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_map.h"
19 #include "btrfs_inode.h"
21 #include "check-integrity.h"
23 #include "rcu-string.h"
27 static struct kmem_cache *extent_state_cache;
28 static struct kmem_cache *extent_buffer_cache;
29 static struct bio_set btrfs_bioset;
31 static inline bool extent_state_in_tree(const struct extent_state *state)
33 return !RB_EMPTY_NODE(&state->rb_node);
36 #ifdef CONFIG_BTRFS_DEBUG
37 static LIST_HEAD(buffers);
38 static LIST_HEAD(states);
40 static DEFINE_SPINLOCK(leak_lock);
43 void btrfs_leak_debug_add(struct list_head *new, struct list_head *head)
47 spin_lock_irqsave(&leak_lock, flags);
49 spin_unlock_irqrestore(&leak_lock, flags);
53 void btrfs_leak_debug_del(struct list_head *entry)
57 spin_lock_irqsave(&leak_lock, flags);
59 spin_unlock_irqrestore(&leak_lock, flags);
63 void btrfs_leak_debug_check(void)
65 struct extent_state *state;
66 struct extent_buffer *eb;
68 while (!list_empty(&states)) {
69 state = list_entry(states.next, struct extent_state, leak_list);
70 pr_err("BTRFS: state leak: start %llu end %llu state %u in tree %d refs %d\n",
71 state->start, state->end, state->state,
72 extent_state_in_tree(state),
73 refcount_read(&state->refs));
74 list_del(&state->leak_list);
75 kmem_cache_free(extent_state_cache, state);
78 while (!list_empty(&buffers)) {
79 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
80 pr_err("BTRFS: buffer leak start %llu len %lu refs %d bflags %lu\n",
81 eb->start, eb->len, atomic_read(&eb->refs), eb->bflags);
82 list_del(&eb->leak_list);
83 kmem_cache_free(extent_buffer_cache, eb);
87 #define btrfs_debug_check_extent_io_range(tree, start, end) \
88 __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end))
89 static inline void __btrfs_debug_check_extent_io_range(const char *caller,
90 struct extent_io_tree *tree, u64 start, u64 end)
92 struct inode *inode = tree->private_data;
95 if (!inode || !is_data_inode(inode))
98 isize = i_size_read(inode);
99 if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) {
100 btrfs_debug_rl(BTRFS_I(inode)->root->fs_info,
101 "%s: ino %llu isize %llu odd range [%llu,%llu]",
102 caller, btrfs_ino(BTRFS_I(inode)), isize, start, end);
106 #define btrfs_leak_debug_add(new, head) do {} while (0)
107 #define btrfs_leak_debug_del(entry) do {} while (0)
108 #define btrfs_leak_debug_check() do {} while (0)
109 #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0)
115 struct rb_node rb_node;
118 struct extent_page_data {
120 struct extent_io_tree *tree;
121 /* tells writepage not to lock the state bits for this range
122 * it still does the unlocking
124 unsigned int extent_locked:1;
126 /* tells the submit_bio code to use REQ_SYNC */
127 unsigned int sync_io:1;
130 static int add_extent_changeset(struct extent_state *state, unsigned bits,
131 struct extent_changeset *changeset,
138 if (set && (state->state & bits) == bits)
140 if (!set && (state->state & bits) == 0)
142 changeset->bytes_changed += state->end - state->start + 1;
143 ret = ulist_add(&changeset->range_changed, state->start, state->end,
148 static int __must_check submit_one_bio(struct bio *bio, int mirror_num,
149 unsigned long bio_flags)
151 blk_status_t ret = 0;
152 struct extent_io_tree *tree = bio->bi_private;
154 bio->bi_private = NULL;
157 ret = tree->ops->submit_bio_hook(tree->private_data, bio,
158 mirror_num, bio_flags);
160 btrfsic_submit_bio(bio);
162 return blk_status_to_errno(ret);
165 /* Cleanup unsubmitted bios */
166 static void end_write_bio(struct extent_page_data *epd, int ret)
169 epd->bio->bi_status = errno_to_blk_status(ret);
176 * Submit bio from extent page data via submit_one_bio
178 * Return 0 if everything is OK.
179 * Return <0 for error.
181 static int __must_check flush_write_bio(struct extent_page_data *epd)
186 ret = submit_one_bio(epd->bio, 0, 0);
188 * Clean up of epd->bio is handled by its endio function.
189 * And endio is either triggered by successful bio execution
190 * or the error handler of submit bio hook.
191 * So at this point, no matter what happened, we don't need
192 * to clean up epd->bio.
199 int __init extent_io_init(void)
201 extent_state_cache = kmem_cache_create("btrfs_extent_state",
202 sizeof(struct extent_state), 0,
203 SLAB_MEM_SPREAD, NULL);
204 if (!extent_state_cache)
207 extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer",
208 sizeof(struct extent_buffer), 0,
209 SLAB_MEM_SPREAD, NULL);
210 if (!extent_buffer_cache)
211 goto free_state_cache;
213 if (bioset_init(&btrfs_bioset, BIO_POOL_SIZE,
214 offsetof(struct btrfs_io_bio, bio),
216 goto free_buffer_cache;
218 if (bioset_integrity_create(&btrfs_bioset, BIO_POOL_SIZE))
224 bioset_exit(&btrfs_bioset);
227 kmem_cache_destroy(extent_buffer_cache);
228 extent_buffer_cache = NULL;
231 kmem_cache_destroy(extent_state_cache);
232 extent_state_cache = NULL;
236 void __cold extent_io_exit(void)
238 btrfs_leak_debug_check();
241 * Make sure all delayed rcu free are flushed before we
245 kmem_cache_destroy(extent_state_cache);
246 kmem_cache_destroy(extent_buffer_cache);
247 bioset_exit(&btrfs_bioset);
250 void extent_io_tree_init(struct btrfs_fs_info *fs_info,
251 struct extent_io_tree *tree, unsigned int owner,
254 tree->fs_info = fs_info;
255 tree->state = RB_ROOT;
257 tree->dirty_bytes = 0;
258 spin_lock_init(&tree->lock);
259 tree->private_data = private_data;
263 void extent_io_tree_release(struct extent_io_tree *tree)
265 spin_lock(&tree->lock);
267 * Do a single barrier for the waitqueue_active check here, the state
268 * of the waitqueue should not change once extent_io_tree_release is
272 while (!RB_EMPTY_ROOT(&tree->state)) {
273 struct rb_node *node;
274 struct extent_state *state;
276 node = rb_first(&tree->state);
277 state = rb_entry(node, struct extent_state, rb_node);
278 rb_erase(&state->rb_node, &tree->state);
279 RB_CLEAR_NODE(&state->rb_node);
281 * btree io trees aren't supposed to have tasks waiting for
282 * changes in the flags of extent states ever.
284 ASSERT(!waitqueue_active(&state->wq));
285 free_extent_state(state);
287 cond_resched_lock(&tree->lock);
289 spin_unlock(&tree->lock);
292 static struct extent_state *alloc_extent_state(gfp_t mask)
294 struct extent_state *state;
297 * The given mask might be not appropriate for the slab allocator,
298 * drop the unsupported bits
300 mask &= ~(__GFP_DMA32|__GFP_HIGHMEM);
301 state = kmem_cache_alloc(extent_state_cache, mask);
305 state->failrec = NULL;
306 RB_CLEAR_NODE(&state->rb_node);
307 btrfs_leak_debug_add(&state->leak_list, &states);
308 refcount_set(&state->refs, 1);
309 init_waitqueue_head(&state->wq);
310 trace_alloc_extent_state(state, mask, _RET_IP_);
314 void free_extent_state(struct extent_state *state)
318 if (refcount_dec_and_test(&state->refs)) {
319 WARN_ON(extent_state_in_tree(state));
320 btrfs_leak_debug_del(&state->leak_list);
321 trace_free_extent_state(state, _RET_IP_);
322 kmem_cache_free(extent_state_cache, state);
326 static struct rb_node *tree_insert(struct rb_root *root,
327 struct rb_node *search_start,
329 struct rb_node *node,
330 struct rb_node ***p_in,
331 struct rb_node **parent_in)
334 struct rb_node *parent = NULL;
335 struct tree_entry *entry;
337 if (p_in && parent_in) {
343 p = search_start ? &search_start : &root->rb_node;
346 entry = rb_entry(parent, struct tree_entry, rb_node);
348 if (offset < entry->start)
350 else if (offset > entry->end)
357 rb_link_node(node, parent, p);
358 rb_insert_color(node, root);
363 * __etree_search - searche @tree for an entry that contains @offset. Such
364 * entry would have entry->start <= offset && entry->end >= offset.
366 * @tree - the tree to search
367 * @offset - offset that should fall within an entry in @tree
368 * @next_ret - pointer to the first entry whose range ends after @offset
369 * @prev - pointer to the first entry whose range begins before @offset
370 * @p_ret - pointer where new node should be anchored (used when inserting an
372 * @parent_ret - points to entry which would have been the parent of the entry,
375 * This function returns a pointer to the entry that contains @offset byte
376 * address. If no such entry exists, then NULL is returned and the other
377 * pointer arguments to the function are filled, otherwise the found entry is
378 * returned and other pointers are left untouched.
380 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
381 struct rb_node **next_ret,
382 struct rb_node **prev_ret,
383 struct rb_node ***p_ret,
384 struct rb_node **parent_ret)
386 struct rb_root *root = &tree->state;
387 struct rb_node **n = &root->rb_node;
388 struct rb_node *prev = NULL;
389 struct rb_node *orig_prev = NULL;
390 struct tree_entry *entry;
391 struct tree_entry *prev_entry = NULL;
395 entry = rb_entry(prev, struct tree_entry, rb_node);
398 if (offset < entry->start)
400 else if (offset > entry->end)
413 while (prev && offset > prev_entry->end) {
414 prev = rb_next(prev);
415 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
422 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
423 while (prev && offset < prev_entry->start) {
424 prev = rb_prev(prev);
425 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
432 static inline struct rb_node *
433 tree_search_for_insert(struct extent_io_tree *tree,
435 struct rb_node ***p_ret,
436 struct rb_node **parent_ret)
438 struct rb_node *next= NULL;
441 ret = __etree_search(tree, offset, &next, NULL, p_ret, parent_ret);
447 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
450 return tree_search_for_insert(tree, offset, NULL, NULL);
454 * utility function to look for merge candidates inside a given range.
455 * Any extents with matching state are merged together into a single
456 * extent in the tree. Extents with EXTENT_IO in their state field
457 * are not merged because the end_io handlers need to be able to do
458 * operations on them without sleeping (or doing allocations/splits).
460 * This should be called with the tree lock held.
462 static void merge_state(struct extent_io_tree *tree,
463 struct extent_state *state)
465 struct extent_state *other;
466 struct rb_node *other_node;
468 if (state->state & (EXTENT_LOCKED | EXTENT_BOUNDARY))
471 other_node = rb_prev(&state->rb_node);
473 other = rb_entry(other_node, struct extent_state, rb_node);
474 if (other->end == state->start - 1 &&
475 other->state == state->state) {
476 if (tree->private_data &&
477 is_data_inode(tree->private_data))
478 btrfs_merge_delalloc_extent(tree->private_data,
480 state->start = other->start;
481 rb_erase(&other->rb_node, &tree->state);
482 RB_CLEAR_NODE(&other->rb_node);
483 free_extent_state(other);
486 other_node = rb_next(&state->rb_node);
488 other = rb_entry(other_node, struct extent_state, rb_node);
489 if (other->start == state->end + 1 &&
490 other->state == state->state) {
491 if (tree->private_data &&
492 is_data_inode(tree->private_data))
493 btrfs_merge_delalloc_extent(tree->private_data,
495 state->end = other->end;
496 rb_erase(&other->rb_node, &tree->state);
497 RB_CLEAR_NODE(&other->rb_node);
498 free_extent_state(other);
503 static void set_state_bits(struct extent_io_tree *tree,
504 struct extent_state *state, unsigned *bits,
505 struct extent_changeset *changeset);
508 * insert an extent_state struct into the tree. 'bits' are set on the
509 * struct before it is inserted.
511 * This may return -EEXIST if the extent is already there, in which case the
512 * state struct is freed.
514 * The tree lock is not taken internally. This is a utility function and
515 * probably isn't what you want to call (see set/clear_extent_bit).
517 static int insert_state(struct extent_io_tree *tree,
518 struct extent_state *state, u64 start, u64 end,
520 struct rb_node **parent,
521 unsigned *bits, struct extent_changeset *changeset)
523 struct rb_node *node;
526 btrfs_err(tree->fs_info,
527 "insert state: end < start %llu %llu", end, start);
530 state->start = start;
533 set_state_bits(tree, state, bits, changeset);
535 node = tree_insert(&tree->state, NULL, end, &state->rb_node, p, parent);
537 struct extent_state *found;
538 found = rb_entry(node, struct extent_state, rb_node);
539 btrfs_err(tree->fs_info,
540 "found node %llu %llu on insert of %llu %llu",
541 found->start, found->end, start, end);
544 merge_state(tree, state);
549 * split a given extent state struct in two, inserting the preallocated
550 * struct 'prealloc' as the newly created second half. 'split' indicates an
551 * offset inside 'orig' where it should be split.
554 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
555 * are two extent state structs in the tree:
556 * prealloc: [orig->start, split - 1]
557 * orig: [ split, orig->end ]
559 * The tree locks are not taken by this function. They need to be held
562 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
563 struct extent_state *prealloc, u64 split)
565 struct rb_node *node;
567 if (tree->private_data && is_data_inode(tree->private_data))
568 btrfs_split_delalloc_extent(tree->private_data, orig, split);
570 prealloc->start = orig->start;
571 prealloc->end = split - 1;
572 prealloc->state = orig->state;
575 node = tree_insert(&tree->state, &orig->rb_node, prealloc->end,
576 &prealloc->rb_node, NULL, NULL);
578 free_extent_state(prealloc);
584 static struct extent_state *next_state(struct extent_state *state)
586 struct rb_node *next = rb_next(&state->rb_node);
588 return rb_entry(next, struct extent_state, rb_node);
594 * utility function to clear some bits in an extent state struct.
595 * it will optionally wake up anyone waiting on this state (wake == 1).
597 * If no bits are set on the state struct after clearing things, the
598 * struct is freed and removed from the tree
600 static struct extent_state *clear_state_bit(struct extent_io_tree *tree,
601 struct extent_state *state,
602 unsigned *bits, int wake,
603 struct extent_changeset *changeset)
605 struct extent_state *next;
606 unsigned bits_to_clear = *bits & ~EXTENT_CTLBITS;
609 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
610 u64 range = state->end - state->start + 1;
611 WARN_ON(range > tree->dirty_bytes);
612 tree->dirty_bytes -= range;
615 if (tree->private_data && is_data_inode(tree->private_data))
616 btrfs_clear_delalloc_extent(tree->private_data, state, bits);
618 ret = add_extent_changeset(state, bits_to_clear, changeset, 0);
620 state->state &= ~bits_to_clear;
623 if (state->state == 0) {
624 next = next_state(state);
625 if (extent_state_in_tree(state)) {
626 rb_erase(&state->rb_node, &tree->state);
627 RB_CLEAR_NODE(&state->rb_node);
628 free_extent_state(state);
633 merge_state(tree, state);
634 next = next_state(state);
639 static struct extent_state *
640 alloc_extent_state_atomic(struct extent_state *prealloc)
643 prealloc = alloc_extent_state(GFP_ATOMIC);
648 static void extent_io_tree_panic(struct extent_io_tree *tree, int err)
650 struct inode *inode = tree->private_data;
652 btrfs_panic(btrfs_sb(inode->i_sb), err,
653 "locking error: extent tree was modified by another thread while locked");
657 * clear some bits on a range in the tree. This may require splitting
658 * or inserting elements in the tree, so the gfp mask is used to
659 * indicate which allocations or sleeping are allowed.
661 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
662 * the given range from the tree regardless of state (ie for truncate).
664 * the range [start, end] is inclusive.
666 * This takes the tree lock, and returns 0 on success and < 0 on error.
668 int __clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
669 unsigned bits, int wake, int delete,
670 struct extent_state **cached_state,
671 gfp_t mask, struct extent_changeset *changeset)
673 struct extent_state *state;
674 struct extent_state *cached;
675 struct extent_state *prealloc = NULL;
676 struct rb_node *node;
681 btrfs_debug_check_extent_io_range(tree, start, end);
682 trace_btrfs_clear_extent_bit(tree, start, end - start + 1, bits);
684 if (bits & EXTENT_DELALLOC)
685 bits |= EXTENT_NORESERVE;
688 bits |= ~EXTENT_CTLBITS;
690 if (bits & (EXTENT_LOCKED | EXTENT_BOUNDARY))
693 if (!prealloc && gfpflags_allow_blocking(mask)) {
695 * Don't care for allocation failure here because we might end
696 * up not needing the pre-allocated extent state at all, which
697 * is the case if we only have in the tree extent states that
698 * cover our input range and don't cover too any other range.
699 * If we end up needing a new extent state we allocate it later.
701 prealloc = alloc_extent_state(mask);
704 spin_lock(&tree->lock);
706 cached = *cached_state;
709 *cached_state = NULL;
713 if (cached && extent_state_in_tree(cached) &&
714 cached->start <= start && cached->end > start) {
716 refcount_dec(&cached->refs);
721 free_extent_state(cached);
724 * this search will find the extents that end after
727 node = tree_search(tree, start);
730 state = rb_entry(node, struct extent_state, rb_node);
732 if (state->start > end)
734 WARN_ON(state->end < start);
735 last_end = state->end;
737 /* the state doesn't have the wanted bits, go ahead */
738 if (!(state->state & bits)) {
739 state = next_state(state);
744 * | ---- desired range ---- |
746 * | ------------- state -------------- |
748 * We need to split the extent we found, and may flip
749 * bits on second half.
