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);
362 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
363 struct rb_node **next_ret,
364 struct rb_node **prev_ret,
365 struct rb_node ***p_ret,
366 struct rb_node **parent_ret)
368 struct rb_root *root = &tree->state;
369 struct rb_node **n = &root->rb_node;
370 struct rb_node *prev = NULL;
371 struct rb_node *orig_prev = NULL;
372 struct tree_entry *entry;
373 struct tree_entry *prev_entry = NULL;
377 entry = rb_entry(prev, struct tree_entry, rb_node);
380 if (offset < entry->start)
382 else if (offset > entry->end)
395 while (prev && offset > prev_entry->end) {
396 prev = rb_next(prev);
397 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
404 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
405 while (prev && offset < prev_entry->start) {
406 prev = rb_prev(prev);
407 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
414 static inline struct rb_node *
415 tree_search_for_insert(struct extent_io_tree *tree,
417 struct rb_node ***p_ret,
418 struct rb_node **parent_ret)
420 struct rb_node *next= NULL;
423 ret = __etree_search(tree, offset, &next, NULL, p_ret, parent_ret);
429 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
432 return tree_search_for_insert(tree, offset, NULL, NULL);
436 * utility function to look for merge candidates inside a given range.
437 * Any extents with matching state are merged together into a single
438 * extent in the tree. Extents with EXTENT_IO in their state field
439 * are not merged because the end_io handlers need to be able to do
440 * operations on them without sleeping (or doing allocations/splits).
442 * This should be called with the tree lock held.
444 static void merge_state(struct extent_io_tree *tree,
445 struct extent_state *state)
447 struct extent_state *other;
448 struct rb_node *other_node;
450 if (state->state & (EXTENT_LOCKED | EXTENT_BOUNDARY))
453 other_node = rb_prev(&state->rb_node);
455 other = rb_entry(other_node, struct extent_state, rb_node);
456 if (other->end == state->start - 1 &&
457 other->state == state->state) {
458 if (tree->private_data &&
459 is_data_inode(tree->private_data))
460 btrfs_merge_delalloc_extent(tree->private_data,
462 state->start = other->start;
463 rb_erase(&other->rb_node, &tree->state);
464 RB_CLEAR_NODE(&other->rb_node);
465 free_extent_state(other);
468 other_node = rb_next(&state->rb_node);
470 other = rb_entry(other_node, struct extent_state, rb_node);
471 if (other->start == state->end + 1 &&
472 other->state == state->state) {
473 if (tree->private_data &&
474 is_data_inode(tree->private_data))
475 btrfs_merge_delalloc_extent(tree->private_data,
477 state->end = other->end;
478 rb_erase(&other->rb_node, &tree->state);
479 RB_CLEAR_NODE(&other->rb_node);
480 free_extent_state(other);
485 static void set_state_bits(struct extent_io_tree *tree,
486 struct extent_state *state, unsigned *bits,
487 struct extent_changeset *changeset);
490 * insert an extent_state struct into the tree. 'bits' are set on the
491 * struct before it is inserted.
493 * This may return -EEXIST if the extent is already there, in which case the
494 * state struct is freed.
496 * The tree lock is not taken internally. This is a utility function and
497 * probably isn't what you want to call (see set/clear_extent_bit).
499 static int insert_state(struct extent_io_tree *tree,
500 struct extent_state *state, u64 start, u64 end,
502 struct rb_node **parent,
503 unsigned *bits, struct extent_changeset *changeset)
505 struct rb_node *node;
508 WARN(1, KERN_ERR "BTRFS: end < start %llu %llu\n",
510 state->start = start;
513 set_state_bits(tree, state, bits, changeset);
515 node = tree_insert(&tree->state, NULL, end, &state->rb_node, p, parent);
517 struct extent_state *found;
518 found = rb_entry(node, struct extent_state, rb_node);
519 pr_err("BTRFS: found node %llu %llu on insert of %llu %llu\n",
520 found->start, found->end, start, end);
523 merge_state(tree, state);
528 * split a given extent state struct in two, inserting the preallocated
529 * struct 'prealloc' as the newly created second half. 'split' indicates an
530 * offset inside 'orig' where it should be split.
533 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
534 * are two extent state structs in the tree:
535 * prealloc: [orig->start, split - 1]
536 * orig: [ split, orig->end ]
538 * The tree locks are not taken by this function. They need to be held
541 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
542 struct extent_state *prealloc, u64 split)
544 struct rb_node *node;
546 if (tree->private_data && is_data_inode(tree->private_data))
547 btrfs_split_delalloc_extent(tree->private_data, orig, split);
549 prealloc->start = orig->start;
550 prealloc->end = split - 1;
551 prealloc->state = orig->state;
554 node = tree_insert(&tree->state, &orig->rb_node, prealloc->end,
555 &prealloc->rb_node, NULL, NULL);
557 free_extent_state(prealloc);
563 static struct extent_state *next_state(struct extent_state *state)
565 struct rb_node *next = rb_next(&state->rb_node);
567 return rb_entry(next, struct extent_state, rb_node);
573 * utility function to clear some bits in an extent state struct.
574 * it will optionally wake up anyone waiting on this state (wake == 1).
576 * If no bits are set on the state struct after clearing things, the
577 * struct is freed and removed from the tree
579 static struct extent_state *clear_state_bit(struct extent_io_tree *tree,
580 struct extent_state *state,
581 unsigned *bits, int wake,
582 struct extent_changeset *changeset)
584 struct extent_state *next;
585 unsigned bits_to_clear = *bits & ~EXTENT_CTLBITS;
588 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
589 u64 range = state->end - state->start + 1;
590 WARN_ON(range > tree->dirty_bytes);
591 tree->dirty_bytes -= range;
594 if (tree->private_data && is_data_inode(tree->private_data))
595 btrfs_clear_delalloc_extent(tree->private_data, state, bits);
597 ret = add_extent_changeset(state, bits_to_clear, changeset, 0);
599 state->state &= ~bits_to_clear;
602 if (state->state == 0) {
603 next = next_state(state);
604 if (extent_state_in_tree(state)) {
605 rb_erase(&state->rb_node, &tree->state);
606 RB_CLEAR_NODE(&state->rb_node);
607 free_extent_state(state);
612 merge_state(tree, state);
613 next = next_state(state);
618 static struct extent_state *
619 alloc_extent_state_atomic(struct extent_state *prealloc)
622 prealloc = alloc_extent_state(GFP_ATOMIC);
627 static void extent_io_tree_panic(struct extent_io_tree *tree, int err)
629 struct inode *inode = tree->private_data;
631 btrfs_panic(btrfs_sb(inode->i_sb), err,
632 "locking error: extent tree was modified by another thread while locked");
636 * clear some bits on a range in the tree. This may require splitting
637 * or inserting elements in the tree, so the gfp mask is used to
638 * indicate which allocations or sleeping are allowed.
640 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
641 * the given range from the tree regardless of state (ie for truncate).
643 * the range [start, end] is inclusive.
645 * This takes the tree lock, and returns 0 on success and < 0 on error.
647 int __clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
648 unsigned bits, int wake, int delete,
649 struct extent_state **cached_state,
650 gfp_t mask, struct extent_changeset *changeset)
652 struct extent_state *state;
653 struct extent_state *cached;
654 struct extent_state *prealloc = NULL;
655 struct rb_node *node;
660 btrfs_debug_check_extent_io_range(tree, start, end);
661 trace_btrfs_clear_extent_bit(tree, start, end - start + 1, bits);
663 if (bits & EXTENT_DELALLOC)
664 bits |= EXTENT_NORESERVE;
667 bits |= ~EXTENT_CTLBITS;
669 if (bits & (EXTENT_LOCKED | EXTENT_BOUNDARY))
672 if (!prealloc && gfpflags_allow_blocking(mask)) {
674 * Don't care for allocation failure here because we might end
675 * up not needing the pre-allocated extent state at all, which
676 * is the case if we only have in the tree extent states that
677 * cover our input range and don't cover too any other range.
678 * If we end up needing a new extent state we allocate it later.
680 prealloc = alloc_extent_state(mask);
683 spin_lock(&tree->lock);
685 cached = *cached_state;
688 *cached_state = NULL;
692 if (cached && extent_state_in_tree(cached) &&
693 cached->start <= start && cached->end > start) {
695 refcount_dec(&cached->refs);
700 free_extent_state(cached);
703 * this search will find the extents that end after
706 node = tree_search(tree, start);
709 state = rb_entry(node, struct extent_state, rb_node);
711 if (state->start > end)
713 WARN_ON(state->end < start);
714 last_end = state->end;
716 /* the state doesn't have the wanted bits, go ahead */
717 if (!(state->state & bits)) {
718 state = next_state(state);
723 * | ---- desired range ---- |
725 * | ------------- state -------------- |
727 * We need to split the extent we found, and may flip
728 * bits on second half.
730 * If the extent we found extends past our range, we
731 * just split and search again. It'll get split again
732 * the next time though.
734 * If the extent we found is inside our range, we clear
735 * the desired bit on it.
738 if (state->start < start) {
739 prealloc = alloc_extent_state_atomic(prealloc);
741 err = split_state(tree, state, prealloc, start);
743 extent_io_tree_panic(tree, err);
748 if (state->end <= end) {
749 state = clear_state_bit(tree, state, &bits, wake,
756 * | ---- desired range ---- |
758 * We need to split the extent, and clear the bit
761 if (state->start <= end && state->end > end) {
762 prealloc = alloc_extent_state_atomic(prealloc);
764 err = split_state(tree, state, prealloc, end + 1);
766 extent_io_tree_panic(tree, err);
771 clear_state_bit(tree, prealloc, &bits, wake, changeset);
777 state = clear_state_bit(tree, state, &bits, wake, changeset);
779 if (last_end == (u64)-1)
781 start = last_end + 1;
782 if (start <= end && state && !need_resched())
788 spin_unlock(&tree->lock);
789 if (gfpflags_allow_blocking(mask))
794 spin_unlock(&tree->lock);
796 free_extent_state(prealloc);
802 static void wait_on_state(struct extent_io_tree *tree,
803 struct extent_state *state)
804 __releases(tree->lock)
805 __acquires(tree->lock)
808 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
809 spin_unlock(&tree->lock);
811 spin_lock(&tree->lock);
812 finish_wait(&state->wq, &wait);
816 * waits for one or more bits to clear on a range in the state tree.
817 * The range [start, end] is inclusive.
818 * The tree lock is taken by this function
820 static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
823 struct extent_state *state;
824 struct rb_node *node;
826 btrfs_debug_check_extent_io_range(tree, start, end);
828 spin_lock(&tree->lock);
832 * this search will find all the extents that end after
835 node = tree_search(tree, start);
840 state = rb_entry(node, struct extent_state, rb_node);
842 if (state->start > end)
845 if (state->state & bits) {
846 start = state->start;
847 refcount_inc(&state->refs);
848 wait_on_state(tree, state);
849 free_extent_state(state);
852 start = state->end + 1;
857 if (!cond_resched_lock(&tree->lock)) {
858 node = rb_next(node);
863 spin_unlock(&tree->lock);
866 static void set_state_bits(struct extent_io_tree *tree,
867 struct extent_state *state,
868 unsigned *bits, struct extent_changeset *changeset)
870 unsigned bits_to_set = *bits & ~EXTENT_CTLBITS;
873 if (tree->private_data && is_data_inode(tree->private_data))
874 btrfs_set_delalloc_extent(tree->private_data, state, bits);
876 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
877 u64 range = state->end - state->start + 1;
878 tree->dirty_bytes += range;
880 ret = add_extent_changeset(state, bits_to_set, changeset, 1);
882 state->state |= bits_to_set;
885 static void cache_state_if_flags(struct extent_state *state,
886 struct extent_state **cached_ptr,
889 if (cached_ptr && !(*cached_ptr)) {
890 if (!flags || (state->state & flags)) {
892 refcount_inc(&state->refs);
897 static void cache_state(struct extent_state *state,
898 struct extent_state **cached_ptr)
900 return cache_state_if_flags(state, cached_ptr,
901 EXTENT_LOCKED | EXTENT_BOUNDARY);
905 * set some bits on a range in the tree. This may require allocations or
906 * sleeping, so the gfp mask is used to indicate what is allowed.
908 * If any of the exclusive bits are set, this will fail with -EEXIST if some
909 * part of the range already has the desired bits set. The start of the
910 * existing range is returned in failed_start in this case.
912 * [start, end] is inclusive This takes the tree lock.
915 static int __must_check
916 __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
917 unsigned bits, unsigned exclusive_bits,
918 u64 *failed_start, struct extent_state **cached_state,
919 gfp_t mask, struct extent_changeset *changeset)
921 struct extent_state *state;
922 struct extent_state *prealloc = NULL;
923 struct rb_node *node;
925 struct rb_node *parent;
930 btrfs_debug_check_extent_io_range(tree, start, end);
931 trace_btrfs_set_extent_bit(tree, start, end - start + 1, bits);
934 if (!prealloc && gfpflags_allow_blocking(mask)) {
936 * Don't care for allocation failure here because we might end
937 * up not needing the pre-allocated extent state at all, which
938 * is the case if we only have in the tree extent states that
939 * cover our input range and don't cover too any other range.
940 * If we end up needing a new extent state we allocate it later.
942 prealloc = alloc_extent_state(mask);
945 spin_lock(&tree->lock);
946 if (cached_state && *cached_state) {
947 state = *cached_state;
948 if (state->start <= start && state->end > start &&
949 extent_state_in_tree(state)) {
950 node = &state->rb_node;
955 * this search will find all the extents that end after
958 node = tree_search_for_insert(tree, start, &p, &parent);
960 prealloc = alloc_extent_state_atomic(prealloc);
962 err = insert_state(tree, prealloc, start, end,
963 &p, &parent, &bits, changeset);
965 extent_io_tree_panic(tree, err);
967 cache_state(prealloc, cached_state);
971 state = rb_entry(node, struct extent_state, rb_node);
973 last_start = state->start;
974 last_end = state->end;
977 * | ---- desired range ---- |
980 * Just lock what we found and keep going
982 if (state->start == start && state->end <= end) {
983 if (state->state & exclusive_bits) {
984 *failed_start = state->start;
989 set_state_bits(tree, state, &bits, changeset);
990 cache_state(state, cached_state);
991 merge_state(tree, state);
992 if (last_end == (u64)-1)
994 start = last_end + 1;
995 state = next_state(state);
996 if (start < end && state && state->start == start &&
1003 * | ---- desired range ---- |
1006 * | ------------- state -------------- |
1008 * We need to split the extent we found, and may flip bits on
1011 * If the extent we found extends past our
1012 * range, we just split and search again. It'll get split
1013 * again the next time though.
1015 * If the extent we found is inside our range, we set the
1016 * desired bit on it.
1018 if (state->start < start) {
1019 if (state->state & exclusive_bits) {
1020 *failed_start = start;
1025 prealloc = alloc_extent_state_atomic(prealloc);
1027 err = split_state(tree, state, prealloc, start);
1029 extent_io_tree_panic(tree, err);
1034 if (state->end <= end) {
1035 set_state_bits(tree, state, &bits, changeset);
1036 cache_state(state, cached_state);
1037 merge_state(tree, state);
1038 if (last_end == (u64)-1)
1040 start = last_end + 1;
1041 state = next_state(state);
1042 if (start < end && state && state->start == start &&
1049 * | ---- desired range ---- |
1050 * | state | or | state |
1052 * There's a hole, we need to insert something in it and
1053 * ignore the extent we found.
1055 if (state->start > start) {
1057 if (end < last_start)
1060 this_end = last_start - 1;
1062 prealloc = alloc_extent_state_atomic(prealloc);
1066 * Avoid to free 'prealloc' if it can be merged with
1069 err = insert_state(tree, prealloc, start, this_end,
1070 NULL, NULL, &bits, changeset);
1072 extent_io_tree_panic(tree, err);
1074 cache_state(prealloc, cached_state);
1076 start = this_end + 1;
1080 * | ---- desired range ---- |
1082 * We need to split the extent, and set the bit
1085 if (state->start <= end && state->end > end) {
1086 if (state->state & exclusive_bits) {
1087 *failed_start = start;
1092 prealloc = alloc_extent_state_atomic(prealloc);
1094 err = split_state(tree, state, prealloc, end + 1);
1096 extent_io_tree_panic(tree, err);
1098 set_state_bits(tree, prealloc, &bits, changeset);
1099 cache_state(prealloc, cached_state);
1100 merge_state(tree, prealloc);
1108 spin_unlock(&tree->lock);
1109 if (gfpflags_allow_blocking(mask))
1114 spin_unlock(&tree->lock);
1116 free_extent_state(prealloc);
1122 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1123 unsigned bits, u64 * failed_start,
1124 struct extent_state **cached_state, gfp_t mask)
1126 return __set_extent_bit(tree, start, end, bits, 0, failed_start,
1127 cached_state, mask, NULL);
1132 * convert_extent_bit - convert all bits in a given range from one bit to
1134 * @tree: the io tree to search
1135 * @start: the start offset in bytes
1136 * @end: the end offset in bytes (inclusive)
1137 * @bits: the bits to set in this range
1138 * @clear_bits: the bits to clear in this range
1139 * @cached_state: state that we're going to cache
1141 * This will go through and set bits for the given range. If any states exist
1142 * already in this range they are set with the given bit and cleared of the
1143 * clear_bits. This is only meant to be used by things that are mergeable, ie
1144 * converting from say DELALLOC to DIRTY. This is not meant to be used with
1145 * boundary bits like LOCK.
