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 btrfs_panic(tree->fs_info, err,
680 "locking error: extent tree was modified by another thread while locked");
684 * clear some bits on a range in the tree. This may require splitting
685 * or inserting elements in the tree, so the gfp mask is used to
686 * indicate which allocations or sleeping are allowed.
688 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
689 * the given range from the tree regardless of state (ie for truncate).
691 * the range [start, end] is inclusive.
693 * This takes the tree lock, and returns 0 on success and < 0 on error.
695 int __clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
696 unsigned bits, int wake, int delete,
697 struct extent_state **cached_state,
698 gfp_t mask, struct extent_changeset *changeset)
700 struct extent_state *state;
701 struct extent_state *cached;
702 struct extent_state *prealloc = NULL;
703 struct rb_node *node;
708 btrfs_debug_check_extent_io_range(tree, start, end);
709 trace_btrfs_clear_extent_bit(tree, start, end - start + 1, bits);
711 if (bits & EXTENT_DELALLOC)
712 bits |= EXTENT_NORESERVE;
715 bits |= ~EXTENT_CTLBITS;
717 if (bits & (EXTENT_LOCKED | EXTENT_BOUNDARY))
720 if (!prealloc && gfpflags_allow_blocking(mask)) {
722 * Don't care for allocation failure here because we might end
723 * up not needing the pre-allocated extent state at all, which
724 * is the case if we only have in the tree extent states that
725 * cover our input range and don't cover too any other range.
726 * If we end up needing a new extent state we allocate it later.
728 prealloc = alloc_extent_state(mask);
731 spin_lock(&tree->lock);
733 cached = *cached_state;
736 *cached_state = NULL;
740 if (cached && extent_state_in_tree(cached) &&
741 cached->start <= start && cached->end > start) {
743 refcount_dec(&cached->refs);
748 free_extent_state(cached);
751 * this search will find the extents that end after
754 node = tree_search(tree, start);
757 state = rb_entry(node, struct extent_state, rb_node);
759 if (state->start > end)
761 WARN_ON(state->end < start);
762 last_end = state->end;
764 /* the state doesn't have the wanted bits, go ahead */
765 if (!(state->state & bits)) {
766 state = next_state(state);
771 * | ---- desired range ---- |
773 * | ------------- state -------------- |
775 * We need to split the extent we found, and may flip
776 * bits on second half.
778 * If the extent we found extends past our range, we
779 * just split and search again. It'll get split again
780 * the next time though.
782 * If the extent we found is inside our range, we clear
783 * the desired bit on it.
786 if (state->start < start) {
787 prealloc = alloc_extent_state_atomic(prealloc);
789 err = split_state(tree, state, prealloc, start);
791 extent_io_tree_panic(tree, err);
796 if (state->end <= end) {
797 state = clear_state_bit(tree, state, &bits, wake,
804 * | ---- desired range ---- |
806 * We need to split the extent, and clear the bit
809 if (state->start <= end && state->end > end) {
810 prealloc = alloc_extent_state_atomic(prealloc);
812 err = split_state(tree, state, prealloc, end + 1);
814 extent_io_tree_panic(tree, err);
819 clear_state_bit(tree, prealloc, &bits, wake, changeset);
825 state = clear_state_bit(tree, state, &bits, wake, changeset);
827 if (last_end == (u64)-1)
829 start = last_end + 1;
830 if (start <= end && state && !need_resched())
836 spin_unlock(&tree->lock);
837 if (gfpflags_allow_blocking(mask))
842 spin_unlock(&tree->lock);
844 free_extent_state(prealloc);
850 static void wait_on_state(struct extent_io_tree *tree,
851 struct extent_state *state)
852 __releases(tree->lock)
853 __acquires(tree->lock)
856 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
857 spin_unlock(&tree->lock);
859 spin_lock(&tree->lock);
860 finish_wait(&state->wq, &wait);
864 * waits for one or more bits to clear on a range in the state tree.
865 * The range [start, end] is inclusive.
866 * The tree lock is taken by this function
868 static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
871 struct extent_state *state;
872 struct rb_node *node;
874 btrfs_debug_check_extent_io_range(tree, start, end);
876 spin_lock(&tree->lock);
880 * this search will find all the extents that end after
883 node = tree_search(tree, start);
888 state = rb_entry(node, struct extent_state, rb_node);
890 if (state->start > end)
893 if (state->state & bits) {
894 start = state->start;
895 refcount_inc(&state->refs);
896 wait_on_state(tree, state);
897 free_extent_state(state);
900 start = state->end + 1;
905 if (!cond_resched_lock(&tree->lock)) {
906 node = rb_next(node);
911 spin_unlock(&tree->lock);
914 static void set_state_bits(struct extent_io_tree *tree,
915 struct extent_state *state,
916 unsigned *bits, struct extent_changeset *changeset)
918 unsigned bits_to_set = *bits & ~EXTENT_CTLBITS;
921 if (tree->private_data && is_data_inode(tree->private_data))
922 btrfs_set_delalloc_extent(tree->private_data, state, bits);
924 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
925 u64 range = state->end - state->start + 1;
926 tree->dirty_bytes += range;
928 ret = add_extent_changeset(state, bits_to_set, changeset, 1);
930 state->state |= bits_to_set;
933 static void cache_state_if_flags(struct extent_state *state,
934 struct extent_state **cached_ptr,
937 if (cached_ptr && !(*cached_ptr)) {
938 if (!flags || (state->state & flags)) {
940 refcount_inc(&state->refs);
945 static void cache_state(struct extent_state *state,
946 struct extent_state **cached_ptr)
948 return cache_state_if_flags(state, cached_ptr,
949 EXTENT_LOCKED | EXTENT_BOUNDARY);
953 * set some bits on a range in the tree. This may require allocations or
954 * sleeping, so the gfp mask is used to indicate what is allowed.
956 * If any of the exclusive bits are set, this will fail with -EEXIST if some
957 * part of the range already has the desired bits set. The start of the
958 * existing range is returned in failed_start in this case.
960 * [start, end] is inclusive This takes the tree lock.
963 static int __must_check
964 __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
965 unsigned bits, unsigned exclusive_bits,
966 u64 *failed_start, struct extent_state **cached_state,
967 gfp_t mask, struct extent_changeset *changeset)
969 struct extent_state *state;
970 struct extent_state *prealloc = NULL;
971 struct rb_node *node;
973 struct rb_node *parent;
978 btrfs_debug_check_extent_io_range(tree, start, end);
979 trace_btrfs_set_extent_bit(tree, start, end - start + 1, bits);
982 if (!prealloc && gfpflags_allow_blocking(mask)) {
984 * Don't care for allocation failure here because we might end
985 * up not needing the pre-allocated extent state at all, which
986 * is the case if we only have in the tree extent states that
987 * cover our input range and don't cover too any other range.
988 * If we end up needing a new extent state we allocate it later.
990 prealloc = alloc_extent_state(mask);
993 spin_lock(&tree->lock);
994 if (cached_state && *cached_state) {
995 state = *cached_state;
996 if (state->start <= start && state->end > start &&
997 extent_state_in_tree(state)) {
998 node = &state->rb_node;
1003 * this search will find all the extents that end after
1006 node = tree_search_for_insert(tree, start, &p, &parent);
1008 prealloc = alloc_extent_state_atomic(prealloc);
1010 err = insert_state(tree, prealloc, start, end,
1011 &p, &parent, &bits, changeset);
1013 extent_io_tree_panic(tree, err);
1015 cache_state(prealloc, cached_state);
1019 state = rb_entry(node, struct extent_state, rb_node);
1021 last_start = state->start;
1022 last_end = state->end;
1025 * | ---- desired range ---- |
1028 * Just lock what we found and keep going
1030 if (state->start == start && state->end <= end) {
1031 if (state->state & exclusive_bits) {
1032 *failed_start = state->start;
1037 set_state_bits(tree, state, &bits, changeset);
1038 cache_state(state, cached_state);
1039 merge_state(tree, state);
1040 if (last_end == (u64)-1)
1042 start = last_end + 1;
1043 state = next_state(state);
1044 if (start < end && state && state->start == start &&
1051 * | ---- desired range ---- |
1054 * | ------------- state -------------- |
1056 * We need to split the extent we found, and may flip bits on
1059 * If the extent we found extends past our
1060 * range, we just split and search again. It'll get split
1061 * again the next time though.
1063 * If the extent we found is inside our range, we set the
1064 * desired bit on it.
1066 if (state->start < start) {
1067 if (state->state & exclusive_bits) {
1068 *failed_start = start;
1074 * If this extent already has all the bits we want set, then
1075 * skip it, not necessary to split it or do anything with it.
1077 if ((state->state & bits) == bits) {
1078 start = state->end + 1;
1079 cache_state(state, cached_state);
1083 prealloc = alloc_extent_state_atomic(prealloc);
1085 err = split_state(tree, state, prealloc, start);
1087 extent_io_tree_panic(tree, err);
1092 if (state->end <= end) {
1093 set_state_bits(tree, state, &bits, changeset);
1094 cache_state(state, cached_state);
1095 merge_state(tree, state);
1096 if (last_end == (u64)-1)
1098 start = last_end + 1;
1099 state = next_state(state);
1100 if (start < end && state && state->start == start &&
1107 * | ---- desired range ---- |
1108 * | state | or | state |
1110 * There's a hole, we need to insert something in it and
1111 * ignore the extent we found.
1113 if (state->start > start) {
1115 if (end < last_start)
1118 this_end = last_start - 1;
1120 prealloc = alloc_extent_state_atomic(prealloc);
1124 * Avoid to free 'prealloc' if it can be merged with
1127 err = insert_state(tree, prealloc, start, this_end,
1128 NULL, NULL, &bits, changeset);
1130 extent_io_tree_panic(tree, err);
1132 cache_state(prealloc, cached_state);
1134 start = this_end + 1;
1138 * | ---- desired range ---- |
1140 * We need to split the extent, and set the bit
1143 if (state->start <= end && state->end > end) {
1144 if (state->state & exclusive_bits) {
1145 *failed_start = start;
1150 prealloc = alloc_extent_state_atomic(prealloc);
1152 err = split_state(tree, state, prealloc, end + 1);
1154 extent_io_tree_panic(tree, err);
1156 set_state_bits(tree, prealloc, &bits, changeset);
1157 cache_state(prealloc, cached_state);
1158 merge_state(tree, prealloc);
1166 spin_unlock(&tree->lock);
1167 if (gfpflags_allow_blocking(mask))
1172 spin_unlock(&tree->lock);
1174 free_extent_state(prealloc);
1180 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1181 unsigned bits, u64 * failed_start,
1182 struct extent_state **cached_state, gfp_t mask)
1184 return __set_extent_bit(tree, start, end, bits, 0, failed_start,
1185 cached_state, mask, NULL);
1190 * convert_extent_bit - convert all bits in a given range from one bit to
1192 * @tree: the io tree to search
1193 * @start: the start offset in bytes
1194 * @end: the end offset in bytes (inclusive)
1195 * @bits: the bits to set in this range
1196 * @clear_bits: the bits to clear in this range
1197 * @cached_state: state that we're going to cache
1199 * This will go through and set bits for the given range. If any states exist
1200 * already in this range they are set with the given bit and cleared of the
1201 * clear_bits. This is only meant to be used by things that are mergeable, ie
1202 * converting from say DELALLOC to DIRTY. This is not meant to be used with
1203 * boundary bits like LOCK.
1205 * All allocations are done with GFP_NOFS.
1207 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1208 unsigned bits, unsigned clear_bits,
1209 struct extent_state **cached_state)
1211 struct extent_state *state;
1212 struct extent_state *prealloc = NULL;
1213 struct rb_node *node;
1215 struct rb_node *parent;
1219 bool first_iteration = true;
1221 btrfs_debug_check_extent_io_range(tree, start, end);
1222 trace_btrfs_convert_extent_bit(tree, start, end - start + 1, bits,
1228 * Best effort, don't worry if extent state allocation fails
1229 * here for the first iteration. We might have a cached state
1230 * that matches exactly the target range, in which case no
1231 * extent state allocations are needed. We'll only know this
1232 * after locking the tree.
1234 prealloc = alloc_extent_state(GFP_NOFS);
1235 if (!prealloc && !first_iteration)
1239 spin_lock(&tree->lock);
1240 if (cached_state && *cached_state) {
1241 state = *cached_state;
1242 if (state->start <= start && state->end > start &&
1243 extent_state_in_tree(state)) {
1244 node = &state->rb_node;
1250 * this search will find all the extents that end after
1253 node = tree_search_for_insert(tree, start, &p, &parent);
1255 prealloc = alloc_extent_state_atomic(prealloc);
1260 err = insert_state(tree, prealloc, start, end,
1261 &p, &parent, &bits, NULL);
1263 extent_io_tree_panic(tree, err);
1264 cache_state(prealloc, cached_state);
1268 state = rb_entry(node, struct extent_state, rb_node);
1270 last_start = state->start;
1271 last_end = state->end;
1274 * | ---- desired range ---- |
1277 * Just lock what we found and keep going
1279 if (state->start == start && state->end <= end) {
1280 set_state_bits(tree, state, &bits, NULL);
1281 cache_state(state, cached_state);
1282 state = clear_state_bit(tree, state, &clear_bits, 0, NULL);
1283 if (last_end == (u64)-1)
1285 start = last_end + 1;
1286 if (start < end && state && state->start == start &&
1293 * | ---- desired range ---- |
1296 * | ------------- state -------------- |
1298 * We need to split the extent we found, and may flip bits on
1301 * If the extent we found extends past our
1302 * range, we just split and search again. It'll get split
1303 * again the next time though.
1305 * If the extent we found is inside our range, we set the
1306 * desired bit on it.
1308 if (state->start < start) {
1309 prealloc = alloc_extent_state_atomic(prealloc);
1314 err = split_state(tree, state, prealloc, start);
1316 extent_io_tree_panic(tree, err);
1320 if (state->end <= end) {
1321 set_state_bits(tree, state, &bits, NULL);
1322 cache_state(state, cached_state);
1323 state = clear_state_bit(tree, state, &clear_bits, 0,
1325 if (last_end == (u64)-1)
1327 start = last_end + 1;
1328 if (start < end && state && state->start == start &&
1335 * | ---- desired range ---- |
1336 * | state | or | state |
1338 * There's a hole, we need to insert something in it and
1339 * ignore the extent we found.
1341 if (state->start > start) {
1343 if (end < last_start)
1346 this_end = last_start - 1;
1348 prealloc = alloc_extent_state_atomic(prealloc);
1355 * Avoid to free 'prealloc' if it can be merged with
1358 err = insert_state(tree, prealloc, start, this_end,
1359 NULL, NULL, &bits, NULL);
1361 extent_io_tree_panic(tree, err);
1362 cache_state(prealloc, cached_state);
1364 start = this_end + 1;
1368 * | ---- desired range ---- |
1370 * We need to split the extent, and set the bit
1373 if (state->start <= end && state->end > end) {
1374 prealloc = alloc_extent_state_atomic(prealloc);
1380 err = split_state(tree, state, prealloc, end + 1);
1382 extent_io_tree_panic(tree, err);
1384 set_state_bits(tree, prealloc, &bits, NULL);
1385 cache_state(prealloc, cached_state);
1386 clear_state_bit(tree, prealloc, &clear_bits, 0, NULL);
1394 spin_unlock(&tree->lock);
1396 first_iteration = false;
1400 spin_unlock(&tree->lock);
1402 free_extent_state(prealloc);
1407 /* wrappers around set/clear extent bit */
1408 int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1409 unsigned bits, struct extent_changeset *changeset)
1412 * We don't support EXTENT_LOCKED yet, as current changeset will
1413 * record any bits changed, so for EXTENT_LOCKED case, it will
1414 * either fail with -EEXIST or changeset will record the whole
1417 BUG_ON(bits & EXTENT_LOCKED);
1419 return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL, GFP_NOFS,
1423 int set_extent_bits_nowait(struct extent_io_tree *tree, u64 start, u64 end,
1426 return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL,
1430 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1431 unsigned bits, int wake, int delete,
1432 struct extent_state **cached)
1434 return __clear_extent_bit(tree, start, end, bits, wake, delete,
1435 cached, GFP_NOFS, NULL);
1438 int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1439 unsigned bits, struct extent_changeset *changeset)
1442 * Don't support EXTENT_LOCKED case, same reason as
1443 * set_record_extent_bits().