751 * If the extent we found extends past our range, we
752 * just split and search again. It'll get split again
753 * the next time though.
755 * If the extent we found is inside our range, we clear
756 * the desired bit on it.
759 if (state->start < start) {
760 prealloc = alloc_extent_state_atomic(prealloc);
762 err = split_state(tree, state, prealloc, start);
764 extent_io_tree_panic(tree, err);
769 if (state->end <= end) {
770 state = clear_state_bit(tree, state, &bits, wake,
777 * | ---- desired range ---- |
779 * We need to split the extent, and clear the bit
782 if (state->start <= end && state->end > end) {
783 prealloc = alloc_extent_state_atomic(prealloc);
785 err = split_state(tree, state, prealloc, end + 1);
787 extent_io_tree_panic(tree, err);
792 clear_state_bit(tree, prealloc, &bits, wake, changeset);
798 state = clear_state_bit(tree, state, &bits, wake, changeset);
800 if (last_end == (u64)-1)
802 start = last_end + 1;
803 if (start <= end && state && !need_resched())
809 spin_unlock(&tree->lock);
810 if (gfpflags_allow_blocking(mask))
815 spin_unlock(&tree->lock);
817 free_extent_state(prealloc);
823 static void wait_on_state(struct extent_io_tree *tree,
824 struct extent_state *state)
825 __releases(tree->lock)
826 __acquires(tree->lock)
829 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
830 spin_unlock(&tree->lock);
832 spin_lock(&tree->lock);
833 finish_wait(&state->wq, &wait);
837 * waits for one or more bits to clear on a range in the state tree.
838 * The range [start, end] is inclusive.
839 * The tree lock is taken by this function
841 static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
844 struct extent_state *state;
845 struct rb_node *node;
847 btrfs_debug_check_extent_io_range(tree, start, end);
849 spin_lock(&tree->lock);
853 * this search will find all the extents that end after
856 node = tree_search(tree, start);
861 state = rb_entry(node, struct extent_state, rb_node);
863 if (state->start > end)
866 if (state->state & bits) {
867 start = state->start;
868 refcount_inc(&state->refs);
869 wait_on_state(tree, state);
870 free_extent_state(state);
873 start = state->end + 1;
878 if (!cond_resched_lock(&tree->lock)) {
879 node = rb_next(node);
884 spin_unlock(&tree->lock);
887 static void set_state_bits(struct extent_io_tree *tree,
888 struct extent_state *state,
889 unsigned *bits, struct extent_changeset *changeset)
891 unsigned bits_to_set = *bits & ~EXTENT_CTLBITS;
894 if (tree->private_data && is_data_inode(tree->private_data))
895 btrfs_set_delalloc_extent(tree->private_data, state, bits);
897 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
898 u64 range = state->end - state->start + 1;
899 tree->dirty_bytes += range;
901 ret = add_extent_changeset(state, bits_to_set, changeset, 1);
903 state->state |= bits_to_set;
906 static void cache_state_if_flags(struct extent_state *state,
907 struct extent_state **cached_ptr,
910 if (cached_ptr && !(*cached_ptr)) {
911 if (!flags || (state->state & flags)) {
913 refcount_inc(&state->refs);
918 static void cache_state(struct extent_state *state,
919 struct extent_state **cached_ptr)
921 return cache_state_if_flags(state, cached_ptr,
922 EXTENT_LOCKED | EXTENT_BOUNDARY);
926 * set some bits on a range in the tree. This may require allocations or
927 * sleeping, so the gfp mask is used to indicate what is allowed.
929 * If any of the exclusive bits are set, this will fail with -EEXIST if some
930 * part of the range already has the desired bits set. The start of the
931 * existing range is returned in failed_start in this case.
933 * [start, end] is inclusive This takes the tree lock.
936 static int __must_check
937 __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
938 unsigned bits, unsigned exclusive_bits,
939 u64 *failed_start, struct extent_state **cached_state,
940 gfp_t mask, struct extent_changeset *changeset)
942 struct extent_state *state;
943 struct extent_state *prealloc = NULL;
944 struct rb_node *node;
946 struct rb_node *parent;
951 btrfs_debug_check_extent_io_range(tree, start, end);
952 trace_btrfs_set_extent_bit(tree, start, end - start + 1, bits);
955 if (!prealloc && gfpflags_allow_blocking(mask)) {
957 * Don't care for allocation failure here because we might end
958 * up not needing the pre-allocated extent state at all, which
959 * is the case if we only have in the tree extent states that
960 * cover our input range and don't cover too any other range.
961 * If we end up needing a new extent state we allocate it later.
963 prealloc = alloc_extent_state(mask);
966 spin_lock(&tree->lock);
967 if (cached_state && *cached_state) {
968 state = *cached_state;
969 if (state->start <= start && state->end > start &&
970 extent_state_in_tree(state)) {
971 node = &state->rb_node;
976 * this search will find all the extents that end after
979 node = tree_search_for_insert(tree, start, &p, &parent);
981 prealloc = alloc_extent_state_atomic(prealloc);
983 err = insert_state(tree, prealloc, start, end,
984 &p, &parent, &bits, changeset);
986 extent_io_tree_panic(tree, err);
988 cache_state(prealloc, cached_state);
992 state = rb_entry(node, struct extent_state, rb_node);
994 last_start = state->start;
995 last_end = state->end;
998 * | ---- desired range ---- |
1001 * Just lock what we found and keep going
1003 if (state->start == start && state->end <= end) {
1004 if (state->state & exclusive_bits) {
1005 *failed_start = state->start;
1010 set_state_bits(tree, state, &bits, changeset);
1011 cache_state(state, cached_state);
1012 merge_state(tree, state);
1013 if (last_end == (u64)-1)
1015 start = last_end + 1;
1016 state = next_state(state);
1017 if (start < end && state && state->start == start &&
1024 * | ---- desired range ---- |
1027 * | ------------- state -------------- |
1029 * We need to split the extent we found, and may flip bits on
1032 * If the extent we found extends past our
1033 * range, we just split and search again. It'll get split
1034 * again the next time though.
1036 * If the extent we found is inside our range, we set the
1037 * desired bit on it.
1039 if (state->start < start) {
1040 if (state->state & exclusive_bits) {
1041 *failed_start = start;
1046 prealloc = alloc_extent_state_atomic(prealloc);
1048 err = split_state(tree, state, prealloc, start);
1050 extent_io_tree_panic(tree, err);
1055 if (state->end <= end) {
1056 set_state_bits(tree, state, &bits, changeset);
1057 cache_state(state, cached_state);
1058 merge_state(tree, state);
1059 if (last_end == (u64)-1)
1061 start = last_end + 1;
1062 state = next_state(state);
1063 if (start < end && state && state->start == start &&
1070 * | ---- desired range ---- |
1071 * | state | or | state |
1073 * There's a hole, we need to insert something in it and
1074 * ignore the extent we found.
1076 if (state->start > start) {
1078 if (end < last_start)
1081 this_end = last_start - 1;
1083 prealloc = alloc_extent_state_atomic(prealloc);
1087 * Avoid to free 'prealloc' if it can be merged with
1090 err = insert_state(tree, prealloc, start, this_end,
1091 NULL, NULL, &bits, changeset);
1093 extent_io_tree_panic(tree, err);
1095 cache_state(prealloc, cached_state);
1097 start = this_end + 1;
1101 * | ---- desired range ---- |
1103 * We need to split the extent, and set the bit
1106 if (state->start <= end && state->end > end) {
1107 if (state->state & exclusive_bits) {
1108 *failed_start = start;
1113 prealloc = alloc_extent_state_atomic(prealloc);
1115 err = split_state(tree, state, prealloc, end + 1);
1117 extent_io_tree_panic(tree, err);
1119 set_state_bits(tree, prealloc, &bits, changeset);
1120 cache_state(prealloc, cached_state);
1121 merge_state(tree, prealloc);
1129 spin_unlock(&tree->lock);
1130 if (gfpflags_allow_blocking(mask))
1135 spin_unlock(&tree->lock);
1137 free_extent_state(prealloc);
1143 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1144 unsigned bits, u64 * failed_start,
1145 struct extent_state **cached_state, gfp_t mask)
1147 return __set_extent_bit(tree, start, end, bits, 0, failed_start,
1148 cached_state, mask, NULL);
1153 * convert_extent_bit - convert all bits in a given range from one bit to
1155 * @tree: the io tree to search
1156 * @start: the start offset in bytes
1157 * @end: the end offset in bytes (inclusive)
1158 * @bits: the bits to set in this range
1159 * @clear_bits: the bits to clear in this range
1160 * @cached_state: state that we're going to cache
1162 * This will go through and set bits for the given range. If any states exist
1163 * already in this range they are set with the given bit and cleared of the
1164 * clear_bits. This is only meant to be used by things that are mergeable, ie
1165 * converting from say DELALLOC to DIRTY. This is not meant to be used with
1166 * boundary bits like LOCK.
1168 * All allocations are done with GFP_NOFS.
1170 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1171 unsigned bits, unsigned clear_bits,
1172 struct extent_state **cached_state)
1174 struct extent_state *state;
1175 struct extent_state *prealloc = NULL;
1176 struct rb_node *node;
1178 struct rb_node *parent;
1182 bool first_iteration = true;
1184 btrfs_debug_check_extent_io_range(tree, start, end);
1185 trace_btrfs_convert_extent_bit(tree, start, end - start + 1, bits,
1191 * Best effort, don't worry if extent state allocation fails
1192 * here for the first iteration. We might have a cached state
1193 * that matches exactly the target range, in which case no
1194 * extent state allocations are needed. We'll only know this
1195 * after locking the tree.
1197 prealloc = alloc_extent_state(GFP_NOFS);
1198 if (!prealloc && !first_iteration)
1202 spin_lock(&tree->lock);
1203 if (cached_state && *cached_state) {
1204 state = *cached_state;
1205 if (state->start <= start && state->end > start &&
1206 extent_state_in_tree(state)) {
1207 node = &state->rb_node;
1213 * this search will find all the extents that end after
1216 node = tree_search_for_insert(tree, start, &p, &parent);
1218 prealloc = alloc_extent_state_atomic(prealloc);
1223 err = insert_state(tree, prealloc, start, end,
1224 &p, &parent, &bits, NULL);
1226 extent_io_tree_panic(tree, err);
1227 cache_state(prealloc, cached_state);
1231 state = rb_entry(node, struct extent_state, rb_node);
1233 last_start = state->start;
1234 last_end = state->end;
1237 * | ---- desired range ---- |
1240 * Just lock what we found and keep going
1242 if (state->start == start && state->end <= end) {
1243 set_state_bits(tree, state, &bits, NULL);
1244 cache_state(state, cached_state);
1245 state = clear_state_bit(tree, state, &clear_bits, 0, NULL);
1246 if (last_end == (u64)-1)
1248 start = last_end + 1;
1249 if (start < end && state && state->start == start &&
1256 * | ---- desired range ---- |
1259 * | ------------- state -------------- |
1261 * We need to split the extent we found, and may flip bits on
1264 * If the extent we found extends past our
1265 * range, we just split and search again. It'll get split
1266 * again the next time though.
1268 * If the extent we found is inside our range, we set the
1269 * desired bit on it.
1271 if (state->start < start) {
1272 prealloc = alloc_extent_state_atomic(prealloc);
1277 err = split_state(tree, state, prealloc, start);
1279 extent_io_tree_panic(tree, err);
1283 if (state->end <= end) {
1284 set_state_bits(tree, state, &bits, NULL);
1285 cache_state(state, cached_state);
1286 state = clear_state_bit(tree, state, &clear_bits, 0,
1288 if (last_end == (u64)-1)
1290 start = last_end + 1;
1291 if (start < end && state && state->start == start &&
1298 * | ---- desired range ---- |
1299 * | state | or | state |
1301 * There's a hole, we need to insert something in it and
1302 * ignore the extent we found.
1304 if (state->start > start) {
1306 if (end < last_start)
1309 this_end = last_start - 1;
1311 prealloc = alloc_extent_state_atomic(prealloc);
1318 * Avoid to free 'prealloc' if it can be merged with
1321 err = insert_state(tree, prealloc, start, this_end,
1322 NULL, NULL, &bits, NULL);
1324 extent_io_tree_panic(tree, err);
1325 cache_state(prealloc, cached_state);
1327 start = this_end + 1;
1331 * | ---- desired range ---- |
1333 * We need to split the extent, and set the bit
1336 if (state->start <= end && state->end > end) {
1337 prealloc = alloc_extent_state_atomic(prealloc);
1343 err = split_state(tree, state, prealloc, end + 1);
1345 extent_io_tree_panic(tree, err);
1347 set_state_bits(tree, prealloc, &bits, NULL);
1348 cache_state(prealloc, cached_state);
1349 clear_state_bit(tree, prealloc, &clear_bits, 0, NULL);
1357 spin_unlock(&tree->lock);
1359 first_iteration = false;
1363 spin_unlock(&tree->lock);
1365 free_extent_state(prealloc);
1370 /* wrappers around set/clear extent bit */
1371 int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1372 unsigned bits, struct extent_changeset *changeset)
1375 * We don't support EXTENT_LOCKED yet, as current changeset will
1376 * record any bits changed, so for EXTENT_LOCKED case, it will
1377 * either fail with -EEXIST or changeset will record the whole
1380 BUG_ON(bits & EXTENT_LOCKED);
1382 return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL, GFP_NOFS,
1386 int set_extent_bits_nowait(struct extent_io_tree *tree, u64 start, u64 end,
1389 return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL,
1393 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1394 unsigned bits, int wake, int delete,
1395 struct extent_state **cached)
1397 return __clear_extent_bit(tree, start, end, bits, wake, delete,
1398 cached, GFP_NOFS, NULL);
1401 int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1402 unsigned bits, struct extent_changeset *changeset)
1405 * Don't support EXTENT_LOCKED case, same reason as
1406 * set_record_extent_bits().
1408 BUG_ON(bits & EXTENT_LOCKED);
1410 return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, GFP_NOFS,
1415 * either insert or lock state struct between start and end use mask to tell
1416 * us if waiting is desired.
1418 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1419 struct extent_state **cached_state)
1425 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED,
1426 EXTENT_LOCKED, &failed_start,
1427 cached_state, GFP_NOFS, NULL);
1428 if (err == -EEXIST) {
1429 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1430 start = failed_start;
1433 WARN_ON(start > end);
1438 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1443 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1444 &failed_start, NULL, GFP_NOFS, NULL);
1445 if (err == -EEXIST) {
1446 if (failed_start > start)
1447 clear_extent_bit(tree, start, failed_start - 1,
1448 EXTENT_LOCKED, 1, 0, NULL);
1454 void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
1456 unsigned long index = start >> PAGE_SHIFT;
1457 unsigned long end_index = end >> PAGE_SHIFT;
1460 while (index <= end_index) {
1461 page = find_get_page(inode->i_mapping, index);
1462 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1463 clear_page_dirty_for_io(page);
1469 void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
1471 unsigned long index = start >> PAGE_SHIFT;
1472 unsigned long end_index = end >> PAGE_SHIFT;
1475 while (index <= end_index) {
1476 page = find_get_page(inode->i_mapping, index);
1477 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1478 __set_page_dirty_nobuffers(page);
1479 account_page_redirty(page);
1485 /* find the first state struct with 'bits' set after 'start', and
1486 * return it. tree->lock must be held. NULL will returned if
1487 * nothing was found after 'start'
1489 static struct extent_state *
1490 find_first_extent_bit_state(struct extent_io_tree *tree,
1491 u64 start, unsigned bits)
1493 struct rb_node *node;
1494 struct extent_state *state;
1497 * this search will find all the extents that end after
1500 node = tree_search(tree, start);
1505 state = rb_entry(node, struct extent_state, rb_node);
1506 if (state->end >= start && (state->state & bits))
1509 node = rb_next(node);
1518 * find the first offset in the io tree with 'bits' set. zero is
1519 * returned if we find something, and *start_ret and *end_ret are
1520 * set to reflect the state struct that was found.
1522 * If nothing was found, 1 is returned. If found something, return 0.
1524 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1525 u64 *start_ret, u64 *end_ret, unsigned bits,
1526 struct extent_state **cached_state)
1528 struct extent_state *state;
1531 spin_lock(&tree->lock);
1532 if (cached_state && *cached_state) {
1533 state = *cached_state;
1534 if (state->end == start - 1 && extent_state_in_tree(state)) {
1535 while ((state = next_state(state)) != NULL) {
1536 if (state->state & bits)
1539 free_extent_state(*cached_state);
1540 *cached_state = NULL;
1543 free_extent_state(*cached_state);
1544 *cached_state = NULL;
1547 state = find_first_extent_bit_state(tree, start, bits);
1550 cache_state_if_flags(state, cached_state, 0);
1551 *start_ret = state->start;
1552 *end_ret = state->end;
1556 spin_unlock(&tree->lock);
1561 * find_first_clear_extent_bit - find the first range that has @bits not set.
1562 * This range could start before @start.
1564 * @tree - the tree to search
1565 * @start - the offset at/after which the found extent should start
1566 * @start_ret - records the beginning of the range
1567 * @end_ret - records the end of the range (inclusive)
1568 * @bits - the set of bits which must be unset
1570 * Since unallocated range is also considered one which doesn't have the bits
1571 * set it's possible that @end_ret contains -1, this happens in case the range
1572 * spans (last_range_end, end of device]. In this case it's up to the caller to
1573 * trim @end_ret to the appropriate size.
