1 #include <linux/bitops.h>
2 #include <linux/slab.h>
5 #include <linux/pagemap.h>
6 #include <linux/page-flags.h>
7 #include <linux/spinlock.h>
8 #include <linux/blkdev.h>
9 #include <linux/swap.h>
10 #include <linux/writeback.h>
11 #include <linux/pagevec.h>
12 #include <linux/prefetch.h>
13 #include <linux/cleancache.h>
14 #include "extent_io.h"
15 #include "extent_map.h"
17 #include "btrfs_inode.h"
19 #include "check-integrity.h"
21 #include "rcu-string.h"
23 #include "transaction.h"
25 static struct kmem_cache *extent_state_cache;
26 static struct kmem_cache *extent_buffer_cache;
27 static struct bio_set *btrfs_bioset;
29 static inline bool extent_state_in_tree(const struct extent_state *state)
31 return !RB_EMPTY_NODE(&state->rb_node);
34 #ifdef CONFIG_BTRFS_DEBUG
35 static LIST_HEAD(buffers);
36 static LIST_HEAD(states);
38 static DEFINE_SPINLOCK(leak_lock);
41 void btrfs_leak_debug_add(struct list_head *new, struct list_head *head)
45 spin_lock_irqsave(&leak_lock, flags);
47 spin_unlock_irqrestore(&leak_lock, flags);
51 void btrfs_leak_debug_del(struct list_head *entry)
55 spin_lock_irqsave(&leak_lock, flags);
57 spin_unlock_irqrestore(&leak_lock, flags);
61 void btrfs_leak_debug_check(void)
63 struct extent_state *state;
64 struct extent_buffer *eb;
66 while (!list_empty(&states)) {
67 state = list_entry(states.next, struct extent_state, leak_list);
68 pr_err("BTRFS: state leak: start %llu end %llu state %u in tree %d refs %d\n",
69 state->start, state->end, state->state,
70 extent_state_in_tree(state),
71 atomic_read(&state->refs));
72 list_del(&state->leak_list);
73 kmem_cache_free(extent_state_cache, state);
76 while (!list_empty(&buffers)) {
77 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
78 pr_err("BTRFS: buffer leak start %llu len %lu refs %d\n",
79 eb->start, eb->len, atomic_read(&eb->refs));
80 list_del(&eb->leak_list);
81 kmem_cache_free(extent_buffer_cache, eb);
85 #define btrfs_debug_check_extent_io_range(tree, start, end) \
86 __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end))
87 static inline void __btrfs_debug_check_extent_io_range(const char *caller,
88 struct extent_io_tree *tree, u64 start, u64 end)
96 inode = tree->mapping->host;
97 isize = i_size_read(inode);
98 if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) {
99 btrfs_debug_rl(BTRFS_I(inode)->root->fs_info,
100 "%s: ino %llu isize %llu odd range [%llu,%llu]",
101 caller, btrfs_ino(inode), isize, start, end);
105 #define btrfs_leak_debug_add(new, head) do {} while (0)
106 #define btrfs_leak_debug_del(entry) do {} while (0)
107 #define btrfs_leak_debug_check() do {} while (0)
108 #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0)
111 #define BUFFER_LRU_MAX 64
116 struct rb_node rb_node;
119 struct extent_page_data {
121 struct extent_io_tree *tree;
122 get_extent_t *get_extent;
123 unsigned long bio_flags;
125 /* tells writepage not to lock the state bits for this range
126 * it still does the unlocking
128 unsigned int extent_locked:1;
130 /* tells the submit_bio code to use a WRITE_SYNC */
131 unsigned int sync_io:1;
134 static void add_extent_changeset(struct extent_state *state, unsigned bits,
135 struct extent_changeset *changeset,
142 if (set && (state->state & bits) == bits)
144 if (!set && (state->state & bits) == 0)
146 changeset->bytes_changed += state->end - state->start + 1;
147 ret = ulist_add(changeset->range_changed, state->start, state->end,
153 static noinline void flush_write_bio(void *data);
154 static inline struct btrfs_fs_info *
155 tree_fs_info(struct extent_io_tree *tree)
159 return btrfs_sb(tree->mapping->host->i_sb);
162 int __init extent_io_init(void)
164 extent_state_cache = kmem_cache_create("btrfs_extent_state",
165 sizeof(struct extent_state), 0,
166 SLAB_MEM_SPREAD, NULL);
167 if (!extent_state_cache)
170 extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer",
171 sizeof(struct extent_buffer), 0,
172 SLAB_MEM_SPREAD, NULL);
173 if (!extent_buffer_cache)
174 goto free_state_cache;
176 btrfs_bioset = bioset_create(BIO_POOL_SIZE,
177 offsetof(struct btrfs_io_bio, bio));
179 goto free_buffer_cache;
181 if (bioset_integrity_create(btrfs_bioset, BIO_POOL_SIZE))
187 bioset_free(btrfs_bioset);
191 kmem_cache_destroy(extent_buffer_cache);
192 extent_buffer_cache = NULL;
195 kmem_cache_destroy(extent_state_cache);
196 extent_state_cache = NULL;
200 void extent_io_exit(void)
202 btrfs_leak_debug_check();
205 * Make sure all delayed rcu free are flushed before we
209 kmem_cache_destroy(extent_state_cache);
210 kmem_cache_destroy(extent_buffer_cache);
212 bioset_free(btrfs_bioset);
215 void extent_io_tree_init(struct extent_io_tree *tree,
216 struct address_space *mapping)
218 tree->state = RB_ROOT;
220 tree->dirty_bytes = 0;
221 spin_lock_init(&tree->lock);
222 tree->mapping = mapping;
225 static struct extent_state *alloc_extent_state(gfp_t mask)
227 struct extent_state *state;
229 state = kmem_cache_alloc(extent_state_cache, mask);
233 state->failrec = NULL;
234 RB_CLEAR_NODE(&state->rb_node);
235 btrfs_leak_debug_add(&state->leak_list, &states);
236 atomic_set(&state->refs, 1);
237 init_waitqueue_head(&state->wq);
238 trace_alloc_extent_state(state, mask, _RET_IP_);
242 void free_extent_state(struct extent_state *state)
246 if (atomic_dec_and_test(&state->refs)) {
247 WARN_ON(extent_state_in_tree(state));
248 btrfs_leak_debug_del(&state->leak_list);
249 trace_free_extent_state(state, _RET_IP_);
250 kmem_cache_free(extent_state_cache, state);
254 static struct rb_node *tree_insert(struct rb_root *root,
255 struct rb_node *search_start,
257 struct rb_node *node,
258 struct rb_node ***p_in,
259 struct rb_node **parent_in)
262 struct rb_node *parent = NULL;
263 struct tree_entry *entry;
265 if (p_in && parent_in) {
271 p = search_start ? &search_start : &root->rb_node;
274 entry = rb_entry(parent, struct tree_entry, rb_node);
276 if (offset < entry->start)
278 else if (offset > entry->end)
285 rb_link_node(node, parent, p);
286 rb_insert_color(node, root);
290 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
291 struct rb_node **prev_ret,
292 struct rb_node **next_ret,
293 struct rb_node ***p_ret,
294 struct rb_node **parent_ret)
296 struct rb_root *root = &tree->state;
297 struct rb_node **n = &root->rb_node;
298 struct rb_node *prev = NULL;
299 struct rb_node *orig_prev = NULL;
300 struct tree_entry *entry;
301 struct tree_entry *prev_entry = NULL;
305 entry = rb_entry(prev, struct tree_entry, rb_node);
308 if (offset < entry->start)
310 else if (offset > entry->end)
323 while (prev && offset > prev_entry->end) {
324 prev = rb_next(prev);
325 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
332 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
333 while (prev && offset < prev_entry->start) {
334 prev = rb_prev(prev);
335 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
342 static inline struct rb_node *
343 tree_search_for_insert(struct extent_io_tree *tree,
345 struct rb_node ***p_ret,
346 struct rb_node **parent_ret)
348 struct rb_node *prev = NULL;
351 ret = __etree_search(tree, offset, &prev, NULL, p_ret, parent_ret);
357 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
360 return tree_search_for_insert(tree, offset, NULL, NULL);
363 static void merge_cb(struct extent_io_tree *tree, struct extent_state *new,
364 struct extent_state *other)
366 if (tree->ops && tree->ops->merge_extent_hook)
367 tree->ops->merge_extent_hook(tree->mapping->host, new,
372 * utility function to look for merge candidates inside a given range.
