is freed.
543 * The tree lock is not taken internally. This is a utility function and
544 * probably isn't what you want to call (see set/clear_extent_bit).
546 static int insert_state(struct extent_io_tree *tree,
547 struct extent_state *state, u64 start, u64 end,
549 struct rb_node **parent,
550 unsigned *bits, struct extent_changeset *changeset)
552 struct rb_node *node;
555 btrfs_err(tree->fs_info,
556 "insert state: end < start %llu %llu", end, start);
559 state->start = start;
562 set_state_bits(tree, state, bits, changeset);
564 node = tree_insert(&tree->state, NULL, end, &state->rb_node, p, parent);
566 struct extent_state *found;
567 found = rb_entry(node, struct extent_state, rb_node);
568 btrfs_err(tree->fs_info,
569 "found node %llu %llu on insert of %llu %llu",
570 found->start, found->end, start, end);
573 merge_state(tree, state);
578 * split a given extent state struct in two, inserting the preallocated
579 * struct 'prealloc' as the newly created second half. 'split' indicates an
580 * offset inside 'orig' where it should be split.
583 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
584 * are two extent state structs in the tree:
585 * prealloc: [orig->start, split - 1]
586 * orig: [ split, orig->end ]
588 * The tree locks are not taken by this function. They need to be held
591 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
592 struct extent_state *prealloc, u64 split)
594 struct rb_node *node;
596 if (tree->private_data && is_data_inode(tree->private_data))
597 btrfs_split_delalloc_extent(tree->private_data, orig, split);
599 prealloc->start = orig->start;
600 prealloc->end = split - 1;
601 prealloc->state = orig->state;
604 node = tree_insert(&tree->state, &orig->rb_node, prealloc->end,
605 &prealloc->rb_node, NULL, NULL);
607 free_extent_state(prealloc);
613 static struct extent_state *next_state(struct extent_state *state)
615 struct rb_node *next = rb_next(&state->rb_node);
617 return rb_entry(next, struct extent_state, rb_node);
623 * utility function to clear some bits in an extent state struct.
624 * it will optionally wake up anyone waiting on this state (wake == 1).
626 * If no bits are set on the state struct after clearing things, the
627 * struct is freed and removed from the tree
629 static struct extent_state *clear_state_bit(struct extent_io_tree *tree,
630 struct extent_state *state,
631 unsigned *bits, int wake,
632 struct extent_changeset *changeset)
634 struct extent_state *next;
635 unsigned bits_to_clear = *bits & ~EXTENT_CTLBITS;
638 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
639 u64 range = state->end - state->start + 1;
640 WARN_ON(range > tree->dirty_bytes);
641 tree->dirty_bytes -= range;
644 if (tree->private_data && is_data_inode(tree->private_data))
645 btrfs_clear_delalloc_extent(tree->private_data, state, bits);
647 ret = add_extent_changeset(state, bits_to_clear, changeset, 0);
649 state->state &= ~bits_to_clear;
652 if (state->state == 0) {
653 next = next_state(state);
654 if (extent_state_in_tree(state)) {
655 rb_erase(&state->rb_node, &tree->state);
656 RB_CLEAR_NODE(&state->rb_node);
657 free_extent_state(state);
662 merge_state(tree, state);
663 next = next_state(state);
668 static struct extent_state *
669 alloc_extent_state_atomic(struct extent_state *prealloc)
672 prealloc = alloc_extent_state(GFP_ATOMIC);
677 static void extent_io_tree_panic(struct extent_io_tree *tree, int err)
679 struct inode *inode = tree->private_data;
681 btrfs_panic(btrfs_sb(inode->i_sb), err,
682 "locking error: extent tree was modified by another thread while locked");
686 * clear some bits on a range in the tree. This may require splitting
687 * or inserting elements in the tree, so the gfp mask is used to
688 * indicate which allocations or sleeping are allowed.
690 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
691 * the given range from the tree regardless of state (ie for truncate).
693 * the range [start, end] is inclusive.
695 * This takes the tree lock, and returns 0 on success and < 0 on error.
697 int __clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
698 unsigned bits, int wake, int delete,
699 struct extent_state **cached_state,
700 gfp_t mask, struct extent_changeset *changeset)
702 struct extent_state *state;
703 struct extent_state *cached;
704 struct extent_state *prealloc = NULL;
705 struct rb_node *node;
710 btrfs_debug_check_extent_io_range(tree, start, end);
711 trace_btrfs_clear_extent_bit(tree, start, end - start + 1, bits);
713 if (bits & EXTENT_DELALLOC)
714 bits |= EXTENT_NORESERVE;
717 bits |= ~EXTENT_CTLBITS;
719 if (bits & (EXTENT_LOCKED | EXTENT_BOUNDARY))
722 if (!prealloc && gfpflags_allow_blocking(mask)) {
724 * Don't care for allocation failure here because we might end
725 * up not needing the pre-allocated extent state at all, which
726 * is the case if we only have in the tree extent states that
727 * cover our input range and don't cover too any other range.
728 * If we end up needing a new extent state we allocate it later.
730 prealloc = alloc_extent_state(mask);
733 spin_lock(&tree->lock);
735 cached = *cached_state;
738 *cached_state = NULL;
742 if (cached && extent_state_in_tree(cached) &&
743 cached->start <= start && cached->end > start) {
745 refcount_dec(&cached->refs);
750 free_extent_state(cached);
753 * this search will find the extents that end after
756 node = tree_search(tree, start);
759 state = rb_entry(node, struct extent_state, rb_node);
761 if (state->start > end)
763 WARN_ON(state->end < start);
764 last_end = state->end;
766 /* the state doesn't have the wanted bits, go ahead */
767 if (!(state->state & bits)) {
768 state = next_state(state);
773 * | ---- desired range ---- |
775 * | ------------- state -------------- |
777 * We need to split the extent we found, and may flip
778 * bits on second half.
780 * If the extent we found extends past our range, we
781 * just split and search again. It'll get split again
782 * the next time though.
784 * If the extent we found is inside our range, we clear
785 * the desired bit on it.
788 if (state->start < start) {
789 prealloc = alloc_extent_state_atomic(prealloc);
791 err = split_state(tree, state, prealloc, start);
793 extent_io_tree_panic(tree, err);
798 if (state->end <= end) {
799 state = clear_state_bit(tree, state, &bits, wake,
806 * | ---- desired range ---- |
808 * We need to split the extent, and clear the bit
811 if (state->start <= end && state->end > end) {
812 prealloc = alloc_extent_state_atomic(prealloc);
814 err = split_state(tree, state, prealloc, end + 1);
816 extent_io_tree_panic(tree, err);
821 clear_state_bit(tree, prealloc, &bits, wake, changeset);
827 state = clear_state_bit(tree, state, &bits, wake, changeset);
829 if (last_end == (u64)-1)
831 start = last_end + 1;
832 if (start <= end && state && !need_resched())
838 spin_unlock(&tree->lock);
839 if (gfpflags_allow_blocking(mask))
844 spin_unlock(&tree->lock);
846 free_extent_state(prealloc);
852 static void wait_on_state(struct extent_io_tree *tree,
853 struct extent_state *state)
854 __releases(tree->lock)
855 __acquires(tree->lock)
858 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
859 spin_unlock(&tree->lock);
861 spin_lock(&tree->lock);
862 finish_wait(&state->wq, &wait);
866 * waits for one or more bits to clear on a range in the state tree.
867 * The range [start, end] is inclusive.
868 * The tree lock is taken by this function
870 static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
873 struct extent_state *state;
874 struct rb_node *node;
876 btrfs_debug_check_extent_io_range(tree, start, end);
878 spin_lock(&tree->lock);
882 * this search will find all the extents that end after
885 node = tree_search(tree, start);
890 state = rb_entry(node, struct extent_state, rb_node);
892 if (state->start > end)
895 if (state->state & bits) {
896 start = state->start;
897 refcount_inc(&state->refs);
898 wait_on_state(tree, state);
899 free_extent_state(state);
902 start = state->end + 1;
907 if (!cond_resched_lock(&tree->lock)) {
908 node = rb_next(node);
913 spin_unlock(&tree->lock);
916 static void set_state_bits(struct extent_io_tree *tree,
917 struct extent_state *state,
918 unsigned *bits, struct extent_changeset *changeset)
920 unsigned bits_to_set = *bits & ~EXTENT_CTLBITS;
923 if (tree->private_data && is_data_inode(tree->private_data))
924 btrfs_set_delalloc_extent(tree->private_data, state, bits);
926 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
927 u64 range = state->end - state->start + 1;
928 tree->dirty_bytes += range;
930 ret = add_extent_changeset(state, bits_to_set, changeset, 1);
932 state->state |= bits_to_set;
935 static void cache_state_if_flags(struct extent_state *state,
936 struct extent_state **cached_ptr,
939 if (cached_ptr && !(*cached_ptr)) {
940 if (!flags || (state->state & flags)) {
942 refcount_inc(&state->refs);
947 static void cache_state(struct extent_state *state,
948 struct extent_state **cached_ptr)
950 return cache_state_if_flags(state, cached_ptr,
951 EXTENT_LOCKED | EXTENT_BOUNDARY);
955 * set some bits on a range in the tree. This may require allocations or
956 * sleeping, so the gfp mask is used to indicate what is allowed.
958 * If any of the exclusive bits are set, this will fail with -EEXIST if some
959 * part of the range already has the desired bits set. The start of the
960 * existing range is returned in failed_start in this case.
962 * [start, end] is inclusive This takes the tree lock.
965 static int __must_check
966 __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
967 unsigned bits, unsigned exclusive_bits,
968 u64 *failed_start, struct extent_state **cached_state,
969 gfp_t mask, struct extent_changeset *changeset)
971 struct extent_state *state;
972 struct extent_state *prealloc = NULL;
973 struct rb_node *node;
975 struct rb_node *parent;
980 btrfs_debug_check_extent_io_range(tree, start, end);
981 trace_btrfs_set_extent_bit(tree, start, end - start + 1, bits);
984 if (!prealloc && gfpflags_allow_blocking(mask)) {
986 * Don't care for allocation failure here because we might end
987 * up not needing the pre-allocated extent state at all, which
988 * is the case if we only have in the tree extent states that
989 * cover our input range and don't cover too any other range.
990 * If we end up needing a new extent state we allocate it later.
992 prealloc = alloc_extent_state(mask);
995 spin_lock(&tree->lock);
996 if (cached_state && *cached_state) {
997 state = *cached_state;
998 if (state->start <= start && state->end > start &&
999 extent_state_in_tree(state)) {
1000 node = &state->rb_node;
1005 * this search will find all the extents that end after
1008 node = tree_search_for_insert(tree, start, &p, &parent);
1010 prealloc = alloc_extent_state_atomic(prealloc);
1012 err = insert_state(tree, prealloc, start, end,
1013 &p, &parent, &bits, changeset);
1015 extent_io_tree_panic(tree, err);
1017 cache_state(prealloc, cached_state);
1021 state = rb_entry(node, struct extent_state, rb_node);
1023 last_start = state->start;
1024 last_end = state->end;
1027 * | ---- desired range ---- |
1030 * Just lock what we found and keep going
1032 if (state->start == start && state->end <= end) {
1033 if (state->state & exclusive_bits) {
1034 *failed_start = state->start;
1039 set_state_bits(tree, state, &bits, changeset);
1040 cache_state(state, cached_state);
1041 merge_state(tree, state);
1042 if (last_end == (u64)-1)
1044 start = last_end + 1;
1045 state = next_state(state);
1046 if (start < end && state && state->start == start &&
1053 * | ---- desired range ---- |
1056 * | ------------- state -------------- |
1058 * We need to split the extent we found, and may flip bits on
1061 * If the extent we found extends past our
1062 * range, we just split and search again. It'll get split
1063 * again the next time though.
1065 * If the extent we found is inside our range, we set the
1066 * desired bit on it.
1068 if (state->start < start) {
1069 if (state->state & exclusive_bits) {
1070 *failed_start = start;
1076 * If this extent already has all the bits we want set, then
1077 * skip it, not necessary to split it or do anything with it.
1079 if ((state->state & bits) == bits) {
1080 start = state->end + 1;
1081 cache_state(state, cached_state);
1085 prealloc = alloc_extent_state_atomic(prealloc);
1087 err = split_state(tree, state, prealloc, start);
1089 extent_io_tree_panic(tree, err);
1094 if (state->end <= end) {
1095 set_state_bits(tree, state, &bits, changeset);
1096 cache_state(state, cached_state);
1097 merge_state(tree, state);
1098 if (last_end == (u64)-1)
1100 start = last_end + 1;
1101 state = next_state(state);
1102 if (start < end && state && state->start == start &&
1109 * | ---- desired range ---- |
1110 * | state | or | state |
1112 * There's a hole, we need to insert something in it and
1113 * ignore the extent we found.
1115 if (state->start > start) {
1117 if (end < last_start)
1120 this_end = last_start - 1;
1122 prealloc = alloc_extent_state_atomic(prealloc);
1126 * Avoid to free 'prealloc' if it can be merged with
1129 err = insert_state(tree, prealloc, start, this_end,
1130 NULL, NULL, &bits, changeset);
1132 extent_io_tree_panic(tree, err);
1134 cache_state(prealloc, cached_state);
1136 start = this_end + 1;
1140 * | ---- desired range ---- |
1142 * We need to split the extent, and set the bit
1145 if (state->start <= end && state->end > end) {
1146 if (state->state & exclusive_bits) {
1147 *failed_start = start;
1152 prealloc = alloc_extent_state_atomic(prealloc);
1154 err = split_state(tree, state, prealloc, end + 1);
1156 extent_io_tree_panic(tree, err);
1158 set_state_bits(tree, prealloc, &bits, changeset);
1159 cache_state(prealloc, cached_state);
1160 merge_state(tree, prealloc);
1168 spin_unlock(&tree->lock);
1169 if (gfpflags_allow_blocking(mask))
1174 spin_unlock(&tree->lock);
1176 free_extent_state(prealloc);
1182 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1183 unsigned bits, u64 * failed_start,
1184 struct extent_state **cached_state, gfp_t mask)
1186 return __set_extent_bit(tree, start, end, bits, 0, failed_start,
1187 cached_state, mask, NULL);
1192 * convert_extent_bit - convert all bits in a given range from one bit to
1194 * @tree: the io tree to search
1195 * @start: the start offset in bytes
1196 * @end: the end offset in bytes (inclusive)
1197 * @bits: the bits to set in this range
1198 * @clear_bits: the bits to clear in this range
1199 * @cached_state: state that we're going to cache
1201 * This will go through and set bits for the given range. If any states exist
1202 * already in this range they are set with the given bit and cleared of the
1203 * clear_bits. This is only meant to be used by things that are mergeable, ie
1204 * converting from say DELALLOC to DIRTY. This is not meant to be used with
1205 * boundary bits like LOCK.
1207 * All allocations are done with GFP_NOFS.
1209 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1210 unsigned bits, unsigned clear_bits,
1211 struct extent_state **cached_state)
1213 struct extent_state *state;
1214 struct extent_state *prealloc = NULL;
1215 struct rb_node *node;
1217 struct rb_node *parent;
1221 bool first_iteration = true;
1223 btrfs_debug_check_extent_io_range(tree, start, end);
1224 trace_btrfs_convert_extent_bit(tree, start, end - start + 1, bits,
1230 * Best effort, don't worry if extent state allocation fails
1231 * here for the first iteration. We might have a cached state
1232 * that matches exactly the target range, in which case no
1233 * extent state allocations are needed. We'll only know this
1234 * after locking the tree.
1236 prealloc = alloc_extent_state(GFP_NOFS);
1237 if (!prealloc && !first_iteration)
1241 spin_lock(&tree->lock);
1242 if (cached_state && *cached_state) {
1243 state = *cached_state;
1244 if (state->start <= start && state->end > start &&
1245 extent_state_in_tree(state)) {
1246 node = &state->rb_node;
1252 * this search will find all the extents that end after
1255 node = tree_search_for_insert(tree, start, &p, &parent);
1257 prealloc = alloc_extent_state_atomic(prealloc);
1262 err = insert_state(tree, prealloc, start, end,
1263 &p, &parent, &bits, NULL);
1265 extent_io_tree_panic(tree, err);
1266 cache_state(prealloc, cached_state);
1270 state = rb_entry(node, struct extent_state, rb_node);
1272 last_start = state->start;
1273 last_end = state->end;
1276 * | ---- desired range ---- |
1279 * Just lock what we found and keep going
1281 if (state->start == start && state->end <= end) {
1282 set_state_bits(tree, state, &bits, NULL);
1283 cache_state(state, cached_state);
1284 state = clear_state_bit(tree, state, &clear_bits, 0, NULL);
1285 if (last_end == (u64)-1)
1287 start = last_end + 1;
1288 if (start < end && state && state->start == start &&
1295 * | ---- desired range ---- |
1298 * | ------------- state -------------- |
1300 * We need to split the extent we found, and may flip bits on
1303 * If the extent we found extends past our
1304 * range, we just split and search again. It'll get split
1305 * again the next time though.