1575 void find_first_clear_extent_bit(struct extent_io_tree *tree, u64 start,
1576 u64 *start_ret, u64 *end_ret, unsigned bits)
1578 struct extent_state *state;
1579 struct rb_node *node, *prev = NULL, *next;
1581 spin_lock(&tree->lock);
1583 /* Find first extent with bits cleared */
1585 node = __etree_search(tree, start, &next, &prev, NULL, NULL);
1590 * We are past the last allocated chunk,
1591 * set start at the end of the last extent. The
1592 * device alloc tree should never be empty so
1593 * prev is always set.
1596 state = rb_entry(prev, struct extent_state, rb_node);
1597 *start_ret = state->end + 1;
1603 * At this point 'node' either contains 'start' or start is
1606 state = rb_entry(node, struct extent_state, rb_node);
1608 if (in_range(start, state->start, state->end - state->start + 1)) {
1609 if (state->state & bits) {
1611 * |--range with bits sets--|
1615 start = state->end + 1;
1618 * 'start' falls within a range that doesn't
1619 * have the bits set, so take its start as
1620 * the beginning of the desired range
1622 * |--range with bits cleared----|
1626 *start_ret = state->start;
1631 * |---prev range---|---hole/unset---|---node range---|
1637 * |---hole/unset--||--first node--|
1642 state = rb_entry(prev, struct extent_state,
1644 *start_ret = state->end + 1;
1653 * Find the longest stretch from start until an entry which has the
1657 state = rb_entry(node, struct extent_state, rb_node);
1658 if (state->end >= start && !(state->state & bits)) {
1659 *end_ret = state->end;
1661 *end_ret = state->start - 1;
1665 node = rb_next(node);
1670 spin_unlock(&tree->lock);
1674 * find a contiguous range of bytes in the file marked as delalloc, not
1675 * more than 'max_bytes'. start and end are used to return the range,
1677 * true is returned if we find something, false if nothing was in the tree
1679 static noinline bool find_delalloc_range(struct extent_io_tree *tree,
1680 u64 *start, u64 *end, u64 max_bytes,
1681 struct extent_state **cached_state)
1683 struct rb_node *node;
1684 struct extent_state *state;
1685 u64 cur_start = *start;
1687 u64 total_bytes = 0;
1689 spin_lock(&tree->lock);
1692 * this search will find all the extents that end after
1695 node = tree_search(tree, cur_start);
1702 state = rb_entry(node, struct extent_state, rb_node);
1703 if (found && (state->start != cur_start ||
1704 (state->state & EXTENT_BOUNDARY))) {
1707 if (!(state->state & EXTENT_DELALLOC)) {
1713 *start = state->start;
1714 *cached_state = state;
1715 refcount_inc(&state->refs);
1719 cur_start = state->end + 1;
1720 node = rb_next(node);
1721 total_bytes += state->end - state->start + 1;
1722 if (total_bytes >= max_bytes)
1728 spin_unlock(&tree->lock);
1732 static int __process_pages_contig(struct address_space *mapping,
1733 struct page *locked_page,
1734 pgoff_t start_index, pgoff_t end_index,
1735 unsigned long page_ops, pgoff_t *index_ret);
1737 static noinline void __unlock_for_delalloc(struct inode *inode,
1738 struct page *locked_page,
1741 unsigned long index = start >> PAGE_SHIFT;
1742 unsigned long end_index = end >> PAGE_SHIFT;
1744 ASSERT(locked_page);
1745 if (index == locked_page->index && end_index == index)
1748 __process_pages_contig(inode->i_mapping, locked_page, index, end_index,
1752 static noinline int lock_delalloc_pages(struct inode *inode,
1753 struct page *locked_page,
1757 unsigned long index = delalloc_start >> PAGE_SHIFT;
1758 unsigned long index_ret = index;
1759 unsigned long end_index = delalloc_end >> PAGE_SHIFT;
1762 ASSERT(locked_page);
1763 if (index == locked_page->index && index == end_index)
1766 ret = __process_pages_contig(inode->i_mapping, locked_page, index,
1767 end_index, PAGE_LOCK, &index_ret);
1769 __unlock_for_delalloc(inode, locked_page, delalloc_start,
1770 (u64)index_ret << PAGE_SHIFT);
1775 * Find and lock a contiguous range of bytes in the file marked as delalloc, no
1776 * more than @max_bytes. @Start and @end are used to return the range,
1778 * Return: true if we find something
1779 * false if nothing was in the tree
1782 noinline_for_stack bool find_lock_delalloc_range(struct inode *inode,
1783 struct page *locked_page, u64 *start,
1786 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
1787 u64 max_bytes = BTRFS_MAX_EXTENT_SIZE;
1791 struct extent_state *cached_state = NULL;
1796 /* step one, find a bunch of delalloc bytes starting at start */
1797 delalloc_start = *start;
1799 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1800 max_bytes, &cached_state);
1801 if (!found || delalloc_end <= *start) {
1802 *start = delalloc_start;
1803 *end = delalloc_end;
1804 free_extent_state(cached_state);
1809 * start comes from the offset of locked_page. We have to lock
1810 * pages in order, so we can't process delalloc bytes before
1813 if (delalloc_start < *start)
1814 delalloc_start = *start;
1817 * make sure to limit the number of pages we try to lock down
1819 if (delalloc_end + 1 - delalloc_start > max_bytes)
1820 delalloc_end = delalloc_start + max_bytes - 1;
1822 /* step two, lock all the pages after the page that has start */
1823 ret = lock_delalloc_pages(inode, locked_page,
1824 delalloc_start, delalloc_end);
1825 ASSERT(!ret || ret == -EAGAIN);
1826 if (ret == -EAGAIN) {
1827 /* some of the pages are gone, lets avoid looping by
1828 * shortening the size of the delalloc range we're searching
1830 free_extent_state(cached_state);
1831 cached_state = NULL;
1833 max_bytes = PAGE_SIZE;
1842 /* step three, lock the state bits for the whole range */
1843 lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state);
1845 /* then test to make sure it is all still delalloc */
1846 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1847 EXTENT_DELALLOC, 1, cached_state);
1849 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1851 __unlock_for_delalloc(inode, locked_page,
1852 delalloc_start, delalloc_end);
1856 free_extent_state(cached_state);
1857 *start = delalloc_start;
1858 *end = delalloc_end;
1863 static int __process_pages_contig(struct address_space *mapping,
1864 struct page *locked_page,
1865 pgoff_t start_index, pgoff_t end_index,
1866 unsigned long page_ops, pgoff_t *index_ret)
1868 unsigned long nr_pages = end_index - start_index + 1;
1869 unsigned long pages_locked = 0;
1870 pgoff_t index = start_index;
1871 struct page *pages[16];
1876 if (page_ops & PAGE_LOCK) {
1877 ASSERT(page_ops == PAGE_LOCK);
1878 ASSERT(index_ret && *index_ret == start_index);
1881 if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0)
1882 mapping_set_error(mapping, -EIO);
1884 while (nr_pages > 0) {
1885 ret = find_get_pages_contig(mapping, index,
1886 min_t(unsigned long,
1887 nr_pages, ARRAY_SIZE(pages)), pages);
1890 * Only if we're going to lock these pages,
1891 * can we find nothing at @index.
1893 ASSERT(page_ops & PAGE_LOCK);
1898 for (i = 0; i < ret; i++) {
1899 if (page_ops & PAGE_SET_PRIVATE2)
1900 SetPagePrivate2(pages[i]);
1902 if (pages[i] == locked_page) {
1907 if (page_ops & PAGE_CLEAR_DIRTY)
1908 clear_page_dirty_for_io(pages[i]);
1909 if (page_ops & PAGE_SET_WRITEBACK)
1910 set_page_writeback(pages[i]);
1911 if (page_ops & PAGE_SET_ERROR)
1912 SetPageError(pages[i]);
1913 if (page_ops & PAGE_END_WRITEBACK)
1914 end_page_writeback(pages[i]);
1915 if (page_ops & PAGE_UNLOCK)
1916 unlock_page(pages[i]);
1917 if (page_ops & PAGE_LOCK) {
1918 lock_page(pages[i]);
1919 if (!PageDirty(pages[i]) ||
1920 pages[i]->mapping != mapping) {
1921 unlock_page(pages[i]);
1935 if (err && index_ret)
1936 *index_ret = start_index + pages_locked - 1;
1940 void extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end,
1941 u64 delalloc_end, struct page *locked_page,
1942 unsigned clear_bits,
1943 unsigned long page_ops)
1945 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, clear_bits, 1, 0,
1948 __process_pages_contig(inode->i_mapping, locked_page,
1949 start >> PAGE_SHIFT, end >> PAGE_SHIFT,
1954 * count the number of bytes in the tree that have a given bit(s)
1955 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1956 * cached. The total number found is returned.
1958 u64 count_range_bits(struct extent_io_tree *tree,
1959 u64 *start, u64 search_end, u64 max_bytes,
1960 unsigned bits, int contig)
1962 struct rb_node *node;
1963 struct extent_state *state;
1964 u64 cur_start = *start;
1965 u64 total_bytes = 0;
1969 if (WARN_ON(search_end <= cur_start))
1972 spin_lock(&tree->lock);
1973 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1974 total_bytes = tree->dirty_bytes;
1978 * this search will find all the extents that end after
1981 node = tree_search(tree, cur_start);
1986 state = rb_entry(node, struct extent_state, rb_node);
1987 if (state->start > search_end)
1989 if (contig && found && state->start > last + 1)
1991 if (state->end >= cur_start && (state->state & bits) == bits) {
1992 total_bytes += min(search_end, state->end) + 1 -
1993 max(cur_start, state->start);
1994 if (total_bytes >= max_bytes)
1997 *start = max(cur_start, state->start);
2001 } else if (contig && found) {
2004 node = rb_next(node);
2009 spin_unlock(&tree->lock);
2014 * set the private field for a given byte offset in the tree. If there isn't
2015 * an extent_state there already, this does nothing.
2017 static noinline int set_state_failrec(struct extent_io_tree *tree, u64 start,
2018 struct io_failure_record *failrec)
2020 struct rb_node *node;
2021 struct extent_state *state;
2024 spin_lock(&tree->lock);
2026 * this search will find all the extents that end after
2029 node = tree_search(tree, start);
2034 state = rb_entry(node, struct extent_state, rb_node);
2035 if (state->start != start) {
2039 state->failrec = failrec;
2041 spin_unlock(&tree->lock);
2045 static noinline int get_state_failrec(struct extent_io_tree *tree, u64 start,
2046 struct io_failure_record **failrec)
2048 struct rb_node *node;
2049 struct extent_state *state;
2052 spin_lock(&tree->lock);
2054 * this search will find all the extents that end after
2057 node = tree_search(tree, start);
2062 state = rb_entry(node, struct extent_state, rb_node);
2063 if (state->start != start) {
2067 *failrec = state->failrec;
2069 spin_unlock(&tree->lock);
2074 * searches a range in the state tree for a given mask.
2075 * If 'filled' == 1, this returns 1 only if every extent in the tree
2076 * has the bits set. Otherwise, 1 is returned if any bit in the
2077 * range is found set.
2079 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
2080 unsigned bits, int filled, struct extent_state *cached)
2082 struct extent_state *state = NULL;
2083 struct rb_node *node;
2086 spin_lock(&tree->lock);
2087 if (cached && extent_state_in_tree(cached) && cached->start <= start &&
2088 cached->end > start)
2089 node = &cached->rb_node;
2091 node = tree_search(tree, start);
2092 while (node && start <= end) {
2093 state = rb_entry(node, struct extent_state, rb_node);
2095 if (filled && state->start > start) {
2100 if (state->start > end)
2103 if (state->state & bits) {
2107 } else if (filled) {
2112 if (state->end == (u64)-1)
2115 start = state->end + 1;
2118 node = rb_next(node);
2125 spin_unlock(&tree->lock);
2130 * helper function to set a given page up to date if all the
2131 * extents in the tree for that page are up to date
2133 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
2135 u64 start = page_offset(page);
2136 u64 end = start + PAGE_SIZE - 1;
2137 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
2138 SetPageUptodate(page);
2141 int free_io_failure(struct extent_io_tree *failure_tree,
2142 struct extent_io_tree *io_tree,
2143 struct io_failure_record *rec)
2148 set_state_failrec(failure_tree, rec->start, NULL);
2149 ret = clear_extent_bits(failure_tree, rec->start,
2150 rec->start + rec->len - 1,
2151 EXTENT_LOCKED | EXTENT_DIRTY);
2155 ret = clear_extent_bits(io_tree, rec->start,
2156 rec->start + rec->len - 1,
2166 * this bypasses the standard btrfs submit functions deliberately, as
2167 * the standard behavior is to write all copies in a raid setup. here we only
2168 * want to write the one bad copy. so we do the mapping for ourselves and issue
2169 * submit_bio directly.
2170 * to avoid any synchronization issues, wait for the data after writing, which
2171 * actually prevents the read that triggered the error from finishing.
2172 * currently, there can be no more than two copies of every data bit. thus,
2173 * exactly one rewrite is required.
2175 int repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start,
2176 u64 length, u64 logical, struct page *page,
2177 unsigned int pg_offset, int mirror_num)
2180 struct btrfs_device *dev;
2183 struct btrfs_bio *bbio = NULL;
2186 ASSERT(!(fs_info->sb->s_flags & SB_RDONLY));
2187 BUG_ON(!mirror_num);
2189 bio = btrfs_io_bio_alloc(1);
2190 bio->bi_iter.bi_size = 0;
2191 map_length = length;
2194 * Avoid races with device replace and make sure our bbio has devices
2195 * associated to its stripes that don't go away while we are doing the
2196 * read repair operation.
2198 btrfs_bio_counter_inc_blocked(fs_info);
2199 if (btrfs_is_parity_mirror(fs_info, logical, length)) {
2201 * Note that we don't use BTRFS_MAP_WRITE because it's supposed
2202 * to update all raid stripes, but here we just want to correct
2203 * bad stripe, thus BTRFS_MAP_READ is abused to only get the bad
2204 * stripe's dev and sector.
2206 ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, logical,
2207 &map_length, &bbio, 0);
2209 btrfs_bio_counter_dec(fs_info);
2213 ASSERT(bbio->mirror_num == 1);
2215 ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical,
2216 &map_length, &bbio, mirror_num);
2218 btrfs_bio_counter_dec(fs_info);
2222 BUG_ON(mirror_num != bbio->mirror_num);
2225 sector = bbio->stripes[bbio->mirror_num - 1].physical >> 9;
2226 bio->bi_iter.bi_sector = sector;
2227 dev = bbio->stripes[bbio->mirror_num - 1].dev;
2228 btrfs_put_bbio(bbio);
2229 if (!dev || !dev->bdev ||
2230 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) {
2231 btrfs_bio_counter_dec(fs_info);
2235 bio_set_dev(bio, dev->bdev);
2236 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC;
2237 bio_add_page(bio, page, length, pg_offset);
2239 if (btrfsic_submit_bio_wait(bio)) {
2240 /* try to remap that extent elsewhere? */
2241 btrfs_bio_counter_dec(fs_info);
2243 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
2247 btrfs_info_rl_in_rcu(fs_info,
2248 "read error corrected: ino %llu off %llu (dev %s sector %llu)",
2250 rcu_str_deref(dev->name), sector);
2251 btrfs_bio_counter_dec(fs_info);
2256 int btrfs_repair_eb_io_failure(struct extent_buffer *eb, int mirror_num)
2258 struct btrfs_fs_info *fs_info = eb->fs_info;
2259 u64 start = eb->start;
2260 int i, num_pages = num_extent_pages(eb);
2263 if (sb_rdonly(fs_info->sb))
2266 for (i = 0; i < num_pages; i++) {
2267 struct page *p = eb->pages[i];
2269 ret = repair_io_failure(fs_info, 0, start, PAGE_SIZE, start, p,
2270 start - page_offset(p), mirror_num);
2280 * each time an IO finishes, we do a fast check in the IO failure tree
2281 * to see if we need to process or clean up an io_failure_record
2283 int clean_io_failure(struct btrfs_fs_info *fs_info,
2284 struct extent_io_tree *failure_tree,
2285 struct extent_io_tree *io_tree, u64 start,
2286 struct page *page, u64 ino, unsigned int pg_offset)
2289 struct io_failure_record *failrec;
2290 struct extent_state *state;
2295 ret = count_range_bits(failure_tree, &private, (u64)-1, 1,
2300 ret = get_state_failrec(failure_tree, start, &failrec);
2304 BUG_ON(!failrec->this_mirror);
2306 if (failrec->in_validation) {
2307 /* there was no real error, just free the record */
2308 btrfs_debug(fs_info,
2309 "clean_io_failure: freeing dummy error at %llu",
2313 if (sb_rdonly(fs_info->sb))
2316 spin_lock(&io_tree->lock);
2317 state = find_first_extent_bit_state(io_tree,
2320 spin_unlock(&io_tree->lock);
2322 if (state && state->start <= failrec->start &&
2323 state->end >= failrec->start + failrec->len - 1) {
2324 num_copies = btrfs_num_copies(fs_info, failrec->logical,
2326 if (num_copies > 1) {
2327 repair_io_failure(fs_info, ino, start, failrec->len,
2328 failrec->logical, page, pg_offset,
2329 failrec->failed_mirror);
2334 free_io_failure(failure_tree, io_tree, failrec);
2340 * Can be called when
2341 * - hold extent lock
2342 * - under ordered extent
2343 * - the inode is freeing
2345 void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start, u64 end)
2347 struct extent_io_tree *failure_tree = &inode->io_failure_tree;
2348 struct io_failure_record *failrec;
2349 struct extent_state *state, *next;
2351 if (RB_EMPTY_ROOT(&failure_tree->state))
2354 spin_lock(&failure_tree->lock);
2355 state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY);
2357 if (state->start > end)
2360 ASSERT(state->end <= end);
2362 next = next_state(state);
2364 failrec = state->failrec;
2365 free_extent_state(state);
2370 spin_unlock(&failure_tree->lock);
2373 int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end,
2374 struct io_failure_record **failrec_ret)
2376 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2377 struct io_failure_record *failrec;
2378 struct extent_map *em;
2379 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2380 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2381 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2385 ret = get_state_failrec(failure_tree, start, &failrec);
2387 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2391 failrec->start = start;
2392 failrec->len = end - start + 1;
2393 failrec->this_mirror = 0;
2394 failrec->bio_flags = 0;
2395 failrec->in_validation = 0;
2397 read_lock(&em_tree->lock);
2398 em = lookup_extent_mapping(em_tree, start, failrec->len);
2400 read_unlock(&em_tree->lock);
2405 if (em->start > start || em->start + em->len <= start) {
2406 free_extent_map(em);
2409 read_unlock(&em_tree->lock);
2415 logical = start - em->start;
2416 logical = em->block_start + logical;
2417 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2418 logical = em->block_start;
2419 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2420 extent_set_compress_type(&failrec->bio_flags,
2424 btrfs_debug(fs_info,
2425 "Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu",
2426 logical, start, failrec->len);
2428 failrec->logical = logical;
2429 free_extent_map(em);
2431 /* set the bits in the private failure tree */
2432 ret = set_extent_bits(failure_tree, start, end,
2433 EXTENT_LOCKED | EXTENT_DIRTY);
2435 ret = set_state_failrec(failure_tree, start, failrec);
2436 /* set the bits in the inode's tree */
2438 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED);
2444 btrfs_debug(fs_info,
2445 "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d",
2446 failrec->logical, failrec->start, failrec->len,
2447 failrec->in_validation);
2449 * when data can be on disk more than twice, add to failrec here
2450 * (e.g. with a list for failed_mirror) to make
2451 * clean_io_failure() clean all those errors at once.