373 * Any extents with matching state are merged together into a single
374 * extent in the tree. Extents with EXTENT_IO in their state field
375 * are not merged because the end_io handlers need to be able to do
376 * operations on them without sleeping (or doing allocations/splits).
378 * This should be called with the tree lock held.
380 static void merge_state(struct extent_io_tree *tree,
381 struct extent_state *state)
383 struct extent_state *other;
384 struct rb_node *other_node;
386 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
389 other_node = rb_prev(&state->rb_node);
391 other = rb_entry(other_node, struct extent_state, rb_node);
392 if (other->end == state->start - 1 &&
393 other->state == state->state) {
394 merge_cb(tree, state, other);
395 state->start = other->start;
396 rb_erase(&other->rb_node, &tree->state);
397 RB_CLEAR_NODE(&other->rb_node);
398 free_extent_state(other);
401 other_node = rb_next(&state->rb_node);
403 other = rb_entry(other_node, struct extent_state, rb_node);
404 if (other->start == state->end + 1 &&
405 other->state == state->state) {
406 merge_cb(tree, state, other);
407 state->end = other->end;
408 rb_erase(&other->rb_node, &tree->state);
409 RB_CLEAR_NODE(&other->rb_node);
410 free_extent_state(other);
415 static void set_state_cb(struct extent_io_tree *tree,
416 struct extent_state *state, unsigned *bits)
418 if (tree->ops && tree->ops->set_bit_hook)
419 tree->ops->set_bit_hook(tree->mapping->host, state, bits);
422 static void clear_state_cb(struct extent_io_tree *tree,
423 struct extent_state *state, unsigned *bits)
425 if (tree->ops && tree->ops->clear_bit_hook)
426 tree->ops->clear_bit_hook(tree->mapping->host, state, bits);
429 static void set_state_bits(struct extent_io_tree *tree,
430 struct extent_state *state, unsigned *bits,
431 struct extent_changeset *changeset);
434 * insert an extent_state struct into the tree. 'bits' are set on the
435 * struct before it is inserted.
437 * This may return -EEXIST if the extent is already there, in which case the
438 * state struct is freed.
440 * The tree lock is not taken internally. This is a utility function and
441 * probably isn't what you want to call (see set/clear_extent_bit).
443 static int insert_state(struct extent_io_tree *tree,
444 struct extent_state *state, u64 start, u64 end,
446 struct rb_node **parent,
447 unsigned *bits, struct extent_changeset *changeset)
449 struct rb_node *node;
452 WARN(1, KERN_ERR "BTRFS: end < start %llu %llu\n",
454 state->start = start;
457 set_state_bits(tree, state, bits, changeset);
459 node = tree_insert(&tree->state, NULL, end, &state->rb_node, p, parent);
461 struct extent_state *found;
462 found = rb_entry(node, struct extent_state, rb_node);
463 pr_err("BTRFS: found node %llu %llu on insert of %llu %llu\n",
464 found->start, found->end, start, end);
467 merge_state(tree, state);
471 static void split_cb(struct extent_io_tree *tree, struct extent_state *orig,
474 if (tree->ops && tree->ops->split_extent_hook)
475 tree->ops->split_extent_hook(tree->mapping->host, orig, split);
479 * split a given extent state struct in two, inserting the preallocated
480 * struct 'prealloc' as the newly created second half. 'split' indicates an
481 * offset inside 'orig' where it should be split.
484 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
485 * are two extent state structs in the tree:
486 * prealloc: [orig->start, split - 1]
487 * orig: [ split, orig->end ]
489 * The tree locks are not taken by this function. They need to be held
492 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
493 struct extent_state *prealloc, u64 split)
495 struct rb_node *node;
497 split_cb(tree, orig, split);
499 prealloc->start = orig->start;
500 prealloc->end = split - 1;
501 prealloc->state = orig->state;
504 node = tree_insert(&tree->state, &orig->rb_node, prealloc->end,
505 &prealloc->rb_node, NULL, NULL);
507 free_extent_state(prealloc);
513 static struct extent_state *next_state(struct extent_state *state)
515 struct rb_node *next = rb_next(&state->rb_node);
517 return rb_entry(next, struct extent_state, rb_node);
523 * utility function to clear some bits in an extent state struct.
524 * it will optionally wake up any one waiting on this state (wake == 1).
526 * If no bits are set on the state struct after clearing things, the
527 * struct is freed and removed from the tree
529 static struct extent_state *clear_state_bit(struct extent_io_tree *tree,
530 struct extent_state *state,
531 unsigned *bits, int wake,
532 struct extent_changeset *changeset)
534 struct extent_state *next;
535 unsigned bits_to_clear = *bits & ~EXTENT_CTLBITS;
537 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
538 u64 range = state->end - state->start + 1;
539 WARN_ON(range > tree->dirty_bytes);
540 tree->dirty_bytes -= range;
542 clear_state_cb(tree, state, bits);
543 add_extent_changeset(state, bits_to_clear, changeset, 0);
544 state->state &= ~bits_to_clear;
547 if (state->state == 0) {
548 next = next_state(state);
549 if (extent_state_in_tree(state)) {
550 rb_erase(&state->rb_node, &tree->state);
551 RB_CLEAR_NODE(&state->rb_node);
552 free_extent_state(state);
557 merge_state(tree, state);
558 next = next_state(state);
563 static struct extent_state *
564 alloc_extent_state_atomic(struct extent_state *prealloc)
567 prealloc = alloc_extent_state(GFP_ATOMIC);
572 static void extent_io_tree_panic(struct extent_io_tree *tree, int err)
574 btrfs_panic(tree_fs_info(tree), err,
575 "Locking error: Extent tree was modified by another thread while locked.");
579 * clear some bits on a range in the tree. This may require splitting
580 * or inserting elements in the tree, so the gfp mask is used to
581 * indicate which allocations or sleeping are allowed.
583 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
584 * the given range from the tree regardless of state (ie for truncate).
586 * the range [start, end] is inclusive.
588 * This takes the tree lock, and returns 0 on success and < 0 on error.