1147 * All allocations are done with GFP_NOFS.
1149 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1150 unsigned bits, unsigned clear_bits,
1151 struct extent_state **cached_state)
1153 struct extent_state *state;
1154 struct extent_state *prealloc = NULL;
1155 struct rb_node *node;
1157 struct rb_node *parent;
1161 bool first_iteration = true;
1163 btrfs_debug_check_extent_io_range(tree, start, end);
1164 trace_btrfs_convert_extent_bit(tree, start, end - start + 1, bits,
1170 * Best effort, don't worry if extent state allocation fails
1171 * here for the first iteration. We might have a cached state
1172 * that matches exactly the target range, in which case no
1173 * extent state allocations are needed. We'll only know this
1174 * after locking the tree.
1176 prealloc = alloc_extent_state(GFP_NOFS);
1177 if (!prealloc && !first_iteration)
1181 spin_lock(&tree->lock);
1182 if (cached_state && *cached_state) {
1183 state = *cached_state;
1184 if (state->start <= start && state->end > start &&
1185 extent_state_in_tree(state)) {
1186 node = &state->rb_node;
1192 * this search will find all the extents that end after
1195 node = tree_search_for_insert(tree, start, &p, &parent);
1197 prealloc = alloc_extent_state_atomic(prealloc);
1202 err = insert_state(tree, prealloc, start, end,
1203 &p, &parent, &bits, NULL);
1205 extent_io_tree_panic(tree, err);
1206 cache_state(prealloc, cached_state);
1210 state = rb_entry(node, struct extent_state, rb_node);
1212 last_start = state->start;
1213 last_end = state->end;
1216 * | ---- desired range ---- |
1219 * Just lock what we found and keep going
1221 if (state->start == start && state->end <= end) {
1222 set_state_bits(tree, state, &bits, NULL);
1223 cache_state(state, cached_state);
1224 state = clear_state_bit(tree, state, &clear_bits, 0, NULL);
1225 if (last_end == (u64)-1)
1227 start = last_end + 1;
1228 if (start < end && state && state->start == start &&
1235 * | ---- desired range ---- |
1238 * | ------------- state -------------- |
1240 * We need to split the extent we found, and may flip bits on
1243 * If the extent we found extends past our
1244 * range, we just split and search again. It'll get split
1245 * again the next time though.
1247 * If the extent we found is inside our range, we set the
1248 * desired bit on it.
1250 if (state->start < start) {
1251 prealloc = alloc_extent_state_atomic(prealloc);
1256 err = split_state(tree, state, prealloc, start);
1258 extent_io_tree_panic(tree, err);
1262 if (state->end <= end) {
1263 set_state_bits(tree, state, &bits, NULL);
1264 cache_state(state, cached_state);
1265 state = clear_state_bit(tree, state, &clear_bits, 0,
1267 if (last_end == (u64)-1)
1269 start = last_end + 1;
1270 if (start < end && state && state->start == start &&
1277 * | ---- desired range ---- |
1278 * | state | or | state |
1280 * There's a hole, we need to insert something in it and
1281 * ignore the extent we found.
1283 if (state->start > start) {
1285 if (end < last_start)
1288 this_end = last_start - 1;
1290 prealloc = alloc_extent_state_atomic(prealloc);
1297 * Avoid to free 'prealloc' if it can be merged with
1300 err = insert_state(tree, prealloc, start, this_end,
1301 NULL, NULL, &bits, NULL);
1303 extent_io_tree_panic(tree, err);
1304 cache_state(prealloc, cached_state);
1306 start = this_end + 1;
1310 * | ---- desired range ---- |
1312 * We need to split the extent, and set the bit
1315 if (state->start <= end && state->end > end) {
1316 prealloc = alloc_extent_state_atomic(prealloc);
1322 err = split_state(tree, state, prealloc, end + 1);
1324 extent_io_tree_panic(tree, err);
1326 set_state_bits(tree, prealloc, &bits, NULL);
1327 cache_state(prealloc, cached_state);
1328 clear_state_bit(tree, prealloc, &clear_bits, 0, NULL);
1336 spin_unlock(&tree->lock);
1338 first_iteration = false;
1342 spin_unlock(&tree->lock);
1344 free_extent_state(prealloc);
1349 /* wrappers around set/clear extent bit */
1350 int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1351 unsigned bits, struct extent_changeset *changeset)
1354 * We don't support EXTENT_LOCKED yet, as current changeset will
1355 * record any bits changed, so for EXTENT_LOCKED case, it will
1356 * either fail with -EEXIST or changeset will record the whole
1359 BUG_ON(bits & EXTENT_LOCKED);
1361 return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL, GFP_NOFS,
1365 int set_extent_bits_nowait(struct extent_io_tree *tree, u64 start, u64 end,
1368 return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL,
1372 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1373 unsigned bits, int wake, int delete,
1374 struct extent_state **cached)
1376 return __clear_extent_bit(tree, start, end, bits, wake, delete,
1377 cached, GFP_NOFS, NULL);
1380 int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1381 unsigned bits, struct extent_changeset *changeset)
1384 * Don't support EXTENT_LOCKED case, same reason as
1385 * set_record_extent_bits().
1387 BUG_ON(bits & EXTENT_LOCKED);
1389 return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, GFP_NOFS,
1394 * either insert or lock state struct between start and end use mask to tell
1395 * us if waiting is desired.
1397 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1398 struct extent_state **cached_state)
1404 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED,
1405 EXTENT_LOCKED, &failed_start,
1406 cached_state, GFP_NOFS, NULL);
1407 if (err == -EEXIST) {
1408 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1409 start = failed_start;
1412 WARN_ON(start > end);
1417 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1422 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1423 &failed_start, NULL, GFP_NOFS, NULL);
1424 if (err == -EEXIST) {
1425 if (failed_start > start)
1426 clear_extent_bit(tree, start, failed_start - 1,
1427 EXTENT_LOCKED, 1, 0, NULL);
1433 void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
1435 unsigned long index = start >> PAGE_SHIFT;
1436 unsigned long end_index = end >> PAGE_SHIFT;
1439 while (index <= end_index) {
1440 page = find_get_page(inode->i_mapping, index);
1441 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1442 clear_page_dirty_for_io(page);
1448 void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
1450 unsigned long index = start >> PAGE_SHIFT;
1451 unsigned long end_index = end >> PAGE_SHIFT;
1454 while (index <= end_index) {
1455 page = find_get_page(inode->i_mapping, index);
1456 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1457 __set_page_dirty_nobuffers(page);
1458 account_page_redirty(page);
1464 /* find the first state struct with 'bits' set after 'start', and
1465 * return it. tree->lock must be held. NULL will returned if
1466 * nothing was found after 'start'
1468 static struct extent_state *
1469 find_first_extent_bit_state(struct extent_io_tree *tree,
1470 u64 start, unsigned bits)
1472 struct rb_node *node;
1473 struct extent_state *state;
1476 * this search will find all the extents that end after
1479 node = tree_search(tree, start);
1484 state = rb_entry(node, struct extent_state, rb_node);
1485 if (state->end >= start && (state->state & bits))
1488 node = rb_next(node);
1497 * find the first offset in the io tree with 'bits' set. zero is
1498 * returned if we find something, and *start_ret and *end_ret are
1499 * set to reflect the state struct that was found.
1501 * If nothing was found, 1 is returned. If found something, return 0.
1503 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1504 u64 *start_ret, u64 *end_ret, unsigned bits,
1505 struct extent_state **cached_state)
1507 struct extent_state *state;
1510 spin_lock(&tree->lock);
1511 if (cached_state && *cached_state) {
1512 state = *cached_state;
1513 if (state->end == start - 1 && extent_state_in_tree(state)) {
1514 while ((state = next_state(state)) != NULL) {
1515 if (state->state & bits)
1518 free_extent_state(*cached_state);
1519 *cached_state = NULL;
1522 free_extent_state(*cached_state);
1523 *cached_state = NULL;
1526 state = find_first_extent_bit_state(tree, start, bits);
1529 cache_state_if_flags(state, cached_state, 0);
1530 *start_ret = state->start;
1531 *end_ret = state->end;
1535 spin_unlock(&tree->lock);
1540 * find_first_clear_extent_bit - find the first range that has @bits not set.
1541 * This range could start before @start.
1543 * @tree - the tree to search
1544 * @start - the offset at/after which the found extent should start
1545 * @start_ret - records the beginning of the range
1546 * @end_ret - records the end of the range (inclusive)
1547 * @bits - the set of bits which must be unset
1549 * Since unallocated range is also considered one which doesn't have the bits
1550 * set it's possible that @end_ret contains -1, this happens in case the range
1551 * spans (last_range_end, end of device]. In this case it's up to the caller to
1552 * trim @end_ret to the appropriate size.
1554 void find_first_clear_extent_bit(struct extent_io_tree *tree, u64 start,
1555 u64 *start_ret, u64 *end_ret, unsigned bits)
1557 struct extent_state *state;
1558 struct rb_node *node, *prev = NULL, *next;
1560 spin_lock(&tree->lock);
1562 /* Find first extent with bits cleared */
1564 node = __etree_search(tree, start, &next, &prev, NULL, NULL);
1569 * We are past the last allocated chunk,
1570 * set start at the end of the last extent. The
1571 * device alloc tree should never be empty so
1572 * prev is always set.
1575 state = rb_entry(prev, struct extent_state, rb_node);
1576 *start_ret = state->end + 1;
1582 * At this point 'node' either contains 'start' or start is
1585 state = rb_entry(node, struct extent_state, rb_node);
1587 if (in_range(start, state->start, state->end - state->start + 1)) {
1588 if (state->state & bits) {
1590 * |--range with bits sets--|
1594 start = state->end + 1;
1597 * 'start' falls within a range that doesn't
1598 * have the bits set, so take its start as
1599 * the beginning of the desired range
1601 * |--range with bits cleared----|
1605 *start_ret = state->start;
1610 * |---prev range---|---hole/unset---|---node range---|
1616 * |---hole/unset--||--first node--|
1621 state = rb_entry(prev, struct extent_state,
1623 *start_ret = state->end + 1;
1632 * Find the longest stretch from start until an entry which has the
1636 state = rb_entry(node, struct extent_state, rb_node);
1637 if (state->end >= start && !(state->state & bits)) {
1638 *end_ret = state->end;
1640 *end_ret = state->start - 1;
1644 node = rb_next(node);
1649 spin_unlock(&tree->lock);
1653 * find a contiguous range of bytes in the file marked as delalloc, not
1654 * more than 'max_bytes'. start and end are used to return the range,
1656 * true is returned if we find something, false if nothing was in the tree
1658 static noinline bool find_delalloc_range(struct extent_io_tree *tree,
1659 u64 *start, u64 *end, u64 max_bytes,
1660 struct extent_state **cached_state)
1662 struct rb_node *node;
1663 struct extent_state *state;
1664 u64 cur_start = *start;
1666 u64 total_bytes = 0;
1668 spin_lock(&tree->lock);
1671 * this search will find all the extents that end after
1674 node = tree_search(tree, cur_start);
1681 state = rb_entry(node, struct extent_state, rb_node);
1682 if (found && (state->start != cur_start ||
1683 (state->state & EXTENT_BOUNDARY))) {
1686 if (!(state->state & EXTENT_DELALLOC)) {
1692 *start = state->start;
1693 *cached_state = state;
1694 refcount_inc(&state->refs);
1698 cur_start = state->end + 1;
1699 node = rb_next(node);
1700 total_bytes += state->end - state->start + 1;
1701 if (total_bytes >= max_bytes)
1707 spin_unlock(&tree->lock);
1711 static int __process_pages_contig(struct address_space *mapping,
1712 struct page *locked_page,
1713 pgoff_t start_index, pgoff_t end_index,
1714 unsigned long page_ops, pgoff_t *index_ret);
1716 static noinline void __unlock_for_delalloc(struct inode *inode,
1717 struct page *locked_page,
1720 unsigned long index = start >> PAGE_SHIFT;
1721 unsigned long end_index = end >> PAGE_SHIFT;
1723 ASSERT(locked_page);
1724 if (index == locked_page->index && end_index == index)
1727 __process_pages_contig(inode->i_mapping, locked_page, index, end_index,
1731 static noinline int lock_delalloc_pages(struct inode *inode,
1732 struct page *locked_page,
1736 unsigned long index = delalloc_start >> PAGE_SHIFT;
1737 unsigned long index_ret = index;
1738 unsigned long end_index = delalloc_end >> PAGE_SHIFT;
1741 ASSERT(locked_page);
1742 if (index == locked_page->index && index == end_index)
1745 ret = __process_pages_contig(inode->i_mapping, locked_page, index,
1746 end_index, PAGE_LOCK, &index_ret);
1748 __unlock_for_delalloc(inode, locked_page, delalloc_start,
1749 (u64)index_ret << PAGE_SHIFT);
1754 * Find and lock a contiguous range of bytes in the file marked as delalloc, no
1755 * more than @max_bytes. @Start and @end are used to return the range,
1757 * Return: true if we find something
1758 * false if nothing was in the tree
1761 noinline_for_stack bool find_lock_delalloc_range(struct inode *inode,
1762 struct extent_io_tree *tree,
1763 struct page *locked_page, u64 *start,
1766 u64 max_bytes = BTRFS_MAX_EXTENT_SIZE;
1770 struct extent_state *cached_state = NULL;
1775 /* step one, find a bunch of delalloc bytes starting at start */
1776 delalloc_start = *start;
1778 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1779 max_bytes, &cached_state);
1780 if (!found || delalloc_end <= *start) {
1781 *start = delalloc_start;
1782 *end = delalloc_end;
1783 free_extent_state(cached_state);
1788 * start comes from the offset of locked_page. We have to lock
1789 * pages in order, so we can't process delalloc bytes before
1792 if (delalloc_start < *start)
1793 delalloc_start = *start;
1796 * make sure to limit the number of pages we try to lock down
1798 if (delalloc_end + 1 - delalloc_start > max_bytes)
1799 delalloc_end = delalloc_start + max_bytes - 1;
1801 /* step two, lock all the pages after the page that has start */
1802 ret = lock_delalloc_pages(inode, locked_page,
1803 delalloc_start, delalloc_end);
1804 ASSERT(!ret || ret == -EAGAIN);
1805 if (ret == -EAGAIN) {
1806 /* some of the pages are gone, lets avoid looping by
1807 * shortening the size of the delalloc range we're searching
1809 free_extent_state(cached_state);
1810 cached_state = NULL;
1812 max_bytes = PAGE_SIZE;
1821 /* step three, lock the state bits for the whole range */
1822 lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state);
1824 /* then test to make sure it is all still delalloc */
1825 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1826 EXTENT_DELALLOC, 1, cached_state);
1828 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1830 __unlock_for_delalloc(inode, locked_page,
1831 delalloc_start, delalloc_end);
1835 free_extent_state(cached_state);
1836 *start = delalloc_start;
1837 *end = delalloc_end;
1842 static int __process_pages_contig(struct address_space *mapping,
1843 struct page *locked_page,
1844 pgoff_t start_index, pgoff_t end_index,
1845 unsigned long page_ops, pgoff_t *index_ret)
1847 unsigned long nr_pages = end_index - start_index + 1;
1848 unsigned long pages_locked = 0;
1849 pgoff_t index = start_index;
1850 struct page *pages[16];
1855 if (page_ops & PAGE_LOCK) {
1856 ASSERT(page_ops == PAGE_LOCK);
1857 ASSERT(index_ret && *index_ret == start_index);
1860 if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0)
1861 mapping_set_error(mapping, -EIO);
1863 while (nr_pages > 0) {
1864 ret = find_get_pages_contig(mapping, index,
1865 min_t(unsigned long,
1866 nr_pages, ARRAY_SIZE(pages)), pages);
1869 * Only if we're going to lock these pages,
1870 * can we find nothing at @index.