2455 *failrec_ret = failrec;
2460 bool btrfs_check_repairable(struct inode *inode, unsigned failed_bio_pages,
2461 struct io_failure_record *failrec, int failed_mirror)
2463 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2466 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
2467 if (num_copies == 1) {
2469 * we only have a single copy of the data, so don't bother with
2470 * all the retry and error correction code that follows. no
2471 * matter what the error is, it is very likely to persist.
2473 btrfs_debug(fs_info,
2474 "Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
2475 num_copies, failrec->this_mirror, failed_mirror);
2480 * there are two premises:
2481 * a) deliver good data to the caller
2482 * b) correct the bad sectors on disk
2484 if (failed_bio_pages > 1) {
2486 * to fulfill b), we need to know the exact failing sectors, as
2487 * we don't want to rewrite any more than the failed ones. thus,
2488 * we need separate read requests for the failed bio
2490 * if the following BUG_ON triggers, our validation request got
2491 * merged. we need separate requests for our algorithm to work.
2493 BUG_ON(failrec->in_validation);
2494 failrec->in_validation = 1;
2495 failrec->this_mirror = failed_mirror;
2498 * we're ready to fulfill a) and b) alongside. get a good copy
2499 * of the failed sector and if we succeed, we have setup
2500 * everything for repair_io_failure to do the rest for us.
2502 if (failrec->in_validation) {
2503 BUG_ON(failrec->this_mirror != failed_mirror);
2504 failrec->in_validation = 0;
2505 failrec->this_mirror = 0;
2507 failrec->failed_mirror = failed_mirror;
2508 failrec->this_mirror++;
2509 if (failrec->this_mirror == failed_mirror)
2510 failrec->this_mirror++;
2513 if (failrec->this_mirror > num_copies) {
2514 btrfs_debug(fs_info,
2515 "Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
2516 num_copies, failrec->this_mirror, failed_mirror);
2524 struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio,
2525 struct io_failure_record *failrec,
2526 struct page *page, int pg_offset, int icsum,
2527 bio_end_io_t *endio_func, void *data)
2529 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2531 struct btrfs_io_bio *btrfs_failed_bio;
2532 struct btrfs_io_bio *btrfs_bio;
2534 bio = btrfs_io_bio_alloc(1);
2535 bio->bi_end_io = endio_func;
2536 bio->bi_iter.bi_sector = failrec->logical >> 9;
2537 bio_set_dev(bio, fs_info->fs_devices->latest_bdev);
2538 bio->bi_iter.bi_size = 0;
2539 bio->bi_private = data;
2541 btrfs_failed_bio = btrfs_io_bio(failed_bio);
2542 if (btrfs_failed_bio->csum) {
2543 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
2545 btrfs_bio = btrfs_io_bio(bio);
2546 btrfs_bio->csum = btrfs_bio->csum_inline;
2548 memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + icsum,
2552 bio_add_page(bio, page, failrec->len, pg_offset);
2558 * This is a generic handler for readpage errors. If other copies exist, read
2559 * those and write back good data to the failed position. Does not investigate
2560 * in remapping the failed extent elsewhere, hoping the device will be smart
2561 * enough to do this as needed
2563 static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset,
2564 struct page *page, u64 start, u64 end,
2567 struct io_failure_record *failrec;
2568 struct inode *inode = page->mapping->host;
2569 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2570 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2573 blk_status_t status;
2575 unsigned failed_bio_pages = failed_bio->bi_iter.bi_size >> PAGE_SHIFT;
2577 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
2579 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
2583 if (!btrfs_check_repairable(inode, failed_bio_pages, failrec,
2585 free_io_failure(failure_tree, tree, failrec);
2589 if (failed_bio_pages > 1)
2590 read_mode |= REQ_FAILFAST_DEV;
2592 phy_offset >>= inode->i_sb->s_blocksize_bits;
2593 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
2594 start - page_offset(page),
2595 (int)phy_offset, failed_bio->bi_end_io,
2597 bio->bi_opf = REQ_OP_READ | read_mode;
2599 btrfs_debug(btrfs_sb(inode->i_sb),
2600 "Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d",
2601 read_mode, failrec->this_mirror, failrec->in_validation);
2603 status = tree->ops->submit_bio_hook(tree->private_data, bio, failrec->this_mirror,
2604 failrec->bio_flags);
2606 free_io_failure(failure_tree, tree, failrec);
2608 ret = blk_status_to_errno(status);
2614 /* lots and lots of room for performance fixes in the end_bio funcs */
2616 void end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2618 int uptodate = (err == 0);
2621 btrfs_writepage_endio_finish_ordered(page, start, end, uptodate);
2624 ClearPageUptodate(page);
2626 ret = err < 0 ? err : -EIO;
2627 mapping_set_error(page->mapping, ret);
2632 * after a writepage IO is done, we need to:
2633 * clear the uptodate bits on error
2634 * clear the writeback bits in the extent tree for this IO
2635 * end_page_writeback if the page has no more pending IO
2637 * Scheduling is not allowed, so the extent state tree is expected
2638 * to have one and only one object corresponding to this IO.
2640 static void end_bio_extent_writepage(struct bio *bio)
2642 int error = blk_status_to_errno(bio->bi_status);
2643 struct bio_vec *bvec;
2646 struct bvec_iter_all iter_all;
2648 ASSERT(!bio_flagged(bio, BIO_CLONED));
2649 bio_for_each_segment_all(bvec, bio, iter_all) {
2650 struct page *page = bvec->bv_page;
2651 struct inode *inode = page->mapping->host;
2652 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2654 /* We always issue full-page reads, but if some block
2655 * in a page fails to read, blk_update_request() will
2656 * advance bv_offset and adjust bv_len to compensate.
2657 * Print a warning for nonzero offsets, and an error
2658 * if they don't add up to a full page. */
2659 if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) {
2660 if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE)
2662 "partial page write in btrfs with offset %u and length %u",
2663 bvec->bv_offset, bvec->bv_len);
2666 "incomplete page write in btrfs with offset %u and length %u",
2667 bvec->bv_offset, bvec->bv_len);
2670 start = page_offset(page);
2671 end = start + bvec->bv_offset + bvec->bv_len - 1;
2673 end_extent_writepage(page, error, start, end);
2674 end_page_writeback(page);
2681 endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len,
2684 struct extent_state *cached = NULL;
2685 u64 end = start + len - 1;
2687 if (uptodate && tree->track_uptodate)
2688 set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC);
2689 unlock_extent_cached_atomic(tree, start, end, &cached);
2693 * after a readpage IO is done, we need to:
2694 * clear the uptodate bits on error
2695 * set the uptodate bits if things worked
2696 * set the page up to date if all extents in the tree are uptodate
2697 * clear the lock bit in the extent tree
2698 * unlock the page if there are no other extents locked for it
2700 * Scheduling is not allowed, so the extent state tree is expected
2701 * to have one and only one object corresponding to this IO.
2703 static void end_bio_extent_readpage(struct bio *bio)
2705 struct bio_vec *bvec;
2706 int uptodate = !bio->bi_status;
2707 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
2708 struct extent_io_tree *tree, *failure_tree;
2713 u64 extent_start = 0;
2717 struct bvec_iter_all iter_all;
2719 ASSERT(!bio_flagged(bio, BIO_CLONED));
2720 bio_for_each_segment_all(bvec, bio, iter_all) {
2721 struct page *page = bvec->bv_page;
2722 struct inode *inode = page->mapping->host;
2723 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2724 bool data_inode = btrfs_ino(BTRFS_I(inode))
2725 != BTRFS_BTREE_INODE_OBJECTID;
2727 btrfs_debug(fs_info,
2728 "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u",
2729 (u64)bio->bi_iter.bi_sector, bio->bi_status,
2730 io_bio->mirror_num);
2731 tree = &BTRFS_I(inode)->io_tree;
2732 failure_tree = &BTRFS_I(inode)->io_failure_tree;
2734 /* We always issue full-page reads, but if some block
2735 * in a page fails to read, blk_update_request() will
2736 * advance bv_offset and adjust bv_len to compensate.
2737 * Print a warning for nonzero offsets, and an error
2738 * if they don't add up to a full page. */
2739 if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) {
2740 if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE)
2742 "partial page read in btrfs with offset %u and length %u",
2743 bvec->bv_offset, bvec->bv_len);
2746 "incomplete page read in btrfs with offset %u and length %u",
2747 bvec->bv_offset, bvec->bv_len);
2750 start = page_offset(page);
2751 end = start + bvec->bv_offset + bvec->bv_len - 1;
2754 mirror = io_bio->mirror_num;
2755 if (likely(uptodate)) {
2756 ret = tree->ops->readpage_end_io_hook(io_bio, offset,
2762 clean_io_failure(BTRFS_I(inode)->root->fs_info,
2763 failure_tree, tree, start,
2765 btrfs_ino(BTRFS_I(inode)), 0);
2768 if (likely(uptodate))
2774 * The generic bio_readpage_error handles errors the
2775 * following way: If possible, new read requests are
2776 * created and submitted and will end up in
2777 * end_bio_extent_readpage as well (if we're lucky,
2778 * not in the !uptodate case). In that case it returns
2779 * 0 and we just go on with the next page in our bio.
2780 * If it can't handle the error it will return -EIO and
2781 * we remain responsible for that page.
2783 ret = bio_readpage_error(bio, offset, page, start, end,
2786 uptodate = !bio->bi_status;
2791 struct extent_buffer *eb;
2793 eb = (struct extent_buffer *)page->private;
2794 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
2795 eb->read_mirror = mirror;
2796 atomic_dec(&eb->io_pages);
2797 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD,
2799 btree_readahead_hook(eb, -EIO);
2802 if (likely(uptodate)) {
2803 loff_t i_size = i_size_read(inode);
2804 pgoff_t end_index = i_size >> PAGE_SHIFT;
2807 /* Zero out the end if this page straddles i_size */
2808 off = offset_in_page(i_size);
2809 if (page->index == end_index && off)
2810 zero_user_segment(page, off, PAGE_SIZE);
2811 SetPageUptodate(page);
2813 ClearPageUptodate(page);
2819 if (unlikely(!uptodate)) {
2821 endio_readpage_release_extent(tree,
2827 endio_readpage_release_extent(tree, start,
2828 end - start + 1, 0);
2829 } else if (!extent_len) {
2830 extent_start = start;
2831 extent_len = end + 1 - start;
2832 } else if (extent_start + extent_len == start) {
2833 extent_len += end + 1 - start;
2835 endio_readpage_release_extent(tree, extent_start,
2836 extent_len, uptodate);
2837 extent_start = start;
2838 extent_len = end + 1 - start;
2843 endio_readpage_release_extent(tree, extent_start, extent_len,
2845 btrfs_io_bio_free_csum(io_bio);
2850 * Initialize the members up to but not including 'bio'. Use after allocating a
2851 * new bio by bio_alloc_bioset as it does not initialize the bytes outside of
2852 * 'bio' because use of __GFP_ZERO is not supported.
2854 static inline void btrfs_io_bio_init(struct btrfs_io_bio *btrfs_bio)
2856 memset(btrfs_bio, 0, offsetof(struct btrfs_io_bio, bio));
2860 * The following helpers allocate a bio. As it's backed by a bioset, it'll
2861 * never fail. We're returning a bio right now but you can call btrfs_io_bio
2862 * for the appropriate container_of magic
2864 struct bio *btrfs_bio_alloc(u64 first_byte)
2868 bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, &btrfs_bioset);
2869 bio->bi_iter.bi_sector = first_byte >> 9;
2870 btrfs_io_bio_init(btrfs_io_bio(bio));
2874 struct bio *btrfs_bio_clone(struct bio *bio)
2876 struct btrfs_io_bio *btrfs_bio;
2879 /* Bio allocation backed by a bioset does not fail */
2880 new = bio_clone_fast(bio, GFP_NOFS, &btrfs_bioset);
2881 btrfs_bio = btrfs_io_bio(new);
2882 btrfs_io_bio_init(btrfs_bio);
2883 btrfs_bio->iter = bio->bi_iter;
2887 struct bio *btrfs_io_bio_alloc(unsigned int nr_iovecs)
2891 /* Bio allocation backed by a bioset does not fail */
2892 bio = bio_alloc_bioset(GFP_NOFS, nr_iovecs, &btrfs_bioset);
2893 btrfs_io_bio_init(btrfs_io_bio(bio));
2897 struct bio *btrfs_bio_clone_partial(struct bio *orig, int offset, int size)
2900 struct btrfs_io_bio *btrfs_bio;
2902 /* this will never fail when it's backed by a bioset */
2903 bio = bio_clone_fast(orig, GFP_NOFS, &btrfs_bioset);
2906 btrfs_bio = btrfs_io_bio(bio);
2907 btrfs_io_bio_init(btrfs_bio);
2909 bio_trim(bio, offset >> 9, size >> 9);
2910 btrfs_bio->iter = bio->bi_iter;
2915 * @opf: bio REQ_OP_* and REQ_* flags as one value
2916 * @tree: tree so we can call our merge_bio hook
2917 * @wbc: optional writeback control for io accounting
2918 * @page: page to add to the bio
2919 * @pg_offset: offset of the new bio or to check whether we are adding
2920 * a contiguous page to the previous one
2921 * @size: portion of page that we want to write
2922 * @offset: starting offset in the page
2923 * @bdev: attach newly created bios to this bdev
2924 * @bio_ret: must be valid pointer, newly allocated bio will be stored there
2925 * @end_io_func: end_io callback for new bio
2926 * @mirror_num: desired mirror to read/write
2927 * @prev_bio_flags: flags of previous bio to see if we can merge the current one
2928 * @bio_flags: flags of the current bio to see if we can merge them
2930 static int submit_extent_page(unsigned int opf, struct extent_io_tree *tree,
2931 struct writeback_control *wbc,
2932 struct page *page, u64 offset,
2933 size_t size, unsigned long pg_offset,
2934 struct block_device *bdev,
2935 struct bio **bio_ret,
2936 bio_end_io_t end_io_func,
2938 unsigned long prev_bio_flags,
2939 unsigned long bio_flags,
2940 bool force_bio_submit)
2944 size_t page_size = min_t(size_t, size, PAGE_SIZE);
2945 sector_t sector = offset >> 9;
2951 bool can_merge = true;
2954 if (prev_bio_flags & EXTENT_BIO_COMPRESSED)
2955 contig = bio->bi_iter.bi_sector == sector;
2957 contig = bio_end_sector(bio) == sector;
2960 if (btrfs_bio_fits_in_stripe(page, page_size, bio, bio_flags))
2963 if (prev_bio_flags != bio_flags || !contig || !can_merge ||
2965 bio_add_page(bio, page, page_size, pg_offset) < page_size) {
2966 ret = submit_one_bio(bio, mirror_num, prev_bio_flags);
2974 wbc_account_cgroup_owner(wbc, page, page_size);
2979 bio = btrfs_bio_alloc(offset);
2980 bio_set_dev(bio, bdev);
2981 bio_add_page(bio, page, page_size, pg_offset);
2982 bio->bi_end_io = end_io_func;
2983 bio->bi_private = tree;
2984 bio->bi_write_hint = page->mapping->host->i_write_hint;
2987 wbc_init_bio(wbc, bio);
2988 wbc_account_cgroup_owner(wbc, page, page_size);
2996 static void attach_extent_buffer_page(struct extent_buffer *eb,
2999 if (!PagePrivate(page)) {
3000 SetPagePrivate(page);
3002 set_page_private(page, (unsigned long)eb);
3004 WARN_ON(page->private != (unsigned long)eb);
3008 void set_page_extent_mapped(struct page *page)
3010 if (!PagePrivate(page)) {
3011 SetPagePrivate(page);
3013 set_page_private(page, EXTENT_PAGE_PRIVATE);
3017 static struct extent_map *
3018 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
3019 u64 start, u64 len, get_extent_t *get_extent,
3020 struct extent_map **em_cached)
3022 struct extent_map *em;
3024 if (em_cached && *em_cached) {
3026 if (extent_map_in_tree(em) && start >= em->start &&
3027 start < extent_map_end(em)) {
3028 refcount_inc(&em->refs);
3032 free_extent_map(em);
3036 em = get_extent(BTRFS_I(inode), page, pg_offset, start, len, 0);
3037 if (em_cached && !IS_ERR_OR_NULL(em)) {
3039 refcount_inc(&em->refs);
3045 * basic readpage implementation. Locked extent state structs are inserted
3046 * into the tree that are removed when the IO is done (by the end_io
3048 * XXX JDM: This needs looking at to ensure proper page locking
3049 * return 0 on success, otherwise return error
3051 static int __do_readpage(struct extent_io_tree *tree,
3053 get_extent_t *get_extent,
3054 struct extent_map **em_cached,
3055 struct bio **bio, int mirror_num,
3056 unsigned long *bio_flags, unsigned int read_flags,
3059 struct inode *inode = page->mapping->host;
3060 u64 start = page_offset(page);
3061 const u64 end = start + PAGE_SIZE - 1;
3064 u64 last_byte = i_size_read(inode);
3067 struct extent_map *em;
3068 struct block_device *bdev;
3071 size_t pg_offset = 0;
3073 size_t disk_io_size;
3074 size_t blocksize = inode->i_sb->s_blocksize;
3075 unsigned long this_bio_flag = 0;
3077 set_page_extent_mapped(page);
3079 if (!PageUptodate(page)) {
3080 if (cleancache_get_page(page) == 0) {
3081 BUG_ON(blocksize != PAGE_SIZE);
3082 unlock_extent(tree, start, end);
3087 if (page->index == last_byte >> PAGE_SHIFT) {
3089 size_t zero_offset = offset_in_page(last_byte);
3092 iosize = PAGE_SIZE - zero_offset;
3093 userpage = kmap_atomic(page);
3094 memset(userpage + zero_offset, 0, iosize);
3095 flush_dcache_page(page);
3096 kunmap_atomic(userpage);
3099 while (cur <= end) {
3100 bool force_bio_submit = false;
3103 if (cur >= last_byte) {
3105 struct extent_state *cached = NULL;
3107 iosize = PAGE_SIZE - pg_offset;
3108 userpage = kmap_atomic(page);
3109 memset(userpage + pg_offset, 0, iosize);
3110 flush_dcache_page(page);
3111 kunmap_atomic(userpage);
3112 set_extent_uptodate(tree, cur, cur + iosize - 1,
3114 unlock_extent_cached(tree, cur,
3115 cur + iosize - 1, &cached);
3118 em = __get_extent_map(inode, page, pg_offset, cur,
3119 end - cur + 1, get_extent, em_cached);
3120 if (IS_ERR_OR_NULL(em)) {
3122 unlock_extent(tree, cur, end);
3125 extent_offset = cur - em->start;
3126 BUG_ON(extent_map_end(em) <= cur);
3129 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
3130 this_bio_flag |= EXTENT_BIO_COMPRESSED;
3131 extent_set_compress_type(&this_bio_flag,
3135 iosize = min(extent_map_end(em) - cur, end - cur + 1);
3136 cur_end = min(extent_map_end(em) - 1, end);
3137 iosize = ALIGN(iosize, blocksize);
3138 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
3139 disk_io_size = em->block_len;
3140 offset = em->block_start;
3142 offset = em->block_start + extent_offset;
3143 disk_io_size = iosize;
3146 block_start = em->block_start;
3147 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
3148 block_start = EXTENT_MAP_HOLE;
3151 * If we have a file range that points to a compressed extent
3152 * and it's followed by a consecutive file range that points to
3153 * to the same compressed extent (possibly with a different
3154 * offset and/or length, so it either points to the whole extent
3155 * or only part of it), we must make sure we do not submit a
3156 * single bio to populate the pages for the 2 ranges because
3157 * this makes the compressed extent read zero out the pages
3158 * belonging to the 2nd range. Imagine the following scenario:
3161 * [0 - 8K] [8K - 24K]
3164 * points to extent X, points to extent X,
3165 * offset 4K, length of 8K offset 0, length 16K
3167 * [extent X, compressed length = 4K uncompressed length = 16K]
3169 * If the bio to read the compressed extent covers both ranges,
3170 * it will decompress extent X into the pages belonging to the
3171 * first range and then it will stop, zeroing out the remaining
3172 * pages that belong to the other range that points to extent X.
