1 // SPDX-License-Identifier: GPL-2.0
3 #include <linux/bitops.h>
4 #include <linux/slab.h>
7 #include <linux/pagemap.h>
8 #include <linux/page-flags.h>
9 #include <linux/spinlock.h>
10 #include <linux/blkdev.h>
11 #include <linux/swap.h>
12 #include <linux/writeback.h>
13 #include <linux/pagevec.h>
14 #include <linux/prefetch.h>
15 #include <linux/cleancache.h>
16 #include "extent_io.h"
17 #include "extent-io-tree.h"
18 #include "extent_map.h"
20 #include "btrfs_inode.h"
22 #include "check-integrity.h"
24 #include "rcu-string.h"
28 static struct kmem_cache *extent_state_cache;
29 static struct kmem_cache *extent_buffer_cache;
30 static struct bio_set btrfs_bioset;
32 static inline bool extent_state_in_tree(const struct extent_state *state)
34 return !RB_EMPTY_NODE(&state->rb_node);
37 #ifdef CONFIG_BTRFS_DEBUG
38 static LIST_HEAD(states);
39 static DEFINE_SPINLOCK(leak_lock);
41 static inline void btrfs_leak_debug_add(spinlock_t *lock,
42 struct list_head *new,
43 struct list_head *head)
47 spin_lock_irqsave(lock, flags);
49 spin_unlock_irqrestore(lock, flags);
52 static inline void btrfs_leak_debug_del(spinlock_t *lock,
53 struct list_head *entry)
57 spin_lock_irqsave(lock, flags);
59 spin_unlock_irqrestore(lock, flags);
62 void btrfs_extent_buffer_leak_debug_check(struct btrfs_fs_info *fs_info)
64 struct extent_buffer *eb;
68 * If we didn't get into open_ctree our allocated_ebs will not be
69 * initialized, so just skip this.
71 if (!fs_info->allocated_ebs.next)
74 spin_lock_irqsave(&fs_info->eb_leak_lock, flags);
75 while (!list_empty(&fs_info->allocated_ebs)) {
76 eb = list_first_entry(&fs_info->allocated_ebs,
77 struct extent_buffer, leak_list);
79 "BTRFS: buffer leak start %llu len %lu refs %d bflags %lu owner %llu\n",
80 eb->start, eb->len, atomic_read(&eb->refs), eb->bflags,
81 btrfs_header_owner(eb));
82 list_del(&eb->leak_list);
83 kmem_cache_free(extent_buffer_cache, eb);
85 spin_unlock_irqrestore(&fs_info->eb_leak_lock, flags);
88 static inline void btrfs_extent_state_leak_debug_check(void)
90 struct extent_state *state;
92 while (!list_empty(&states)) {
93 state = list_entry(states.next, struct extent_state, leak_list);
94 pr_err("BTRFS: state leak: start %llu end %llu state %u in tree %d refs %d\n",
95 state->start, state->end, state->state,
96 extent_state_in_tree(state),
97 refcount_read(&state->refs));
98 list_del(&state->leak_list);
99 kmem_cache_free(extent_state_cache, state);
103 #define btrfs_debug_check_extent_io_range(tree, start, end) \
104 __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end))
105 static inline void __btrfs_debug_check_extent_io_range(const char *caller,
106 struct extent_io_tree *tree, u64 start, u64 end)
108 struct inode *inode = tree->private_data;
111 if (!inode || !is_data_inode(inode))
114 isize = i_size_read(inode);
115 if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) {
116 btrfs_debug_rl(BTRFS_I(inode)->root->fs_info,
117 "%s: ino %llu isize %llu odd range [%llu,%llu]",
118 caller, btrfs_ino(BTRFS_I(inode)), isize, start, end);
122 #define btrfs_leak_debug_add(lock, new, head) do {} while (0)
123 #define btrfs_leak_debug_del(lock, entry) do {} while (0)
124 #define btrfs_extent_state_leak_debug_check() do {} while (0)
125 #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0)
131 struct rb_node rb_node;
134 struct extent_page_data {
136 /* tells writepage not to lock the state bits for this range
137 * it still does the unlocking
139 unsigned int extent_locked:1;
141 /* tells the submit_bio code to use REQ_SYNC */
142 unsigned int sync_io:1;
145 static int add_extent_changeset(struct extent_state *state, unsigned bits,
146 struct extent_changeset *changeset,
153 if (set && (state->state & bits) == bits)
155 if (!set && (state->state & bits) == 0)
157 changeset->bytes_changed += state->end - state->start + 1;
158 ret = ulist_add(&changeset->range_changed, state->start, state->end,
163 int __must_check submit_one_bio(struct bio *bio, int mirror_num,
164 unsigned long bio_flags)
166 blk_status_t ret = 0;
167 struct extent_io_tree *tree = bio->bi_private;
169 bio->bi_private = NULL;
171 if (is_data_inode(tree->private_data))
172 ret = btrfs_submit_data_bio(tree->private_data, bio, mirror_num,
175 ret = btrfs_submit_metadata_bio(tree->private_data, bio,
176 mirror_num, bio_flags);
178 return blk_status_to_errno(ret);
181 /* Cleanup unsubmitted bios */
182 static void end_write_bio(struct extent_page_data *epd, int ret)
185 epd->bio->bi_status = errno_to_blk_status(ret);
192 * Submit bio from extent page data via submit_one_bio
194 * Return 0 if everything is OK.
195 * Return <0 for error.
197 static int __must_check flush_write_bio(struct extent_page_data *epd)
202 ret = submit_one_bio(epd->bio, 0, 0);
204 * Clean up of epd->bio is handled by its endio function.
205 * And endio is either triggered by successful bio execution
206 * or the error handler of submit bio hook.
207 * So at this point, no matter what happened, we don't need
208 * to clean up epd->bio.
215 int __init extent_state_cache_init(void)
217 extent_state_cache = kmem_cache_create("btrfs_extent_state",
218 sizeof(struct extent_state), 0,
219 SLAB_MEM_SPREAD, NULL);
220 if (!extent_state_cache)
225 int __init extent_io_init(void)
227 extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer",
228 sizeof(struct extent_buffer), 0,
229 SLAB_MEM_SPREAD, NULL);
230 if (!extent_buffer_cache)
233 if (bioset_init(&btrfs_bioset, BIO_POOL_SIZE,
234 offsetof(struct btrfs_io_bio, bio),
236 goto free_buffer_cache;
238 if (bioset_integrity_create(&btrfs_bioset, BIO_POOL_SIZE))
244 bioset_exit(&btrfs_bioset);
247 kmem_cache_destroy(extent_buffer_cache);
248 extent_buffer_cache = NULL;
252 void __cold extent_state_cache_exit(void)
254 btrfs_extent_state_leak_debug_check();
255 kmem_cache_destroy(extent_state_cache);
258 void __cold extent_io_exit(void)
261 * Make sure all delayed rcu free are flushed before we
265 kmem_cache_destroy(extent_buffer_cache);
266 bioset_exit(&btrfs_bioset);
270 * For the file_extent_tree, we want to hold the inode lock when we lookup and
271 * update the disk_i_size, but lockdep will complain because our io_tree we hold
272 * the tree lock and get the inode lock when setting delalloc. These two things
273 * are unrelated, so make a class for the file_extent_tree so we don't get the
274 * two locking patterns mixed up.
276 static struct lock_class_key file_extent_tree_class;
278 void extent_io_tree_init(struct btrfs_fs_info *fs_info,
279 struct extent_io_tree *tree, unsigned int owner,
282 tree->fs_info = fs_info;
283 tree->state = RB_ROOT;
284 tree->dirty_bytes = 0;
285 spin_lock_init(&tree->lock);
286 tree->private_data = private_data;
288 if (owner == IO_TREE_INODE_FILE_EXTENT)
289 lockdep_set_class(&tree->lock, &file_extent_tree_class);
292 void extent_io_tree_release(struct extent_io_tree *tree)
294 spin_lock(&tree->lock);
296 * Do a single barrier for the waitqueue_active check here, the state
297 * of the waitqueue should not change once extent_io_tree_release is
301 while (!RB_EMPTY_ROOT(&tree->state)) {
302 struct rb_node *node;
303 struct extent_state *state;
305 node = rb_first(&tree->state);
306 state = rb_entry(node, struct extent_state, rb_node);
307 rb_erase(&state->rb_node, &tree->state);
308 RB_CLEAR_NODE(&state->rb_node);
310 * btree io trees aren't supposed to have tasks waiting for
311 * changes in the flags of extent states ever.
313 ASSERT(!waitqueue_active(&state->wq));
314 free_extent_state(state);
316 cond_resched_lock(&tree->lock);
318 spin_unlock(&tree->lock);
321 static struct extent_state *alloc_extent_state(gfp_t mask)
323 struct extent_state *state;
326 * The given mask might be not appropriate for the slab allocator,
327 * drop the unsupported bits
329 mask &= ~(__GFP_DMA32|__GFP_HIGHMEM);
330 state = kmem_cache_alloc(extent_state_cache, mask);
334 state->failrec = NULL;
335 RB_CLEAR_NODE(&state->rb_node);
336 btrfs_leak_debug_add(&leak_lock, &state->leak_list, &states);
337 refcount_set(&state->refs, 1);
338 init_waitqueue_head(&state->wq);
339 trace_alloc_extent_state(state, mask, _RET_IP_);
343 void free_extent_state(struct extent_state *state)
347 if (refcount_dec_and_test(&state->refs)) {
348 WARN_ON(extent_state_in_tree(state));
349 btrfs_leak_debug_del(&leak_lock, &state->leak_list);
350 trace_free_extent_state(state, _RET_IP_);
351 kmem_cache_free(extent_state_cache, state);
355 static struct rb_node *tree_insert(struct rb_root *root,
356 struct rb_node *search_start,
358 struct rb_node *node,
359 struct rb_node ***p_in,
360 struct rb_node **parent_in)
363 struct rb_node *parent = NULL;
364 struct tree_entry *entry;
366 if (p_in && parent_in) {
372 p = search_start ? &search_start : &root->rb_node;
375 entry = rb_entry(parent, struct tree_entry, rb_node);
377 if (offset < entry->start)
379 else if (offset > entry->end)
386 rb_link_node(node, parent, p);
387 rb_insert_color(node, root);
392 * __etree_search - searche @tree for an entry that contains @offset. Such
393 * entry would have entry->start <= offset && entry->end >= offset.
395 * @tree - the tree to search
396 * @offset - offset that should fall within an entry in @tree
397 * @next_ret - pointer to the first entry whose range ends after @offset
398 * @prev - pointer to the first entry whose range begins before @offset
399 * @p_ret - pointer where new node should be anchored (used when inserting an
401 * @parent_ret - points to entry which would have been the parent of the entry,
404 * This function returns a pointer to the entry that contains @offset byte
405 * address. If no such entry exists, then NULL is returned and the other
406 * pointer arguments to the function are filled, otherwise the found entry is
407 * returned and other pointers are left untouched.
409 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
410 struct rb_node **next_ret,
411 struct rb_node **prev_ret,
412 struct rb_node ***p_ret,
413 struct rb_node **parent_ret)
415 struct rb_root *root = &tree->state;
416 struct rb_node **n = &root->rb_node;
417 struct rb_node *prev = NULL;
418 struct rb_node *orig_prev = NULL;
419 struct tree_entry *entry;
420 struct tree_entry *prev_entry = NULL;
424 entry = rb_entry(prev, struct tree_entry, rb_node);
427 if (offset < entry->start)
429 else if (offset > entry->end)
442 while (prev && offset > prev_entry->end) {
443 prev = rb_next(prev);
444 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
451 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
452 while (prev && offset < prev_entry->start) {
453 prev = rb_prev(prev);
454 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
461 static inline struct rb_node *
462 tree_search_for_insert(struct extent_io_tree *tree,
464 struct rb_node ***p_ret,
465 struct rb_node **parent_ret)
467 struct rb_node *next= NULL;
470 ret = __etree_search(tree, offset, &next, NULL, p_ret, parent_ret);
476 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
479 return tree_search_for_insert(tree, offset, NULL, NULL);
483 * utility function to look for merge candidates inside a given range.
484 * Any extents with matching state are merged together into a single
485 * extent in the tree. Extents with EXTENT_IO in their state field
486 * are not merged because the end_io handlers need to be able to do
487 * operations on them without sleeping (or doing allocations/splits).
489 * This should be called with the tree lock held.
491 static void merge_state(struct extent_io_tree *tree,
492 struct extent_state *state)
494 struct extent_state *other;
495 struct rb_node *other_node;
497 if (state->state & (EXTENT_LOCKED | EXTENT_BOUNDARY))
500 other_node = rb_prev(&state->rb_node);
502 other = rb_entry(other_node, struct extent_state, rb_node);
503 if (other->end == state->start - 1 &&
504 other->state == state->state) {
505 if (tree->private_data &&
506 is_data_inode(tree->private_data))
507 btrfs_merge_delalloc_extent(tree->private_data,
509 state->start = other->start;
510 rb_erase(&other->rb_node, &tree->state);
511 RB_CLEAR_NODE(&other->rb_node);
512 free_extent_state(other);
515 other_node = rb_next(&state->rb_node);
517 other = rb_entry(other_node, struct extent_state, rb_node);
518 if (other->start == state->end + 1 &&
519 other->state == state->state) {
520 if (tree->private_data &&
521 is_data_inode(tree->private_data))
522 btrfs_merge_delalloc_extent(tree->private_data,
524 state->end = other->end;
525 rb_erase(&other->rb_node, &tree->state);
526 RB_CLEAR_NODE(&other->rb_node);
527 free_extent_state(other);
532 static void set_state_bits(struct extent_io_tree *tree,
533 struct extent_state *state, unsigned *bits,
534 struct extent_changeset *changeset);
537 * insert an extent_state struct into the tree. 'bits' are set on the
538 * struct before it is inserted.
540 * This may return -EEXIST if the extent is already there, in which case the
541 * state struct is freed.
543 * The tree lock is not taken internally. This is a utility function and
544 * probably isn't what you want to call (see set/clear_extent_bit).
546 static int insert_state(struct extent_io_tree *tree,
547 struct extent_state *state, u64 start, u64 end,
549 struct rb_node **parent,
550 unsigned *bits, struct extent_changeset *changeset)
552 struct rb_node *node;
555 btrfs_err(tree->fs_info,
556 "insert state: end < start %llu %llu", end, start);
559 state->start = start;
562 set_state_bits(tree, state, bits, changeset);
564 node = tree_insert(&tree->state, NULL, end, &state->rb_node, p, parent);
566 struct extent_state *found;
567 found = rb_entry(node, struct extent_state, rb_node);
568 btrfs_err(tree->fs_info,
569 "found node %llu %llu on insert of %llu %llu",
570 found->start, found->end, start, end);
573 merge_state(tree, state);
578 * split a given extent state struct in two, inserting the preallocated
579 * struct 'prealloc' as the newly created second half. 'split' indicates an
580 * offset inside 'orig' where it should be split.
583 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
584 * are two extent state structs in the tree:
585 * prealloc: [orig->start, split - 1]
586 * orig: [ split, orig->end ]
588 * The tree locks are not taken by this function. They need to be held
591 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
592 struct extent_state *prealloc, u64 split)
594 struct rb_node *node;
596 if (tree->private_data && is_data_inode(tree->private_data))
597 btrfs_split_delalloc_extent(tree->private_data, orig, split);
599 prealloc->start = orig->start;
600 prealloc->end = split - 1;
601 prealloc->state = orig->state;
604 node = tree_insert(&tree->state, &orig->rb_node, prealloc->end,
605 &prealloc->rb_node, NULL, NULL);
607 free_extent_state(prealloc);
613 static struct extent_state *next_state(struct extent_state *state)
615 struct rb_node *next = rb_next(&state->rb_node);
617 return rb_entry(next, struct extent_state, rb_node);
623 * utility function to clear some bits in an extent state struct.
624 * it will optionally wake up anyone waiting on this state (wake == 1).
626 * If no bits are set on the state struct after clearing things, the
627 * struct is freed and removed from the tree
629 static struct extent_state *clear_state_bit(struct extent_io_tree *tree,
630 struct extent_state *state,
631 unsigned *bits, int wake,
632 struct extent_changeset *changeset)
634 struct extent_state *next;
635 unsigned bits_to_clear = *bits & ~EXTENT_CTLBITS;
638 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
639 u64 range = state->end - state->start + 1;
640 WARN_ON(range > tree->dirty_bytes);
641 tree->dirty_bytes -= range;
644 if (tree->private_data && is_data_inode(tree->private_data))
645 btrfs_clear_delalloc_extent(tree->private_data, state, bits);
647 ret = add_extent_changeset(state, bits_to_clear, changeset, 0);
649 state->state &= ~bits_to_clear;
652 if (state->state == 0) {
653 next = next_state(state);
654 if (extent_state_in_tree(state)) {
655 rb_erase(&state->rb_node, &tree->state);
656 RB_CLEAR_NODE(&state->rb_node);
657 free_extent_state(state);
662 merge_state(tree, state);
663 next = next_state(state);
668 static struct extent_state *
669 alloc_extent_state_atomic(struct extent_state *prealloc)
672 prealloc = alloc_extent_state(GFP_ATOMIC);
677 static void extent_io_tree_panic(struct extent_io_tree *tree, int err)
679 struct inode *inode = tree->private_data;
681 btrfs_panic(btrfs_sb(inode->i_sb), err,
682 "locking error: extent tree was modified by another thread while locked");
686 * clear some bits on a range in the tree. This may require splitting
687 * or inserting elements in the tree, so the gfp mask is used to
688 * indicate which allocations or sleeping are allowed.
690 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
691 * the given range from the tree regardless of state (ie for truncate).
693 * the range [start, end] is inclusive.
695 * This takes the tree lock, and returns 0 on success and < 0 on error.
697 int __clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
698 unsigned bits, int wake, int delete,
699 struct extent_state **cached_state,
700 gfp_t mask, struct extent_changeset *changeset)
702 struct extent_state *state;
703 struct extent_state *cached;
704 struct extent_state *prealloc = NULL;
705 struct rb_node *node;
710 btrfs_debug_check_extent_io_range(tree, start, end);
711 trace_btrfs_clear_extent_bit(tree, start, end - start + 1, bits);
713 if (bits & EXTENT_DELALLOC)
714 bits |= EXTENT_NORESERVE;
717 bits |= ~EXTENT_CTLBITS;
719 if (bits & (EXTENT_LOCKED | EXTENT_BOUNDARY))
722 if (!prealloc && gfpflags_allow_blocking(mask)) {
724 * Don't care for allocation failure here because we might end
725 * up not needing the pre-allocated extent state at all, which
726 * is the case if we only have in the tree extent states that
727 * cover our input range and don't cover too any other range.