1872 ASSERT(page_ops & PAGE_LOCK);
1877 for (i = 0; i < ret; i++) {
1878 if (page_ops & PAGE_SET_PRIVATE2)
1879 SetPagePrivate2(pages[i]);
1881 if (pages[i] == locked_page) {
1886 if (page_ops & PAGE_CLEAR_DIRTY)
1887 clear_page_dirty_for_io(pages[i]);
1888 if (page_ops & PAGE_SET_WRITEBACK)
1889 set_page_writeback(pages[i]);
1890 if (page_ops & PAGE_SET_ERROR)
1891 SetPageError(pages[i]);
1892 if (page_ops & PAGE_END_WRITEBACK)
1893 end_page_writeback(pages[i]);
1894 if (page_ops & PAGE_UNLOCK)
1895 unlock_page(pages[i]);
1896 if (page_ops & PAGE_LOCK) {
1897 lock_page(pages[i]);
1898 if (!PageDirty(pages[i]) ||
1899 pages[i]->mapping != mapping) {
1900 unlock_page(pages[i]);
1914 if (err && index_ret)
1915 *index_ret = start_index + pages_locked - 1;
1919 void extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end,
1920 u64 delalloc_end, struct page *locked_page,
1921 unsigned clear_bits,
1922 unsigned long page_ops)
1924 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, clear_bits, 1, 0,
1927 __process_pages_contig(inode->i_mapping, locked_page,
1928 start >> PAGE_SHIFT, end >> PAGE_SHIFT,
1933 * count the number of bytes in the tree that have a given bit(s)
1934 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1935 * cached. The total number found is returned.
1937 u64 count_range_bits(struct extent_io_tree *tree,
1938 u64 *start, u64 search_end, u64 max_bytes,
1939 unsigned bits, int contig)
1941 struct rb_node *node;
1942 struct extent_state *state;
1943 u64 cur_start = *start;
1944 u64 total_bytes = 0;
1948 if (WARN_ON(search_end <= cur_start))
1951 spin_lock(&tree->lock);
1952 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1953 total_bytes = tree->dirty_bytes;
1957 * this search will find all the extents that end after
1960 node = tree_search(tree, cur_start);
1965 state = rb_entry(node, struct extent_state, rb_node);
1966 if (state->start > search_end)
1968 if (contig && found && state->start > last + 1)
1970 if (state->end >= cur_start && (state->state & bits) == bits) {
1971 total_bytes += min(search_end, state->end) + 1 -
1972 max(cur_start, state->start);
1973 if (total_bytes >= max_bytes)
1976 *start = max(cur_start, state->start);
1980 } else if (contig && found) {
1983 node = rb_next(node);
1988 spin_unlock(&tree->lock);
1993 * set the private field for a given byte offset in the tree. If there isn't
1994 * an extent_state there already, this does nothing.
1996 static noinline int set_state_failrec(struct extent_io_tree *tree, u64 start,
1997 struct io_failure_record *failrec)
1999 struct rb_node *node;
2000 struct extent_state *state;
2003 spin_lock(&tree->lock);
2005 * this search will find all the extents that end after
2008 node = tree_search(tree, start);
2013 state = rb_entry(node, struct extent_state, rb_node);
2014 if (state->start != start) {
2018 state->failrec = failrec;
2020 spin_unlock(&tree->lock);
2024 static noinline int get_state_failrec(struct extent_io_tree *tree, u64 start,
2025 struct io_failure_record **failrec)
2027 struct rb_node *node;
2028 struct extent_state *state;
2031 spin_lock(&tree->lock);
2033 * this search will find all the extents that end after
2036 node = tree_search(tree, start);
2041 state = rb_entry(node, struct extent_state, rb_node);
2042 if (state->start != start) {
2046 *failrec = state->failrec;
2048 spin_unlock(&tree->lock);
2053 * searches a range in the state tree for a given mask.
2054 * If 'filled' == 1, this returns 1 only if every extent in the tree
2055 * has the bits set. Otherwise, 1 is returned if any bit in the
2056 * range is found set.
2058 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
2059 unsigned bits, int filled, struct extent_state *cached)
2061 struct extent_state *state = NULL;
2062 struct rb_node *node;
2065 spin_lock(&tree->lock);
2066 if (cached && extent_state_in_tree(cached) && cached->start <= start &&
2067 cached->end > start)
2068 node = &cached->rb_node;
2070 node = tree_search(tree, start);
2071 while (node && start <= end) {
2072 state = rb_entry(node, struct extent_state, rb_node);
2074 if (filled && state->start > start) {
2079 if (state->start > end)
2082 if (state->state & bits) {
2086 } else if (filled) {
2091 if (state->end == (u64)-1)
2094 start = state->end + 1;
2097 node = rb_next(node);
2104 spin_unlock(&tree->lock);
2109 * helper function to set a given page up to date if all the
2110 * extents in the tree for that page are up to date
2112 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
2114 u64 start = page_offset(page);
2115 u64 end = start + PAGE_SIZE - 1;
2116 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
2117 SetPageUptodate(page);
2120 int free_io_failure(struct extent_io_tree *failure_tree,
2121 struct extent_io_tree *io_tree,
2122 struct io_failure_record *rec)
2127 set_state_failrec(failure_tree, rec->start, NULL);
2128 ret = clear_extent_bits(failure_tree, rec->start,
2129 rec->start + rec->len - 1,
2130 EXTENT_LOCKED | EXTENT_DIRTY);
2134 ret = clear_extent_bits(io_tree, rec->start,
2135 rec->start + rec->len - 1,
2145 * this bypasses the standard btrfs submit functions deliberately, as
2146 * the standard behavior is to write all copies in a raid setup. here we only
2147 * want to write the one bad copy. so we do the mapping for ourselves and issue
2148 * submit_bio directly.
2149 * to avoid any synchronization issues, wait for the data after writing, which
2150 * actually prevents the read that triggered the error from finishing.
2151 * currently, there can be no more than two copies of every data bit. thus,
2152 * exactly one rewrite is required.
2154 int repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start,
2155 u64 length, u64 logical, struct page *page,
2156 unsigned int pg_offset, int mirror_num)
2159 struct btrfs_device *dev;
2162 struct btrfs_bio *bbio = NULL;
2165 ASSERT(!(fs_info->sb->s_flags & SB_RDONLY));
2166 BUG_ON(!mirror_num);
2168 bio = btrfs_io_bio_alloc(1);
2169 bio->bi_iter.bi_size = 0;
2170 map_length = length;
2173 * Avoid races with device replace and make sure our bbio has devices
2174 * associated to its stripes that don't go away while we are doing the
2175 * read repair operation.
2177 btrfs_bio_counter_inc_blocked(fs_info);
2178 if (btrfs_is_parity_mirror(fs_info, logical, length)) {
2180 * Note that we don't use BTRFS_MAP_WRITE because it's supposed
2181 * to update all raid stripes, but here we just want to correct
2182 * bad stripe, thus BTRFS_MAP_READ is abused to only get the bad
2183 * stripe's dev and sector.
2185 ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, logical,
2186 &map_length, &bbio, 0);
2188 btrfs_bio_counter_dec(fs_info);
2192 ASSERT(bbio->mirror_num == 1);
2194 ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical,
2195 &map_length, &bbio, mirror_num);
2197 btrfs_bio_counter_dec(fs_info);
2201 BUG_ON(mirror_num != bbio->mirror_num);
2204 sector = bbio->stripes[bbio->mirror_num - 1].physical >> 9;
2205 bio->bi_iter.bi_sector = sector;
2206 dev = bbio->stripes[bbio->mirror_num - 1].dev;
2207 btrfs_put_bbio(bbio);
2208 if (!dev || !dev->bdev ||
2209 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) {
2210 btrfs_bio_counter_dec(fs_info);
2214 bio_set_dev(bio, dev->bdev);
2215 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC;
2216 bio_add_page(bio, page, length, pg_offset);
2218 if (btrfsic_submit_bio_wait(bio)) {
2219 /* try to remap that extent elsewhere? */
2220 btrfs_bio_counter_dec(fs_info);
2222 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
2226 btrfs_info_rl_in_rcu(fs_info,
2227 "read error corrected: ino %llu off %llu (dev %s sector %llu)",
2229 rcu_str_deref(dev->name), sector);
2230 btrfs_bio_counter_dec(fs_info);
2235 int btrfs_repair_eb_io_failure(struct extent_buffer *eb, int mirror_num)
2237 struct btrfs_fs_info *fs_info = eb->fs_info;
2238 u64 start = eb->start;
2239 int i, num_pages = num_extent_pages(eb);
2242 if (sb_rdonly(fs_info->sb))
2245 for (i = 0; i < num_pages; i++) {
2246 struct page *p = eb->pages[i];
2248 ret = repair_io_failure(fs_info, 0, start, PAGE_SIZE, start, p,
2249 start - page_offset(p), mirror_num);
2259 * each time an IO finishes, we do a fast check in the IO failure tree
2260 * to see if we need to process or clean up an io_failure_record
2262 int clean_io_failure(struct btrfs_fs_info *fs_info,
2263 struct extent_io_tree *failure_tree,
2264 struct extent_io_tree *io_tree, u64 start,
2265 struct page *page, u64 ino, unsigned int pg_offset)
2268 struct io_failure_record *failrec;
2269 struct extent_state *state;
2274 ret = count_range_bits(failure_tree, &private, (u64)-1, 1,
2279 ret = get_state_failrec(failure_tree, start, &failrec);
2283 BUG_ON(!failrec->this_mirror);
2285 if (failrec->in_validation) {
2286 /* there was no real error, just free the record */
2287 btrfs_debug(fs_info,
2288 "clean_io_failure: freeing dummy error at %llu",
2292 if (sb_rdonly(fs_info->sb))
2295 spin_lock(&io_tree->lock);
2296 state = find_first_extent_bit_state(io_tree,
2299 spin_unlock(&io_tree->lock);
2301 if (state && state->start <= failrec->start &&
2302 state->end >= failrec->start + failrec->len - 1) {
2303 num_copies = btrfs_num_copies(fs_info, failrec->logical,
2305 if (num_copies > 1) {
2306 repair_io_failure(fs_info, ino, start, failrec->len,
2307 failrec->logical, page, pg_offset,
2308 failrec->failed_mirror);
2313 free_io_failure(failure_tree, io_tree, failrec);
2319 * Can be called when
2320 * - hold extent lock
2321 * - under ordered extent
2322 * - the inode is freeing
2324 void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start, u64 end)
2326 struct extent_io_tree *failure_tree = &inode->io_failure_tree;
2327 struct io_failure_record *failrec;
2328 struct extent_state *state, *next;
2330 if (RB_EMPTY_ROOT(&failure_tree->state))
2333 spin_lock(&failure_tree->lock);
2334 state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY);
2336 if (state->start > end)
2339 ASSERT(state->end <= end);
2341 next = next_state(state);
2343 failrec = state->failrec;
2344 free_extent_state(state);
2349 spin_unlock(&failure_tree->lock);
2352 int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end,
2353 struct io_failure_record **failrec_ret)
2355 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2356 struct io_failure_record *failrec;
2357 struct extent_map *em;
2358 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2359 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2360 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2364 ret = get_state_failrec(failure_tree, start, &failrec);
2366 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2370 failrec->start = start;
2371 failrec->len = end - start + 1;
2372 failrec->this_mirror = 0;
2373 failrec->bio_flags = 0;
2374 failrec->in_validation = 0;
2376 read_lock(&em_tree->lock);
2377 em = lookup_extent_mapping(em_tree, start, failrec->len);
2379 read_unlock(&em_tree->lock);
2384 if (em->start > start || em->start + em->len <= start) {
2385 free_extent_map(em);
2388 read_unlock(&em_tree->lock);
2394 logical = start - em->start;
2395 logical = em->block_start + logical;
2396 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2397 logical = em->block_start;
2398 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2399 extent_set_compress_type(&failrec->bio_flags,
2403 btrfs_debug(fs_info,
2404 "Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu",
2405 logical, start, failrec->len);
2407 failrec->logical = logical;
2408 free_extent_map(em);
2410 /* set the bits in the private failure tree */
2411 ret = set_extent_bits(failure_tree, start, end,
2412 EXTENT_LOCKED | EXTENT_DIRTY);
2414 ret = set_state_failrec(failure_tree, start, failrec);
2415 /* set the bits in the inode's tree */
2417 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED);
2423 btrfs_debug(fs_info,
2424 "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d",
2425 failrec->logical, failrec->start, failrec->len,
2426 failrec->in_validation);
2428 * when data can be on disk more than twice, add to failrec here
2429 * (e.g. with a list for failed_mirror) to make
2430 * clean_io_failure() clean all those errors at once.
2434 *failrec_ret = failrec;
2439 bool btrfs_check_repairable(struct inode *inode, unsigned failed_bio_pages,
2440 struct io_failure_record *failrec, int failed_mirror)
2442 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2445 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
2446 if (num_copies == 1) {
2448 * we only have a single copy of the data, so don't bother with
2449 * all the retry and error correction code that follows. no
2450 * matter what the error is, it is very likely to persist.
2452 btrfs_debug(fs_info,
2453 "Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
2454 num_copies, failrec->this_mirror, failed_mirror);
2459 * there are two premises:
2460 * a) deliver good data to the caller
2461 * b) correct the bad sectors on disk
2463 if (failed_bio_pages > 1) {
2465 * to fulfill b), we need to know the exact failing sectors, as
2466 * we don't want to rewrite any more than the failed ones. thus,
2467 * we need separate read requests for the failed bio
2469 * if the following BUG_ON triggers, our validation request got
2470 * merged. we need separate requests for our algorithm to work.
2472 BUG_ON(failrec->in_validation);
2473 failrec->in_validation = 1;
2474 failrec->this_mirror = failed_mirror;
2477 * we're ready to fulfill a) and b) alongside. get a good copy
2478 * of the failed sector and if we succeed, we have setup
2479 * everything for repair_io_failure to do the rest for us.
2481 if (failrec->in_validation) {
2482 BUG_ON(failrec->this_mirror != failed_mirror);
2483 failrec->in_validation = 0;
2484 failrec->this_mirror = 0;
2486 failrec->failed_mirror = failed_mirror;
2487 failrec->this_mirror++;
2488 if (failrec->this_mirror == failed_mirror)
2489 failrec->this_mirror++;
2492 if (failrec->this_mirror > num_copies) {
2493 btrfs_debug(fs_info,
2494 "Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
2495 num_copies, failrec->this_mirror, failed_mirror);
2503 struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio,
2504 struct io_failure_record *failrec,
2505 struct page *page, int pg_offset, int icsum,
2506 bio_end_io_t *endio_func, void *data)
2508 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2510 struct btrfs_io_bio *btrfs_failed_bio;
2511 struct btrfs_io_bio *btrfs_bio;
2513 bio = btrfs_io_bio_alloc(1);
2514 bio->bi_end_io = endio_func;
2515 bio->bi_iter.bi_sector = failrec->logical >> 9;
2516 bio_set_dev(bio, fs_info->fs_devices->latest_bdev);
2517 bio->bi_iter.bi_size = 0;
2518 bio->bi_private = data;
2520 btrfs_failed_bio = btrfs_io_bio(failed_bio);
2521 if (btrfs_failed_bio->csum) {
2522 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
2524 btrfs_bio = btrfs_io_bio(bio);
2525 btrfs_bio->csum = btrfs_bio->csum_inline;
2527 memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + icsum,
2531 bio_add_page(bio, page, failrec->len, pg_offset);
2537 * This is a generic handler for readpage errors. If other copies exist, read
2538 * those and write back good data to the failed position. Does not investigate
2539 * in remapping the failed extent elsewhere, hoping the device will be smart
2540 * enough to do this as needed
2542 static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset,
2543 struct page *page, u64 start, u64 end,
2546 struct io_failure_record *failrec;
2547 struct inode *inode = page->mapping->host;
2548 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2549 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2552 blk_status_t status;
2554 unsigned failed_bio_pages = failed_bio->bi_iter.bi_size >> PAGE_SHIFT;
2556 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
2558 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
2562 if (!btrfs_check_repairable(inode, failed_bio_pages, failrec,
2564 free_io_failure(failure_tree, tree, failrec);
2568 if (failed_bio_pages > 1)
2569 read_mode |= REQ_FAILFAST_DEV;
2571 phy_offset >>= inode->i_sb->s_blocksize_bits;
2572 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
2573 start - page_offset(page),
2574 (int)phy_offset, failed_bio->bi_end_io,
2576 bio->bi_opf = REQ_OP_READ | read_mode;
2578 btrfs_debug(btrfs_sb(inode->i_sb),
2579 "Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d",
2580 read_mode, failrec->this_mirror, failrec->in_validation);
2582 status = tree->ops->submit_bio_hook(tree->private_data, bio, failrec->this_mirror,
2583 failrec->bio_flags);
2585 free_io_failure(failure_tree, tree, failrec);
2587 ret = blk_status_to_errno(status);
2593 /* lots and lots of room for performance fixes in the end_bio funcs */
2595 void end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2597 int uptodate = (err == 0);
2600 btrfs_writepage_endio_finish_ordered(page, start, end, uptodate);
2603 ClearPageUptodate(page);
2605 ret = err < 0 ? err : -EIO;
2606 mapping_set_error(page->mapping, ret);
2611 * after a writepage IO is done, we need to:
2612 * clear the uptodate bits on error
2613 * clear the writeback bits in the extent tree for this IO
2614 * end_page_writeback if the page has no more pending IO
2616 * Scheduling is not allowed, so the extent state tree is expected
2617 * to have one and only one object corresponding to this IO.