3173 * So here we make sure we submit 2 bios, one for the first
3174 * range and another one for the third range. Both will target
3175 * the same physical extent from disk, but we can't currently
3176 * make the compressed bio endio callback populate the pages
3177 * for both ranges because each compressed bio is tightly
3178 * coupled with a single extent map, and each range can have
3179 * an extent map with a different offset value relative to the
3180 * uncompressed data of our extent and different lengths. This
3181 * is a corner case so we prioritize correctness over
3182 * non-optimal behavior (submitting 2 bios for the same extent).
3184 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) &&
3185 prev_em_start && *prev_em_start != (u64)-1 &&
3186 *prev_em_start != em->start)
3187 force_bio_submit = true;
3190 *prev_em_start = em->start;
3192 free_extent_map(em);
3195 /* we've found a hole, just zero and go on */
3196 if (block_start == EXTENT_MAP_HOLE) {
3198 struct extent_state *cached = NULL;
3200 userpage = kmap_atomic(page);
3201 memset(userpage + pg_offset, 0, iosize);
3202 flush_dcache_page(page);
3203 kunmap_atomic(userpage);
3205 set_extent_uptodate(tree, cur, cur + iosize - 1,
3207 unlock_extent_cached(tree, cur,
3208 cur + iosize - 1, &cached);
3210 pg_offset += iosize;
3213 /* the get_extent function already copied into the page */
3214 if (test_range_bit(tree, cur, cur_end,
3215 EXTENT_UPTODATE, 1, NULL)) {
3216 check_page_uptodate(tree, page);
3217 unlock_extent(tree, cur, cur + iosize - 1);
3219 pg_offset += iosize;
3222 /* we have an inline extent but it didn't get marked up
3223 * to date. Error out
3225 if (block_start == EXTENT_MAP_INLINE) {
3227 unlock_extent(tree, cur, cur + iosize - 1);
3229 pg_offset += iosize;
3233 ret = submit_extent_page(REQ_OP_READ | read_flags, tree, NULL,
3234 page, offset, disk_io_size,
3235 pg_offset, bdev, bio,
3236 end_bio_extent_readpage, mirror_num,
3242 *bio_flags = this_bio_flag;
3245 unlock_extent(tree, cur, cur + iosize - 1);
3249 pg_offset += iosize;
3253 if (!PageError(page))
3254 SetPageUptodate(page);
3260 static inline void contiguous_readpages(struct extent_io_tree *tree,
3261 struct page *pages[], int nr_pages,
3263 struct extent_map **em_cached,
3265 unsigned long *bio_flags,
3268 struct btrfs_inode *inode = BTRFS_I(pages[0]->mapping->host);
3271 btrfs_lock_and_flush_ordered_range(tree, inode, start, end, NULL);
3273 for (index = 0; index < nr_pages; index++) {
3274 __do_readpage(tree, pages[index], btrfs_get_extent, em_cached,
3275 bio, 0, bio_flags, REQ_RAHEAD, prev_em_start);
3276 put_page(pages[index]);
3280 static int __extent_read_full_page(struct extent_io_tree *tree,
3282 get_extent_t *get_extent,
3283 struct bio **bio, int mirror_num,
3284 unsigned long *bio_flags,
3285 unsigned int read_flags)
3287 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
3288 u64 start = page_offset(page);
3289 u64 end = start + PAGE_SIZE - 1;
3292 btrfs_lock_and_flush_ordered_range(tree, inode, start, end, NULL);
3294 ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num,
3295 bio_flags, read_flags, NULL);
3299 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
3300 get_extent_t *get_extent, int mirror_num)
3302 struct bio *bio = NULL;
3303 unsigned long bio_flags = 0;
3306 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
3309 ret = submit_one_bio(bio, mirror_num, bio_flags);
3313 static void update_nr_written(struct writeback_control *wbc,
3314 unsigned long nr_written)
3316 wbc->nr_to_write -= nr_written;
3320 * helper for __extent_writepage, doing all of the delayed allocation setup.
3322 * This returns 1 if btrfs_run_delalloc_range function did all the work required
3323 * to write the page (copy into inline extent). In this case the IO has
3324 * been started and the page is already unlocked.
3326 * This returns 0 if all went well (page still locked)
3327 * This returns < 0 if there were errors (page still locked)
3329 static noinline_for_stack int writepage_delalloc(struct inode *inode,
3330 struct page *page, struct writeback_control *wbc,
3331 u64 delalloc_start, unsigned long *nr_written)
3333 u64 page_end = delalloc_start + PAGE_SIZE - 1;
3335 u64 delalloc_to_write = 0;
3336 u64 delalloc_end = 0;
3338 int page_started = 0;
3341 while (delalloc_end < page_end) {
3342 found = find_lock_delalloc_range(inode, page,
3346 delalloc_start = delalloc_end + 1;
3349 ret = btrfs_run_delalloc_range(inode, page, delalloc_start,
3350 delalloc_end, &page_started, nr_written, wbc);
3354 * btrfs_run_delalloc_range should return < 0 for error
3355 * but just in case, we use > 0 here meaning the IO is
3356 * started, so we don't want to return > 0 unless
3357 * things are going well.
3359 ret = ret < 0 ? ret : -EIO;
3363 * delalloc_end is already one less than the total length, so
3364 * we don't subtract one from PAGE_SIZE
3366 delalloc_to_write += (delalloc_end - delalloc_start +
3367 PAGE_SIZE) >> PAGE_SHIFT;
3368 delalloc_start = delalloc_end + 1;
3370 if (wbc->nr_to_write < delalloc_to_write) {
3373 if (delalloc_to_write < thresh * 2)
3374 thresh = delalloc_to_write;
3375 wbc->nr_to_write = min_t(u64, delalloc_to_write,
3379 /* did the fill delalloc function already unlock and start
3384 * we've unlocked the page, so we can't update
3385 * the mapping's writeback index, just update
3388 wbc->nr_to_write -= *nr_written;
3399 * helper for __extent_writepage. This calls the writepage start hooks,
3400 * and does the loop to map the page into extents and bios.
3402 * We return 1 if the IO is started and the page is unlocked,
3403 * 0 if all went well (page still locked)
3404 * < 0 if there were errors (page still locked)
3406 static noinline_for_stack int __extent_writepage_io(struct inode *inode,
3408 struct writeback_control *wbc,
3409 struct extent_page_data *epd,
3411 unsigned long nr_written,
3412 unsigned int write_flags, int *nr_ret)
3414 struct extent_io_tree *tree = epd->tree;
3415 u64 start = page_offset(page);
3416 u64 page_end = start + PAGE_SIZE - 1;
3422 struct extent_map *em;
3423 struct block_device *bdev;
3424 size_t pg_offset = 0;
3430 ret = btrfs_writepage_cow_fixup(page, start, page_end);
3432 /* Fixup worker will requeue */
3434 wbc->pages_skipped++;
3436 redirty_page_for_writepage(wbc, page);
3438 update_nr_written(wbc, nr_written);
3444 * we don't want to touch the inode after unlocking the page,
3445 * so we update the mapping writeback index now
3447 update_nr_written(wbc, nr_written + 1);
3450 if (i_size <= start) {
3451 btrfs_writepage_endio_finish_ordered(page, start, page_end, 1);
3455 blocksize = inode->i_sb->s_blocksize;
3457 while (cur <= end) {
3461 if (cur >= i_size) {
3462 btrfs_writepage_endio_finish_ordered(page, cur,
3466 em = btrfs_get_extent(BTRFS_I(inode), page, pg_offset, cur,
3468 if (IS_ERR_OR_NULL(em)) {
3470 ret = PTR_ERR_OR_ZERO(em);
3474 extent_offset = cur - em->start;
3475 em_end = extent_map_end(em);
3476 BUG_ON(em_end <= cur);
3478 iosize = min(em_end - cur, end - cur + 1);
3479 iosize = ALIGN(iosize, blocksize);
3480 offset = em->block_start + extent_offset;
3482 block_start = em->block_start;
3483 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3484 free_extent_map(em);
3488 * compressed and inline extents are written through other
3491 if (compressed || block_start == EXTENT_MAP_HOLE ||
3492 block_start == EXTENT_MAP_INLINE) {
3494 * end_io notification does not happen here for
3495 * compressed extents
3498 btrfs_writepage_endio_finish_ordered(page, cur,
3501 else if (compressed) {
3502 /* we don't want to end_page_writeback on
3503 * a compressed extent. this happens
3510 pg_offset += iosize;
3514 btrfs_set_range_writeback(tree, cur, cur + iosize - 1);
3515 if (!PageWriteback(page)) {
3516 btrfs_err(BTRFS_I(inode)->root->fs_info,
3517 "page %lu not writeback, cur %llu end %llu",
3518 page->index, cur, end);
3521 ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc,
3522 page, offset, iosize, pg_offset,
3524 end_bio_extent_writepage,
3528 if (PageWriteback(page))
3529 end_page_writeback(page);
3533 pg_offset += iosize;
3542 * the writepage semantics are similar to regular writepage. extent
3543 * records are inserted to lock ranges in the tree, and as dirty areas
3544 * are found, they are marked writeback. Then the lock bits are removed
3545 * and the end_io handler clears the writeback ranges
3547 * Return 0 if everything goes well.
3548 * Return <0 for error.
3550 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
3551 struct extent_page_data *epd)
3553 struct inode *inode = page->mapping->host;
3554 u64 start = page_offset(page);
3555 u64 page_end = start + PAGE_SIZE - 1;
3558 size_t pg_offset = 0;
3559 loff_t i_size = i_size_read(inode);
3560 unsigned long end_index = i_size >> PAGE_SHIFT;
3561 unsigned int write_flags = 0;
3562 unsigned long nr_written = 0;
3564 write_flags = wbc_to_write_flags(wbc);
3566 trace___extent_writepage(page, inode, wbc);
3568 WARN_ON(!PageLocked(page));
3570 ClearPageError(page);
3572 pg_offset = offset_in_page(i_size);
3573 if (page->index > end_index ||
3574 (page->index == end_index && !pg_offset)) {
3575 page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
3580 if (page->index == end_index) {
3583 userpage = kmap_atomic(page);
3584 memset(userpage + pg_offset, 0,
3585 PAGE_SIZE - pg_offset);
3586 kunmap_atomic(userpage);
3587 flush_dcache_page(page);
3592 set_page_extent_mapped(page);
3594 if (!epd->extent_locked) {
3595 ret = writepage_delalloc(inode, page, wbc, start, &nr_written);
3602 ret = __extent_writepage_io(inode, page, wbc, epd,
3603 i_size, nr_written, write_flags, &nr);
3609 /* make sure the mapping tag for page dirty gets cleared */
3610 set_page_writeback(page);
3611 end_page_writeback(page);
3613 if (PageError(page)) {
3614 ret = ret < 0 ? ret : -EIO;
3615 end_extent_writepage(page, ret, start, page_end);
3625 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3627 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
3628 TASK_UNINTERRUPTIBLE);
3632 * Lock eb pages and flush the bio if we can't the locks
3634 * Return 0 if nothing went wrong
3635 * Return >0 is same as 0, except bio is not submitted
3636 * Return <0 if something went wrong, no page is locked
3638 static noinline_for_stack int lock_extent_buffer_for_io(struct extent_buffer *eb,
3639 struct extent_page_data *epd)
3641 struct btrfs_fs_info *fs_info = eb->fs_info;
3642 int i, num_pages, failed_page_nr;
3646 if (!btrfs_try_tree_write_lock(eb)) {
3647 ret = flush_write_bio(epd);
3651 btrfs_tree_lock(eb);
3654 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3655 btrfs_tree_unlock(eb);
3659 ret = flush_write_bio(epd);
3665 wait_on_extent_buffer_writeback(eb);
3666 btrfs_tree_lock(eb);
3667 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3669 btrfs_tree_unlock(eb);
3674 * We need to do this to prevent races in people who check if the eb is
3675 * under IO since we can end up having no IO bits set for a short period
3678 spin_lock(&eb->refs_lock);
3679 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3680 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3681 spin_unlock(&eb->refs_lock);
3682 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3683 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
3685 fs_info->dirty_metadata_batch);
3688 spin_unlock(&eb->refs_lock);
3691 btrfs_tree_unlock(eb);
3696 num_pages = num_extent_pages(eb);
3697 for (i = 0; i < num_pages; i++) {
3698 struct page *p = eb->pages[i];
3700 if (!trylock_page(p)) {
3702 ret = flush_write_bio(epd);
3715 /* Unlock already locked pages */
3716 for (i = 0; i < failed_page_nr; i++)
3717 unlock_page(eb->pages[i]);
3721 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3723 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3724 smp_mb__after_atomic();
3725 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3728 static void set_btree_ioerr(struct page *page)
3730 struct extent_buffer *eb = (struct extent_buffer *)page->private;
3733 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))
3737 * If writeback for a btree extent that doesn't belong to a log tree
3738 * failed, increment the counter transaction->eb_write_errors.