1445 BUG_ON(bits & EXTENT_LOCKED);
1447 return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, GFP_NOFS,
1452 * either insert or lock state struct between start and end use mask to tell
1453 * us if waiting is desired.
1455 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1456 struct extent_state **cached_state)
1462 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED,
1463 EXTENT_LOCKED, &failed_start,
1464 cached_state, GFP_NOFS, NULL);
1465 if (err == -EEXIST) {
1466 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1467 start = failed_start;
1470 WARN_ON(start > end);
1475 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1480 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1481 &failed_start, NULL, GFP_NOFS, NULL);
1482 if (err == -EEXIST) {
1483 if (failed_start > start)
1484 clear_extent_bit(tree, start, failed_start - 1,
1485 EXTENT_LOCKED, 1, 0, NULL);
1491 void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
1493 unsigned long index = start >> PAGE_SHIFT;
1494 unsigned long end_index = end >> PAGE_SHIFT;
1497 while (index <= end_index) {
1498 page = find_get_page(inode->i_mapping, index);
1499 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1500 clear_page_dirty_for_io(page);
1506 void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
1508 unsigned long index = start >> PAGE_SHIFT;
1509 unsigned long end_index = end >> PAGE_SHIFT;
1512 while (index <= end_index) {
1513 page = find_get_page(inode->i_mapping, index);
1514 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1515 __set_page_dirty_nobuffers(page);
1516 account_page_redirty(page);
1522 /* find the first state struct with 'bits' set after 'start', and
1523 * return it. tree->lock must be held. NULL will returned if
1524 * nothing was found after 'start'
1526 static struct extent_state *
1527 find_first_extent_bit_state(struct extent_io_tree *tree,
1528 u64 start, unsigned bits)
1530 struct rb_node *node;
1531 struct extent_state *state;
1534 * this search will find all the extents that end after
1537 node = tree_search(tree, start);
1542 state = rb_entry(node, struct extent_state, rb_node);
1543 if (state->end >= start && (state->state & bits))
1546 node = rb_next(node);
1555 * find the first offset in the io tree with 'bits' set. zero is
1556 * returned if we find something, and *start_ret and *end_ret are
1557 * set to reflect the state struct that was found.
1559 * If nothing was found, 1 is returned. If found something, return 0.
1561 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1562 u64 *start_ret, u64 *end_ret, unsigned bits,
1563 struct extent_state **cached_state)
1565 struct extent_state *state;
1568 spin_lock(&tree->lock);
1569 if (cached_state && *cached_state) {
1570 state = *cached_state;
1571 if (state->end == start - 1 && extent_state_in_tree(state)) {
1572 while ((state = next_state(state)) != NULL) {
1573 if (state->state & bits)
1576 free_extent_state(*cached_state);
1577 *cached_state = NULL;
1580 free_extent_state(*cached_state);
1581 *cached_state = NULL;
1584 state = find_first_extent_bit_state(tree, start, bits);
1587 cache_state_if_flags(state, cached_state, 0);
1588 *start_ret = state->start;
1589 *end_ret = state->end;
1593 spin_unlock(&tree->lock);
1598 * find_contiguous_extent_bit: find a contiguous area of bits
1599 * @tree - io tree to check
1600 * @start - offset to start the search from
1601 * @start_ret - the first offset we found with the bits set
1602 * @end_ret - the final contiguous range of the bits that were set
1603 * @bits - bits to look for
1605 * set_extent_bit and clear_extent_bit can temporarily split contiguous ranges
1606 * to set bits appropriately, and then merge them again. During this time it
1607 * will drop the tree->lock, so use this helper if you want to find the actual
1608 * contiguous area for given bits. We will search to the first bit we find, and
1609 * then walk down the tree until we find a non-contiguous area. The area
1610 * returned will be the full contiguous area with the bits set.
1612 int find_contiguous_extent_bit(struct extent_io_tree *tree, u64 start,
1613 u64 *start_ret, u64 *end_ret, unsigned bits)
1615 struct extent_state *state;
1618 spin_lock(&tree->lock);
1619 state = find_first_extent_bit_state(tree, start, bits);
1621 *start_ret = state->start;
1622 *end_ret = state->end;
1623 while ((state = next_state(state)) != NULL) {
1624 if (state->start > (*end_ret + 1))
1626 *end_ret = state->end;
1630 spin_unlock(&tree->lock);
1635 * find_first_clear_extent_bit - find the first range that has @bits not set.
1636 * This range could start before @start.
1638 * @tree - the tree to search
1639 * @start - the offset at/after which the found extent should start
1640 * @start_ret - records the beginning of the range
1641 * @end_ret - records the end of the range (inclusive)
1642 * @bits - the set of bits which must be unset
1644 * Since unallocated range is also considered one which doesn't have the bits
1645 * set it's possible that @end_ret contains -1, this happens in case the range
1646 * spans (last_range_end, end of device]. In this case it's up to the caller to
1647 * trim @end_ret to the appropriate size.
1649 void find_first_clear_extent_bit(struct extent_io_tree *tree, u64 start,
1650 u64 *start_ret, u64 *end_ret, unsigned bits)
1652 struct extent_state *state;
1653 struct rb_node *node, *prev = NULL, *next;
1655 spin_lock(&tree->lock);
1657 /* Find first extent with bits cleared */
1659 node = __etree_search(tree, start, &next, &prev, NULL, NULL);
1660 if (!node && !next && !prev) {
1662 * Tree is completely empty, send full range and let
1663 * caller deal with it
1668 } else if (!node && !next) {
1670 * We are past the last allocated chunk, set start at
1671 * the end of the last extent.
1673 state = rb_entry(prev, struct extent_state, rb_node);
1674 *start_ret = state->end + 1;
1681 * At this point 'node' either contains 'start' or start is
1684 state = rb_entry(node, struct extent_state, rb_node);
1686 if (in_range(start, state->start, state->end - state->start + 1)) {
1687 if (state->state & bits) {
1689 * |--range with bits sets--|
1693 start = state->end + 1;
1696 * 'start' falls within a range that doesn't
1697 * have the bits set, so take its start as
1698 * the beginning of the desired range
1700 * |--range with bits cleared----|
1704 *start_ret = state->start;
1709 * |---prev range---|---hole/unset---|---node range---|
1715 * |---hole/unset--||--first node--|
1720 state = rb_entry(prev, struct extent_state,
1722 *start_ret = state->end + 1;
1731 * Find the longest stretch from start until an entry which has the
1735 state = rb_entry(node, struct extent_state, rb_node);
1736 if (state->end >= start && !(state->state & bits)) {
1737 *end_ret = state->end;
1739 *end_ret = state->start - 1;
1743 node = rb_next(node);
1748 spin_unlock(&tree->lock);
1752 * find a contiguous range of bytes in the file marked as delalloc, not
1753 * more than 'max_bytes'. start and end are used to return the range,
1755 * true is returned if we find something, false if nothing was in the tree
1757 bool btrfs_find_delalloc_range(struct extent_io_tree *tree, u64 *start,
1758 u64 *end, u64 max_bytes,
1759 struct extent_state **cached_state)
1761 struct rb_node *node;
1762 struct extent_state *state;
1763 u64 cur_start = *start;
1765 u64 total_bytes = 0;
1767 spin_lock(&tree->lock);
1770 * this search will find all the extents that end after
1773 node = tree_search(tree, cur_start);
1780 state = rb_entry(node, struct extent_state, rb_node);
1781 if (found && (state->start != cur_start ||
1782 (state->state & EXTENT_BOUNDARY))) {
1785 if (!(state->state & EXTENT_DELALLOC)) {
1791 *start = state->start;
1792 *cached_state = state;
1793 refcount_inc(&state->refs);
1797 cur_start = state->end + 1;
1798 node = rb_next(node);
1799 total_bytes += state->end - state->start + 1;
1800 if (total_bytes >= max_bytes)
1806 spin_unlock(&tree->lock);
1810 static int __process_pages_contig(struct address_space *mapping,
1811 struct page *locked_page,
1812 pgoff_t start_index, pgoff_t end_index,
1813 unsigned long page_ops, pgoff_t *index_ret);
1815 static noinline void __unlock_for_delalloc(struct inode *inode,
1816 struct page *locked_page,
1819 unsigned long index = start >> PAGE_SHIFT;
1820 unsigned long end_index = end >> PAGE_SHIFT;
1822 ASSERT(locked_page);
1823 if (index == locked_page->index && end_index == index)
1826 __process_pages_contig(inode->i_mapping, locked_page, index, end_index,
1830 static noinline int lock_delalloc_pages(struct inode *inode,
1831 struct page *locked_page,
1835 unsigned long index = delalloc_start >> PAGE_SHIFT;
1836 unsigned long index_ret = index;
1837 unsigned long end_index = delalloc_end >> PAGE_SHIFT;
1840 ASSERT(locked_page);
1841 if (index == locked_page->index && index == end_index)
1844 ret = __process_pages_contig(inode->i_mapping, locked_page, index,
1845 end_index, PAGE_LOCK, &index_ret);
1847 __unlock_for_delalloc(inode, locked_page, delalloc_start,
1848 (u64)index_ret << PAGE_SHIFT);
1853 * Find and lock a contiguous range of bytes in the file marked as delalloc, no
1854 * more than @max_bytes. @Start and @end are used to return the range,
1856 * Return: true if we find something
1857 * false if nothing was in the tree
1860 noinline_for_stack bool find_lock_delalloc_range(struct inode *inode,
1861 struct page *locked_page, u64 *start,
1864 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
1865 u64 max_bytes = BTRFS_MAX_EXTENT_SIZE;
1869 struct extent_state *cached_state = NULL;
1874 /* step one, find a bunch of delalloc bytes starting at start */
1875 delalloc_start = *start;
1877 found = btrfs_find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1878 max_bytes, &cached_state);
1879 if (!found || delalloc_end <= *start) {
1880 *start = delalloc_start;
1881 *end = delalloc_end;
1882 free_extent_state(cached_state);
1887 * start comes from the offset of locked_page. We have to lock
1888 * pages in order, so we can't process delalloc bytes before
1891 if (delalloc_start < *start)
1892 delalloc_start = *start;
1895 * make sure to limit the number of pages we try to lock down
1897 if (delalloc_end + 1 - delalloc_start > max_bytes)
1898 delalloc_end = delalloc_start + max_bytes - 1;
1900 /* step two, lock all the pages after the page that has start */
1901 ret = lock_delalloc_pages(inode, locked_page,
1902 delalloc_start, delalloc_end);
1903 ASSERT(!ret || ret == -EAGAIN);
1904 if (ret == -EAGAIN) {
1905 /* some of the pages are gone, lets avoid looping by
1906 * shortening the size of the delalloc range we're searching
1908 free_extent_state(cached_state);
1909 cached_state = NULL;
1911 max_bytes = PAGE_SIZE;
1920 /* step three, lock the state bits for the whole range */
1921 lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state);
1923 /* then test to make sure it is all still delalloc */
1924 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1925 EXTENT_DELALLOC, 1, cached_state);
1927 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1929 __unlock_for_delalloc(inode, locked_page,
1930 delalloc_start, delalloc_end);
1934 free_extent_state(cached_state);
1935 *start = delalloc_start;
1936 *end = delalloc_end;
1941 static int __process_pages_contig(struct address_space *mapping,
1942 struct page *locked_page,
1943 pgoff_t start_index, pgoff_t end_index,
1944 unsigned long page_ops, pgoff_t *index_ret)
1946 unsigned long nr_pages = end_index - start_index + 1;
1947 unsigned long pages_locked = 0;
1948 pgoff_t index = start_index;
1949 struct page *pages[16];
1954 if (page_ops & PAGE_LOCK) {
1955 ASSERT(page_ops == PAGE_LOCK);
1956 ASSERT(index_ret && *index_ret == start_index);
1959 if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0)
1960 mapping_set_error(mapping, -EIO);
1962 while (nr_pages > 0) {
1963 ret = find_get_pages_contig(mapping, index,
1964 min_t(unsigned long,
1965 nr_pages, ARRAY_SIZE(pages)), pages);
1968 * Only if we're going to lock these pages,
1969 * can we find nothing at @index.
1971 ASSERT(page_ops & PAGE_LOCK);
1976 for (i = 0; i < ret; i++) {
1977 if (page_ops & PAGE_SET_PRIVATE2)
1978 SetPagePrivate2(pages[i]);
1980 if (locked_page && pages[i] == locked_page) {
1985 if (page_ops & PAGE_CLEAR_DIRTY)
1986 clear_page_dirty_for_io(pages[i]);
1987 if (page_ops & PAGE_SET_WRITEBACK)
1988 set_page_writeback(pages[i]);
1989 if (page_ops & PAGE_SET_ERROR)
1990 SetPageError(pages[i]);
1991 if (page_ops & PAGE_END_WRITEBACK)
1992 end_page_writeback(pages[i]);
1993 if (page_ops & PAGE_UNLOCK)
1994 unlock_page(pages[i]);
1995 if (page_ops & PAGE_LOCK) {
1996 lock_page(pages[i]);
1997 if (!PageDirty(pages[i]) ||
1998 pages[i]->mapping != mapping) {
1999 unlock_page(pages[i]);
2000 for (; i < ret; i++)
2014 if (err && index_ret)
2015 *index_ret = start_index + pages_locked - 1;
2019 void extent_clear_unlock_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2020 struct page *locked_page,
2021 unsigned clear_bits,
2022 unsigned long page_ops)
2024 clear_extent_bit(&inode->io_tree, start, end, clear_bits, 1, 0, NULL);
2026 __process_pages_contig(inode->vfs_inode.i_mapping, locked_page,
2027 start >> PAGE_SHIFT, end >> PAGE_SHIFT,
2032 * count the number of bytes in the tree that have a given bit(s)
2033 * set. This can be fairly slow, except for EXTENT_DIRTY which is
2034 * cached. The total number found is returned.
2036 u64 count_range_bits(struct extent_io_tree *tree,
2037 u64 *start, u64 search_end, u64 max_bytes,
2038 unsigned bits, int contig)
2040 struct rb_node *node;
2041 struct extent_state *state;
2042 u64 cur_start = *start;
2043 u64 total_bytes = 0;
2047 if (WARN_ON(search_end <= cur_start))
2050 spin_lock(&tree->lock);
2051 if (cur_start == 0 && bits == EXTENT_DIRTY) {
2052 total_bytes = tree->dirty_bytes;
2056 * this search will find all the extents that end after
2059 node = tree_search(tree, cur_start);
2064 state = rb_entry(node, struct extent_state, rb_node);
2065 if (state->start > search_end)
2067 if (contig && found && state->start > last + 1)
2069 if (state->end >= cur_start && (state->state & bits) == bits) {
2070 total_bytes += min(search_end, state->end) + 1 -
2071 max(cur_start, state->start);
2072 if (total_bytes >= max_bytes)
2075 *start = max(cur_start, state->start);
2079 } else if (contig && found) {
2082 node = rb_next(node);
2087 spin_unlock(&tree->lock);
2092 * set the private field for a given byte offset in the tree. If there isn't
2093 * an extent_state there already, this does nothing.
2095 int set_state_failrec(struct extent_io_tree *tree, u64 start,
2096 struct io_failure_record *failrec)
2098 struct rb_node *node;
2099 struct extent_state *state;
2102 spin_lock(&tree->lock);
2104 * this search will find all the extents that end after
2107 node = tree_search(tree, start);
2112 state = rb_entry(node, struct extent_state, rb_node);
2113 if (state->start != start) {
2117 state->failrec = failrec;
2119 spin_unlock(&tree->lock);
2123 struct io_failure_record *get_state_failrec(struct extent_io_tree *tree, u64 start)
2125 struct rb_node *node;
2126 struct extent_state *state;
2127 struct io_failure_record *failrec;
2129 spin_lock(&tree->lock);
2131 * this search will find all the extents that end after
2134 node = tree_search(tree, start);
2136 failrec = ERR_PTR(-ENOENT);
2139 state = rb_entry(node, struct extent_state, rb_node);
2140 if (state->start != start) {
2141 failrec = ERR_PTR(-ENOENT);
2145 failrec = state->failrec;
2147 spin_unlock(&tree->lock);
2152 * searches a range in the state tree for a given mask.
2153 * If 'filled' == 1, this returns 1 only if every extent in the tree
2154 * has the bits set. Otherwise, 1 is returned if any bit in the
2155 * range is found set.