1307 * If the extent we found is inside our range, we set the
1308 * desired bit on it.
1310 if (state->start < start) {
1311 prealloc = alloc_extent_state_atomic(prealloc);
1316 err = split_state(tree, state, prealloc, start);
1318 extent_io_tree_panic(tree, err);
1322 if (state->end <= end) {
1323 set_state_bits(tree, state, &bits, NULL);
1324 cache_state(state, cached_state);
1325 state = clear_state_bit(tree, state, &clear_bits, 0,
1327 if (last_end == (u64)-1)
1329 start = last_end + 1;
1330 if (start < end && state && state->start == start &&
1337 * | ---- desired range ---- |
1338 * | state | or | state |
1340 * There's a hole, we need to insert something in it and
1341 * ignore the extent we found.
1343 if (state->start > start) {
1345 if (end < last_start)
1348 this_end = last_start - 1;
1350 prealloc = alloc_extent_state_atomic(prealloc);
1357 * Avoid to free 'prealloc' if it can be merged with
1360 err = insert_state(tree, prealloc, start, this_end,
1361 NULL, NULL, &bits, NULL);
1363 extent_io_tree_panic(tree, err);
1364 cache_state(prealloc, cached_state);
1366 start = this_end + 1;
1370 * | ---- desired range ---- |
1372 * We need to split the extent, and set the bit
1375 if (state->start <= end && state->end > end) {
1376 prealloc = alloc_extent_state_atomic(prealloc);
1382 err = split_state(tree, state, prealloc, end + 1);
1384 extent_io_tree_panic(tree, err);
1386 set_state_bits(tree, prealloc, &bits, NULL);
1387 cache_state(prealloc, cached_state);
1388 clear_state_bit(tree, prealloc, &clear_bits, 0, NULL);
1396 spin_unlock(&tree->lock);
1398 first_iteration = false;
1402 spin_unlock(&tree->lock);
1404 free_extent_state(prealloc);
1409 /* wrappers around set/clear extent bit */
1410 int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1411 unsigned bits, struct extent_changeset *changeset)
1414 * We don't support EXTENT_LOCKED yet, as current changeset will
1415 * record any bits changed, so for EXTENT_LOCKED case, it will
1416 * either fail with -EEXIST or changeset will record the whole
1419 BUG_ON(bits & EXTENT_LOCKED);
1421 return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL, GFP_NOFS,
1425 int set_extent_bits_nowait(struct extent_io_tree *tree, u64 start, u64 end,
1428 return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL,
1432 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1433 unsigned bits, int wake, int delete,
1434 struct extent_state **cached)
1436 return __clear_extent_bit(tree, start, end, bits, wake, delete,
1437 cached, GFP_NOFS, NULL);
1440 int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1441 unsigned bits, struct extent_changeset *changeset)
1444 * Don't support EXTENT_LOCKED case, same reason as
1445 * set_record_extent_bits().
1447 BUG_ON(bits & EXTENT_LOCKED);
1449 return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, GFP_NOFS,
1454 * either insert or lock state struct between start and end use mask to tell
1455 * us if waiting is desired.
1457 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1458 struct extent_state **cached_state)
1464 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED,
1465 EXTENT_LOCKED, &failed_start,
1466 cached_state, GFP_NOFS, NULL);
1467 if (err == -EEXIST) {
1468 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1469 start = failed_start;
1472 WARN_ON(start > end);
1477 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1482 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1483 &failed_start, NULL, GFP_NOFS, NULL);
1484 if (err == -EEXIST) {
1485 if (failed_start > start)
1486 clear_extent_bit(tree, start, failed_start - 1,
1487 EXTENT_LOCKED, 1, 0, NULL);
1493 void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
1495 unsigned long index = start >> PAGE_SHIFT;
1496 unsigned long end_index = end >> PAGE_SHIFT;
1499 while (index <= end_index) {
1500 page = find_get_page(inode->i_mapping, index);
1501 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1502 clear_page_dirty_for_io(page);
1508 void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
1510 unsigned long index = start >> PAGE_SHIFT;
1511 unsigned long end_index = end >> PAGE_SHIFT;
1514 while (index <= end_index) {
1515 page = find_get_page(inode->i_mapping, index);
1516 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1517 __set_page_dirty_nobuffers(page);
1518 account_page_redirty(page);
1524 /* find the first state struct with 'bits' set after 'start', and
1525 * return it. tree->lock must be held. NULL will returned if
1526 * nothing was found after 'start'
1528 static struct extent_state *
1529 find_first_extent_bit_state(struct extent_io_tree *tree,
1530 u64 start, unsigned bits)
1532 struct rb_node *node;
1533 struct extent_state *state;
1536 * this search will find all the extents that end after
1539 node = tree_search(tree, start);
1544 state = rb_entry(node, struct extent_state, rb_node);
1545 if (state->end >= start && (state->state & bits))
1548 node = rb_next(node);
1557 * find the first offset in the io tree with 'bits' set. zero is
1558 * returned if we find something, and *start_ret and *end_ret are
1559 * set to reflect the state struct that was found.
1561 * If nothing was found, 1 is returned. If found something, return 0.
1563 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1564 u64 *start_ret, u64 *end_ret, unsigned bits,
1565 struct extent_state **cached_state)
1567 struct extent_state *state;
1570 spin_lock(&tree->lock);
1571 if (cached_state && *cached_state) {
1572 state = *cached_state;
1573 if (state->end == start - 1 && extent_state_in_tree(state)) {
1574 while ((state = next_state(state)) != NULL) {
1575 if (state->state & bits)
1578 free_extent_state(*cached_state);
1579 *cached_state = NULL;
1582 free_extent_state(*cached_state);
1583 *cached_state = NULL;
1586 state = find_first_extent_bit_state(tree, start, bits);
1589 cache_state_if_flags(state, cached_state, 0);
1590 *start_ret = state->start;
1591 *end_ret = state->end;
1595 spin_unlock(&tree->lock);
1600 * find_contiguous_extent_bit: find a contiguous area of bits
1601 * @tree - io tree to check
1602 * @start - offset to start the search from
1603 * @start_ret - the first offset we found with the bits set
1604 * @end_ret - the final contiguous range of the bits that were set
1605 * @bits - bits to look for
1607 * set_extent_bit and clear_extent_bit can temporarily split contiguous ranges
1608 * to set bits appropriately, and then merge them again. During this time it
1609 * will drop the tree->lock, so use this helper if you want to find the actual
1610 * contiguous area for given bits. We will search to the first bit we find, and
1611 * then walk down the tree until we find a non-contiguous area. The area
1612 * returned will be the full contiguous area with the bits set.
1614 int find_contiguous_extent_bit(struct extent_io_tree *tree, u64 start,
1615 u64 *start_ret, u64 *end_ret, unsigned bits)
1617 struct extent_state *state;
1620 spin_lock(&tree->lock);
1621 state = find_first_extent_bit_state(tree, start, bits);
1623 *start_ret = state->start;
1624 *end_ret = state->end;
1625 while ((state = next_state(state)) != NULL) {
1626 if (state->start > (*end_ret + 1))
1628 *end_ret = state->end;
1632 spin_unlock(&tree->lock);
1637 * find_first_clear_extent_bit - find the first range that has @bits not set.
1638 * This range could start before @start.
1640 * @tree - the tree to search
1641 * @start - the offset at/after which the found extent should start
1642 * @start_ret - records the beginning of the range
1643 * @end_ret - records the end of the range (inclusive)
1644 * @bits - the set of bits which must be unset
1646 * Since unallocated range is also considered one which doesn't have the bits
1647 * set it's possible that @end_ret contains -1, this happens in case the range
1648 * spans (last_range_end, end of device]. In this case it's up to the caller to
1649 * trim @end_ret to the appropriate size.
1651 void find_first_clear_extent_bit(struct extent_io_tree *tree, u64 start,
1652 u64 *start_ret, u64 *end_ret, unsigned bits)
1654 struct extent_state *state;
1655 struct rb_node *node, *prev = NULL, *next;
1657 spin_lock(&tree->lock);
1659 /* Find first extent with bits cleared */
1661 node = __etree_search(tree, start, &next, &prev, NULL, NULL);
1662 if (!node && !next && !prev) {
1664 * Tree is completely empty, send full range and let
1665 * caller deal with it
1670 } else if (!node && !next) {
1672 * We are past the last allocated chunk, set start at
1673 * the end of the last extent.
1675 state = rb_entry(prev, struct extent_state, rb_node);
1676 *start_ret = state->end + 1;
1683 * At this point 'node' either contains 'start' or start is
1686 state = rb_entry(node, struct extent_state, rb_node);
1688 if (in_range(start, state->start, state->end - state->start + 1)) {
1689 if (state->state & bits) {
1691 * |--range with bits sets--|
1695 start = state->end + 1;
1698 * 'start' falls within a range that doesn't
1699 * have the bits set, so take its start as
1700 * the beginning of the desired range
1702 * |--range with bits cleared----|
1706 *start_ret = state->start;
1711 * |---prev range---|---hole/unset---|---node range---|
1717 * |---hole/unset--||--first node--|
1722 state = rb_entry(prev, struct extent_state,
1724 *start_ret = state->end + 1;
1733 * Find the longest stretch from start until an entry which has the
1737 state = rb_entry(node, struct extent_state, rb_node);
1738 if (state->end >= start && !(state->state & bits)) {
1739 *end_ret = state->end;
1741 *end_ret = state->start - 1;
1745 node = rb_next(node);
1750 spin_unlock(&tree->lock);
1754 * find a contiguous range of bytes in the file marked as delalloc, not
1755 * more than 'max_bytes'. start and end are used to return the range,
1757 * true is returned if we find something, false if nothing was in the tree
1759 bool btrfs_find_delalloc_range(struct extent_io_tree *tree, u64 *start,
1760 u64 *end, u64 max_bytes,
1761 struct extent_state **cached_state)
1763 struct rb_node *node;
1764 struct extent_state *state;
1765 u64 cur_start = *start;
1767 u64 total_bytes = 0;
1769 spin_lock(&tree->lock);
1772 * this search will find all the extents that end after
1775 node = tree_search(tree, cur_start);
1782 state = rb_entry(node, struct extent_state, rb_node);
1783 if (found && (state->start != cur_start ||
1784 (state->state & EXTENT_BOUNDARY))) {
1787 if (!(state->state & EXTENT_DELALLOC)) {
1793 *start = state->start;
1794 *cached_state = state;
1795 refcount_inc(&state->refs);
1799 cur_start = state->end + 1;
1800 node = rb_next(node);
1801 total_bytes += state->end - state->start + 1;
1802 if (total_bytes >= max_bytes)
1808 spin_unlock(&tree->lock);
1812 static int __process_pages_contig(struct address_space *mapping,
1813 struct page *locked_page,
1814 pgoff_t start_index, pgoff_t end_index,
1815 unsigned long page_ops, pgoff_t *index_ret);
1817 static noinline void __unlock_for_delalloc(struct inode *inode,
1818 struct page *locked_page,
1821 unsigned long index = start >> PAGE_SHIFT;
1822 unsigned long end_index = end >> PAGE_SHIFT;
1824 ASSERT(locked_page);
1825 if (index == locked_page->index && end_index == index)
1828 __process_pages_contig(inode->i_mapping, locked_page, index, end_index,
1832 static noinline int lock_delalloc_pages(struct inode *inode,
1833 struct page *locked_page,
1837 unsigned long index = delalloc_start >> PAGE_SHIFT;
1838 unsigned long index_ret = index;
1839 unsigned long end_index = delalloc_end >> PAGE_SHIFT;
1842 ASSERT(locked_page);
1843 if (index == locked_page->index && index == end_index)
1846 ret = __process_pages_contig(inode->i_mapping, locked_page, index,
1847 end_index, PAGE_LOCK, &index_ret);
1849 __unlock_for_delalloc(inode, locked_page, delalloc_start,
1850 (u64)index_ret << PAGE_SHIFT);
1855 * Find and lock a contiguous range of bytes in the file marked as delalloc, no
1856 * more than @max_bytes. @Start and @end are used to return the range,
1858 * Return: true if we find something
1859 * false if nothing was in the tree
1862 noinline_for_stack bool find_lock_delalloc_range(struct inode *inode,
1863 struct page *locked_page, u64 *start,
1866 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
1867 u64 max_bytes = BTRFS_MAX_EXTENT_SIZE;
1871 struct extent_state *cached_state = NULL;
1876 /* step one, find a bunch of delalloc bytes starting at start */
1877 delalloc_start = *start;
1879 found = btrfs_find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1880 max_bytes, &cached_state);
1881 if (!found || delalloc_end <= *start) {
1882 *start = delalloc_start;
1883 *end = delalloc_end;
1884 free_extent_state(cached_state);
1889 * start comes from the offset of locked_page. We have to lock
1890 * pages in order, so we can't process delalloc bytes before
1893 if (delalloc_start < *start)
1894 delalloc_start = *start;
1897 * make sure to limit the number of pages we try to lock down
1899 if (delalloc_end + 1 - delalloc_start > max_bytes)
1900 delalloc_end = delalloc_start + max_bytes - 1;
1902 /* step two, lock all the pages after the page that has start */
1903 ret = lock_delalloc_pages(inode, locked_page,
1904 delalloc_start, delalloc_end);
1905 ASSERT(!ret || ret == -EAGAIN);
1906 if (ret == -EAGAIN) {
1907 /* some of the pages are gone, lets avoid looping by
1908 * shortening the size of the delalloc range we're searching
1910 free_extent_state(cached_state);
1911 cached_state = NULL;
1913 max_bytes = PAGE_SIZE;
1922 /* step three, lock the state bits for the whole range */
1923 lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state);
1925 /* then test to make sure it is all still delalloc */
1926 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1927 EXTENT_DELALLOC, 1, cached_state);
1929 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1931 __unlock_for_delalloc(inode, locked_page,
1932 delalloc_start, delalloc_end);
1936 free_extent_state(cached_state);
1937 *start = delalloc_start;
1938 *end = delalloc_end;
1943 static int __process_pages_contig(struct address_space *mapping,
1944 struct page *locked_page,
1945 pgoff_t start_index, pgoff_t end_index,
1946 unsigned long page_ops, pgoff_t *index_ret)
1948 unsigned long nr_pages = end_index - start_index + 1;
1949 unsigned long pages_locked = 0;
1950 pgoff_t index = start_index;
1951 struct page *pages[16];
1956 if (page_ops & PAGE_LOCK) {
1957 ASSERT(page_ops == PAGE_LOCK);
1958 ASSERT(index_ret && *index_ret == start_index);
1961 if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0)
1962 mapping_set_error(mapping, -EIO);
1964 while (nr_pages > 0) {
1965 ret = find_get_pages_contig(mapping, index,
1966 min_t(unsigned long,
1967 nr_pages, ARRAY_SIZE(pages)), pages);
1970 * Only if we're going to lock these pages,
1971 * can we find nothing at @index.
1973 ASSERT(page_ops & PAGE_LOCK);
1978 for (i = 0; i < ret; i++) {
1979 if (page_ops & PAGE_SET_PRIVATE2)
1980 SetPagePrivate2(pages[i]);
1982 if (locked_page && pages[i] == locked_page) {
1987 if (page_ops & PAGE_CLEAR_DIRTY)
1988 clear_page_dirty_for_io(pages[i]);
1989 if (page_ops & PAGE_SET_WRITEBACK)
1990 set_page_writeback(pages[i]);
1991 if (page_ops & PAGE_SET_ERROR)
1992 SetPageError(pages[i]);
1993 if (page_ops & PAGE_END_WRITEBACK)
1994 end_page_writeback(pages[i]);
1995 if (page_ops & PAGE_UNLOCK)
1996 unlock_page(pages[i]);
1997 if (page_ops & PAGE_LOCK) {
1998 lock_page(pages[i]);
1999 if (!PageDirty(pages[i]) ||
2000 pages[i]->mapping != mapping) {
2001 unlock_page(pages[i]);
2002 for (; i < ret; i++)
2016 if (err && index_ret)
2017 *index_ret = start_index + pages_locked - 1;
2021 void extent_clear_unlock_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2022 struct page *locked_page,
2023 unsigned clear_bits,
2024 unsigned long page_ops)
2026 clear_extent_bit(&inode->io_tree, start, end, clear_bits, 1, 0, NULL);
2028 __process_pages_contig(inode->vfs_inode.i_mapping, locked_page,
2029 start >> PAGE_SHIFT, end >> PAGE_SHIFT,
2034 * count the number of bytes in the tree that have a given bit(s)
2035 * set. This can be fairly slow, except for EXTENT_DIRTY which is
2036 * cached. The total number found is returned.