2619 static void end_bio_extent_writepage(struct bio *bio)
2621 int error = blk_status_to_errno(bio->bi_status);
2622 struct bio_vec *bvec;
2625 struct bvec_iter_all iter_all;
2627 ASSERT(!bio_flagged(bio, BIO_CLONED));
2628 bio_for_each_segment_all(bvec, bio, iter_all) {
2629 struct page *page = bvec->bv_page;
2630 struct inode *inode = page->mapping->host;
2631 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2633 /* We always issue full-page reads, but if some block
2634 * in a page fails to read, blk_update_request() will
2635 * advance bv_offset and adjust bv_len to compensate.
2636 * Print a warning for nonzero offsets, and an error
2637 * if they don't add up to a full page. */
2638 if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) {
2639 if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE)
2641 "partial page write in btrfs with offset %u and length %u",
2642 bvec->bv_offset, bvec->bv_len);
2645 "incomplete page write in btrfs with offset %u and length %u",
2646 bvec->bv_offset, bvec->bv_len);
2649 start = page_offset(page);
2650 end = start + bvec->bv_offset + bvec->bv_len - 1;
2652 end_extent_writepage(page, error, start, end);
2653 end_page_writeback(page);
2660 endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len,
2663 struct extent_state *cached = NULL;
2664 u64 end = start + len - 1;
2666 if (uptodate && tree->track_uptodate)
2667 set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC);
2668 unlock_extent_cached_atomic(tree, start, end, &cached);
2672 * after a readpage IO is done, we need to:
2673 * clear the uptodate bits on error
2674 * set the uptodate bits if things worked
2675 * set the page up to date if all extents in the tree are uptodate
2676 * clear the lock bit in the extent tree
2677 * unlock the page if there are no other extents locked for it
2679 * Scheduling is not allowed, so the extent state tree is expected
2680 * to have one and only one object corresponding to this IO.
2682 static void end_bio_extent_readpage(struct bio *bio)
2684 struct bio_vec *bvec;
2685 int uptodate = !bio->bi_status;
2686 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
2687 struct extent_io_tree *tree, *failure_tree;
2692 u64 extent_start = 0;
2696 struct bvec_iter_all iter_all;
2698 ASSERT(!bio_flagged(bio, BIO_CLONED));
2699 bio_for_each_segment_all(bvec, bio, iter_all) {
2700 struct page *page = bvec->bv_page;
2701 struct inode *inode = page->mapping->host;
2702 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2703 bool data_inode = btrfs_ino(BTRFS_I(inode))
2704 != BTRFS_BTREE_INODE_OBJECTID;
2706 btrfs_debug(fs_info,
2707 "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u",
2708 (u64)bio->bi_iter.bi_sector, bio->bi_status,
2709 io_bio->mirror_num);
2710 tree = &BTRFS_I(inode)->io_tree;
2711 failure_tree = &BTRFS_I(inode)->io_failure_tree;
2713 /* We always issue full-page reads, but if some block
2714 * in a page fails to read, blk_update_request() will
2715 * advance bv_offset and adjust bv_len to compensate.
2716 * Print a warning for nonzero offsets, and an error
2717 * if they don't add up to a full page. */
2718 if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) {
2719 if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE)
2721 "partial page read in btrfs with offset %u and length %u",
2722 bvec->bv_offset, bvec->bv_len);
2725 "incomplete page read in btrfs with offset %u and length %u",
2726 bvec->bv_offset, bvec->bv_len);
2729 start = page_offset(page);
2730 end = start + bvec->bv_offset + bvec->bv_len - 1;
2733 mirror = io_bio->mirror_num;
2734 if (likely(uptodate)) {
2735 ret = tree->ops->readpage_end_io_hook(io_bio, offset,
2741 clean_io_failure(BTRFS_I(inode)->root->fs_info,
2742 failure_tree, tree, start,
2744 btrfs_ino(BTRFS_I(inode)), 0);
2747 if (likely(uptodate))
2753 * The generic bio_readpage_error handles errors the
2754 * following way: If possible, new read requests are
2755 * created and submitted and will end up in
2756 * end_bio_extent_readpage as well (if we're lucky,
2757 * not in the !uptodate case). In that case it returns
2758 * 0 and we just go on with the next page in our bio.
2759 * If it can't handle the error it will return -EIO and
2760 * we remain responsible for that page.
2762 ret = bio_readpage_error(bio, offset, page, start, end,
2765 uptodate = !bio->bi_status;
2770 struct extent_buffer *eb;
2772 eb = (struct extent_buffer *)page->private;
2773 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
2774 eb->read_mirror = mirror;
2775 atomic_dec(&eb->io_pages);
2776 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD,
2778 btree_readahead_hook(eb, -EIO);
2781 if (likely(uptodate)) {
2782 loff_t i_size = i_size_read(inode);
2783 pgoff_t end_index = i_size >> PAGE_SHIFT;
2786 /* Zero out the end if this page straddles i_size */
2787 off = offset_in_page(i_size);
2788 if (page->index == end_index && off)
2789 zero_user_segment(page, off, PAGE_SIZE);
2790 SetPageUptodate(page);
2792 ClearPageUptodate(page);
2798 if (unlikely(!uptodate)) {
2800 endio_readpage_release_extent(tree,
2806 endio_readpage_release_extent(tree, start,
2807 end - start + 1, 0);
2808 } else if (!extent_len) {
2809 extent_start = start;
2810 extent_len = end + 1 - start;
2811 } else if (extent_start + extent_len == start) {
2812 extent_len += end + 1 - start;
2814 endio_readpage_release_extent(tree, extent_start,
2815 extent_len, uptodate);
2816 extent_start = start;
2817 extent_len = end + 1 - start;
2822 endio_readpage_release_extent(tree, extent_start, extent_len,
2824 btrfs_io_bio_free_csum(io_bio);
2829 * Initialize the members up to but not including 'bio'. Use after allocating a
2830 * new bio by bio_alloc_bioset as it does not initialize the bytes outside of
2831 * 'bio' because use of __GFP_ZERO is not supported.
2833 static inline void btrfs_io_bio_init(struct btrfs_io_bio *btrfs_bio)
2835 memset(btrfs_bio, 0, offsetof(struct btrfs_io_bio, bio));
2839 * The following helpers allocate a bio. As it's backed by a bioset, it'll
2840 * never fail. We're returning a bio right now but you can call btrfs_io_bio
2841 * for the appropriate container_of magic
2843 struct bio *btrfs_bio_alloc(struct block_device *bdev, u64 first_byte)
2847 bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, &btrfs_bioset);
2848 bio_set_dev(bio, bdev);
2849 bio->bi_iter.bi_sector = first_byte >> 9;
2850 btrfs_io_bio_init(btrfs_io_bio(bio));
2854 struct bio *btrfs_bio_clone(struct bio *bio)
2856 struct btrfs_io_bio *btrfs_bio;
2859 /* Bio allocation backed by a bioset does not fail */
2860 new = bio_clone_fast(bio, GFP_NOFS, &btrfs_bioset);
2861 btrfs_bio = btrfs_io_bio(new);
2862 btrfs_io_bio_init(btrfs_bio);
2863 btrfs_bio->iter = bio->bi_iter;
2867 struct bio *btrfs_io_bio_alloc(unsigned int nr_iovecs)
2871 /* Bio allocation backed by a bioset does not fail */
2872 bio = bio_alloc_bioset(GFP_NOFS, nr_iovecs, &btrfs_bioset);
2873 btrfs_io_bio_init(btrfs_io_bio(bio));
2877 struct bio *btrfs_bio_clone_partial(struct bio *orig, int offset, int size)
2880 struct btrfs_io_bio *btrfs_bio;
2882 /* this will never fail when it's backed by a bioset */
2883 bio = bio_clone_fast(orig, GFP_NOFS, &btrfs_bioset);
2886 btrfs_bio = btrfs_io_bio(bio);
2887 btrfs_io_bio_init(btrfs_bio);
2889 bio_trim(bio, offset >> 9, size >> 9);
2890 btrfs_bio->iter = bio->bi_iter;
2895 * @opf: bio REQ_OP_* and REQ_* flags as one value
2896 * @tree: tree so we can call our merge_bio hook
2897 * @wbc: optional writeback control for io accounting
2898 * @page: page to add to the bio
2899 * @pg_offset: offset of the new bio or to check whether we are adding
2900 * a contiguous page to the previous one
2901 * @size: portion of page that we want to write
2902 * @offset: starting offset in the page
2903 * @bdev: attach newly created bios to this bdev
2904 * @bio_ret: must be valid pointer, newly allocated bio will be stored there
2905 * @end_io_func: end_io callback for new bio
2906 * @mirror_num: desired mirror to read/write
2907 * @prev_bio_flags: flags of previous bio to see if we can merge the current one
2908 * @bio_flags: flags of the current bio to see if we can merge them
2910 static int submit_extent_page(unsigned int opf, struct extent_io_tree *tree,
2911 struct writeback_control *wbc,
2912 struct page *page, u64 offset,
2913 size_t size, unsigned long pg_offset,
2914 struct block_device *bdev,
2915 struct bio **bio_ret,
2916 bio_end_io_t end_io_func,
2918 unsigned long prev_bio_flags,
2919 unsigned long bio_flags,
2920 bool force_bio_submit)
2924 size_t page_size = min_t(size_t, size, PAGE_SIZE);
2925 sector_t sector = offset >> 9;
2931 bool can_merge = true;
2934 if (prev_bio_flags & EXTENT_BIO_COMPRESSED)
2935 contig = bio->bi_iter.bi_sector == sector;
2937 contig = bio_end_sector(bio) == sector;
2940 if (btrfs_bio_fits_in_stripe(page, page_size, bio, bio_flags))
2943 if (prev_bio_flags != bio_flags || !contig || !can_merge ||
2945 bio_add_page(bio, page, page_size, pg_offset) < page_size) {
2946 ret = submit_one_bio(bio, mirror_num, prev_bio_flags);
2954 wbc_account_io(wbc, page, page_size);
2959 bio = btrfs_bio_alloc(bdev, offset);
2960 bio_add_page(bio, page, page_size, pg_offset);
2961 bio->bi_end_io = end_io_func;
2962 bio->bi_private = tree;
2963 bio->bi_write_hint = page->mapping->host->i_write_hint;
2966 wbc_init_bio(wbc, bio);
2967 wbc_account_io(wbc, page, page_size);
2975 static void attach_extent_buffer_page(struct extent_buffer *eb,
2978 if (!PagePrivate(page)) {
2979 SetPagePrivate(page);
2981 set_page_private(page, (unsigned long)eb);
2983 WARN_ON(page->private != (unsigned long)eb);
2987 void set_page_extent_mapped(struct page *page)
2989 if (!PagePrivate(page)) {
2990 SetPagePrivate(page);
2992 set_page_private(page, EXTENT_PAGE_PRIVATE);
2996 static struct extent_map *
2997 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
2998 u64 start, u64 len, get_extent_t *get_extent,
2999 struct extent_map **em_cached)
3001 struct extent_map *em;
3003 if (em_cached && *em_cached) {
3005 if (extent_map_in_tree(em) && start >= em->start &&
3006 start < extent_map_end(em)) {
3007 refcount_inc(&em->refs);
3011 free_extent_map(em);
3015 em = get_extent(BTRFS_I(inode), page, pg_offset, start, len, 0);
3016 if (em_cached && !IS_ERR_OR_NULL(em)) {
3018 refcount_inc(&em->refs);
3024 * basic readpage implementation. Locked extent state structs are inserted
3025 * into the tree that are removed when the IO is done (by the end_io
3027 * XXX JDM: This needs looking at to ensure proper page locking
3028 * return 0 on success, otherwise return error
3030 static int __do_readpage(struct extent_io_tree *tree,
3032 get_extent_t *get_extent,
3033 struct extent_map **em_cached,
3034 struct bio **bio, int mirror_num,
3035 unsigned long *bio_flags, unsigned int read_flags,
3038 struct inode *inode = page->mapping->host;
3039 u64 start = page_offset(page);
3040 const u64 end = start + PAGE_SIZE - 1;
3043 u64 last_byte = i_size_read(inode);
3046 struct extent_map *em;
3047 struct block_device *bdev;
3050 size_t pg_offset = 0;
3052 size_t disk_io_size;
3053 size_t blocksize = inode->i_sb->s_blocksize;
3054 unsigned long this_bio_flag = 0;
3056 set_page_extent_mapped(page);
3058 if (!PageUptodate(page)) {
3059 if (cleancache_get_page(page) == 0) {
3060 BUG_ON(blocksize != PAGE_SIZE);
3061 unlock_extent(tree, start, end);
3066 if (page->index == last_byte >> PAGE_SHIFT) {
3068 size_t zero_offset = offset_in_page(last_byte);
3071 iosize = PAGE_SIZE - zero_offset;
3072 userpage = kmap_atomic(page);
3073 memset(userpage + zero_offset, 0, iosize);
3074 flush_dcache_page(page);
3075 kunmap_atomic(userpage);
3078 while (cur <= end) {
3079 bool force_bio_submit = false;
3082 if (cur >= last_byte) {
3084 struct extent_state *cached = NULL;
3086 iosize = PAGE_SIZE - pg_offset;
3087 userpage = kmap_atomic(page);
3088 memset(userpage + pg_offset, 0, iosize);
3089 flush_dcache_page(page);
3090 kunmap_atomic(userpage);
3091 set_extent_uptodate(tree, cur, cur + iosize - 1,
3093 unlock_extent_cached(tree, cur,
3094 cur + iosize - 1, &cached);
3097 em = __get_extent_map(inode, page, pg_offset, cur,
3098 end - cur + 1, get_extent, em_cached);
3099 if (IS_ERR_OR_NULL(em)) {
3101 unlock_extent(tree, cur, end);
3104 extent_offset = cur - em->start;
3105 BUG_ON(extent_map_end(em) <= cur);
3108 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
3109 this_bio_flag |= EXTENT_BIO_COMPRESSED;
3110 extent_set_compress_type(&this_bio_flag,
3114 iosize = min(extent_map_end(em) - cur, end - cur + 1);
3115 cur_end = min(extent_map_end(em) - 1, end);
3116 iosize = ALIGN(iosize, blocksize);
3117 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
3118 disk_io_size = em->block_len;
3119 offset = em->block_start;
3121 offset = em->block_start + extent_offset;
3122 disk_io_size = iosize;
3125 block_start = em->block_start;
3126 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
3127 block_start = EXTENT_MAP_HOLE;
3130 * If we have a file range that points to a compressed extent
3131 * and it's followed by a consecutive file range that points to
3132 * to the same compressed extent (possibly with a different
3133 * offset and/or length, so it either points to the whole extent
3134 * or only part of it), we must make sure we do not submit a
3135 * single bio to populate the pages for the 2 ranges because
3136 * this makes the compressed extent read zero out the pages
3137 * belonging to the 2nd range. Imagine the following scenario:
3140 * [0 - 8K] [8K - 24K]
3143 * points to extent X, points to extent X,
3144 * offset 4K, length of 8K offset 0, length 16K
3146 * [extent X, compressed length = 4K uncompressed length = 16K]
3148 * If the bio to read the compressed extent covers both ranges,
3149 * it will decompress extent X into the pages belonging to the
3150 * first range and then it will stop, zeroing out the remaining
3151 * pages that belong to the other range that points to extent X.