2157 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
2158 unsigned bits, int filled, struct extent_state *cached)
2160 struct extent_state *state = NULL;
2161 struct rb_node *node;
2164 spin_lock(&tree->lock);
2165 if (cached && extent_state_in_tree(cached) && cached->start <= start &&
2166 cached->end > start)
2167 node = &cached->rb_node;
2169 node = tree_search(tree, start);
2170 while (node && start <= end) {
2171 state = rb_entry(node, struct extent_state, rb_node);
2173 if (filled && state->start > start) {
2178 if (state->start > end)
2181 if (state->state & bits) {
2185 } else if (filled) {
2190 if (state->end == (u64)-1)
2193 start = state->end + 1;
2196 node = rb_next(node);
2203 spin_unlock(&tree->lock);
2208 * helper function to set a given page up to date if all the
2209 * extents in the tree for that page are up to date
2211 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
2213 u64 start = page_offset(page);
2214 u64 end = start + PAGE_SIZE - 1;
2215 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
2216 SetPageUptodate(page);
2219 int free_io_failure(struct extent_io_tree *failure_tree,
2220 struct extent_io_tree *io_tree,
2221 struct io_failure_record *rec)
2226 set_state_failrec(failure_tree, rec->start, NULL);
2227 ret = clear_extent_bits(failure_tree, rec->start,
2228 rec->start + rec->len - 1,
2229 EXTENT_LOCKED | EXTENT_DIRTY);
2233 ret = clear_extent_bits(io_tree, rec->start,
2234 rec->start + rec->len - 1,
2244 * this bypasses the standard btrfs submit functions deliberately, as
2245 * the standard behavior is to write all copies in a raid setup. here we only
2246 * want to write the one bad copy. so we do the mapping for ourselves and issue
2247 * submit_bio directly.
2248 * to avoid any synchronization issues, wait for the data after writing, which
2249 * actually prevents the read that triggered the error from finishing.
2250 * currently, there can be no more than two copies of every data bit. thus,
2251 * exactly one rewrite is required.
2253 int repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start,
2254 u64 length, u64 logical, struct page *page,
2255 unsigned int pg_offset, int mirror_num)
2258 struct btrfs_device *dev;
2261 struct btrfs_bio *bbio = NULL;
2264 ASSERT(!(fs_info->sb->s_flags & SB_RDONLY));
2265 BUG_ON(!mirror_num);
2267 bio = btrfs_io_bio_alloc(1);
2268 bio->bi_iter.bi_size = 0;
2269 map_length = length;
2272 * Avoid races with device replace and make sure our bbio has devices
2273 * associated to its stripes that don't go away while we are doing the
2274 * read repair operation.
2276 btrfs_bio_counter_inc_blocked(fs_info);
2277 if (btrfs_is_parity_mirror(fs_info, logical, length)) {
2279 * Note that we don't use BTRFS_MAP_WRITE because it's supposed
2280 * to update all raid stripes, but here we just want to correct
2281 * bad stripe, thus BTRFS_MAP_READ is abused to only get the bad
2282 * stripe's dev and sector.
2284 ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, logical,
2285 &map_length, &bbio, 0);
2287 btrfs_bio_counter_dec(fs_info);
2291 ASSERT(bbio->mirror_num == 1);
2293 ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical,
2294 &map_length, &bbio, mirror_num);
2296 btrfs_bio_counter_dec(fs_info);
2300 BUG_ON(mirror_num != bbio->mirror_num);
2303 sector = bbio->stripes[bbio->mirror_num - 1].physical >> 9;
2304 bio->bi_iter.bi_sector = sector;
2305 dev = bbio->stripes[bbio->mirror_num - 1].dev;
2306 btrfs_put_bbio(bbio);
2307 if (!dev || !dev->bdev ||
2308 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) {
2309 btrfs_bio_counter_dec(fs_info);
2313 bio_set_dev(bio, dev->bdev);
2314 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC;
2315 bio_add_page(bio, page, length, pg_offset);
2317 if (btrfsic_submit_bio_wait(bio)) {
2318 /* try to remap that extent elsewhere? */
2319 btrfs_bio_counter_dec(fs_info);
2321 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
2325 btrfs_info_rl_in_rcu(fs_info,
2326 "read error corrected: ino %llu off %llu (dev %s sector %llu)",
2328 rcu_str_deref(dev->name), sector);
2329 btrfs_bio_counter_dec(fs_info);
2334 int btrfs_repair_eb_io_failure(const struct extent_buffer *eb, int mirror_num)
2336 struct btrfs_fs_info *fs_info = eb->fs_info;
2337 u64 start = eb->start;
2338 int i, num_pages = num_extent_pages(eb);
2341 if (sb_rdonly(fs_info->sb))
2344 for (i = 0; i < num_pages; i++) {
2345 struct page *p = eb->pages[i];
2347 ret = repair_io_failure(fs_info, 0, start, PAGE_SIZE, start, p,
2348 start - page_offset(p), mirror_num);
2358 * each time an IO finishes, we do a fast check in the IO failure tree
2359 * to see if we need to process or clean up an io_failure_record
2361 int clean_io_failure(struct btrfs_fs_info *fs_info,
2362 struct extent_io_tree *failure_tree,
2363 struct extent_io_tree *io_tree, u64 start,
2364 struct page *page, u64 ino, unsigned int pg_offset)
2367 struct io_failure_record *failrec;
2368 struct extent_state *state;
2373 ret = count_range_bits(failure_tree, &private, (u64)-1, 1,
2378 failrec = get_state_failrec(failure_tree, start);
2379 if (IS_ERR(failrec))
2382 BUG_ON(!failrec->this_mirror);
2384 if (failrec->in_validation) {
2385 /* there was no real error, just free the record */
2386 btrfs_debug(fs_info,
2387 "clean_io_failure: freeing dummy error at %llu",
2391 if (sb_rdonly(fs_info->sb))
2394 spin_lock(&io_tree->lock);
2395 state = find_first_extent_bit_state(io_tree,
2398 spin_unlock(&io_tree->lock);
2400 if (state && state->start <= failrec->start &&
2401 state->end >= failrec->start + failrec->len - 1) {
2402 num_copies = btrfs_num_copies(fs_info, failrec->logical,
2404 if (num_copies > 1) {
2405 repair_io_failure(fs_info, ino, start, failrec->len,
2406 failrec->logical, page, pg_offset,
2407 failrec->failed_mirror);
2412 free_io_failure(failure_tree, io_tree, failrec);
2418 * Can be called when
2419 * - hold extent lock
2420 * - under ordered extent
2421 * - the inode is freeing
2423 void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start, u64 end)
2425 struct extent_io_tree *failure_tree = &inode->io_failure_tree;
2426 struct io_failure_record *failrec;
2427 struct extent_state *state, *next;
2429 if (RB_EMPTY_ROOT(&failure_tree->state))
2432 spin_lock(&failure_tree->lock);
2433 state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY);
2435 if (state->start > end)
2438 ASSERT(state->end <= end);
2440 next = next_state(state);
2442 failrec = state->failrec;
2443 free_extent_state(state);
2448 spin_unlock(&failure_tree->lock);
2451 static struct io_failure_record *btrfs_get_io_failure_record(struct inode *inode,
2454 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2455 struct io_failure_record *failrec;
2456 struct extent_map *em;
2457 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2458 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2459 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2463 failrec = get_state_failrec(failure_tree, start);
2464 if (!IS_ERR(failrec)) {
2465 btrfs_debug(fs_info,
2466 "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d",
2467 failrec->logical, failrec->start, failrec->len,
2468 failrec->in_validation);
2470 * when data can be on disk more than twice, add to failrec here
2471 * (e.g. with a list for failed_mirror) to make
2472 * clean_io_failure() clean all those errors at once.
2478 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2480 return ERR_PTR(-ENOMEM);
2482 failrec->start = start;
2483 failrec->len = end - start + 1;
2484 failrec->this_mirror = 0;
2485 failrec->bio_flags = 0;
2486 failrec->in_validation = 0;
2488 read_lock(&em_tree->lock);
2489 em = lookup_extent_mapping(em_tree, start, failrec->len);
2491 read_unlock(&em_tree->lock);
2493 return ERR_PTR(-EIO);
2496 if (em->start > start || em->start + em->len <= start) {
2497 free_extent_map(em);
2500 read_unlock(&em_tree->lock);
2503 return ERR_PTR(-EIO);
2506 logical = start - em->start;
2507 logical = em->block_start + logical;
2508 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2509 logical = em->block_start;
2510 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2511 extent_set_compress_type(&failrec->bio_flags, em->compress_type);
2514 btrfs_debug(fs_info,
2515 "Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu",
2516 logical, start, failrec->len);
2518 failrec->logical = logical;
2519 free_extent_map(em);
2521 /* Set the bits in the private failure tree */
2522 ret = set_extent_bits(failure_tree, start, end,
2523 EXTENT_LOCKED | EXTENT_DIRTY);
2525 ret = set_state_failrec(failure_tree, start, failrec);
2526 /* Set the bits in the inode's tree */
2527 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED);
2528 } else if (ret < 0) {
2530 return ERR_PTR(ret);
2536 static bool btrfs_check_repairable(struct inode *inode, bool needs_validation,
2537 struct io_failure_record *failrec,
2540 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2543 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
2544 if (num_copies == 1) {
2546 * we only have a single copy of the data, so don't bother with
2547 * all the retry and error correction code that follows. no
2548 * matter what the error is, it is very likely to persist.
2550 btrfs_debug(fs_info,
2551 "Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
2552 num_copies, failrec->this_mirror, failed_mirror);
2557 * there are two premises:
2558 * a) deliver good data to the caller
2559 * b) correct the bad sectors on disk
2561 if (needs_validation) {
2563 * to fulfill b), we need to know the exact failing sectors, as
2564 * we don't want to rewrite any more than the failed ones. thus,
2565 * we need separate read requests for the failed bio
2567 * if the following BUG_ON triggers, our validation request got
2568 * merged. we need separate requests for our algorithm to work.
2570 BUG_ON(failrec->in_validation);
2571 failrec->in_validation = 1;
2572 failrec->this_mirror = failed_mirror;
2575 * we're ready to fulfill a) and b) alongside. get a good copy
2576 * of the failed sector and if we succeed, we have setup
2577 * everything for repair_io_failure to do the rest for us.
2579 if (failrec->in_validation) {
2580 BUG_ON(failrec->this_mirror != failed_mirror);
2581 failrec->in_validation = 0;
2582 failrec->this_mirror = 0;
2584 failrec->failed_mirror = failed_mirror;
2585 failrec->this_mirror++;
2586 if (failrec->this_mirror == failed_mirror)
2587 failrec->this_mirror++;
2590 if (failrec->this_mirror > num_copies) {
2591 btrfs_debug(fs_info,
2592 "Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
2593 num_copies, failrec->this_mirror, failed_mirror);
2600 static bool btrfs_io_needs_validation(struct inode *inode, struct bio *bio)
2603 const u32 blocksize = inode->i_sb->s_blocksize;
2606 * If bi_status is BLK_STS_OK, then this was a checksum error, not an
2607 * I/O error. In this case, we already know exactly which sector was
2608 * bad, so we don't need to validate.
2610 if (bio->bi_status == BLK_STS_OK)
2614 * We need to validate each sector individually if the failed I/O was
2615 * for multiple sectors.
2617 * There are a few possible bios that can end up here:
2618 * 1. A buffered read bio, which is not cloned.
2619 * 2. A direct I/O read bio, which is cloned.
2620 * 3. A (buffered or direct) repair bio, which is not cloned.
2622 * For cloned bios (case 2), we can get the size from
2623 * btrfs_io_bio->iter; for non-cloned bios (cases 1 and 3), we can get
2624 * it from the bvecs.
2626 if (bio_flagged(bio, BIO_CLONED)) {
2627 if (btrfs_io_bio(bio)->iter.bi_size > blocksize)
2630 struct bio_vec *bvec;
2633 bio_for_each_bvec_all(bvec, bio, i) {
2634 len += bvec->bv_len;
2635 if (len > blocksize)
2642 blk_status_t btrfs_submit_read_repair(struct inode *inode,
2643 struct bio *failed_bio, u64 phy_offset,
2644 struct page *page, unsigned int pgoff,
2645 u64 start, u64 end, int failed_mirror,
2646 submit_bio_hook_t *submit_bio_hook)
2648 struct io_failure_record *failrec;
2649 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2650 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2651 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2652 struct btrfs_io_bio *failed_io_bio = btrfs_io_bio(failed_bio);
2653 const int icsum = phy_offset >> inode->i_sb->s_blocksize_bits;
2654 bool need_validation;
2655 struct bio *repair_bio;
2656 struct btrfs_io_bio *repair_io_bio;
2657 blk_status_t status;
2659 btrfs_debug(fs_info,
2660 "repair read error: read error at %llu", start);
2662 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
2664 failrec = btrfs_get_io_failure_record(inode, start, end);
2665 if (IS_ERR(failrec))
2666 return errno_to_blk_status(PTR_ERR(failrec));
2668 need_validation = btrfs_io_needs_validation(inode, failed_bio);
2670 if (!btrfs_check_repairable(inode, need_validation, failrec,
2672 free_io_failure(failure_tree, tree, failrec);
2673 return BLK_STS_IOERR;
2676 repair_bio = btrfs_io_bio_alloc(1);
2677 repair_io_bio = btrfs_io_bio(repair_bio);
2678 repair_bio->bi_opf = REQ_OP_READ;
2679 if (need_validation)
2680 repair_bio->bi_opf |= REQ_FAILFAST_DEV;
2681 repair_bio->bi_end_io = failed_bio->bi_end_io;
2682 repair_bio->bi_iter.bi_sector = failrec->logical >> 9;
2683 repair_bio->bi_private = failed_bio->bi_private;
2685 if (failed_io_bio->csum) {
2686 const u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
2688 repair_io_bio->csum = repair_io_bio->csum_inline;
2689 memcpy(repair_io_bio->csum,
2690 failed_io_bio->csum + csum_size * icsum, csum_size);
2693 bio_add_page(repair_bio, page, failrec->len, pgoff);
2694 repair_io_bio->logical = failrec->start;
2695 repair_io_bio->iter = repair_bio->bi_iter;
2697 btrfs_debug(btrfs_sb(inode->i_sb),
2698 "repair read error: submitting new read to mirror %d, in_validation=%d",
2699 failrec->this_mirror, failrec->in_validation);
2701 status = submit_bio_hook(inode, repair_bio, failrec->this_mirror,
2702 failrec->bio_flags);
2704 free_io_failure(failure_tree, tree, failrec);
2705 bio_put(repair_bio);
2710 /* lots and lots of room for performance fixes in the end_bio funcs */
2712 void end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2714 int uptodate = (err == 0);
2717 btrfs_writepage_endio_finish_ordered(page, start, end, uptodate);
2720 ClearPageUptodate(page);
2722 ret = err < 0 ? err : -EIO;
2723 mapping_set_error(page->mapping, ret);
2728 * after a writepage IO is done, we need to:
2729 * clear the uptodate bits on error
2730 * clear the writeback bits in the extent tree for this IO
2731 * end_page_writeback if the page has no more pending IO
2733 * Scheduling is not allowed, so the extent state tree is expected
2734 * to have one and only one object corresponding to this IO.
2736 static void end_bio_extent_writepage(struct bio *bio)
2738 int error = blk_status_to_errno(bio->bi_status);
2739 struct bio_vec *bvec;
2742 struct bvec_iter_all iter_all;
2744 ASSERT(!bio_flagged(bio, BIO_CLONED));
2745 bio_for_each_segment_all(bvec, bio, iter_all) {
2746 struct page *page = bvec->bv_page;
2747 struct inode *inode = page->mapping->host;
2748 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2750 /* We always issue full-page reads, but if some block
2751 * in a page fails to read, blk_update_request() will
2752 * advance bv_offset and adjust bv_len to compensate.
2753 * Print a warning for nonzero offsets, and an error
2754 * if they don't add up to a full page. */
2755 if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) {
2756 if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE)
2758 "partial page write in btrfs with offset %u and length %u",
2759 bvec->bv_offset, bvec->bv_len);
2762 "incomplete page write in btrfs with offset %u and length %u",
2763 bvec->bv_offset, bvec->bv_len);
2766 start = page_offset(page);
2767 end = start + bvec->bv_offset + bvec->bv_len - 1;
2769 end_extent_writepage(page, error, start, end);
2770 end_page_writeback(page);
2777 endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len,
2780 struct extent_state *cached = NULL;
2781 u64 end = start + len - 1;
2783 if (uptodate && tree->track_uptodate)
2784 set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC);
2785 unlock_extent_cached_atomic(tree, start, end, &cached);
2789 * after a readpage IO is done, we need to:
2790 * clear the uptodate bits on error
2791 * set the uptodate bits if things worked
2792 * set the page up to date if all extents in the tree are uptodate
2793 * clear the lock bit in the extent tree
2794 * unlock the page if there are no other extents locked for it
2796 * Scheduling is not allowed, so the extent state tree is expected
2797 * to have one and only one object corresponding to this IO.