2038 u64 count_range_bits(struct extent_io_tree *tree,
2039 u64 *start, u64 search_end, u64 max_bytes,
2040 unsigned bits, int contig)
2042 struct rb_node *node;
2043 struct extent_state *state;
2044 u64 cur_start = *start;
2045 u64 total_bytes = 0;
2049 if (WARN_ON(search_end <= cur_start))
2052 spin_lock(&tree->lock);
2053 if (cur_start == 0 && bits == EXTENT_DIRTY) {
2054 total_bytes = tree->dirty_bytes;
2058 * this search will find all the extents that end after
2061 node = tree_search(tree, cur_start);
2066 state = rb_entry(node, struct extent_state, rb_node);
2067 if (state->start > search_end)
2069 if (contig && found && state->start > last + 1)
2071 if (state->end >= cur_start && (state->state & bits) == bits) {
2072 total_bytes += min(search_end, state->end) + 1 -
2073 max(cur_start, state->start);
2074 if (total_bytes >= max_bytes)
2077 *start = max(cur_start, state->start);
2081 } else if (contig && found) {
2084 node = rb_next(node);
2089 spin_unlock(&tree->lock);
2094 * set the private field for a given byte offset in the tree. If there isn't
2095 * an extent_state there already, this does nothing.
2097 int set_state_failrec(struct extent_io_tree *tree, u64 start,
2098 struct io_failure_record *failrec)
2100 struct rb_node *node;
2101 struct extent_state *state;
2104 spin_lock(&tree->lock);
2106 * this search will find all the extents that end after
2109 node = tree_search(tree, start);
2114 state = rb_entry(node, struct extent_state, rb_node);
2115 if (state->start != start) {
2119 state->failrec = failrec;
2121 spin_unlock(&tree->lock);
2125 struct io_failure_record *get_state_failrec(struct extent_io_tree *tree, u64 start)
2127 struct rb_node *node;
2128 struct extent_state *state;
2129 struct io_failure_record *failrec;
2131 spin_lock(&tree->lock);
2133 * this search will find all the extents that end after
2136 node = tree_search(tree, start);
2138 failrec = ERR_PTR(-ENOENT);
2141 state = rb_entry(node, struct extent_state, rb_node);
2142 if (state->start != start) {
2143 failrec = ERR_PTR(-ENOENT);
2147 failrec = state->failrec;
2149 spin_unlock(&tree->lock);
2154 * searches a range in the state tree for a given mask.
2155 * If 'filled' == 1, this returns 1 only if every extent in the tree
2156 * has the bits set. Otherwise, 1 is returned if any bit in the
2157 * range is found set.
2159 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
2160 unsigned bits, int filled, struct extent_state *cached)
2162 struct extent_state *state = NULL;
2163 struct rb_node *node;
2166 spin_lock(&tree->lock);
2167 if (cached && extent_state_in_tree(cached) && cached->start <= start &&
2168 cached->end > start)
2169 node = &cached->rb_node;
2171 node = tree_search(tree, start);
2172 while (node && start <= end) {
2173 state = rb_entry(node, struct extent_state, rb_node);
2175 if (filled && state->start > start) {
2180 if (state->start > end)
2183 if (state->state & bits) {
2187 } else if (filled) {
2192 if (state->end == (u64)-1)
2195 start = state->end + 1;
2198 node = rb_next(node);
2205 spin_unlock(&tree->lock);
2210 * helper function to set a given page up to date if all the
2211 * extents in the tree for that page are up to date
2213 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
2215 u64 start = page_offset(page);
2216 u64 end = start + PAGE_SIZE - 1;
2217 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
2218 SetPageUptodate(page);
2221 int free_io_failure(struct extent_io_tree *failure_tree,
2222 struct extent_io_tree *io_tree,
2223 struct io_failure_record *rec)
2228 set_state_failrec(failure_tree, rec->start, NULL);
2229 ret = clear_extent_bits(failure_tree, rec->start,
2230 rec->start + rec->len - 1,
2231 EXTENT_LOCKED | EXTENT_DIRTY);
2235 ret = clear_extent_bits(io_tree, rec->start,
2236 rec->start + rec->len - 1,
2246 * this bypasses the standard btrfs submit functions deliberately, as
2247 * the standard behavior is to write all copies in a raid setup. here we only
2248 * want to write the one bad copy. so we do the mapping for ourselves and issue
2249 * submit_bio directly.
2250 * to avoid any synchronization issues, wait for the data after writing, which
2251 * actually prevents the read that triggered the error from finishing.
2252 * currently, there can be no more than two copies of every data bit. thus,
2253 * exactly one rewrite is required.
2255 int repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start,
2256 u64 length, u64 logical, struct page *page,
2257 unsigned int pg_offset, int mirror_num)
2260 struct btrfs_device *dev;
2263 struct btrfs_bio *bbio = NULL;
2266 ASSERT(!(fs_info->sb->s_flags & SB_RDONLY));
2267 BUG_ON(!mirror_num);
2269 bio = btrfs_io_bio_alloc(1);
2270 bio->bi_iter.bi_size = 0;
2271 map_length = length;
2274 * Avoid races with device replace and make sure our bbio has devices
2275 * associated to its stripes that don't go away while we are doing the
2276 * read repair operation.
2278 btrfs_bio_counter_inc_blocked(fs_info);
2279 if (btrfs_is_parity_mirror(fs_info, logical, length)) {
2281 * Note that we don't use BTRFS_MAP_WRITE because it's supposed
2282 * to update all raid stripes, but here we just want to correct
2283 * bad stripe, thus BTRFS_MAP_READ is abused to only get the bad
2284 * stripe's dev and sector.
2286 ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, logical,
2287 &map_length, &bbio, 0);
2289 btrfs_bio_counter_dec(fs_info);
2293 ASSERT(bbio->mirror_num == 1);
2295 ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical,
2296 &map_length, &bbio, mirror_num);
2298 btrfs_bio_counter_dec(fs_info);
2302 BUG_ON(mirror_num != bbio->mirror_num);
2305 sector = bbio->stripes[bbio->mirror_num - 1].physical >> 9;
2306 bio->bi_iter.bi_sector = sector;
2307 dev = bbio->stripes[bbio->mirror_num - 1].dev;
2308 btrfs_put_bbio(bbio);
2309 if (!dev || !dev->bdev ||
2310 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) {
2311 btrfs_bio_counter_dec(fs_info);
2315 bio_set_dev(bio, dev->bdev);
2316 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC;
2317 bio_add_page(bio, page, length, pg_offset);
2319 if (btrfsic_submit_bio_wait(bio)) {
2320 /* try to remap that extent elsewhere? */
2321 btrfs_bio_counter_dec(fs_info);
2323 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
2327 btrfs_info_rl_in_rcu(fs_info,
2328 "read error corrected: ino %llu off %llu (dev %s sector %llu)",
2330 rcu_str_deref(dev->name), sector);
2331 btrfs_bio_counter_dec(fs_info);
2336 int btrfs_repair_eb_io_failure(const struct extent_buffer *eb, int mirror_num)
2338 struct btrfs_fs_info *fs_info = eb->fs_info;
2339 u64 start = eb->start;
2340 int i, num_pages = num_extent_pages(eb);
2343 if (sb_rdonly(fs_info->sb))
2346 for (i = 0; i < num_pages; i++) {
2347 struct page *p = eb->pages[i];
2349 ret = repair_io_failure(fs_info, 0, start, PAGE_SIZE, start, p,
2350 start - page_offset(p), mirror_num);
2360 * each time an IO finishes, we do a fast check in the IO failure tree
2361 * to see if we need to process or clean up an io_failure_record
2363 int clean_io_failure(struct btrfs_fs_info *fs_info,
2364 struct extent_io_tree *failure_tree,
2365 struct extent_io_tree *io_tree, u64 start,
2366 struct page *page, u64 ino, unsigned int pg_offset)
2369 struct io_failure_record *failrec;
2370 struct extent_state *state;
2375 ret = count_range_bits(failure_tree, &private, (u64)-1, 1,
2380 failrec = get_state_failrec(failure_tree, start);
2381 if (IS_ERR(failrec))
2384 BUG_ON(!failrec->this_mirror);
2386 if (failrec->in_validation) {
2387 /* there was no real error, just free the record */
2388 btrfs_debug(fs_info,
2389 "clean_io_failure: freeing dummy error at %llu",
2393 if (sb_rdonly(fs_info->sb))
2396 spin_lock(&io_tree->lock);
2397 state = find_first_extent_bit_state(io_tree,
2400 spin_unlock(&io_tree->lock);
2402 if (state && state->start <= failrec->start &&
2403 state->end >= failrec->start + failrec->len - 1) {
2404 num_copies = btrfs_num_copies(fs_info, failrec->logical,
2406 if (num_copies > 1) {
2407 repair_io_failure(fs_info, ino, start, failrec->len,
2408 failrec->logical, page, pg_offset,
2409 failrec->failed_mirror);
2414 free_io_failure(failure_tree, io_tree, failrec);
2420 * Can be called when
2421 * - hold extent lock
2422 * - under ordered extent
2423 * - the inode is freeing
2425 void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start, u64 end)
2427 struct extent_io_tree *failure_tree = &inode->io_failure_tree;
2428 struct io_failure_record *failrec;
2429 struct extent_state *state, *next;
2431 if (RB_EMPTY_ROOT(&failure_tree->state))
2434 spin_lock(&failure_tree->lock);
2435 state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY);
2437 if (state->start > end)
2440 ASSERT(state->end <= end);
2442 next = next_state(state);
2444 failrec = state->failrec;
2445 free_extent_state(state);
2450 spin_unlock(&failure_tree->lock);
2453 static struct io_failure_record *btrfs_get_io_failure_record(struct inode *inode,
2456 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2457 struct io_failure_record *failrec;
2458 struct extent_map *em;
2459 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2460 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2461 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2465 failrec = get_state_failrec(failure_tree, start);
2466 if (!IS_ERR(failrec)) {
2467 btrfs_debug(fs_info,
2468 "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d",
2469 failrec->logical, failrec->start, failrec->len,
2470 failrec->in_validation);
2472 * when data can be on disk more than twice, add to failrec here
2473 * (e.g. with a list for failed_mirror) to make
2474 * clean_io_failure() clean all those errors at once.
2480 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2482 return ERR_PTR(-ENOMEM);
2484 failrec->start = start;
2485 failrec->len = end - start + 1;
2486 failrec->this_mirror = 0;
2487 failrec->bio_flags = 0;
2488 failrec->in_validation = 0;
2490 read_lock(&em_tree->lock);
2491 em = lookup_extent_mapping(em_tree, start, failrec->len);
2493 read_unlock(&em_tree->lock);
2495 return ERR_PTR(-EIO);
2498 if (em->start > start || em->start + em->len <= start) {
2499 free_extent_map(em);
2502 read_unlock(&em_tree->lock);
2505 return ERR_PTR(-EIO);
2508 logical = start - em->start;
2509 logical = em->block_start + logical;
2510 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2511 logical = em->block_start;
2512 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2513 extent_set_compress_type(&failrec->bio_flags, em->compress_type);
2516 btrfs_debug(fs_info,
2517 "Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu",
2518 logical, start, failrec->len);
2520 failrec->logical = logical;
2521 free_extent_map(em);
2523 /* Set the bits in the private failure tree */
2524 ret = set_extent_bits(failure_tree, start, end,
2525 EXTENT_LOCKED | EXTENT_DIRTY);
2527 ret = set_state_failrec(failure_tree, start, failrec);
2528 /* Set the bits in the inode's tree */
2529 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED);
2530 } else if (ret < 0) {
2532 return ERR_PTR(ret);
2538 static bool btrfs_check_repairable(struct inode *inode, bool needs_validation,
2539 struct io_failure_record *failrec,
2542 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2545 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
2546 if (num_copies == 1) {
2548 * we only have a single copy of the data, so don't bother with
2549 * all the retry and error correction code that follows. no
2550 * matter what the error is, it is very likely to persist.
2552 btrfs_debug(fs_info,
2553 "Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
2554 num_copies, failrec->this_mirror, failed_mirror);
2559 * there are two premises:
2560 * a) deliver good data to the caller
2561 * b) correct the bad sectors on disk
2563 if (needs_validation) {
2565 * to fulfill b), we need to know the exact failing sectors, as
2566 * we don't want to rewrite any more than the failed ones. thus,
2567 * we need separate read requests for the failed bio
2569 * if the following BUG_ON triggers, our validation request got
2570 * merged. we need separate requests for our algorithm to work.
2572 BUG_ON(failrec->in_validation);
2573 failrec->in_validation = 1;
2574 failrec->this_mirror = failed_mirror;
2577 * we're ready to fulfill a) and b) alongside. get a good copy
2578 * of the failed sector and if we succeed, we have setup
2579 * everything for repair_io_failure to do the rest for us.
2581 if (failrec->in_validation) {
2582 BUG_ON(failrec->this_mirror != failed_mirror);
2583 failrec->in_validation = 0;
2584 failrec->this_mirror = 0;
2586 failrec->failed_mirror = failed_mirror;
2587 failrec->this_mirror++;
2588 if (failrec->this_mirror == failed_mirror)
2589 failrec->this_mirror++;
2592 if (failrec->this_mirror > num_copies) {
2593 btrfs_debug(fs_info,
2594 "Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
2595 num_copies, failrec->this_mirror, failed_mirror);
2602 static bool btrfs_io_needs_validation(struct inode *inode, struct bio *bio)
2605 const u32 blocksize = inode->i_sb->s_blocksize;
2608 * If bi_status is BLK_STS_OK, then this was a checksum error, not an
2609 * I/O error. In this case, we already know exactly which sector was
2610 * bad, so we don't need to validate.
2612 if (bio->bi_status == BLK_STS_OK)
2616 * We need to validate each sector individually if the failed I/O was
2617 * for multiple sectors.
2619 * There are a few possible bios that can end up here:
2620 * 1. A buffered read bio, which is not cloned.
2621 * 2. A direct I/O read bio, which is cloned.
2622 * 3. A (buffered or direct) repair bio, which is not cloned.
2624 * For cloned bios (case 2), we can get the size from
2625 * btrfs_io_bio->iter; for non-cloned bios (cases 1 and 3), we can get
2626 * it from the bvecs.
2628 if (bio_flagged(bio, BIO_CLONED)) {
2629 if (btrfs_io_bio(bio)->iter.bi_size > blocksize)
2632 struct bio_vec *bvec;
2635 bio_for_each_bvec_all(bvec, bio, i) {
2636 len += bvec->bv_len;
2637 if (len > blocksize)
2644 blk_status_t btrfs_submit_read_repair(struct inode *inode,
2645 struct bio *failed_bio, u64 phy_offset,
2646 struct page *page, unsigned int pgoff,
2647 u64 start, u64 end, int failed_mirror,
2648 submit_bio_hook_t *submit_bio_hook)
2650 struct io_failure_record *failrec;
2651 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2652 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2653 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2654 struct btrfs_io_bio *failed_io_bio = btrfs_io_bio(failed_bio);
2655 const int icsum = phy_offset >> inode->i_sb->s_blocksize_bits;
2656 bool need_validation;
2657 struct bio *repair_bio;
2658 struct btrfs_io_bio *repair_io_bio;
2659 blk_status_t status;
2661 btrfs_debug(fs_info,
2662 "repair read error: read error at %llu", start);
2664 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
2666 failrec = btrfs_get_io_failure_record(inode, start, end);
2667 if (IS_ERR(failrec))
2668 return errno_to_blk_status(PTR_ERR(failrec));
2670 need_validation = btrfs_io_needs_validation(inode, failed_bio);
2672 if (!btrfs_check_repairable(inode, need_validation, failrec,
2674 free_io_failure(failure_tree, tree, failrec);
2675 return BLK_STS_IOERR;
2678 repair_bio = btrfs_io_bio_alloc(1);
2679 repair_io_bio = btrfs_io_bio(repair_bio);
2680 repair_bio->bi_opf = REQ_OP_READ;
2681 if (need_validation)
2682 repair_bio->bi_opf |= REQ_FAILFAST_DEV;
2683 repair_bio->bi_end_io = failed_bio->bi_end_io;
2684 repair_bio->bi_iter.bi_sector = failrec->logical >> 9;
2685 repair_bio->bi_private = failed_bio->bi_private;
2687 if (failed_io_bio->csum) {
2688 const u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
2690 repair_io_bio->csum = repair_io_bio->csum_inline;
2691 memcpy(repair_io_bio->csum,
2692 failed_io_bio->csum + csum_size * icsum, csum_size);
2695 bio_add_page(repair_bio, page, failrec->len, pgoff);
2696 repair_io_bio->logical = failrec->start;
2697 repair_io_bio->iter = repair_bio->bi_iter;
2699 btrfs_debug(btrfs_sb(inode->i_sb),
2700 "repair read error: submitting new read to mirror %d, in_validation=%d",
2701 failrec->this_mirror, failrec->in_validation);
2703 status = submit_bio_hook(inode, repair_bio, failrec->this_mirror,
2704 failrec->bio_flags);
2706 free_io_failure(failure_tree, tree, failrec);
2707 bio_put(repair_bio);
2712 /* lots and lots of room for performance fixes in the end_bio funcs */
2714 void end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2716 int uptodate = (err == 0);
2719 btrfs_writepage_endio_finish_ordered(page, start, end, uptodate);
2722 ClearPageUptodate(page);
2724 ret = err < 0 ? err : -EIO;
2725 mapping_set_error(page->mapping, ret);
2730 * after a writepage IO is done, we need to:
2731 * clear the uptodate bits on error
2732 * clear the writeback bits in the extent tree for this IO
2733 * end_page_writeback if the page has no more pending IO
2735 * Scheduling is not allowed, so the extent state tree is expected
2736 * to have one and only one object corresponding to this IO.