3152 * So here we make sure we submit 2 bios, one for the first
3153 * range and another one for the third range. Both will target
3154 * the same physical extent from disk, but we can't currently
3155 * make the compressed bio endio callback populate the pages
3156 * for both ranges because each compressed bio is tightly
3157 * coupled with a single extent map, and each range can have
3158 * an extent map with a different offset value relative to the
3159 * uncompressed data of our extent and different lengths. This
3160 * is a corner case so we prioritize correctness over
3161 * non-optimal behavior (submitting 2 bios for the same extent).
3163 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) &&
3164 prev_em_start && *prev_em_start != (u64)-1 &&
3165 *prev_em_start != em->start)
3166 force_bio_submit = true;
3169 *prev_em_start = em->start;
3171 free_extent_map(em);
3174 /* we've found a hole, just zero and go on */
3175 if (block_start == EXTENT_MAP_HOLE) {
3177 struct extent_state *cached = NULL;
3179 userpage = kmap_atomic(page);
3180 memset(userpage + pg_offset, 0, iosize);
3181 flush_dcache_page(page);
3182 kunmap_atomic(userpage);
3184 set_extent_uptodate(tree, cur, cur + iosize - 1,
3186 unlock_extent_cached(tree, cur,
3187 cur + iosize - 1, &cached);
3189 pg_offset += iosize;
3192 /* the get_extent function already copied into the page */
3193 if (test_range_bit(tree, cur, cur_end,
3194 EXTENT_UPTODATE, 1, NULL)) {
3195 check_page_uptodate(tree, page);
3196 unlock_extent(tree, cur, cur + iosize - 1);
3198 pg_offset += iosize;
3201 /* we have an inline extent but it didn't get marked up
3202 * to date. Error out
3204 if (block_start == EXTENT_MAP_INLINE) {
3206 unlock_extent(tree, cur, cur + iosize - 1);
3208 pg_offset += iosize;
3212 ret = submit_extent_page(REQ_OP_READ | read_flags, tree, NULL,
3213 page, offset, disk_io_size,
3214 pg_offset, bdev, bio,
3215 end_bio_extent_readpage, mirror_num,
3221 *bio_flags = this_bio_flag;
3224 unlock_extent(tree, cur, cur + iosize - 1);
3228 pg_offset += iosize;
3232 if (!PageError(page))
3233 SetPageUptodate(page);
3239 static inline void contiguous_readpages(struct extent_io_tree *tree,
3240 struct page *pages[], int nr_pages,
3242 struct extent_map **em_cached,
3244 unsigned long *bio_flags,
3247 struct btrfs_inode *inode = BTRFS_I(pages[0]->mapping->host);
3250 btrfs_lock_and_flush_ordered_range(tree, inode, start, end, NULL);
3252 for (index = 0; index < nr_pages; index++) {
3253 __do_readpage(tree, pages[index], btrfs_get_extent, em_cached,
3254 bio, 0, bio_flags, REQ_RAHEAD, prev_em_start);
3255 put_page(pages[index]);
3259 static int __extent_read_full_page(struct extent_io_tree *tree,
3261 get_extent_t *get_extent,
3262 struct bio **bio, int mirror_num,
3263 unsigned long *bio_flags,
3264 unsigned int read_flags)
3266 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
3267 u64 start = page_offset(page);
3268 u64 end = start + PAGE_SIZE - 1;
3271 btrfs_lock_and_flush_ordered_range(tree, inode, start, end, NULL);
3273 ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num,
3274 bio_flags, read_flags, NULL);
3278 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
3279 get_extent_t *get_extent, int mirror_num)
3281 struct bio *bio = NULL;
3282 unsigned long bio_flags = 0;
3285 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
3288 ret = submit_one_bio(bio, mirror_num, bio_flags);
3292 static void update_nr_written(struct writeback_control *wbc,
3293 unsigned long nr_written)
3295 wbc->nr_to_write -= nr_written;
3299 * helper for __extent_writepage, doing all of the delayed allocation setup.
3301 * This returns 1 if btrfs_run_delalloc_range function did all the work required
3302 * to write the page (copy into inline extent). In this case the IO has
3303 * been started and the page is already unlocked.
3305 * This returns 0 if all went well (page still locked)
3306 * This returns < 0 if there were errors (page still locked)
3308 static noinline_for_stack int writepage_delalloc(struct inode *inode,
3309 struct page *page, struct writeback_control *wbc,
3310 u64 delalloc_start, unsigned long *nr_written)
3312 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
3313 u64 page_end = delalloc_start + PAGE_SIZE - 1;
3315 u64 delalloc_to_write = 0;
3316 u64 delalloc_end = 0;
3318 int page_started = 0;
3321 while (delalloc_end < page_end) {
3322 found = find_lock_delalloc_range(inode, tree,
3327 delalloc_start = delalloc_end + 1;
3330 ret = btrfs_run_delalloc_range(inode, page, delalloc_start,
3331 delalloc_end, &page_started, nr_written, wbc);
3335 * btrfs_run_delalloc_range should return < 0 for error
3336 * but just in case, we use > 0 here meaning the IO is
3337 * started, so we don't want to return > 0 unless
3338 * things are going well.
3340 ret = ret < 0 ? ret : -EIO;
3344 * delalloc_end is already one less than the total length, so
3345 * we don't subtract one from PAGE_SIZE
3347 delalloc_to_write += (delalloc_end - delalloc_start +
3348 PAGE_SIZE) >> PAGE_SHIFT;
3349 delalloc_start = delalloc_end + 1;
3351 if (wbc->nr_to_write < delalloc_to_write) {
3354 if (delalloc_to_write < thresh * 2)
3355 thresh = delalloc_to_write;
3356 wbc->nr_to_write = min_t(u64, delalloc_to_write,
3360 /* did the fill delalloc function already unlock and start
3365 * we've unlocked the page, so we can't update
3366 * the mapping's writeback index, just update
3369 wbc->nr_to_write -= *nr_written;
3380 * helper for __extent_writepage. This calls the writepage start hooks,
3381 * and does the loop to map the page into extents and bios.
3383 * We return 1 if the IO is started and the page is unlocked,
3384 * 0 if all went well (page still locked)
3385 * < 0 if there were errors (page still locked)
3387 static noinline_for_stack int __extent_writepage_io(struct inode *inode,
3389 struct writeback_control *wbc,
3390 struct extent_page_data *epd,
3392 unsigned long nr_written,
3393 unsigned int write_flags, int *nr_ret)
3395 struct extent_io_tree *tree = epd->tree;
3396 u64 start = page_offset(page);
3397 u64 page_end = start + PAGE_SIZE - 1;
3403 struct extent_map *em;
3404 struct block_device *bdev;
3405 size_t pg_offset = 0;
3411 ret = btrfs_writepage_cow_fixup(page, start, page_end);
3413 /* Fixup worker will requeue */
3415 wbc->pages_skipped++;
3417 redirty_page_for_writepage(wbc, page);
3419 update_nr_written(wbc, nr_written);
3425 * we don't want to touch the inode after unlocking the page,
3426 * so we update the mapping writeback index now
3428 update_nr_written(wbc, nr_written + 1);
3431 if (i_size <= start) {
3432 btrfs_writepage_endio_finish_ordered(page, start, page_end, 1);
3436 blocksize = inode->i_sb->s_blocksize;
3438 while (cur <= end) {
3442 if (cur >= i_size) {
3443 btrfs_writepage_endio_finish_ordered(page, cur,
3447 em = btrfs_get_extent(BTRFS_I(inode), page, pg_offset, cur,
3449 if (IS_ERR_OR_NULL(em)) {
3451 ret = PTR_ERR_OR_ZERO(em);
3455 extent_offset = cur - em->start;
3456 em_end = extent_map_end(em);
3457 BUG_ON(em_end <= cur);
3459 iosize = min(em_end - cur, end - cur + 1);
3460 iosize = ALIGN(iosize, blocksize);
3461 offset = em->block_start + extent_offset;
3463 block_start = em->block_start;
3464 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3465 free_extent_map(em);
3469 * compressed and inline extents are written through other
3472 if (compressed || block_start == EXTENT_MAP_HOLE ||
3473 block_start == EXTENT_MAP_INLINE) {
3475 * end_io notification does not happen here for
3476 * compressed extents
3479 btrfs_writepage_endio_finish_ordered(page, cur,
3482 else if (compressed) {
3483 /* we don't want to end_page_writeback on
3484 * a compressed extent. this happens
3491 pg_offset += iosize;
3495 btrfs_set_range_writeback(tree, cur, cur + iosize - 1);
3496 if (!PageWriteback(page)) {
3497 btrfs_err(BTRFS_I(inode)->root->fs_info,
3498 "page %lu not writeback, cur %llu end %llu",
3499 page->index, cur, end);
3502 ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc,
3503 page, offset, iosize, pg_offset,
3505 end_bio_extent_writepage,
3509 if (PageWriteback(page))
3510 end_page_writeback(page);
3514 pg_offset += iosize;
3523 * the writepage semantics are similar to regular writepage. extent
3524 * records are inserted to lock ranges in the tree, and as dirty areas
3525 * are found, they are marked writeback. Then the lock bits are removed
3526 * and the end_io handler clears the writeback ranges
3528 * Return 0 if everything goes well.
3529 * Return <0 for error.
3531 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
3532 struct extent_page_data *epd)
3534 struct inode *inode = page->mapping->host;
3535 u64 start = page_offset(page);
3536 u64 page_end = start + PAGE_SIZE - 1;
3539 size_t pg_offset = 0;
3540 loff_t i_size = i_size_read(inode);
3541 unsigned long end_index = i_size >> PAGE_SHIFT;
3542 unsigned int write_flags = 0;
3543 unsigned long nr_written = 0;
3545 write_flags = wbc_to_write_flags(wbc);
3547 trace___extent_writepage(page, inode, wbc);
3549 WARN_ON(!PageLocked(page));
3551 ClearPageError(page);
3553 pg_offset = offset_in_page(i_size);
3554 if (page->index > end_index ||
3555 (page->index == end_index && !pg_offset)) {
3556 page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
3561 if (page->index == end_index) {
3564 userpage = kmap_atomic(page);
3565 memset(userpage + pg_offset, 0,
3566 PAGE_SIZE - pg_offset);
3567 kunmap_atomic(userpage);
3568 flush_dcache_page(page);
3573 set_page_extent_mapped(page);
3575 if (!epd->extent_locked) {
3576 ret = writepage_delalloc(inode, page, wbc, start, &nr_written);
3583 ret = __extent_writepage_io(inode, page, wbc, epd,
3584 i_size, nr_written, write_flags, &nr);
3590 /* make sure the mapping tag for page dirty gets cleared */
3591 set_page_writeback(page);
3592 end_page_writeback(page);
3594 if (PageError(page)) {
3595 ret = ret < 0 ? ret : -EIO;
3596 end_extent_writepage(page, ret, start, page_end);
3606 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3608 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
3609 TASK_UNINTERRUPTIBLE);
3613 * Lock eb pages and flush the bio if we can't the locks
3615 * Return 0 if nothing went wrong
3616 * Return >0 is same as 0, except bio is not submitted
3617 * Return <0 if something went wrong, no page is locked
3619 static noinline_for_stack int lock_extent_buffer_for_io(struct extent_buffer *eb,
3620 struct extent_page_data *epd)
3622 struct btrfs_fs_info *fs_info = eb->fs_info;
3623 int i, num_pages, failed_page_nr;
3627 if (!btrfs_try_tree_write_lock(eb)) {
3628 ret = flush_write_bio(epd);
3632 btrfs_tree_lock(eb);
3635 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3636 btrfs_tree_unlock(eb);
3640 ret = flush_write_bio(epd);
3646 wait_on_extent_buffer_writeback(eb);
3647 btrfs_tree_lock(eb);
3648 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3650 btrfs_tree_unlock(eb);
3655 * We need to do this to prevent races in people who check if the eb is
3656 * under IO since we can end up having no IO bits set for a short period
3659 spin_lock(&eb->refs_lock);
3660 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3661 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3662 spin_unlock(&eb->refs_lock);
3663 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3664 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
3666 fs_info->dirty_metadata_batch);
3669 spin_unlock(&eb->refs_lock);
3672 btrfs_tree_unlock(eb);
3677 num_pages = num_extent_pages(eb);
3678 for (i = 0; i < num_pages; i++) {
3679 struct page *p = eb->pages[i];
3681 if (!trylock_page(p)) {
3683 ret = flush_write_bio(epd);
3696 /* Unlock already locked pages */
3697 for (i = 0; i < failed_page_nr; i++)
3698 unlock_page(eb->pages[i]);
3702 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3704 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3705 smp_mb__after_atomic();
3706 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3709 static void set_btree_ioerr(struct page *page)
3711 struct extent_buffer *eb = (struct extent_buffer *)page->private;
3714 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))
3718 * If writeback for a btree extent that doesn't belong to a log tree
3719 * failed, increment the counter transaction->eb_write_errors.
3720 * We do this because while the transaction is running and before it's
3721 * committing (when we call filemap_fdata[write|wait]_range against
3722 * the btree inode), we might have
3723 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
3724 * returns an error or an error happens during writeback, when we're
3725 * committing the transaction we wouldn't know about it, since the pages
3726 * can be no longer dirty nor marked anymore for writeback (if a
3727 * subsequent modification to the extent buffer didn't happen before the
3728 * transaction commit), which makes filemap_fdata[write|wait]_range not
3729 * able to find the pages tagged with SetPageError at transaction
3730 * commit time. So if this happens we must abort the transaction,
3731 * otherwise we commit a super block with btree roots that point to
3732 * btree nodes/leafs whose content on disk is invalid - either garbage
3733 * or the content of some node/leaf from a past generation that got
3734 * cowed or deleted and is no longer valid.
3736 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
3737 * not be enough - we need to distinguish between log tree extents vs
3738 * non-log tree extents, and the next filemap_fdatawait_range() call
3739 * will catch and clear such errors in the mapping - and that call might
3740 * be from a log sync and not from a transaction commit. Also, checking
3741 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
3742 * not done and would not be reliable - the eb might have been released
3743 * from memory and reading it back again means that flag would not be
3744 * set (since it's a runtime flag, not persisted on disk).
3746 * Using the flags below in the btree inode also makes us achieve the
3747 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
3748 * writeback for all dirty pages and before filemap_fdatawait_range()
3749 * is called, the writeback for all dirty pages had already finished
3750 * with errors - because we were not using AS_EIO/AS_ENOSPC,
3751 * filemap_fdatawait_range() would return success, as it could not know
3752 * that writeback errors happened (the pages were no longer tagged for
3755 switch (eb->log_index) {
3757 set_bit(BTRFS_FS_BTREE_ERR, &eb->fs_info->flags);
3760 set_bit(BTRFS_FS_LOG1_ERR, &eb->fs_info->flags);
3763 set_bit(BTRFS_FS_LOG2_ERR, &eb->fs_info->flags);
3766 BUG(); /* unexpected, logic error */
3770 static void end_bio_extent_buffer_writepage(struct bio *bio)
3772 struct bio_vec *bvec;
3773 struct extent_buffer *eb;
3775 struct bvec_iter_all iter_all;
3777 ASSERT(!bio_flagged(bio, BIO_CLONED));
3778 bio_for_each_segment_all(bvec, bio, iter_all) {
3779 struct page *page = bvec->bv_page;
3781 eb = (struct extent_buffer *)page->private;
3783 done = atomic_dec_and_test(&eb->io_pages);
3785 if (bio->bi_status ||
3786 test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) {
3787 ClearPageUptodate(page);
3788 set_btree_ioerr(page);
3791 end_page_writeback(page);
3796 end_extent_buffer_writeback(eb);
3802 static noinline_for_stack int write_one_eb(struct extent_buffer *eb,
3803 struct writeback_control *wbc,
3804 struct extent_page_data *epd)
3806 struct btrfs_fs_info *fs_info = eb->fs_info;
3807 struct block_device *bdev = fs_info->fs_devices->latest_bdev;
3808 struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
3809 u64 offset = eb->start;
3812 unsigned long start, end;
3813 unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META;
3816 clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
3817 num_pages = num_extent_pages(eb);
3818 atomic_set(&eb->io_pages, num_pages);
3820 /* set btree blocks beyond nritems with 0 to avoid stale content. */
3821 nritems = btrfs_header_nritems(eb);
3822 if (btrfs_header_level(eb) > 0) {
3823 end = btrfs_node_key_ptr_offset(nritems);
3825 memzero_extent_buffer(eb, end, eb->len - end);
3829 * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0
3831 start = btrfs_item_nr_offset(nritems);
3832 end = BTRFS_LEAF_DATA_OFFSET + leaf_data_end(eb);
3833 memzero_extent_buffer(eb, start, end - start);
3836 for (i = 0; i < num_pages; i++) {
3837 struct page *p = eb->pages[i];
3839 clear_page_dirty_for_io(p);
3840 set_page_writeback(p);
3841 ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc,
3842 p, offset, PAGE_SIZE, 0, bdev,
3844 end_bio_extent_buffer_writepage,
3848 if (PageWriteback(p))
3849 end_page_writeback(p);
3850 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3851 end_extent_buffer_writeback(eb);
3855 offset += PAGE_SIZE;
3856 update_nr_written(wbc, 1);
3860 if (unlikely(ret)) {
3861 for (; i < num_pages; i++) {
3862 struct page *p = eb->pages[i];
3863 clear_page_dirty_for_io(p);
3871 int btree_write_cache_pages(struct address_space *mapping,
3872 struct writeback_control *wbc)
3874 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
3875 struct extent_buffer *eb, *prev_eb = NULL;
3876 struct extent_page_data epd = {
3880 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3884 int nr_to_write_done = 0;
3885 struct pagevec pvec;
3888 pgoff_t end; /* Inclusive */
3892 pagevec_init(&pvec);
3893 if (wbc->range_cyclic) {
3894 index = mapping->writeback_index; /* Start from prev offset */
3897 index = wbc->range_start >> PAGE_SHIFT;
3898 end = wbc->range_end >> PAGE_SHIFT;
3901 if (wbc->sync_mode == WB_SYNC_ALL)
3902 tag = PAGECACHE_TAG_TOWRITE;
3904 tag = PAGECACHE_TAG_DIRTY;
3906 if (wbc->sync_mode == WB_SYNC_ALL)
3907 tag_pages_for_writeback(mapping, index, end);
3908 while (!done && !nr_to_write_done && (index <= end) &&
3909 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
3914 for (i = 0; i < nr_pages; i++) {
3915 struct page *page = pvec.pages[i];
3917 if (!PagePrivate(page))
3920 spin_lock(&mapping->private_lock);
3921 if (!PagePrivate(page)) {
3922 spin_unlock(&mapping->private_lock);
3926 eb = (struct extent_buffer *)page->private;
3929 * Shouldn't happen and normally this would be a BUG_ON
3930 * but no sense in crashing the users box for something
3931 * we can survive anyway.