728 * If we end up needing a new extent state we allocate it later.
730 prealloc = alloc_extent_state(mask);
733 spin_lock(&tree->lock);
735 cached = *cached_state;
738 *cached_state = NULL;
742 if (cached && extent_state_in_tree(cached) &&
743 cached->start <= start && cached->end > start) {
745 refcount_dec(&cached->refs);
750 free_extent_state(cached);
753 * this search will find the extents that end after
756 node = tree_search(tree, start);
759 state = rb_entry(node, struct extent_state, rb_node);
761 if (state->start > end)
763 WARN_ON(state->end < start);
764 last_end = state->end;
766 /* the state doesn't have the wanted bits, go ahead */
767 if (!(state->state & bits)) {
768 state = next_state(state);
773 * | ---- desired range ---- |
775 * | ------------- state -------------- |
777 * We need to split the extent we found, and may flip
778 * bits on second half.
780 * If the extent we found extends past our range, we
781 * just split and search again. It'll get split again
782 * the next time though.
784 * If the extent we found is inside our range, we clear
785 * the desired bit on it.
788 if (state->start < start) {
789 prealloc = alloc_extent_state_atomic(prealloc);
791 err = split_state(tree, state, prealloc, start);
793 extent_io_tree_panic(tree, err);
798 if (state->end <= end) {
799 state = clear_state_bit(tree, state, &bits, wake,
806 * | ---- desired range ---- |
808 * We need to split the extent, and clear the bit
811 if (state->start <= end && state->end > end) {
812 prealloc = alloc_extent_state_atomic(prealloc);
814 err = split_state(tree, state, prealloc, end + 1);
816 extent_io_tree_panic(tree, err);
821 clear_state_bit(tree, prealloc, &bits, wake, changeset);
827 state = clear_state_bit(tree, state, &bits, wake, changeset);
829 if (last_end == (u64)-1)
831 start = last_end + 1;
832 if (start <= end && state && !need_resched())
838 spin_unlock(&tree->lock);
839 if (gfpflags_allow_blocking(mask))
844 spin_unlock(&tree->lock);
846 free_extent_state(prealloc);
852 static void wait_on_state(struct extent_io_tree *tree,
853 struct extent_state *state)
854 __releases(tree->lock)
855 __acquires(tree->lock)
858 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
859 spin_unlock(&tree->lock);
861 spin_lock(&tree->lock);
862 finish_wait(&state->wq, &wait);
866 * waits for one or more bits to clear on a range in the state tree.
867 * The range [start, end] is inclusive.
868 * The tree lock is taken by this function
870 static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
873 struct extent_state *state;
874 struct rb_node *node;
876 btrfs_debug_check_extent_io_range(tree, start, end);
878 spin_lock(&tree->lock);
882 * this search will find all the extents that end after
885 node = tree_search(tree, start);
890 state = rb_entry(node, struct extent_state, rb_node);
892 if (state->start > end)
895 if (state->state & bits) {
896 start = state->start;
897 refcount_inc(&state->refs);
898 wait_on_state(tree, state);
899 free_extent_state(state);
902 start = state->end + 1;
907 if (!cond_resched_lock(&tree->lock)) {
908 node = rb_next(node);
913 spin_unlock(&tree->lock);
916 static void set_state_bits(struct extent_io_tree *tree,
917 struct extent_state *state,
918 unsigned *bits, struct extent_changeset *changeset)
920 unsigned bits_to_set = *bits & ~EXTENT_CTLBITS;
923 if (tree->private_data && is_data_inode(tree->private_data))
924 btrfs_set_delalloc_extent(tree->private_data, state, bits);
926 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
927 u64 range = state->end - state->start + 1;
928 tree->dirty_bytes += range;
930 ret = add_extent_changeset(state, bits_to_set, changeset, 1);
932 state->state |= bits_to_set;
935 static void cache_state_if_flags(struct extent_state *state,
936 struct extent_state **cached_ptr,
939 if (cached_ptr && !(*cached_ptr)) {
940 if (!flags || (state->state & flags)) {
942 refcount_inc(&state->refs);
947 static void cache_state(struct extent_state *state,
948 struct extent_state **cached_ptr)
950 return cache_state_if_flags(state, cached_ptr,
951 EXTENT_LOCKED | EXTENT_BOUNDARY);
955 * set some bits on a range in the tree. This may require allocations or
956 * sleeping, so the gfp mask is used to indicate what is allowed.
958 * If any of the exclusive bits are set, this will fail with -EEXIST if some
959 * part of the range already has the desired bits set. The start of the
960 * existing range is returned in failed_start in this case.
962 * [start, end] is inclusive This takes the tree lock.
965 static int __must_check
966 __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
967 unsigned bits, unsigned exclusive_bits,
968 u64 *failed_start, struct extent_state **cached_state,
969 gfp_t mask, struct extent_changeset *changeset)
971 struct extent_state *state;
972 struct extent_state *prealloc = NULL;
973 struct rb_node *node;
975 struct rb_node *parent;
980 btrfs_debug_check_extent_io_range(tree, start, end);
981 trace_btrfs_set_extent_bit(tree, start, end - start + 1, bits);
984 ASSERT(failed_start);
986 ASSERT(failed_start == NULL);
988 if (!prealloc && gfpflags_allow_blocking(mask)) {
990 * Don't care for allocation failure here because we might end
991 * up not needing the pre-allocated extent state at all, which
992 * is the case if we only have in the tree extent states that
993 * cover our input range and don't cover too any other range.
994 * If we end up needing a new extent state we allocate it later.
996 prealloc = alloc_extent_state(mask);
999 spin_lock(&tree->lock);
1000 if (cached_state && *cached_state) {
1001 state = *cached_state;
1002 if (state->start <= start && state->end > start &&
1003 extent_state_in_tree(state)) {
1004 node = &state->rb_node;
1009 * this search will find all the extents that end after
1012 node = tree_search_for_insert(tree, start, &p, &parent);
1014 prealloc = alloc_extent_state_atomic(prealloc);
1016 err = insert_state(tree, prealloc, start, end,
1017 &p, &parent, &bits, changeset);
1019 extent_io_tree_panic(tree, err);
1021 cache_state(prealloc, cached_state);
1025 state = rb_entry(node, struct extent_state, rb_node);
1027 last_start = state->start;
1028 last_end = state->end;
1031 * | ---- desired range ---- |
1034 * Just lock what we found and keep going
1036 if (state->start == start && state->end <= end) {
1037 if (state->state & exclusive_bits) {
1038 *failed_start = state->start;
1043 set_state_bits(tree, state, &bits, changeset);
1044 cache_state(state, cached_state);
1045 merge_state(tree, state);
1046 if (last_end == (u64)-1)
1048 start = last_end + 1;
1049 state = next_state(state);
1050 if (start < end && state && state->start == start &&
1057 * | ---- desired range ---- |
1060 * | ------------- state -------------- |
1062 * We need to split the extent we found, and may flip bits on
1065 * If the extent we found extends past our
1066 * range, we just split and search again. It'll get split
1067 * again the next time though.
1069 * If the extent we found is inside our range, we set the
1070 * desired bit on it.
1072 if (state->start < start) {
1073 if (state->state & exclusive_bits) {
1074 *failed_start = start;
1080 * If this extent already has all the bits we want set, then
1081 * skip it, not necessary to split it or do anything with it.
1083 if ((state->state & bits) == bits) {
1084 start = state->end + 1;
1085 cache_state(state, cached_state);
1089 prealloc = alloc_extent_state_atomic(prealloc);
1091 err = split_state(tree, state, prealloc, start);
1093 extent_io_tree_panic(tree, err);
1098 if (state->end <= end) {
1099 set_state_bits(tree, state, &bits, changeset);
1100 cache_state(state, cached_state);
1101 merge_state(tree, state);
1102 if (last_end == (u64)-1)
1104 start = last_end + 1;
1105 state = next_state(state);
1106 if (start < end && state && state->start == start &&
1113 * | ---- desired range ---- |
1114 * | state | or | state |
1116 * There's a hole, we need to insert something in it and
1117 * ignore the extent we found.
1119 if (state->start > start) {
1121 if (end < last_start)
1124 this_end = last_start - 1;
1126 prealloc = alloc_extent_state_atomic(prealloc);
1130 * Avoid to free 'prealloc' if it can be merged with
1133 err = insert_state(tree, prealloc, start, this_end,
1134 NULL, NULL, &bits, changeset);
1136 extent_io_tree_panic(tree, err);
1138 cache_state(prealloc, cached_state);
1140 start = this_end + 1;
1144 * | ---- desired range ---- |
1146 * We need to split the extent, and set the bit
1149 if (state->start <= end && state->end > end) {
1150 if (state->state & exclusive_bits) {
1151 *failed_start = start;
1156 prealloc = alloc_extent_state_atomic(prealloc);
1158 err = split_state(tree, state, prealloc, end + 1);
1160 extent_io_tree_panic(tree, err);
1162 set_state_bits(tree, prealloc, &bits, changeset);
1163 cache_state(prealloc, cached_state);
1164 merge_state(tree, prealloc);
1172 spin_unlock(&tree->lock);
1173 if (gfpflags_allow_blocking(mask))
1178 spin_unlock(&tree->lock);
1180 free_extent_state(prealloc);
1186 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1187 unsigned bits, struct extent_state **cached_state, gfp_t mask)
1189 return __set_extent_bit(tree, start, end, bits, 0, NULL, cached_state,
1195 * convert_extent_bit - convert all bits in a given range from one bit to
1197 * @tree: the io tree to search
1198 * @start: the start offset in bytes
1199 * @end: the end offset in bytes (inclusive)
1200 * @bits: the bits to set in this range
1201 * @clear_bits: the bits to clear in this range
1202 * @cached_state: state that we're going to cache
1204 * This will go through and set bits for the given range. If any states exist
1205 * already in this range they are set with the given bit and cleared of the
1206 * clear_bits. This is only meant to be used by things that are mergeable, ie
1207 * converting from say DELALLOC to DIRTY. This is not meant to be used with
1208 * boundary bits like LOCK.
1210 * All allocations are done with GFP_NOFS.
1212 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1213 unsigned bits, unsigned clear_bits,
1214 struct extent_state **cached_state)
1216 struct extent_state *state;
1217 struct extent_state *prealloc = NULL;
1218 struct rb_node *node;
1220 struct rb_node *parent;
1224 bool first_iteration = true;
1226 btrfs_debug_check_extent_io_range(tree, start, end);
1227 trace_btrfs_convert_extent_bit(tree, start, end - start + 1, bits,
1233 * Best effort, don't worry if extent state allocation fails
1234 * here for the first iteration. We might have a cached state
1235 * that matches exactly the target range, in which case no
1236 * extent state allocations are needed. We'll only know this
1237 * after locking the tree.
1239 prealloc = alloc_extent_state(GFP_NOFS);
1240 if (!prealloc && !first_iteration)
1244 spin_lock(&tree->lock);
1245 if (cached_state && *cached_state) {
1246 state = *cached_state;
1247 if (state->start <= start && state->end > start &&
1248 extent_state_in_tree(state)) {
1249 node = &state->rb_node;
1255 * this search will find all the extents that end after
1258 node = tree_search_for_insert(tree, start, &p, &parent);
1260 prealloc = alloc_extent_state_atomic(prealloc);
1265 err = insert_state(tree, prealloc, start, end,
1266 &p, &parent, &bits, NULL);
1268 extent_io_tree_panic(tree, err);
1269 cache_state(prealloc, cached_state);
1273 state = rb_entry(node, struct extent_state, rb_node);
1275 last_start = state->start;
1276 last_end = state->end;
1279 * | ---- desired range ---- |
1282 * Just lock what we found and keep going
1284 if (state->start == start && state->end <= end) {
1285 set_state_bits(tree, state, &bits, NULL);
1286 cache_state(state, cached_state);
1287 state = clear_state_bit(tree, state, &clear_bits, 0, NULL);
1288 if (last_end == (u64)-1)
1290 start = last_end + 1;
1291 if (start < end && state && state->start == start &&
1298 * | ---- desired range ---- |
1301 * | ------------- state -------------- |
1303 * We need to split the extent we found, and may flip bits on
1306 * If the extent we found extends past our
1307 * range, we just split and search again. It'll get split
1308 * again the next time though.
1310 * If the extent we found is inside our range, we set the
1311 * desired bit on it.
1313 if (state->start < start) {
1314 prealloc = alloc_extent_state_atomic(prealloc);
1319 err = split_state(tree, state, prealloc, start);
1321 extent_io_tree_panic(tree, err);
1325 if (state->end <= end) {
1326 set_state_bits(tree, state, &bits, NULL);
1327 cache_state(state, cached_state);
1328 state = clear_state_bit(tree, state, &clear_bits, 0,
1330 if (last_end == (u64)-1)
1332 start = last_end + 1;
1333 if (start < end && state && state->start == start &&
1340 * | ---- desired range ---- |
1341 * | state | or | state |
1343 * There's a hole, we need to insert something in it and
1344 * ignore the extent we found.
1346 if (state->start > start) {
1348 if (end < last_start)
1351 this_end = last_start - 1;
1353 prealloc = alloc_extent_state_atomic(prealloc);
1360 * Avoid to free 'prealloc' if it can be merged with
1363 err = insert_state(tree, prealloc, start, this_end,
1364 NULL, NULL, &bits, NULL);
1366 extent_io_tree_panic(tree, err);
1367 cache_state(prealloc, cached_state);
1369 start = this_end + 1;
1373 * | ---- desired range ---- |
1375 * We need to split the extent, and set the bit
1378 if (state->start <= end && state->end > end) {
1379 prealloc = alloc_extent_state_atomic(prealloc);
1385 err = split_state(tree, state, prealloc, end + 1);
1387 extent_io_tree_panic(tree, err);
1389 set_state_bits(tree, prealloc, &bits, NULL);
1390 cache_state(prealloc, cached_state);
1391 clear_state_bit(tree, prealloc, &clear_bits, 0, NULL);
1399 spin_unlock(&tree->lock);
1401 first_iteration = false;
1405 spin_unlock(&tree->lock);
1407 free_extent_state(prealloc);
1412 /* wrappers around set/clear extent bit */
1413 int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1414 unsigned bits, struct extent_changeset *changeset)
1417 * We don't support EXTENT_LOCKED yet, as current changeset will
1418 * record any bits changed, so for EXTENT_LOCKED case, it will
1419 * either fail with -EEXIST or changeset will record the whole
1422 BUG_ON(bits & EXTENT_LOCKED);
1424 return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL, GFP_NOFS,
1428 int set_extent_bits_nowait(struct extent_io_tree *tree, u64 start, u64 end,
1431 return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL,
1435 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1436 unsigned bits, int wake, int delete,
1437 struct extent_state **cached)
1439 return __clear_extent_bit(tree, start, end, bits, wake, delete,
1440 cached, GFP_NOFS, NULL);
1443 int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1444 unsigned bits, struct extent_changeset *changeset)
1447 * Don't support EXTENT_LOCKED case, same reason as
1448 * set_record_extent_bits().
1450 BUG_ON(bits & EXTENT_LOCKED);
1452 return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, GFP_NOFS,
1457 * either insert or lock state struct between start and end use mask to tell
1458 * us if waiting is desired.
1460 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1461 struct extent_state **cached_state)
1467 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED,
1468 EXTENT_LOCKED, &failed_start,
1469 cached_state, GFP_NOFS, NULL);
1470 if (err == -EEXIST) {
1471 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1472 start = failed_start;
1475 WARN_ON(start > end);
1480 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1485 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1486 &failed_start, NULL, GFP_NOFS, NULL);
1487 if (err == -EEXIST) {
1488 if (failed_start > start)
1489 clear_extent_bit(tree, start, failed_start - 1,
1490 EXTENT_LOCKED, 1, 0, NULL);
1496 void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
1498 unsigned long index = start >> PAGE_SHIFT;
1499 unsigned long end_index = end >> PAGE_SHIFT;
1502 while (index <= end_index) {
1503 page = find_get_page(inode->i_mapping, index);
1504 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1505 clear_page_dirty_for_io(page);
1511 void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
1513 unsigned long index = start >> PAGE_SHIFT;
1514 unsigned long end_index = end >> PAGE_SHIFT;
1517 while (index <= end_index) {
1518 page = find_get_page(inode->i_mapping, index);
1519 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1520 __set_page_dirty_nobuffers(page);
1521 account_page_redirty(page);
1527 /* find the first state struct with 'bits' set after 'start', and
1528 * return it. tree->lock must be held. NULL will returned if
1529 * nothing was found after 'start'
1531 static struct extent_state *
1532 find_first_extent_bit_state(struct extent_io_tree *tree,
1533 u64 start, unsigned bits)
1535 struct rb_node *node;
1536 struct extent_state *state;
1539 * this search will find all the extents that end after
1542 node = tree_search(tree, start);
1547 state = rb_entry(node, struct extent_state, rb_node);
1548 if (state->end >= start && (state->state & bits))
1551 node = rb_next(node);
1560 * Find the first offset in the io tree with one or more @bits set.
1562 * Note: If there are multiple bits set in @bits, any of them will match.
1564 * Return 0 if we find something, and update @start_ret and @end_ret.
1565 * Return 1 if we found nothing.
1567 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1568 u64 *start_ret, u64 *end_ret, unsigned bits,
1569 struct extent_state **cached_state)
1571 struct extent_state *state;
1574 spin_lock(&tree->lock);
1575 if (cached_state && *cached_state) {
1576 state = *cached_state;
1577 if (state->end == start - 1 && extent_state_in_tree(state)) {
1578 while ((state = next_state(state)) != NULL) {
1579 if (state->state & bits)
1582 free_extent_state(*cached_state);
1583 *cached_state = NULL;
1586 free_extent_state(*cached_state);
1587 *cached_state = NULL;
1590 state = find_first_extent_bit_state(tree, start, bits);
1593 cache_state_if_flags(state, cached_state, 0);
1594 *start_ret = state->start;
1595 *end_ret = state->end;
1599 spin_unlock(&tree->lock);
1604 * find_contiguous_extent_bit: find a contiguous area of bits
1605 * @tree - io tree to check
1606 * @start - offset to start the search from
1607 * @start_ret - the first offset we found with the bits set
1608 * @end_ret - the final contiguous range of the bits that were set
1609 * @bits - bits to look for
1611 * set_extent_bit and clear_extent_bit can temporarily split contiguous ranges
1612 * to set bits appropriately, and then merge them again. During this time it
1613 * will drop the tree->lock, so use this helper if you want to find the actual
1614 * contiguous area for given bits. We will search to the first bit we find, and
1615 * then walk down the tree until we find a non-contiguous area. The area
1616 * returned will be the full contiguous area with the bits set.