2738 static void end_bio_extent_writepage(struct bio *bio)
2740 int error = blk_status_to_errno(bio->bi_status);
2741 struct bio_vec *bvec;
2744 struct bvec_iter_all iter_all;
2746 ASSERT(!bio_flagged(bio, BIO_CLONED));
2747 bio_for_each_segment_all(bvec, bio, iter_all) {
2748 struct page *page = bvec->bv_page;
2749 struct inode *inode = page->mapping->host;
2750 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2752 /* We always issue full-page reads, but if some block
2753 * in a page fails to read, blk_update_request() will
2754 * advance bv_offset and adjust bv_len to compensate.
2755 * Print a warning for nonzero offsets, and an error
2756 * if they don't add up to a full page. */
2757 if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) {
2758 if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE)
2760 "partial page write in btrfs with offset %u and length %u",
2761 bvec->bv_offset, bvec->bv_len);
2764 "incomplete page write in btrfs with offset %u and length %u",
2765 bvec->bv_offset, bvec->bv_len);
2768 start = page_offset(page);
2769 end = start + bvec->bv_offset + bvec->bv_len - 1;
2771 end_extent_writepage(page, error, start, end);
2772 end_page_writeback(page);
2779 endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len,
2782 struct extent_state *cached = NULL;
2783 u64 end = start + len - 1;
2785 if (uptodate && tree->track_uptodate)
2786 set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC);
2787 unlock_extent_cached_atomic(tree, start, end, &cached);
2791 * after a readpage IO is done, we need to:
2792 * clear the uptodate bits on error
2793 * set the uptodate bits if things worked
2794 * set the page up to date if all extents in the tree are uptodate
2795 * clear the lock bit in the extent tree
2796 * unlock the page if there are no other extents locked for it
2798 * Scheduling is not allowed, so the extent state tree is expected
2799 * to have one and only one object corresponding to this IO.
2801 static void end_bio_extent_readpage(struct bio *bio)
2803 struct bio_vec *bvec;
2804 int uptodate = !bio->bi_status;
2805 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
2806 struct extent_io_tree *tree, *failure_tree;
2811 u64 extent_start = 0;
2815 struct bvec_iter_all iter_all;
2817 ASSERT(!bio_flagged(bio, BIO_CLONED));
2818 bio_for_each_segment_all(bvec, bio, iter_all) {
2819 struct page *page = bvec->bv_page;
2820 struct inode *inode = page->mapping->host;
2821 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2823 btrfs_debug(fs_info,
2824 "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u",
2825 (u64)bio->bi_iter.bi_sector, bio->bi_status,
2826 io_bio->mirror_num);
2827 tree = &BTRFS_I(inode)->io_tree;
2828 failure_tree = &BTRFS_I(inode)->io_failure_tree;
2830 /* We always issue full-page reads, but if some block
2831 * in a page fails to read, blk_update_request() will
2832 * advance bv_offset and adjust bv_len to compensate.
2833 * Print a warning for nonzero offsets, and an error
2834 * if they don't add up to a full page. */
2835 if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) {
2836 if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE)
2838 "partial page read in btrfs with offset %u and length %u",
2839 bvec->bv_offset, bvec->bv_len);
2842 "incomplete page read in btrfs with offset %u and length %u",
2843 bvec->bv_offset, bvec->bv_len);
2846 start = page_offset(page);
2847 end = start + bvec->bv_offset + bvec->bv_len - 1;
2850 mirror = io_bio->mirror_num;
2851 if (likely(uptodate)) {
2852 if (is_data_inode(inode))
2853 ret = btrfs_verify_data_csum(io_bio, offset, page,
2854 start, end, mirror);
2856 ret = btrfs_validate_metadata_buffer(io_bio,
2857 offset, page, start, end, mirror);
2861 clean_io_failure(BTRFS_I(inode)->root->fs_info,
2862 failure_tree, tree, start,
2864 btrfs_ino(BTRFS_I(inode)), 0);
2867 if (likely(uptodate))
2870 if (is_data_inode(inode)) {
2873 * The generic bio_readpage_error handles errors the
2874 * following way: If possible, new read requests are
2875 * created and submitted and will end up in
2876 * end_bio_extent_readpage as well (if we're lucky,
2877 * not in the !uptodate case). In that case it returns
2878 * 0 and we just go on with the next page in our bio.
2879 * If it can't handle the error it will return -EIO and
2880 * we remain responsible for that page.
2882 if (!btrfs_submit_read_repair(inode, bio, offset, page,
2883 start - page_offset(page),
2885 btrfs_submit_data_bio)) {
2886 uptodate = !bio->bi_status;
2891 struct extent_buffer *eb;
2893 eb = (struct extent_buffer *)page->private;
2894 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
2895 eb->read_mirror = mirror;
2896 atomic_dec(&eb->io_pages);
2897 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD,
2899 btree_readahead_hook(eb, -EIO);
2902 if (likely(uptodate)) {
2903 loff_t i_size = i_size_read(inode);
2904 pgoff_t end_index = i_size >> PAGE_SHIFT;
2907 /* Zero out the end if this page straddles i_size */
2908 off = offset_in_page(i_size);
2909 if (page->index == end_index && off)
2910 zero_user_segment(page, off, PAGE_SIZE);
2911 SetPageUptodate(page);
2913 ClearPageUptodate(page);
2919 if (unlikely(!uptodate)) {
2921 endio_readpage_release_extent(tree,
2927 endio_readpage_release_extent(tree, start,
2928 end - start + 1, 0);
2929 } else if (!extent_len) {
2930 extent_start = start;
2931 extent_len = end + 1 - start;
2932 } else if (extent_start + extent_len == start) {
2933 extent_len += end + 1 - start;
2935 endio_readpage_release_extent(tree, extent_start,
2936 extent_len, uptodate);
2937 extent_start = start;
2938 extent_len = end + 1 - start;
2943 endio_readpage_release_extent(tree, extent_start, extent_len,
2945 btrfs_io_bio_free_csum(io_bio);
2950 * Initialize the members up to but not including 'bio'. Use after allocating a
2951 * new bio by bio_alloc_bioset as it does not initialize the bytes outside of
2952 * 'bio' because use of __GFP_ZERO is not supported.
2954 static inline void btrfs_io_bio_init(struct btrfs_io_bio *btrfs_bio)
2956 memset(btrfs_bio, 0, offsetof(struct btrfs_io_bio, bio));
2960 * The following helpers allocate a bio. As it's backed by a bioset, it'll
2961 * never fail. We're returning a bio right now but you can call btrfs_io_bio
2962 * for the appropriate container_of magic
2964 struct bio *btrfs_bio_alloc(u64 first_byte)
2968 bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, &btrfs_bioset);
2969 bio->bi_iter.bi_sector = first_byte >> 9;
2970 btrfs_io_bio_init(btrfs_io_bio(bio));
2974 struct bio *btrfs_bio_clone(struct bio *bio)
2976 struct btrfs_io_bio *btrfs_bio;
2979 /* Bio allocation backed by a bioset does not fail */
2980 new = bio_clone_fast(bio, GFP_NOFS, &btrfs_bioset);
2981 btrfs_bio = btrfs_io_bio(new);
2982 btrfs_io_bio_init(btrfs_bio);
2983 btrfs_bio->iter = bio->bi_iter;
2987 struct bio *btrfs_io_bio_alloc(unsigned int nr_iovecs)
2991 /* Bio allocation backed by a bioset does not fail */
2992 bio = bio_alloc_bioset(GFP_NOFS, nr_iovecs, &btrfs_bioset);
2993 btrfs_io_bio_init(btrfs_io_bio(bio));
2997 struct bio *btrfs_bio_clone_partial(struct bio *orig, int offset, int size)
3000 struct btrfs_io_bio *btrfs_bio;
3002 /* this will never fail when it's backed by a bioset */
3003 bio = bio_clone_fast(orig, GFP_NOFS, &btrfs_bioset);
3006 btrfs_bio = btrfs_io_bio(bio);
3007 btrfs_io_bio_init(btrfs_bio);
3009 bio_trim(bio, offset >> 9, size >> 9);
3010 btrfs_bio->iter = bio->bi_iter;
3015 * @opf: bio REQ_OP_* and REQ_* flags as one value
3016 * @wbc: optional writeback control for io accounting
3017 * @page: page to add to the bio
3018 * @pg_offset: offset of the new bio or to check whether we are adding
3019 * a contiguous page to the previous one
3020 * @size: portion of page that we want to write
3021 * @offset: starting offset in the page
3022 * @bio_ret: must be valid pointer, newly allocated bio will be stored there
3023 * @end_io_func: end_io callback for new bio
3024 * @mirror_num: desired mirror to read/write
3025 * @prev_bio_flags: flags of previous bio to see if we can merge the current one
3026 * @bio_flags: flags of the current bio to see if we can merge them
3028 static int submit_extent_page(unsigned int opf,
3029 struct writeback_control *wbc,
3030 struct page *page, u64 offset,
3031 size_t size, unsigned long pg_offset,
3032 struct bio **bio_ret,
3033 bio_end_io_t end_io_func,
3035 unsigned long prev_bio_flags,
3036 unsigned long bio_flags,
3037 bool force_bio_submit)
3041 size_t page_size = min_t(size_t, size, PAGE_SIZE);
3042 sector_t sector = offset >> 9;
3043 struct extent_io_tree *tree = &BTRFS_I(page->mapping->host)->io_tree;
3049 bool can_merge = true;
3052 if (prev_bio_flags & EXTENT_BIO_COMPRESSED)
3053 contig = bio->bi_iter.bi_sector == sector;
3055 contig = bio_end_sector(bio) == sector;
3057 if (btrfs_bio_fits_in_stripe(page, page_size, bio, bio_flags))
3060 if (prev_bio_flags != bio_flags || !contig || !can_merge ||
3062 bio_add_page(bio, page, page_size, pg_offset) < page_size) {
3063 ret = submit_one_bio(bio, mirror_num, prev_bio_flags);
3071 wbc_account_cgroup_owner(wbc, page, page_size);
3076 bio = btrfs_bio_alloc(offset);
3077 bio_add_page(bio, page, page_size, pg_offset);
3078 bio->bi_end_io = end_io_func;
3079 bio->bi_private = tree;
3080 bio->bi_write_hint = page->mapping->host->i_write_hint;
3083 struct block_device *bdev;
3085 bdev = BTRFS_I(page->mapping->host)->root->fs_info->fs_devices->latest_bdev;
3086 bio_set_dev(bio, bdev);
3087 wbc_init_bio(wbc, bio);
3088 wbc_account_cgroup_owner(wbc, page, page_size);
3096 static void attach_extent_buffer_page(struct extent_buffer *eb,
3099 if (!PagePrivate(page))
3100 attach_page_private(page, eb);
3102 WARN_ON(page->private != (unsigned long)eb);
3105 void set_page_extent_mapped(struct page *page)
3107 if (!PagePrivate(page))
3108 attach_page_private(page, (void *)EXTENT_PAGE_PRIVATE);
3111 static struct extent_map *
3112 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
3113 u64 start, u64 len, struct extent_map **em_cached)
3115 struct extent_map *em;
3117 if (em_cached && *em_cached) {
3119 if (extent_map_in_tree(em) && start >= em->start &&
3120 start < extent_map_end(em)) {
3121 refcount_inc(&em->refs);
3125 free_extent_map(em);
3129 em = btrfs_get_extent(BTRFS_I(inode), page, pg_offset, start, len);
3130 if (em_cached && !IS_ERR_OR_NULL(em)) {
3132 refcount_inc(&em->refs);
3138 * basic readpage implementation. Locked extent state structs are inserted
3139 * into the tree that are removed when the IO is done (by the end_io
3141 * XXX JDM: This needs looking at to ensure proper page locking
3142 * return 0 on success, otherwise return error
3144 int btrfs_do_readpage(struct page *page, struct extent_map **em_cached,
3145 struct bio **bio, unsigned long *bio_flags,
3146 unsigned int read_flags, u64 *prev_em_start)
3148 struct inode *inode = page->mapping->host;
3149 u64 start = page_offset(page);
3150 const u64 end = start + PAGE_SIZE - 1;
3153 u64 last_byte = i_size_read(inode);
3156 struct extent_map *em;
3159 size_t pg_offset = 0;
3161 size_t disk_io_size;
3162 size_t blocksize = inode->i_sb->s_blocksize;
3163 unsigned long this_bio_flag = 0;
3164 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
3166 set_page_extent_mapped(page);
3168 if (!PageUptodate(page)) {
3169 if (cleancache_get_page(page) == 0) {
3170 BUG_ON(blocksize != PAGE_SIZE);
3171 unlock_extent(tree, start, end);
3176 if (page->index == last_byte >> PAGE_SHIFT) {
3178 size_t zero_offset = offset_in_page(last_byte);
3181 iosize = PAGE_SIZE - zero_offset;
3182 userpage = kmap_atomic(page);
3183 memset(userpage + zero_offset, 0, iosize);
3184 flush_dcache_page(page);
3185 kunmap_atomic(userpage);
3188 while (cur <= end) {
3189 bool force_bio_submit = false;
3192 if (cur >= last_byte) {
3194 struct extent_state *cached = NULL;
3196 iosize = PAGE_SIZE - pg_offset;
3197 userpage = kmap_atomic(page);
3198 memset(userpage + pg_offset, 0, iosize);
3199 flush_dcache_page(page);
3200 kunmap_atomic(userpage);
3201 set_extent_uptodate(tree, cur, cur + iosize - 1,
3203 unlock_extent_cached(tree, cur,
3204 cur + iosize - 1, &cached);
3207 em = __get_extent_map(inode, page, pg_offset, cur,
3208 end - cur + 1, em_cached);
3209 if (IS_ERR_OR_NULL(em)) {
3211 unlock_extent(tree, cur, end);
3214 extent_offset = cur - em->start;
3215 BUG_ON(extent_map_end(em) <= cur);
3218 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
3219 this_bio_flag |= EXTENT_BIO_COMPRESSED;
3220 extent_set_compress_type(&this_bio_flag,
3224 iosize = min(extent_map_end(em) - cur, end - cur + 1);
3225 cur_end = min(extent_map_end(em) - 1, end);
3226 iosize = ALIGN(iosize, blocksize);
3227 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
3228 disk_io_size = em->block_len;
3229 offset = em->block_start;
3231 offset = em->block_start + extent_offset;
3232 disk_io_size = iosize;
3234 block_start = em->block_start;
3235 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
3236 block_start = EXTENT_MAP_HOLE;
3239 * If we have a file range that points to a compressed extent
3240 * and it's followed by a consecutive file range that points
3241 * to the same compressed extent (possibly with a different
3242 * offset and/or length, so it either points to the whole extent
3243 * or only part of it), we must make sure we do not submit a
3244 * single bio to populate the pages for the 2 ranges because
3245 * this makes the compressed extent read zero out the pages
3246 * belonging to the 2nd range. Imagine the following scenario:
3249 * [0 - 8K] [8K - 24K]
3252 * points to extent X, points to extent X,
3253 * offset 4K, length of 8K offset 0, length 16K
3255 * [extent X, compressed length = 4K uncompressed length = 16K]
3257 * If the bio to read the compressed extent covers both ranges,
3258 * it will decompress extent X into the pages belonging to the
3259 * first range and then it will stop, zeroing out the remaining
3260 * pages that belong to the other range that points to extent X.
3261 * So here we make sure we submit 2 bios, one for the first
3262 * range and another one for the third range. Both will target
3263 * the same physical extent from disk, but we can't currently
3264 * make the compressed bio endio callback populate the pages
3265 * for both ranges because each compressed bio is tightly
3266 * coupled with a single extent map, and each range can have
3267 * an extent map with a different offset value relative to the
3268 * uncompressed data of our extent and different lengths. This
3269 * is a corner case so we prioritize correctness over
3270 * non-optimal behavior (submitting 2 bios for the same extent).