3934 spin_unlock(&mapping->private_lock);
3938 if (eb == prev_eb) {
3939 spin_unlock(&mapping->private_lock);
3943 ret = atomic_inc_not_zero(&eb->refs);
3944 spin_unlock(&mapping->private_lock);
3949 ret = lock_extent_buffer_for_io(eb, &epd);
3951 free_extent_buffer(eb);
3955 ret = write_one_eb(eb, wbc, &epd);
3958 free_extent_buffer(eb);
3961 free_extent_buffer(eb);
3964 * the filesystem may choose to bump up nr_to_write.
3965 * We have to make sure to honor the new nr_to_write
3968 nr_to_write_done = wbc->nr_to_write <= 0;
3970 pagevec_release(&pvec);
3973 if (!scanned && !done) {
3975 * We hit the last page and there is more work to be done: wrap
3976 * back to the start of the file
3984 end_write_bio(&epd, ret);
3987 ret = flush_write_bio(&epd);
3992 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3993 * @mapping: address space structure to write
3994 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3995 * @data: data passed to __extent_writepage function
3997 * If a page is already under I/O, write_cache_pages() skips it, even
3998 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3999 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
4000 * and msync() need to guarantee that all the data which was dirty at the time
4001 * the call was made get new I/O started against them. If wbc->sync_mode is
4002 * WB_SYNC_ALL then we were called for data integrity and we must wait for
4003 * existing IO to complete.
4005 static int extent_write_cache_pages(struct address_space *mapping,
4006 struct writeback_control *wbc,
4007 struct extent_page_data *epd)
4009 struct inode *inode = mapping->host;
4012 int nr_to_write_done = 0;
4013 struct pagevec pvec;
4016 pgoff_t end; /* Inclusive */
4018 int range_whole = 0;
4023 * We have to hold onto the inode so that ordered extents can do their
4024 * work when the IO finishes. The alternative to this is failing to add
4025 * an ordered extent if the igrab() fails there and that is a huge pain
4026 * to deal with, so instead just hold onto the inode throughout the
4027 * writepages operation. If it fails here we are freeing up the inode
4028 * anyway and we'd rather not waste our time writing out stuff that is
4029 * going to be truncated anyway.
4034 pagevec_init(&pvec);
4035 if (wbc->range_cyclic) {
4036 index = mapping->writeback_index; /* Start from prev offset */
4039 index = wbc->range_start >> PAGE_SHIFT;
4040 end = wbc->range_end >> PAGE_SHIFT;
4041 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
4047 * We do the tagged writepage as long as the snapshot flush bit is set
4048 * and we are the first one who do the filemap_flush() on this inode.
4050 * The nr_to_write == LONG_MAX is needed to make sure other flushers do
4051 * not race in and drop the bit.
4053 if (range_whole && wbc->nr_to_write == LONG_MAX &&
4054 test_and_clear_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
4055 &BTRFS_I(inode)->runtime_flags))
4056 wbc->tagged_writepages = 1;
4058 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
4059 tag = PAGECACHE_TAG_TOWRITE;
4061 tag = PAGECACHE_TAG_DIRTY;
4063 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
4064 tag_pages_for_writeback(mapping, index, end);
4066 while (!done && !nr_to_write_done && (index <= end) &&
4067 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping,
4068 &index, end, tag))) {
4072 for (i = 0; i < nr_pages; i++) {
4073 struct page *page = pvec.pages[i];
4075 done_index = page->index;
4077 * At this point we hold neither the i_pages lock nor
4078 * the page lock: the page may be truncated or
4079 * invalidated (changing page->mapping to NULL),
4080 * or even swizzled back from swapper_space to
4081 * tmpfs file mapping
4083 if (!trylock_page(page)) {
4084 ret = flush_write_bio(epd);
4089 if (unlikely(page->mapping != mapping)) {
4094 if (wbc->sync_mode != WB_SYNC_NONE) {
4095 if (PageWriteback(page)) {
4096 ret = flush_write_bio(epd);
4099 wait_on_page_writeback(page);
4102 if (PageWriteback(page) ||
4103 !clear_page_dirty_for_io(page)) {
4108 ret = __extent_writepage(page, wbc, epd);
4111 * done_index is set past this page,
4112 * so media errors will not choke
4113 * background writeout for the entire
4114 * file. This has consequences for
4115 * range_cyclic semantics (ie. it may
4116 * not be suitable for data integrity
4119 done_index = page->index + 1;
4125 * the filesystem may choose to bump up nr_to_write.
4126 * We have to make sure to honor the new nr_to_write
4129 nr_to_write_done = wbc->nr_to_write <= 0;
4131 pagevec_release(&pvec);
4134 if (!scanned && !done) {
4136 * We hit the last page and there is more work to be done: wrap
4137 * back to the start of the file
4144 if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole))
4145 mapping->writeback_index = done_index;
4147 btrfs_add_delayed_iput(inode);
4151 int extent_write_full_page(struct page *page, struct writeback_control *wbc)
4154 struct extent_page_data epd = {
4156 .tree = &BTRFS_I(page->mapping->host)->io_tree,
4158 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4161 ret = __extent_writepage(page, wbc, &epd);
4164 end_write_bio(&epd, ret);
4168 ret = flush_write_bio(&epd);
4173 int extent_write_locked_range(struct inode *inode, u64 start, u64 end,
4177 struct address_space *mapping = inode->i_mapping;
4178 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
4180 unsigned long nr_pages = (end - start + PAGE_SIZE) >>
4183 struct extent_page_data epd = {
4187 .sync_io = mode == WB_SYNC_ALL,
4189 struct writeback_control wbc_writepages = {
4191 .nr_to_write = nr_pages * 2,
4192 .range_start = start,
4193 .range_end = end + 1,
4196 while (start <= end) {
4197 page = find_get_page(mapping, start >> PAGE_SHIFT);
4198 if (clear_page_dirty_for_io(page))
4199 ret = __extent_writepage(page, &wbc_writepages, &epd);
4201 btrfs_writepage_endio_finish_ordered(page, start,
4202 start + PAGE_SIZE - 1, 1);
4211 end_write_bio(&epd, ret);
4214 ret = flush_write_bio(&epd);
4218 int extent_writepages(struct address_space *mapping,
4219 struct writeback_control *wbc)
4222 struct extent_page_data epd = {
4224 .tree = &BTRFS_I(mapping->host)->io_tree,
4226 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4229 ret = extent_write_cache_pages(mapping, wbc, &epd);
4232 end_write_bio(&epd, ret);
4235 ret = flush_write_bio(&epd);
4239 int extent_readpages(struct address_space *mapping, struct list_head *pages,
4242 struct bio *bio = NULL;
4243 unsigned long bio_flags = 0;
4244 struct page *pagepool[16];
4245 struct extent_map *em_cached = NULL;
4246 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
4248 u64 prev_em_start = (u64)-1;
4250 while (!list_empty(pages)) {
4253 for (nr = 0; nr < ARRAY_SIZE(pagepool) && !list_empty(pages);) {
4254 struct page *page = lru_to_page(pages);
4256 prefetchw(&page->flags);
4257 list_del(&page->lru);
4258 if (add_to_page_cache_lru(page, mapping, page->index,
4259 readahead_gfp_mask(mapping))) {
4264 pagepool[nr++] = page;
4265 contig_end = page_offset(page) + PAGE_SIZE - 1;
4269 u64 contig_start = page_offset(pagepool[0]);
4271 ASSERT(contig_start + nr * PAGE_SIZE - 1 == contig_end);
4273 contiguous_readpages(tree, pagepool, nr, contig_start,
4274 contig_end, &em_cached, &bio, &bio_flags,
4280 free_extent_map(em_cached);
4283 return submit_one_bio(bio, 0, bio_flags);
4288 * basic invalidatepage code, this waits on any locked or writeback
4289 * ranges corresponding to the page, and then deletes any extent state
4290 * records from the tree
4292 int extent_invalidatepage(struct extent_io_tree *tree,
4293 struct page *page, unsigned long offset)
4295 struct extent_state *cached_state = NULL;
4296 u64 start = page_offset(page);
4297 u64 end = start + PAGE_SIZE - 1;
4298 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
4300 start += ALIGN(offset, blocksize);
4304 lock_extent_bits(tree, start, end, &cached_state);
4305 wait_on_page_writeback(page);
4306 clear_extent_bit(tree, start, end,
4307 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4308 EXTENT_DO_ACCOUNTING,
4309 1, 1, &cached_state);
4314 * a helper for releasepage, this tests for areas of the page that
4315 * are locked or under IO and drops the related state bits if it is safe
4318 static int try_release_extent_state(struct extent_io_tree *tree,
4319 struct page *page, gfp_t mask)
4321 u64 start = page_offset(page);
4322 u64 end = start + PAGE_SIZE - 1;
4325 if (test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL)) {
4329 * at this point we can safely clear everything except the
4330 * locked bit and the nodatasum bit
4332 ret = __clear_extent_bit(tree, start, end,
4333 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
4334 0, 0, NULL, mask, NULL);
4336 /* if clear_extent_bit failed for enomem reasons,
4337 * we can't allow the release to continue.
4348 * a helper for releasepage. As long as there are no locked extents
4349 * in the range corresponding to the page, both state records and extent
4350 * map records are removed
4352 int try_release_extent_mapping(struct page *page, gfp_t mask)
4354 struct extent_map *em;
4355 u64 start = page_offset(page);
4356 u64 end = start + PAGE_SIZE - 1;
4357 struct btrfs_inode *btrfs_inode = BTRFS_I(page->mapping->host);
4358 struct extent_io_tree *tree = &btrfs_inode->io_tree;
4359 struct extent_map_tree *map = &btrfs_inode->extent_tree;
4361 if (gfpflags_allow_blocking(mask) &&
4362 page->mapping->host->i_size > SZ_16M) {
4364 while (start <= end) {
4365 len = end - start + 1;
4366 write_lock(&map->lock);
4367 em = lookup_extent_mapping(map, start, len);
4369 write_unlock(&map->lock);
4372 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
4373 em->start != start) {
4374 write_unlock(&map->lock);
4375 free_extent_map(em);
4378 if (!test_range_bit(tree, em->start,
4379 extent_map_end(em) - 1,
4380 EXTENT_LOCKED, 0, NULL)) {
4381 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4382 &btrfs_inode->runtime_flags);
4383 remove_extent_mapping(map, em);
4384 /* once for the rb tree */
4385 free_extent_map(em);
4387 start = extent_map_end(em);
4388 write_unlock(&map->lock);
4391 free_extent_map(em);
4394 return try_release_extent_state(tree, page, mask);
4398 * helper function for fiemap, which doesn't want to see any holes.
4399 * This maps until we find something past 'last'
4401 static struct extent_map *get_extent_skip_holes(struct inode *inode,
4402 u64 offset, u64 last)
4404 u64 sectorsize = btrfs_inode_sectorsize(inode);
4405 struct extent_map *em;
4412 len = last - offset;
4415 len = ALIGN(len, sectorsize);
4416 em = btrfs_get_extent_fiemap(BTRFS_I(inode), offset, len);
4417 if (IS_ERR_OR_NULL(em))
4420 /* if this isn't a hole return it */
4421 if (em->block_start != EXTENT_MAP_HOLE)
4424 /* this is a hole, advance to the next extent */
4425 offset = extent_map_end(em);
4426 free_extent_map(em);
4434 * To cache previous fiemap extent
4436 * Will be used for merging fiemap extent
4438 struct fiemap_cache {
4447 * Helper to submit fiemap extent.
4449 * Will try to merge current fiemap extent specified by @offset, @phys,
4450 * @len and @flags with cached one.
4451 * And only when we fails to merge, cached one will be submitted as
4454 * Return value is the same as fiemap_fill_next_extent().
4456 static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
4457 struct fiemap_cache *cache,
4458 u64 offset, u64 phys, u64 len, u32 flags)
4466 * Sanity check, extent_fiemap() should have ensured that new
4467 * fiemap extent won't overlap with cached one.
4470 * NOTE: Physical address can overlap, due to compression
4472 if (cache->offset + cache->len > offset) {
4478 * Only merges fiemap extents if
4479 * 1) Their logical addresses are continuous
4481 * 2) Their physical addresses are continuous
4482 * So truly compressed (physical size smaller than logical size)
4483 * extents won't get merged with each other
4485 * 3) Share same flags except FIEMAP_EXTENT_LAST
4486 * So regular extent won't get merged with prealloc extent
4488 if (cache->offset + cache->len == offset &&
4489 cache->phys + cache->len == phys &&
4490 (cache->flags & ~FIEMAP_EXTENT_LAST) ==
4491 (flags & ~FIEMAP_EXTENT_LAST)) {
4493 cache->flags |= flags;
4494 goto try_submit_last;
4497 /* Not mergeable, need to submit cached one */
4498 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
4499 cache->len, cache->flags);
4500 cache->cached = false;
4504 cache->cached = true;
4505 cache->offset = offset;
4508 cache->flags = flags;
4510 if (cache->flags & FIEMAP_EXTENT_LAST) {
4511 ret = fiemap_fill_next_extent(fieinfo, cache->offset,
4512 cache->phys, cache->len, cache->flags);
4513 cache->cached = false;
4519 * Emit last fiemap cache
4521 * The last fiemap cache may still be cached in the following case:
4523 * |<- Fiemap range ->|
4524 * |<------------ First extent ----------->|
4526 * In this case, the first extent range will be cached but not emitted.
4527 * So we must emit it before ending extent_fiemap().