1618 int find_contiguous_extent_bit(struct extent_io_tree *tree, u64 start,
1619 u64 *start_ret, u64 *end_ret, unsigned bits)
1621 struct extent_state *state;
1624 spin_lock(&tree->lock);
1625 state = find_first_extent_bit_state(tree, start, bits);
1627 *start_ret = state->start;
1628 *end_ret = state->end;
1629 while ((state = next_state(state)) != NULL) {
1630 if (state->start > (*end_ret + 1))
1632 *end_ret = state->end;
1636 spin_unlock(&tree->lock);
1641 * find_first_clear_extent_bit - find the first range that has @bits not set.
1642 * This range could start before @start.
1644 * @tree - the tree to search
1645 * @start - the offset at/after which the found extent should start
1646 * @start_ret - records the beginning of the range
1647 * @end_ret - records the end of the range (inclusive)
1648 * @bits - the set of bits which must be unset
1650 * Since unallocated range is also considered one which doesn't have the bits
1651 * set it's possible that @end_ret contains -1, this happens in case the range
1652 * spans (last_range_end, end of device]. In this case it's up to the caller to
1653 * trim @end_ret to the appropriate size.
1655 void find_first_clear_extent_bit(struct extent_io_tree *tree, u64 start,
1656 u64 *start_ret, u64 *end_ret, unsigned bits)
1658 struct extent_state *state;
1659 struct rb_node *node, *prev = NULL, *next;
1661 spin_lock(&tree->lock);
1663 /* Find first extent with bits cleared */
1665 node = __etree_search(tree, start, &next, &prev, NULL, NULL);
1666 if (!node && !next && !prev) {
1668 * Tree is completely empty, send full range and let
1669 * caller deal with it
1674 } else if (!node && !next) {
1676 * We are past the last allocated chunk, set start at
1677 * the end of the last extent.
1679 state = rb_entry(prev, struct extent_state, rb_node);
1680 *start_ret = state->end + 1;
1687 * At this point 'node' either contains 'start' or start is
1690 state = rb_entry(node, struct extent_state, rb_node);
1692 if (in_range(start, state->start, state->end - state->start + 1)) {
1693 if (state->state & bits) {
1695 * |--range with bits sets--|
1699 start = state->end + 1;
1702 * 'start' falls within a range that doesn't
1703 * have the bits set, so take its start as
1704 * the beginning of the desired range
1706 * |--range with bits cleared----|
1710 *start_ret = state->start;
1715 * |---prev range---|---hole/unset---|---node range---|
1721 * |---hole/unset--||--first node--|
1726 state = rb_entry(prev, struct extent_state,
1728 *start_ret = state->end + 1;
1737 * Find the longest stretch from start until an entry which has the
1741 state = rb_entry(node, struct extent_state, rb_node);
1742 if (state->end >= start && !(state->state & bits)) {
1743 *end_ret = state->end;
1745 *end_ret = state->start - 1;
1749 node = rb_next(node);
1754 spin_unlock(&tree->lock);
1758 * find a contiguous range of bytes in the file marked as delalloc, not
1759 * more than 'max_bytes'. start and end are used to return the range,
1761 * true is returned if we find something, false if nothing was in the tree
1763 bool btrfs_find_delalloc_range(struct extent_io_tree *tree, u64 *start,
1764 u64 *end, u64 max_bytes,
1765 struct extent_state **cached_state)
1767 struct rb_node *node;
1768 struct extent_state *state;
1769 u64 cur_start = *start;
1771 u64 total_bytes = 0;
1773 spin_lock(&tree->lock);
1776 * this search will find all the extents that end after
1779 node = tree_search(tree, cur_start);
1786 state = rb_entry(node, struct extent_state, rb_node);
1787 if (found && (state->start != cur_start ||
1788 (state->state & EXTENT_BOUNDARY))) {
1791 if (!(state->state & EXTENT_DELALLOC)) {
1797 *start = state->start;
1798 *cached_state = state;
1799 refcount_inc(&state->refs);
1803 cur_start = state->end + 1;
1804 node = rb_next(node);
1805 total_bytes += state->end - state->start + 1;
1806 if (total_bytes >= max_bytes)
1812 spin_unlock(&tree->lock);
1816 static int __process_pages_contig(struct address_space *mapping,
1817 struct page *locked_page,
1818 pgoff_t start_index, pgoff_t end_index,
1819 unsigned long page_ops, pgoff_t *index_ret);
1821 static noinline void __unlock_for_delalloc(struct inode *inode,
1822 struct page *locked_page,
1825 unsigned long index = start >> PAGE_SHIFT;
1826 unsigned long end_index = end >> PAGE_SHIFT;
1828 ASSERT(locked_page);
1829 if (index == locked_page->index && end_index == index)
1832 __process_pages_contig(inode->i_mapping, locked_page, index, end_index,
1836 static noinline int lock_delalloc_pages(struct inode *inode,
1837 struct page *locked_page,
1841 unsigned long index = delalloc_start >> PAGE_SHIFT;
1842 unsigned long index_ret = index;
1843 unsigned long end_index = delalloc_end >> PAGE_SHIFT;
1846 ASSERT(locked_page);
1847 if (index == locked_page->index && index == end_index)
1850 ret = __process_pages_contig(inode->i_mapping, locked_page, index,
1851 end_index, PAGE_LOCK, &index_ret);
1853 __unlock_for_delalloc(inode, locked_page, delalloc_start,
1854 (u64)index_ret << PAGE_SHIFT);
1859 * Find and lock a contiguous range of bytes in the file marked as delalloc, no
1860 * more than @max_bytes. @Start and @end are used to return the range,
1862 * Return: true if we find something
1863 * false if nothing was in the tree
1866 noinline_for_stack bool find_lock_delalloc_range(struct inode *inode,
1867 struct page *locked_page, u64 *start,
1870 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
1871 u64 max_bytes = BTRFS_MAX_EXTENT_SIZE;
1875 struct extent_state *cached_state = NULL;
1880 /* step one, find a bunch of delalloc bytes starting at start */
1881 delalloc_start = *start;
1883 found = btrfs_find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1884 max_bytes, &cached_state);
1885 if (!found || delalloc_end <= *start) {
1886 *start = delalloc_start;
1887 *end = delalloc_end;
1888 free_extent_state(cached_state);
1893 * start comes from the offset of locked_page. We have to lock
1894 * pages in order, so we can't process delalloc bytes before
1897 if (delalloc_start < *start)
1898 delalloc_start = *start;
1901 * make sure to limit the number of pages we try to lock down
1903 if (delalloc_end + 1 - delalloc_start > max_bytes)
1904 delalloc_end = delalloc_start + max_bytes - 1;
1906 /* step two, lock all the pages after the page that has start */
1907 ret = lock_delalloc_pages(inode, locked_page,
1908 delalloc_start, delalloc_end);
1909 ASSERT(!ret || ret == -EAGAIN);
1910 if (ret == -EAGAIN) {
1911 /* some of the pages are gone, lets avoid looping by
1912 * shortening the size of the delalloc range we're searching
1914 free_extent_state(cached_state);
1915 cached_state = NULL;
1917 max_bytes = PAGE_SIZE;
1926 /* step three, lock the state bits for the whole range */
1927 lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state);
1929 /* then test to make sure it is all still delalloc */
1930 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1931 EXTENT_DELALLOC, 1, cached_state);
1933 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1935 __unlock_for_delalloc(inode, locked_page,
1936 delalloc_start, delalloc_end);
1940 free_extent_state(cached_state);
1941 *start = delalloc_start;
1942 *end = delalloc_end;
1947 static int __process_pages_contig(struct address_space *mapping,
1948 struct page *locked_page,
1949 pgoff_t start_index, pgoff_t end_index,
1950 unsigned long page_ops, pgoff_t *index_ret)
1952 unsigned long nr_pages = end_index - start_index + 1;
1953 unsigned long pages_processed = 0;
1954 pgoff_t index = start_index;
1955 struct page *pages[16];
1960 if (page_ops & PAGE_LOCK) {
1961 ASSERT(page_ops == PAGE_LOCK);
1962 ASSERT(index_ret && *index_ret == start_index);
1965 if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0)
1966 mapping_set_error(mapping, -EIO);
1968 while (nr_pages > 0) {
1969 ret = find_get_pages_contig(mapping, index,
1970 min_t(unsigned long,
1971 nr_pages, ARRAY_SIZE(pages)), pages);
1974 * Only if we're going to lock these pages,
1975 * can we find nothing at @index.
1977 ASSERT(page_ops & PAGE_LOCK);
1982 for (i = 0; i < ret; i++) {
1983 if (page_ops & PAGE_SET_PRIVATE2)
1984 SetPagePrivate2(pages[i]);
1986 if (locked_page && pages[i] == locked_page) {
1991 if (page_ops & PAGE_CLEAR_DIRTY)
1992 clear_page_dirty_for_io(pages[i]);
1993 if (page_ops & PAGE_SET_WRITEBACK)
1994 set_page_writeback(pages[i]);
1995 if (page_ops & PAGE_SET_ERROR)
1996 SetPageError(pages[i]);
1997 if (page_ops & PAGE_END_WRITEBACK)
1998 end_page_writeback(pages[i]);
1999 if (page_ops & PAGE_UNLOCK)
2000 unlock_page(pages[i]);
2001 if (page_ops & PAGE_LOCK) {
2002 lock_page(pages[i]);
2003 if (!PageDirty(pages[i]) ||
2004 pages[i]->mapping != mapping) {
2005 unlock_page(pages[i]);
2006 for (; i < ret; i++)
2020 if (err && index_ret)
2021 *index_ret = start_index + pages_processed - 1;
2025 void extent_clear_unlock_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2026 struct page *locked_page,
2027 unsigned clear_bits,
2028 unsigned long page_ops)
2030 clear_extent_bit(&inode->io_tree, start, end, clear_bits, 1, 0, NULL);
2032 __process_pages_contig(inode->vfs_inode.i_mapping, locked_page,
2033 start >> PAGE_SHIFT, end >> PAGE_SHIFT,
2038 * count the number of bytes in the tree that have a given bit(s)
2039 * set. This can be fairly slow, except for EXTENT_DIRTY which is
2040 * cached. The total number found is returned.
2042 u64 count_range_bits(struct extent_io_tree *tree,
2043 u64 *start, u64 search_end, u64 max_bytes,
2044 unsigned bits, int contig)
2046 struct rb_node *node;
2047 struct extent_state *state;
2048 u64 cur_start = *start;
2049 u64 total_bytes = 0;
2053 if (WARN_ON(search_end <= cur_start))
2056 spin_lock(&tree->lock);
2057 if (cur_start == 0 && bits == EXTENT_DIRTY) {
2058 total_bytes = tree->dirty_bytes;
2062 * this search will find all the extents that end after
2065 node = tree_search(tree, cur_start);
2070 state = rb_entry(node, struct extent_state, rb_node);
2071 if (state->start > search_end)
2073 if (contig && found && state->start > last + 1)
2075 if (state->end >= cur_start && (state->state & bits) == bits) {
2076 total_bytes += min(search_end, state->end) + 1 -
2077 max(cur_start, state->start);
2078 if (total_bytes >= max_bytes)
2081 *start = max(cur_start, state->start);
2085 } else if (contig && found) {
2088 node = rb_next(node);
2093 spin_unlock(&tree->lock);
2098 * set the private field for a given byte offset in the tree. If there isn't
2099 * an extent_state there already, this does nothing.
2101 int set_state_failrec(struct extent_io_tree *tree, u64 start,
2102 struct io_failure_record *failrec)
2104 struct rb_node *node;
2105 struct extent_state *state;
2108 spin_lock(&tree->lock);
2110 * this search will find all the extents that end after
2113 node = tree_search(tree, start);
2118 state = rb_entry(node, struct extent_state, rb_node);
2119 if (state->start != start) {
2123 state->failrec = failrec;
2125 spin_unlock(&tree->lock);
2129 struct io_failure_record *get_state_failrec(struct extent_io_tree *tree, u64 start)
2131 struct rb_node *node;
2132 struct extent_state *state;
2133 struct io_failure_record *failrec;
2135 spin_lock(&tree->lock);
2137 * this search will find all the extents that end after
2140 node = tree_search(tree, start);
2142 failrec = ERR_PTR(-ENOENT);
2145 state = rb_entry(node, struct extent_state, rb_node);
2146 if (state->start != start) {
2147 failrec = ERR_PTR(-ENOENT);
2151 failrec = state->failrec;
2153 spin_unlock(&tree->lock);
2158 * searches a range in the state tree for a given mask.
2159 * If 'filled' == 1, this returns 1 only if every extent in the tree
2160 * has the bits set. Otherwise, 1 is returned if any bit in the
2161 * range is found set.
2163 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
2164 unsigned bits, int filled, struct extent_state *cached)
2166 struct extent_state *state = NULL;
2167 struct rb_node *node;
2170 spin_lock(&tree->lock);
2171 if (cached && extent_state_in_tree(cached) && cached->start <= start &&
2172 cached->end > start)
2173 node = &cached->rb_node;
2175 node = tree_search(tree, start);
2176 while (node && start <= end) {
2177 state = rb_entry(node, struct extent_state, rb_node);
2179 if (filled && state->start > start) {
2184 if (state->start > end)
2187 if (state->state & bits) {
2191 } else if (filled) {
2196 if (state->end == (u64)-1)
2199 start = state->end + 1;
2202 node = rb_next(node);
2209 spin_unlock(&tree->lock);
2214 * helper function to set a given page up to date if all the
2215 * extents in the tree for that page are up to date
2217 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
2219 u64 start = page_offset(page);
2220 u64 end = start + PAGE_SIZE - 1;
2221 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
2222 SetPageUptodate(page);
2225 int free_io_failure(struct extent_io_tree *failure_tree,
2226 struct extent_io_tree *io_tree,
2227 struct io_failure_record *rec)
2232 set_state_failrec(failure_tree, rec->start, NULL);
2233 ret = clear_extent_bits(failure_tree, rec->start,
2234 rec->start + rec->len - 1,
2235 EXTENT_LOCKED | EXTENT_DIRTY);
2239 ret = clear_extent_bits(io_tree, rec->start,
2240 rec->start + rec->len - 1,
2250 * this bypasses the standard btrfs submit functions deliberately, as
2251 * the standard behavior is to write all copies in a raid setup. here we only
2252 * want to write the one bad copy. so we do the mapping for ourselves and issue
2253 * submit_bio directly.
2254 * to avoid any synchronization issues, wait for the data after writing, which
2255 * actually prevents the read that triggered the error from finishing.
2256 * currently, there can be no more than two copies of every data bit. thus,
2257 * exactly one rewrite is required.
2259 int repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start,
2260 u64 length, u64 logical, struct page *page,
2261 unsigned int pg_offset, int mirror_num)
2264 struct btrfs_device *dev;
2267 struct btrfs_bio *bbio = NULL;
2270 ASSERT(!(fs_info->sb->s_flags & SB_RDONLY));
2271 BUG_ON(!mirror_num);
2273 bio = btrfs_io_bio_alloc(1);
2274 bio->bi_iter.bi_size = 0;
2275 map_length = length;
2278 * Avoid races with device replace and make sure our bbio has devices
2279 * associated to its stripes that don't go away while we are doing the
2280 * read repair operation.
2282 btrfs_bio_counter_inc_blocked(fs_info);
2283 if (btrfs_is_parity_mirror(fs_info, logical, length)) {
2285 * Note that we don't use BTRFS_MAP_WRITE because it's supposed
2286 * to update all raid stripes, but here we just want to correct
2287 * bad stripe, thus BTRFS_MAP_READ is abused to only get the bad
2288 * stripe's dev and sector.
2290 ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, logical,
2291 &map_length, &bbio, 0);
2293 btrfs_bio_counter_dec(fs_info);
2297 ASSERT(bbio->mirror_num == 1);
2299 ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical,
2300 &map_length, &bbio, mirror_num);
2302 btrfs_bio_counter_dec(fs_info);
2306 BUG_ON(mirror_num != bbio->mirror_num);
2309 sector = bbio->stripes[bbio->mirror_num - 1].physical >> 9;
2310 bio->bi_iter.bi_sector = sector;
2311 dev = bbio->stripes[bbio->mirror_num - 1].dev;
2312 btrfs_put_bbio(bbio);
2313 if (!dev || !dev->bdev ||
2314 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) {
2315 btrfs_bio_counter_dec(fs_info);
2319 bio_set_dev(bio, dev->bdev);
2320 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC;
2321 bio_add_page(bio, page, length, pg_offset);
2323 if (btrfsic_submit_bio_wait(bio)) {
2324 /* try to remap that extent elsewhere? */
2325 btrfs_bio_counter_dec(fs_info);
2327 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
2331 btrfs_info_rl_in_rcu(fs_info,
2332 "read error corrected: ino %llu off %llu (dev %s sector %llu)",
2334 rcu_str_deref(dev->name), sector);
2335 btrfs_bio_counter_dec(fs_info);
2340 int btrfs_repair_eb_io_failure(const struct extent_buffer *eb, int mirror_num)
2342 struct btrfs_fs_info *fs_info = eb->fs_info;
2343 u64 start = eb->start;
2344 int i, num_pages = num_extent_pages(eb);
2347 if (sb_rdonly(fs_info->sb))
2350 for (i = 0; i < num_pages; i++) {
2351 struct page *p = eb->pages[i];
2353 ret = repair_io_failure(fs_info, 0, start, PAGE_SIZE, start, p,
2354 start - page_offset(p), mirror_num);
2364 * each time an IO finishes, we do a fast check in the IO failure tree
2365 * to see if we need to process or clean up an io_failure_record
2367 int clean_io_failure(struct btrfs_fs_info *fs_info,
2368 struct extent_io_tree *failure_tree,
2369 struct extent_io_tree *io_tree, u64 start,
2370 struct page *page, u64 ino, unsigned int pg_offset)
2373 struct io_failure_record *failrec;
2374 struct extent_state *state;
2379 ret = count_range_bits(failure_tree, &private, (u64)-1, 1,
2384 failrec = get_state_failrec(failure_tree, start);
2385 if (IS_ERR(failrec))
2388 BUG_ON(!failrec->this_mirror);
2390 if (failrec->in_validation) {
2391 /* there was no real error, just free the record */
2392 btrfs_debug(fs_info,
2393 "clean_io_failure: freeing dummy error at %llu",
2397 if (sb_rdonly(fs_info->sb))
2400 spin_lock(&io_tree->lock);
2401 state = find_first_extent_bit_state(io_tree,
2404 spin_unlock(&io_tree->lock);
2406 if (state && state->start <= failrec->start &&
2407 state->end >= failrec->start + failrec->len - 1) {
2408 num_copies = btrfs_num_copies(fs_info, failrec->logical,
2410 if (num_copies > 1) {
2411 repair_io_failure(fs_info, ino, start, failrec->len,
2412 failrec->logical, page, pg_offset,
2413 failrec->failed_mirror);
2418 free_io_failure(failure_tree, io_tree, failrec);
2424 * Can be called when
2425 * - hold extent lock
2426 * - under ordered extent
2427 * - the inode is freeing
2429 void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start, u64 end)
2431 struct extent_io_tree *failure_tree = &inode->io_failure_tree;
2432 struct io_failure_record *failrec;
2433 struct extent_state *state, *next;
2435 if (RB_EMPTY_ROOT(&failure_tree->state))
2438 spin_lock(&failure_tree->lock);
2439 state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY);
2441 if (state->start > end)
2444 ASSERT(state->end <= end);
2446 next = next_state(state);
2448 failrec = state->failrec;
2449 free_extent_state(state);
2454 spin_unlock(&failure_tree->lock);
2457 static struct io_failure_record *btrfs_get_io_failure_record(struct inode *inode,
2460 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2461 struct io_failure_record *failrec;
2462 struct extent_map *em;
2463 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2464 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2465 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2469 failrec = get_state_failrec(failure_tree, start);
2470 if (!IS_ERR(failrec)) {
2471 btrfs_debug(fs_info,
2472 "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d",
2473 failrec->logical, failrec->start, failrec->len,
2474 failrec->in_validation);
2476 * when data can be on disk more than twice, add to failrec here
2477 * (e.g. with a list for failed_mirror) to make
2478 * clean_io_failure() clean all those errors at once.