3272 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) &&
3273 prev_em_start && *prev_em_start != (u64)-1 &&
3274 *prev_em_start != em->start)
3275 force_bio_submit = true;
3278 *prev_em_start = em->start;
3280 free_extent_map(em);
3283 /* we've found a hole, just zero and go on */
3284 if (block_start == EXTENT_MAP_HOLE) {
3286 struct extent_state *cached = NULL;
3288 userpage = kmap_atomic(page);
3289 memset(userpage + pg_offset, 0, iosize);
3290 flush_dcache_page(page);
3291 kunmap_atomic(userpage);
3293 set_extent_uptodate(tree, cur, cur + iosize - 1,
3295 unlock_extent_cached(tree, cur,
3296 cur + iosize - 1, &cached);
3298 pg_offset += iosize;
3301 /* the get_extent function already copied into the page */
3302 if (test_range_bit(tree, cur, cur_end,
3303 EXTENT_UPTODATE, 1, NULL)) {
3304 check_page_uptodate(tree, page);
3305 unlock_extent(tree, cur, cur + iosize - 1);
3307 pg_offset += iosize;
3310 /* we have an inline extent but it didn't get marked up
3311 * to date. Error out
3313 if (block_start == EXTENT_MAP_INLINE) {
3315 unlock_extent(tree, cur, cur + iosize - 1);
3317 pg_offset += iosize;
3321 ret = submit_extent_page(REQ_OP_READ | read_flags, NULL,
3322 page, offset, disk_io_size,
3324 end_bio_extent_readpage, 0,
3330 *bio_flags = this_bio_flag;
3333 unlock_extent(tree, cur, cur + iosize - 1);
3337 pg_offset += iosize;
3341 if (!PageError(page))
3342 SetPageUptodate(page);
3348 static inline void contiguous_readpages(struct page *pages[], int nr_pages,
3350 struct extent_map **em_cached,
3352 unsigned long *bio_flags,
3355 struct btrfs_inode *inode = BTRFS_I(pages[0]->mapping->host);
3358 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
3360 for (index = 0; index < nr_pages; index++) {
3361 btrfs_do_readpage(pages[index], em_cached, bio, bio_flags,
3362 REQ_RAHEAD, prev_em_start);
3363 put_page(pages[index]);
3367 static void update_nr_written(struct writeback_control *wbc,
3368 unsigned long nr_written)
3370 wbc->nr_to_write -= nr_written;
3374 * helper for __extent_writepage, doing all of the delayed allocation setup.
3376 * This returns 1 if btrfs_run_delalloc_range function did all the work required
3377 * to write the page (copy into inline extent). In this case the IO has
3378 * been started and the page is already unlocked.
3380 * This returns 0 if all went well (page still locked)
3381 * This returns < 0 if there were errors (page still locked)
3383 static noinline_for_stack int writepage_delalloc(struct btrfs_inode *inode,
3384 struct page *page, struct writeback_control *wbc,
3385 u64 delalloc_start, unsigned long *nr_written)
3387 u64 page_end = delalloc_start + PAGE_SIZE - 1;
3389 u64 delalloc_to_write = 0;
3390 u64 delalloc_end = 0;
3392 int page_started = 0;
3395 while (delalloc_end < page_end) {
3396 found = find_lock_delalloc_range(&inode->vfs_inode, page,
3400 delalloc_start = delalloc_end + 1;
3403 ret = btrfs_run_delalloc_range(inode, page, delalloc_start,
3404 delalloc_end, &page_started, nr_written, wbc);
3408 * btrfs_run_delalloc_range should return < 0 for error
3409 * but just in case, we use > 0 here meaning the IO is
3410 * started, so we don't want to return > 0 unless
3411 * things are going well.
3413 return ret < 0 ? ret : -EIO;
3416 * delalloc_end is already one less than the total length, so
3417 * we don't subtract one from PAGE_SIZE
3419 delalloc_to_write += (delalloc_end - delalloc_start +
3420 PAGE_SIZE) >> PAGE_SHIFT;
3421 delalloc_start = delalloc_end + 1;
3423 if (wbc->nr_to_write < delalloc_to_write) {
3426 if (delalloc_to_write < thresh * 2)
3427 thresh = delalloc_to_write;
3428 wbc->nr_to_write = min_t(u64, delalloc_to_write,
3432 /* did the fill delalloc function already unlock and start
3437 * we've unlocked the page, so we can't update
3438 * the mapping's writeback index, just update
3441 wbc->nr_to_write -= *nr_written;
3449 * helper for __extent_writepage. This calls the writepage start hooks,
3450 * and does the loop to map the page into extents and bios.
3452 * We return 1 if the IO is started and the page is unlocked,
3453 * 0 if all went well (page still locked)
3454 * < 0 if there were errors (page still locked)
3456 static noinline_for_stack int __extent_writepage_io(struct btrfs_inode *inode,
3458 struct writeback_control *wbc,
3459 struct extent_page_data *epd,
3461 unsigned long nr_written,
3464 struct extent_io_tree *tree = &inode->io_tree;
3465 u64 start = page_offset(page);
3466 u64 page_end = start + PAGE_SIZE - 1;
3472 struct extent_map *em;
3473 size_t pg_offset = 0;
3477 const unsigned int write_flags = wbc_to_write_flags(wbc);
3480 ret = btrfs_writepage_cow_fixup(page, start, page_end);
3482 /* Fixup worker will requeue */
3483 redirty_page_for_writepage(wbc, page);
3484 update_nr_written(wbc, nr_written);
3490 * we don't want to touch the inode after unlocking the page,
3491 * so we update the mapping writeback index now
3493 update_nr_written(wbc, nr_written + 1);
3496 blocksize = inode->vfs_inode.i_sb->s_blocksize;
3498 while (cur <= end) {
3502 if (cur >= i_size) {
3503 btrfs_writepage_endio_finish_ordered(page, cur,
3507 em = btrfs_get_extent(inode, NULL, 0, cur, end - cur + 1);
3508 if (IS_ERR_OR_NULL(em)) {
3510 ret = PTR_ERR_OR_ZERO(em);
3514 extent_offset = cur - em->start;
3515 em_end = extent_map_end(em);
3516 BUG_ON(em_end <= cur);
3518 iosize = min(em_end - cur, end - cur + 1);
3519 iosize = ALIGN(iosize, blocksize);
3520 offset = em->block_start + extent_offset;
3521 block_start = em->block_start;
3522 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3523 free_extent_map(em);
3527 * compressed and inline extents are written through other
3530 if (compressed || block_start == EXTENT_MAP_HOLE ||
3531 block_start == EXTENT_MAP_INLINE) {
3535 btrfs_writepage_endio_finish_ordered(page, cur,
3536 cur + iosize - 1, 1);
3538 pg_offset += iosize;
3542 btrfs_set_range_writeback(tree, cur, cur + iosize - 1);
3543 if (!PageWriteback(page)) {
3544 btrfs_err(inode->root->fs_info,
3545 "page %lu not writeback, cur %llu end %llu",
3546 page->index, cur, end);
3549 ret = submit_extent_page(REQ_OP_WRITE | write_flags, wbc,
3550 page, offset, iosize, pg_offset,
3552 end_bio_extent_writepage,
3556 if (PageWriteback(page))
3557 end_page_writeback(page);
3561 pg_offset += iosize;
3569 * the writepage semantics are similar to regular writepage. extent
3570 * records are inserted to lock ranges in the tree, and as dirty areas
3571 * are found, they are marked writeback. Then the lock bits are removed
3572 * and the end_io handler clears the writeback ranges
3574 * Return 0 if everything goes well.
3575 * Return <0 for error.
3577 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
3578 struct extent_page_data *epd)
3580 struct inode *inode = page->mapping->host;
3581 u64 start = page_offset(page);
3582 u64 page_end = start + PAGE_SIZE - 1;
3586 loff_t i_size = i_size_read(inode);
3587 unsigned long end_index = i_size >> PAGE_SHIFT;
3588 unsigned long nr_written = 0;
3590 trace___extent_writepage(page, inode, wbc);
3592 WARN_ON(!PageLocked(page));
3594 ClearPageError(page);
3596 pg_offset = offset_in_page(i_size);
3597 if (page->index > end_index ||
3598 (page->index == end_index && !pg_offset)) {
3599 page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
3604 if (page->index == end_index) {
3607 userpage = kmap_atomic(page);
3608 memset(userpage + pg_offset, 0,
3609 PAGE_SIZE - pg_offset);
3610 kunmap_atomic(userpage);
3611 flush_dcache_page(page);
3614 set_page_extent_mapped(page);
3616 if (!epd->extent_locked) {
3617 ret = writepage_delalloc(BTRFS_I(inode), page, wbc, start,
3625 ret = __extent_writepage_io(BTRFS_I(inode), page, wbc, epd, i_size,
3632 /* make sure the mapping tag for page dirty gets cleared */
3633 set_page_writeback(page);
3634 end_page_writeback(page);
3636 if (PageError(page)) {
3637 ret = ret < 0 ? ret : -EIO;
3638 end_extent_writepage(page, ret, start, page_end);
3645 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3647 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
3648 TASK_UNINTERRUPTIBLE);
3651 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3653 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3654 smp_mb__after_atomic();
3655 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3659 * Lock eb pages and flush the bio if we can't the locks
3661 * Return 0 if nothing went wrong
3662 * Return >0 is same as 0, except bio is not submitted
3663 * Return <0 if something went wrong, no page is locked
3665 static noinline_for_stack int lock_extent_buffer_for_io(struct extent_buffer *eb,
3666 struct extent_page_data *epd)
3668 struct btrfs_fs_info *fs_info = eb->fs_info;
3669 int i, num_pages, failed_page_nr;
3673 if (!btrfs_try_tree_write_lock(eb)) {
3674 ret = flush_write_bio(epd);
3678 btrfs_tree_lock(eb);
3681 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3682 btrfs_tree_unlock(eb);
3686 ret = flush_write_bio(epd);
3692 wait_on_extent_buffer_writeback(eb);
3693 btrfs_tree_lock(eb);
3694 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3696 btrfs_tree_unlock(eb);
3701 * We need to do this to prevent races in people who check if the eb is
3702 * under IO since we can end up having no IO bits set for a short period
3705 spin_lock(&eb->refs_lock);
3706 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3707 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3708 spin_unlock(&eb->refs_lock);
3709 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3710 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
3712 fs_info->dirty_metadata_batch);
3715 spin_unlock(&eb->refs_lock);
3718 btrfs_tree_unlock(eb);
3723 num_pages = num_extent_pages(eb);
3724 for (i = 0; i < num_pages; i++) {
3725 struct page *p = eb->pages[i];
3727 if (!trylock_page(p)) {
3731 err = flush_write_bio(epd);
3745 /* Unlock already locked pages */
3746 for (i = 0; i < failed_page_nr; i++)
3747 unlock_page(eb->pages[i]);
3749 * Clear EXTENT_BUFFER_WRITEBACK and wake up anyone waiting on it.
3750 * Also set back EXTENT_BUFFER_DIRTY so future attempts to this eb can
3751 * be made and undo everything done before.
3753 btrfs_tree_lock(eb);
3754 spin_lock(&eb->refs_lock);
3755 set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
3756 end_extent_buffer_writeback(eb);
3757 spin_unlock(&eb->refs_lock);
3758 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, eb->len,
3759 fs_info->dirty_metadata_batch);
3760 btrfs_clear_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3761 btrfs_tree_unlock(eb);
3765 static void set_btree_ioerr(struct page *page)
3767 struct extent_buffer *eb = (struct extent_buffer *)page->private;
3768 struct btrfs_fs_info *fs_info;
3771 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))
3775 * If we error out, we should add back the dirty_metadata_bytes
3776 * to make it consistent.
3778 fs_info = eb->fs_info;
3779 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
3780 eb->len, fs_info->dirty_metadata_batch);
3783 * If writeback for a btree extent that doesn't belong to a log tree
3784 * failed, increment the counter transaction->eb_write_errors.
3785 * We do this because while the transaction is running and before it's
3786 * committing (when we call filemap_fdata[write|wait]_range against
3787 * the btree inode), we might have
3788 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
3789 * returns an error or an error happens during writeback, when we're
3790 * committing the transaction we wouldn't know about it, since the pages
3791 * can be no longer dirty nor marked anymore for writeback (if a
3792 * subsequent modification to the extent buffer didn't happen before the
3793 * transaction commit), which makes filemap_fdata[write|wait]_range not
3794 * able to find the pages tagged with SetPageError at transaction
3795 * commit time. So if this happens we must abort the transaction,
3796 * otherwise we commit a super block with btree roots that point to
3797 * btree nodes/leafs whose content on disk is invalid - either garbage
3798 * or the content of some node/leaf from a past generation that got
3799 * cowed or deleted and is no longer valid.
3801 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
3802 * not be enough - we need to distinguish between log tree extents vs
3803 * non-log tree extents, and the next filemap_fdatawait_range() call
3804 * will catch and clear such errors in the mapping - and that call might
3805 * be from a log sync and not from a transaction commit. Also, checking
3806 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
3807 * not done and would not be reliable - the eb might have been released
3808 * from memory and reading it back again means that flag would not be
3809 * set (since it's a runtime flag, not persisted on disk).
3811 * Using the flags below in the btree inode also makes us achieve the
3812 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
3813 * writeback for all dirty pages and before filemap_fdatawait_range()
3814 * is called, the writeback for all dirty pages had already finished
3815 * with errors - because we were not using AS_EIO/AS_ENOSPC,
3816 * filemap_fdatawait_range() would return success, as it could not know
3817 * that writeback errors happened (the pages were no longer tagged for
3820 switch (eb->log_index) {
3822 set_bit(BTRFS_FS_BTREE_ERR, &eb->fs_info->flags);
3825 set_bit(BTRFS_FS_LOG1_ERR, &eb->fs_info->flags);
3828 set_bit(BTRFS_FS_LOG2_ERR, &eb->fs_info->flags);
3831 BUG(); /* unexpected, logic error */
3835 static void end_bio_extent_buffer_writepage(struct bio *bio)
3837 struct bio_vec *bvec;
3838 struct extent_buffer *eb;
3840 struct bvec_iter_all iter_all;
3842 ASSERT(!bio_flagged(bio, BIO_CLONED));
3843 bio_for_each_segment_all(bvec, bio, iter_all) {
3844 struct page *page = bvec->bv_page;
3846 eb = (struct extent_buffer *)page->private;
3848 done = atomic_dec_and_test(&eb->io_pages);
3850 if (bio->bi_status ||
3851 test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) {
3852 ClearPageUptodate(page);
3853 set_btree_ioerr(page);
3856 end_page_writeback(page);
3861 end_extent_buffer_writeback(eb);
3867 static noinline_for_stack int write_one_eb(struct extent_buffer *eb,
3868 struct writeback_control *wbc,
3869 struct extent_page_data *epd)
3871 u64 offset = eb->start;
3874 unsigned long start, end;
3875 unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META;
3878 clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
3879 num_pages = num_extent_pages(eb);
3880 atomic_set(&eb->io_pages, num_pages);
3882 /* set btree blocks beyond nritems with 0 to avoid stale content. */
3883 nritems = btrfs_header_nritems(eb);
3884 if (btrfs_header_level(eb) > 0) {
3885 end = btrfs_node_key_ptr_offset(nritems);
3887 memzero_extent_buffer(eb, end, eb->len - end);
3891 * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0
3893 start = btrfs_item_nr_offset(nritems);
3894 end = BTRFS_LEAF_DATA_OFFSET + leaf_data_end(eb);
3895 memzero_extent_buffer(eb, start, end - start);
3898 for (i = 0; i < num_pages; i++) {
3899 struct page *p = eb->pages[i];
3901 clear_page_dirty_for_io(p);
3902 set_page_writeback(p);
3903 ret = submit_extent_page(REQ_OP_WRITE | write_flags, wbc,
3904 p, offset, PAGE_SIZE, 0,
3906 end_bio_extent_buffer_writepage,
3910 if (PageWriteback(p))
3911 end_page_writeback(p);
3912 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3913 end_extent_buffer_writeback(eb);
3917 offset += PAGE_SIZE;
3918 update_nr_written(wbc, 1);
3922 if (unlikely(ret)) {
3923 for (; i < num_pages; i++) {
3924 struct page *p = eb->pages[i];
3925 clear_page_dirty_for_io(p);
3933 int btree_write_cache_pages(struct address_space *mapping,
3934 struct writeback_control *wbc)
3936 struct extent_buffer *eb, *prev_eb = NULL;
3937 struct extent_page_data epd = {
3940 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3942 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
3945 int nr_to_write_done = 0;
3946 struct pagevec pvec;
3949 pgoff_t end; /* Inclusive */
3953 pagevec_init(&pvec);
3954 if (wbc->range_cyclic) {
3955 index = mapping->writeback_index; /* Start from prev offset */
3958 * Start from the beginning does not need to cycle over the
3959 * range, mark it as scanned.