4529 static int emit_last_fiemap_cache(struct fiemap_extent_info *fieinfo,
4530 struct fiemap_cache *cache)
4537 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
4538 cache->len, cache->flags);
4539 cache->cached = false;
4545 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4546 __u64 start, __u64 len)
4550 u64 max = start + len;
4554 u64 last_for_get_extent = 0;
4556 u64 isize = i_size_read(inode);
4557 struct btrfs_key found_key;
4558 struct extent_map *em = NULL;
4559 struct extent_state *cached_state = NULL;
4560 struct btrfs_path *path;
4561 struct btrfs_root *root = BTRFS_I(inode)->root;
4562 struct fiemap_cache cache = { 0 };
4563 struct ulist *roots;
4564 struct ulist *tmp_ulist;
4573 path = btrfs_alloc_path();
4576 path->leave_spinning = 1;
4578 roots = ulist_alloc(GFP_KERNEL);
4579 tmp_ulist = ulist_alloc(GFP_KERNEL);
4580 if (!roots || !tmp_ulist) {
4582 goto out_free_ulist;
4585 start = round_down(start, btrfs_inode_sectorsize(inode));
4586 len = round_up(max, btrfs_inode_sectorsize(inode)) - start;
4589 * lookup the last file extent. We're not using i_size here
4590 * because there might be preallocation past i_size
4592 ret = btrfs_lookup_file_extent(NULL, root, path,
4593 btrfs_ino(BTRFS_I(inode)), -1, 0);
4595 btrfs_free_path(path);
4596 goto out_free_ulist;
4604 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
4605 found_type = found_key.type;
4607 /* No extents, but there might be delalloc bits */
4608 if (found_key.objectid != btrfs_ino(BTRFS_I(inode)) ||
4609 found_type != BTRFS_EXTENT_DATA_KEY) {
4610 /* have to trust i_size as the end */
4612 last_for_get_extent = isize;
4615 * remember the start of the last extent. There are a
4616 * bunch of different factors that go into the length of the
4617 * extent, so its much less complex to remember where it started
4619 last = found_key.offset;
4620 last_for_get_extent = last + 1;
4622 btrfs_release_path(path);
4625 * we might have some extents allocated but more delalloc past those
4626 * extents. so, we trust isize unless the start of the last extent is
4631 last_for_get_extent = isize;
4634 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4637 em = get_extent_skip_holes(inode, start, last_for_get_extent);
4646 u64 offset_in_extent = 0;
4648 /* break if the extent we found is outside the range */
4649 if (em->start >= max || extent_map_end(em) < off)
4653 * get_extent may return an extent that starts before our
4654 * requested range. We have to make sure the ranges
4655 * we return to fiemap always move forward and don't
4656 * overlap, so adjust the offsets here
4658 em_start = max(em->start, off);
4661 * record the offset from the start of the extent
4662 * for adjusting the disk offset below. Only do this if the
4663 * extent isn't compressed since our in ram offset may be past
4664 * what we have actually allocated on disk.
4666 if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4667 offset_in_extent = em_start - em->start;
4668 em_end = extent_map_end(em);
4669 em_len = em_end - em_start;
4671 if (em->block_start < EXTENT_MAP_LAST_BYTE)
4672 disko = em->block_start + offset_in_extent;
4677 * bump off for our next call to get_extent
4679 off = extent_map_end(em);
4683 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
4685 flags |= FIEMAP_EXTENT_LAST;
4686 } else if (em->block_start == EXTENT_MAP_INLINE) {
4687 flags |= (FIEMAP_EXTENT_DATA_INLINE |
4688 FIEMAP_EXTENT_NOT_ALIGNED);
4689 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
4690 flags |= (FIEMAP_EXTENT_DELALLOC |
4691 FIEMAP_EXTENT_UNKNOWN);
4692 } else if (fieinfo->fi_extents_max) {
4693 u64 bytenr = em->block_start -
4694 (em->start - em->orig_start);
4697 * As btrfs supports shared space, this information
4698 * can be exported to userspace tools via
4699 * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0
4700 * then we're just getting a count and we can skip the
4703 ret = btrfs_check_shared(root,
4704 btrfs_ino(BTRFS_I(inode)),
4705 bytenr, roots, tmp_ulist);
4709 flags |= FIEMAP_EXTENT_SHARED;
4712 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4713 flags |= FIEMAP_EXTENT_ENCODED;
4714 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4715 flags |= FIEMAP_EXTENT_UNWRITTEN;
4717 free_extent_map(em);
4719 if ((em_start >= last) || em_len == (u64)-1 ||
4720 (last == (u64)-1 && isize <= em_end)) {
4721 flags |= FIEMAP_EXTENT_LAST;
4725 /* now scan forward to see if this is really the last extent. */
4726 em = get_extent_skip_holes(inode, off, last_for_get_extent);
4732 flags |= FIEMAP_EXTENT_LAST;
4735 ret = emit_fiemap_extent(fieinfo, &cache, em_start, disko,
4745 ret = emit_last_fiemap_cache(fieinfo, &cache);
4746 free_extent_map(em);
4748 btrfs_free_path(path);
4749 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4754 ulist_free(tmp_ulist);
4758 static void __free_extent_buffer(struct extent_buffer *eb)
4760 btrfs_leak_debug_del(&eb->leak_list);
4761 kmem_cache_free(extent_buffer_cache, eb);
4764 int extent_buffer_under_io(struct extent_buffer *eb)
4766 return (atomic_read(&eb->io_pages) ||
4767 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4768 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4772 * Release all pages attached to the extent buffer.
4774 static void btrfs_release_extent_buffer_pages(struct extent_buffer *eb)
4778 int mapped = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
4780 BUG_ON(extent_buffer_under_io(eb));
4782 num_pages = num_extent_pages(eb);
4783 for (i = 0; i < num_pages; i++) {
4784 struct page *page = eb->pages[i];
4789 spin_lock(&page->mapping->private_lock);
4791 * We do this since we'll remove the pages after we've
4792 * removed the eb from the radix tree, so we could race
4793 * and have this page now attached to the new eb. So
4794 * only clear page_private if it's still connected to
4797 if (PagePrivate(page) &&
4798 page->private == (unsigned long)eb) {
4799 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4800 BUG_ON(PageDirty(page));
4801 BUG_ON(PageWriteback(page));
4803 * We need to make sure we haven't be attached
4806 ClearPagePrivate(page);
4807 set_page_private(page, 0);
4808 /* One for the page private */
4813 spin_unlock(&page->mapping->private_lock);
4815 /* One for when we allocated the page */
4821 * Helper for releasing the extent buffer.
4823 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4825 btrfs_release_extent_buffer_pages(eb);
4826 __free_extent_buffer(eb);
4829 static struct extent_buffer *
4830 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
4833 struct extent_buffer *eb = NULL;
4835 eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL);
4838 eb->fs_info = fs_info;
4840 rwlock_init(&eb->lock);
4841 atomic_set(&eb->blocking_readers, 0);
4842 atomic_set(&eb->blocking_writers, 0);
4843 eb->lock_nested = false;
4844 init_waitqueue_head(&eb->write_lock_wq);
4845 init_waitqueue_head(&eb->read_lock_wq);
4847 btrfs_leak_debug_add(&eb->leak_list, &buffers);
4849 spin_lock_init(&eb->refs_lock);
4850 atomic_set(&eb->refs, 1);
4851 atomic_set(&eb->io_pages, 0);
4854 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4856 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4857 > MAX_INLINE_EXTENT_BUFFER_SIZE);
4858 BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE);
4860 #ifdef CONFIG_BTRFS_DEBUG
4861 atomic_set(&eb->spinning_writers, 0);
4862 atomic_set(&eb->spinning_readers, 0);
4863 atomic_set(&eb->read_locks, 0);
4864 atomic_set(&eb->write_locks, 0);
4870 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src)
4874 struct extent_buffer *new;
4875 int num_pages = num_extent_pages(src);
4877 new = __alloc_extent_buffer(src->fs_info, src->start, src->len);
4881 for (i = 0; i < num_pages; i++) {
4882 p = alloc_page(GFP_NOFS);
4884 btrfs_release_extent_buffer(new);
4887 attach_extent_buffer_page(new, p);
4888 WARN_ON(PageDirty(p));
4891 copy_page(page_address(p), page_address(src->pages[i]));
4894 set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
4895 set_bit(EXTENT_BUFFER_UNMAPPED, &new->bflags);
4900 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4901 u64 start, unsigned long len)
4903 struct extent_buffer *eb;
4907 eb = __alloc_extent_buffer(fs_info, start, len);
4911 num_pages = num_extent_pages(eb);
4912 for (i = 0; i < num_pages; i++) {
4913 eb->pages[i] = alloc_page(GFP_NOFS);
4917 set_extent_buffer_uptodate(eb);
4918 btrfs_set_header_nritems(eb, 0);
4919 set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
4924 __free_page(eb->pages[i - 1]);
4925 __free_extent_buffer(eb);
4929 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4932 return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize);
4935 static void check_buffer_tree_ref(struct extent_buffer *eb)
4938 /* the ref bit is tricky. We have to make sure it is set
4939 * if we have the buffer dirty. Otherwise the
4940 * code to free a buffer can end up dropping a dirty
4943 * Once the ref bit is set, it won't go away while the
4944 * buffer is dirty or in writeback, and it also won't
4945 * go away while we have the reference count on the
4948 * We can't just set the ref bit without bumping the
4949 * ref on the eb because free_extent_buffer might
4950 * see the ref bit and try to clear it. If this happens
4951 * free_extent_buffer might end up dropping our original
4952 * ref by mistake and freeing the page before we are able
4953 * to add one more ref.
4955 * So bump the ref count first, then set the bit. If someone
4956 * beat us to it, drop the ref we added.
4958 refs = atomic_read(&eb->refs);
4959 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4962 spin_lock(&eb->refs_lock);
4963 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4964 atomic_inc(&eb->refs);
4965 spin_unlock(&eb->refs_lock);
4968 static void mark_extent_buffer_accessed(struct extent_buffer *eb,
4969 struct page *accessed)
4973 check_buffer_tree_ref(eb);
4975 num_pages = num_extent_pages(eb);
4976 for (i = 0; i < num_pages; i++) {
4977 struct page *p = eb->pages[i];
4980 mark_page_accessed(p);
4984 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
4987 struct extent_buffer *eb;
4990 eb = radix_tree_lookup(&fs_info->buffer_radix,
4991 start >> PAGE_SHIFT);
4992 if (eb && atomic_inc_not_zero(&eb->refs)) {
4995 * Lock our eb's refs_lock to avoid races with
4996 * free_extent_buffer. When we get our eb it might be flagged
4997 * with EXTENT_BUFFER_STALE and another task running
4998 * free_extent_buffer might have seen that flag set,
4999 * eb->refs == 2, that the buffer isn't under IO (dirty and
5000 * writeback flags not set) and it's still in the tree (flag
5001 * EXTENT_BUFFER_TREE_REF set), therefore being in the process
5002 * of decrementing the extent buffer's reference count twice.
5003 * So here we could race and increment the eb's reference count,
5004 * clear its stale flag, mark it as dirty and drop our reference
5005 * before the other task finishes executing free_extent_buffer,
5006 * which would later result in an attempt to free an extent
5007 * buffer that is dirty.
5009 if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) {
5010 spin_lock(&eb->refs_lock);
5011 spin_unlock(&eb->refs_lock);
5013 mark_extent_buffer_accessed(eb, NULL);
5021 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5022 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
5025 struct extent_buffer *eb, *exists = NULL;
5028 eb = find_extent_buffer(fs_info, start);
5031 eb = alloc_dummy_extent_buffer(fs_info, start);
5034 eb->fs_info = fs_info;
5036 ret = radix_tree_preload(GFP_NOFS);
5039 spin_lock(&fs_info->buffer_lock);
5040 ret = radix_tree_insert(&fs_info->buffer_radix,
5041 start >> PAGE_SHIFT, eb);
5042 spin_unlock(&fs_info->buffer_lock);
5043 radix_tree_preload_end();
5044 if (ret == -EEXIST) {
5045 exists = find_extent_buffer(fs_info, start);
5051 check_buffer_tree_ref(eb);
5052 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
5056 btrfs_release_extent_buffer(eb);
5061 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
5064 unsigned long len = fs_info->nodesize;
5067 unsigned long index = start >> PAGE_SHIFT;
5068 struct extent_buffer *eb;
5069 struct extent_buffer *exists = NULL;
5071 struct address_space *mapping = fs_info->btree_inode->i_mapping;
5075 if (!IS_ALIGNED(start, fs_info->sectorsize)) {
5076 btrfs_err(fs_info, "bad tree block start %llu", start);
5077 return ERR_PTR(-EINVAL);
5080 eb = find_extent_buffer(fs_info, start);
5084 eb = __alloc_extent_buffer(fs_info, start, len);
5086 return ERR_PTR(-ENOMEM);
5088 num_pages = num_extent_pages(eb);
5089 for (i = 0; i < num_pages; i++, index++) {
5090 p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL);
5092 exists = ERR_PTR(-ENOMEM);
5096 spin_lock(&mapping->private_lock);
5097 if (PagePrivate(p)) {
5099 * We could have already allocated an eb for this page
5100 * and attached one so lets see if we can get a ref on
5101 * the existing eb, and if we can we know it's good and
5102 * we can just return that one, else we know we can just
5103 * overwrite page->private.
5105 exists = (struct extent_buffer *)p->private;
5106 if (atomic_inc_not_zero(&exists->refs)) {
5107 spin_unlock(&mapping->private_lock);
5110 mark_extent_buffer_accessed(exists, p);
5116 * Do this so attach doesn't complain and we need to
5117 * drop the ref the old guy had.
5119 ClearPagePrivate(p);
5120 WARN_ON(PageDirty(p));
5123 attach_extent_buffer_page(eb, p);
5124 spin_unlock(&mapping->private_lock);
5125 WARN_ON(PageDirty(p));
5127 if (!PageUptodate(p))
5131 * We can't unlock the pages just yet since the extent buffer
5132 * hasn't been properly inserted in the radix tree, this
5133 * opens a race with btree_releasepage which can free a page
5134 * while we are still filling in all pages for the buffer and
5139 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5141 ret = radix_tree_preload(GFP_NOFS);
5143 exists = ERR_PTR(ret);
5147 spin_lock(&fs_info->buffer_lock);
5148 ret = radix_tree_insert(&fs_info->buffer_radix,
5149 start >> PAGE_SHIFT, eb);
5150 spin_unlock(&fs_info->buffer_lock);
5151 radix_tree_preload_end();
5152 if (ret == -EEXIST) {
5153 exists = find_extent_buffer(fs_info, start);
5159 /* add one reference for the tree */
5160 check_buffer_tree_ref(eb);
5161 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
5164 * Now it's safe to unlock the pages because any calls to
5165 * btree_releasepage will correctly detect that a page belongs to a
5166 * live buffer and won't free them prematurely.
5168 for (i = 0; i < num_pages; i++)
5169 unlock_page(eb->pages[i]);
5173 WARN_ON(!atomic_dec_and_test(&eb->refs));
5174 for (i = 0; i < num_pages; i++) {
5176 unlock_page(eb->pages[i]);
5179 btrfs_release_extent_buffer(eb);
5183 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
5185 struct extent_buffer *eb =
5186 container_of(head, struct extent_buffer, rcu_head);
5188 __free_extent_buffer(eb);
5191 static int release_extent_buffer(struct extent_buffer *eb)
5193 lockdep_assert_held(&eb->refs_lock);
5195 WARN_ON(atomic_read(&eb->refs) == 0);
5196 if (atomic_dec_and_test(&eb->refs)) {
5197 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
5198 struct btrfs_fs_info *fs_info = eb->fs_info;
5200 spin_unlock(&eb->refs_lock);
5202 spin_lock(&fs_info->buffer_lock);
5203 radix_tree_delete(&fs_info->buffer_radix,
5204 eb->start >> PAGE_SHIFT);
5205 spin_unlock(&fs_info->buffer_lock);
5207 spin_unlock(&eb->refs_lock);
5210 /* Should be safe to release our pages at this point */
5211 btrfs_release_extent_buffer_pages(eb);
5212 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5213 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags))) {
5214 __free_extent_buffer(eb);
5218 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
5221 spin_unlock(&eb->refs_lock);
5226 void free_extent_buffer(struct extent_buffer *eb)
5234 refs = atomic_read(&eb->refs);
5235 if ((!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && refs <= 3)
5236 || (test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) &&
5239 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
5244 spin_lock(&eb->refs_lock);
5245 if (atomic_read(&eb->refs) == 2 &&
5246 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
5247 !extent_buffer_under_io(eb) &&
5248 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5249 atomic_dec(&eb->refs);
5252 * I know this is terrible, but it's temporary until we stop tracking
5253 * the uptodate bits and such for the extent buffers.