590 static int __clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
591 unsigned bits, int wake, int delete,
592 struct extent_state **cached_state,
593 gfp_t mask, struct extent_changeset *changeset)
595 struct extent_state *state;
596 struct extent_state *cached;
597 struct extent_state *prealloc = NULL;
598 struct rb_node *node;
603 btrfs_debug_check_extent_io_range(tree, start, end);
605 if (bits & EXTENT_DELALLOC)
606 bits |= EXTENT_NORESERVE;
609 bits |= ~EXTENT_CTLBITS;
610 bits |= EXTENT_FIRST_DELALLOC;
612 if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY))
615 if (!prealloc && gfpflags_allow_blocking(mask)) {
617 * Don't care for allocation failure here because we might end
618 * up not needing the pre-allocated extent state at all, which
619 * is the case if we only have in the tree extent states that
620 * cover our input range and don't cover too any other range.
621 * If we end up needing a new extent state we allocate it later.
623 prealloc = alloc_extent_state(mask);
626 spin_lock(&tree->lock);
628 cached = *cached_state;
631 *cached_state = NULL;
635 if (cached && extent_state_in_tree(cached) &&
636 cached->start <= start && cached->end > start) {
638 atomic_dec(&cached->refs);
643 free_extent_state(cached);
646 * this search will find the extents that end after
649 node = tree_search(tree, start);
652 state = rb_entry(node, struct extent_state, rb_node);
654 if (state->start > end)
656 WARN_ON(state->end < start);
657 last_end = state->end;
659 /* the state doesn't have the wanted bits, go ahead */
660 if (!(state->state & bits)) {
661 state = next_state(state);
666 * | ---- desired range ---- |
668 * | ------------- state -------------- |
670 * We need to split the extent we found, and may flip
671 * bits on second half.
673 * If the extent we found extends past our range, we
674 * just split and search again. It'll get split again
675 * the next time though.
677 * If the extent we found is inside our range, we clear
678 * the desired bit on it.
681 if (state->start < start) {
682 prealloc = alloc_extent_state_atomic(prealloc);
684 err = split_state(tree, state, prealloc, start);
686 extent_io_tree_panic(tree, err);
691 if (state->end <= end) {
692 state = clear_state_bit(tree, state, &bits, wake,
699 * | ---- desired range ---- |
701 * We need to split the extent, and clear the bit
704 if (state->start <= end && state->end > end) {
705 prealloc = alloc_extent_state_atomic(prealloc);
707 err = split_state(tree, state, prealloc, end + 1);
709 extent_io_tree_panic(tree, err);
714 clear_state_bit(tree, prealloc, &bits, wake, changeset);
720 state = clear_state_bit(tree, state, &bits, wake, changeset);
722 if (last_end == (u64)-1)
724 start = last_end + 1;
725 if (start <= end && state && !need_resched())
731 spin_unlock(&tree->lock);
732 if (gfpflags_allow_blocking(mask))
737 spin_unlock(&tree->lock);
739 free_extent_state(prealloc);
745 static void wait_on_state(struct extent_io_tree *tree,
746 struct extent_state *state)
747 __releases(tree->lock)
748 __acquires(tree->lock)
751 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
752 spin_unlock(&tree->lock);
754 spin_lock(&tree->lock);
755 finish_wait(&state->wq, &wait);
759 * waits for one or more bits to clear on a range in the state tree.
760 * The range [start, end] is inclusive.
761 * The tree lock is taken by this function
763 static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
766 struct extent_state *state;
767 struct rb_node *node;
769 btrfs_debug_check_extent_io_range(tree, start, end);
771 spin_lock(&tree->lock);
775 * this search will find all the extents that end after
778 node = tree_search(tree, start);
783 state = rb_entry(node, struct extent_state, rb_node);
785 if (state->start > end)
788 if (state->state & bits) {
789 start = state->start;
790 atomic_inc(&state->refs);
791 wait_on_state(tree, state);
792 free_extent_state(state);
795 start = state->end + 1;
800 if (!cond_resched_lock(&tree->lock)) {
801 node = rb_next(node);
806 spin_unlock(&tree->lock);
809 static void set_state_bits(struct extent_io_tree *tree,
810 struct extent_state *state,
811 unsigned *bits, struct extent_changeset *changeset)
813 unsigned bits_to_set = *bits & ~EXTENT_CTLBITS;
815 set_state_cb(tree, state, bits);
816 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
817 u64 range = state->end - state->start + 1;
818 tree->dirty_bytes += range;
820 add_extent_changeset(state, bits_to_set, changeset, 1);
821 state->state |= bits_to_set;
824 static void cache_state_if_flags(struct extent_state *state,
825 struct extent_state **cached_ptr,
828 if (cached_ptr && !(*cached_ptr)) {
829 if (!flags || (state->state & flags)) {
831 atomic_inc(&state->refs);
836 static void cache_state(struct extent_state *state,
837 struct extent_state **cached_ptr)
839 return cache_state_if_flags(state, cached_ptr,
840 EXTENT_IOBITS | EXTENT_BOUNDARY);
844 * set some bits on a range in the tree. This may require allocations or
845 * sleeping, so the gfp mask is used to indicate what is allowed.
847 * If any of the exclusive bits are set, this will fail with -EEXIST if some
848 * part of the range already has the desired bits set. The start of the
849 * existing range is returned in failed_start in this case.
851 * [start, end] is inclusive This takes the tree lock.
854 static int __must_check
855 __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
856 unsigned bits, unsigned exclusive_bits,
857 u64 *failed_start, struct extent_state **cached_state,
858 gfp_t mask, struct extent_changeset *changeset)
860 struct extent_state *state;
861 struct extent_state *prealloc = NULL;
862 struct rb_node *node;
864 struct rb_node *parent;
869 btrfs_debug_check_extent_io_range(tree, start, end);
871 bits |= EXTENT_FIRST_DELALLOC;
873 if (!prealloc && gfpflags_allow_blocking(mask)) {
875 * Don't care for allocation failure here because we might end
876 * up not needing the pre-allocated extent state at all, which
877 * is the case if we only have in the tree extent states that
878 * cover our input range and don't cover too any other range.
879 * If we end up needing a new extent state we allocate it later.
881 prealloc = alloc_extent_state(mask);
884 spin_lock(&tree->lock);
885 if (cached_state && *cached_state) {
886 state = *cached_state;
887 if (state->start <= start && state->end > start &&
888 extent_state_in_tree(state)) {
889 node = &state->rb_node;
894 * this search will find all the extents that end after
897 node = tree_search_for_insert(tree, start, &p, &parent);
899 prealloc = alloc_extent_state_atomic(prealloc);
901 err = insert_state(tree, prealloc, start, end,
902 &p, &parent, &bits, changeset);
904 extent_io_tree_panic(tree, err);
906 cache_state(prealloc, cached_state);
910 state = rb_entry(node, struct extent_state, rb_node);
912 last_start = state->start;
913 last_end = state->end;
916 * | ---- desired range ---- |
919 * Just lock what we found and keep going
921 if (state->start == start && state->end <= end) {
922 if (state->state & exclusive_bits) {
923 *failed_start = state->start;
928 set_state_bits(tree, state, &bits, changeset);
929 cache_state(state, cached_state);
930 merge_state(tree, state);
931 if (last_end == (u64)-1)
933 start = last_end + 1;
934 state = next_state(state);
935 if (start < end && state && state->start == start &&
942 * | ---- desired range ---- |
945 * | ------------- state -------------- |
947 * We need to split the extent we found, and may flip bits on
950 * If the extent we found extends past our
951 * range, we just split and search again. It'll get split
952 * again the next time though.