2484 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2486 return ERR_PTR(-ENOMEM);
2488 failrec->start = start;
2489 failrec->len = end - start + 1;
2490 failrec->this_mirror = 0;
2491 failrec->bio_flags = 0;
2492 failrec->in_validation = 0;
2494 read_lock(&em_tree->lock);
2495 em = lookup_extent_mapping(em_tree, start, failrec->len);
2497 read_unlock(&em_tree->lock);
2499 return ERR_PTR(-EIO);
2502 if (em->start > start || em->start + em->len <= start) {
2503 free_extent_map(em);
2506 read_unlock(&em_tree->lock);
2509 return ERR_PTR(-EIO);
2512 logical = start - em->start;
2513 logical = em->block_start + logical;
2514 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2515 logical = em->block_start;
2516 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2517 extent_set_compress_type(&failrec->bio_flags, em->compress_type);
2520 btrfs_debug(fs_info,
2521 "Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu",
2522 logical, start, failrec->len);
2524 failrec->logical = logical;
2525 free_extent_map(em);
2527 /* Set the bits in the private failure tree */
2528 ret = set_extent_bits(failure_tree, start, end,
2529 EXTENT_LOCKED | EXTENT_DIRTY);
2531 ret = set_state_failrec(failure_tree, start, failrec);
2532 /* Set the bits in the inode's tree */
2533 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED);
2534 } else if (ret < 0) {
2536 return ERR_PTR(ret);
2542 static bool btrfs_check_repairable(struct inode *inode, bool needs_validation,
2543 struct io_failure_record *failrec,
2546 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2549 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
2550 if (num_copies == 1) {
2552 * we only have a single copy of the data, so don't bother with
2553 * all the retry and error correction code that follows. no
2554 * matter what the error is, it is very likely to persist.
2556 btrfs_debug(fs_info,
2557 "Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
2558 num_copies, failrec->this_mirror, failed_mirror);
2563 * there are two premises:
2564 * a) deliver good data to the caller
2565 * b) correct the bad sectors on disk
2567 if (needs_validation) {
2569 * to fulfill b), we need to know the exact failing sectors, as
2570 * we don't want to rewrite any more than the failed ones. thus,
2571 * we need separate read requests for the failed bio
2573 * if the following BUG_ON triggers, our validation request got
2574 * merged. we need separate requests for our algorithm to work.
2576 BUG_ON(failrec->in_validation);
2577 failrec->in_validation = 1;
2578 failrec->this_mirror = failed_mirror;
2581 * we're ready to fulfill a) and b) alongside. get a good copy
2582 * of the failed sector and if we succeed, we have setup
2583 * everything for repair_io_failure to do the rest for us.
2585 if (failrec->in_validation) {
2586 BUG_ON(failrec->this_mirror != failed_mirror);
2587 failrec->in_validation = 0;
2588 failrec->this_mirror = 0;
2590 failrec->failed_mirror = failed_mirror;
2591 failrec->this_mirror++;
2592 if (failrec->this_mirror == failed_mirror)
2593 failrec->this_mirror++;
2596 if (failrec->this_mirror > num_copies) {
2597 btrfs_debug(fs_info,
2598 "Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
2599 num_copies, failrec->this_mirror, failed_mirror);
2606 static bool btrfs_io_needs_validation(struct inode *inode, struct bio *bio)
2609 const u32 blocksize = inode->i_sb->s_blocksize;
2612 * If bi_status is BLK_STS_OK, then this was a checksum error, not an
2613 * I/O error. In this case, we already know exactly which sector was
2614 * bad, so we don't need to validate.
2616 if (bio->bi_status == BLK_STS_OK)
2620 * We need to validate each sector individually if the failed I/O was
2621 * for multiple sectors.
2623 * There are a few possible bios that can end up here:
2624 * 1. A buffered read bio, which is not cloned.
2625 * 2. A direct I/O read bio, which is cloned.
2626 * 3. A (buffered or direct) repair bio, which is not cloned.
2628 * For cloned bios (case 2), we can get the size from
2629 * btrfs_io_bio->iter; for non-cloned bios (cases 1 and 3), we can get
2630 * it from the bvecs.
2632 if (bio_flagged(bio, BIO_CLONED)) {
2633 if (btrfs_io_bio(bio)->iter.bi_size > blocksize)
2636 struct bio_vec *bvec;
2639 bio_for_each_bvec_all(bvec, bio, i) {
2640 len += bvec->bv_len;
2641 if (len > blocksize)
2648 blk_status_t btrfs_submit_read_repair(struct inode *inode,
2649 struct bio *failed_bio, u64 phy_offset,
2650 struct page *page, unsigned int pgoff,
2651 u64 start, u64 end, int failed_mirror,
2652 submit_bio_hook_t *submit_bio_hook)
2654 struct io_failure_record *failrec;
2655 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2656 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2657 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2658 struct btrfs_io_bio *failed_io_bio = btrfs_io_bio(failed_bio);
2659 const int icsum = phy_offset >> fs_info->sectorsize_bits;
2660 bool need_validation;
2661 struct bio *repair_bio;
2662 struct btrfs_io_bio *repair_io_bio;
2663 blk_status_t status;
2665 btrfs_debug(fs_info,
2666 "repair read error: read error at %llu", start);
2668 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
2670 failrec = btrfs_get_io_failure_record(inode, start, end);
2671 if (IS_ERR(failrec))
2672 return errno_to_blk_status(PTR_ERR(failrec));
2674 need_validation = btrfs_io_needs_validation(inode, failed_bio);
2676 if (!btrfs_check_repairable(inode, need_validation, failrec,
2678 free_io_failure(failure_tree, tree, failrec);
2679 return BLK_STS_IOERR;
2682 repair_bio = btrfs_io_bio_alloc(1);
2683 repair_io_bio = btrfs_io_bio(repair_bio);
2684 repair_bio->bi_opf = REQ_OP_READ;
2685 if (need_validation)
2686 repair_bio->bi_opf |= REQ_FAILFAST_DEV;
2687 repair_bio->bi_end_io = failed_bio->bi_end_io;
2688 repair_bio->bi_iter.bi_sector = failrec->logical >> 9;
2689 repair_bio->bi_private = failed_bio->bi_private;
2691 if (failed_io_bio->csum) {
2692 const u32 csum_size = fs_info->csum_size;
2694 repair_io_bio->csum = repair_io_bio->csum_inline;
2695 memcpy(repair_io_bio->csum,
2696 failed_io_bio->csum + csum_size * icsum, csum_size);
2699 bio_add_page(repair_bio, page, failrec->len, pgoff);
2700 repair_io_bio->logical = failrec->start;
2701 repair_io_bio->iter = repair_bio->bi_iter;
2703 btrfs_debug(btrfs_sb(inode->i_sb),
2704 "repair read error: submitting new read to mirror %d, in_validation=%d",
2705 failrec->this_mirror, failrec->in_validation);
2707 status = submit_bio_hook(inode, repair_bio, failrec->this_mirror,
2708 failrec->bio_flags);
2710 free_io_failure(failure_tree, tree, failrec);
2711 bio_put(repair_bio);
2716 /* lots and lots of room for performance fixes in the end_bio funcs */
2718 void end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2720 int uptodate = (err == 0);
2723 btrfs_writepage_endio_finish_ordered(page, start, end, uptodate);
2726 ClearPageUptodate(page);
2728 ret = err < 0 ? err : -EIO;
2729 mapping_set_error(page->mapping, ret);
2734 * after a writepage IO is done, we need to:
2735 * clear the uptodate bits on error
2736 * clear the writeback bits in the extent tree for this IO
2737 * end_page_writeback if the page has no more pending IO
2739 * Scheduling is not allowed, so the extent state tree is expected
2740 * to have one and only one object corresponding to this IO.
2742 static void end_bio_extent_writepage(struct bio *bio)
2744 int error = blk_status_to_errno(bio->bi_status);
2745 struct bio_vec *bvec;
2748 struct bvec_iter_all iter_all;
2750 ASSERT(!bio_flagged(bio, BIO_CLONED));
2751 bio_for_each_segment_all(bvec, bio, iter_all) {
2752 struct page *page = bvec->bv_page;
2753 struct inode *inode = page->mapping->host;
2754 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2756 /* We always issue full-page reads, but if some block
2757 * in a page fails to read, blk_update_request() will
2758 * advance bv_offset and adjust bv_len to compensate.
2759 * Print a warning for nonzero offsets, and an error
2760 * if they don't add up to a full page. */
2761 if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) {
2762 if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE)
2764 "partial page write in btrfs with offset %u and length %u",
2765 bvec->bv_offset, bvec->bv_len);
2768 "incomplete page write in btrfs with offset %u and length %u",
2769 bvec->bv_offset, bvec->bv_len);
2772 start = page_offset(page);
2773 end = start + bvec->bv_offset + bvec->bv_len - 1;
2775 end_extent_writepage(page, error, start, end);
2776 end_page_writeback(page);
2783 endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len,
2786 struct extent_state *cached = NULL;
2787 u64 end = start + len - 1;
2789 if (uptodate && tree->track_uptodate)
2790 set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC);
2791 unlock_extent_cached_atomic(tree, start, end, &cached);
2795 * after a readpage IO is done, we need to:
2796 * clear the uptodate bits on error
2797 * set the uptodate bits if things worked
2798 * set the page up to date if all extents in the tree are uptodate
2799 * clear the lock bit in the extent tree
2800 * unlock the page if there are no other extents locked for it
2802 * Scheduling is not allowed, so the extent state tree is expected
2803 * to have one and only one object corresponding to this IO.
2805 static void end_bio_extent_readpage(struct bio *bio)
2807 struct bio_vec *bvec;
2808 int uptodate = !bio->bi_status;
2809 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
2810 struct extent_io_tree *tree, *failure_tree;
2815 u64 extent_start = 0;
2819 struct bvec_iter_all iter_all;
2821 ASSERT(!bio_flagged(bio, BIO_CLONED));
2822 bio_for_each_segment_all(bvec, bio, iter_all) {
2823 struct page *page = bvec->bv_page;
2824 struct inode *inode = page->mapping->host;
2825 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2826 u32 sectorsize = fs_info->sectorsize;
2828 btrfs_debug(fs_info,
2829 "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u",
2830 (u64)bio->bi_iter.bi_sector, bio->bi_status,
2831 io_bio->mirror_num);
2832 tree = &BTRFS_I(inode)->io_tree;
2833 failure_tree = &BTRFS_I(inode)->io_failure_tree;
2836 * We always issue full-sector reads, but if some block in a
2837 * page fails to read, blk_update_request() will advance
2838 * bv_offset and adjust bv_len to compensate. Print a warning
2839 * for unaligned offsets, and an error if they don't add up to
2842 if (!IS_ALIGNED(bvec->bv_offset, sectorsize))
2844 "partial page read in btrfs with offset %u and length %u",
2845 bvec->bv_offset, bvec->bv_len);
2846 else if (!IS_ALIGNED(bvec->bv_offset + bvec->bv_len,
2849 "incomplete page read with offset %u and length %u",
2850 bvec->bv_offset, bvec->bv_len);
2852 start = page_offset(page) + bvec->bv_offset;
2853 end = start + bvec->bv_len - 1;
2856 mirror = io_bio->mirror_num;
2857 if (likely(uptodate)) {
2858 if (is_data_inode(inode))
2859 ret = btrfs_verify_data_csum(io_bio, offset, page,
2860 start, end, mirror);
2862 ret = btrfs_validate_metadata_buffer(io_bio,
2863 offset, page, start, end, mirror);
2867 clean_io_failure(BTRFS_I(inode)->root->fs_info,
2868 failure_tree, tree, start,
2870 btrfs_ino(BTRFS_I(inode)), 0);
2873 if (likely(uptodate))
2876 if (is_data_inode(inode)) {
2879 * The generic bio_readpage_error handles errors the
2880 * following way: If possible, new read requests are
2881 * created and submitted and will end up in
2882 * end_bio_extent_readpage as well (if we're lucky,
2883 * not in the !uptodate case). In that case it returns
2884 * 0 and we just go on with the next page in our bio.
2885 * If it can't handle the error it will return -EIO and
2886 * we remain responsible for that page.
2888 if (!btrfs_submit_read_repair(inode, bio, offset, page,
2889 start - page_offset(page),
2891 btrfs_submit_data_bio)) {
2892 uptodate = !bio->bi_status;
2897 struct extent_buffer *eb;
2899 eb = (struct extent_buffer *)page->private;
2900 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
2901 eb->read_mirror = mirror;
2902 atomic_dec(&eb->io_pages);
2903 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD,
2905 btree_readahead_hook(eb, -EIO);
2908 if (likely(uptodate)) {
2909 loff_t i_size = i_size_read(inode);
2910 pgoff_t end_index = i_size >> PAGE_SHIFT;
2913 /* Zero out the end if this page straddles i_size */
2914 off = offset_in_page(i_size);
2915 if (page->index == end_index && off)
2916 zero_user_segment(page, off, PAGE_SIZE);
2917 SetPageUptodate(page);
2919 ClearPageUptodate(page);
2925 if (unlikely(!uptodate)) {
2927 endio_readpage_release_extent(tree,
2933 endio_readpage_release_extent(tree, start,
2934 end - start + 1, 0);
2935 } else if (!extent_len) {
2936 extent_start = start;
2937 extent_len = end + 1 - start;
2938 } else if (extent_start + extent_len == start) {
2939 extent_len += end + 1 - start;
2941 endio_readpage_release_extent(tree, extent_start,
2942 extent_len, uptodate);
2943 extent_start = start;
2944 extent_len = end + 1 - start;
2949 endio_readpage_release_extent(tree, extent_start, extent_len,
2951 btrfs_io_bio_free_csum(io_bio);
2956 * Initialize the members up to but not including 'bio'. Use after allocating a
2957 * new bio by bio_alloc_bioset as it does not initialize the bytes outside of
2958 * 'bio' because use of __GFP_ZERO is not supported.
2960 static inline void btrfs_io_bio_init(struct btrfs_io_bio *btrfs_bio)
2962 memset(btrfs_bio, 0, offsetof(struct btrfs_io_bio, bio));
2966 * The following helpers allocate a bio. As it's backed by a bioset, it'll
2967 * never fail. We're returning a bio right now but you can call btrfs_io_bio
2968 * for the appropriate container_of magic
2970 struct bio *btrfs_bio_alloc(u64 first_byte)
2974 bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, &btrfs_bioset);
2975 bio->bi_iter.bi_sector = first_byte >> 9;
2976 btrfs_io_bio_init(btrfs_io_bio(bio));
2980 struct bio *btrfs_bio_clone(struct bio *bio)
2982 struct btrfs_io_bio *btrfs_bio;
2985 /* Bio allocation backed by a bioset does not fail */
2986 new = bio_clone_fast(bio, GFP_NOFS, &btrfs_bioset);
2987 btrfs_bio = btrfs_io_bio(new);
2988 btrfs_io_bio_init(btrfs_bio);
2989 btrfs_bio->iter = bio->bi_iter;
2993 struct bio *btrfs_io_bio_alloc(unsigned int nr_iovecs)
2997 /* Bio allocation backed by a bioset does not fail */
2998 bio = bio_alloc_bioset(GFP_NOFS, nr_iovecs, &btrfs_bioset);
2999 btrfs_io_bio_init(btrfs_io_bio(bio));
3003 struct bio *btrfs_bio_clone_partial(struct bio *orig, int offset, int size)
3006 struct btrfs_io_bio *btrfs_bio;
3008 /* this will never fail when it's backed by a bioset */
3009 bio = bio_clone_fast(orig, GFP_NOFS, &btrfs_bioset);
3012 btrfs_bio = btrfs_io_bio(bio);
3013 btrfs_io_bio_init(btrfs_bio);
3015 bio_trim(bio, offset >> 9, size >> 9);
3016 btrfs_bio->iter = bio->bi_iter;
3021 * @opf: bio REQ_OP_* and REQ_* flags as one value
3022 * @wbc: optional writeback control for io accounting
3023 * @page: page to add to the bio
3024 * @pg_offset: offset of the new bio or to check whether we are adding
3025 * a contiguous page to the previous one
3026 * @size: portion of page that we want to write
3027 * @offset: starting offset in the page
3028 * @bio_ret: must be valid pointer, newly allocated bio will be stored there
3029 * @end_io_func: end_io callback for new bio
3030 * @mirror_num: desired mirror to read/write
3031 * @prev_bio_flags: flags of previous bio to see if we can merge the current one
3032 * @bio_flags: flags of the current bio to see if we can merge them
3034 static int submit_extent_page(unsigned int opf,
3035 struct writeback_control *wbc,
3036 struct page *page, u64 offset,
3037 size_t size, unsigned long pg_offset,
3038 struct bio **bio_ret,
3039 bio_end_io_t end_io_func,
3041 unsigned long prev_bio_flags,
3042 unsigned long bio_flags,
3043 bool force_bio_submit)
3047 size_t io_size = min_t(size_t, size, PAGE_SIZE);
3048 sector_t sector = offset >> 9;
3049 struct extent_io_tree *tree = &BTRFS_I(page->mapping->host)->io_tree;
3055 bool can_merge = true;
3058 if (prev_bio_flags & EXTENT_BIO_COMPRESSED)
3059 contig = bio->bi_iter.bi_sector == sector;
3061 contig = bio_end_sector(bio) == sector;
3063 if (btrfs_bio_fits_in_stripe(page, io_size, bio, bio_flags))
3066 if (prev_bio_flags != bio_flags || !contig || !can_merge ||
3068 bio_add_page(bio, page, io_size, pg_offset) < io_size) {
3069 ret = submit_one_bio(bio, mirror_num, prev_bio_flags);
3077 wbc_account_cgroup_owner(wbc, page, io_size);
3082 bio = btrfs_bio_alloc(offset);
3083 bio_add_page(bio, page, io_size, pg_offset);
3084 bio->bi_end_io = end_io_func;
3085 bio->bi_private = tree;
3086 bio->bi_write_hint = page->mapping->host->i_write_hint;
3089 struct block_device *bdev;
3091 bdev = BTRFS_I(page->mapping->host)->root->fs_info->fs_devices->latest_bdev;
3092 bio_set_dev(bio, bdev);
3093 wbc_init_bio(wbc, bio);
3094 wbc_account_cgroup_owner(wbc, page, io_size);
3102 static void attach_extent_buffer_page(struct extent_buffer *eb,
3106 * If the page is mapped to btree inode, we should hold the private
3107 * lock to prevent race.