5255 release_extent_buffer(eb);
5258 void free_extent_buffer_stale(struct extent_buffer *eb)
5263 spin_lock(&eb->refs_lock);
5264 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
5266 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
5267 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5268 atomic_dec(&eb->refs);
5269 release_extent_buffer(eb);
5272 void clear_extent_buffer_dirty(struct extent_buffer *eb)
5278 num_pages = num_extent_pages(eb);
5280 for (i = 0; i < num_pages; i++) {
5281 page = eb->pages[i];
5282 if (!PageDirty(page))
5286 WARN_ON(!PagePrivate(page));
5288 clear_page_dirty_for_io(page);
5289 xa_lock_irq(&page->mapping->i_pages);
5290 if (!PageDirty(page))
5291 __xa_clear_mark(&page->mapping->i_pages,
5292 page_index(page), PAGECACHE_TAG_DIRTY);
5293 xa_unlock_irq(&page->mapping->i_pages);
5294 ClearPageError(page);
5297 WARN_ON(atomic_read(&eb->refs) == 0);
5300 bool set_extent_buffer_dirty(struct extent_buffer *eb)
5306 check_buffer_tree_ref(eb);
5308 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
5310 num_pages = num_extent_pages(eb);
5311 WARN_ON(atomic_read(&eb->refs) == 0);
5312 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
5315 for (i = 0; i < num_pages; i++)
5316 set_page_dirty(eb->pages[i]);
5318 #ifdef CONFIG_BTRFS_DEBUG
5319 for (i = 0; i < num_pages; i++)
5320 ASSERT(PageDirty(eb->pages[i]));
5326 void clear_extent_buffer_uptodate(struct extent_buffer *eb)
5332 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5333 num_pages = num_extent_pages(eb);
5334 for (i = 0; i < num_pages; i++) {
5335 page = eb->pages[i];
5337 ClearPageUptodate(page);
5341 void set_extent_buffer_uptodate(struct extent_buffer *eb)
5347 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5348 num_pages = num_extent_pages(eb);
5349 for (i = 0; i < num_pages; i++) {
5350 page = eb->pages[i];
5351 SetPageUptodate(page);
5355 int read_extent_buffer_pages(struct extent_buffer *eb, int wait, int mirror_num)
5361 int locked_pages = 0;
5362 int all_uptodate = 1;
5364 unsigned long num_reads = 0;
5365 struct bio *bio = NULL;
5366 unsigned long bio_flags = 0;
5367 struct extent_io_tree *tree = &BTRFS_I(eb->fs_info->btree_inode)->io_tree;
5369 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
5372 num_pages = num_extent_pages(eb);
5373 for (i = 0; i < num_pages; i++) {
5374 page = eb->pages[i];
5375 if (wait == WAIT_NONE) {
5376 if (!trylock_page(page))
5384 * We need to firstly lock all pages to make sure that
5385 * the uptodate bit of our pages won't be affected by
5386 * clear_extent_buffer_uptodate().
5388 for (i = 0; i < num_pages; i++) {
5389 page = eb->pages[i];
5390 if (!PageUptodate(page)) {
5397 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5401 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
5402 eb->read_mirror = 0;
5403 atomic_set(&eb->io_pages, num_reads);
5404 for (i = 0; i < num_pages; i++) {
5405 page = eb->pages[i];
5407 if (!PageUptodate(page)) {
5409 atomic_dec(&eb->io_pages);
5414 ClearPageError(page);
5415 err = __extent_read_full_page(tree, page,
5416 btree_get_extent, &bio,
5417 mirror_num, &bio_flags,
5422 * We use &bio in above __extent_read_full_page,
5423 * so we ensure that if it returns error, the
5424 * current page fails to add itself to bio and
5425 * it's been unlocked.
5427 * We must dec io_pages by ourselves.
5429 atomic_dec(&eb->io_pages);
5437 err = submit_one_bio(bio, mirror_num, bio_flags);
5442 if (ret || wait != WAIT_COMPLETE)
5445 for (i = 0; i < num_pages; i++) {
5446 page = eb->pages[i];
5447 wait_on_page_locked(page);
5448 if (!PageUptodate(page))
5455 while (locked_pages > 0) {
5457 page = eb->pages[locked_pages];
5463 void read_extent_buffer(const struct extent_buffer *eb, void *dstv,
5464 unsigned long start, unsigned long len)
5470 char *dst = (char *)dstv;
5471 size_t start_offset = offset_in_page(eb->start);
5472 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5474 if (start + len > eb->len) {
5475 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n",
5476 eb->start, eb->len, start, len);
5477 memset(dst, 0, len);
5481 offset = offset_in_page(start_offset + start);
5484 page = eb->pages[i];
5486 cur = min(len, (PAGE_SIZE - offset));
5487 kaddr = page_address(page);
5488 memcpy(dst, kaddr + offset, cur);
5497 int read_extent_buffer_to_user(const struct extent_buffer *eb,
5499 unsigned long start, unsigned long len)
5505 char __user *dst = (char __user *)dstv;
5506 size_t start_offset = offset_in_page(eb->start);
5507 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5510 WARN_ON(start > eb->len);
5511 WARN_ON(start + len > eb->start + eb->len);
5513 offset = offset_in_page(start_offset + start);
5516 page = eb->pages[i];
5518 cur = min(len, (PAGE_SIZE - offset));
5519 kaddr = page_address(page);
5520 if (copy_to_user(dst, kaddr + offset, cur)) {
5535 * return 0 if the item is found within a page.
5536 * return 1 if the item spans two pages.
5537 * return -EINVAL otherwise.
5539 int map_private_extent_buffer(const struct extent_buffer *eb,
5540 unsigned long start, unsigned long min_len,
5541 char **map, unsigned long *map_start,
5542 unsigned long *map_len)
5547 size_t start_offset = offset_in_page(eb->start);
5548 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5549 unsigned long end_i = (start_offset + start + min_len - 1) >>
5552 if (start + min_len > eb->len) {
5553 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n",
5554 eb->start, eb->len, start, min_len);
5562 offset = start_offset;
5566 *map_start = ((u64)i << PAGE_SHIFT) - start_offset;
5570 kaddr = page_address(p);
5571 *map = kaddr + offset;
5572 *map_len = PAGE_SIZE - offset;
5576 int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,
5577 unsigned long start, unsigned long len)
5583 char *ptr = (char *)ptrv;
5584 size_t start_offset = offset_in_page(eb->start);
5585 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5588 WARN_ON(start > eb->len);
5589 WARN_ON(start + len > eb->start + eb->len);
5591 offset = offset_in_page(start_offset + start);
5594 page = eb->pages[i];
5596 cur = min(len, (PAGE_SIZE - offset));
5598 kaddr = page_address(page);
5599 ret = memcmp(ptr, kaddr + offset, cur);
5611 void write_extent_buffer_chunk_tree_uuid(struct extent_buffer *eb,
5616 WARN_ON(!PageUptodate(eb->pages[0]));
5617 kaddr = page_address(eb->pages[0]);
5618 memcpy(kaddr + offsetof(struct btrfs_header, chunk_tree_uuid), srcv,
5622 void write_extent_buffer_fsid(struct extent_buffer *eb, const void *srcv)
5626 WARN_ON(!PageUptodate(eb->pages[0]));
5627 kaddr = page_address(eb->pages[0]);
5628 memcpy(kaddr + offsetof(struct btrfs_header, fsid), srcv,
5632 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
5633 unsigned long start, unsigned long len)
5639 char *src = (char *)srcv;
5640 size_t start_offset = offset_in_page(eb->start);
5641 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5643 WARN_ON(start > eb->len);
5644 WARN_ON(start + len > eb->start + eb->len);
5646 offset = offset_in_page(start_offset + start);
5649 page = eb->pages[i];
5650 WARN_ON(!PageUptodate(page));
5652 cur = min(len, PAGE_SIZE - offset);
5653 kaddr = page_address(page);
5654 memcpy(kaddr + offset, src, cur);
5663 void memzero_extent_buffer(struct extent_buffer *eb, unsigned long start,
5670 size_t start_offset = offset_in_page(eb->start);
5671 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5673 WARN_ON(start > eb->len);
5674 WARN_ON(start + len > eb->start + eb->len);
5676 offset = offset_in_page(start_offset + start);
5679 page = eb->pages[i];
5680 WARN_ON(!PageUptodate(page));
5682 cur = min(len, PAGE_SIZE - offset);
5683 kaddr = page_address(page);
5684 memset(kaddr + offset, 0, cur);
5692 void copy_extent_buffer_full(struct extent_buffer *dst,
5693 struct extent_buffer *src)
5698 ASSERT(dst->len == src->len);
5700 num_pages = num_extent_pages(dst);
5701 for (i = 0; i < num_pages; i++)
5702 copy_page(page_address(dst->pages[i]),
5703 page_address(src->pages[i]));
5706 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
5707 unsigned long dst_offset, unsigned long src_offset,
5710 u64 dst_len = dst->len;
5715 size_t start_offset = offset_in_page(dst->start);
5716 unsigned long i = (start_offset + dst_offset) >> PAGE_SHIFT;
5718 WARN_ON(src->len != dst_len);
5720 offset = offset_in_page(start_offset + dst_offset);
5723 page = dst->pages[i];
5724 WARN_ON(!PageUptodate(page));
5726 cur = min(len, (unsigned long)(PAGE_SIZE - offset));
5728 kaddr = page_address(page);
5729 read_extent_buffer(src, kaddr + offset, src_offset, cur);
5739 * eb_bitmap_offset() - calculate the page and offset of the byte containing the
5741 * @eb: the extent buffer
5742 * @start: offset of the bitmap item in the extent buffer
5744 * @page_index: return index of the page in the extent buffer that contains the
5746 * @page_offset: return offset into the page given by page_index
5748 * This helper hides the ugliness of finding the byte in an extent buffer which
5749 * contains a given bit.
5751 static inline void eb_bitmap_offset(struct extent_buffer *eb,
5752 unsigned long start, unsigned long nr,
5753 unsigned long *page_index,
5754 size_t *page_offset)
5756 size_t start_offset = offset_in_page(eb->start);
5757 size_t byte_offset = BIT_BYTE(nr);
5761 * The byte we want is the offset of the extent buffer + the offset of
5762 * the bitmap item in the extent buffer + the offset of the byte in the
5765 offset = start_offset + start + byte_offset;
5767 *page_index = offset >> PAGE_SHIFT;
5768 *page_offset = offset_in_page(offset);
5772 * extent_buffer_test_bit - determine whether a bit in a bitmap item is set
5773 * @eb: the extent buffer
5774 * @start: offset of the bitmap item in the extent buffer
5775 * @nr: bit number to test
5777 int extent_buffer_test_bit(struct extent_buffer *eb, unsigned long start,
5785 eb_bitmap_offset(eb, start, nr, &i, &offset);
5786 page = eb->pages[i];
5787 WARN_ON(!PageUptodate(page));
5788 kaddr = page_address(page);
5789 return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1)));
5793 * extent_buffer_bitmap_set - set an area of a bitmap
5794 * @eb: the extent buffer
5795 * @start: offset of the bitmap item in the extent buffer
5796 * @pos: bit number of the first bit
5797 * @len: number of bits to set
5799 void extent_buffer_bitmap_set(struct extent_buffer *eb, unsigned long start,
5800 unsigned long pos, unsigned long len)
5806 const unsigned int size = pos + len;
5807 int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5808 u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos);
5810 eb_bitmap_offset(eb, start, pos, &i, &offset);
5811 page = eb->pages[i];
5812 WARN_ON(!PageUptodate(page));
5813 kaddr = page_address(page);
5815 while (len >= bits_to_set) {
5816 kaddr[offset] |= mask_to_set;
5818 bits_to_set = BITS_PER_BYTE;
5820 if (++offset >= PAGE_SIZE && len > 0) {
5822 page = eb->pages[++i];
5823 WARN_ON(!PageUptodate(page));
5824 kaddr = page_address(page);
5828 mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
5829 kaddr[offset] |= mask_to_set;
5835 * extent_buffer_bitmap_clear - clear an area of a bitmap
5836 * @eb: the extent buffer
5837 * @start: offset of the bitmap item in the extent buffer
5838 * @pos: bit number of the first bit
5839 * @len: number of bits to clear
5841 void extent_buffer_bitmap_clear(struct extent_buffer *eb, unsigned long start,
5842 unsigned long pos, unsigned long len)
5848 const unsigned int size = pos + len;
5849 int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5850 u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos);
5852 eb_bitmap_offset(eb, start, pos, &i, &offset);
5853 page = eb->pages[i];
5854 WARN_ON(!PageUptodate(page));
5855 kaddr = page_address(page);
5857 while (len >= bits_to_clear) {
5858 kaddr[offset] &= ~mask_to_clear;
5859 len -= bits_to_clear;
5860 bits_to_clear = BITS_PER_BYTE;
5862 if (++offset >= PAGE_SIZE && len > 0) {
5864 page = eb->pages[++i];
5865 WARN_ON(!PageUptodate(page));
5866 kaddr = page_address(page);
5870 mask_to_clear &= BITMAP_LAST_BYTE_MASK(size);
5871 kaddr[offset] &= ~mask_to_clear;
5875 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
5877 unsigned long distance = (src > dst) ? src - dst : dst - src;
5878 return distance < len;
5881 static void copy_pages(struct page *dst_page, struct page *src_page,
5882 unsigned long dst_off, unsigned long src_off,
5885 char *dst_kaddr = page_address(dst_page);
5887 int must_memmove = 0;
5889 if (dst_page != src_page) {
5890 src_kaddr = page_address(src_page);
5892 src_kaddr = dst_kaddr;
5893 if (areas_overlap(src_off, dst_off, len))
5898 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
5900 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
5903 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5904 unsigned long src_offset, unsigned long len)
5906 struct btrfs_fs_info *fs_info = dst->fs_info;
5908 size_t dst_off_in_page;
5909 size_t src_off_in_page;
5910 size_t start_offset = offset_in_page(dst->start);
5911 unsigned long dst_i;
5912 unsigned long src_i;
5914 if (src_offset + len > dst->len) {
5916 "memmove bogus src_offset %lu move len %lu dst len %lu",
5917 src_offset, len, dst->len);
5920 if (dst_offset + len > dst->len) {
5922 "memmove bogus dst_offset %lu move len %lu dst len %lu",
5923 dst_offset, len, dst->len);
5928 dst_off_in_page = offset_in_page(start_offset + dst_offset);
5929 src_off_in_page = offset_in_page(start_offset + src_offset);
5931 dst_i = (start_offset + dst_offset) >> PAGE_SHIFT;
5932 src_i = (start_offset + src_offset) >> PAGE_SHIFT;
5934 cur = min(len, (unsigned long)(PAGE_SIZE -
5936 cur = min_t(unsigned long, cur,
5937 (unsigned long)(PAGE_SIZE - dst_off_in_page));
5939 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5940 dst_off_in_page, src_off_in_page, cur);
5948 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5949 unsigned long src_offset, unsigned long len)
5951 struct btrfs_fs_info *fs_info = dst->fs_info;
5953 size_t dst_off_in_page;
5954 size_t src_off_in_page;
5955 unsigned long dst_end = dst_offset + len - 1;
5956 unsigned long src_end = src_offset + len - 1;
5957 size_t start_offset = offset_in_page(dst->start);
5958 unsigned long dst_i;
5959 unsigned long src_i;
5961 if (src_offset + len > dst->len) {
5963 "memmove bogus src_offset %lu move len %lu len %lu",
5964 src_offset, len, dst->len);
5967 if (dst_offset + len > dst->len) {
5969 "memmove bogus dst_offset %lu move len %lu len %lu",
5970 dst_offset, len, dst->len);
5973 if (dst_offset < src_offset) {
5974 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
5978 dst_i = (start_offset + dst_end) >> PAGE_SHIFT;
5979 src_i = (start_offset + src_end) >> PAGE_SHIFT;
5981 dst_off_in_page = offset_in_page(start_offset + dst_end);
5982 src_off_in_page = offset_in_page(start_offset + src_end);
5984 cur = min_t(unsigned long, len, src_off_in_page + 1);
5985 cur = min(cur, dst_off_in_page + 1);
5986 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5987 dst_off_in_page - cur + 1,
5988 src_off_in_page - cur + 1, cur);
5996 int try_release_extent_buffer(struct page *page)
5998 struct extent_buffer *eb;
6001 * We need to make sure nobody is attaching this page to an eb right
6004 spin_lock(&page->mapping->private_lock);
6005 if (!PagePrivate(page)) {
6006 spin_unlock(&page->mapping->private_lock);
6010 eb = (struct extent_buffer *)page->private;
6014 * This is a little awful but should be ok, we need to make sure that
6015 * the eb doesn't disappear out from under us while we're looking at
6018 spin_lock(&eb->refs_lock);
6019 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
6020 spin_unlock(&eb->refs_lock);
6021 spin_unlock(&page->mapping->private_lock);
6024 spin_unlock(&page->mapping->private_lock);
6027 * If tree ref isn't set then we know the ref on this eb is a real ref,
6028 * so just return, this page will likely be freed soon anyway.
6030 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
6031 spin_unlock(&eb->refs_lock);
6035 return release_extent_buffer(eb);