3961 scanned = (index == 0);
3963 index = wbc->range_start >> PAGE_SHIFT;
3964 end = wbc->range_end >> PAGE_SHIFT;
3967 if (wbc->sync_mode == WB_SYNC_ALL)
3968 tag = PAGECACHE_TAG_TOWRITE;
3970 tag = PAGECACHE_TAG_DIRTY;
3972 if (wbc->sync_mode == WB_SYNC_ALL)
3973 tag_pages_for_writeback(mapping, index, end);
3974 while (!done && !nr_to_write_done && (index <= end) &&
3975 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
3979 for (i = 0; i < nr_pages; i++) {
3980 struct page *page = pvec.pages[i];
3982 if (!PagePrivate(page))
3985 spin_lock(&mapping->private_lock);
3986 if (!PagePrivate(page)) {
3987 spin_unlock(&mapping->private_lock);
3991 eb = (struct extent_buffer *)page->private;
3994 * Shouldn't happen and normally this would be a BUG_ON
3995 * but no sense in crashing the users box for something
3996 * we can survive anyway.
3999 spin_unlock(&mapping->private_lock);
4003 if (eb == prev_eb) {
4004 spin_unlock(&mapping->private_lock);
4008 ret = atomic_inc_not_zero(&eb->refs);
4009 spin_unlock(&mapping->private_lock);
4014 ret = lock_extent_buffer_for_io(eb, &epd);
4016 free_extent_buffer(eb);
4018 } else if (ret < 0) {
4020 free_extent_buffer(eb);
4024 ret = write_one_eb(eb, wbc, &epd);
4027 free_extent_buffer(eb);
4030 free_extent_buffer(eb);
4033 * the filesystem may choose to bump up nr_to_write.
4034 * We have to make sure to honor the new nr_to_write
4037 nr_to_write_done = wbc->nr_to_write <= 0;
4039 pagevec_release(&pvec);
4042 if (!scanned && !done) {
4044 * We hit the last page and there is more work to be done: wrap
4045 * back to the start of the file
4053 end_write_bio(&epd, ret);
4057 * If something went wrong, don't allow any metadata write bio to be
4060 * This would prevent use-after-free if we had dirty pages not
4061 * cleaned up, which can still happen by fuzzed images.
4064 * Allowing existing tree block to be allocated for other trees.
4066 * - Log tree operations
4067 * Exiting tree blocks get allocated to log tree, bumps its
4068 * generation, then get cleaned in tree re-balance.
4069 * Such tree block will not be written back, since it's clean,
4070 * thus no WRITTEN flag set.
4071 * And after log writes back, this tree block is not traced by
4072 * any dirty extent_io_tree.
4074 * - Offending tree block gets re-dirtied from its original owner
4075 * Since it has bumped generation, no WRITTEN flag, it can be
4076 * reused without COWing. This tree block will not be traced
4077 * by btrfs_transaction::dirty_pages.
4079 * Now such dirty tree block will not be cleaned by any dirty
4080 * extent io tree. Thus we don't want to submit such wild eb
4081 * if the fs already has error.
4083 if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
4084 ret = flush_write_bio(&epd);
4087 end_write_bio(&epd, ret);
4093 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
4094 * @mapping: address space structure to write
4095 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
4096 * @data: data passed to __extent_writepage function
4098 * If a page is already under I/O, write_cache_pages() skips it, even
4099 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
4100 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
4101 * and msync() need to guarantee that all the data which was dirty at the time
4102 * the call was made get new I/O started against them. If wbc->sync_mode is
4103 * WB_SYNC_ALL then we were called for data integrity and we must wait for
4104 * existing IO to complete.
4106 static int extent_write_cache_pages(struct address_space *mapping,
4107 struct writeback_control *wbc,
4108 struct extent_page_data *epd)
4110 struct inode *inode = mapping->host;
4113 int nr_to_write_done = 0;
4114 struct pagevec pvec;
4117 pgoff_t end; /* Inclusive */
4119 int range_whole = 0;
4124 * We have to hold onto the inode so that ordered extents can do their
4125 * work when the IO finishes. The alternative to this is failing to add
4126 * an ordered extent if the igrab() fails there and that is a huge pain
4127 * to deal with, so instead just hold onto the inode throughout the
4128 * writepages operation. If it fails here we are freeing up the inode
4129 * anyway and we'd rather not waste our time writing out stuff that is
4130 * going to be truncated anyway.
4135 pagevec_init(&pvec);
4136 if (wbc->range_cyclic) {
4137 index = mapping->writeback_index; /* Start from prev offset */
4140 * Start from the beginning does not need to cycle over the
4141 * range, mark it as scanned.
4143 scanned = (index == 0);
4145 index = wbc->range_start >> PAGE_SHIFT;
4146 end = wbc->range_end >> PAGE_SHIFT;
4147 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
4153 * We do the tagged writepage as long as the snapshot flush bit is set
4154 * and we are the first one who do the filemap_flush() on this inode.
4156 * The nr_to_write == LONG_MAX is needed to make sure other flushers do
4157 * not race in and drop the bit.
4159 if (range_whole && wbc->nr_to_write == LONG_MAX &&
4160 test_and_clear_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
4161 &BTRFS_I(inode)->runtime_flags))
4162 wbc->tagged_writepages = 1;
4164 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
4165 tag = PAGECACHE_TAG_TOWRITE;
4167 tag = PAGECACHE_TAG_DIRTY;
4169 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
4170 tag_pages_for_writeback(mapping, index, end);
4172 while (!done && !nr_to_write_done && (index <= end) &&
4173 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping,
4174 &index, end, tag))) {
4177 for (i = 0; i < nr_pages; i++) {
4178 struct page *page = pvec.pages[i];
4180 done_index = page->index + 1;
4182 * At this point we hold neither the i_pages lock nor
4183 * the page lock: the page may be truncated or
4184 * invalidated (changing page->mapping to NULL),
4185 * or even swizzled back from swapper_space to
4186 * tmpfs file mapping
4188 if (!trylock_page(page)) {
4189 ret = flush_write_bio(epd);
4194 if (unlikely(page->mapping != mapping)) {
4199 if (wbc->sync_mode != WB_SYNC_NONE) {
4200 if (PageWriteback(page)) {
4201 ret = flush_write_bio(epd);
4204 wait_on_page_writeback(page);
4207 if (PageWriteback(page) ||
4208 !clear_page_dirty_for_io(page)) {
4213 ret = __extent_writepage(page, wbc, epd);
4220 * the filesystem may choose to bump up nr_to_write.
4221 * We have to make sure to honor the new nr_to_write
4224 nr_to_write_done = wbc->nr_to_write <= 0;
4226 pagevec_release(&pvec);
4229 if (!scanned && !done) {
4231 * We hit the last page and there is more work to be done: wrap
4232 * back to the start of the file
4238 * If we're looping we could run into a page that is locked by a
4239 * writer and that writer could be waiting on writeback for a
4240 * page in our current bio, and thus deadlock, so flush the
4243 ret = flush_write_bio(epd);
4248 if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole))
4249 mapping->writeback_index = done_index;
4251 btrfs_add_delayed_iput(inode);
4255 int extent_write_full_page(struct page *page, struct writeback_control *wbc)
4258 struct extent_page_data epd = {
4261 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4264 ret = __extent_writepage(page, wbc, &epd);
4267 end_write_bio(&epd, ret);
4271 ret = flush_write_bio(&epd);
4276 int extent_write_locked_range(struct inode *inode, u64 start, u64 end,
4280 struct address_space *mapping = inode->i_mapping;
4282 unsigned long nr_pages = (end - start + PAGE_SIZE) >>
4285 struct extent_page_data epd = {
4288 .sync_io = mode == WB_SYNC_ALL,
4290 struct writeback_control wbc_writepages = {
4292 .nr_to_write = nr_pages * 2,
4293 .range_start = start,
4294 .range_end = end + 1,
4295 /* We're called from an async helper function */
4296 .punt_to_cgroup = 1,
4297 .no_cgroup_owner = 1,
4300 wbc_attach_fdatawrite_inode(&wbc_writepages, inode);
4301 while (start <= end) {
4302 page = find_get_page(mapping, start >> PAGE_SHIFT);
4303 if (clear_page_dirty_for_io(page))
4304 ret = __extent_writepage(page, &wbc_writepages, &epd);
4306 btrfs_writepage_endio_finish_ordered(page, start,
4307 start + PAGE_SIZE - 1, 1);
4316 ret = flush_write_bio(&epd);
4318 end_write_bio(&epd, ret);
4320 wbc_detach_inode(&wbc_writepages);
4324 int extent_writepages(struct address_space *mapping,
4325 struct writeback_control *wbc)
4328 struct extent_page_data epd = {
4331 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4334 ret = extent_write_cache_pages(mapping, wbc, &epd);
4337 end_write_bio(&epd, ret);
4340 ret = flush_write_bio(&epd);
4344 void extent_readahead(struct readahead_control *rac)
4346 struct bio *bio = NULL;
4347 unsigned long bio_flags = 0;
4348 struct page *pagepool[16];
4349 struct extent_map *em_cached = NULL;
4350 u64 prev_em_start = (u64)-1;
4353 while ((nr = readahead_page_batch(rac, pagepool))) {
4354 u64 contig_start = page_offset(pagepool[0]);
4355 u64 contig_end = page_offset(pagepool[nr - 1]) + PAGE_SIZE - 1;
4357 ASSERT(contig_start + nr * PAGE_SIZE - 1 == contig_end);
4359 contiguous_readpages(pagepool, nr, contig_start, contig_end,
4360 &em_cached, &bio, &bio_flags, &prev_em_start);
4364 free_extent_map(em_cached);
4367 if (submit_one_bio(bio, 0, bio_flags))
4373 * basic invalidatepage code, this waits on any locked or writeback
4374 * ranges corresponding to the page, and then deletes any extent state
4375 * records from the tree
4377 int extent_invalidatepage(struct extent_io_tree *tree,
4378 struct page *page, unsigned long offset)
4380 struct extent_state *cached_state = NULL;
4381 u64 start = page_offset(page);
4382 u64 end = start + PAGE_SIZE - 1;
4383 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
4385 start += ALIGN(offset, blocksize);
4389 lock_extent_bits(tree, start, end, &cached_state);
4390 wait_on_page_writeback(page);
4391 clear_extent_bit(tree, start, end, EXTENT_LOCKED | EXTENT_DELALLOC |
4392 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state);
4397 * a helper for releasepage, this tests for areas of the page that
4398 * are locked or under IO and drops the related state bits if it is safe
4401 static int try_release_extent_state(struct extent_io_tree *tree,
4402 struct page *page, gfp_t mask)
4404 u64 start = page_offset(page);
4405 u64 end = start + PAGE_SIZE - 1;
4408 if (test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL)) {
4412 * at this point we can safely clear everything except the
4413 * locked bit and the nodatasum bit
4415 ret = __clear_extent_bit(tree, start, end,
4416 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
4417 0, 0, NULL, mask, NULL);
4419 /* if clear_extent_bit failed for enomem reasons,
4420 * we can't allow the release to continue.
4431 * a helper for releasepage. As long as there are no locked extents
4432 * in the range corresponding to the page, both state records and extent
4433 * map records are removed
4435 int try_release_extent_mapping(struct page *page, gfp_t mask)
4437 struct extent_map *em;
4438 u64 start = page_offset(page);
4439 u64 end = start + PAGE_SIZE - 1;
4440 struct btrfs_inode *btrfs_inode = BTRFS_I(page->mapping->host);
4441 struct extent_io_tree *tree = &btrfs_inode->io_tree;
4442 struct extent_map_tree *map = &btrfs_inode->extent_tree;
4444 if (gfpflags_allow_blocking(mask) &&
4445 page->mapping->host->i_size > SZ_16M) {
4447 while (start <= end) {
4448 struct btrfs_fs_info *fs_info;
4451 len = end - start + 1;
4452 write_lock(&map->lock);
4453 em = lookup_extent_mapping(map, start, len);
4455 write_unlock(&map->lock);
4458 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
4459 em->start != start) {
4460 write_unlock(&map->lock);
4461 free_extent_map(em);
4464 if (test_range_bit(tree, em->start,
4465 extent_map_end(em) - 1,
4466 EXTENT_LOCKED, 0, NULL))
4469 * If it's not in the list of modified extents, used
4470 * by a fast fsync, we can remove it. If it's being
4471 * logged we can safely remove it since fsync took an
4472 * extra reference on the em.
4474 if (list_empty(&em->list) ||
4475 test_bit(EXTENT_FLAG_LOGGING, &em->flags))
4478 * If it's in the list of modified extents, remove it
4479 * only if its generation is older then the current one,
4480 * in which case we don't need it for a fast fsync.
4481 * Otherwise don't remove it, we could be racing with an
4482 * ongoing fast fsync that could miss the new extent.
4484 fs_info = btrfs_inode->root->fs_info;
4485 spin_lock(&fs_info->trans_lock);
4486 cur_gen = fs_info->generation;
4487 spin_unlock(&fs_info->trans_lock);
4488 if (em->generation >= cur_gen)
4492 * We only remove extent maps that are not in the list of
4493 * modified extents or that are in the list but with a
4494 * generation lower then the current generation, so there
4495 * is no need to set the full fsync flag on the inode (it
4496 * hurts the fsync performance for workloads with a data
4497 * size that exceeds or is close to the system's memory).
4499 remove_extent_mapping(map, em);
4500 /* once for the rb tree */
4501 free_extent_map(em);
4503 start = extent_map_end(em);
4504 write_unlock(&map->lock);
4507 free_extent_map(em);
4509 cond_resched(); /* Allow large-extent preemption. */
4512 return try_release_extent_state(tree, page, mask);
4516 * helper function for fiemap, which doesn't want to see any holes.
4517 * This maps until we find something past 'last'
4519 static struct extent_map *get_extent_skip_holes(struct btrfs_inode *inode,
4520 u64 offset, u64 last)
4522 u64 sectorsize = btrfs_inode_sectorsize(inode);
4523 struct extent_map *em;
4530 len = last - offset;
4533 len = ALIGN(len, sectorsize);
4534 em = btrfs_get_extent_fiemap(inode, offset, len);
4535 if (IS_ERR_OR_NULL(em))
4538 /* if this isn't a hole return it */
4539 if (em->block_start != EXTENT_MAP_HOLE)
4542 /* this is a hole, advance to the next extent */
4543 offset = extent_map_end(em);
4544 free_extent_map(em);
4552 * To cache previous fiemap extent
4554 * Will be used for merging fiemap extent
4556 struct fiemap_cache {
4565 * Helper to submit fiemap extent.
4567 * Will try to merge current fiemap extent specified by @offset, @phys,
4568 * @len and @flags with cached one.
4569 * And only when we fails to merge, cached one will be submitted as
4572 * Return value is the same as fiemap_fill_next_extent().
4574 static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
4575 struct fiemap_cache *cache,
4576 u64 offset, u64 phys, u64 len, u32 flags)
4584 * Sanity check, extent_fiemap() should have ensured that new
4585 * fiemap extent won't overlap with cached one.
4588 * NOTE: Physical address can overlap, due to compression
4590 if (cache->offset + cache->len > offset) {
4596 * Only merges fiemap extents if
4597 * 1) Their logical addresses are continuous
4599 * 2) Their physical addresses are continuous
4600 * So truly compressed (physical size smaller than logical size)
4601 * extents won't get merged with each other
4603 * 3) Share same flags except FIEMAP_EXTENT_LAST
4604 * So regular extent won't get merged with prealloc extent
4606 if (cache->offset + cache->len == offset &&
4607 cache->phys + cache->len == phys &&
4608 (cache->flags & ~FIEMAP_EXTENT_LAST) ==
4609 (flags & ~FIEMAP_EXTENT_LAST)) {
4611 cache->flags |= flags;
4612 goto try_submit_last;
4615 /* Not mergeable, need to submit cached one */
4616 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
4617 cache->len, cache->flags);
4618 cache->cached = false;
4622 cache->cached = true;
4623 cache->offset = offset;
4626 cache->flags = flags;
4628 if (cache->flags & FIEMAP_EXTENT_LAST) {
4629 ret = fiemap_fill_next_extent(fieinfo, cache->offset,
4630 cache->phys, cache->len, cache->flags);
4631 cache->cached = false;
4637 * Emit last fiemap cache
4639 * The last fiemap cache may still be cached in the following case:
4641 * |<- Fiemap range ->|
4642 * |<------------ First extent ----------->|
4644 * In this case, the first extent range will be cached but not emitted.
4645 * So we must emit it before ending extent_fiemap().