954 * If the extent we found is inside our range, we set the
957 if (state->start < start) {
958 if (state->state & exclusive_bits) {
959 *failed_start = start;
964 prealloc = alloc_extent_state_atomic(prealloc);
966 err = split_state(tree, state, prealloc, start);
968 extent_io_tree_panic(tree, err);
973 if (state->end <= end) {
974 set_state_bits(tree, state, &bits, changeset);
975 cache_state(state, cached_state);
976 merge_state(tree, state);
977 if (last_end == (u64)-1)
979 start = last_end + 1;
980 state = next_state(state);
981 if (start < end && state && state->start == start &&
988 * | ---- desired range ---- |
989 * | state | or | state |
991 * There's a hole, we need to insert something in it and
992 * ignore the extent we found.
994 if (state->start > start) {
996 if (end < last_start)
999 this_end = last_start - 1;
1001 prealloc = alloc_extent_state_atomic(prealloc);
1005 * Avoid to free 'prealloc' if it can be merged with
1008 err = insert_state(tree, prealloc, start, this_end,
1009 NULL, NULL, &bits, changeset);
1011 extent_io_tree_panic(tree, err);
1013 cache_state(prealloc, cached_state);
1015 start = this_end + 1;
1019 * | ---- desired range ---- |
1021 * We need to split the extent, and set the bit
1024 if (state->start <= end && state->end > end) {
1025 if (state->state & exclusive_bits) {
1026 *failed_start = start;
1031 prealloc = alloc_extent_state_atomic(prealloc);
1033 err = split_state(tree, state, prealloc, end + 1);
1035 extent_io_tree_panic(tree, err);
1037 set_state_bits(tree, prealloc, &bits, changeset);
1038 cache_state(prealloc, cached_state);
1039 merge_state(tree, prealloc);
1047 spin_unlock(&tree->lock);
1048 if (gfpflags_allow_blocking(mask))
1053 spin_unlock(&tree->lock);
1055 free_extent_state(prealloc);
1061 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1062 unsigned bits, u64 * failed_start,
1063 struct extent_state **cached_state, gfp_t mask)
1065 return __set_extent_bit(tree, start, end, bits, 0, failed_start,
1066 cached_state, mask, NULL);
1071 * convert_extent_bit - convert all bits in a given range from one bit to
1073 * @tree: the io tree to search
1074 * @start: the start offset in bytes
1075 * @end: the end offset in bytes (inclusive)
1076 * @bits: the bits to set in this range
1077 * @clear_bits: the bits to clear in this range
1078 * @cached_state: state that we're going to cache
1080 * This will go through and set bits for the given range. If any states exist
1081 * already in this range they are set with the given bit and cleared of the
1082 * clear_bits. This is only meant to be used by things that are mergeable, ie
1083 * converting from say DELALLOC to DIRTY. This is not meant to be used with
1084 * boundary bits like LOCK.
1086 * All allocations are done with GFP_NOFS.
1088 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1089 unsigned bits, unsigned clear_bits,
1090 struct extent_state **cached_state)
1092 struct extent_state *state;
1093 struct extent_state *prealloc = NULL;
1094 struct rb_node *node;
1096 struct rb_node *parent;
1100 bool first_iteration = true;
1102 btrfs_debug_check_extent_io_range(tree, start, end);
1107 * Best effort, don't worry if extent state allocation fails
1108 * here for the first iteration. We might have a cached state
1109 * that matches exactly the target range, in which case no
1110 * extent state allocations are needed. We'll only know this
1111 * after locking the tree.
1113 prealloc = alloc_extent_state(GFP_NOFS);
1114 if (!prealloc && !first_iteration)
1118 spin_lock(&tree->lock);
1119 if (cached_state && *cached_state) {
1120 state = *cached_state;
1121 if (state->start <= start && state->end > start &&
1122 extent_state_in_tree(state)) {
1123 node = &state->rb_node;
1129 * this search will find all the extents that end after
1132 node = tree_search_for_insert(tree, start, &p, &parent);
1134 prealloc = alloc_extent_state_atomic(prealloc);
1139 err = insert_state(tree, prealloc, start, end,
1140 &p, &parent, &bits, NULL);
1142 extent_io_tree_panic(tree, err);
1143 cache_state(prealloc, cached_state);
1147 state = rb_entry(node, struct extent_state, rb_node);
1149 last_start = state->start;
1150 last_end = state->end;
1153 * | ---- desired range ---- |
1156 * Just lock what we found and keep going
1158 if (state->start == start && state->end <= end) {
1159 set_state_bits(tree, state, &bits, NULL);
1160 cache_state(state, cached_state);
1161 state = clear_state_bit(tree, state, &clear_bits, 0, NULL);
1162 if (last_end == (u64)-1)
1164 start = last_end + 1;
1165 if (start < end && state && state->start == start &&
1172 * | ---- desired range ---- |
1175 * | ------------- state -------------- |
1177 * We need to split the extent we found, and may flip bits on
1180 * If the extent we found extends past our
1181 * range, we just split and search again. It'll get split
1182 * again the next time though.
1184 * If the extent we found is inside our range, we set the
1185 * desired bit on it.
1187 if (state->start < start) {
1188 prealloc = alloc_extent_state_atomic(prealloc);
1193 err = split_state(tree, state, prealloc, start);
1195 extent_io_tree_panic(tree, err);
1199 if (state->end <= end) {
1200 set_state_bits(tree, state, &bits, NULL);
1201 cache_state(state, cached_state);
1202 state = clear_state_bit(tree, state, &clear_bits, 0,
1204 if (last_end == (u64)-1)
1206 start = last_end + 1;
1207 if (start < end && state && state->start == start &&
1214 * | ---- desired range ---- |
1215 * | state | or | state |
1217 * There's a hole, we need to insert something in it and
1218 * ignore the extent we found.
1220 if (state->start > start) {
1222 if (end < last_start)
1225 this_end = last_start - 1;
1227 prealloc = alloc_extent_state_atomic(prealloc);
1234 * Avoid to free 'prealloc' if it can be merged with
1237 err = insert_state(tree, prealloc, start, this_end,
1238 NULL, NULL, &bits, NULL);
1240 extent_io_tree_panic(tree, err);
1241 cache_state(prealloc, cached_state);
1243 start = this_end + 1;
1247 * | ---- desired range ---- |
1249 * We need to split the extent, and set the bit
1252 if (state->start <= end && state->end > end) {
1253 prealloc = alloc_extent_state_atomic(prealloc);
1259 err = split_state(tree, state, prealloc, end + 1);
1261 extent_io_tree_panic(tree, err);
1263 set_state_bits(tree, prealloc, &bits, NULL);
1264 cache_state(prealloc, cached_state);
1265 clear_state_bit(tree, prealloc, &clear_bits, 0, NULL);
1273 spin_unlock(&tree->lock);
1275 first_iteration = false;
1279 spin_unlock(&tree->lock);
1281 free_extent_state(prealloc);
1286 /* wrappers around set/clear extent bit */
1287 int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1288 unsigned bits, struct extent_changeset *changeset)
1291 * We don't support EXTENT_LOCKED yet, as current changeset will
1292 * record any bits changed, so for EXTENT_LOCKED case, it will
1293 * either fail with -EEXIST or changeset will record the whole
1296 BUG_ON(bits & EXTENT_LOCKED);
1298 return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL, GFP_NOFS,
1302 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1303 unsigned bits, int wake, int delete,
1304 struct extent_state **cached, gfp_t mask)
1306 return __clear_extent_bit(tree, start, end, bits, wake, delete,
1307 cached, mask, NULL);
1310 int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1311 unsigned bits, struct extent_changeset *changeset)
1314 * Don't support EXTENT_LOCKED case, same reason as
1315 * set_record_extent_bits().