3739 * We do this because while the transaction is running and before it's
3740 * committing (when we call filemap_fdata[write|wait]_range against
3741 * the btree inode), we might have
3742 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
3743 * returns an error or an error happens during writeback, when we're
3744 * committing the transaction we wouldn't know about it, since the pages
3745 * can be no longer dirty nor marked anymore for writeback (if a
3746 * subsequent modification to the extent buffer didn't happen before the
3747 * transaction commit), which makes filemap_fdata[write|wait]_range not
3748 * able to find the pages tagged with SetPageError at transaction
3749 * commit time. So if this happens we must abort the transaction,
3750 * otherwise we commit a super block with btree roots that point to
3751 * btree nodes/leafs whose content on disk is invalid - either garbage
3752 * or the content of some node/leaf from a past generation that got
3753 * cowed or deleted and is no longer valid.
3755 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
3756 * not be enough - we need to distinguish between log tree extents vs
3757 * non-log tree extents, and the next filemap_fdatawait_range() call
3758 * will catch and clear such errors in the mapping - and that call might
3759 * be from a log sync and not from a transaction commit. Also, checking
3760 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
3761 * not done and would not be reliable - the eb might have been released
3762 * from memory and reading it back again means that flag would not be
3763 * set (since it's a runtime flag, not persisted on disk).
3765 * Using the flags below in the btree inode also makes us achieve the
3766 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
3767 * writeback for all dirty pages and before filemap_fdatawait_range()
3768 * is called, the writeback for all dirty pages had already finished
3769 * with errors - because we were not using AS_EIO/AS_ENOSPC,
3770 * filemap_fdatawait_range() would return success, as it could not know
3771 * that writeback errors happened (the pages were no longer tagged for
3774 switch (eb->log_index) {
3776 set_bit(BTRFS_FS_BTREE_ERR, &eb->fs_info->flags);
3779 set_bit(BTRFS_FS_LOG1_ERR, &eb->fs_info->flags);
3782 set_bit(BTRFS_FS_LOG2_ERR, &eb->fs_info->flags);
3785 BUG(); /* unexpected, logic error */
3789 static void end_bio_extent_buffer_writepage(struct bio *bio)
3791 struct bio_vec *bvec;
3792 struct extent_buffer *eb;
3794 struct bvec_iter_all iter_all;
3796 ASSERT(!bio_flagged(bio, BIO_CLONED));
3797 bio_for_each_segment_all(bvec, bio, iter_all) {
3798 struct page *page = bvec->bv_page;
3800 eb = (struct extent_buffer *)page->private;
3802 done = atomic_dec_and_test(&eb->io_pages);
3804 if (bio->bi_status ||
3805 test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) {
3806 ClearPageUptodate(page);
3807 set_btree_ioerr(page);
3810 end_page_writeback(page);
3815 end_extent_buffer_writeback(eb);
3821 static noinline_for_stack int write_one_eb(struct extent_buffer *eb,
3822 struct writeback_control *wbc,
3823 struct extent_page_data *epd)
3825 struct btrfs_fs_info *fs_info = eb->fs_info;
3826 struct block_device *bdev = fs_info->fs_devices->latest_bdev;
3827 struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
3828 u64 offset = eb->start;
3831 unsigned long start, end;
3832 unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META;
3835 clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
3836 num_pages = num_extent_pages(eb);
3837 atomic_set(&eb->io_pages, num_pages);
3839 /* set btree blocks beyond nritems with 0 to avoid stale content. */
3840 nritems = btrfs_header_nritems(eb);
3841 if (btrfs_header_level(eb) > 0) {
3842 end = btrfs_node_key_ptr_offset(nritems);
3844 memzero_extent_buffer(eb, end, eb->len - end);
3848 * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0
3850 start = btrfs_item_nr_offset(nritems);
3851 end = BTRFS_LEAF_DATA_OFFSET + leaf_data_end(eb);
3852 memzero_extent_buffer(eb, start, end - start);
3855 for (i = 0; i < num_pages; i++) {
3856 struct page *p = eb->pages[i];
3858 clear_page_dirty_for_io(p);
3859 set_page_writeback(p);
3860 ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc,
3861 p, offset, PAGE_SIZE, 0, bdev,
3863 end_bio_extent_buffer_writepage,
3867 if (PageWriteback(p))
3868 end_page_writeback(p);
3869 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3870 end_extent_buffer_writeback(eb);
3874 offset += PAGE_SIZE;
3875 update_nr_written(wbc, 1);
3879 if (unlikely(ret)) {
3880 for (; i < num_pages; i++) {
3881 struct page *p = eb->pages[i];
3882 clear_page_dirty_for_io(p);
3890 int btree_write_cache_pages(struct address_space *mapping,
3891 struct writeback_control *wbc)
3893 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
3894 struct extent_buffer *eb, *prev_eb = NULL;
3895 struct extent_page_data epd = {
3899 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3903 int nr_to_write_done = 0;
3904 struct pagevec pvec;
3907 pgoff_t end; /* Inclusive */
3911 pagevec_init(&pvec);
3912 if (wbc->range_cyclic) {
3913 index = mapping->writeback_index; /* Start from prev offset */
3916 index = wbc->range_start >> PAGE_SHIFT;
3917 end = wbc->range_end >> PAGE_SHIFT;
3920 if (wbc->sync_mode == WB_SYNC_ALL)
3921 tag = PAGECACHE_TAG_TOWRITE;
3923 tag = PAGECACHE_TAG_DIRTY;
3925 if (wbc->sync_mode == WB_SYNC_ALL)
3926 tag_pages_for_writeback(mapping, index, end);
3927 while (!done && !nr_to_write_done && (index <= end) &&
3928 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
3933 for (i = 0; i < nr_pages; i++) {
3934 struct page *page = pvec.pages[i];
3936 if (!PagePrivate(page))
3939 spin_lock(&mapping->private_lock);
3940 if (!PagePrivate(page)) {
3941 spin_unlock(&mapping->private_lock);
3945 eb = (struct extent_buffer *)page->private;
3948 * Shouldn't happen and normally this would be a BUG_ON
3949 * but no sense in crashing the users box for something
3950 * we can survive anyway.
3953 spin_unlock(&mapping->private_lock);
3957 if (eb == prev_eb) {
3958 spin_unlock(&mapping->private_lock);
3962 ret = atomic_inc_not_zero(&eb->refs);
3963 spin_unlock(&mapping->private_lock);
3968 ret = lock_extent_buffer_for_io(eb, &epd);
3970 free_extent_buffer(eb);
3974 ret = write_one_eb(eb, wbc, &epd);
3977 free_extent_buffer(eb);
3980 free_extent_buffer(eb);
3983 * the filesystem may choose to bump up nr_to_write.
3984 * We have to make sure to honor the new nr_to_write
3987 nr_to_write_done = wbc->nr_to_write <= 0;
3989 pagevec_release(&pvec);
3992 if (!scanned && !done) {
3994 * We hit the last page and there is more work to be done: wrap
3995 * back to the start of the file
4003 end_write_bio(&epd, ret);
4006 ret = flush_write_bio(&epd);
4011 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
4012 * @mapping: address space structure to write
4013 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
4014 * @data: data passed to __extent_writepage function
4016 * If a page is already under I/O, write_cache_pages() skips it, even
4017 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
4018 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
4019 * and msync() need to guarantee that all the data which was dirty at the time
4020 * the call was made get new I/O started against them. If wbc->sync_mode is
4021 * WB_SYNC_ALL then we were called for data integrity and we must wait for
4022 * existing IO to complete.
4024 static int extent_write_cache_pages(struct address_space *mapping,
4025 struct writeback_control *wbc,
4026 struct extent_page_data *epd)
4028 struct inode *inode = mapping->host;
4031 int nr_to_write_done = 0;
4032 struct pagevec pvec;
4035 pgoff_t end; /* Inclusive */
4037 int range_whole = 0;
4042 * We have to hold onto the inode so that ordered extents can do their
4043 * work when the IO finishes. The alternative to this is failing to add
4044 * an ordered extent if the igrab() fails there and that is a huge pain
4045 * to deal with, so instead just hold onto the inode throughout the
4046 * writepages operation. If it fails here we are freeing up the inode
4047 * anyway and we'd rather not waste our time writing out stuff that is
4048 * going to be truncated anyway.
4053 pagevec_init(&pvec);
4054 if (wbc->range_cyclic) {
4055 index = mapping->writeback_index; /* Start from prev offset */
4058 index = wbc->range_start >> PAGE_SHIFT;
4059 end = wbc->range_end >> PAGE_SHIFT;
4060 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
4066 * We do the tagged writepage as long as the snapshot flush bit is set
4067 * and we are the first one who do the filemap_flush() on this inode.
4069 * The nr_to_write == LONG_MAX is needed to make sure other flushers do
4070 * not race in and drop the bit.
4072 if (range_whole && wbc->nr_to_write == LONG_MAX &&
4073 test_and_clear_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
4074 &BTRFS_I(inode)->runtime_flags))
4075 wbc->tagged_writepages = 1;
4077 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
4078 tag = PAGECACHE_TAG_TOWRITE;
4080 tag = PAGECACHE_TAG_DIRTY;
4082 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
4083 tag_pages_for_writeback(mapping, index, end);
4085 while (!done && !nr_to_write_done && (index <= end) &&
4086 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping,
4087 &index, end, tag))) {
4091 for (i = 0; i < nr_pages; i++) {
4092 struct page *page = pvec.pages[i];
4094 done_index = page->index;
4096 * At this point we hold neither the i_pages lock nor
4097 * the page lock: the page may be truncated or
4098 * invalidated (changing page->mapping to NULL),
4099 * or even swizzled back from swapper_space to
4100 * tmpfs file mapping
4102 if (!trylock_page(page)) {
4103 ret = flush_write_bio(epd);
4108 if (unlikely(page->mapping != mapping)) {
4113 if (wbc->sync_mode != WB_SYNC_NONE) {
4114 if (PageWriteback(page)) {
4115 ret = flush_write_bio(epd);
4118 wait_on_page_writeback(page);
4121 if (PageWriteback(page) ||
4122 !clear_page_dirty_for_io(page)) {
4127 ret = __extent_writepage(page, wbc, epd);
4130 * done_index is set past this page,
4131 * so media errors will not choke
4132 * background writeout for the entire
4133 * file. This has consequences for
4134 * range_cyclic semantics (ie. it may
4135 * not be suitable for data integrity
4138 done_index = page->index + 1;
4144 * the filesystem may choose to bump up nr_to_write.
4145 * We have to make sure to honor the new nr_to_write
4148 nr_to_write_done = wbc->nr_to_write <= 0;
4150 pagevec_release(&pvec);
4153 if (!scanned && !done) {
4155 * We hit the last page and there is more work to be done: wrap
4156 * back to the start of the file
4163 if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole))
4164 mapping->writeback_index = done_index;
4166 btrfs_add_delayed_iput(inode);
4170 int extent_write_full_page(struct page *page, struct writeback_control *wbc)
4173 struct extent_page_data epd = {
4175 .tree = &BTRFS_I(page->mapping->host)->io_tree,
4177 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4180 ret = __extent_writepage(page, wbc, &epd);
4183 end_write_bio(&epd, ret);
4187 ret = flush_write_bio(&epd);
4192 int extent_write_locked_range(struct inode *inode, u64 start, u64 end,
4196 struct address_space *mapping = inode->i_mapping;
4197 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
4199 unsigned long nr_pages = (end - start + PAGE_SIZE) >>
4202 struct extent_page_data epd = {
4206 .sync_io = mode == WB_SYNC_ALL,
4208 struct writeback_control wbc_writepages = {
4210 .nr_to_write = nr_pages * 2,
4211 .range_start = start,
4212 .range_end = end + 1,
4215 while (start <= end) {
4216 page = find_get_page(mapping, start >> PAGE_SHIFT);
4217 if (clear_page_dirty_for_io(page))
4218 ret = __extent_writepage(page, &wbc_writepages, &epd);
4220 btrfs_writepage_endio_finish_ordered(page, start,
4221 start + PAGE_SIZE - 1, 1);
4230 end_write_bio(&epd, ret);
4233 ret = flush_write_bio(&epd);
4237 int extent_writepages(struct address_space *mapping,
4238 struct writeback_control *wbc)
4241 struct extent_page_data epd = {
4243 .tree = &BTRFS_I(mapping->host)->io_tree,
4245 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4248 ret = extent_write_cache_pages(mapping, wbc, &epd);
4251 end_write_bio(&epd, ret);
4254 ret = flush_write_bio(&epd);
4258 int extent_readpages(struct address_space *mapping, struct list_head *pages,
4261 struct bio *bio = NULL;
4262 unsigned long bio_flags = 0;
4263 struct page *pagepool[16];
4264 struct extent_map *em_cached = NULL;
4265 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
4267 u64 prev_em_start = (u64)-1;
4269 while (!list_empty(pages)) {
4272 for (nr = 0; nr < ARRAY_SIZE(pagepool) && !list_empty(pages);) {
4273 struct page *page = lru_to_page(pages);
4275 prefetchw(&page->flags);
4276 list_del(&page->lru);
4277 if (add_to_page_cache_lru(page, mapping, page->index,
4278 readahead_gfp_mask(mapping))) {
4283 pagepool[nr++] = page;
4284 contig_end = page_offset(page) + PAGE_SIZE - 1;
4288 u64 contig_start = page_offset(pagepool[0]);
4290 ASSERT(contig_start + nr * PAGE_SIZE - 1 == contig_end);
4292 contiguous_readpages(tree, pagepool, nr, contig_start,
4293 contig_end, &em_cached, &bio, &bio_flags,
4299 free_extent_map(em_cached);
4302 return submit_one_bio(bio, 0, bio_flags);
4307 * basic invalidatepage code, this waits on any locked or writeback
4308 * ranges corresponding to the page, and then deletes any extent state
4309 * records from the tree
4311 int extent_invalidatepage(struct extent_io_tree *tree,
4312 struct page *page, unsigned long offset)
4314 struct extent_state *cached_state = NULL;
4315 u64 start = page_offset(page);
4316 u64 end = start + PAGE_SIZE - 1;
4317 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
4319 start += ALIGN(offset, blocksize);
4323 lock_extent_bits(tree, start, end, &cached_state);
4324 wait_on_page_writeback(page);
4325 clear_extent_bit(tree, start, end,
4326 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4327 EXTENT_DO_ACCOUNTING,
4328 1, 1, &cached_state);
4333 * a helper for releasepage, this tests for areas of the page that
4334 * are locked or under IO and drops the related state bits if it is safe
4337 static int try_release_extent_state(struct extent_io_tree *tree,
4338 struct page *page, gfp_t mask)
4340 u64 start = page_offset(page);
4341 u64 end = start + PAGE_SIZE - 1;
4344 if (test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL)) {
4348 * at this point we can safely clear everything except the
4349 * locked bit and the nodatasum bit
4351 ret = __clear_extent_bit(tree, start, end,
4352 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
4353 0, 0, NULL, mask, NULL);
4355 /* if clear_extent_bit failed for enomem reasons,
4356 * we can't allow the release to continue.
4367 * a helper for releasepage. As long as there are no locked extents
4368 * in the range corresponding to the page, both state records and extent
4369 * map records are removed
4371 int try_release_extent_mapping(struct page *page, gfp_t mask)
4373 struct extent_map *em;
4374 u64 start = page_offset(page);
4375 u64 end = start + PAGE_SIZE - 1;
4376 struct btrfs_inode *btrfs_inode = BTRFS_I(page->mapping->host);
4377 struct extent_io_tree *tree = &btrfs_inode->io_tree;
4378 struct extent_map_tree *map = &btrfs_inode->extent_tree;
4380 if (gfpflags_allow_blocking(mask) &&
4381 page->mapping->host->i_size > SZ_16M) {
4383 while (start <= end) {
4384 len = end - start + 1;
4385 write_lock(&map->lock);
4386 em = lookup_extent_mapping(map, start, len);
4388 write_unlock(&map->lock);
4391 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
4392 em->start != start) {
4393 write_unlock(&map->lock);
4394 free_extent_map(em);
4397 if (!test_range_bit(tree, em->start,
4398 extent_map_end(em) - 1,
4399 EXTENT_LOCKED, 0, NULL)) {
4400 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4401 &btrfs_inode->runtime_flags);
4402 remove_extent_mapping(map, em);
4403 /* once for the rb tree */
4404 free_extent_map(em);
4406 start = extent_map_end(em);
4407 write_unlock(&map->lock);
4410 free_extent_map(em);
4413 return try_release_extent_state(tree, page, mask);
4417 * helper function for fiemap, which doesn't want to see any holes.
4418 * This maps until we find something past 'last'
4420 static struct extent_map *get_extent_skip_holes(struct inode *inode,
4421 u64 offset, u64 last)
4423 u64 sectorsize = btrfs_inode_sectorsize(inode);
4424 struct extent_map *em;
4431 len = last - offset;
4434 len = ALIGN(len, sectorsize);
4435 em = btrfs_get_extent_fiemap(BTRFS_I(inode), offset, len);
4436 if (IS_ERR_OR_NULL(em))
4439 /* if this isn't a hole return it */
4440 if (em->block_start != EXTENT_MAP_HOLE)
4443 /* this is a hole, advance to the next extent */
4444 offset = extent_map_end(em);
4445 free_extent_map(em);
4453 * To cache previous fiemap extent
4455 * Will be used for merging fiemap extent
4457 struct fiemap_cache {
4466 * Helper to submit fiemap extent.
4468 * Will try to merge current fiemap extent specified by @offset, @phys,
4469 * @len and @flags with cached one.
4470 * And only when we fails to merge, cached one will be submitted as
4473 * Return value is the same as fiemap_fill_next_extent().
4475 static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
4476 struct fiemap_cache *cache,
4477 u64 offset, u64 phys, u64 len, u32 flags)
4485 * Sanity check, extent_fiemap() should have ensured that new
4486 * fiemap extent won't overlap with cached one.