4647 static int emit_last_fiemap_cache(struct fiemap_extent_info *fieinfo,
4648 struct fiemap_cache *cache)
4655 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
4656 cache->len, cache->flags);
4657 cache->cached = false;
4663 int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,
4668 u64 max = start + len;
4672 u64 last_for_get_extent = 0;
4674 u64 isize = i_size_read(&inode->vfs_inode);
4675 struct btrfs_key found_key;
4676 struct extent_map *em = NULL;
4677 struct extent_state *cached_state = NULL;
4678 struct btrfs_path *path;
4679 struct btrfs_root *root = inode->root;
4680 struct fiemap_cache cache = { 0 };
4681 struct ulist *roots;
4682 struct ulist *tmp_ulist;
4691 path = btrfs_alloc_path();
4694 path->leave_spinning = 1;
4696 roots = ulist_alloc(GFP_KERNEL);
4697 tmp_ulist = ulist_alloc(GFP_KERNEL);
4698 if (!roots || !tmp_ulist) {
4700 goto out_free_ulist;
4703 start = round_down(start, btrfs_inode_sectorsize(inode));
4704 len = round_up(max, btrfs_inode_sectorsize(inode)) - start;
4707 * lookup the last file extent. We're not using i_size here
4708 * because there might be preallocation past i_size
4710 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode), -1,
4713 goto out_free_ulist;
4721 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
4722 found_type = found_key.type;
4724 /* No extents, but there might be delalloc bits */
4725 if (found_key.objectid != btrfs_ino(inode) ||
4726 found_type != BTRFS_EXTENT_DATA_KEY) {
4727 /* have to trust i_size as the end */
4729 last_for_get_extent = isize;
4732 * remember the start of the last extent. There are a
4733 * bunch of different factors that go into the length of the
4734 * extent, so its much less complex to remember where it started
4736 last = found_key.offset;
4737 last_for_get_extent = last + 1;
4739 btrfs_release_path(path);
4742 * we might have some extents allocated but more delalloc past those
4743 * extents. so, we trust isize unless the start of the last extent is
4748 last_for_get_extent = isize;
4751 lock_extent_bits(&inode->io_tree, start, start + len - 1,
4754 em = get_extent_skip_holes(inode, start, last_for_get_extent);
4763 u64 offset_in_extent = 0;
4765 /* break if the extent we found is outside the range */
4766 if (em->start >= max || extent_map_end(em) < off)
4770 * get_extent may return an extent that starts before our
4771 * requested range. We have to make sure the ranges
4772 * we return to fiemap always move forward and don't
4773 * overlap, so adjust the offsets here
4775 em_start = max(em->start, off);
4778 * record the offset from the start of the extent
4779 * for adjusting the disk offset below. Only do this if the
4780 * extent isn't compressed since our in ram offset may be past
4781 * what we have actually allocated on disk.
4783 if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4784 offset_in_extent = em_start - em->start;
4785 em_end = extent_map_end(em);
4786 em_len = em_end - em_start;
4788 if (em->block_start < EXTENT_MAP_LAST_BYTE)
4789 disko = em->block_start + offset_in_extent;
4794 * bump off for our next call to get_extent
4796 off = extent_map_end(em);
4800 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
4802 flags |= FIEMAP_EXTENT_LAST;
4803 } else if (em->block_start == EXTENT_MAP_INLINE) {
4804 flags |= (FIEMAP_EXTENT_DATA_INLINE |
4805 FIEMAP_EXTENT_NOT_ALIGNED);
4806 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
4807 flags |= (FIEMAP_EXTENT_DELALLOC |
4808 FIEMAP_EXTENT_UNKNOWN);
4809 } else if (fieinfo->fi_extents_max) {
4810 u64 bytenr = em->block_start -
4811 (em->start - em->orig_start);
4814 * As btrfs supports shared space, this information
4815 * can be exported to userspace tools via
4816 * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0
4817 * then we're just getting a count and we can skip the
4820 ret = btrfs_check_shared(root, btrfs_ino(inode),
4821 bytenr, roots, tmp_ulist);
4825 flags |= FIEMAP_EXTENT_SHARED;
4828 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4829 flags |= FIEMAP_EXTENT_ENCODED;
4830 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4831 flags |= FIEMAP_EXTENT_UNWRITTEN;
4833 free_extent_map(em);
4835 if ((em_start >= last) || em_len == (u64)-1 ||
4836 (last == (u64)-1 && isize <= em_end)) {
4837 flags |= FIEMAP_EXTENT_LAST;
4841 /* now scan forward to see if this is really the last extent. */
4842 em = get_extent_skip_holes(inode, off, last_for_get_extent);
4848 flags |= FIEMAP_EXTENT_LAST;
4851 ret = emit_fiemap_extent(fieinfo, &cache, em_start, disko,
4861 ret = emit_last_fiemap_cache(fieinfo, &cache);
4862 free_extent_map(em);
4864 unlock_extent_cached(&inode->io_tree, start, start + len - 1,
4868 btrfs_free_path(path);
4870 ulist_free(tmp_ulist);
4874 static void __free_extent_buffer(struct extent_buffer *eb)
4876 kmem_cache_free(extent_buffer_cache, eb);
4879 int extent_buffer_under_io(const struct extent_buffer *eb)
4881 return (atomic_read(&eb->io_pages) ||
4882 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4883 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4887 * Release all pages attached to the extent buffer.
4889 static void btrfs_release_extent_buffer_pages(struct extent_buffer *eb)
4893 int mapped = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
4895 BUG_ON(extent_buffer_under_io(eb));
4897 num_pages = num_extent_pages(eb);
4898 for (i = 0; i < num_pages; i++) {
4899 struct page *page = eb->pages[i];
4904 spin_lock(&page->mapping->private_lock);
4906 * We do this since we'll remove the pages after we've
4907 * removed the eb from the radix tree, so we could race
4908 * and have this page now attached to the new eb. So
4909 * only clear page_private if it's still connected to
4912 if (PagePrivate(page) &&
4913 page->private == (unsigned long)eb) {
4914 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4915 BUG_ON(PageDirty(page));
4916 BUG_ON(PageWriteback(page));
4918 * We need to make sure we haven't be attached
4921 detach_page_private(page);
4925 spin_unlock(&page->mapping->private_lock);
4927 /* One for when we allocated the page */
4933 * Helper for releasing the extent buffer.
4935 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4937 btrfs_release_extent_buffer_pages(eb);
4938 btrfs_leak_debug_del(&eb->fs_info->eb_leak_lock, &eb->leak_list);
4939 __free_extent_buffer(eb);
4942 static struct extent_buffer *
4943 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
4946 struct extent_buffer *eb = NULL;
4948 eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL);
4951 eb->fs_info = fs_info;
4953 rwlock_init(&eb->lock);
4954 atomic_set(&eb->blocking_readers, 0);
4955 eb->blocking_writers = 0;
4956 eb->lock_recursed = false;
4957 init_waitqueue_head(&eb->write_lock_wq);
4958 init_waitqueue_head(&eb->read_lock_wq);
4960 btrfs_leak_debug_add(&fs_info->eb_leak_lock, &eb->leak_list,
4961 &fs_info->allocated_ebs);
4963 spin_lock_init(&eb->refs_lock);
4964 atomic_set(&eb->refs, 1);
4965 atomic_set(&eb->io_pages, 0);
4968 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4970 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4971 > MAX_INLINE_EXTENT_BUFFER_SIZE);
4972 BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE);
4974 #ifdef CONFIG_BTRFS_DEBUG
4975 eb->spinning_writers = 0;
4976 atomic_set(&eb->spinning_readers, 0);
4977 atomic_set(&eb->read_locks, 0);
4978 eb->write_locks = 0;
4984 struct extent_buffer *btrfs_clone_extent_buffer(const struct extent_buffer *src)
4988 struct extent_buffer *new;
4989 int num_pages = num_extent_pages(src);
4991 new = __alloc_extent_buffer(src->fs_info, src->start, src->len);
4995 for (i = 0; i < num_pages; i++) {
4996 p = alloc_page(GFP_NOFS);
4998 btrfs_release_extent_buffer(new);
5001 attach_extent_buffer_page(new, p);
5002 WARN_ON(PageDirty(p));
5005 copy_page(page_address(p), page_address(src->pages[i]));
5008 set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
5009 set_bit(EXTENT_BUFFER_UNMAPPED, &new->bflags);
5014 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
5015 u64 start, unsigned long len)
5017 struct extent_buffer *eb;
5021 eb = __alloc_extent_buffer(fs_info, start, len);
5025 num_pages = num_extent_pages(eb);
5026 for (i = 0; i < num_pages; i++) {
5027 eb->pages[i] = alloc_page(GFP_NOFS);
5031 set_extent_buffer_uptodate(eb);
5032 btrfs_set_header_nritems(eb, 0);
5033 set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
5038 __free_page(eb->pages[i - 1]);
5039 __free_extent_buffer(eb);
5043 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
5046 return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize);
5049 static void check_buffer_tree_ref(struct extent_buffer *eb)
5053 * The TREE_REF bit is first set when the extent_buffer is added
5054 * to the radix tree. It is also reset, if unset, when a new reference
5055 * is created by find_extent_buffer.
5057 * It is only cleared in two cases: freeing the last non-tree
5058 * reference to the extent_buffer when its STALE bit is set or
5059 * calling releasepage when the tree reference is the only reference.
5061 * In both cases, care is taken to ensure that the extent_buffer's
5062 * pages are not under io. However, releasepage can be concurrently
5063 * called with creating new references, which is prone to race
5064 * conditions between the calls to check_buffer_tree_ref in those
5065 * codepaths and clearing TREE_REF in try_release_extent_buffer.
5067 * The actual lifetime of the extent_buffer in the radix tree is
5068 * adequately protected by the refcount, but the TREE_REF bit and
5069 * its corresponding reference are not. To protect against this
5070 * class of races, we call check_buffer_tree_ref from the codepaths
5071 * which trigger io after they set eb->io_pages. Note that once io is
5072 * initiated, TREE_REF can no longer be cleared, so that is the
5073 * moment at which any such race is best fixed.
5075 refs = atomic_read(&eb->refs);
5076 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5079 spin_lock(&eb->refs_lock);
5080 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5081 atomic_inc(&eb->refs);
5082 spin_unlock(&eb->refs_lock);
5085 static void mark_extent_buffer_accessed(struct extent_buffer *eb,
5086 struct page *accessed)
5090 check_buffer_tree_ref(eb);
5092 num_pages = num_extent_pages(eb);
5093 for (i = 0; i < num_pages; i++) {
5094 struct page *p = eb->pages[i];
5097 mark_page_accessed(p);
5101 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
5104 struct extent_buffer *eb;
5107 eb = radix_tree_lookup(&fs_info->buffer_radix,
5108 start >> PAGE_SHIFT);
5109 if (eb && atomic_inc_not_zero(&eb->refs)) {
5112 * Lock our eb's refs_lock to avoid races with
5113 * free_extent_buffer. When we get our eb it might be flagged
5114 * with EXTENT_BUFFER_STALE and another task running
5115 * free_extent_buffer might have seen that flag set,
5116 * eb->refs == 2, that the buffer isn't under IO (dirty and
5117 * writeback flags not set) and it's still in the tree (flag
5118 * EXTENT_BUFFER_TREE_REF set), therefore being in the process
5119 * of decrementing the extent buffer's reference count twice.
5120 * So here we could race and increment the eb's reference count,
5121 * clear its stale flag, mark it as dirty and drop our reference
5122 * before the other task finishes executing free_extent_buffer,
5123 * which would later result in an attempt to free an extent
5124 * buffer that is dirty.
5126 if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) {
5127 spin_lock(&eb->refs_lock);
5128 spin_unlock(&eb->refs_lock);
5130 mark_extent_buffer_accessed(eb, NULL);
5138 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5139 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
5142 struct extent_buffer *eb, *exists = NULL;
5145 eb = find_extent_buffer(fs_info, start);
5148 eb = alloc_dummy_extent_buffer(fs_info, start);
5150 return ERR_PTR(-ENOMEM);
5151 eb->fs_info = fs_info;
5153 ret = radix_tree_preload(GFP_NOFS);
5155 exists = ERR_PTR(ret);
5158 spin_lock(&fs_info->buffer_lock);
5159 ret = radix_tree_insert(&fs_info->buffer_radix,
5160 start >> PAGE_SHIFT, eb);
5161 spin_unlock(&fs_info->buffer_lock);
5162 radix_tree_preload_end();
5163 if (ret == -EEXIST) {
5164 exists = find_extent_buffer(fs_info, start);
5170 check_buffer_tree_ref(eb);
5171 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
5175 btrfs_release_extent_buffer(eb);
5180 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
5183 unsigned long len = fs_info->nodesize;
5186 unsigned long index = start >> PAGE_SHIFT;
5187 struct extent_buffer *eb;
5188 struct extent_buffer *exists = NULL;
5190 struct address_space *mapping = fs_info->btree_inode->i_mapping;
5194 if (!IS_ALIGNED(start, fs_info->sectorsize)) {
5195 btrfs_err(fs_info, "bad tree block start %llu", start);
5196 return ERR_PTR(-EINVAL);
5199 eb = find_extent_buffer(fs_info, start);
5203 eb = __alloc_extent_buffer(fs_info, start, len);
5205 return ERR_PTR(-ENOMEM);
5207 num_pages = num_extent_pages(eb);
5208 for (i = 0; i < num_pages; i++, index++) {
5209 p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL);
5211 exists = ERR_PTR(-ENOMEM);
5215 spin_lock(&mapping->private_lock);
5216 if (PagePrivate(p)) {
5218 * We could have already allocated an eb for this page
5219 * and attached one so lets see if we can get a ref on
5220 * the existing eb, and if we can we know it's good and
5221 * we can just return that one, else we know we can just
5222 * overwrite page->private.
5224 exists = (struct extent_buffer *)p->private;
5225 if (atomic_inc_not_zero(&exists->refs)) {
5226 spin_unlock(&mapping->private_lock);
5229 mark_extent_buffer_accessed(exists, p);
5235 * Do this so attach doesn't complain and we need to
5236 * drop the ref the old guy had.
5238 ClearPagePrivate(p);
5239 WARN_ON(PageDirty(p));
5242 attach_extent_buffer_page(eb, p);
5243 spin_unlock(&mapping->private_lock);
5244 WARN_ON(PageDirty(p));
5246 if (!PageUptodate(p))
5250 * We can't unlock the pages just yet since the extent buffer
5251 * hasn't been properly inserted in the radix tree, this
5252 * opens a race with btree_releasepage which can free a page
5253 * while we are still filling in all pages for the buffer and
5258 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5260 ret = radix_tree_preload(GFP_NOFS);
5262 exists = ERR_PTR(ret);
5266 spin_lock(&fs_info->buffer_lock);
5267 ret = radix_tree_insert(&fs_info->buffer_radix,
5268 start >> PAGE_SHIFT, eb);
5269 spin_unlock(&fs_info->buffer_lock);
5270 radix_tree_preload_end();
5271 if (ret == -EEXIST) {
5272 exists = find_extent_buffer(fs_info, start);
5278 /* add one reference for the tree */
5279 check_buffer_tree_ref(eb);
5280 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
5283 * Now it's safe to unlock the pages because any calls to
5284 * btree_releasepage will correctly detect that a page belongs to a
5285 * live buffer and won't free them prematurely.
5287 for (i = 0; i < num_pages; i++)
5288 unlock_page(eb->pages[i]);
5292 WARN_ON(!atomic_dec_and_test(&eb->refs));
5293 for (i = 0; i < num_pages; i++) {
5295 unlock_page(eb->pages[i]);
5298 btrfs_release_extent_buffer(eb);
5302 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
5304 struct extent_buffer *eb =
5305 container_of(head, struct extent_buffer, rcu_head);
5307 __free_extent_buffer(eb);
5310 static int release_extent_buffer(struct extent_buffer *eb)
5311 __releases(&eb->refs_lock)
5313 lockdep_assert_held(&eb->refs_lock);
5315 WARN_ON(atomic_read(&eb->refs) == 0);
5316 if (atomic_dec_and_test(&eb->refs)) {
5317 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
5318 struct btrfs_fs_info *fs_info = eb->fs_info;
5320 spin_unlock(&eb->refs_lock);
5322 spin_lock(&fs_info->buffer_lock);
5323 radix_tree_delete(&fs_info->buffer_radix,
5324 eb->start >> PAGE_SHIFT);
5325 spin_unlock(&fs_info->buffer_lock);
5327 spin_unlock(&eb->refs_lock);
5330 btrfs_leak_debug_del(&eb->fs_info->eb_leak_lock, &eb->leak_list);
5331 /* Should be safe to release our pages at this point */
5332 btrfs_release_extent_buffer_pages(eb);
5333 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5334 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags))) {
5335 __free_extent_buffer(eb);
5339 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
5342 spin_unlock(&eb->refs_lock);
5347 void free_extent_buffer(struct extent_buffer *eb)
5355 refs = atomic_read(&eb->refs);
5356 if ((!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && refs <= 3)
5357 || (test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) &&
5360 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
5365 spin_lock(&eb->refs_lock);
5366 if (atomic_read(&eb->refs) == 2 &&
5367 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
5368 !extent_buffer_under_io(eb) &&
5369 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5370 atomic_dec(&eb->refs);
5373 * I know this is terrible, but it's temporary until we stop tracking
5374 * the uptodate bits and such for the extent buffers.