1317 BUG_ON(bits & EXTENT_LOCKED);
1319 return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, GFP_NOFS,
1324 * either insert or lock state struct between start and end use mask to tell
1325 * us if waiting is desired.
1327 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1328 struct extent_state **cached_state)
1334 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED,
1335 EXTENT_LOCKED, &failed_start,
1336 cached_state, GFP_NOFS, NULL);
1337 if (err == -EEXIST) {
1338 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1339 start = failed_start;
1342 WARN_ON(start > end);
1347 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1352 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1353 &failed_start, NULL, GFP_NOFS, NULL);
1354 if (err == -EEXIST) {
1355 if (failed_start > start)
1356 clear_extent_bit(tree, start, failed_start - 1,
1357 EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS);
1363 void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
1365 unsigned long index = start >> PAGE_SHIFT;
1366 unsigned long end_index = end >> PAGE_SHIFT;
1369 while (index <= end_index) {
1370 page = find_get_page(inode->i_mapping, index);
1371 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1372 clear_page_dirty_for_io(page);
1378 void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
1380 unsigned long index = start >> PAGE_SHIFT;
1381 unsigned long end_index = end >> PAGE_SHIFT;
1384 while (index <= end_index) {
1385 page = find_get_page(inode->i_mapping, index);
1386 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1387 __set_page_dirty_nobuffers(page);
1388 account_page_redirty(page);
1395 * helper function to set both pages and extents in the tree writeback
1397 static void set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
1399 unsigned long index = start >> PAGE_SHIFT;
1400 unsigned long end_index = end >> PAGE_SHIFT;
1403 while (index <= end_index) {
1404 page = find_get_page(tree->mapping, index);
1405 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1406 set_page_writeback(page);
1412 /* find the first state struct with 'bits' set after 'start', and
1413 * return it. tree->lock must be held. NULL will returned if
1414 * nothing was found after 'start'
1416 static struct extent_state *
1417 find_first_extent_bit_state(struct extent_io_tree *tree,
1418 u64 start, unsigned bits)
1420 struct rb_node *node;
1421 struct extent_state *state;
1424 * this search will find all the extents that end after
1427 node = tree_search(tree, start);
1432 state = rb_entry(node, struct extent_state, rb_node);
1433 if (state->end >= start && (state->state & bits))
1436 node = rb_next(node);
1445 * find the first offset in the io tree with 'bits' set. zero is
1446 * returned if we find something, and *start_ret and *end_ret are
1447 * set to reflect the state struct that was found.
1449 * If nothing was found, 1 is returned. If found something, return 0.
1451 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1452 u64 *start_ret, u64 *end_ret, unsigned bits,
1453 struct extent_state **cached_state)
1455 struct extent_state *state;
1459 spin_lock(&tree->lock);
1460 if (cached_state && *cached_state) {
1461 state = *cached_state;
1462 if (state->end == start - 1 && extent_state_in_tree(state)) {
1463 n = rb_next(&state->rb_node);
1465 state = rb_entry(n, struct extent_state,
1467 if (state->state & bits)
1471 free_extent_state(*cached_state);
1472 *cached_state = NULL;
1475 free_extent_state(*cached_state);
1476 *cached_state = NULL;
1479 state = find_first_extent_bit_state(tree, start, bits);
1482 cache_state_if_flags(state, cached_state, 0);
1483 *start_ret = state->start;
1484 *end_ret = state->end;
1488 spin_unlock(&tree->lock);
1493 * find a contiguous range of bytes in the file marked as delalloc, not
1494 * more than 'max_bytes'. start and end are used to return the range,
1496 * 1 is returned if we find something, 0 if nothing was in the tree
1498 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1499 u64 *start, u64 *end, u64 max_bytes,
1500 struct extent_state **cached_state)
1502 struct rb_node *node;
1503 struct extent_state *state;
1504 u64 cur_start = *start;
1506 u64 total_bytes = 0;
1508 spin_lock(&tree->lock);
1511 * this search will find all the extents that end after
1514 node = tree_search(tree, cur_start);
1522 state = rb_entry(node, struct extent_state, rb_node);
1523 if (found && (state->start != cur_start ||
1524 (state->state & EXTENT_BOUNDARY))) {
1527 if (!(state->state & EXTENT_DELALLOC)) {
1533 *start = state->start;
1534 *cached_state = state;
1535 atomic_inc(&state->refs);
1539 cur_start = state->end + 1;
1540 node = rb_next(node);
1541 total_bytes += state->end - state->start + 1;
1542 if (total_bytes >= max_bytes)
1548 spin_unlock(&tree->lock);
1552 static noinline void __unlock_for_delalloc(struct inode *inode,
1553 struct page *locked_page,
1557 struct page *pages[16];
1558 unsigned long index = start >> PAGE_SHIFT;
1559 unsigned long end_index = end >> PAGE_SHIFT;
1560 unsigned long nr_pages = end_index - index + 1;
1563 if (index == locked_page->index && end_index == index)
1566 while (nr_pages > 0) {
1567 ret = find_get_pages_contig(inode->i_mapping, index,
1568 min_t(unsigned long, nr_pages,
1569 ARRAY_SIZE(pages)), pages);
1570 for (i = 0; i < ret; i++) {
1571 if (pages[i] != locked_page)
1572 unlock_page(pages[i]);
1581 static noinline int lock_delalloc_pages(struct inode *inode,
1582 struct page *locked_page,
1586 unsigned long index = delalloc_start >> PAGE_SHIFT;
1587 unsigned long start_index = index;
1588 unsigned long end_index = delalloc_end >> PAGE_SHIFT;
1589 unsigned long pages_locked = 0;
1590 struct page *pages[16];
1591 unsigned long nrpages;
1595 /* the caller is responsible for locking the start index */
1596 if (index == locked_page->index && index == end_index)
1599 /* skip the page at the start index */
1600 nrpages = end_index - index + 1;
1601 while (nrpages > 0) {
1602 ret = find_get_pages_contig(inode->i_mapping, index,
1603 min_t(unsigned long,
1604 nrpages, ARRAY_SIZE(pages)), pages);
1609 /* now we have an array of pages, lock them all */
1610 for (i = 0; i < ret; i++) {
1612 * the caller is taking responsibility for
1615 if (pages[i] != locked_page) {
1616 lock_page(pages[i]);
1617 if (!PageDirty(pages[i]) ||
1618 pages[i]->mapping != inode->i_mapping) {
1620 unlock_page(pages[i]);
1634 if (ret && pages_locked) {
1635 __unlock_for_delalloc(inode, locked_page,
1637 ((u64)(start_index + pages_locked - 1)) <<
1644 * find a contiguous range of bytes in the file marked as delalloc, not
1645 * more than 'max_bytes'. start and end are used to return the range,
1647 * 1 is returned if we find something, 0 if nothing was in the tree
1649 STATIC u64 find_lock_delalloc_range(struct inode *inode,
1650 struct extent_io_tree *tree,
1651 struct page *locked_page, u64 *start,
1652 u64 *end, u64 max_bytes)
1657 struct extent_state *cached_state = NULL;
1662 /* step one, find a bunch of delalloc bytes starting at start */
1663 delalloc_start = *start;