4489 * NOTE: Physical address can overlap, due to compression
4491 if (cache->offset + cache->len > offset) {
4497 * Only merges fiemap extents if
4498 * 1) Their logical addresses are continuous
4500 * 2) Their physical addresses are continuous
4501 * So truly compressed (physical size smaller than logical size)
4502 * extents won't get merged with each other
4504 * 3) Share same flags except FIEMAP_EXTENT_LAST
4505 * So regular extent won't get merged with prealloc extent
4507 if (cache->offset + cache->len == offset &&
4508 cache->phys + cache->len == phys &&
4509 (cache->flags & ~FIEMAP_EXTENT_LAST) ==
4510 (flags & ~FIEMAP_EXTENT_LAST)) {
4512 cache->flags |= flags;
4513 goto try_submit_last;
4516 /* Not mergeable, need to submit cached one */
4517 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
4518 cache->len, cache->flags);
4519 cache->cached = false;
4523 cache->cached = true;
4524 cache->offset = offset;
4527 cache->flags = flags;
4529 if (cache->flags & FIEMAP_EXTENT_LAST) {
4530 ret = fiemap_fill_next_extent(fieinfo, cache->offset,
4531 cache->phys, cache->len, cache->flags);
4532 cache->cached = false;
4538 * Emit last fiemap cache
4540 * The last fiemap cache may still be cached in the following case:
4542 * |<- Fiemap range ->|
4543 * |<------------ First extent ----------->|
4545 * In this case, the first extent range will be cached but not emitted.
4546 * So we must emit it before ending extent_fiemap().
4548 static int emit_last_fiemap_cache(struct fiemap_extent_info *fieinfo,
4549 struct fiemap_cache *cache)
4556 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
4557 cache->len, cache->flags);
4558 cache->cached = false;
4564 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4565 __u64 start, __u64 len)
4569 u64 max = start + len;
4573 u64 last_for_get_extent = 0;
4575 u64 isize = i_size_read(inode);
4576 struct btrfs_key found_key;
4577 struct extent_map *em = NULL;
4578 struct extent_state *cached_state = NULL;
4579 struct btrfs_path *path;
4580 struct btrfs_root *root = BTRFS_I(inode)->root;
4581 struct fiemap_cache cache = { 0 };
4582 struct ulist *roots;
4583 struct ulist *tmp_ulist;
4592 path = btrfs_alloc_path();
4595 path->leave_spinning = 1;
4597 roots = ulist_alloc(GFP_KERNEL);
4598 tmp_ulist = ulist_alloc(GFP_KERNEL);
4599 if (!roots || !tmp_ulist) {
4601 goto out_free_ulist;
4604 start = round_down(start, btrfs_inode_sectorsize(inode));
4605 len = round_up(max, btrfs_inode_sectorsize(inode)) - start;
4608 * lookup the last file extent. We're not using i_size here
4609 * because there might be preallocation past i_size
4611 ret = btrfs_lookup_file_extent(NULL, root, path,
4612 btrfs_ino(BTRFS_I(inode)), -1, 0);
4614 goto out_free_ulist;
4622 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
4623 found_type = found_key.type;
4625 /* No extents, but there might be delalloc bits */
4626 if (found_key.objectid != btrfs_ino(BTRFS_I(inode)) ||
4627 found_type != BTRFS_EXTENT_DATA_KEY) {
4628 /* have to trust i_size as the end */
4630 last_for_get_extent = isize;
4633 * remember the start of the last extent. There are a
4634 * bunch of different factors that go into the length of the
4635 * extent, so its much less complex to remember where it started
4637 last = found_key.offset;
4638 last_for_get_extent = last + 1;
4640 btrfs_release_path(path);
4643 * we might have some extents allocated but more delalloc past those
4644 * extents. so, we trust isize unless the start of the last extent is
4649 last_for_get_extent = isize;
4652 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4655 em = get_extent_skip_holes(inode, start, last_for_get_extent);
4664 u64 offset_in_extent = 0;
4666 /* break if the extent we found is outside the range */
4667 if (em->start >= max || extent_map_end(em) < off)
4671 * get_extent may return an extent that starts before our
4672 * requested range. We have to make sure the ranges
4673 * we return to fiemap always move forward and don't
4674 * overlap, so adjust the offsets here
4676 em_start = max(em->start, off);
4679 * record the offset from the start of the extent
4680 * for adjusting the disk offset below. Only do this if the
4681 * extent isn't compressed since our in ram offset may be past
4682 * what we have actually allocated on disk.
4684 if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4685 offset_in_extent = em_start - em->start;
4686 em_end = extent_map_end(em);
4687 em_len = em_end - em_start;
4689 if (em->block_start < EXTENT_MAP_LAST_BYTE)
4690 disko = em->block_start + offset_in_extent;
4695 * bump off for our next call to get_extent
4697 off = extent_map_end(em);
4701 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
4703 flags |= FIEMAP_EXTENT_LAST;
4704 } else if (em->block_start == EXTENT_MAP_INLINE) {
4705 flags |= (FIEMAP_EXTENT_DATA_INLINE |
4706 FIEMAP_EXTENT_NOT_ALIGNED);
4707 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
4708 flags |= (FIEMAP_EXTENT_DELALLOC |
4709 FIEMAP_EXTENT_UNKNOWN);
4710 } else if (fieinfo->fi_extents_max) {
4711 u64 bytenr = em->block_start -
4712 (em->start - em->orig_start);
4715 * As btrfs supports shared space, this information
4716 * can be exported to userspace tools via
4717 * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0
4718 * then we're just getting a count and we can skip the
4721 ret = btrfs_check_shared(root,
4722 btrfs_ino(BTRFS_I(inode)),
4723 bytenr, roots, tmp_ulist);
4727 flags |= FIEMAP_EXTENT_SHARED;
4730 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4731 flags |= FIEMAP_EXTENT_ENCODED;
4732 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4733 flags |= FIEMAP_EXTENT_UNWRITTEN;
4735 free_extent_map(em);
4737 if ((em_start >= last) || em_len == (u64)-1 ||
4738 (last == (u64)-1 && isize <= em_end)) {
4739 flags |= FIEMAP_EXTENT_LAST;
4743 /* now scan forward to see if this is really the last extent. */
4744 em = get_extent_skip_holes(inode, off, last_for_get_extent);
4750 flags |= FIEMAP_EXTENT_LAST;
4753 ret = emit_fiemap_extent(fieinfo, &cache, em_start, disko,
4763 ret = emit_last_fiemap_cache(fieinfo, &cache);
4764 free_extent_map(em);
4766 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4770 btrfs_free_path(path);
4772 ulist_free(tmp_ulist);
4776 static void __free_extent_buffer(struct extent_buffer *eb)
4778 btrfs_leak_debug_del(&eb->leak_list);
4779 kmem_cache_free(extent_buffer_cache, eb);
4782 int extent_buffer_under_io(struct extent_buffer *eb)
4784 return (atomic_read(&eb->io_pages) ||
4785 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4786 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4790 * Release all pages attached to the extent buffer.
4792 static void btrfs_release_extent_buffer_pages(struct extent_buffer *eb)
4796 int mapped = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
4798 BUG_ON(extent_buffer_under_io(eb));
4800 num_pages = num_extent_pages(eb);
4801 for (i = 0; i < num_pages; i++) {
4802 struct page *page = eb->pages[i];
4807 spin_lock(&page->mapping->private_lock);
4809 * We do this since we'll remove the pages after we've
4810 * removed the eb from the radix tree, so we could race
4811 * and have this page now attached to the new eb. So
4812 * only clear page_private if it's still connected to
4815 if (PagePrivate(page) &&
4816 page->private == (unsigned long)eb) {
4817 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4818 BUG_ON(PageDirty(page));
4819 BUG_ON(PageWriteback(page));
4821 * We need to make sure we haven't be attached
4824 ClearPagePrivate(page);
4825 set_page_private(page, 0);
4826 /* One for the page private */
4831 spin_unlock(&page->mapping->private_lock);
4833 /* One for when we allocated the page */
4839 * Helper for releasing the extent buffer.
4841 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4843 btrfs_release_extent_buffer_pages(eb);
4844 __free_extent_buffer(eb);
4847 static struct extent_buffer *
4848 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
4851 struct extent_buffer *eb = NULL;
4853 eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL);
4856 eb->fs_info = fs_info;
4858 rwlock_init(&eb->lock);
4859 atomic_set(&eb->blocking_readers, 0);
4860 eb->blocking_writers = 0;
4861 eb->lock_nested = false;
4862 init_waitqueue_head(&eb->write_lock_wq);
4863 init_waitqueue_head(&eb->read_lock_wq);
4865 btrfs_leak_debug_add(&eb->leak_list, &buffers);
4867 spin_lock_init(&eb->refs_lock);
4868 atomic_set(&eb->refs, 1);
4869 atomic_set(&eb->io_pages, 0);
4872 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4874 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4875 > MAX_INLINE_EXTENT_BUFFER_SIZE);
4876 BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE);
4878 #ifdef CONFIG_BTRFS_DEBUG
4879 eb->spinning_writers = 0;
4880 atomic_set(&eb->spinning_readers, 0);
4881 atomic_set(&eb->read_locks, 0);
4882 eb->write_locks = 0;
4888 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src)
4892 struct extent_buffer *new;
4893 int num_pages = num_extent_pages(src);
4895 new = __alloc_extent_buffer(src->fs_info, src->start, src->len);
4899 for (i = 0; i < num_pages; i++) {
4900 p = alloc_page(GFP_NOFS);
4902 btrfs_release_extent_buffer(new);
4905 attach_extent_buffer_page(new, p);
4906 WARN_ON(PageDirty(p));
4909 copy_page(page_address(p), page_address(src->pages[i]));
4912 set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
4913 set_bit(EXTENT_BUFFER_UNMAPPED, &new->bflags);
4918 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4919 u64 start, unsigned long len)
4921 struct extent_buffer *eb;
4925 eb = __alloc_extent_buffer(fs_info, start, len);
4929 num_pages = num_extent_pages(eb);
4930 for (i = 0; i < num_pages; i++) {
4931 eb->pages[i] = alloc_page(GFP_NOFS);
4935 set_extent_buffer_uptodate(eb);
4936 btrfs_set_header_nritems(eb, 0);
4937 set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
4942 __free_page(eb->pages[i - 1]);
4943 __free_extent_buffer(eb);
4947 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4950 return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize);
4953 static void check_buffer_tree_ref(struct extent_buffer *eb)
4956 /* the ref bit is tricky. We have to make sure it is set
4957 * if we have the buffer dirty. Otherwise the
4958 * code to free a buffer can end up dropping a dirty
4961 * Once the ref bit is set, it won't go away while the
4962 * buffer is dirty or in writeback, and it also won't
4963 * go away while we have the reference count on the
4966 * We can't just set the ref bit without bumping the
4967 * ref on the eb because free_extent_buffer might
4968 * see the ref bit and try to clear it. If this happens
4969 * free_extent_buffer might end up dropping our original
4970 * ref by mistake and freeing the page before we are able
4971 * to add one more ref.
4973 * So bump the ref count first, then set the bit. If someone
4974 * beat us to it, drop the ref we added.
4976 refs = atomic_read(&eb->refs);
4977 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4980 spin_lock(&eb->refs_lock);
4981 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4982 atomic_inc(&eb->refs);
4983 spin_unlock(&eb->refs_lock);
4986 static void mark_extent_buffer_accessed(struct extent_buffer *eb,
4987 struct page *accessed)
4991 check_buffer_tree_ref(eb);
4993 num_pages = num_extent_pages(eb);
4994 for (i = 0; i < num_pages; i++) {
4995 struct page *p = eb->pages[i];
4998 mark_page_accessed(p);
5002 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
5005 struct extent_buffer *eb;
5008 eb = radix_tree_lookup(&fs_info->buffer_radix,
5009 start >> PAGE_SHIFT);
5010 if (eb && atomic_inc_not_zero(&eb->refs)) {
5013 * Lock our eb's refs_lock to avoid races with
5014 * free_extent_buffer. When we get our eb it might be flagged
5015 * with EXTENT_BUFFER_STALE and another task running
5016 * free_extent_buffer might have seen that flag set,
5017 * eb->refs == 2, that the buffer isn't under IO (dirty and
5018 * writeback flags not set) and it's still in the tree (flag
5019 * EXTENT_BUFFER_TREE_REF set), therefore being in the process
5020 * of decrementing the extent buffer's reference count twice.
5021 * So here we could race and increment the eb's reference count,
5022 * clear its stale flag, mark it as dirty and drop our reference
5023 * before the other task finishes executing free_extent_buffer,
5024 * which would later result in an attempt to free an extent
5025 * buffer that is dirty.
5027 if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) {
5028 spin_lock(&eb->refs_lock);
5029 spin_unlock(&eb->refs_lock);
5031 mark_extent_buffer_accessed(eb, NULL);
5039 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5040 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
5043 struct extent_buffer *eb, *exists = NULL;
5046 eb = find_extent_buffer(fs_info, start);
5049 eb = alloc_dummy_extent_buffer(fs_info, start);
5052 eb->fs_info = fs_info;
5054 ret = radix_tree_preload(GFP_NOFS);
5057 spin_lock(&fs_info->buffer_lock);
5058 ret = radix_tree_insert(&fs_info->buffer_radix,
5059 start >> PAGE_SHIFT, eb);
5060 spin_unlock(&fs_info->buffer_lock);
5061 radix_tree_preload_end();
5062 if (ret == -EEXIST) {
5063 exists = find_extent_buffer(fs_info, start);
5069 check_buffer_tree_ref(eb);
5070 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
5074 btrfs_release_extent_buffer(eb);
5079 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
5082 unsigned long len = fs_info->nodesize;
5085 unsigned long index = start >> PAGE_SHIFT;
5086 struct extent_buffer *eb;
5087 struct extent_buffer *exists = NULL;
5089 struct address_space *mapping = fs_info->btree_inode->i_mapping;
5093 if (!IS_ALIGNED(start, fs_info->sectorsize)) {
5094 btrfs_err(fs_info, "bad tree block start %llu", start);
5095 return ERR_PTR(-EINVAL);
5098 eb = find_extent_buffer(fs_info, start);
5102 eb = __alloc_extent_buffer(fs_info, start, len);
5104 return ERR_PTR(-ENOMEM);
5106 num_pages = num_extent_pages(eb);
5107 for (i = 0; i < num_pages; i++, index++) {
5108 p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL);
5110 exists = ERR_PTR(-ENOMEM);
5114 spin_lock(&mapping->private_lock);
5115 if (PagePrivate(p)) {
5117 * We could have already allocated an eb for this page
5118 * and attached one so lets see if we can get a ref on
5119 * the existing eb, and if we can we know it's good and
5120 * we can just return that one, else we know we can just
5121 * overwrite page->private.
5123 exists = (struct extent_buffer *)p->private;
5124 if (atomic_inc_not_zero(&exists->refs)) {
5125 spin_unlock(&mapping->private_lock);
5128 mark_extent_buffer_accessed(exists, p);
5134 * Do this so attach doesn't complain and we need to
5135 * drop the ref the old guy had.
5137 ClearPagePrivate(p);
5138 WARN_ON(PageDirty(p));
5141 attach_extent_buffer_page(eb, p);
5142 spin_unlock(&mapping->private_lock);
5143 WARN_ON(PageDirty(p));
5145 if (!PageUptodate(p))
5149 * We can't unlock the pages just yet since the extent buffer
5150 * hasn't been properly inserted in the radix tree, this
5151 * opens a race with btree_releasepage which can free a page
5152 * while we are still filling in all pages for the buffer and
5157 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5159 ret = radix_tree_preload(GFP_NOFS);
5161 exists = ERR_PTR(ret);
5165 spin_lock(&fs_info->buffer_lock);
5166 ret = radix_tree_insert(&fs_info->buffer_radix,
5167 start >> PAGE_SHIFT, eb);
5168 spin_unlock(&fs_info->buffer_lock);
5169 radix_tree_preload_end();
5170 if (ret == -EEXIST) {
5171 exists = find_extent_buffer(fs_info, start);
5177 /* add one reference for the tree */
5178 check_buffer_tree_ref(eb);
5179 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
5182 * Now it's safe to unlock the pages because any calls to
5183 * btree_releasepage will correctly detect that a page belongs to a
5184 * live buffer and won't free them prematurely.
5186 for (i = 0; i < num_pages; i++)
5187 unlock_page(eb->pages[i]);
5191 WARN_ON(!atomic_dec_and_test(&eb->refs));
5192 for (i = 0; i < num_pages; i++) {
5194 unlock_page(eb->pages[i]);
5197 btrfs_release_extent_buffer(eb);
5201 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
5203 struct extent_buffer *eb =
5204 container_of(head, struct extent_buffer, rcu_head);
5206 __free_extent_buffer(eb);
5209 static int release_extent_buffer(struct extent_buffer *eb)
5211 lockdep_assert_held(&eb->refs_lock);
5213 WARN_ON(atomic_read(&eb->refs) == 0);
5214 if (atomic_dec_and_test(&eb->refs)) {
5215 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
5216 struct btrfs_fs_info *fs_info = eb->fs_info;
5218 spin_unlock(&eb->refs_lock);
5220 spin_lock(&fs_info->buffer_lock);
5221 radix_tree_delete(&fs_info->buffer_radix,
5222 eb->start >> PAGE_SHIFT);
5223 spin_unlock(&fs_info->buffer_lock);
5225 spin_unlock(&eb->refs_lock);
5228 /* Should be safe to release our pages at this point */
5229 btrfs_release_extent_buffer_pages(eb);
5230 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5231 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags))) {
5232 __free_extent_buffer(eb);
5236 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
5239 spin_unlock(&eb->refs_lock);
5244 void free_extent_buffer(struct extent_buffer *eb)
5252 refs = atomic_read(&eb->refs);
5253 if ((!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && refs <= 3)
5254 || (test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) &&
5257 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
5262 spin_lock(&eb->refs_lock);
5263 if (atomic_read(&eb->refs) == 2 &&
5264 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
5265 !extent_buffer_under_io(eb) &&
5266 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5267 atomic_dec(&eb->refs);
5270 * I know this is terrible, but it's temporary until we stop tracking
5271 * the uptodate bits and such for the extent buffers.