3108 * For cloned or dummy extent buffers, their pages are not mapped and
3109 * will not race with any other ebs.
3112 lockdep_assert_held(&page->mapping->private_lock);
3114 if (!PagePrivate(page))
3115 attach_page_private(page, eb);
3117 WARN_ON(page->private != (unsigned long)eb);
3120 void set_page_extent_mapped(struct page *page)
3122 if (!PagePrivate(page))
3123 attach_page_private(page, (void *)EXTENT_PAGE_PRIVATE);
3126 static struct extent_map *
3127 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
3128 u64 start, u64 len, struct extent_map **em_cached)
3130 struct extent_map *em;
3132 if (em_cached && *em_cached) {
3134 if (extent_map_in_tree(em) && start >= em->start &&
3135 start < extent_map_end(em)) {
3136 refcount_inc(&em->refs);
3140 free_extent_map(em);
3144 em = btrfs_get_extent(BTRFS_I(inode), page, pg_offset, start, len);
3145 if (em_cached && !IS_ERR_OR_NULL(em)) {
3147 refcount_inc(&em->refs);
3153 * basic readpage implementation. Locked extent state structs are inserted
3154 * into the tree that are removed when the IO is done (by the end_io
3156 * XXX JDM: This needs looking at to ensure proper page locking
3157 * return 0 on success, otherwise return error
3159 int btrfs_do_readpage(struct page *page, struct extent_map **em_cached,
3160 struct bio **bio, unsigned long *bio_flags,
3161 unsigned int read_flags, u64 *prev_em_start)
3163 struct inode *inode = page->mapping->host;
3164 u64 start = page_offset(page);
3165 const u64 end = start + PAGE_SIZE - 1;
3168 u64 last_byte = i_size_read(inode);
3171 struct extent_map *em;
3174 size_t pg_offset = 0;
3176 size_t blocksize = inode->i_sb->s_blocksize;
3177 unsigned long this_bio_flag = 0;
3178 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
3180 set_page_extent_mapped(page);
3182 if (!PageUptodate(page)) {
3183 if (cleancache_get_page(page) == 0) {
3184 BUG_ON(blocksize != PAGE_SIZE);
3185 unlock_extent(tree, start, end);
3190 if (page->index == last_byte >> PAGE_SHIFT) {
3192 size_t zero_offset = offset_in_page(last_byte);
3195 iosize = PAGE_SIZE - zero_offset;
3196 userpage = kmap_atomic(page);
3197 memset(userpage + zero_offset, 0, iosize);
3198 flush_dcache_page(page);
3199 kunmap_atomic(userpage);
3202 while (cur <= end) {
3203 bool force_bio_submit = false;
3206 if (cur >= last_byte) {
3208 struct extent_state *cached = NULL;
3210 iosize = PAGE_SIZE - pg_offset;
3211 userpage = kmap_atomic(page);
3212 memset(userpage + pg_offset, 0, iosize);
3213 flush_dcache_page(page);
3214 kunmap_atomic(userpage);
3215 set_extent_uptodate(tree, cur, cur + iosize - 1,
3217 unlock_extent_cached(tree, cur,
3218 cur + iosize - 1, &cached);
3221 em = __get_extent_map(inode, page, pg_offset, cur,
3222 end - cur + 1, em_cached);
3223 if (IS_ERR_OR_NULL(em)) {
3225 unlock_extent(tree, cur, end);
3228 extent_offset = cur - em->start;
3229 BUG_ON(extent_map_end(em) <= cur);
3232 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
3233 this_bio_flag |= EXTENT_BIO_COMPRESSED;
3234 extent_set_compress_type(&this_bio_flag,
3238 iosize = min(extent_map_end(em) - cur, end - cur + 1);
3239 cur_end = min(extent_map_end(em) - 1, end);
3240 iosize = ALIGN(iosize, blocksize);
3241 if (this_bio_flag & EXTENT_BIO_COMPRESSED)
3242 offset = em->block_start;
3244 offset = em->block_start + extent_offset;
3245 block_start = em->block_start;
3246 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
3247 block_start = EXTENT_MAP_HOLE;
3250 * If we have a file range that points to a compressed extent
3251 * and it's followed by a consecutive file range that points
3252 * to the same compressed extent (possibly with a different
3253 * offset and/or length, so it either points to the whole extent
3254 * or only part of it), we must make sure we do not submit a
3255 * single bio to populate the pages for the 2 ranges because
3256 * this makes the compressed extent read zero out the pages
3257 * belonging to the 2nd range. Imagine the following scenario:
3260 * [0 - 8K] [8K - 24K]
3263 * points to extent X, points to extent X,
3264 * offset 4K, length of 8K offset 0, length 16K
3266 * [extent X, compressed length = 4K uncompressed length = 16K]
3268 * If the bio to read the compressed extent covers both ranges,
3269 * it will decompress extent X into the pages belonging to the
3270 * first range and then it will stop, zeroing out the remaining
3271 * pages that belong to the other range that points to extent X.
3272 * So here we make sure we submit 2 bios, one for the first
3273 * range and another one for the third range. Both will target
3274 * the same physical extent from disk, but we can't currently
3275 * make the compressed bio endio callback populate the pages
3276 * for both ranges because each compressed bio is tightly
3277 * coupled with a single extent map, and each range can have
3278 * an extent map with a different offset value relative to the
3279 * uncompressed data of our extent and different lengths. This
3280 * is a corner case so we prioritize correctness over
3281 * non-optimal behavior (submitting 2 bios for the same extent).
3283 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) &&
3284 prev_em_start && *prev_em_start != (u64)-1 &&
3285 *prev_em_start != em->start)
3286 force_bio_submit = true;
3289 *prev_em_start = em->start;
3291 free_extent_map(em);
3294 /* we've found a hole, just zero and go on */
3295 if (block_start == EXTENT_MAP_HOLE) {
3297 struct extent_state *cached = NULL;
3299 userpage = kmap_atomic(page);
3300 memset(userpage + pg_offset, 0, iosize);
3301 flush_dcache_page(page);
3302 kunmap_atomic(userpage);
3304 set_extent_uptodate(tree, cur, cur + iosize - 1,
3306 unlock_extent_cached(tree, cur,
3307 cur + iosize - 1, &cached);
3309 pg_offset += iosize;
3312 /* the get_extent function already copied into the page */
3313 if (test_range_bit(tree, cur, cur_end,
3314 EXTENT_UPTODATE, 1, NULL)) {
3315 check_page_uptodate(tree, page);
3316 unlock_extent(tree, cur, cur + iosize - 1);
3318 pg_offset += iosize;
3321 /* we have an inline extent but it didn't get marked up
3322 * to date. Error out
3324 if (block_start == EXTENT_MAP_INLINE) {
3326 unlock_extent(tree, cur, cur + iosize - 1);
3328 pg_offset += iosize;
3332 ret = submit_extent_page(REQ_OP_READ | read_flags, NULL,
3333 page, offset, iosize,
3335 end_bio_extent_readpage, 0,
3341 *bio_flags = this_bio_flag;
3344 unlock_extent(tree, cur, cur + iosize - 1);
3348 pg_offset += iosize;
3352 if (!PageError(page))
3353 SetPageUptodate(page);
3359 static inline void contiguous_readpages(struct page *pages[], int nr_pages,
3361 struct extent_map **em_cached,
3363 unsigned long *bio_flags,
3366 struct btrfs_inode *inode = BTRFS_I(pages[0]->mapping->host);
3369 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
3371 for (index = 0; index < nr_pages; index++) {
3372 btrfs_do_readpage(pages[index], em_cached, bio, bio_flags,
3373 REQ_RAHEAD, prev_em_start);
3374 put_page(pages[index]);
3378 static void update_nr_written(struct writeback_control *wbc,
3379 unsigned long nr_written)
3381 wbc->nr_to_write -= nr_written;
3385 * helper for __extent_writepage, doing all of the delayed allocation setup.
3387 * This returns 1 if btrfs_run_delalloc_range function did all the work required
3388 * to write the page (copy into inline extent). In this case the IO has
3389 * been started and the page is already unlocked.
3391 * This returns 0 if all went well (page still locked)
3392 * This returns < 0 if there were errors (page still locked)
3394 static noinline_for_stack int writepage_delalloc(struct btrfs_inode *inode,
3395 struct page *page, struct writeback_control *wbc,
3396 u64 delalloc_start, unsigned long *nr_written)
3398 u64 page_end = delalloc_start + PAGE_SIZE - 1;
3400 u64 delalloc_to_write = 0;
3401 u64 delalloc_end = 0;
3403 int page_started = 0;
3406 while (delalloc_end < page_end) {
3407 found = find_lock_delalloc_range(&inode->vfs_inode, page,
3411 delalloc_start = delalloc_end + 1;
3414 ret = btrfs_run_delalloc_range(inode, page, delalloc_start,
3415 delalloc_end, &page_started, nr_written, wbc);
3419 * btrfs_run_delalloc_range should return < 0 for error
3420 * but just in case, we use > 0 here meaning the IO is
3421 * started, so we don't want to return > 0 unless
3422 * things are going well.
3424 return ret < 0 ? ret : -EIO;
3427 * delalloc_end is already one less than the total length, so
3428 * we don't subtract one from PAGE_SIZE
3430 delalloc_to_write += (delalloc_end - delalloc_start +
3431 PAGE_SIZE) >> PAGE_SHIFT;
3432 delalloc_start = delalloc_end + 1;
3434 if (wbc->nr_to_write < delalloc_to_write) {
3437 if (delalloc_to_write < thresh * 2)
3438 thresh = delalloc_to_write;
3439 wbc->nr_to_write = min_t(u64, delalloc_to_write,
3443 /* did the fill delalloc function already unlock and start
3448 * we've unlocked the page, so we can't update
3449 * the mapping's writeback index, just update
3452 wbc->nr_to_write -= *nr_written;
3460 * helper for __extent_writepage. This calls the writepage start hooks,
3461 * and does the loop to map the page into extents and bios.
3463 * We return 1 if the IO is started and the page is unlocked,
3464 * 0 if all went well (page still locked)
3465 * < 0 if there were errors (page still locked)
3467 static noinline_for_stack int __extent_writepage_io(struct btrfs_inode *inode,
3469 struct writeback_control *wbc,
3470 struct extent_page_data *epd,
3472 unsigned long nr_written,
3475 struct extent_io_tree *tree = &inode->io_tree;
3476 u64 start = page_offset(page);
3477 u64 page_end = start + PAGE_SIZE - 1;
3483 struct extent_map *em;
3484 size_t pg_offset = 0;
3488 const unsigned int write_flags = wbc_to_write_flags(wbc);
3491 ret = btrfs_writepage_cow_fixup(page, start, page_end);
3493 /* Fixup worker will requeue */
3494 redirty_page_for_writepage(wbc, page);
3495 update_nr_written(wbc, nr_written);
3501 * we don't want to touch the inode after unlocking the page,
3502 * so we update the mapping writeback index now
3504 update_nr_written(wbc, nr_written + 1);
3507 blocksize = inode->vfs_inode.i_sb->s_blocksize;
3509 while (cur <= end) {
3513 if (cur >= i_size) {
3514 btrfs_writepage_endio_finish_ordered(page, cur,
3518 em = btrfs_get_extent(inode, NULL, 0, cur, end - cur + 1);
3519 if (IS_ERR_OR_NULL(em)) {
3521 ret = PTR_ERR_OR_ZERO(em);
3525 extent_offset = cur - em->start;
3526 em_end = extent_map_end(em);
3527 BUG_ON(em_end <= cur);
3529 iosize = min(em_end - cur, end - cur + 1);
3530 iosize = ALIGN(iosize, blocksize);
3531 offset = em->block_start + extent_offset;
3532 block_start = em->block_start;
3533 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3534 free_extent_map(em);
3538 * compressed and inline extents are written through other
3541 if (compressed || block_start == EXTENT_MAP_HOLE ||
3542 block_start == EXTENT_MAP_INLINE) {
3546 btrfs_writepage_endio_finish_ordered(page, cur,
3547 cur + iosize - 1, 1);
3549 pg_offset += iosize;
3553 btrfs_set_range_writeback(tree, cur, cur + iosize - 1);
3554 if (!PageWriteback(page)) {
3555 btrfs_err(inode->root->fs_info,
3556 "page %lu not writeback, cur %llu end %llu",
3557 page->index, cur, end);
3560 ret = submit_extent_page(REQ_OP_WRITE | write_flags, wbc,
3561 page, offset, iosize, pg_offset,
3563 end_bio_extent_writepage,
3567 if (PageWriteback(page))
3568 end_page_writeback(page);
3572 pg_offset += iosize;
3580 * the writepage semantics are similar to regular writepage. extent
3581 * records are inserted to lock ranges in the tree, and as dirty areas
3582 * are found, they are marked writeback. Then the lock bits are removed
3583 * and the end_io handler clears the writeback ranges
3585 * Return 0 if everything goes well.
3586 * Return <0 for error.
3588 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
3589 struct extent_page_data *epd)
3591 struct inode *inode = page->mapping->host;
3592 u64 start = page_offset(page);
3593 u64 page_end = start + PAGE_SIZE - 1;
3597 loff_t i_size = i_size_read(inode);
3598 unsigned long end_index = i_size >> PAGE_SHIFT;
3599 unsigned long nr_written = 0;
3601 trace___extent_writepage(page, inode, wbc);
3603 WARN_ON(!PageLocked(page));
3605 ClearPageError(page);
3607 pg_offset = offset_in_page(i_size);
3608 if (page->index > end_index ||
3609 (page->index == end_index && !pg_offset)) {
3610 page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
3615 if (page->index == end_index) {
3618 userpage = kmap_atomic(page);
3619 memset(userpage + pg_offset, 0,
3620 PAGE_SIZE - pg_offset);
3621 kunmap_atomic(userpage);
3622 flush_dcache_page(page);
3625 set_page_extent_mapped(page);
3627 if (!epd->extent_locked) {
3628 ret = writepage_delalloc(BTRFS_I(inode), page, wbc, start,
3636 ret = __extent_writepage_io(BTRFS_I(inode), page, wbc, epd, i_size,
3643 /* make sure the mapping tag for page dirty gets cleared */
3644 set_page_writeback(page);
3645 end_page_writeback(page);
3647 if (PageError(page)) {
3648 ret = ret < 0 ? ret : -EIO;
3649 end_extent_writepage(page, ret, start, page_end);
3656 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3658 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
3659 TASK_UNINTERRUPTIBLE);
3662 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3664 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3665 smp_mb__after_atomic();
3666 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3670 * Lock extent buffer status and pages for writeback.
3672 * May try to flush write bio if we can't get the lock.
3674 * Return 0 if the extent buffer doesn't need to be submitted.
3675 * (E.g. the extent buffer is not dirty)
3676 * Return >0 is the extent buffer is submitted to bio.
3677 * Return <0 if something went wrong, no page is locked.
3679 static noinline_for_stack int lock_extent_buffer_for_io(struct extent_buffer *eb,
3680 struct extent_page_data *epd)
3682 struct btrfs_fs_info *fs_info = eb->fs_info;
3683 int i, num_pages, failed_page_nr;
3687 if (!btrfs_try_tree_write_lock(eb)) {
3688 ret = flush_write_bio(epd);
3692 btrfs_tree_lock(eb);
3695 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3696 btrfs_tree_unlock(eb);
3700 ret = flush_write_bio(epd);
3706 wait_on_extent_buffer_writeback(eb);
3707 btrfs_tree_lock(eb);
3708 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3710 btrfs_tree_unlock(eb);
3715 * We need to do this to prevent races in people who check if the eb is
3716 * under IO since we can end up having no IO bits set for a short period
3719 spin_lock(&eb->refs_lock);
3720 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3721 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3722 spin_unlock(&eb->refs_lock);
3723 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3724 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
3726 fs_info->dirty_metadata_batch);
3729 spin_unlock(&eb->refs_lock);
3732 btrfs_tree_unlock(eb);
3737 num_pages = num_extent_pages(eb);
3738 for (i = 0; i < num_pages; i++) {
3739 struct page *p = eb->pages[i];
3741 if (!trylock_page(p)) {
3745 err = flush_write_bio(epd);
3759 /* Unlock already locked pages */
3760 for (i = 0; i < failed_page_nr; i++)
3761 unlock_page(eb->pages[i]);
3763 * Clear EXTENT_BUFFER_WRITEBACK and wake up anyone waiting on it.
3764 * Also set back EXTENT_BUFFER_DIRTY so future attempts to this eb can
3765 * be made and undo everything done before.