1665 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1666 max_bytes, &cached_state);
1667 if (!found || delalloc_end <= *start) {
1668 *start = delalloc_start;
1669 *end = delalloc_end;
1670 free_extent_state(cached_state);
1675 * start comes from the offset of locked_page. We have to lock
1676 * pages in order, so we can't process delalloc bytes before
1679 if (delalloc_start < *start)
1680 delalloc_start = *start;
1683 * make sure to limit the number of pages we try to lock down
1685 if (delalloc_end + 1 - delalloc_start > max_bytes)
1686 delalloc_end = delalloc_start + max_bytes - 1;
1688 /* step two, lock all the pages after the page that has start */
1689 ret = lock_delalloc_pages(inode, locked_page,
1690 delalloc_start, delalloc_end);
1691 if (ret == -EAGAIN) {
1692 /* some of the pages are gone, lets avoid looping by
1693 * shortening the size of the delalloc range we're searching
1695 free_extent_state(cached_state);
1696 cached_state = NULL;
1698 max_bytes = PAGE_SIZE;
1706 BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */
1708 /* step three, lock the state bits for the whole range */
1709 lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state);
1711 /* then test to make sure it is all still delalloc */
1712 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1713 EXTENT_DELALLOC, 1, cached_state);
1715 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1716 &cached_state, GFP_NOFS);
1717 __unlock_for_delalloc(inode, locked_page,
1718 delalloc_start, delalloc_end);
1722 free_extent_state(cached_state);
1723 *start = delalloc_start;
1724 *end = delalloc_end;
1729 void extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end,
1730 u64 delalloc_end, struct page *locked_page,
1731 unsigned clear_bits,
1732 unsigned long page_ops)
1734 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
1736 struct page *pages[16];
1737 unsigned long index = start >> PAGE_SHIFT;
1738 unsigned long end_index = end >> PAGE_SHIFT;
1739 unsigned long nr_pages = end_index - index + 1;
1742 clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS);
1746 if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0)
1747 mapping_set_error(inode->i_mapping, -EIO);
1749 while (nr_pages > 0) {
1750 ret = find_get_pages_contig(inode->i_mapping, index,
1751 min_t(unsigned long,
1752 nr_pages, ARRAY_SIZE(pages)), pages);
1753 for (i = 0; i < ret; i++) {
1755 if (page_ops & PAGE_SET_PRIVATE2)
1756 SetPagePrivate2(pages[i]);
1758 if (pages[i] == locked_page) {
1762 if (page_ops & PAGE_CLEAR_DIRTY)
1763 clear_page_dirty_for_io(pages[i]);
1764 if (page_ops & PAGE_SET_WRITEBACK)
1765 set_page_writeback(pages[i]);
1766 if (page_ops & PAGE_SET_ERROR)
1767 SetPageError(pages[i]);
1768 if (page_ops & PAGE_END_WRITEBACK)
1769 end_page_writeback(pages[i]);
1770 if (page_ops & PAGE_UNLOCK)
1771 unlock_page(pages[i]);
1781 * count the number of bytes in the tree that have a given bit(s)
1782 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1783 * cached. The total number found is returned.
1785 u64 count_range_bits(struct extent_io_tree *tree,
1786 u64 *start, u64 search_end, u64 max_bytes,
1787 unsigned bits, int contig)
1789 struct rb_node *node;
1790 struct extent_state *state;
1791 u64 cur_start = *start;
1792 u64 total_bytes = 0;
1796 if (WARN_ON(search_end <= cur_start))
1799 spin_lock(&tree->lock);
1800 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1801 total_bytes = tree->dirty_bytes;
1805 * this search will find all the extents that end after
1808 node = tree_search(tree, cur_start);
1813 state = rb_entry(node, struct extent_state, rb_node);
1814 if (state->start > search_end)
1816 if (contig && found && state->start > last + 1)
1818 if (state->end >= cur_start && (state->state & bits) == bits) {
1819 total_bytes += min(search_end, state->end) + 1 -
1820 max(cur_start, state->start);
1821 if (total_bytes >= max_bytes)
1824 *start = max(cur_start, state->start);
1828 } else if (contig && found) {
1831 node = rb_next(node);
1836 spin_unlock(&tree->lock);
1841 * set the private field for a given byte offset in the tree. If there isn't
1842 * an extent_state there already, this does nothing.
1844 static noinline int set_state_failrec(struct extent_io_tree *tree, u64 start,
1845 struct io_failure_record *failrec)
1847 struct rb_node *node;
1848 struct extent_state *state;
1851 spin_lock(&tree->lock);
1853 * this search will find all the extents that end after
1856 node = tree_search(tree, start);
1861 state = rb_entry(node, struct extent_state, rb_node);
1862 if (state->start != start) {
1866 state->failrec = failrec;
1868 spin_unlock(&tree->lock);
1872 static noinline int get_state_failrec(struct extent_io_tree *tree, u64 start,
1873 struct io_failure_record **failrec)
1875 struct rb_node *node;
1876 struct extent_state *state;
1879 spin_lock(&tree->lock);
1881 * this search will find all the extents that end after
1884 node = tree_search(tree, start);
1889 state = rb_entry(node, struct extent_state, rb_node);
1890 if (state->start != start) {
1894 *failrec = state->failrec;
1896 spin_unlock(&tree->lock);
1901 * searches a range in the state tree for a given mask.
1902 * If 'filled' == 1, this returns 1 only if every extent in the tree
1903 * has the bits set. Otherwise, 1 is returned if any bit in the
1904 * range is found set.
1906 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1907 unsigned bits, int filled, struct extent_state *cached)
1909 struct extent_state *state = NULL;
1910 struct rb_node *node;
1913 spin_lock(&tree->lock);
1914 if (cached && extent_state_in_tree(cached) && cached->start <= start &&
1915 cached->end > start)
1916 node = &cached->rb_node;
1918 node = tree_search(tree, start);
1919 while (node && start <= end) {
1920 state = rb_entry(node, struct extent_state, rb_node);
1922 if (filled && state->start > start) {
1927 if (state->start > end)
1930 if (state->state & bits) {
1934 } else if (filled) {
1939 if (state->end == (u64)-1)
1942 start = state->end + 1;
1945 node = rb_next(node);
1952 spin_unlock(&tree->lock);
1957 * helper function to set a given page up to date if all the
1958 * extents in the tree for that page are up to date
1960 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
1962 u64 start = page_offset(page);
1963 u64 end = start + PAGE_SIZE - 1;
1964 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1965 SetPageUptodate(page);
1968 int free_io_failure(struct inode *inode, struct io_failure_record *rec)
1972 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1974 set_state_failrec(failure_tree, rec->start, NULL);
1975 ret = clear_extent_bits(failure_tree, rec->start,
1976 rec->start + rec->len - 1,
1977 EXTENT_LOCKED | EXTENT_DIRTY);
1981 ret = clear_extent_bits(&BTRFS_I(inode)->io_tree, rec->start,
1982 rec->start + rec->len - 1,
1992 * this bypasses the standard btrfs submit functions deliberately, as
1993 * the standard behavior is to write all copies in a raid setup. here we only
1994 * want to write the one bad copy. so we do the mapping for ourselves and issue
1995 * submit_bio directly.