2799 static void end_bio_extent_readpage(struct bio *bio)
2801 struct bio_vec *bvec;
2802 int uptodate = !bio->bi_status;
2803 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
2804 struct extent_io_tree *tree, *failure_tree;
2809 u64 extent_start = 0;
2813 struct bvec_iter_all iter_all;
2815 ASSERT(!bio_flagged(bio, BIO_CLONED));
2816 bio_for_each_segment_all(bvec, bio, iter_all) {
2817 struct page *page = bvec->bv_page;
2818 struct inode *inode = page->mapping->host;
2819 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2821 btrfs_debug(fs_info,
2822 "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u",
2823 (u64)bio->bi_iter.bi_sector, bio->bi_status,
2824 io_bio->mirror_num);
2825 tree = &BTRFS_I(inode)->io_tree;
2826 failure_tree = &BTRFS_I(inode)->io_failure_tree;
2828 /* We always issue full-page reads, but if some block
2829 * in a page fails to read, blk_update_request() will
2830 * advance bv_offset and adjust bv_len to compensate.
2831 * Print a warning for nonzero offsets, and an error
2832 * if they don't add up to a full page. */
2833 if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) {
2834 if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE)
2836 "partial page read in btrfs with offset %u and length %u",
2837 bvec->bv_offset, bvec->bv_len);
2840 "incomplete page read in btrfs with offset %u and length %u",
2841 bvec->bv_offset, bvec->bv_len);
2844 start = page_offset(page);
2845 end = start + bvec->bv_offset + bvec->bv_len - 1;
2848 mirror = io_bio->mirror_num;
2849 if (likely(uptodate)) {
2850 if (is_data_inode(inode))
2851 ret = btrfs_verify_data_csum(io_bio, offset, page,
2852 start, end, mirror);
2854 ret = btrfs_validate_metadata_buffer(io_bio,
2855 offset, page, start, end, mirror);
2859 clean_io_failure(BTRFS_I(inode)->root->fs_info,
2860 failure_tree, tree, start,
2862 btrfs_ino(BTRFS_I(inode)), 0);
2865 if (likely(uptodate))
2868 if (is_data_inode(inode)) {
2871 * The generic bio_readpage_error handles errors the
2872 * following way: If possible, new read requests are
2873 * created and submitted and will end up in
2874 * end_bio_extent_readpage as well (if we're lucky,
2875 * not in the !uptodate case). In that case it returns
2876 * 0 and we just go on with the next page in our bio.
2877 * If it can't handle the error it will return -EIO and
2878 * we remain responsible for that page.
2880 if (!btrfs_submit_read_repair(inode, bio, offset, page,
2881 start - page_offset(page),
2883 btrfs_submit_data_bio)) {
2884 uptodate = !bio->bi_status;
2889 struct extent_buffer *eb;
2891 eb = (struct extent_buffer *)page->private;
2892 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
2893 eb->read_mirror = mirror;
2894 atomic_dec(&eb->io_pages);
2895 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD,
2897 btree_readahead_hook(eb, -EIO);
2900 if (likely(uptodate)) {
2901 loff_t i_size = i_size_read(inode);
2902 pgoff_t end_index = i_size >> PAGE_SHIFT;
2905 /* Zero out the end if this page straddles i_size */
2906 off = offset_in_page(i_size);
2907 if (page->index == end_index && off)
2908 zero_user_segment(page, off, PAGE_SIZE);
2909 SetPageUptodate(page);
2911 ClearPageUptodate(page);
2917 if (unlikely(!uptodate)) {
2919 endio_readpage_release_extent(tree,
2925 endio_readpage_release_extent(tree, start,
2926 end - start + 1, 0);
2927 } else if (!extent_len) {
2928 extent_start = start;
2929 extent_len = end + 1 - start;
2930 } else if (extent_start + extent_len == start) {
2931 extent_len += end + 1 - start;
2933 endio_readpage_release_extent(tree, extent_start,
2934 extent_len, uptodate);
2935 extent_start = start;
2936 extent_len = end + 1 - start;
2941 endio_readpage_release_extent(tree, extent_start, extent_len,
2943 btrfs_io_bio_free_csum(io_bio);
2948 * Initialize the members up to but not including 'bio'. Use after allocating a
2949 * new bio by bio_alloc_bioset as it does not initialize the bytes outside of
2950 * 'bio' because use of __GFP_ZERO is not supported.
2952 static inline void btrfs_io_bio_init(struct btrfs_io_bio *btrfs_bio)
2954 memset(btrfs_bio, 0, offsetof(struct btrfs_io_bio, bio));
2958 * The following helpers allocate a bio. As it's backed by a bioset, it'll
2959 * never fail. We're returning a bio right now but you can call btrfs_io_bio
2960 * for the appropriate container_of magic
2962 struct bio *btrfs_bio_alloc(u64 first_byte)
2966 bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, &btrfs_bioset);
2967 bio->bi_iter.bi_sector = first_byte >> 9;
2968 btrfs_io_bio_init(btrfs_io_bio(bio));
2972 struct bio *btrfs_bio_clone(struct bio *bio)
2974 struct btrfs_io_bio *btrfs_bio;
2977 /* Bio allocation backed by a bioset does not fail */
2978 new = bio_clone_fast(bio, GFP_NOFS, &btrfs_bioset);
2979 btrfs_bio = btrfs_io_bio(new);
2980 btrfs_io_bio_init(btrfs_bio);
2981 btrfs_bio->iter = bio->bi_iter;
2985 struct bio *btrfs_io_bio_alloc(unsigned int nr_iovecs)
2989 /* Bio allocation backed by a bioset does not fail */
2990 bio = bio_alloc_bioset(GFP_NOFS, nr_iovecs, &btrfs_bioset);
2991 btrfs_io_bio_init(btrfs_io_bio(bio));
2995 struct bio *btrfs_bio_clone_partial(struct bio *orig, int offset, int size)
2998 struct btrfs_io_bio *btrfs_bio;
3000 /* this will never fail when it's backed by a bioset */
3001 bio = bio_clone_fast(orig, GFP_NOFS, &btrfs_bioset);
3004 btrfs_bio = btrfs_io_bio(bio);
3005 btrfs_io_bio_init(btrfs_bio);
3007 bio_trim(bio, offset >> 9, size >> 9);
3008 btrfs_bio->iter = bio->bi_iter;
3013 * @opf: bio REQ_OP_* and REQ_* flags as one value
3014 * @wbc: optional writeback control for io accounting
3015 * @page: page to add to the bio
3016 * @pg_offset: offset of the new bio or to check whether we are adding
3017 * a contiguous page to the previous one
3018 * @size: portion of page that we want to write
3019 * @offset: starting offset in the page
3020 * @bio_ret: must be valid pointer, newly allocated bio will be stored there
3021 * @end_io_func: end_io callback for new bio
3022 * @mirror_num: desired mirror to read/write
3023 * @prev_bio_flags: flags of previous bio to see if we can merge the current one
3024 * @bio_flags: flags of the current bio to see if we can merge them
3026 static int submit_extent_page(unsigned int opf,
3027 struct writeback_control *wbc,
3028 struct page *page, u64 offset,
3029 size_t size, unsigned long pg_offset,
3030 struct bio **bio_ret,
3031 bio_end_io_t end_io_func,
3033 unsigned long prev_bio_flags,
3034 unsigned long bio_flags,
3035 bool force_bio_submit)
3039 size_t page_size = min_t(size_t, size, PAGE_SIZE);
3040 sector_t sector = offset >> 9;
3041 struct extent_io_tree *tree = &BTRFS_I(page->mapping->host)->io_tree;
3047 bool can_merge = true;
3050 if (prev_bio_flags & EXTENT_BIO_COMPRESSED)
3051 contig = bio->bi_iter.bi_sector == sector;
3053 contig = bio_end_sector(bio) == sector;
3055 if (btrfs_bio_fits_in_stripe(page, page_size, bio, bio_flags))
3058 if (prev_bio_flags != bio_flags || !contig || !can_merge ||
3060 bio_add_page(bio, page, page_size, pg_offset) < page_size) {
3061 ret = submit_one_bio(bio, mirror_num, prev_bio_flags);
3069 wbc_account_cgroup_owner(wbc, page, page_size);
3074 bio = btrfs_bio_alloc(offset);
3075 bio_add_page(bio, page, page_size, pg_offset);
3076 bio->bi_end_io = end_io_func;
3077 bio->bi_private = tree;
3078 bio->bi_write_hint = page->mapping->host->i_write_hint;
3081 struct block_device *bdev;
3083 bdev = BTRFS_I(page->mapping->host)->root->fs_info->fs_devices->latest_bdev;
3084 bio_set_dev(bio, bdev);
3085 wbc_init_bio(wbc, bio);
3086 wbc_account_cgroup_owner(wbc, page, page_size);
3094 static void attach_extent_buffer_page(struct extent_buffer *eb,
3097 if (!PagePrivate(page))
3098 attach_page_private(page, eb);
3100 WARN_ON(page->private != (unsigned long)eb);
3103 void set_page_extent_mapped(struct page *page)
3105 if (!PagePrivate(page))
3106 attach_page_private(page, (void *)EXTENT_PAGE_PRIVATE);
3109 static struct extent_map *
3110 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
3111 u64 start, u64 len, struct extent_map **em_cached)
3113 struct extent_map *em;
3115 if (em_cached && *em_cached) {
3117 if (extent_map_in_tree(em) && start >= em->start &&
3118 start < extent_map_end(em)) {
3119 refcount_inc(&em->refs);
3123 free_extent_map(em);
3127 em = btrfs_get_extent(BTRFS_I(inode), page, pg_offset, start, len);
3128 if (em_cached && !IS_ERR_OR_NULL(em)) {
3130 refcount_inc(&em->refs);
3136 * basic readpage implementation. Locked extent state structs are inserted
3137 * into the tree that are removed when the IO is done (by the end_io
3139 * XXX JDM: This needs looking at to ensure proper page locking
3140 * return 0 on success, otherwise return error
3142 int btrfs_do_readpage(struct page *page, struct extent_map **em_cached,
3143 struct bio **bio, unsigned long *bio_flags,
3144 unsigned int read_flags, u64 *prev_em_start)
3146 struct inode *inode = page->mapping->host;
3147 u64 start = page_offset(page);
3148 const u64 end = start + PAGE_SIZE - 1;
3151 u64 last_byte = i_size_read(inode);
3154 struct extent_map *em;
3157 size_t pg_offset = 0;
3159 size_t disk_io_size;
3160 size_t blocksize = inode->i_sb->s_blocksize;
3161 unsigned long this_bio_flag = 0;
3162 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
3164 set_page_extent_mapped(page);
3166 if (!PageUptodate(page)) {
3167 if (cleancache_get_page(page) == 0) {
3168 BUG_ON(blocksize != PAGE_SIZE);
3169 unlock_extent(tree, start, end);
3174 if (page->index == last_byte >> PAGE_SHIFT) {
3176 size_t zero_offset = offset_in_page(last_byte);
3179 iosize = PAGE_SIZE - zero_offset;
3180 userpage = kmap_atomic(page);
3181 memset(userpage + zero_offset, 0, iosize);
3182 flush_dcache_page(page);
3183 kunmap_atomic(userpage);
3186 while (cur <= end) {
3187 bool force_bio_submit = false;
3190 if (cur >= last_byte) {
3192 struct extent_state *cached = NULL;
3194 iosize = PAGE_SIZE - pg_offset;
3195 userpage = kmap_atomic(page);
3196 memset(userpage + pg_offset, 0, iosize);
3197 flush_dcache_page(page);
3198 kunmap_atomic(userpage);
3199 set_extent_uptodate(tree, cur, cur + iosize - 1,
3201 unlock_extent_cached(tree, cur,
3202 cur + iosize - 1, &cached);
3205 em = __get_extent_map(inode, page, pg_offset, cur,
3206 end - cur + 1, em_cached);
3207 if (IS_ERR_OR_NULL(em)) {
3209 unlock_extent(tree, cur, end);
3212 extent_offset = cur - em->start;
3213 BUG_ON(extent_map_end(em) <= cur);
3216 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
3217 this_bio_flag |= EXTENT_BIO_COMPRESSED;
3218 extent_set_compress_type(&this_bio_flag,
3222 iosize = min(extent_map_end(em) - cur, end - cur + 1);
3223 cur_end = min(extent_map_end(em) - 1, end);
3224 iosize = ALIGN(iosize, blocksize);
3225 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
3226 disk_io_size = em->block_len;
3227 offset = em->block_start;
3229 offset = em->block_start + extent_offset;
3230 disk_io_size = iosize;
3232 block_start = em->block_start;
3233 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
3234 block_start = EXTENT_MAP_HOLE;
3237 * If we have a file range that points to a compressed extent
3238 * and it's followed by a consecutive file range that points
3239 * to the same compressed extent (possibly with a different
3240 * offset and/or length, so it either points to the whole extent
3241 * or only part of it), we must make sure we do not submit a
3242 * single bio to populate the pages for the 2 ranges because
3243 * this makes the compressed extent read zero out the pages
3244 * belonging to the 2nd range. Imagine the following scenario:
3247 * [0 - 8K] [8K - 24K]
3250 * points to extent X, points to extent X,
3251 * offset 4K, length of 8K offset 0, length 16K
3253 * [extent X, compressed length = 4K uncompressed length = 16K]
3255 * If the bio to read the compressed extent covers both ranges,
3256 * it will decompress extent X into the pages belonging to the
3257 * first range and then it will stop, zeroing out the remaining
3258 * pages that belong to the other range that points to extent X.
3259 * So here we make sure we submit 2 bios, one for the first
3260 * range and another one for the third range. Both will target
3261 * the same physical extent from disk, but we can't currently
3262 * make the compressed bio endio callback populate the pages
3263 * for both ranges because each compressed bio is tightly
3264 * coupled with a single extent map, and each range can have
3265 * an extent map with a different offset value relative to the
3266 * uncompressed data of our extent and different lengths. This
3267 * is a corner case so we prioritize correctness over
3268 * non-optimal behavior (submitting 2 bios for the same extent).
3270 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) &&
3271 prev_em_start && *prev_em_start != (u64)-1 &&
3272 *prev_em_start != em->start)
3273 force_bio_submit = true;
3276 *prev_em_start = em->start;
3278 free_extent_map(em);
3281 /* we've found a hole, just zero and go on */
3282 if (block_start == EXTENT_MAP_HOLE) {
3284 struct extent_state *cached = NULL;
3286 userpage = kmap_atomic(page);
3287 memset(userpage + pg_offset, 0, iosize);
3288 flush_dcache_page(page);
3289 kunmap_atomic(userpage);
3291 set_extent_uptodate(tree, cur, cur + iosize - 1,
3293 unlock_extent_cached(tree, cur,
3294 cur + iosize - 1, &cached);
3296 pg_offset += iosize;
3299 /* the get_extent function already copied into the page */
3300 if (test_range_bit(tree, cur, cur_end,
3301 EXTENT_UPTODATE, 1, NULL)) {
3302 check_page_uptodate(tree, page);
3303 unlock_extent(tree, cur, cur + iosize - 1);
3305 pg_offset += iosize;
3308 /* we have an inline extent but it didn't get marked up
3309 * to date. Error out
3311 if (block_start == EXTENT_MAP_INLINE) {
3313 unlock_extent(tree, cur, cur + iosize - 1);
3315 pg_offset += iosize;
3319 ret = submit_extent_page(REQ_OP_READ | read_flags, NULL,
3320 page, offset, disk_io_size,
3322 end_bio_extent_readpage, 0,
3328 *bio_flags = this_bio_flag;
3331 unlock_extent(tree, cur, cur + iosize - 1);
3335 pg_offset += iosize;
3339 if (!PageError(page))
3340 SetPageUptodate(page);
3346 static inline void contiguous_readpages(struct page *pages[], int nr_pages,
3348 struct extent_map **em_cached,
3350 unsigned long *bio_flags,
3353 struct btrfs_inode *inode = BTRFS_I(pages[0]->mapping->host);
3356 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
3358 for (index = 0; index < nr_pages; index++) {
3359 btrfs_do_readpage(pages[index], em_cached, bio, bio_flags,
3360 REQ_RAHEAD, prev_em_start);
3361 put_page(pages[index]);
3365 static void update_nr_written(struct writeback_control *wbc,
3366 unsigned long nr_written)
3368 wbc->nr_to_write -= nr_written;
3372 * helper for __extent_writepage, doing all of the delayed allocation setup.