3767 btrfs_tree_lock(eb);
3768 spin_lock(&eb->refs_lock);
3769 set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
3770 end_extent_buffer_writeback(eb);
3771 spin_unlock(&eb->refs_lock);
3772 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, eb->len,
3773 fs_info->dirty_metadata_batch);
3774 btrfs_clear_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3775 btrfs_tree_unlock(eb);
3779 static void set_btree_ioerr(struct page *page)
3781 struct extent_buffer *eb = (struct extent_buffer *)page->private;
3782 struct btrfs_fs_info *fs_info;
3785 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))
3789 * If we error out, we should add back the dirty_metadata_bytes
3790 * to make it consistent.
3792 fs_info = eb->fs_info;
3793 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
3794 eb->len, fs_info->dirty_metadata_batch);
3797 * If writeback for a btree extent that doesn't belong to a log tree
3798 * failed, increment the counter transaction->eb_write_errors.
3799 * We do this because while the transaction is running and before it's
3800 * committing (when we call filemap_fdata[write|wait]_range against
3801 * the btree inode), we might have
3802 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
3803 * returns an error or an error happens during writeback, when we're
3804 * committing the transaction we wouldn't know about it, since the pages
3805 * can be no longer dirty nor marked anymore for writeback (if a
3806 * subsequent modification to the extent buffer didn't happen before the
3807 * transaction commit), which makes filemap_fdata[write|wait]_range not
3808 * able to find the pages tagged with SetPageError at transaction
3809 * commit time. So if this happens we must abort the transaction,
3810 * otherwise we commit a super block with btree roots that point to
3811 * btree nodes/leafs whose content on disk is invalid - either garbage
3812 * or the content of some node/leaf from a past generation that got
3813 * cowed or deleted and is no longer valid.
3815 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
3816 * not be enough - we need to distinguish between log tree extents vs
3817 * non-log tree extents, and the next filemap_fdatawait_range() call
3818 * will catch and clear such errors in the mapping - and that call might
3819 * be from a log sync and not from a transaction commit. Also, checking
3820 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
3821 * not done and would not be reliable - the eb might have been released
3822 * from memory and reading it back again means that flag would not be
3823 * set (since it's a runtime flag, not persisted on disk).
3825 * Using the flags below in the btree inode also makes us achieve the
3826 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
3827 * writeback for all dirty pages and before filemap_fdatawait_range()
3828 * is called, the writeback for all dirty pages had already finished
3829 * with errors - because we were not using AS_EIO/AS_ENOSPC,
3830 * filemap_fdatawait_range() would return success, as it could not know
3831 * that writeback errors happened (the pages were no longer tagged for
3834 switch (eb->log_index) {
3836 set_bit(BTRFS_FS_BTREE_ERR, &eb->fs_info->flags);
3839 set_bit(BTRFS_FS_LOG1_ERR, &eb->fs_info->flags);
3842 set_bit(BTRFS_FS_LOG2_ERR, &eb->fs_info->flags);
3845 BUG(); /* unexpected, logic error */
3849 static void end_bio_extent_buffer_writepage(struct bio *bio)
3851 struct bio_vec *bvec;
3852 struct extent_buffer *eb;
3854 struct bvec_iter_all iter_all;
3856 ASSERT(!bio_flagged(bio, BIO_CLONED));
3857 bio_for_each_segment_all(bvec, bio, iter_all) {
3858 struct page *page = bvec->bv_page;
3860 eb = (struct extent_buffer *)page->private;
3862 done = atomic_dec_and_test(&eb->io_pages);
3864 if (bio->bi_status ||
3865 test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) {
3866 ClearPageUptodate(page);
3867 set_btree_ioerr(page);
3870 end_page_writeback(page);
3875 end_extent_buffer_writeback(eb);
3881 static noinline_for_stack int write_one_eb(struct extent_buffer *eb,
3882 struct writeback_control *wbc,
3883 struct extent_page_data *epd)
3885 u64 offset = eb->start;
3888 unsigned long start, end;
3889 unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META;
3892 clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
3893 num_pages = num_extent_pages(eb);
3894 atomic_set(&eb->io_pages, num_pages);
3896 /* set btree blocks beyond nritems with 0 to avoid stale content. */
3897 nritems = btrfs_header_nritems(eb);
3898 if (btrfs_header_level(eb) > 0) {
3899 end = btrfs_node_key_ptr_offset(nritems);
3901 memzero_extent_buffer(eb, end, eb->len - end);
3905 * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0
3907 start = btrfs_item_nr_offset(nritems);
3908 end = BTRFS_LEAF_DATA_OFFSET + leaf_data_end(eb);
3909 memzero_extent_buffer(eb, start, end - start);
3912 for (i = 0; i < num_pages; i++) {
3913 struct page *p = eb->pages[i];
3915 clear_page_dirty_for_io(p);
3916 set_page_writeback(p);
3917 ret = submit_extent_page(REQ_OP_WRITE | write_flags, wbc,
3918 p, offset, PAGE_SIZE, 0,
3920 end_bio_extent_buffer_writepage,
3924 if (PageWriteback(p))
3925 end_page_writeback(p);
3926 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3927 end_extent_buffer_writeback(eb);
3931 offset += PAGE_SIZE;
3932 update_nr_written(wbc, 1);
3936 if (unlikely(ret)) {
3937 for (; i < num_pages; i++) {
3938 struct page *p = eb->pages[i];
3939 clear_page_dirty_for_io(p);
3947 int btree_write_cache_pages(struct address_space *mapping,
3948 struct writeback_control *wbc)
3950 struct extent_buffer *eb, *prev_eb = NULL;
3951 struct extent_page_data epd = {
3954 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3956 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
3959 int nr_to_write_done = 0;
3960 struct pagevec pvec;
3963 pgoff_t end; /* Inclusive */
3967 pagevec_init(&pvec);
3968 if (wbc->range_cyclic) {
3969 index = mapping->writeback_index; /* Start from prev offset */
3972 * Start from the beginning does not need to cycle over the
3973 * range, mark it as scanned.
3975 scanned = (index == 0);
3977 index = wbc->range_start >> PAGE_SHIFT;
3978 end = wbc->range_end >> PAGE_SHIFT;
3981 if (wbc->sync_mode == WB_SYNC_ALL)
3982 tag = PAGECACHE_TAG_TOWRITE;
3984 tag = PAGECACHE_TAG_DIRTY;
3986 if (wbc->sync_mode == WB_SYNC_ALL)
3987 tag_pages_for_writeback(mapping, index, end);
3988 while (!done && !nr_to_write_done && (index <= end) &&
3989 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
3993 for (i = 0; i < nr_pages; i++) {
3994 struct page *page = pvec.pages[i];
3996 if (!PagePrivate(page))
3999 spin_lock(&mapping->private_lock);
4000 if (!PagePrivate(page)) {
4001 spin_unlock(&mapping->private_lock);
4005 eb = (struct extent_buffer *)page->private;
4008 * Shouldn't happen and normally this would be a BUG_ON
4009 * but no sense in crashing the users box for something
4010 * we can survive anyway.
4013 spin_unlock(&mapping->private_lock);
4017 if (eb == prev_eb) {
4018 spin_unlock(&mapping->private_lock);
4022 ret = atomic_inc_not_zero(&eb->refs);
4023 spin_unlock(&mapping->private_lock);
4028 ret = lock_extent_buffer_for_io(eb, &epd);
4030 free_extent_buffer(eb);
4032 } else if (ret < 0) {
4034 free_extent_buffer(eb);
4038 ret = write_one_eb(eb, wbc, &epd);
4041 free_extent_buffer(eb);
4044 free_extent_buffer(eb);
4047 * the filesystem may choose to bump up nr_to_write.
4048 * We have to make sure to honor the new nr_to_write
4051 nr_to_write_done = wbc->nr_to_write <= 0;
4053 pagevec_release(&pvec);
4056 if (!scanned && !done) {
4058 * We hit the last page and there is more work to be done: wrap
4059 * back to the start of the file
4067 end_write_bio(&epd, ret);
4071 * If something went wrong, don't allow any metadata write bio to be
4074 * This would prevent use-after-free if we had dirty pages not
4075 * cleaned up, which can still happen by fuzzed images.
4078 * Allowing existing tree block to be allocated for other trees.
4080 * - Log tree operations
4081 * Exiting tree blocks get allocated to log tree, bumps its
4082 * generation, then get cleaned in tree re-balance.
4083 * Such tree block will not be written back, since it's clean,
4084 * thus no WRITTEN flag set.
4085 * And after log writes back, this tree block is not traced by
4086 * any dirty extent_io_tree.
4088 * - Offending tree block gets re-dirtied from its original owner
4089 * Since it has bumped generation, no WRITTEN flag, it can be
4090 * reused without COWing. This tree block will not be traced
4091 * by btrfs_transaction::dirty_pages.
4093 * Now such dirty tree block will not be cleaned by any dirty
4094 * extent io tree. Thus we don't want to submit such wild eb
4095 * if the fs already has error.
4097 if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
4098 ret = flush_write_bio(&epd);
4101 end_write_bio(&epd, ret);
4107 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
4108 * @mapping: address space structure to write
4109 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
4110 * @data: data passed to __extent_writepage function
4112 * If a page is already under I/O, write_cache_pages() skips it, even
4113 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
4114 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
4115 * and msync() need to guarantee that all the data which was dirty at the time
4116 * the call was made get new I/O started against them. If wbc->sync_mode is
4117 * WB_SYNC_ALL then we were called for data integrity and we must wait for
4118 * existing IO to complete.
4120 static int extent_write_cache_pages(struct address_space *mapping,
4121 struct writeback_control *wbc,
4122 struct extent_page_data *epd)
4124 struct inode *inode = mapping->host;
4127 int nr_to_write_done = 0;
4128 struct pagevec pvec;
4131 pgoff_t end; /* Inclusive */
4133 int range_whole = 0;
4138 * We have to hold onto the inode so that ordered extents can do their
4139 * work when the IO finishes. The alternative to this is failing to add
4140 * an ordered extent if the igrab() fails there and that is a huge pain
4141 * to deal with, so instead just hold onto the inode throughout the
4142 * writepages operation. If it fails here we are freeing up the inode
4143 * anyway and we'd rather not waste our time writing out stuff that is
4144 * going to be truncated anyway.
4149 pagevec_init(&pvec);
4150 if (wbc->range_cyclic) {
4151 index = mapping->writeback_index; /* Start from prev offset */
4154 * Start from the beginning does not need to cycle over the
4155 * range, mark it as scanned.
4157 scanned = (index == 0);
4159 index = wbc->range_start >> PAGE_SHIFT;
4160 end = wbc->range_end >> PAGE_SHIFT;
4161 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
4167 * We do the tagged writepage as long as the snapshot flush bit is set
4168 * and we are the first one who do the filemap_flush() on this inode.
4170 * The nr_to_write == LONG_MAX is needed to make sure other flushers do
4171 * not race in and drop the bit.
4173 if (range_whole && wbc->nr_to_write == LONG_MAX &&
4174 test_and_clear_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
4175 &BTRFS_I(inode)->runtime_flags))
4176 wbc->tagged_writepages = 1;
4178 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
4179 tag = PAGECACHE_TAG_TOWRITE;
4181 tag = PAGECACHE_TAG_DIRTY;
4183 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
4184 tag_pages_for_writeback(mapping, index, end);
4186 while (!done && !nr_to_write_done && (index <= end) &&
4187 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping,
4188 &index, end, tag))) {
4191 for (i = 0; i < nr_pages; i++) {
4192 struct page *page = pvec.pages[i];
4194 done_index = page->index + 1;
4196 * At this point we hold neither the i_pages lock nor
4197 * the page lock: the page may be truncated or
4198 * invalidated (changing page->mapping to NULL),
4199 * or even swizzled back from swapper_space to
4200 * tmpfs file mapping
4202 if (!trylock_page(page)) {
4203 ret = flush_write_bio(epd);
4208 if (unlikely(page->mapping != mapping)) {
4213 if (wbc->sync_mode != WB_SYNC_NONE) {
4214 if (PageWriteback(page)) {
4215 ret = flush_write_bio(epd);
4218 wait_on_page_writeback(page);
4221 if (PageWriteback(page) ||
4222 !clear_page_dirty_for_io(page)) {
4227 ret = __extent_writepage(page, wbc, epd);
4234 * the filesystem may choose to bump up nr_to_write.
4235 * We have to make sure to honor the new nr_to_write
4238 nr_to_write_done = wbc->nr_to_write <= 0;
4240 pagevec_release(&pvec);
4243 if (!scanned && !done) {
4245 * We hit the last page and there is more work to be done: wrap
4246 * back to the start of the file
4252 * If we're looping we could run into a page that is locked by a
4253 * writer and that writer could be waiting on writeback for a
4254 * page in our current bio, and thus deadlock, so flush the
4257 ret = flush_write_bio(epd);
4262 if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole))
4263 mapping->writeback_index = done_index;
4265 btrfs_add_delayed_iput(inode);
4269 int extent_write_full_page(struct page *page, struct writeback_control *wbc)
4272 struct extent_page_data epd = {
4275 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4278 ret = __extent_writepage(page, wbc, &epd);
4281 end_write_bio(&epd, ret);
4285 ret = flush_write_bio(&epd);
4290 int extent_write_locked_range(struct inode *inode, u64 start, u64 end,
4294 struct address_space *mapping = inode->i_mapping;
4296 unsigned long nr_pages = (end - start + PAGE_SIZE) >>
4299 struct extent_page_data epd = {
4302 .sync_io = mode == WB_SYNC_ALL,
4304 struct writeback_control wbc_writepages = {
4306 .nr_to_write = nr_pages * 2,
4307 .range_start = start,
4308 .range_end = end + 1,
4309 /* We're called from an async helper function */
4310 .punt_to_cgroup = 1,
4311 .no_cgroup_owner = 1,
4314 wbc_attach_fdatawrite_inode(&wbc_writepages, inode);
4315 while (start <= end) {
4316 page = find_get_page(mapping, start >> PAGE_SHIFT);
4317 if (clear_page_dirty_for_io(page))
4318 ret = __extent_writepage(page, &wbc_writepages, &epd);
4320 btrfs_writepage_endio_finish_ordered(page, start,
4321 start + PAGE_SIZE - 1, 1);
4330 ret = flush_write_bio(&epd);
4332 end_write_bio(&epd, ret);
4334 wbc_detach_inode(&wbc_writepages);
4338 int extent_writepages(struct address_space *mapping,
4339 struct writeback_control *wbc)
4342 struct extent_page_data epd = {
4345 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4348 ret = extent_write_cache_pages(mapping, wbc, &epd);
4351 end_write_bio(&epd, ret);
4354 ret = flush_write_bio(&epd);
4358 void extent_readahead(struct readahead_control *rac)
4360 struct bio *bio = NULL;
4361 unsigned long bio_flags = 0;
4362 struct page *pagepool[16];
4363 struct extent_map *em_cached = NULL;
4364 u64 prev_em_start = (u64)-1;
4367 while ((nr = readahead_page_batch(rac, pagepool))) {
4368 u64 contig_start = page_offset(pagepool[0]);
4369 u64 contig_end = page_offset(pagepool[nr - 1]) + PAGE_SIZE - 1;
4371 ASSERT(contig_start + nr * PAGE_SIZE - 1 == contig_end);
4373 contiguous_readpages(pagepool, nr, contig_start, contig_end,
4374 &em_cached, &bio, &bio_flags, &prev_em_start);
4378 free_extent_map(em_cached);
4381 if (submit_one_bio(bio, 0, bio_flags))
4387 * basic invalidatepage code, this waits on any locked or writeback
4388 * ranges corresponding to the page, and then deletes any extent state
4389 * records from the tree
4391 int extent_invalidatepage(struct extent_io_tree *tree,
4392 struct page *page, unsigned long offset)
4394 struct extent_state *cached_state = NULL;
4395 u64 start = page_offset(page);
4396 u64 end = start + PAGE_SIZE - 1;
4397 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
4399 start += ALIGN(offset, blocksize);
4403 lock_extent_bits(tree, start, end, &cached_state);
4404 wait_on_page_writeback(page);
4405 clear_extent_bit(tree, start, end, EXTENT_LOCKED | EXTENT_DELALLOC |
4406 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state);
4411 * a helper for releasepage, this tests for areas of the page that
4412 * are locked or under IO and drops the related state bits if it is safe
4415 static int try_release_extent_state(struct extent_io_tree *tree,
4416 struct page *page, gfp_t mask)
4418 u64 start = page_offset(page);
4419 u64 end = start + PAGE_SIZE - 1;
4422 if (test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL)) {
4426 * at this point we can safely clear everything except the
4427 * locked bit and the nodatasum bit
4429 ret = __clear_extent_bit(tree, start, end,
4430 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
4431 0, 0, NULL, mask, NULL);
4433 /* if clear_extent_bit failed for enomem reasons,
4434 * we can't allow the release to continue.
4445 * a helper for releasepage. As long as there are no locked extents
4446 * in the range corresponding to the page, both state records and extent
4447 * map records are removed
4449 int try_release_extent_mapping(struct page *page, gfp_t mask)
4451 struct extent_map *em;
4452 u64 start = page_offset(page);
4453 u64 end = start + PAGE_SIZE - 1;
4454 struct btrfs_inode *btrfs_inode = BTRFS_I(page->mapping->host);
4455 struct extent_io_tree *tree = &btrfs_inode->io_tree;
4456 struct extent_map_tree *map = &btrfs_inode->extent_tree;
4458 if (gfpflags_allow_blocking(mask) &&
4459 page->mapping->host->i_size > SZ_16M) {
4461 while (start <= end) {
4462 struct btrfs_fs_info *fs_info;
4465 len = end - start + 1;
4466 write_lock(&map->lock);
4467 em = lookup_extent_mapping(map, start, len);
4469 write_unlock(&map->lock);
4472 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
4473 em->start != start) {
4474 write_unlock(&map->lock);
4475 free_extent_map(em);
4478 if (test_range_bit(tree, em->start,
4479 extent_map_end(em) - 1,
4480 EXTENT_LOCKED, 0, NULL))
4483 * If it's not in the list of modified extents, used
4484 * by a fast fsync, we can remove it. If it's being
4485 * logged we can safely remove it since fsync took an
4486 * extra reference on the em.
4488 if (list_empty(&em->list) ||
4489 test_bit(EXTENT_FLAG_LOGGING, &em->flags))
4492 * If it's in the list of modified extents, remove it
4493 * only if its generation is older then the current one,
4494 * in which case we don't need it for a fast fsync.
4495 * Otherwise don't remove it, we could be racing with an
4496 * ongoing fast fsync that could miss the new extent.