1996 * to avoid any synchronization issues, wait for the data after writing, which
1997 * actually prevents the read that triggered the error from finishing.
1998 * currently, there can be no more than two copies of every data bit. thus,
1999 * exactly one rewrite is required.
2001 int repair_io_failure(struct inode *inode, u64 start, u64 length, u64 logical,
2002 struct page *page, unsigned int pg_offset, int mirror_num)
2004 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2006 struct btrfs_device *dev;
2009 struct btrfs_bio *bbio = NULL;
2010 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
2013 ASSERT(!(fs_info->sb->s_flags & MS_RDONLY));
2014 BUG_ON(!mirror_num);
2016 /* we can't repair anything in raid56 yet */
2017 if (btrfs_is_parity_mirror(map_tree, logical, length, mirror_num))
2020 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
2023 bio->bi_iter.bi_size = 0;
2024 map_length = length;
2027 * Avoid races with device replace and make sure our bbio has devices
2028 * associated to its stripes that don't go away while we are doing the
2029 * read repair operation.
2031 btrfs_bio_counter_inc_blocked(fs_info);
2032 ret = btrfs_map_block(fs_info, WRITE, logical,
2033 &map_length, &bbio, mirror_num);
2035 btrfs_bio_counter_dec(fs_info);
2039 BUG_ON(mirror_num != bbio->mirror_num);
2040 sector = bbio->stripes[mirror_num-1].physical >> 9;
2041 bio->bi_iter.bi_sector = sector;
2042 dev = bbio->stripes[mirror_num-1].dev;
2043 btrfs_put_bbio(bbio);
2044 if (!dev || !dev->bdev || !dev->writeable) {
2045 btrfs_bio_counter_dec(fs_info);
2049 bio->bi_bdev = dev->bdev;
2050 bio_set_op_attrs(bio, REQ_OP_WRITE, WRITE_SYNC);
2051 bio_add_page(bio, page, length, pg_offset);
2053 if (btrfsic_submit_bio_wait(bio)) {
2054 /* try to remap that extent elsewhere? */
2055 btrfs_bio_counter_dec(fs_info);
2057 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
2061 btrfs_info_rl_in_rcu(fs_info,
2062 "read error corrected: ino %llu off %llu (dev %s sector %llu)",
2063 btrfs_ino(inode), start,
2064 rcu_str_deref(dev->name), sector);
2065 btrfs_bio_counter_dec(fs_info);
2070 int repair_eb_io_failure(struct btrfs_root *root, struct extent_buffer *eb,
2073 u64 start = eb->start;
2074 unsigned long i, num_pages = num_extent_pages(eb->start, eb->len);
2077 if (root->fs_info->sb->s_flags & MS_RDONLY)
2080 for (i = 0; i < num_pages; i++) {
2081 struct page *p = eb->pages[i];
2083 ret = repair_io_failure(root->fs_info->btree_inode, start,
2084 PAGE_SIZE, start, p,
2085 start - page_offset(p), mirror_num);
2095 * each time an IO finishes, we do a fast check in the IO failure tree
2096 * to see if we need to process or clean up an io_failure_record
2098 int clean_io_failure(struct inode *inode, u64 start, struct page *page,
2099 unsigned int pg_offset)
2102 struct io_failure_record *failrec;
2103 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2104 struct extent_state *state;
2109 ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
2110 (u64)-1, 1, EXTENT_DIRTY, 0);
2114 ret = get_state_failrec(&BTRFS_I(inode)->io_failure_tree, start,
2119 BUG_ON(!failrec->this_mirror);
2121 if (failrec->in_validation) {
2122 /* there was no real error, just free the record */
2123 btrfs_debug(fs_info,
2124 "clean_io_failure: freeing dummy error at %llu",
2128 if (fs_info->sb->s_flags & MS_RDONLY)
2131 spin_lock(&BTRFS_I(inode)->io_tree.lock);
2132 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
2135 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
2137 if (state && state->start <= failrec->start &&
2138 state->end >= failrec->start + failrec->len - 1) {
2139 num_copies = btrfs_num_copies(fs_info, failrec->logical,
2141 if (num_copies > 1) {
2142 repair_io_failure(inode, start, failrec->len,
2143 failrec->logical, page,
2144 pg_offset, failrec->failed_mirror);
2149 free_io_failure(inode, failrec);
2155 * Can be called when
2156 * - hold extent lock
2157 * - under ordered extent
2158 * - the inode is freeing
2160 void btrfs_free_io_failure_record(struct inode *inode, u64 start, u64 end)
2162 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2163 struct io_failure_record *failrec;
2164 struct extent_state *state, *next;
2166 if (RB_EMPTY_ROOT(&failure_tree->state))
2169 spin_lock(&failure_tree->lock);
2170 state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY);
2172 if (state->start > end)
2175 ASSERT(state->end <= end);
2177 next = next_state(state);
2179 failrec = state->failrec;
2180 free_extent_state(state);
2185 spin_unlock(&failure_tree->lock);
2188 int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end,
2189 struct io_failure_record **failrec_ret)
2191 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2192 struct io_failure_record *failrec;
2193 struct extent_map *em;
2194 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2195 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2196 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2200 ret = get_state_failrec(failure_tree, start, &failrec);
2202 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2206 failrec->start = start;
2207 failrec->len = end - start + 1;
2208 failrec->this_mirror = 0;
2209 failrec->bio_flags = 0;
2210 failrec->in_validation = 0;
2212 read_lock(&em_tree->lock);
2213 em = lookup_extent_mapping(em_tree, start, failrec->len);
2215 read_unlock(&em_tree->lock);
2220 if (em->start > start || em->start + em->len <= start) {
2221 free_extent_map(em);
2224 read_unlock(&em_tree->lock);
2230 logical = start - em->start;
2231 logical = em->block_start + logical;
2232 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2233 logical = em->block_start;
2234 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2235 extent_set_compress_type(&failrec->bio_flags,
2239 btrfs_debug(fs_info,
2240 "Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu",
2241 logical, start, failrec->len);
2243 failrec->logical = logical;
2244 free_extent_map(em);
2246 /* set the bits in the private failure tree */
2247 ret = set_extent_bits(failure_tree, start, end,
2248 EXTENT_LOCKED | EXTENT_DIRTY);
2250 ret = set_state_failrec(failure_tree, start, failrec);
2251 /* set the bits in the inode's tree */
2253 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED);
2259 btrfs_debug(fs_info,
2260 "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d",
2261 failrec->logical, failrec->start, failrec->len,
2262 failrec->in_validation);
2264 * when data can be on disk more than twice, add to failrec here
2265 * (e.g. with a list for failed_mirror) to make
2266 * clean_io_failure() clean all those errors at once.