3374 * This returns 1 if btrfs_run_delalloc_range function did all the work required
3375 * to write the page (copy into inline extent). In this case the IO has
3376 * been started and the page is already unlocked.
3378 * This returns 0 if all went well (page still locked)
3379 * This returns < 0 if there were errors (page still locked)
3381 static noinline_for_stack int writepage_delalloc(struct btrfs_inode *inode,
3382 struct page *page, struct writeback_control *wbc,
3383 u64 delalloc_start, unsigned long *nr_written)
3385 u64 page_end = delalloc_start + PAGE_SIZE - 1;
3387 u64 delalloc_to_write = 0;
3388 u64 delalloc_end = 0;
3390 int page_started = 0;
3393 while (delalloc_end < page_end) {
3394 found = find_lock_delalloc_range(&inode->vfs_inode, page,
3398 delalloc_start = delalloc_end + 1;
3401 ret = btrfs_run_delalloc_range(inode, page, delalloc_start,
3402 delalloc_end, &page_started, nr_written, wbc);
3406 * btrfs_run_delalloc_range should return < 0 for error
3407 * but just in case, we use > 0 here meaning the IO is
3408 * started, so we don't want to return > 0 unless
3409 * things are going well.
3411 return ret < 0 ? ret : -EIO;
3414 * delalloc_end is already one less than the total length, so
3415 * we don't subtract one from PAGE_SIZE
3417 delalloc_to_write += (delalloc_end - delalloc_start +
3418 PAGE_SIZE) >> PAGE_SHIFT;
3419 delalloc_start = delalloc_end + 1;
3421 if (wbc->nr_to_write < delalloc_to_write) {
3424 if (delalloc_to_write < thresh * 2)
3425 thresh = delalloc_to_write;
3426 wbc->nr_to_write = min_t(u64, delalloc_to_write,
3430 /* did the fill delalloc function already unlock and start
3435 * we've unlocked the page, so we can't update
3436 * the mapping's writeback index, just update
3439 wbc->nr_to_write -= *nr_written;
3447 * helper for __extent_writepage. This calls the writepage start hooks,
3448 * and does the loop to map the page into extents and bios.
3450 * We return 1 if the IO is started and the page is unlocked,
3451 * 0 if all went well (page still locked)
3452 * < 0 if there were errors (page still locked)
3454 static noinline_for_stack int __extent_writepage_io(struct btrfs_inode *inode,
3456 struct writeback_control *wbc,
3457 struct extent_page_data *epd,
3459 unsigned long nr_written,
3462 struct extent_io_tree *tree = &inode->io_tree;
3463 u64 start = page_offset(page);
3464 u64 page_end = start + PAGE_SIZE - 1;
3470 struct extent_map *em;
3471 size_t pg_offset = 0;
3475 const unsigned int write_flags = wbc_to_write_flags(wbc);
3478 ret = btrfs_writepage_cow_fixup(page, start, page_end);
3480 /* Fixup worker will requeue */
3481 redirty_page_for_writepage(wbc, page);
3482 update_nr_written(wbc, nr_written);
3488 * we don't want to touch the inode after unlocking the page,
3489 * so we update the mapping writeback index now
3491 update_nr_written(wbc, nr_written + 1);
3494 blocksize = inode->vfs_inode.i_sb->s_blocksize;
3496 while (cur <= end) {
3500 if (cur >= i_size) {
3501 btrfs_writepage_endio_finish_ordered(page, cur,
3505 em = btrfs_get_extent(inode, NULL, 0, cur, end - cur + 1);
3506 if (IS_ERR_OR_NULL(em)) {
3508 ret = PTR_ERR_OR_ZERO(em);
3512 extent_offset = cur - em->start;
3513 em_end = extent_map_end(em);
3514 BUG_ON(em_end <= cur);
3516 iosize = min(em_end - cur, end - cur + 1);
3517 iosize = ALIGN(iosize, blocksize);
3518 offset = em->block_start + extent_offset;
3519 block_start = em->block_start;
3520 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3521 free_extent_map(em);
3525 * compressed and inline extents are written through other
3528 if (compressed || block_start == EXTENT_MAP_HOLE ||
3529 block_start == EXTENT_MAP_INLINE) {
3533 btrfs_writepage_endio_finish_ordered(page, cur,
3534 cur + iosize - 1, 1);
3536 pg_offset += iosize;
3540 btrfs_set_range_writeback(tree, cur, cur + iosize - 1);
3541 if (!PageWriteback(page)) {
3542 btrfs_err(inode->root->fs_info,
3543 "page %lu not writeback, cur %llu end %llu",
3544 page->index, cur, end);
3547 ret = submit_extent_page(REQ_OP_WRITE | write_flags, wbc,
3548 page, offset, iosize, pg_offset,
3550 end_bio_extent_writepage,
3554 if (PageWriteback(page))
3555 end_page_writeback(page);
3559 pg_offset += iosize;
3567 * the writepage semantics are similar to regular writepage. extent
3568 * records are inserted to lock ranges in the tree, and as dirty areas
3569 * are found, they are marked writeback. Then the lock bits are removed
3570 * and the end_io handler clears the writeback ranges
3572 * Return 0 if everything goes well.
3573 * Return <0 for error.
3575 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
3576 struct extent_page_data *epd)
3578 struct inode *inode = page->mapping->host;
3579 u64 start = page_offset(page);
3580 u64 page_end = start + PAGE_SIZE - 1;
3584 loff_t i_size = i_size_read(inode);
3585 unsigned long end_index = i_size >> PAGE_SHIFT;
3586 unsigned long nr_written = 0;
3588 trace___extent_writepage(page, inode, wbc);
3590 WARN_ON(!PageLocked(page));
3592 ClearPageError(page);
3594 pg_offset = offset_in_page(i_size);
3595 if (page->index > end_index ||
3596 (page->index == end_index && !pg_offset)) {
3597 page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
3602 if (page->index == end_index) {
3605 userpage = kmap_atomic(page);
3606 memset(userpage + pg_offset, 0,
3607 PAGE_SIZE - pg_offset);
3608 kunmap_atomic(userpage);
3609 flush_dcache_page(page);
3612 set_page_extent_mapped(page);
3614 if (!epd->extent_locked) {
3615 ret = writepage_delalloc(BTRFS_I(inode), page, wbc, start,
3623 ret = __extent_writepage_io(BTRFS_I(inode), page, wbc, epd, i_size,
3630 /* make sure the mapping tag for page dirty gets cleared */
3631 set_page_writeback(page);
3632 end_page_writeback(page);
3634 if (PageError(page)) {
3635 ret = ret < 0 ? ret : -EIO;
3636 end_extent_writepage(page, ret, start, page_end);
3643 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3645 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
3646 TASK_UNINTERRUPTIBLE);
3649 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3651 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3652 smp_mb__after_atomic();
3653 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3657 * Lock eb pages and flush the bio if we can't the locks
3659 * Return 0 if nothing went wrong
3660 * Return >0 is same as 0, except bio is not submitted
3661 * Return <0 if something went wrong, no page is locked
3663 static noinline_for_stack int lock_extent_buffer_for_io(struct extent_buffer *eb,
3664 struct extent_page_data *epd)
3666 struct btrfs_fs_info *fs_info = eb->fs_info;
3667 int i, num_pages, failed_page_nr;
3671 if (!btrfs_try_tree_write_lock(eb)) {
3672 ret = flush_write_bio(epd);
3676 btrfs_tree_lock(eb);
3679 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3680 btrfs_tree_unlock(eb);
3684 ret = flush_write_bio(epd);
3690 wait_on_extent_buffer_writeback(eb);
3691 btrfs_tree_lock(eb);
3692 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3694 btrfs_tree_unlock(eb);
3699 * We need to do this to prevent races in people who check if the eb is
3700 * under IO since we can end up having no IO bits set for a short period
3703 spin_lock(&eb->refs_lock);
3704 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3705 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3706 spin_unlock(&eb->refs_lock);
3707 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3708 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
3710 fs_info->dirty_metadata_batch);
3713 spin_unlock(&eb->refs_lock);
3716 btrfs_tree_unlock(eb);
3721 num_pages = num_extent_pages(eb);
3722 for (i = 0; i < num_pages; i++) {
3723 struct page *p = eb->pages[i];
3725 if (!trylock_page(p)) {
3729 err = flush_write_bio(epd);
3743 /* Unlock already locked pages */
3744 for (i = 0; i < failed_page_nr; i++)
3745 unlock_page(eb->pages[i]);
3747 * Clear EXTENT_BUFFER_WRITEBACK and wake up anyone waiting on it.
3748 * Also set back EXTENT_BUFFER_DIRTY so future attempts to this eb can
3749 * be made and undo everything done before.
3751 btrfs_tree_lock(eb);
3752 spin_lock(&eb->refs_lock);
3753 set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
3754 end_extent_buffer_writeback(eb);
3755 spin_unlock(&eb->refs_lock);
3756 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, eb->len,
3757 fs_info->dirty_metadata_batch);
3758 btrfs_clear_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3759 btrfs_tree_unlock(eb);
3763 static void set_btree_ioerr(struct page *page)
3765 struct extent_buffer *eb = (struct extent_buffer *)page->private;
3766 struct btrfs_fs_info *fs_info;
3769 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))
3773 * A read may stumble upon this buffer later, make sure that it gets an
3774 * error and knows there was an error.
3776 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3779 * If we error out, we should add back the dirty_metadata_bytes
3780 * to make it consistent.
3782 fs_info = eb->fs_info;
3783 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
3784 eb->len, fs_info->dirty_metadata_batch);
3787 * If writeback for a btree extent that doesn't belong to a log tree
3788 * failed, increment the counter transaction->eb_write_errors.
3789 * We do this because while the transaction is running and before it's
3790 * committing (when we call filemap_fdata[write|wait]_range against
3791 * the btree inode), we might have
3792 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
3793 * returns an error or an error happens during writeback, when we're
3794 * committing the transaction we wouldn't know about it, since the pages
3795 * can be no longer dirty nor marked anymore for writeback (if a
3796 * subsequent modification to the extent buffer didn't happen before the
3797 * transaction commit), which makes filemap_fdata[write|wait]_range not
3798 * able to find the pages tagged with SetPageError at transaction
3799 * commit time. So if this happens we must abort the transaction,
3800 * otherwise we commit a super block with btree roots that point to
3801 * btree nodes/leafs whose content on disk is invalid - either garbage
3802 * or the content of some node/leaf from a past generation that got
3803 * cowed or deleted and is no longer valid.
3805 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
3806 * not be enough - we need to distinguish between log tree extents vs
3807 * non-log tree extents, and the next filemap_fdatawait_range() call
3808 * will catch and clear such errors in the mapping - and that call might
3809 * be from a log sync and not from a transaction commit. Also, checking
3810 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
3811 * not done and would not be reliable - the eb might have been released
3812 * from memory and reading it back again means that flag would not be
3813 * set (since it's a runtime flag, not persisted on disk).
3815 * Using the flags below in the btree inode also makes us achieve the
3816 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
3817 * writeback for all dirty pages and before filemap_fdatawait_range()
3818 * is called, the writeback for all dirty pages had already finished
3819 * with errors - because we were not using AS_EIO/AS_ENOSPC,
3820 * filemap_fdatawait_range() would return success, as it could not know
3821 * that writeback errors happened (the pages were no longer tagged for
3824 switch (eb->log_index) {
3826 set_bit(BTRFS_FS_BTREE_ERR, &eb->fs_info->flags);
3829 set_bit(BTRFS_FS_LOG1_ERR, &eb->fs_info->flags);
3832 set_bit(BTRFS_FS_LOG2_ERR, &eb->fs_info->flags);
3835 BUG(); /* unexpected, logic error */
3839 static void end_bio_extent_buffer_writepage(struct bio *bio)
3841 struct bio_vec *bvec;
3842 struct extent_buffer *eb;
3844 struct bvec_iter_all iter_all;
3846 ASSERT(!bio_flagged(bio, BIO_CLONED));
3847 bio_for_each_segment_all(bvec, bio, iter_all) {
3848 struct page *page = bvec->bv_page;
3850 eb = (struct extent_buffer *)page->private;
3852 done = atomic_dec_and_test(&eb->io_pages);
3854 if (bio->bi_status ||
3855 test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) {
3856 ClearPageUptodate(page);
3857 set_btree_ioerr(page);
3860 end_page_writeback(page);
3865 end_extent_buffer_writeback(eb);
3871 static noinline_for_stack int write_one_eb(struct extent_buffer *eb,
3872 struct writeback_control *wbc,
3873 struct extent_page_data *epd)
3875 u64 offset = eb->start;
3878 unsigned long start, end;
3879 unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META;
3882 clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
3883 num_pages = num_extent_pages(eb);
3884 atomic_set(&eb->io_pages, num_pages);
3886 /* set btree blocks beyond nritems with 0 to avoid stale content. */
3887 nritems = btrfs_header_nritems(eb);
3888 if (btrfs_header_level(eb) > 0) {
3889 end = btrfs_node_key_ptr_offset(nritems);
3891 memzero_extent_buffer(eb, end, eb->len - end);
3895 * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0
3897 start = btrfs_item_nr_offset(nritems);
3898 end = BTRFS_LEAF_DATA_OFFSET + leaf_data_end(eb);
3899 memzero_extent_buffer(eb, start, end - start);
3902 for (i = 0; i < num_pages; i++) {
3903 struct page *p = eb->pages[i];
3905 clear_page_dirty_for_io(p);
3906 set_page_writeback(p);
3907 ret = submit_extent_page(REQ_OP_WRITE | write_flags, wbc,
3908 p, offset, PAGE_SIZE, 0,
3910 end_bio_extent_buffer_writepage,
3914 if (PageWriteback(p))
3915 end_page_writeback(p);
3916 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3917 end_extent_buffer_writeback(eb);
3921 offset += PAGE_SIZE;
3922 update_nr_written(wbc, 1);
3926 if (unlikely(ret)) {
3927 for (; i < num_pages; i++) {
3928 struct page *p = eb->pages[i];
3929 clear_page_dirty_for_io(p);
3937 int btree_write_cache_pages(struct address_space *mapping,
3938 struct writeback_control *wbc)
3940 struct extent_buffer *eb, *prev_eb = NULL;
3941 struct extent_page_data epd = {
3944 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3946 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
3949 int nr_to_write_done = 0;
3950 struct pagevec pvec;
3953 pgoff_t end; /* Inclusive */
3957 pagevec_init(&pvec);
3958 if (wbc->range_cyclic) {
3959 index = mapping->writeback_index; /* Start from prev offset */
3962 * Start from the beginning does not need to cycle over the
3963 * range, mark it as scanned.
3965 scanned = (index == 0);
3967 index = wbc->range_start >> PAGE_SHIFT;
3968 end = wbc->range_end >> PAGE_SHIFT;
3971 if (wbc->sync_mode == WB_SYNC_ALL)
3972 tag = PAGECACHE_TAG_TOWRITE;
3974 tag = PAGECACHE_TAG_DIRTY;
3976 if (wbc->sync_mode == WB_SYNC_ALL)
3977 tag_pages_for_writeback(mapping, index, end);
3978 while (!done && !nr_to_write_done && (index <= end) &&
3979 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
3983 for (i = 0; i < nr_pages; i++) {
3984 struct page *page = pvec.pages[i];
3986 if (!PagePrivate(page))
3989 spin_lock(&mapping->private_lock);
3990 if (!PagePrivate(page)) {
3991 spin_unlock(&mapping->private_lock);
3995 eb = (struct extent_buffer *)page->private;
3998 * Shouldn't happen and normally this would be a BUG_ON
3999 * but no sense in crashing the users box for something
4000 * we can survive anyway.
4003 spin_unlock(&mapping->private_lock);
4007 if (eb == prev_eb) {
4008 spin_unlock(&mapping->private_lock);
4012 ret = atomic_inc_not_zero(&eb->refs);
4013 spin_unlock(&mapping->private_lock);
4018 ret = lock_extent_buffer_for_io(eb, &epd);
4020 free_extent_buffer(eb);
4022 } else if (ret < 0) {
4024 free_extent_buffer(eb);
4028 ret = write_one_eb(eb, wbc, &epd);
4031 free_extent_buffer(eb);
4034 free_extent_buffer(eb);
4037 * The filesystem may choose to bump up nr_to_write.