5376 release_extent_buffer(eb);
5379 void free_extent_buffer_stale(struct extent_buffer *eb)
5384 spin_lock(&eb->refs_lock);
5385 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
5387 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
5388 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5389 atomic_dec(&eb->refs);
5390 release_extent_buffer(eb);
5393 void clear_extent_buffer_dirty(const struct extent_buffer *eb)
5399 num_pages = num_extent_pages(eb);
5401 for (i = 0; i < num_pages; i++) {
5402 page = eb->pages[i];
5403 if (!PageDirty(page))
5407 WARN_ON(!PagePrivate(page));
5409 clear_page_dirty_for_io(page);
5410 xa_lock_irq(&page->mapping->i_pages);
5411 if (!PageDirty(page))
5412 __xa_clear_mark(&page->mapping->i_pages,
5413 page_index(page), PAGECACHE_TAG_DIRTY);
5414 xa_unlock_irq(&page->mapping->i_pages);
5415 ClearPageError(page);
5418 WARN_ON(atomic_read(&eb->refs) == 0);
5421 bool set_extent_buffer_dirty(struct extent_buffer *eb)
5427 check_buffer_tree_ref(eb);
5429 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
5431 num_pages = num_extent_pages(eb);
5432 WARN_ON(atomic_read(&eb->refs) == 0);
5433 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
5436 for (i = 0; i < num_pages; i++)
5437 set_page_dirty(eb->pages[i]);
5439 #ifdef CONFIG_BTRFS_DEBUG
5440 for (i = 0; i < num_pages; i++)
5441 ASSERT(PageDirty(eb->pages[i]));
5447 void clear_extent_buffer_uptodate(struct extent_buffer *eb)
5453 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5454 num_pages = num_extent_pages(eb);
5455 for (i = 0; i < num_pages; i++) {
5456 page = eb->pages[i];
5458 ClearPageUptodate(page);
5462 void set_extent_buffer_uptodate(struct extent_buffer *eb)
5468 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5469 num_pages = num_extent_pages(eb);
5470 for (i = 0; i < num_pages; i++) {
5471 page = eb->pages[i];
5472 SetPageUptodate(page);
5476 int read_extent_buffer_pages(struct extent_buffer *eb, int wait, int mirror_num)
5482 int locked_pages = 0;
5483 int all_uptodate = 1;
5485 unsigned long num_reads = 0;
5486 struct bio *bio = NULL;
5487 unsigned long bio_flags = 0;
5489 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
5492 num_pages = num_extent_pages(eb);
5493 for (i = 0; i < num_pages; i++) {
5494 page = eb->pages[i];
5495 if (wait == WAIT_NONE) {
5496 if (!trylock_page(page))
5504 * We need to firstly lock all pages to make sure that
5505 * the uptodate bit of our pages won't be affected by
5506 * clear_extent_buffer_uptodate().
5508 for (i = 0; i < num_pages; i++) {
5509 page = eb->pages[i];
5510 if (!PageUptodate(page)) {
5517 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5521 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
5522 eb->read_mirror = 0;
5523 atomic_set(&eb->io_pages, num_reads);
5525 * It is possible for releasepage to clear the TREE_REF bit before we
5526 * set io_pages. See check_buffer_tree_ref for a more detailed comment.
5528 check_buffer_tree_ref(eb);
5529 for (i = 0; i < num_pages; i++) {
5530 page = eb->pages[i];
5532 if (!PageUptodate(page)) {
5534 atomic_dec(&eb->io_pages);
5539 ClearPageError(page);
5540 err = submit_extent_page(REQ_OP_READ | REQ_META, NULL,
5541 page, page_offset(page), PAGE_SIZE, 0,
5542 &bio, end_bio_extent_readpage,
5543 mirror_num, 0, 0, false);
5546 * We failed to submit the bio so it's the
5547 * caller's responsibility to perform cleanup
5548 * i.e unlock page/set error bit.
5553 atomic_dec(&eb->io_pages);
5561 err = submit_one_bio(bio, mirror_num, bio_flags);
5566 if (ret || wait != WAIT_COMPLETE)
5569 for (i = 0; i < num_pages; i++) {
5570 page = eb->pages[i];
5571 wait_on_page_locked(page);
5572 if (!PageUptodate(page))
5579 while (locked_pages > 0) {
5581 page = eb->pages[locked_pages];
5587 static bool report_eb_range(const struct extent_buffer *eb, unsigned long start,
5590 btrfs_warn(eb->fs_info,
5591 "access to eb bytenr %llu len %lu out of range start %lu len %lu",
5592 eb->start, eb->len, start, len);
5593 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
5599 * Check if the [start, start + len) range is valid before reading/writing
5601 * NOTE: @start and @len are offset inside the eb, not logical address.
5603 * Caller should not touch the dst/src memory if this function returns error.
5605 static inline int check_eb_range(const struct extent_buffer *eb,
5606 unsigned long start, unsigned long len)
5608 unsigned long offset;
5610 /* start, start + len should not go beyond eb->len nor overflow */
5611 if (unlikely(check_add_overflow(start, len, &offset) || offset > eb->len))
5612 return report_eb_range(eb, start, len);
5617 void read_extent_buffer(const struct extent_buffer *eb, void *dstv,
5618 unsigned long start, unsigned long len)
5624 char *dst = (char *)dstv;
5625 unsigned long i = start >> PAGE_SHIFT;
5627 if (check_eb_range(eb, start, len))
5630 offset = offset_in_page(start);
5633 page = eb->pages[i];
5635 cur = min(len, (PAGE_SIZE - offset));
5636 kaddr = page_address(page);
5637 memcpy(dst, kaddr + offset, cur);
5646 int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb,
5648 unsigned long start, unsigned long len)
5654 char __user *dst = (char __user *)dstv;
5655 unsigned long i = start >> PAGE_SHIFT;
5658 WARN_ON(start > eb->len);
5659 WARN_ON(start + len > eb->start + eb->len);
5661 offset = offset_in_page(start);
5664 page = eb->pages[i];
5666 cur = min(len, (PAGE_SIZE - offset));
5667 kaddr = page_address(page);
5668 if (copy_to_user_nofault(dst, kaddr + offset, cur)) {
5682 int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,
5683 unsigned long start, unsigned long len)
5689 char *ptr = (char *)ptrv;
5690 unsigned long i = start >> PAGE_SHIFT;
5693 if (check_eb_range(eb, start, len))
5696 offset = offset_in_page(start);
5699 page = eb->pages[i];
5701 cur = min(len, (PAGE_SIZE - offset));
5703 kaddr = page_address(page);
5704 ret = memcmp(ptr, kaddr + offset, cur);
5716 void write_extent_buffer_chunk_tree_uuid(const struct extent_buffer *eb,
5721 WARN_ON(!PageUptodate(eb->pages[0]));
5722 kaddr = page_address(eb->pages[0]);
5723 memcpy(kaddr + offsetof(struct btrfs_header, chunk_tree_uuid), srcv,
5727 void write_extent_buffer_fsid(const struct extent_buffer *eb, const void *srcv)
5731 WARN_ON(!PageUptodate(eb->pages[0]));
5732 kaddr = page_address(eb->pages[0]);
5733 memcpy(kaddr + offsetof(struct btrfs_header, fsid), srcv,
5737 void write_extent_buffer(const struct extent_buffer *eb, const void *srcv,
5738 unsigned long start, unsigned long len)
5744 char *src = (char *)srcv;
5745 unsigned long i = start >> PAGE_SHIFT;
5747 if (check_eb_range(eb, start, len))
5750 offset = offset_in_page(start);
5753 page = eb->pages[i];
5754 WARN_ON(!PageUptodate(page));
5756 cur = min(len, PAGE_SIZE - offset);
5757 kaddr = page_address(page);
5758 memcpy(kaddr + offset, src, cur);
5767 void memzero_extent_buffer(const struct extent_buffer *eb, unsigned long start,
5774 unsigned long i = start >> PAGE_SHIFT;
5776 if (check_eb_range(eb, start, len))
5779 offset = offset_in_page(start);
5782 page = eb->pages[i];
5783 WARN_ON(!PageUptodate(page));
5785 cur = min(len, PAGE_SIZE - offset);
5786 kaddr = page_address(page);
5787 memset(kaddr + offset, 0, cur);
5795 void copy_extent_buffer_full(const struct extent_buffer *dst,
5796 const struct extent_buffer *src)
5801 ASSERT(dst->len == src->len);
5803 num_pages = num_extent_pages(dst);
5804 for (i = 0; i < num_pages; i++)
5805 copy_page(page_address(dst->pages[i]),
5806 page_address(src->pages[i]));
5809 void copy_extent_buffer(const struct extent_buffer *dst,
5810 const struct extent_buffer *src,
5811 unsigned long dst_offset, unsigned long src_offset,
5814 u64 dst_len = dst->len;
5819 unsigned long i = dst_offset >> PAGE_SHIFT;
5821 if (check_eb_range(dst, dst_offset, len) ||
5822 check_eb_range(src, src_offset, len))
5825 WARN_ON(src->len != dst_len);
5827 offset = offset_in_page(dst_offset);
5830 page = dst->pages[i];
5831 WARN_ON(!PageUptodate(page));
5833 cur = min(len, (unsigned long)(PAGE_SIZE - offset));
5835 kaddr = page_address(page);
5836 read_extent_buffer(src, kaddr + offset, src_offset, cur);
5846 * eb_bitmap_offset() - calculate the page and offset of the byte containing the
5848 * @eb: the extent buffer
5849 * @start: offset of the bitmap item in the extent buffer
5851 * @page_index: return index of the page in the extent buffer that contains the
5853 * @page_offset: return offset into the page given by page_index
5855 * This helper hides the ugliness of finding the byte in an extent buffer which
5856 * contains a given bit.
5858 static inline void eb_bitmap_offset(const struct extent_buffer *eb,
5859 unsigned long start, unsigned long nr,
5860 unsigned long *page_index,
5861 size_t *page_offset)
5863 size_t byte_offset = BIT_BYTE(nr);
5867 * The byte we want is the offset of the extent buffer + the offset of
5868 * the bitmap item in the extent buffer + the offset of the byte in the
5871 offset = start + byte_offset;
5873 *page_index = offset >> PAGE_SHIFT;
5874 *page_offset = offset_in_page(offset);
5878 * extent_buffer_test_bit - determine whether a bit in a bitmap item is set
5879 * @eb: the extent buffer
5880 * @start: offset of the bitmap item in the extent buffer
5881 * @nr: bit number to test
5883 int extent_buffer_test_bit(const struct extent_buffer *eb, unsigned long start,
5891 eb_bitmap_offset(eb, start, nr, &i, &offset);
5892 page = eb->pages[i];
5893 WARN_ON(!PageUptodate(page));
5894 kaddr = page_address(page);
5895 return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1)));
5899 * extent_buffer_bitmap_set - set an area of a bitmap
5900 * @eb: the extent buffer
5901 * @start: offset of the bitmap item in the extent buffer
5902 * @pos: bit number of the first bit
5903 * @len: number of bits to set
5905 void extent_buffer_bitmap_set(const struct extent_buffer *eb, unsigned long start,
5906 unsigned long pos, unsigned long len)
5912 const unsigned int size = pos + len;
5913 int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5914 u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos);
5916 eb_bitmap_offset(eb, start, pos, &i, &offset);
5917 page = eb->pages[i];
5918 WARN_ON(!PageUptodate(page));
5919 kaddr = page_address(page);
5921 while (len >= bits_to_set) {
5922 kaddr[offset] |= mask_to_set;
5924 bits_to_set = BITS_PER_BYTE;
5926 if (++offset >= PAGE_SIZE && len > 0) {
5928 page = eb->pages[++i];
5929 WARN_ON(!PageUptodate(page));
5930 kaddr = page_address(page);
5934 mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
5935 kaddr[offset] |= mask_to_set;
5941 * extent_buffer_bitmap_clear - clear an area of a bitmap
5942 * @eb: the extent buffer
5943 * @start: offset of the bitmap item in the extent buffer
5944 * @pos: bit number of the first bit
5945 * @len: number of bits to clear
5947 void extent_buffer_bitmap_clear(const struct extent_buffer *eb,
5948 unsigned long start, unsigned long pos,
5955 const unsigned int size = pos + len;
5956 int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5957 u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos);
5959 eb_bitmap_offset(eb, start, pos, &i, &offset);
5960 page = eb->pages[i];
5961 WARN_ON(!PageUptodate(page));
5962 kaddr = page_address(page);
5964 while (len >= bits_to_clear) {
5965 kaddr[offset] &= ~mask_to_clear;
5966 len -= bits_to_clear;
5967 bits_to_clear = BITS_PER_BYTE;
5969 if (++offset >= PAGE_SIZE && len > 0) {
5971 page = eb->pages[++i];
5972 WARN_ON(!PageUptodate(page));
5973 kaddr = page_address(page);
5977 mask_to_clear &= BITMAP_LAST_BYTE_MASK(size);
5978 kaddr[offset] &= ~mask_to_clear;
5982 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
5984 unsigned long distance = (src > dst) ? src - dst : dst - src;
5985 return distance < len;
5988 static void copy_pages(struct page *dst_page, struct page *src_page,
5989 unsigned long dst_off, unsigned long src_off,
5992 char *dst_kaddr = page_address(dst_page);
5994 int must_memmove = 0;
5996 if (dst_page != src_page) {
5997 src_kaddr = page_address(src_page);
5999 src_kaddr = dst_kaddr;
6000 if (areas_overlap(src_off, dst_off, len))
6005 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
6007 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
6010 void memcpy_extent_buffer(const struct extent_buffer *dst,
6011 unsigned long dst_offset, unsigned long src_offset,
6015 size_t dst_off_in_page;
6016 size_t src_off_in_page;
6017 unsigned long dst_i;
6018 unsigned long src_i;
6020 if (check_eb_range(dst, dst_offset, len) ||
6021 check_eb_range(dst, src_offset, len))
6025 dst_off_in_page = offset_in_page(dst_offset);
6026 src_off_in_page = offset_in_page(src_offset);
6028 dst_i = dst_offset >> PAGE_SHIFT;
6029 src_i = src_offset >> PAGE_SHIFT;
6031 cur = min(len, (unsigned long)(PAGE_SIZE -
6033 cur = min_t(unsigned long, cur,
6034 (unsigned long)(PAGE_SIZE - dst_off_in_page));
6036 copy_pages(dst->pages[dst_i], dst->pages[src_i],
6037 dst_off_in_page, src_off_in_page, cur);
6045 void memmove_extent_buffer(const struct extent_buffer *dst,
6046 unsigned long dst_offset, unsigned long src_offset,
6050 size_t dst_off_in_page;
6051 size_t src_off_in_page;
6052 unsigned long dst_end = dst_offset + len - 1;
6053 unsigned long src_end = src_offset + len - 1;
6054 unsigned long dst_i;
6055 unsigned long src_i;
6057 if (check_eb_range(dst, dst_offset, len) ||
6058 check_eb_range(dst, src_offset, len))
6060 if (dst_offset < src_offset) {
6061 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
6065 dst_i = dst_end >> PAGE_SHIFT;
6066 src_i = src_end >> PAGE_SHIFT;
6068 dst_off_in_page = offset_in_page(dst_end);
6069 src_off_in_page = offset_in_page(src_end);
6071 cur = min_t(unsigned long, len, src_off_in_page + 1);
6072 cur = min(cur, dst_off_in_page + 1);
6073 copy_pages(dst->pages[dst_i], dst->pages[src_i],
6074 dst_off_in_page - cur + 1,
6075 src_off_in_page - cur + 1, cur);
6083 int try_release_extent_buffer(struct page *page)
6085 struct extent_buffer *eb;
6088 * We need to make sure nobody is attaching this page to an eb right
6091 spin_lock(&page->mapping->private_lock);
6092 if (!PagePrivate(page)) {
6093 spin_unlock(&page->mapping->private_lock);
6097 eb = (struct extent_buffer *)page->private;
6101 * This is a little awful but should be ok, we need to make sure that
6102 * the eb doesn't disappear out from under us while we're looking at
6105 spin_lock(&eb->refs_lock);
6106 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
6107 spin_unlock(&eb->refs_lock);
6108 spin_unlock(&page->mapping->private_lock);
6111 spin_unlock(&page->mapping->private_lock);
6114 * If tree ref isn't set then we know the ref on this eb is a real ref,
6115 * so just return, this page will likely be freed soon anyway.
6117 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
6118 spin_unlock(&eb->refs_lock);
6122 return release_extent_buffer(eb);