4498 fs_info = btrfs_inode->root->fs_info;
4499 spin_lock(&fs_info->trans_lock);
4500 cur_gen = fs_info->generation;
4501 spin_unlock(&fs_info->trans_lock);
4502 if (em->generation >= cur_gen)
4506 * We only remove extent maps that are not in the list of
4507 * modified extents or that are in the list but with a
4508 * generation lower then the current generation, so there
4509 * is no need to set the full fsync flag on the inode (it
4510 * hurts the fsync performance for workloads with a data
4511 * size that exceeds or is close to the system's memory).
4513 remove_extent_mapping(map, em);
4514 /* once for the rb tree */
4515 free_extent_map(em);
4517 start = extent_map_end(em);
4518 write_unlock(&map->lock);
4521 free_extent_map(em);
4523 cond_resched(); /* Allow large-extent preemption. */
4526 return try_release_extent_state(tree, page, mask);
4530 * helper function for fiemap, which doesn't want to see any holes.
4531 * This maps until we find something past 'last'
4533 static struct extent_map *get_extent_skip_holes(struct btrfs_inode *inode,
4534 u64 offset, u64 last)
4536 u64 sectorsize = btrfs_inode_sectorsize(inode);
4537 struct extent_map *em;
4544 len = last - offset;
4547 len = ALIGN(len, sectorsize);
4548 em = btrfs_get_extent_fiemap(inode, offset, len);
4549 if (IS_ERR_OR_NULL(em))
4552 /* if this isn't a hole return it */
4553 if (em->block_start != EXTENT_MAP_HOLE)
4556 /* this is a hole, advance to the next extent */
4557 offset = extent_map_end(em);
4558 free_extent_map(em);
4566 * To cache previous fiemap extent
4568 * Will be used for merging fiemap extent
4570 struct fiemap_cache {
4579 * Helper to submit fiemap extent.
4581 * Will try to merge current fiemap extent specified by @offset, @phys,
4582 * @len and @flags with cached one.
4583 * And only when we fails to merge, cached one will be submitted as
4586 * Return value is the same as fiemap_fill_next_extent().
4588 static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
4589 struct fiemap_cache *cache,
4590 u64 offset, u64 phys, u64 len, u32 flags)
4598 * Sanity check, extent_fiemap() should have ensured that new
4599 * fiemap extent won't overlap with cached one.
4602 * NOTE: Physical address can overlap, due to compression
4604 if (cache->offset + cache->len > offset) {
4610 * Only merges fiemap extents if
4611 * 1) Their logical addresses are continuous
4613 * 2) Their physical addresses are continuous
4614 * So truly compressed (physical size smaller than logical size)
4615 * extents won't get merged with each other
4617 * 3) Share same flags except FIEMAP_EXTENT_LAST
4618 * So regular extent won't get merged with prealloc extent
4620 if (cache->offset + cache->len == offset &&
4621 cache->phys + cache->len == phys &&
4622 (cache->flags & ~FIEMAP_EXTENT_LAST) ==
4623 (flags & ~FIEMAP_EXTENT_LAST)) {
4625 cache->flags |= flags;
4626 goto try_submit_last;
4629 /* Not mergeable, need to submit cached one */
4630 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
4631 cache->len, cache->flags);
4632 cache->cached = false;
4636 cache->cached = true;
4637 cache->offset = offset;
4640 cache->flags = flags;
4642 if (cache->flags & FIEMAP_EXTENT_LAST) {
4643 ret = fiemap_fill_next_extent(fieinfo, cache->offset,
4644 cache->phys, cache->len, cache->flags);
4645 cache->cached = false;
4651 * Emit last fiemap cache
4653 * The last fiemap cache may still be cached in the following case:
4655 * |<- Fiemap range ->|
4656 * |<------------ First extent ----------->|
4658 * In this case, the first extent range will be cached but not emitted.
4659 * So we must emit it before ending extent_fiemap().
4661 static int emit_last_fiemap_cache(struct fiemap_extent_info *fieinfo,
4662 struct fiemap_cache *cache)
4669 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
4670 cache->len, cache->flags);
4671 cache->cached = false;
4677 int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,
4682 u64 max = start + len;
4686 u64 last_for_get_extent = 0;
4688 u64 isize = i_size_read(&inode->vfs_inode);
4689 struct btrfs_key found_key;
4690 struct extent_map *em = NULL;
4691 struct extent_state *cached_state = NULL;
4692 struct btrfs_path *path;
4693 struct btrfs_root *root = inode->root;
4694 struct fiemap_cache cache = { 0 };
4695 struct ulist *roots;
4696 struct ulist *tmp_ulist;
4705 path = btrfs_alloc_path();
4709 roots = ulist_alloc(GFP_KERNEL);
4710 tmp_ulist = ulist_alloc(GFP_KERNEL);
4711 if (!roots || !tmp_ulist) {
4713 goto out_free_ulist;
4716 start = round_down(start, btrfs_inode_sectorsize(inode));
4717 len = round_up(max, btrfs_inode_sectorsize(inode)) - start;
4720 * lookup the last file extent. We're not using i_size here
4721 * because there might be preallocation past i_size
4723 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode), -1,
4726 goto out_free_ulist;
4734 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
4735 found_type = found_key.type;
4737 /* No extents, but there might be delalloc bits */
4738 if (found_key.objectid != btrfs_ino(inode) ||
4739 found_type != BTRFS_EXTENT_DATA_KEY) {
4740 /* have to trust i_size as the end */
4742 last_for_get_extent = isize;
4745 * remember the start of the last extent. There are a
4746 * bunch of different factors that go into the length of the
4747 * extent, so its much less complex to remember where it started
4749 last = found_key.offset;
4750 last_for_get_extent = last + 1;
4752 btrfs_release_path(path);
4755 * we might have some extents allocated but more delalloc past those
4756 * extents. so, we trust isize unless the start of the last extent is
4761 last_for_get_extent = isize;
4764 lock_extent_bits(&inode->io_tree, start, start + len - 1,
4767 em = get_extent_skip_holes(inode, start, last_for_get_extent);
4776 u64 offset_in_extent = 0;
4778 /* break if the extent we found is outside the range */
4779 if (em->start >= max || extent_map_end(em) < off)
4783 * get_extent may return an extent that starts before our
4784 * requested range. We have to make sure the ranges
4785 * we return to fiemap always move forward and don't
4786 * overlap, so adjust the offsets here
4788 em_start = max(em->start, off);
4791 * record the offset from the start of the extent
4792 * for adjusting the disk offset below. Only do this if the
4793 * extent isn't compressed since our in ram offset may be past
4794 * what we have actually allocated on disk.
4796 if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4797 offset_in_extent = em_start - em->start;
4798 em_end = extent_map_end(em);
4799 em_len = em_end - em_start;
4801 if (em->block_start < EXTENT_MAP_LAST_BYTE)
4802 disko = em->block_start + offset_in_extent;
4807 * bump off for our next call to get_extent
4809 off = extent_map_end(em);
4813 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
4815 flags |= FIEMAP_EXTENT_LAST;
4816 } else if (em->block_start == EXTENT_MAP_INLINE) {
4817 flags |= (FIEMAP_EXTENT_DATA_INLINE |
4818 FIEMAP_EXTENT_NOT_ALIGNED);
4819 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
4820 flags |= (FIEMAP_EXTENT_DELALLOC |
4821 FIEMAP_EXTENT_UNKNOWN);
4822 } else if (fieinfo->fi_extents_max) {
4823 u64 bytenr = em->block_start -
4824 (em->start - em->orig_start);
4827 * As btrfs supports shared space, this information
4828 * can be exported to userspace tools via
4829 * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0
4830 * then we're just getting a count and we can skip the
4833 ret = btrfs_check_shared(root, btrfs_ino(inode),
4834 bytenr, roots, tmp_ulist);
4838 flags |= FIEMAP_EXTENT_SHARED;
4841 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4842 flags |= FIEMAP_EXTENT_ENCODED;
4843 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4844 flags |= FIEMAP_EXTENT_UNWRITTEN;
4846 free_extent_map(em);
4848 if ((em_start >= last) || em_len == (u64)-1 ||
4849 (last == (u64)-1 && isize <= em_end)) {
4850 flags |= FIEMAP_EXTENT_LAST;
4854 /* now scan forward to see if this is really the last extent. */
4855 em = get_extent_skip_holes(inode, off, last_for_get_extent);
4861 flags |= FIEMAP_EXTENT_LAST;
4864 ret = emit_fiemap_extent(fieinfo, &cache, em_start, disko,
4874 ret = emit_last_fiemap_cache(fieinfo, &cache);
4875 free_extent_map(em);
4877 unlock_extent_cached(&inode->io_tree, start, start + len - 1,
4881 btrfs_free_path(path);
4883 ulist_free(tmp_ulist);
4887 static void __free_extent_buffer(struct extent_buffer *eb)
4889 kmem_cache_free(extent_buffer_cache, eb);
4892 int extent_buffer_under_io(const struct extent_buffer *eb)
4894 return (atomic_read(&eb->io_pages) ||
4895 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4896 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4900 * Release all pages attached to the extent buffer.
4902 static void btrfs_release_extent_buffer_pages(struct extent_buffer *eb)
4906 int mapped = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
4908 BUG_ON(extent_buffer_under_io(eb));
4910 num_pages = num_extent_pages(eb);
4911 for (i = 0; i < num_pages; i++) {
4912 struct page *page = eb->pages[i];
4917 spin_lock(&page->mapping->private_lock);
4919 * We do this since we'll remove the pages after we've
4920 * removed the eb from the radix tree, so we could race
4921 * and have this page now attached to the new eb. So
4922 * only clear page_private if it's still connected to
4925 if (PagePrivate(page) &&
4926 page->private == (unsigned long)eb) {
4927 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4928 BUG_ON(PageDirty(page));
4929 BUG_ON(PageWriteback(page));
4931 * We need to make sure we haven't be attached
4934 detach_page_private(page);
4938 spin_unlock(&page->mapping->private_lock);
4940 /* One for when we allocated the page */
4946 * Helper for releasing the extent buffer.
4948 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4950 btrfs_release_extent_buffer_pages(eb);
4951 btrfs_leak_debug_del(&eb->fs_info->eb_leak_lock, &eb->leak_list);
4952 __free_extent_buffer(eb);
4955 static struct extent_buffer *
4956 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
4959 struct extent_buffer *eb = NULL;
4961 eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL);
4964 eb->fs_info = fs_info;
4966 init_rwsem(&eb->lock);
4967 eb->lock_recursed = false;
4969 btrfs_leak_debug_add(&fs_info->eb_leak_lock, &eb->leak_list,
4970 &fs_info->allocated_ebs);
4972 spin_lock_init(&eb->refs_lock);
4973 atomic_set(&eb->refs, 1);
4974 atomic_set(&eb->io_pages, 0);
4977 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4979 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4980 > MAX_INLINE_EXTENT_BUFFER_SIZE);
4981 BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE);
4986 struct extent_buffer *btrfs_clone_extent_buffer(const struct extent_buffer *src)
4990 struct extent_buffer *new;
4991 int num_pages = num_extent_pages(src);
4993 new = __alloc_extent_buffer(src->fs_info, src->start, src->len);
4997 for (i = 0; i < num_pages; i++) {
4998 p = alloc_page(GFP_NOFS);
5000 btrfs_release_extent_buffer(new);
5003 attach_extent_buffer_page(new, p);
5004 WARN_ON(PageDirty(p));
5007 copy_page(page_address(p), page_address(src->pages[i]));
5010 set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
5011 set_bit(EXTENT_BUFFER_UNMAPPED, &new->bflags);
5016 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
5017 u64 start, unsigned long len)
5019 struct extent_buffer *eb;
5023 eb = __alloc_extent_buffer(fs_info, start, len);
5027 num_pages = num_extent_pages(eb);
5028 for (i = 0; i < num_pages; i++) {
5029 eb->pages[i] = alloc_page(GFP_NOFS);
5033 set_extent_buffer_uptodate(eb);
5034 btrfs_set_header_nritems(eb, 0);
5035 set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
5040 __free_page(eb->pages[i - 1]);
5041 __free_extent_buffer(eb);
5045 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
5048 return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize);
5051 static void check_buffer_tree_ref(struct extent_buffer *eb)
5055 * The TREE_REF bit is first set when the extent_buffer is added
5056 * to the radix tree. It is also reset, if unset, when a new reference
5057 * is created by find_extent_buffer.
5059 * It is only cleared in two cases: freeing the last non-tree
5060 * reference to the extent_buffer when its STALE bit is set or
5061 * calling releasepage when the tree reference is the only reference.
5063 * In both cases, care is taken to ensure that the extent_buffer's
5064 * pages are not under io. However, releasepage can be concurrently
5065 * called with creating new references, which is prone to race
5066 * conditions between the calls to check_buffer_tree_ref in those
5067 * codepaths and clearing TREE_REF in try_release_extent_buffer.
5069 * The actual lifetime of the extent_buffer in the radix tree is
5070 * adequately protected by the refcount, but the TREE_REF bit and
5071 * its corresponding reference are not. To protect against this
5072 * class of races, we call check_buffer_tree_ref from the codepaths
5073 * which trigger io after they set eb->io_pages. Note that once io is
5074 * initiated, TREE_REF can no longer be cleared, so that is the
5075 * moment at which any such race is best fixed.
5077 refs = atomic_read(&eb->refs);
5078 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5081 spin_lock(&eb->refs_lock);
5082 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5083 atomic_inc(&eb->refs);
5084 spin_unlock(&eb->refs_lock);
5087 static void mark_extent_buffer_accessed(struct extent_buffer *eb,
5088 struct page *accessed)
5092 check_buffer_tree_ref(eb);
5094 num_pages = num_extent_pages(eb);
5095 for (i = 0; i < num_pages; i++) {
5096 struct page *p = eb->pages[i];
5099 mark_page_accessed(p);
5103 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
5106 struct extent_buffer *eb;
5109 eb = radix_tree_lookup(&fs_info->buffer_radix,
5110 start >> fs_info->sectorsize_bits);
5111 if (eb && atomic_inc_not_zero(&eb->refs)) {
5114 * Lock our eb's refs_lock to avoid races with
5115 * free_extent_buffer. When we get our eb it might be flagged
5116 * with EXTENT_BUFFER_STALE and another task running
5117 * free_extent_buffer might have seen that flag set,
5118 * eb->refs == 2, that the buffer isn't under IO (dirty and
5119 * writeback flags not set) and it's still in the tree (flag
5120 * EXTENT_BUFFER_TREE_REF set), therefore being in the process
5121 * of decrementing the extent buffer's reference count twice.
5122 * So here we could race and increment the eb's reference count,
5123 * clear its stale flag, mark it as dirty and drop our reference
5124 * before the other task finishes executing free_extent_buffer,
5125 * which would later result in an attempt to free an extent
5126 * buffer that is dirty.
5128 if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) {
5129 spin_lock(&eb->refs_lock);
5130 spin_unlock(&eb->refs_lock);
5132 mark_extent_buffer_accessed(eb, NULL);
5140 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5141 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
5144 struct extent_buffer *eb, *exists = NULL;
5147 eb = find_extent_buffer(fs_info, start);
5150 eb = alloc_dummy_extent_buffer(fs_info, start);
5152 return ERR_PTR(-ENOMEM);
5153 eb->fs_info = fs_info;
5155 ret = radix_tree_preload(GFP_NOFS);
5157 exists = ERR_PTR(ret);
5160 spin_lock(&fs_info->buffer_lock);
5161 ret = radix_tree_insert(&fs_info->buffer_radix,
5162 start >> fs_info->sectorsize_bits, eb);
5163 spin_unlock(&fs_info->buffer_lock);
5164 radix_tree_preload_end();
5165 if (ret == -EEXIST) {
5166 exists = find_extent_buffer(fs_info, start);
5172 check_buffer_tree_ref(eb);
5173 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
5177 btrfs_release_extent_buffer(eb);
5182 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
5185 unsigned long len = fs_info->nodesize;
5188 unsigned long index = start >> PAGE_SHIFT;
5189 struct extent_buffer *eb;
5190 struct extent_buffer *exists = NULL;
5192 struct address_space *mapping = fs_info->btree_inode->i_mapping;
5196 if (!IS_ALIGNED(start, fs_info->sectorsize)) {
5197 btrfs_err(fs_info, "bad tree block start %llu", start);
5198 return ERR_PTR(-EINVAL);
5201 eb = find_extent_buffer(fs_info, start);
5205 eb = __alloc_extent_buffer(fs_info, start, len);
5207 return ERR_PTR(-ENOMEM);
5209 num_pages = num_extent_pages(eb);
5210 for (i = 0; i < num_pages; i++, index++) {
5211 p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL);
5213 exists = ERR_PTR(-ENOMEM);
5217 spin_lock(&mapping->private_lock);
5218 if (PagePrivate(p)) {
5220 * We could have already allocated an eb for this page
5221 * and attached one so lets see if we can get a ref on
5222 * the existing eb, and if we can we know it's good and
5223 * we can just return that one, else we know we can just
5224 * overwrite page->private.
5226 exists = (struct extent_buffer *)p->private;
5227 if (atomic_inc_not_zero(&exists->refs)) {
5228 spin_unlock(&mapping->private_lock);
5231 mark_extent_buffer_accessed(exists, p);
5237 * Do this so attach doesn't complain and we need to
5238 * drop the ref the old guy had.
5240 ClearPagePrivate(p);
5241 WARN_ON(PageDirty(p));
5244 attach_extent_buffer_page(eb, p);
5245 spin_unlock(&mapping->private_lock);
5246 WARN_ON(PageDirty(p));
5248 if (!PageUptodate(p))
5252 * We can't unlock the pages just yet since the extent buffer
5253 * hasn't been properly inserted in the radix tree, this
5254 * opens a race with btree_releasepage which can free a page
5255 * while we are still filling in all pages for the buffer and
5260 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5262 ret = radix_tree_preload(GFP_NOFS);
5264 exists = ERR_PTR(ret);
5268 spin_lock(&fs_info->buffer_lock);
5269 ret = radix_tree_insert(&fs_info->buffer_radix,
5270 start >> fs_info->sectorsize_bits, eb);
5271 spin_unlock(&fs_info->buffer_lock);
5272 radix_tree_preload_end();
5273 if (ret == -EEXIST) {
5274 exists = find_extent_buffer(fs_info, start);
5280 /* add one reference for the tree */
5281 check_buffer_tree_ref(eb);
5282 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
5285 * Now it's safe to unlock the pages because any calls to
5286 * btree_releasepage will correctly detect that a page belongs to a
5287 * live buffer and won't free them prematurely.