4038 * We have to make sure to honor the new nr_to_write
4041 nr_to_write_done = (wbc->sync_mode == WB_SYNC_NONE &&
4042 wbc->nr_to_write <= 0);
4044 pagevec_release(&pvec);
4047 if (!scanned && !done) {
4049 * We hit the last page and there is more work to be done: wrap
4050 * back to the start of the file
4058 end_write_bio(&epd, ret);
4062 * If something went wrong, don't allow any metadata write bio to be
4065 * This would prevent use-after-free if we had dirty pages not
4066 * cleaned up, which can still happen by fuzzed images.
4069 * Allowing existing tree block to be allocated for other trees.
4071 * - Log tree operations
4072 * Exiting tree blocks get allocated to log tree, bumps its
4073 * generation, then get cleaned in tree re-balance.
4074 * Such tree block will not be written back, since it's clean,
4075 * thus no WRITTEN flag set.
4076 * And after log writes back, this tree block is not traced by
4077 * any dirty extent_io_tree.
4079 * - Offending tree block gets re-dirtied from its original owner
4080 * Since it has bumped generation, no WRITTEN flag, it can be
4081 * reused without COWing. This tree block will not be traced
4082 * by btrfs_transaction::dirty_pages.
4084 * Now such dirty tree block will not be cleaned by any dirty
4085 * extent io tree. Thus we don't want to submit such wild eb
4086 * if the fs already has error.
4088 if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
4089 ret = flush_write_bio(&epd);
4092 end_write_bio(&epd, ret);
4098 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
4099 * @mapping: address space structure to write
4100 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
4101 * @data: data passed to __extent_writepage function
4103 * If a page is already under I/O, write_cache_pages() skips it, even
4104 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
4105 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
4106 * and msync() need to guarantee that all the data which was dirty at the time
4107 * the call was made get new I/O started against them. If wbc->sync_mode is
4108 * WB_SYNC_ALL then we were called for data integrity and we must wait for
4109 * existing IO to complete.
4111 static int extent_write_cache_pages(struct address_space *mapping,
4112 struct writeback_control *wbc,
4113 struct extent_page_data *epd)
4115 struct inode *inode = mapping->host;
4118 int nr_to_write_done = 0;
4119 struct pagevec pvec;
4122 pgoff_t end; /* Inclusive */
4124 int range_whole = 0;
4129 * We have to hold onto the inode so that ordered extents can do their
4130 * work when the IO finishes. The alternative to this is failing to add
4131 * an ordered extent if the igrab() fails there and that is a huge pain
4132 * to deal with, so instead just hold onto the inode throughout the
4133 * writepages operation. If it fails here we are freeing up the inode
4134 * anyway and we'd rather not waste our time writing out stuff that is
4135 * going to be truncated anyway.
4140 pagevec_init(&pvec);
4141 if (wbc->range_cyclic) {
4142 index = mapping->writeback_index; /* Start from prev offset */
4145 * Start from the beginning does not need to cycle over the
4146 * range, mark it as scanned.
4148 scanned = (index == 0);
4150 index = wbc->range_start >> PAGE_SHIFT;
4151 end = wbc->range_end >> PAGE_SHIFT;
4152 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
4158 * We do the tagged writepage as long as the snapshot flush bit is set
4159 * and we are the first one who do the filemap_flush() on this inode.
4161 * The nr_to_write == LONG_MAX is needed to make sure other flushers do
4162 * not race in and drop the bit.
4164 if (range_whole && wbc->nr_to_write == LONG_MAX &&
4165 test_and_clear_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
4166 &BTRFS_I(inode)->runtime_flags))
4167 wbc->tagged_writepages = 1;
4169 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
4170 tag = PAGECACHE_TAG_TOWRITE;
4172 tag = PAGECACHE_TAG_DIRTY;
4174 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
4175 tag_pages_for_writeback(mapping, index, end);
4177 while (!done && !nr_to_write_done && (index <= end) &&
4178 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping,
4179 &index, end, tag))) {
4182 for (i = 0; i < nr_pages; i++) {
4183 struct page *page = pvec.pages[i];
4185 done_index = page->index + 1;
4187 * At this point we hold neither the i_pages lock nor
4188 * the page lock: the page may be truncated or
4189 * invalidated (changing page->mapping to NULL),
4190 * or even swizzled back from swapper_space to
4191 * tmpfs file mapping
4193 if (!trylock_page(page)) {
4194 ret = flush_write_bio(epd);
4199 if (unlikely(page->mapping != mapping)) {
4204 if (wbc->sync_mode != WB_SYNC_NONE) {
4205 if (PageWriteback(page)) {
4206 ret = flush_write_bio(epd);
4209 wait_on_page_writeback(page);
4212 if (PageWriteback(page) ||
4213 !clear_page_dirty_for_io(page)) {
4218 ret = __extent_writepage(page, wbc, epd);
4225 * the filesystem may choose to bump up nr_to_write.
4226 * We have to make sure to honor the new nr_to_write
4229 nr_to_write_done = wbc->nr_to_write <= 0;
4231 pagevec_release(&pvec);
4234 if (!scanned && !done) {
4236 * We hit the last page and there is more work to be done: wrap
4237 * back to the start of the file
4243 * If we're looping we could run into a page that is locked by a
4244 * writer and that writer could be waiting on writeback for a
4245 * page in our current bio, and thus deadlock, so flush the
4248 ret = flush_write_bio(epd);
4253 if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole))
4254 mapping->writeback_index = done_index;
4256 btrfs_add_delayed_iput(inode);
4260 int extent_write_full_page(struct page *page, struct writeback_control *wbc)
4263 struct extent_page_data epd = {
4266 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4269 ret = __extent_writepage(page, wbc, &epd);
4272 end_write_bio(&epd, ret);
4276 ret = flush_write_bio(&epd);
4281 int extent_write_locked_range(struct inode *inode, u64 start, u64 end,
4285 struct address_space *mapping = inode->i_mapping;
4287 unsigned long nr_pages = (end - start + PAGE_SIZE) >>
4290 struct extent_page_data epd = {
4293 .sync_io = mode == WB_SYNC_ALL,
4295 struct writeback_control wbc_writepages = {
4297 .nr_to_write = nr_pages * 2,
4298 .range_start = start,
4299 .range_end = end + 1,
4300 /* We're called from an async helper function */
4301 .punt_to_cgroup = 1,
4302 .no_cgroup_owner = 1,
4305 wbc_attach_fdatawrite_inode(&wbc_writepages, inode);
4306 while (start <= end) {
4307 page = find_get_page(mapping, start >> PAGE_SHIFT);
4308 if (clear_page_dirty_for_io(page))
4309 ret = __extent_writepage(page, &wbc_writepages, &epd);
4311 btrfs_writepage_endio_finish_ordered(page, start,
4312 start + PAGE_SIZE - 1, 1);
4321 ret = flush_write_bio(&epd);
4323 end_write_bio(&epd, ret);
4325 wbc_detach_inode(&wbc_writepages);
4329 int extent_writepages(struct address_space *mapping,
4330 struct writeback_control *wbc)
4333 struct extent_page_data epd = {
4336 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4339 ret = extent_write_cache_pages(mapping, wbc, &epd);
4342 end_write_bio(&epd, ret);
4345 ret = flush_write_bio(&epd);
4349 void extent_readahead(struct readahead_control *rac)
4351 struct bio *bio = NULL;
4352 unsigned long bio_flags = 0;
4353 struct page *pagepool[16];
4354 struct extent_map *em_cached = NULL;
4355 u64 prev_em_start = (u64)-1;
4358 while ((nr = readahead_page_batch(rac, pagepool))) {
4359 u64 contig_start = page_offset(pagepool[0]);
4360 u64 contig_end = page_offset(pagepool[nr - 1]) + PAGE_SIZE - 1;
4362 ASSERT(contig_start + nr * PAGE_SIZE - 1 == contig_end);
4364 contiguous_readpages(pagepool, nr, contig_start, contig_end,
4365 &em_cached, &bio, &bio_flags, &prev_em_start);
4369 free_extent_map(em_cached);
4372 if (submit_one_bio(bio, 0, bio_flags))
4378 * basic invalidatepage code, this waits on any locked or writeback
4379 * ranges corresponding to the page, and then deletes any extent state
4380 * records from the tree
4382 int extent_invalidatepage(struct extent_io_tree *tree,
4383 struct page *page, unsigned long offset)
4385 struct extent_state *cached_state = NULL;
4386 u64 start = page_offset(page);
4387 u64 end = start + PAGE_SIZE - 1;
4388 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
4390 start += ALIGN(offset, blocksize);
4394 lock_extent_bits(tree, start, end, &cached_state);
4395 wait_on_page_writeback(page);
4396 clear_extent_bit(tree, start, end, EXTENT_LOCKED | EXTENT_DELALLOC |
4397 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state);
4402 * a helper for releasepage, this tests for areas of the page that
4403 * are locked or under IO and drops the related state bits if it is safe
4406 static int try_release_extent_state(struct extent_io_tree *tree,
4407 struct page *page, gfp_t mask)
4409 u64 start = page_offset(page);
4410 u64 end = start + PAGE_SIZE - 1;
4413 if (test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL)) {
4417 * at this point we can safely clear everything except the
4418 * locked bit and the nodatasum bit
4420 ret = __clear_extent_bit(tree, start, end,
4421 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
4422 0, 0, NULL, mask, NULL);
4424 /* if clear_extent_bit failed for enomem reasons,
4425 * we can't allow the release to continue.
4436 * a helper for releasepage. As long as there are no locked extents
4437 * in the range corresponding to the page, both state records and extent
4438 * map records are removed
4440 int try_release_extent_mapping(struct page *page, gfp_t mask)
4442 struct extent_map *em;
4443 u64 start = page_offset(page);
4444 u64 end = start + PAGE_SIZE - 1;
4445 struct btrfs_inode *btrfs_inode = BTRFS_I(page->mapping->host);
4446 struct extent_io_tree *tree = &btrfs_inode->io_tree;
4447 struct extent_map_tree *map = &btrfs_inode->extent_tree;
4449 if (gfpflags_allow_blocking(mask) &&
4450 page->mapping->host->i_size > SZ_16M) {
4452 while (start <= end) {
4453 struct btrfs_fs_info *fs_info;
4456 len = end - start + 1;
4457 write_lock(&map->lock);
4458 em = lookup_extent_mapping(map, start, len);
4460 write_unlock(&map->lock);
4463 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
4464 em->start != start) {
4465 write_unlock(&map->lock);
4466 free_extent_map(em);
4469 if (test_range_bit(tree, em->start,
4470 extent_map_end(em) - 1,
4471 EXTENT_LOCKED, 0, NULL))
4474 * If it's not in the list of modified extents, used
4475 * by a fast fsync, we can remove it. If it's being
4476 * logged we can safely remove it since fsync took an
4477 * extra reference on the em.
4479 if (list_empty(&em->list) ||
4480 test_bit(EXTENT_FLAG_LOGGING, &em->flags))
4483 * If it's in the list of modified extents, remove it
4484 * only if its generation is older then the current one,
4485 * in which case we don't need it for a fast fsync.
4486 * Otherwise don't remove it, we could be racing with an
4487 * ongoing fast fsync that could miss the new extent.
4489 fs_info = btrfs_inode->root->fs_info;
4490 spin_lock(&fs_info->trans_lock);
4491 cur_gen = fs_info->generation;
4492 spin_unlock(&fs_info->trans_lock);
4493 if (em->generation >= cur_gen)
4497 * We only remove extent maps that are not in the list of
4498 * modified extents or that are in the list but with a
4499 * generation lower then the current generation, so there
4500 * is no need to set the full fsync flag on the inode (it
4501 * hurts the fsync performance for workloads with a data
4502 * size that exceeds or is close to the system's memory).
4504 remove_extent_mapping(map, em);
4505 /* once for the rb tree */
4506 free_extent_map(em);
4508 start = extent_map_end(em);
4509 write_unlock(&map->lock);
4512 free_extent_map(em);
4514 cond_resched(); /* Allow large-extent preemption. */
4517 return try_release_extent_state(tree, page, mask);
4521 * helper function for fiemap, which doesn't want to see any holes.
4522 * This maps until we find something past 'last'
4524 static struct extent_map *get_extent_skip_holes(struct btrfs_inode *inode,
4525 u64 offset, u64 last)
4527 u64 sectorsize = btrfs_inode_sectorsize(inode);
4528 struct extent_map *em;
4535 len = last - offset;
4538 len = ALIGN(len, sectorsize);
4539 em = btrfs_get_extent_fiemap(inode, offset, len);
4540 if (IS_ERR_OR_NULL(em))
4543 /* if this isn't a hole return it */
4544 if (em->block_start != EXTENT_MAP_HOLE)
4547 /* this is a hole, advance to the next extent */
4548 offset = extent_map_end(em);
4549 free_extent_map(em);
4557 * To cache previous fiemap extent
4559 * Will be used for merging fiemap extent
4561 struct fiemap_cache {
4570 * Helper to submit fiemap extent.
4572 * Will try to merge current fiemap extent specified by @offset, @phys,
4573 * @len and @flags with cached one.
4574 * And only when we fails to merge, cached one will be submitted as
4577 * Return value is the same as fiemap_fill_next_extent().
4579 static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
4580 struct fiemap_cache *cache,
4581 u64 offset, u64 phys, u64 len, u32 flags)
4589 * Sanity check, extent_fiemap() should have ensured that new
4590 * fiemap extent won't overlap with cached one.
4593 * NOTE: Physical address can overlap, due to compression
4595 if (cache->offset + cache->len > offset) {
4601 * Only merges fiemap extents if
4602 * 1) Their logical addresses are continuous
4604 * 2) Their physical addresses are continuous
4605 * So truly compressed (physical size smaller than logical size)
4606 * extents won't get merged with each other
4608 * 3) Share same flags except FIEMAP_EXTENT_LAST
4609 * So regular extent won't get merged with prealloc extent
4611 if (cache->offset + cache->len == offset &&
4612 cache->phys + cache->len == phys &&
4613 (cache->flags & ~FIEMAP_EXTENT_LAST) ==
4614 (flags & ~FIEMAP_EXTENT_LAST)) {
4616 cache->flags |= flags;
4617 goto try_submit_last;
4620 /* Not mergeable, need to submit cached one */
4621 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
4622 cache->len, cache->flags);
4623 cache->cached = false;
4627 cache->cached = true;
4628 cache->offset = offset;
4631 cache->flags = flags;
4633 if (cache->flags & FIEMAP_EXTENT_LAST) {
4634 ret = fiemap_fill_next_extent(fieinfo, cache->offset,
4635 cache->phys, cache->len, cache->flags);
4636 cache->cached = false;
4642 * Emit last fiemap cache
4644 * The last fiemap cache may still be cached in the following case:
4646 * |<- Fiemap range ->|
4647 * |<------------ First extent ----------->|
4649 * In this case, the first extent range will be cached but not emitted.
4650 * So we must emit it before ending extent_fiemap().