5273 release_extent_buffer(eb);
5276 void free_extent_buffer_stale(struct extent_buffer *eb)
5281 spin_lock(&eb->refs_lock);
5282 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
5284 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
5285 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5286 atomic_dec(&eb->refs);
5287 release_extent_buffer(eb);
5290 void clear_extent_buffer_dirty(struct extent_buffer *eb)
5296 num_pages = num_extent_pages(eb);
5298 for (i = 0; i < num_pages; i++) {
5299 page = eb->pages[i];
5300 if (!PageDirty(page))
5304 WARN_ON(!PagePrivate(page));
5306 clear_page_dirty_for_io(page);
5307 xa_lock_irq(&page->mapping->i_pages);
5308 if (!PageDirty(page))
5309 __xa_clear_mark(&page->mapping->i_pages,
5310 page_index(page), PAGECACHE_TAG_DIRTY);
5311 xa_unlock_irq(&page->mapping->i_pages);
5312 ClearPageError(page);
5315 WARN_ON(atomic_read(&eb->refs) == 0);
5318 bool set_extent_buffer_dirty(struct extent_buffer *eb)
5324 check_buffer_tree_ref(eb);
5326 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
5328 num_pages = num_extent_pages(eb);
5329 WARN_ON(atomic_read(&eb->refs) == 0);
5330 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
5333 for (i = 0; i < num_pages; i++)
5334 set_page_dirty(eb->pages[i]);
5336 #ifdef CONFIG_BTRFS_DEBUG
5337 for (i = 0; i < num_pages; i++)
5338 ASSERT(PageDirty(eb->pages[i]));
5344 void clear_extent_buffer_uptodate(struct extent_buffer *eb)
5350 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5351 num_pages = num_extent_pages(eb);
5352 for (i = 0; i < num_pages; i++) {
5353 page = eb->pages[i];
5355 ClearPageUptodate(page);
5359 void set_extent_buffer_uptodate(struct extent_buffer *eb)
5365 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5366 num_pages = num_extent_pages(eb);
5367 for (i = 0; i < num_pages; i++) {
5368 page = eb->pages[i];
5369 SetPageUptodate(page);
5373 int read_extent_buffer_pages(struct extent_buffer *eb, int wait, int mirror_num)
5379 int locked_pages = 0;
5380 int all_uptodate = 1;
5382 unsigned long num_reads = 0;
5383 struct bio *bio = NULL;
5384 unsigned long bio_flags = 0;
5385 struct extent_io_tree *tree = &BTRFS_I(eb->fs_info->btree_inode)->io_tree;
5387 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
5390 num_pages = num_extent_pages(eb);
5391 for (i = 0; i < num_pages; i++) {
5392 page = eb->pages[i];
5393 if (wait == WAIT_NONE) {
5394 if (!trylock_page(page))
5402 * We need to firstly lock all pages to make sure that
5403 * the uptodate bit of our pages won't be affected by
5404 * clear_extent_buffer_uptodate().
5406 for (i = 0; i < num_pages; i++) {
5407 page = eb->pages[i];
5408 if (!PageUptodate(page)) {
5415 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5419 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
5420 eb->read_mirror = 0;
5421 atomic_set(&eb->io_pages, num_reads);
5422 for (i = 0; i < num_pages; i++) {
5423 page = eb->pages[i];
5425 if (!PageUptodate(page)) {
5427 atomic_dec(&eb->io_pages);
5432 ClearPageError(page);
5433 err = __extent_read_full_page(tree, page,
5434 btree_get_extent, &bio,
5435 mirror_num, &bio_flags,
5440 * We use &bio in above __extent_read_full_page,
5441 * so we ensure that if it returns error, the
5442 * current page fails to add itself to bio and
5443 * it's been unlocked.
5445 * We must dec io_pages by ourselves.
5447 atomic_dec(&eb->io_pages);
5455 err = submit_one_bio(bio, mirror_num, bio_flags);
5460 if (ret || wait != WAIT_COMPLETE)
5463 for (i = 0; i < num_pages; i++) {
5464 page = eb->pages[i];
5465 wait_on_page_locked(page);
5466 if (!PageUptodate(page))
5473 while (locked_pages > 0) {
5475 page = eb->pages[locked_pages];
5481 void read_extent_buffer(const struct extent_buffer *eb, void *dstv,
5482 unsigned long start, unsigned long len)
5488 char *dst = (char *)dstv;
5489 size_t start_offset = offset_in_page(eb->start);
5490 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5492 if (start + len > eb->len) {
5493 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n",
5494 eb->start, eb->len, start, len);
5495 memset(dst, 0, len);
5499 offset = offset_in_page(start_offset + start);
5502 page = eb->pages[i];
5504 cur = min(len, (PAGE_SIZE - offset));
5505 kaddr = page_address(page);
5506 memcpy(dst, kaddr + offset, cur);
5515 int read_extent_buffer_to_user(const struct extent_buffer *eb,
5517 unsigned long start, unsigned long len)
5523 char __user *dst = (char __user *)dstv;
5524 size_t start_offset = offset_in_page(eb->start);
5525 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5528 WARN_ON(start > eb->len);
5529 WARN_ON(start + len > eb->start + eb->len);
5531 offset = offset_in_page(start_offset + start);
5534 page = eb->pages[i];
5536 cur = min(len, (PAGE_SIZE - offset));
5537 kaddr = page_address(page);
5538 if (copy_to_user(dst, kaddr + offset, cur)) {
5553 * return 0 if the item is found within a page.
5554 * return 1 if the item spans two pages.
5555 * return -EINVAL otherwise.
5557 int map_private_extent_buffer(const struct extent_buffer *eb,
5558 unsigned long start, unsigned long min_len,
5559 char **map, unsigned long *map_start,
5560 unsigned long *map_len)
5565 size_t start_offset = offset_in_page(eb->start);
5566 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5567 unsigned long end_i = (start_offset + start + min_len - 1) >>
5570 if (start + min_len > eb->len) {
5571 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n",
5572 eb->start, eb->len, start, min_len);
5580 offset = start_offset;
5584 *map_start = ((u64)i << PAGE_SHIFT) - start_offset;
5588 kaddr = page_address(p);
5589 *map = kaddr + offset;
5590 *map_len = PAGE_SIZE - offset;
5594 int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,
5595 unsigned long start, unsigned long len)
5601 char *ptr = (char *)ptrv;
5602 size_t start_offset = offset_in_page(eb->start);
5603 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5606 WARN_ON(start > eb->len);
5607 WARN_ON(start + len > eb->start + eb->len);
5609 offset = offset_in_page(start_offset + start);
5612 page = eb->pages[i];
5614 cur = min(len, (PAGE_SIZE - offset));
5616 kaddr = page_address(page);
5617 ret = memcmp(ptr, kaddr + offset, cur);
5629 void write_extent_buffer_chunk_tree_uuid(struct extent_buffer *eb,
5634 WARN_ON(!PageUptodate(eb->pages[0]));
5635 kaddr = page_address(eb->pages[0]);
5636 memcpy(kaddr + offsetof(struct btrfs_header, chunk_tree_uuid), srcv,
5640 void write_extent_buffer_fsid(struct extent_buffer *eb, const void *srcv)
5644 WARN_ON(!PageUptodate(eb->pages[0]));
5645 kaddr = page_address(eb->pages[0]);
5646 memcpy(kaddr + offsetof(struct btrfs_header, fsid), srcv,
5650 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
5651 unsigned long start, unsigned long len)
5657 char *src = (char *)srcv;
5658 size_t start_offset = offset_in_page(eb->start);
5659 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5661 WARN_ON(start > eb->len);
5662 WARN_ON(start + len > eb->start + eb->len);
5664 offset = offset_in_page(start_offset + start);
5667 page = eb->pages[i];
5668 WARN_ON(!PageUptodate(page));
5670 cur = min(len, PAGE_SIZE - offset);
5671 kaddr = page_address(page);
5672 memcpy(kaddr + offset, src, cur);
5681 void memzero_extent_buffer(struct extent_buffer *eb, unsigned long start,
5688 size_t start_offset = offset_in_page(eb->start);
5689 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5691 WARN_ON(start > eb->len);
5692 WARN_ON(start + len > eb->start + eb->len);
5694 offset = offset_in_page(start_offset + start);
5697 page = eb->pages[i];
5698 WARN_ON(!PageUptodate(page));
5700 cur = min(len, PAGE_SIZE - offset);
5701 kaddr = page_address(page);
5702 memset(kaddr + offset, 0, cur);
5710 void copy_extent_buffer_full(struct extent_buffer *dst,
5711 struct extent_buffer *src)
5716 ASSERT(dst->len == src->len);
5718 num_pages = num_extent_pages(dst);
5719 for (i = 0; i < num_pages; i++)
5720 copy_page(page_address(dst->pages[i]),
5721 page_address(src->pages[i]));
5724 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
5725 unsigned long dst_offset, unsigned long src_offset,
5728 u64 dst_len = dst->len;
5733 size_t start_offset = offset_in_page(dst->start);
5734 unsigned long i = (start_offset + dst_offset) >> PAGE_SHIFT;
5736 WARN_ON(src->len != dst_len);
5738 offset = offset_in_page(start_offset + dst_offset);
5741 page = dst->pages[i];
5742 WARN_ON(!PageUptodate(page));
5744 cur = min(len, (unsigned long)(PAGE_SIZE - offset));
5746 kaddr = page_address(page);
5747 read_extent_buffer(src, kaddr + offset, src_offset, cur);
5757 * eb_bitmap_offset() - calculate the page and offset of the byte containing the
5759 * @eb: the extent buffer
5760 * @start: offset of the bitmap item in the extent buffer
5762 * @page_index: return index of the page in the extent buffer that contains the
5764 * @page_offset: return offset into the page given by page_index
5766 * This helper hides the ugliness of finding the byte in an extent buffer which
5767 * contains a given bit.
5769 static inline void eb_bitmap_offset(struct extent_buffer *eb,
5770 unsigned long start, unsigned long nr,
5771 unsigned long *page_index,
5772 size_t *page_offset)
5774 size_t start_offset = offset_in_page(eb->start);
5775 size_t byte_offset = BIT_BYTE(nr);
5779 * The byte we want is the offset of the extent buffer + the offset of
5780 * the bitmap item in the extent buffer + the offset of the byte in the
5783 offset = start_offset + start + byte_offset;
5785 *page_index = offset >> PAGE_SHIFT;
5786 *page_offset = offset_in_page(offset);
5790 * extent_buffer_test_bit - determine whether a bit in a bitmap item is set
5791 * @eb: the extent buffer
5792 * @start: offset of the bitmap item in the extent buffer
5793 * @nr: bit number to test
5795 int extent_buffer_test_bit(struct extent_buffer *eb, unsigned long start,
5803 eb_bitmap_offset(eb, start, nr, &i, &offset);
5804 page = eb->pages[i];
5805 WARN_ON(!PageUptodate(page));
5806 kaddr = page_address(page);
5807 return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1)));
5811 * extent_buffer_bitmap_set - set an area of a bitmap
5812 * @eb: the extent buffer
5813 * @start: offset of the bitmap item in the extent buffer
5814 * @pos: bit number of the first bit
5815 * @len: number of bits to set
5817 void extent_buffer_bitmap_set(struct extent_buffer *eb, unsigned long start,
5818 unsigned long pos, unsigned long len)
5824 const unsigned int size = pos + len;
5825 int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5826 u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos);
5828 eb_bitmap_offset(eb, start, pos, &i, &offset);
5829 page = eb->pages[i];
5830 WARN_ON(!PageUptodate(page));
5831 kaddr = page_address(page);
5833 while (len >= bits_to_set) {
5834 kaddr[offset] |= mask_to_set;
5836 bits_to_set = BITS_PER_BYTE;
5838 if (++offset >= PAGE_SIZE && len > 0) {
5840 page = eb->pages[++i];
5841 WARN_ON(!PageUptodate(page));
5842 kaddr = page_address(page);
5846 mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
5847 kaddr[offset] |= mask_to_set;
5853 * extent_buffer_bitmap_clear - clear an area of a bitmap
5854 * @eb: the extent buffer
5855 * @start: offset of the bitmap item in the extent buffer
5856 * @pos: bit number of the first bit
5857 * @len: number of bits to clear
5859 void extent_buffer_bitmap_clear(struct extent_buffer *eb, unsigned long start,
5860 unsigned long pos, unsigned long len)
5866 const unsigned int size = pos + len;
5867 int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5868 u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos);
5870 eb_bitmap_offset(eb, start, pos, &i, &offset);
5871 page = eb->pages[i];
5872 WARN_ON(!PageUptodate(page));
5873 kaddr = page_address(page);
5875 while (len >= bits_to_clear) {
5876 kaddr[offset] &= ~mask_to_clear;
5877 len -= bits_to_clear;
5878 bits_to_clear = BITS_PER_BYTE;
5880 if (++offset >= PAGE_SIZE && len > 0) {
5882 page = eb->pages[++i];
5883 WARN_ON(!PageUptodate(page));
5884 kaddr = page_address(page);
5888 mask_to_clear &= BITMAP_LAST_BYTE_MASK(size);
5889 kaddr[offset] &= ~mask_to_clear;
5893 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
5895 unsigned long distance = (src > dst) ? src - dst : dst - src;
5896 return distance < len;
5899 static void copy_pages(struct page *dst_page, struct page *src_page,
5900 unsigned long dst_off, unsigned long src_off,
5903 char *dst_kaddr = page_address(dst_page);
5905 int must_memmove = 0;
5907 if (dst_page != src_page) {
5908 src_kaddr = page_address(src_page);
5910 src_kaddr = dst_kaddr;
5911 if (areas_overlap(src_off, dst_off, len))
5916 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
5918 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
5921 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5922 unsigned long src_offset, unsigned long len)
5924 struct btrfs_fs_info *fs_info = dst->fs_info;
5926 size_t dst_off_in_page;
5927 size_t src_off_in_page;
5928 size_t start_offset = offset_in_page(dst->start);
5929 unsigned long dst_i;
5930 unsigned long src_i;
5932 if (src_offset + len > dst->len) {
5934 "memmove bogus src_offset %lu move len %lu dst len %lu",
5935 src_offset, len, dst->len);
5938 if (dst_offset + len > dst->len) {
5940 "memmove bogus dst_offset %lu move len %lu dst len %lu",
5941 dst_offset, len, dst->len);
5946 dst_off_in_page = offset_in_page(start_offset + dst_offset);
5947 src_off_in_page = offset_in_page(start_offset + src_offset);
5949 dst_i = (start_offset + dst_offset) >> PAGE_SHIFT;
5950 src_i = (start_offset + src_offset) >> PAGE_SHIFT;
5952 cur = min(len, (unsigned long)(PAGE_SIZE -
5954 cur = min_t(unsigned long, cur,
5955 (unsigned long)(PAGE_SIZE - dst_off_in_page));
5957 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5958 dst_off_in_page, src_off_in_page, cur);
5966 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5967 unsigned long src_offset, unsigned long len)
5969 struct btrfs_fs_info *fs_info = dst->fs_info;
5971 size_t dst_off_in_page;
5972 size_t src_off_in_page;
5973 unsigned long dst_end = dst_offset + len - 1;
5974 unsigned long src_end = src_offset + len - 1;
5975 size_t start_offset = offset_in_page(dst->start);
5976 unsigned long dst_i;
5977 unsigned long src_i;
5979 if (src_offset + len > dst->len) {
5981 "memmove bogus src_offset %lu move len %lu len %lu",
5982 src_offset, len, dst->len);
5985 if (dst_offset + len > dst->len) {
5987 "memmove bogus dst_offset %lu move len %lu len %lu",
5988 dst_offset, len, dst->len);
5991 if (dst_offset < src_offset) {
5992 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
5996 dst_i = (start_offset + dst_end) >> PAGE_SHIFT;
5997 src_i = (start_offset + src_end) >> PAGE_SHIFT;
5999 dst_off_in_page = offset_in_page(start_offset + dst_end);
6000 src_off_in_page = offset_in_page(start_offset + src_end);
6002 cur = min_t(unsigned long, len, src_off_in_page + 1);
6003 cur = min(cur, dst_off_in_page + 1);
6004 copy_pages(dst->pages[dst_i], dst->pages[src_i],
6005 dst_off_in_page - cur + 1,
6006 src_off_in_page - cur + 1, cur);
6014 int try_release_extent_buffer(struct page *page)
6016 struct extent_buffer *eb;
6019 * We need to make sure nobody is attaching this page to an eb right
6022 spin_lock(&page->mapping->private_lock);
6023 if (!PagePrivate(page)) {
6024 spin_unlock(&page->mapping->private_lock);
6028 eb = (struct extent_buffer *)page->private;
6032 * This is a little awful but should be ok, we need to make sure that
6033 * the eb doesn't disappear out from under us while we're looking at
6036 spin_lock(&eb->refs_lock);
6037 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
6038 spin_unlock(&eb->refs_lock);
6039 spin_unlock(&page->mapping->private_lock);
6042 spin_unlock(&page->mapping->private_lock);
6045 * If tree ref isn't set then we know the ref on this eb is a real ref,
6046 * so just return, this page will likely be freed soon anyway.
6048 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
6049 spin_unlock(&eb->refs_lock);
6053 return release_extent_buffer(eb);