5289 for (i = 0; i < num_pages; i++)
5290 unlock_page(eb->pages[i]);
5294 WARN_ON(!atomic_dec_and_test(&eb->refs));
5295 for (i = 0; i < num_pages; i++) {
5297 unlock_page(eb->pages[i]);
5300 btrfs_release_extent_buffer(eb);
5304 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
5306 struct extent_buffer *eb =
5307 container_of(head, struct extent_buffer, rcu_head);
5309 __free_extent_buffer(eb);
5312 static int release_extent_buffer(struct extent_buffer *eb)
5313 __releases(&eb->refs_lock)
5315 lockdep_assert_held(&eb->refs_lock);
5317 WARN_ON(atomic_read(&eb->refs) == 0);
5318 if (atomic_dec_and_test(&eb->refs)) {
5319 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
5320 struct btrfs_fs_info *fs_info = eb->fs_info;
5322 spin_unlock(&eb->refs_lock);
5324 spin_lock(&fs_info->buffer_lock);
5325 radix_tree_delete(&fs_info->buffer_radix,
5326 eb->start >> fs_info->sectorsize_bits);
5327 spin_unlock(&fs_info->buffer_lock);
5329 spin_unlock(&eb->refs_lock);
5332 btrfs_leak_debug_del(&eb->fs_info->eb_leak_lock, &eb->leak_list);
5333 /* Should be safe to release our pages at this point */
5334 btrfs_release_extent_buffer_pages(eb);
5335 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5336 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags))) {
5337 __free_extent_buffer(eb);
5341 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
5344 spin_unlock(&eb->refs_lock);
5349 void free_extent_buffer(struct extent_buffer *eb)
5357 refs = atomic_read(&eb->refs);
5358 if ((!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && refs <= 3)
5359 || (test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) &&
5362 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
5367 spin_lock(&eb->refs_lock);
5368 if (atomic_read(&eb->refs) == 2 &&
5369 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
5370 !extent_buffer_under_io(eb) &&
5371 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5372 atomic_dec(&eb->refs);
5375 * I know this is terrible, but it's temporary until we stop tracking
5376 * the uptodate bits and such for the extent buffers.
5378 release_extent_buffer(eb);
5381 void free_extent_buffer_stale(struct extent_buffer *eb)
5386 spin_lock(&eb->refs_lock);
5387 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
5389 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
5390 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5391 atomic_dec(&eb->refs);
5392 release_extent_buffer(eb);
5395 void clear_extent_buffer_dirty(const struct extent_buffer *eb)
5401 num_pages = num_extent_pages(eb);
5403 for (i = 0; i < num_pages; i++) {
5404 page = eb->pages[i];
5405 if (!PageDirty(page))
5409 WARN_ON(!PagePrivate(page));
5411 clear_page_dirty_for_io(page);
5412 xa_lock_irq(&page->mapping->i_pages);
5413 if (!PageDirty(page))
5414 __xa_clear_mark(&page->mapping->i_pages,
5415 page_index(page), PAGECACHE_TAG_DIRTY);
5416 xa_unlock_irq(&page->mapping->i_pages);
5417 ClearPageError(page);
5420 WARN_ON(atomic_read(&eb->refs) == 0);
5423 bool set_extent_buffer_dirty(struct extent_buffer *eb)
5429 check_buffer_tree_ref(eb);
5431 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
5433 num_pages = num_extent_pages(eb);
5434 WARN_ON(atomic_read(&eb->refs) == 0);
5435 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
5438 for (i = 0; i < num_pages; i++)
5439 set_page_dirty(eb->pages[i]);
5441 #ifdef CONFIG_BTRFS_DEBUG
5442 for (i = 0; i < num_pages; i++)
5443 ASSERT(PageDirty(eb->pages[i]));
5449 void clear_extent_buffer_uptodate(struct extent_buffer *eb)
5455 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5456 num_pages = num_extent_pages(eb);
5457 for (i = 0; i < num_pages; i++) {
5458 page = eb->pages[i];
5460 ClearPageUptodate(page);
5464 void set_extent_buffer_uptodate(struct extent_buffer *eb)
5470 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5471 num_pages = num_extent_pages(eb);
5472 for (i = 0; i < num_pages; i++) {
5473 page = eb->pages[i];
5474 SetPageUptodate(page);
5478 int read_extent_buffer_pages(struct extent_buffer *eb, int wait, int mirror_num)
5484 int locked_pages = 0;
5485 int all_uptodate = 1;
5487 unsigned long num_reads = 0;
5488 struct bio *bio = NULL;
5489 unsigned long bio_flags = 0;
5491 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
5494 num_pages = num_extent_pages(eb);
5495 for (i = 0; i < num_pages; i++) {
5496 page = eb->pages[i];
5497 if (wait == WAIT_NONE) {
5498 if (!trylock_page(page))
5506 * We need to firstly lock all pages to make sure that
5507 * the uptodate bit of our pages won't be affected by
5508 * clear_extent_buffer_uptodate().
5510 for (i = 0; i < num_pages; i++) {
5511 page = eb->pages[i];
5512 if (!PageUptodate(page)) {
5519 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5523 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
5524 eb->read_mirror = 0;
5525 atomic_set(&eb->io_pages, num_reads);
5527 * It is possible for releasepage to clear the TREE_REF bit before we
5528 * set io_pages. See check_buffer_tree_ref for a more detailed comment.
5530 check_buffer_tree_ref(eb);
5531 for (i = 0; i < num_pages; i++) {
5532 page = eb->pages[i];
5534 if (!PageUptodate(page)) {
5536 atomic_dec(&eb->io_pages);
5541 ClearPageError(page);
5542 err = submit_extent_page(REQ_OP_READ | REQ_META, NULL,
5543 page, page_offset(page), PAGE_SIZE, 0,
5544 &bio, end_bio_extent_readpage,
5545 mirror_num, 0, 0, false);
5548 * We failed to submit the bio so it's the
5549 * caller's responsibility to perform cleanup
5550 * i.e unlock page/set error bit.
5555 atomic_dec(&eb->io_pages);
5563 err = submit_one_bio(bio, mirror_num, bio_flags);
5568 if (ret || wait != WAIT_COMPLETE)
5571 for (i = 0; i < num_pages; i++) {
5572 page = eb->pages[i];
5573 wait_on_page_locked(page);
5574 if (!PageUptodate(page))
5581 while (locked_pages > 0) {
5583 page = eb->pages[locked_pages];
5589 static bool report_eb_range(const struct extent_buffer *eb, unsigned long start,
5592 btrfs_warn(eb->fs_info,
5593 "access to eb bytenr %llu len %lu out of range start %lu len %lu",
5594 eb->start, eb->len, start, len);
5595 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
5601 * Check if the [start, start + len) range is valid before reading/writing
5603 * NOTE: @start and @len are offset inside the eb, not logical address.
5605 * Caller should not touch the dst/src memory if this function returns error.
5607 static inline int check_eb_range(const struct extent_buffer *eb,
5608 unsigned long start, unsigned long len)
5610 unsigned long offset;
5612 /* start, start + len should not go beyond eb->len nor overflow */
5613 if (unlikely(check_add_overflow(start, len, &offset) || offset > eb->len))
5614 return report_eb_range(eb, start, len);
5619 void read_extent_buffer(const struct extent_buffer *eb, void *dstv,
5620 unsigned long start, unsigned long len)
5626 char *dst = (char *)dstv;
5627 unsigned long i = start >> PAGE_SHIFT;
5629 if (check_eb_range(eb, start, len))
5632 offset = offset_in_page(start);
5635 page = eb->pages[i];
5637 cur = min(len, (PAGE_SIZE - offset));
5638 kaddr = page_address(page);
5639 memcpy(dst, kaddr + offset, cur);
5648 int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb,
5650 unsigned long start, unsigned long len)
5656 char __user *dst = (char __user *)dstv;
5657 unsigned long i = start >> PAGE_SHIFT;
5660 WARN_ON(start > eb->len);
5661 WARN_ON(start + len > eb->start + eb->len);
5663 offset = offset_in_page(start);
5666 page = eb->pages[i];
5668 cur = min(len, (PAGE_SIZE - offset));
5669 kaddr = page_address(page);
5670 if (copy_to_user_nofault(dst, kaddr + offset, cur)) {
5684 int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,
5685 unsigned long start, unsigned long len)
5691 char *ptr = (char *)ptrv;
5692 unsigned long i = start >> PAGE_SHIFT;
5695 if (check_eb_range(eb, start, len))
5698 offset = offset_in_page(start);
5701 page = eb->pages[i];
5703 cur = min(len, (PAGE_SIZE - offset));
5705 kaddr = page_address(page);
5706 ret = memcmp(ptr, kaddr + offset, cur);
5718 void write_extent_buffer_chunk_tree_uuid(const struct extent_buffer *eb,
5723 WARN_ON(!PageUptodate(eb->pages[0]));
5724 kaddr = page_address(eb->pages[0]);
5725 memcpy(kaddr + offsetof(struct btrfs_header, chunk_tree_uuid), srcv,
5729 void write_extent_buffer_fsid(const struct extent_buffer *eb, const void *srcv)
5733 WARN_ON(!PageUptodate(eb->pages[0]));
5734 kaddr = page_address(eb->pages[0]);
5735 memcpy(kaddr + offsetof(struct btrfs_header, fsid), srcv,
5739 void write_extent_buffer(const struct extent_buffer *eb, const void *srcv,
5740 unsigned long start, unsigned long len)
5746 char *src = (char *)srcv;
5747 unsigned long i = start >> PAGE_SHIFT;
5749 if (check_eb_range(eb, start, len))
5752 offset = offset_in_page(start);
5755 page = eb->pages[i];
5756 WARN_ON(!PageUptodate(page));
5758 cur = min(len, PAGE_SIZE - offset);
5759 kaddr = page_address(page);
5760 memcpy(kaddr + offset, src, cur);
5769 void memzero_extent_buffer(const struct extent_buffer *eb, unsigned long start,
5776 unsigned long i = start >> PAGE_SHIFT;
5778 if (check_eb_range(eb, start, len))
5781 offset = offset_in_page(start);
5784 page = eb->pages[i];
5785 WARN_ON(!PageUptodate(page));
5787 cur = min(len, PAGE_SIZE - offset);
5788 kaddr = page_address(page);
5789 memset(kaddr + offset, 0, cur);
5797 void copy_extent_buffer_full(const struct extent_buffer *dst,
5798 const struct extent_buffer *src)
5803 ASSERT(dst->len == src->len);
5805 num_pages = num_extent_pages(dst);
5806 for (i = 0; i < num_pages; i++)
5807 copy_page(page_address(dst->pages[i]),
5808 page_address(src->pages[i]));
5811 void copy_extent_buffer(const struct extent_buffer *dst,
5812 const struct extent_buffer *src,
5813 unsigned long dst_offset, unsigned long src_offset,
5816 u64 dst_len = dst->len;
5821 unsigned long i = dst_offset >> PAGE_SHIFT;
5823 if (check_eb_range(dst, dst_offset, len) ||
5824 check_eb_range(src, src_offset, len))
5827 WARN_ON(src->len != dst_len);
5829 offset = offset_in_page(dst_offset);
5832 page = dst->pages[i];
5833 WARN_ON(!PageUptodate(page));
5835 cur = min(len, (unsigned long)(PAGE_SIZE - offset));
5837 kaddr = page_address(page);
5838 read_extent_buffer(src, kaddr + offset, src_offset, cur);
5848 * eb_bitmap_offset() - calculate the page and offset of the byte containing the
5850 * @eb: the extent buffer
5851 * @start: offset of the bitmap item in the extent buffer
5853 * @page_index: return index of the page in the extent buffer that contains the
5855 * @page_offset: return offset into the page given by page_index
5857 * This helper hides the ugliness of finding the byte in an extent buffer which
5858 * contains a given bit.
5860 static inline void eb_bitmap_offset(const struct extent_buffer *eb,
5861 unsigned long start, unsigned long nr,
5862 unsigned long *page_index,
5863 size_t *page_offset)
5865 size_t byte_offset = BIT_BYTE(nr);
5869 * The byte we want is the offset of the extent buffer + the offset of
5870 * the bitmap item in the extent buffer + the offset of the byte in the
5873 offset = start + byte_offset;
5875 *page_index = offset >> PAGE_SHIFT;
5876 *page_offset = offset_in_page(offset);
5880 * extent_buffer_test_bit - determine whether a bit in a bitmap item is set
5881 * @eb: the extent buffer
5882 * @start: offset of the bitmap item in the extent buffer
5883 * @nr: bit number to test
5885 int extent_buffer_test_bit(const struct extent_buffer *eb, unsigned long start,
5893 eb_bitmap_offset(eb, start, nr, &i, &offset);
5894 page = eb->pages[i];
5895 WARN_ON(!PageUptodate(page));
5896 kaddr = page_address(page);
5897 return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1)));
5901 * extent_buffer_bitmap_set - set an area of a bitmap
5902 * @eb: the extent buffer
5903 * @start: offset of the bitmap item in the extent buffer
5904 * @pos: bit number of the first bit
5905 * @len: number of bits to set
5907 void extent_buffer_bitmap_set(const struct extent_buffer *eb, unsigned long start,
5908 unsigned long pos, unsigned long len)
5914 const unsigned int size = pos + len;
5915 int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5916 u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos);
5918 eb_bitmap_offset(eb, start, pos, &i, &offset);
5919 page = eb->pages[i];
5920 WARN_ON(!PageUptodate(page));
5921 kaddr = page_address(page);
5923 while (len >= bits_to_set) {
5924 kaddr[offset] |= mask_to_set;
5926 bits_to_set = BITS_PER_BYTE;
5928 if (++offset >= PAGE_SIZE && len > 0) {
5930 page = eb->pages[++i];
5931 WARN_ON(!PageUptodate(page));
5932 kaddr = page_address(page);
5936 mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
5937 kaddr[offset] |= mask_to_set;
5943 * extent_buffer_bitmap_clear - clear an area of a bitmap
5944 * @eb: the extent buffer
5945 * @start: offset of the bitmap item in the extent buffer
5946 * @pos: bit number of the first bit
5947 * @len: number of bits to clear
5949 void extent_buffer_bitmap_clear(const struct extent_buffer *eb,
5950 unsigned long start, unsigned long pos,
5957 const unsigned int size = pos + len;
5958 int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5959 u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos);
5961 eb_bitmap_offset(eb, start, pos, &i, &offset);
5962 page = eb->pages[i];
5963 WARN_ON(!PageUptodate(page));
5964 kaddr = page_address(page);
5966 while (len >= bits_to_clear) {
5967 kaddr[offset] &= ~mask_to_clear;
5968 len -= bits_to_clear;
5969 bits_to_clear = BITS_PER_BYTE;
5971 if (++offset >= PAGE_SIZE && len > 0) {
5973 page = eb->pages[++i];
5974 WARN_ON(!PageUptodate(page));
5975 kaddr = page_address(page);
5979 mask_to_clear &= BITMAP_LAST_BYTE_MASK(size);
5980 kaddr[offset] &= ~mask_to_clear;
5984 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
5986 unsigned long distance = (src > dst) ? src - dst : dst - src;
5987 return distance < len;
5990 static void copy_pages(struct page *dst_page, struct page *src_page,
5991 unsigned long dst_off, unsigned long src_off,
5994 char *dst_kaddr = page_address(dst_page);
5996 int must_memmove = 0;
5998 if (dst_page != src_page) {
5999 src_kaddr = page_address(src_page);
6001 src_kaddr = dst_kaddr;
6002 if (areas_overlap(src_off, dst_off, len))
6007 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
6009 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
6012 void memcpy_extent_buffer(const struct extent_buffer *dst,
6013 unsigned long dst_offset, unsigned long src_offset,
6017 size_t dst_off_in_page;
6018 size_t src_off_in_page;
6019 unsigned long dst_i;
6020 unsigned long src_i;
6022 if (check_eb_range(dst, dst_offset, len) ||
6023 check_eb_range(dst, src_offset, len))
6027 dst_off_in_page = offset_in_page(dst_offset);
6028 src_off_in_page = offset_in_page(src_offset);
6030 dst_i = dst_offset >> PAGE_SHIFT;
6031 src_i = src_offset >> PAGE_SHIFT;
6033 cur = min(len, (unsigned long)(PAGE_SIZE -
6035 cur = min_t(unsigned long, cur,
6036 (unsigned long)(PAGE_SIZE - dst_off_in_page));
6038 copy_pages(dst->pages[dst_i], dst->pages[src_i],
6039 dst_off_in_page, src_off_in_page, cur);
6047 void memmove_extent_buffer(const struct extent_buffer *dst,
6048 unsigned long dst_offset, unsigned long src_offset,
6052 size_t dst_off_in_page;
6053 size_t src_off_in_page;
6054 unsigned long dst_end = dst_offset + len - 1;
6055 unsigned long src_end = src_offset + len - 1;
6056 unsigned long dst_i;
6057 unsigned long src_i;
6059 if (check_eb_range(dst, dst_offset, len) ||
6060 check_eb_range(dst, src_offset, len))
6062 if (dst_offset < src_offset) {
6063 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
6067 dst_i = dst_end >> PAGE_SHIFT;
6068 src_i = src_end >> PAGE_SHIFT;
6070 dst_off_in_page = offset_in_page(dst_end);
6071 src_off_in_page = offset_in_page(src_end);
6073 cur = min_t(unsigned long, len, src_off_in_page + 1);
6074 cur = min(cur, dst_off_in_page + 1);
6075 copy_pages(dst->pages[dst_i], dst->pages[src_i],
6076 dst_off_in_page - cur + 1,
6077 src_off_in_page - cur + 1, cur);
6085 int try_release_extent_buffer(struct page *page)
6087 struct extent_buffer *eb;
6090 * We need to make sure nobody is attaching this page to an eb right
6093 spin_lock(&page->mapping->private_lock);
6094 if (!PagePrivate(page)) {
6095 spin_unlock(&page->mapping->private_lock);
6099 eb = (struct extent_buffer *)page->private;
6103 * This is a little awful but should be ok, we need to make sure that
6104 * the eb doesn't disappear out from under us while we're looking at
6107 spin_lock(&eb->refs_lock);
6108 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
6109 spin_unlock(&eb->refs_lock);
6110 spin_unlock(&page->mapping->private_lock);
6113 spin_unlock(&page->mapping->private_lock);
6116 * If tree ref isn't set then we know the ref on this eb is a real ref,
6117 * so just return, this page will likely be freed soon anyway.
6119 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
6120 spin_unlock(&eb->refs_lock);
6124 return release_extent_buffer(eb);