4652 static int emit_last_fiemap_cache(struct fiemap_extent_info *fieinfo,
4653 struct fiemap_cache *cache)
4660 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
4661 cache->len, cache->flags);
4662 cache->cached = false;
4668 int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,
4673 u64 max = start + len;
4677 u64 last_for_get_extent = 0;
4679 u64 isize = i_size_read(&inode->vfs_inode);
4680 struct btrfs_key found_key;
4681 struct extent_map *em = NULL;
4682 struct extent_state *cached_state = NULL;
4683 struct btrfs_path *path;
4684 struct btrfs_root *root = inode->root;
4685 struct fiemap_cache cache = { 0 };
4686 struct ulist *roots;
4687 struct ulist *tmp_ulist;
4696 path = btrfs_alloc_path();
4699 path->leave_spinning = 1;
4701 roots = ulist_alloc(GFP_KERNEL);
4702 tmp_ulist = ulist_alloc(GFP_KERNEL);
4703 if (!roots || !tmp_ulist) {
4705 goto out_free_ulist;
4709 * We can't initialize that to 'start' as this could miss extents due
4710 * to extent item merging
4713 start = round_down(start, btrfs_inode_sectorsize(inode));
4714 len = round_up(max, btrfs_inode_sectorsize(inode)) - start;
4717 * lookup the last file extent. We're not using i_size here
4718 * because there might be preallocation past i_size
4720 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode), -1,
4723 goto out_free_ulist;
4731 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
4732 found_type = found_key.type;
4734 /* No extents, but there might be delalloc bits */
4735 if (found_key.objectid != btrfs_ino(inode) ||
4736 found_type != BTRFS_EXTENT_DATA_KEY) {
4737 /* have to trust i_size as the end */
4739 last_for_get_extent = isize;
4742 * remember the start of the last extent. There are a
4743 * bunch of different factors that go into the length of the
4744 * extent, so its much less complex to remember where it started
4746 last = found_key.offset;
4747 last_for_get_extent = last + 1;
4749 btrfs_release_path(path);
4752 * we might have some extents allocated but more delalloc past those
4753 * extents. so, we trust isize unless the start of the last extent is
4758 last_for_get_extent = isize;
4761 lock_extent_bits(&inode->io_tree, start, start + len - 1,
4764 em = get_extent_skip_holes(inode, start, last_for_get_extent);
4773 u64 offset_in_extent = 0;
4775 /* break if the extent we found is outside the range */
4776 if (em->start >= max || extent_map_end(em) < off)
4780 * get_extent may return an extent that starts before our
4781 * requested range. We have to make sure the ranges
4782 * we return to fiemap always move forward and don't
4783 * overlap, so adjust the offsets here
4785 em_start = max(em->start, off);
4788 * record the offset from the start of the extent
4789 * for adjusting the disk offset below. Only do this if the
4790 * extent isn't compressed since our in ram offset may be past
4791 * what we have actually allocated on disk.
4793 if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4794 offset_in_extent = em_start - em->start;
4795 em_end = extent_map_end(em);
4796 em_len = em_end - em_start;
4798 if (em->block_start < EXTENT_MAP_LAST_BYTE)
4799 disko = em->block_start + offset_in_extent;
4804 * bump off for our next call to get_extent
4806 off = extent_map_end(em);
4810 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
4812 flags |= FIEMAP_EXTENT_LAST;
4813 } else if (em->block_start == EXTENT_MAP_INLINE) {
4814 flags |= (FIEMAP_EXTENT_DATA_INLINE |
4815 FIEMAP_EXTENT_NOT_ALIGNED);
4816 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
4817 flags |= (FIEMAP_EXTENT_DELALLOC |
4818 FIEMAP_EXTENT_UNKNOWN);
4819 } else if (fieinfo->fi_extents_max) {
4820 u64 bytenr = em->block_start -
4821 (em->start - em->orig_start);
4824 * As btrfs supports shared space, this information
4825 * can be exported to userspace tools via
4826 * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0
4827 * then we're just getting a count and we can skip the
4830 ret = btrfs_check_shared(root, btrfs_ino(inode),
4831 bytenr, roots, tmp_ulist);
4835 flags |= FIEMAP_EXTENT_SHARED;
4838 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4839 flags |= FIEMAP_EXTENT_ENCODED;
4840 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4841 flags |= FIEMAP_EXTENT_UNWRITTEN;
4843 free_extent_map(em);
4845 if ((em_start >= last) || em_len == (u64)-1 ||
4846 (last == (u64)-1 && isize <= em_end)) {
4847 flags |= FIEMAP_EXTENT_LAST;
4851 /* now scan forward to see if this is really the last extent. */
4852 em = get_extent_skip_holes(inode, off, last_for_get_extent);
4858 flags |= FIEMAP_EXTENT_LAST;
4861 ret = emit_fiemap_extent(fieinfo, &cache, em_start, disko,
4871 ret = emit_last_fiemap_cache(fieinfo, &cache);
4872 free_extent_map(em);
4874 unlock_extent_cached(&inode->io_tree, start, start + len - 1,
4878 btrfs_free_path(path);
4880 ulist_free(tmp_ulist);
4884 static void __free_extent_buffer(struct extent_buffer *eb)
4886 kmem_cache_free(extent_buffer_cache, eb);
4889 int extent_buffer_under_io(const struct extent_buffer *eb)
4891 return (atomic_read(&eb->io_pages) ||
4892 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4893 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4897 * Release all pages attached to the extent buffer.
4899 static void btrfs_release_extent_buffer_pages(struct extent_buffer *eb)
4903 int mapped = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
4905 BUG_ON(extent_buffer_under_io(eb));
4907 num_pages = num_extent_pages(eb);
4908 for (i = 0; i < num_pages; i++) {
4909 struct page *page = eb->pages[i];
4914 spin_lock(&page->mapping->private_lock);
4916 * We do this since we'll remove the pages after we've
4917 * removed the eb from the radix tree, so we could race
4918 * and have this page now attached to the new eb. So
4919 * only clear page_private if it's still connected to
4922 if (PagePrivate(page) &&
4923 page->private == (unsigned long)eb) {
4924 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4925 BUG_ON(PageDirty(page));
4926 BUG_ON(PageWriteback(page));
4928 * We need to make sure we haven't be attached
4931 detach_page_private(page);
4935 spin_unlock(&page->mapping->private_lock);
4937 /* One for when we allocated the page */
4943 * Helper for releasing the extent buffer.
4945 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4947 btrfs_release_extent_buffer_pages(eb);
4948 btrfs_leak_debug_del(&eb->fs_info->eb_leak_lock, &eb->leak_list);
4949 __free_extent_buffer(eb);
4952 static struct extent_buffer *
4953 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
4956 struct extent_buffer *eb = NULL;
4958 eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL);
4961 eb->fs_info = fs_info;
4963 rwlock_init(&eb->lock);
4964 atomic_set(&eb->blocking_readers, 0);
4965 eb->blocking_writers = 0;
4966 eb->lock_recursed = false;
4967 init_waitqueue_head(&eb->write_lock_wq);
4968 init_waitqueue_head(&eb->read_lock_wq);
4970 btrfs_leak_debug_add(&fs_info->eb_leak_lock, &eb->leak_list,
4971 &fs_info->allocated_ebs);
4973 spin_lock_init(&eb->refs_lock);
4974 atomic_set(&eb->refs, 1);
4975 atomic_set(&eb->io_pages, 0);
4978 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4980 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4981 > MAX_INLINE_EXTENT_BUFFER_SIZE);
4982 BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE);
4984 #ifdef CONFIG_BTRFS_DEBUG
4985 eb->spinning_writers = 0;
4986 atomic_set(&eb->spinning_readers, 0);
4987 atomic_set(&eb->read_locks, 0);
4988 eb->write_locks = 0;
4994 struct extent_buffer *btrfs_clone_extent_buffer(const struct extent_buffer *src)
4998 struct extent_buffer *new;
4999 int num_pages = num_extent_pages(src);
5001 new = __alloc_extent_buffer(src->fs_info, src->start, src->len);
5005 for (i = 0; i < num_pages; i++) {
5006 p = alloc_page(GFP_NOFS);
5008 btrfs_release_extent_buffer(new);
5011 attach_extent_buffer_page(new, p);
5012 WARN_ON(PageDirty(p));
5015 copy_page(page_address(p), page_address(src->pages[i]));
5018 set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
5019 set_bit(EXTENT_BUFFER_UNMAPPED, &new->bflags);
5024 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
5025 u64 start, unsigned long len)
5027 struct extent_buffer *eb;
5031 eb = __alloc_extent_buffer(fs_info, start, len);
5035 num_pages = num_extent_pages(eb);
5036 for (i = 0; i < num_pages; i++) {
5037 eb->pages[i] = alloc_page(GFP_NOFS);
5041 set_extent_buffer_uptodate(eb);
5042 btrfs_set_header_nritems(eb, 0);
5043 set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
5048 __free_page(eb->pages[i - 1]);
5049 __free_extent_buffer(eb);
5053 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
5056 return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize);
5059 static void check_buffer_tree_ref(struct extent_buffer *eb)
5063 * The TREE_REF bit is first set when the extent_buffer is added
5064 * to the radix tree. It is also reset, if unset, when a new reference
5065 * is created by find_extent_buffer.
5067 * It is only cleared in two cases: freeing the last non-tree
5068 * reference to the extent_buffer when its STALE bit is set or
5069 * calling releasepage when the tree reference is the only reference.
5071 * In both cases, care is taken to ensure that the extent_buffer's
5072 * pages are not under io. However, releasepage can be concurrently
5073 * called with creating new references, which is prone to race
5074 * conditions between the calls to check_buffer_tree_ref in those
5075 * codepaths and clearing TREE_REF in try_release_extent_buffer.
5077 * The actual lifetime of the extent_buffer in the radix tree is
5078 * adequately protected by the refcount, but the TREE_REF bit and
5079 * its corresponding reference are not. To protect against this
5080 * class of races, we call check_buffer_tree_ref from the codepaths
5081 * which trigger io after they set eb->io_pages. Note that once io is
5082 * initiated, TREE_REF can no longer be cleared, so that is the
5083 * moment at which any such race is best fixed.
5085 refs = atomic_read(&eb->refs);
5086 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5089 spin_lock(&eb->refs_lock);
5090 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5091 atomic_inc(&eb->refs);
5092 spin_unlock(&eb->refs_lock);
5095 static void mark_extent_buffer_accessed(struct extent_buffer *eb,
5096 struct page *accessed)
5100 check_buffer_tree_ref(eb);
5102 num_pages = num_extent_pages(eb);
5103 for (i = 0; i < num_pages; i++) {
5104 struct page *p = eb->pages[i];
5107 mark_page_accessed(p);
5111 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
5114 struct extent_buffer *eb;
5117 eb = radix_tree_lookup(&fs_info->buffer_radix,
5118 start >> PAGE_SHIFT);
5119 if (eb && atomic_inc_not_zero(&eb->refs)) {
5122 * Lock our eb's refs_lock to avoid races with
5123 * free_extent_buffer. When we get our eb it might be flagged
5124 * with EXTENT_BUFFER_STALE and another task running
5125 * free_extent_buffer might have seen that flag set,
5126 * eb->refs == 2, that the buffer isn't under IO (dirty and
5127 * writeback flags not set) and it's still in the tree (flag
5128 * EXTENT_BUFFER_TREE_REF set), therefore being in the process
5129 * of decrementing the extent buffer's reference count twice.
5130 * So here we could race and increment the eb's reference count,
5131 * clear its stale flag, mark it as dirty and drop our reference
5132 * before the other task finishes executing free_extent_buffer,
5133 * which would later result in an attempt to free an extent
5134 * buffer that is dirty.
5136 if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) {
5137 spin_lock(&eb->refs_lock);
5138 spin_unlock(&eb->refs_lock);
5140 mark_extent_buffer_accessed(eb, NULL);
5148 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5149 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
5152 struct extent_buffer *eb, *exists = NULL;
5155 eb = find_extent_buffer(fs_info, start);
5158 eb = alloc_dummy_extent_buffer(fs_info, start);
5160 return ERR_PTR(-ENOMEM);
5161 eb->fs_info = fs_info;
5163 ret = radix_tree_preload(GFP_NOFS);
5165 exists = ERR_PTR(ret);
5168 spin_lock(&fs_info->buffer_lock);
5169 ret = radix_tree_insert(&fs_info->buffer_radix,
5170 start >> PAGE_SHIFT, eb);
5171 spin_unlock(&fs_info->buffer_lock);
5172 radix_tree_preload_end();
5173 if (ret == -EEXIST) {
5174 exists = find_extent_buffer(fs_info, start);
5180 check_buffer_tree_ref(eb);
5181 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
5185 btrfs_release_extent_buffer(eb);
5190 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
5193 unsigned long len = fs_info->nodesize;
5196 unsigned long index = start >> PAGE_SHIFT;
5197 struct extent_buffer *eb;
5198 struct extent_buffer *exists = NULL;
5200 struct address_space *mapping = fs_info->btree_inode->i_mapping;
5204 if (!IS_ALIGNED(start, fs_info->sectorsize)) {
5205 btrfs_err(fs_info, "bad tree block start %llu", start);
5206 return ERR_PTR(-EINVAL);
5209 eb = find_extent_buffer(fs_info, start);
5213 eb = __alloc_extent_buffer(fs_info, start, len);
5215 return ERR_PTR(-ENOMEM);
5217 num_pages = num_extent_pages(eb);
5218 for (i = 0; i < num_pages; i++, index++) {
5219 p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL);
5221 exists = ERR_PTR(-ENOMEM);
5225 spin_lock(&mapping->private_lock);
5226 if (PagePrivate(p)) {
5228 * We could have already allocated an eb for this page
5229 * and attached one so lets see if we can get a ref on
5230 * the existing eb, and if we can we know it's good and
5231 * we can just return that one, else we know we can just
5232 * overwrite page->private.
5234 exists = (struct extent_buffer *)p->private;
5235 if (atomic_inc_not_zero(&exists->refs)) {
5236 spin_unlock(&mapping->private_lock);
5239 mark_extent_buffer_accessed(exists, p);
5245 * Do this so attach doesn't complain and we need to
5246 * drop the ref the old guy had.
5248 ClearPagePrivate(p);
5249 WARN_ON(PageDirty(p));
5252 attach_extent_buffer_page(eb, p);
5253 spin_unlock(&mapping->private_lock);
5254 WARN_ON(PageDirty(p));
5256 if (!PageUptodate(p))
5260 * We can't unlock the pages just yet since the extent buffer
5261 * hasn't been properly inserted in the radix tree, this
5262 * opens a race with btree_releasepage which can free a page
5263 * while we are still filling in all pages for the buffer and
5268 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5270 ret = radix_tree_preload(GFP_NOFS);
5272 exists = ERR_PTR(ret);
5276 spin_lock(&fs_info->buffer_lock);
5277 ret = radix_tree_insert(&fs_info->buffer_radix,
5278 start >> PAGE_SHIFT, eb);
5279 spin_unlock(&fs_info->buffer_lock);
5280 radix_tree_preload_end();
5281 if (ret == -EEXIST) {
5282 exists = find_extent_buffer(fs_info, start);
5288 /* add one reference for the tree */
5289 check_buffer_tree_ref(eb);
5290 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
5293 * Now it's safe to unlock the pages because any calls to
5294 * btree_releasepage will correctly detect that a page belongs to a
5295 * live buffer and won't free them prematurely.
5297 for (i = 0; i < num_pages; i++)
5298 unlock_page(eb->pages[i]);
5302 WARN_ON(!atomic_dec_and_test(&eb->refs));
5303 for (i = 0; i < num_pages; i++) {
5305 unlock_page(eb->pages[i]);
5308 btrfs_release_extent_buffer(eb);
5312 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
5314 struct extent_buffer *eb =
5315 container_of(head, struct extent_buffer, rcu_head);
5317 __free_extent_buffer(eb);
5320 static int release_extent_buffer(struct extent_buffer *eb)
5321 __releases(&eb->refs_lock)
5323 lockdep_assert_held(&eb->refs_lock);
5325 WARN_ON(atomic_read(&eb->refs) == 0);
5326 if (atomic_dec_and_test(&eb->refs)) {
5327 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
5328 struct btrfs_fs_info *fs_info = eb->fs_info;
5330 spin_unlock(&eb->refs_lock);
5332 spin_lock(&fs_info->buffer_lock);
5333 radix_tree_delete(&fs_info->buffer_radix,
5334 eb->start >> PAGE_SHIFT);
5335 spin_unlock(&fs_info->buffer_lock);
5337 spin_unlock(&eb->refs_lock);
5340 btrfs_leak_debug_del(&eb->fs_info->eb_leak_lock, &eb->leak_list);
5341 /* Should be safe to release our pages at this point */
5342 btrfs_release_extent_buffer_pages(eb);
5343 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5344 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags))) {
5345 __free_extent_buffer(eb);
5349 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
5352 spin_unlock(&eb->refs_lock);
5357 void free_extent_buffer(struct extent_buffer *eb)
5365 refs = atomic_read(&eb->refs);
5366 if ((!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && refs <= 3)
5367 || (test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) &&
5370 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
5375 spin_lock(&eb->refs_lock);
5376 if (atomic_read(&eb->refs) == 2 &&
5377 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
5378 !extent_buffer_under_io(eb) &&
5379 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5380 atomic_dec(&eb->refs);
5383 * I know this is terrible, but it's temporary until we stop tracking
5384 * the uptodate bits and such for the extent buffers.
5386 release_extent_buffer(eb);
5389 void free_extent_buffer_stale(struct extent_buffer *eb)
5394 spin_lock(&eb->refs_lock);
5395 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
5397 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
5398 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5399 atomic_dec(&eb->refs);
5400 release_extent_buffer(eb);
5403 void clear_extent_buffer_dirty(const struct extent_buffer *eb)
5409 num_pages = num_extent_pages(eb);
5411 for (i = 0; i < num_pages; i++) {
5412 page = eb->pages[i];
5413 if (!PageDirty(page))
5417 WARN_ON(!PagePrivate(page));
5419 clear_page_dirty_for_io(page);
5420 xa_lock_irq(&page->mapping->i_pages);
5421 if (!PageDirty(page))
5422 __xa_clear_mark(&page->mapping->i_pages,
5423 page_index(page), PAGECACHE_TAG_DIRTY);
5424 xa_unlock_irq(&page->mapping->i_pages);
5425 ClearPageError(page);
5428 WARN_ON(atomic_read(&eb->refs) == 0);
5431 bool set_extent_buffer_dirty(struct extent_buffer *eb)
5437 check_buffer_tree_ref(eb);
5439 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
5441 num_pages = num_extent_pages(eb);
5442 WARN_ON(atomic_read(&eb->refs) == 0);
5443 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
5446 for (i = 0; i < num_pages; i++)
5447 set_page_dirty(eb->pages[i]);
5449 #ifdef CONFIG_BTRFS_DEBUG
5450 for (i = 0; i < num_pages; i++)
5451 ASSERT(PageDirty(eb->pages[i]));
5457 void clear_extent_buffer_uptodate(struct extent_buffer *eb)
5463 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5464 num_pages = num_extent_pages(eb);
5465 for (i = 0; i < num_pages; i++) {
5466 page = eb->pages[i];
5468 ClearPageUptodate(page);
5472 void set_extent_buffer_uptodate(struct extent_buffer *eb)
5478 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5479 num_pages = num_extent_pages(eb);
5480 for (i = 0; i < num_pages; i++) {
5481 page = eb->pages[i];
5482 SetPageUptodate(page);
5486 int read_extent_buffer_pages(struct extent_buffer *eb, int wait, int mirror_num)
5492 int locked_pages = 0;
5493 int all_uptodate = 1;
5495 unsigned long num_reads = 0;
5496 struct bio *bio = NULL;
5497 unsigned long bio_flags = 0;
5499 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
5502 num_pages = num_extent_pages(eb);
5503 for (i = 0; i < num_pages; i++) {
5504 page = eb->pages[i];
5505 if (wait == WAIT_NONE) {
5506 if (!trylock_page(page))
5514 * We need to firstly lock all pages to make sure that
5515 * the uptodate bit of our pages won't be affected by
5516 * clear_extent_buffer_uptodate().
5518 for (i = 0; i < num_pages; i++) {
5519 page = eb->pages[i];
5520 if (!PageUptodate(page)) {
5527 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5531 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
5532 eb->read_mirror = 0;
5533 atomic_set(&eb->io_pages, num_reads);
5535 * It is possible for releasepage to clear the TREE_REF bit before we
5536 * set io_pages. See check_buffer_tree_ref for a more detailed comment.
5538 check_buffer_tree_ref(eb);
5539 for (i = 0; i < num_pages; i++) {
5540 page = eb->pages[i];
5542 if (!PageUptodate(page)) {
5544 atomic_dec(&eb->io_pages);
5549 ClearPageError(page);
5550 err = submit_extent_page(REQ_OP_READ | REQ_META, NULL,
5551 page, page_offset(page), PAGE_SIZE, 0,
5552 &bio, end_bio_extent_readpage,
5553 mirror_num, 0, 0, false);
5556 * We failed to submit the bio so it's the
5557 * caller's responsibility to perform cleanup
5558 * i.e unlock page/set error bit.
5563 atomic_dec(&eb->io_pages);
5571 err = submit_one_bio(bio, mirror_num, bio_flags);
5576 if (ret || wait != WAIT_COMPLETE)
5579 for (i = 0; i < num_pages; i++) {
5580 page = eb->pages[i];
5581 wait_on_page_locked(page);
5582 if (!PageUptodate(page))
5589 while (locked_pages > 0) {
5591 page = eb->pages[locked_pages];
5597 static bool report_eb_range(const struct extent_buffer *eb, unsigned long start,
5600 btrfs_warn(eb->fs_info,
5601 "access to eb bytenr %llu len %lu out of range start %lu len %lu",
5602 eb->start, eb->len, start, len);
5603 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
5609 * Check if the [start, start + len) range is valid before reading/writing
5611 * NOTE: @start and @len are offset inside the eb, not logical address.
5613 * Caller should not touch the dst/src memory if this function returns error.
5615 static inline int check_eb_range(const struct extent_buffer *eb,
5616 unsigned long start, unsigned long len)
5618 unsigned long offset;
5620 /* start, start + len should not go beyond eb->len nor overflow */
5621 if (unlikely(check_add_overflow(start, len, &offset) || offset > eb->len))
5622 return report_eb_range(eb, start, len);
5627 void read_extent_buffer(const struct extent_buffer *eb, void *dstv,
5628 unsigned long start, unsigned long len)
5634 char *dst = (char *)dstv;
5635 unsigned long i = start >> PAGE_SHIFT;
5637 if (check_eb_range(eb, start, len)) {
5639 * Invalid range hit, reset the memory, so callers won't get
5640 * some random garbage for their uninitialzed memory.
5642 memset(dstv, 0, len);
5646 offset = offset_in_page(start);
5649 page = eb->pages[i];
5651 cur = min(len, (PAGE_SIZE - offset));
5652 kaddr = page_address(page);
5653 memcpy(dst, kaddr + offset, cur);
5662 int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb,
5664 unsigned long start, unsigned long len)
5670 char __user *dst = (char __user *)dstv;
5671 unsigned long i = start >> PAGE_SHIFT;
5674 WARN_ON(start > eb->len);
5675 WARN_ON(start + len > eb->start + eb->len);
5677 offset = offset_in_page(start);
5680 page = eb->pages[i];
5682 cur = min(len, (PAGE_SIZE - offset));
5683 kaddr = page_address(page);
5684 if (copy_to_user_nofault(dst, kaddr + offset, cur)) {
5698 int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,
5699 unsigned long start, unsigned long len)
5705 char *ptr = (char *)ptrv;
5706 unsigned long i = start >> PAGE_SHIFT;
5709 if (check_eb_range(eb, start, len))
5712 offset = offset_in_page(start);
5715 page = eb->pages[i];
5717 cur = min(len, (PAGE_SIZE - offset));
5719 kaddr = page_address(page);
5720 ret = memcmp(ptr, kaddr + offset, cur);
5732 void write_extent_buffer_chunk_tree_uuid(const struct extent_buffer *eb,
5737 WARN_ON(!PageUptodate(eb->pages[0]));
5738 kaddr = page_address(eb->pages[0]);
5739 memcpy(kaddr + offsetof(struct btrfs_header, chunk_tree_uuid), srcv,
5743 void write_extent_buffer_fsid(const struct extent_buffer *eb, const void *srcv)
5747 WARN_ON(!PageUptodate(eb->pages[0]));
5748 kaddr = page_address(eb->pages[0]);
5749 memcpy(kaddr + offsetof(struct btrfs_header, fsid), srcv,
5753 void write_extent_buffer(const struct extent_buffer *eb, const void *srcv,
5754 unsigned long start, unsigned long len)
5760 char *src = (char *)srcv;
5761 unsigned long i = start >> PAGE_SHIFT;
5763 if (check_eb_range(eb, start, len))
5766 offset = offset_in_page(start);
5769 page = eb->pages[i];
5770 WARN_ON(!PageUptodate(page));
5772 cur = min(len, PAGE_SIZE - offset);
5773 kaddr = page_address(page);
5774 memcpy(kaddr + offset, src, cur);
5783 void memzero_extent_buffer(const struct extent_buffer *eb, unsigned long start,
5790 unsigned long i = start >> PAGE_SHIFT;
5792 if (check_eb_range(eb, start, len))
5795 offset = offset_in_page(start);
5798 page = eb->pages[i];
5799 WARN_ON(!PageUptodate(page));
5801 cur = min(len, PAGE_SIZE - offset);
5802 kaddr = page_address(page);
5803 memset(kaddr + offset, 0, cur);
5811 void copy_extent_buffer_full(const struct extent_buffer *dst,
5812 const struct extent_buffer *src)
5817 ASSERT(dst->len == src->len);
5819 num_pages = num_extent_pages(dst);
5820 for (i = 0; i < num_pages; i++)
5821 copy_page(page_address(dst->pages[i]),
5822 page_address(src->pages[i]));
5825 void copy_extent_buffer(const struct extent_buffer *dst,
5826 const struct extent_buffer *src,
5827 unsigned long dst_offset, unsigned long src_offset,
5830 u64 dst_len = dst->len;
5835 unsigned long i = dst_offset >> PAGE_SHIFT;
5837 if (check_eb_range(dst, dst_offset, len) ||
5838 check_eb_range(src, src_offset, len))
5841 WARN_ON(src->len != dst_len);
5843 offset = offset_in_page(dst_offset);
5846 page = dst->pages[i];
5847 WARN_ON(!PageUptodate(page));
5849 cur = min(len, (unsigned long)(PAGE_SIZE - offset));
5851 kaddr = page_address(page);
5852 read_extent_buffer(src, kaddr + offset, src_offset, cur);
5862 * eb_bitmap_offset() - calculate the page and offset of the byte containing the
5864 * @eb: the extent buffer
5865 * @start: offset of the bitmap item in the extent buffer
5867 * @page_index: return index of the page in the extent buffer that contains the
5869 * @page_offset: return offset into the page given by page_index
5871 * This helper hides the ugliness of finding the byte in an extent buffer which
5872 * contains a given bit.
5874 static inline void eb_bitmap_offset(const struct extent_buffer *eb,
5875 unsigned long start, unsigned long nr,
5876 unsigned long *page_index,
5877 size_t *page_offset)
5879 size_t byte_offset = BIT_BYTE(nr);
5883 * The byte we want is the offset of the extent buffer + the offset of
5884 * the bitmap item in the extent buffer + the offset of the byte in the
5887 offset = start + byte_offset;
5889 *page_index = offset >> PAGE_SHIFT;
5890 *page_offset = offset_in_page(offset);
5894 * extent_buffer_test_bit - determine whether a bit in a bitmap item is set
5895 * @eb: the extent buffer
5896 * @start: offset of the bitmap item in the extent buffer
5897 * @nr: bit number to test
5899 int extent_buffer_test_bit(const struct extent_buffer *eb, unsigned long start,
5907 eb_bitmap_offset(eb, start, nr, &i, &offset);
5908 page = eb->pages[i];
5909 WARN_ON(!PageUptodate(page));
5910 kaddr = page_address(page);
5911 return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1)));
5915 * extent_buffer_bitmap_set - set an area of a bitmap
5916 * @eb: the extent buffer
5917 * @start: offset of the bitmap item in the extent buffer
5918 * @pos: bit number of the first bit
5919 * @len: number of bits to set
5921 void extent_buffer_bitmap_set(const struct extent_buffer *eb, unsigned long start,
5922 unsigned long pos, unsigned long len)
5928 const unsigned int size = pos + len;
5929 int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5930 u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos);
5932 eb_bitmap_offset(eb, start, pos, &i, &offset);
5933 page = eb->pages[i];
5934 WARN_ON(!PageUptodate(page));
5935 kaddr = page_address(page);
5937 while (len >= bits_to_set) {
5938 kaddr[offset] |= mask_to_set;
5940 bits_to_set = BITS_PER_BYTE;
5942 if (++offset >= PAGE_SIZE && len > 0) {
5944 page = eb->pages[++i];
5945 WARN_ON(!PageUptodate(page));
5946 kaddr = page_address(page);
5950 mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
5951 kaddr[offset] |= mask_to_set;
5957 * extent_buffer_bitmap_clear - clear an area of a bitmap
5958 * @eb: the extent buffer
5959 * @start: offset of the bitmap item in the extent buffer
5960 * @pos: bit number of the first bit
5961 * @len: number of bits to clear
5963 void extent_buffer_bitmap_clear(const struct extent_buffer *eb,
5964 unsigned long start, unsigned long pos,
5971 const unsigned int size = pos + len;
5972 int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5973 u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos);
5975 eb_bitmap_offset(eb, start, pos, &i, &offset);
5976 page = eb->pages[i];
5977 WARN_ON(!PageUptodate(page));
5978 kaddr = page_address(page);
5980 while (len >= bits_to_clear) {
5981 kaddr[offset] &= ~mask_to_clear;
5982 len -= bits_to_clear;
5983 bits_to_clear = BITS_PER_BYTE;
5985 if (++offset >= PAGE_SIZE && len > 0) {
5987 page = eb->pages[++i];
5988 WARN_ON(!PageUptodate(page));
5989 kaddr = page_address(page);
5993 mask_to_clear &= BITMAP_LAST_BYTE_MASK(size);
5994 kaddr[offset] &= ~mask_to_clear;
5998 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
6000 unsigned long distance = (src > dst) ? src - dst : dst - src;
6001 return distance < len;
6004 static void copy_pages(struct page *dst_page, struct page *src_page,
6005 unsigned long dst_off, unsigned long src_off,
6008 char *dst_kaddr = page_address(dst_page);
6010 int must_memmove = 0;
6012 if (dst_page != src_page) {
6013 src_kaddr = page_address(src_page);
6015 src_kaddr = dst_kaddr;
6016 if (areas_overlap(src_off, dst_off, len))
6021 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
6023 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
6026 void memcpy_extent_buffer(const struct extent_buffer *dst,
6027 unsigned long dst_offset, unsigned long src_offset,
6031 size_t dst_off_in_page;
6032 size_t src_off_in_page;
6033 unsigned long dst_i;
6034 unsigned long src_i;
6036 if (check_eb_range(dst, dst_offset, len) ||
6037 check_eb_range(dst, src_offset, len))
6041 dst_off_in_page = offset_in_page(dst_offset);
6042 src_off_in_page = offset_in_page(src_offset);
6044 dst_i = dst_offset >> PAGE_SHIFT;
6045 src_i = src_offset >> PAGE_SHIFT;
6047 cur = min(len, (unsigned long)(PAGE_SIZE -
6049 cur = min_t(unsigned long, cur,
6050 (unsigned long)(PAGE_SIZE - dst_off_in_page));
6052 copy_pages(dst->pages[dst_i], dst->pages[src_i],
6053 dst_off_in_page, src_off_in_page, cur);
6061 void memmove_extent_buffer(const struct extent_buffer *dst,
6062 unsigned long dst_offset, unsigned long src_offset,
6066 size_t dst_off_in_page;
6067 size_t src_off_in_page;
6068 unsigned long dst_end = dst_offset + len - 1;
6069 unsigned long src_end = src_offset + len - 1;
6070 unsigned long dst_i;
6071 unsigned long src_i;
6073 if (check_eb_range(dst, dst_offset, len) ||
6074 check_eb_range(dst, src_offset, len))
6076 if (dst_offset < src_offset) {
6077 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
6081 dst_i = dst_end >> PAGE_SHIFT;
6082 src_i = src_end >> PAGE_SHIFT;
6084 dst_off_in_page = offset_in_page(dst_end);
6085 src_off_in_page = offset_in_page(src_end);
6087 cur = min_t(unsigned long, len, src_off_in_page + 1);
6088 cur = min(cur, dst_off_in_page + 1);
6089 copy_pages(dst->pages[dst_i], dst->pages[src_i],
6090 dst_off_in_page - cur + 1,
6091 src_off_in_page - cur + 1, cur);
6099 int try_release_extent_buffer(struct page *page)
6101 struct extent_buffer *eb;
6104 * We need to make sure nobody is attaching this page to an eb right
6107 spin_lock(&page->mapping->private_lock);
6108 if (!PagePrivate(page)) {
6109 spin_unlock(&page->mapping->private_lock);
6113 eb = (struct extent_buffer *)page->private;
6117 * This is a little awful but should be ok, we need to make sure that
6118 * the eb doesn't disappear out from under us while we're looking at
6121 spin_lock(&eb->refs_lock);
6122 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
6123 spin_unlock(&eb->refs_lock);
6124 spin_unlock(&page->mapping->private_lock);
6127 spin_unlock(&page->mapping->private_lock);
6130 * If tree ref isn't set then we know the ref on this eb is a real ref,
6131 * so just return, this page will likely be freed soon anyway.
6133 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
6134 spin_unlock(&eb->refs_lock);
6138 return release_extent_buffer(eb);