2270 *failrec_ret = failrec;
2275 int btrfs_check_repairable(struct inode *inode, struct bio *failed_bio,
2276 struct io_failure_record *failrec, int failed_mirror)
2278 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2281 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
2282 if (num_copies == 1) {
2284 * we only have a single copy of the data, so don't bother with
2285 * all the retry and error correction code that follows. no
2286 * matter what the error is, it is very likely to persist.
2288 btrfs_debug(fs_info,
2289 "Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
2290 num_copies, failrec->this_mirror, failed_mirror);
2295 * there are two premises:
2296 * a) deliver good data to the caller
2297 * b) correct the bad sectors on disk
2299 if (failed_bio->bi_vcnt > 1) {
2301 * to fulfill b), we need to know the exact failing sectors, as
2302 * we don't want to rewrite any more than the failed ones. thus,
2303 * we need separate read requests for the failed bio
2305 * if the following BUG_ON triggers, our validation request got
2306 * merged. we need separate requests for our algorithm to work.
2308 BUG_ON(failrec->in_validation);
2309 failrec->in_validation = 1;
2310 failrec->this_mirror = failed_mirror;
2313 * we're ready to fulfill a) and b) alongside. get a good copy
2314 * of the failed sector and if we succeed, we have setup
2315 * everything for repair_io_failure to do the rest for us.
2317 if (failrec->in_validation) {
2318 BUG_ON(failrec->this_mirror != failed_mirror);
2319 failrec->in_validation = 0;
2320 failrec->this_mirror = 0;
2322 failrec->failed_mirror = failed_mirror;
2323 failrec->this_mirror++;
2324 if (failrec->this_mirror == failed_mirror)
2325 failrec->this_mirror++;
2328 if (failrec->this_mirror > num_copies) {
2329 btrfs_debug(fs_info,
2330 "Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
2331 num_copies, failrec->this_mirror, failed_mirror);
2339 struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio,
2340 struct io_failure_record *failrec,
2341 struct page *page, int pg_offset, int icsum,
2342 bio_end_io_t *endio_func, void *data)
2345 struct btrfs_io_bio *btrfs_failed_bio;
2346 struct btrfs_io_bio *btrfs_bio;
2348 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
2352 bio->bi_end_io = endio_func;
2353 bio->bi_iter.bi_sector = failrec->logical >> 9;
2354 bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
2355 bio->bi_iter.bi_size = 0;
2356 bio->bi_private = data;
2358 btrfs_failed_bio = btrfs_io_bio(failed_bio);
2359 if (btrfs_failed_bio->csum) {
2360 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2361 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
2363 btrfs_bio = btrfs_io_bio(bio);
2364 btrfs_bio->csum = btrfs_bio->csum_inline;
2366 memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + icsum,
2370 bio_add_page(bio, page, failrec->len, pg_offset);
2376 * this is a generic handler for readpage errors (default
2377 * readpage_io_failed_hook). if other copies exist, read those and write back
2378 * good data to the failed position. does not investigate in remapping the
2379 * failed extent elsewhere, hoping the device will be smart enough to do this as
2383 static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset,
2384 struct page *page, u64 start, u64 end,
2387 struct io_failure_record *failrec;
2388 struct inode *inode = page->mapping->host;
2389 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2394 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
2396 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
2400 ret = btrfs_check_repairable(inode, failed_bio, failrec, failed_mirror);
2402 free_io_failure(inode, failrec);
2406 if (failed_bio->bi_vcnt > 1)
2407 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
2409 read_mode = READ_SYNC;
2411 phy_offset >>= inode->i_sb->s_blocksize_bits;
2412 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
2413 start - page_offset(page),
2414 (int)phy_offset, failed_bio->bi_end_io,
2417 free_io_failure(inode, failrec);
2420 bio_set_op_attrs(bio, REQ_OP_READ, read_mode);
2422 btrfs_debug(btrfs_sb(inode->i_sb),
2423 "Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d",
2424 read_mode, failrec->this_mirror, failrec->in_validation);
2426 ret = tree->ops->submit_bio_hook(inode, bio, failrec->this_mirror,
2427 failrec->bio_flags, 0);
2429 free_io_failure(inode, failrec);
2436 /* lots and lots of room for performance fixes in the end_bio funcs */
2438 void end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2440 int uptodate = (err == 0);
2441 struct extent_io_tree *tree;
2444 tree = &BTRFS_I(page->mapping->host)->io_tree;
2446 if (tree->ops && tree->ops->writepage_end_io_hook) {
2447 ret = tree->ops->writepage_end_io_hook(page, start,
2448 end, NULL, uptodate);
2454 ClearPageUptodate(page);
2456 ret = err < 0 ? err : -EIO;
2457 mapping_set_error(page->mapping, ret);
2462 * after a writepage IO is done, we need to:
2463 * clear the uptodate bits on error
2464 * clear the writeback bits in the extent tree for this IO
2465 * end_page_writeback if the page has no more pending IO
2467 * Scheduling is not allowed, so the extent state tree is expected
2468 * to have one and only one object corresponding to this IO.
2470 static void end_bio_extent_writepage(struct bio *bio)
2472 struct bio_vec *bvec;
2477 bio_for_each_segment_all(bvec, bio, i) {
2478 struct page *page = bvec->bv_page;
2480 /* We always issue full-page reads, but if some block
2481 * in a page fails to read, blk_update_request() will
2482 * advance bv_offset and adjust bv_len to compensate.
2483 * Print a warning for nonzero offsets, and an error
2484 * if they don't add up to a full page. */
2485 if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) {
2486 if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE)
2487 btrfs_err(BTRFS_I(page->mapping->host)->root->fs_info,
2488 "partial page write in btrfs with offset %u and length %u",
2489 bvec->bv_offset, bvec->bv_len);
2491 btrfs_info(BTRFS_I(page->mapping->host)->root->fs_info,
2492 "incomplete page write in btrfs with offset %u and length %u",
2493 bvec->bv_offset, bvec->bv_len);
2496 start = page_offset(page);
2497 end = start + bvec->bv_offset + bvec->bv_len - 1;
2499 end_extent_writepage(page, bio->bi_error, start, end);
2500 end_page_writeback(page);
2507 endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len,
2510 struct extent_state *cached = NULL;
2511 u64 end = start + len - 1;
2513 if (uptodate && tree->track_uptodate)
2514 set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC);
2515 unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC);
2519 * after a readpage IO is done, we need to:
2520 * clear the uptodate bits on error
2521 * set the uptodate bits if things worked
2522 * set the page up to date if all extents in the tree are uptodate
2523 * clear the lock bit in the extent tree
2524 * unlock the page if there are no other extents locked for it
2526 * Scheduling is not allowed, so the extent state tree is expected
2527 * to have one and only one object corresponding to this IO.
2529 static void end_bio_extent_readpage(struct bio *bio)
2531 struct bio_vec *bvec;
2532 int uptodate = !bio->bi_error;
2533 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
2534 struct extent_io_tree *tree;
2539 u64 extent_start = 0;
2545 bio_for_each_segment_all(bvec, bio, i) {
2546 struct page *page = bvec->bv_page;
2547 struct inode *inode = page->mapping->host;
2548 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2550 btrfs_debug(fs_info,
2551 "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u",
2552 (u64)bio->bi_iter.bi_sector, bio->bi_error,
2553 io_bio->mirror_num);
2554 tree = &BTRFS_I(inode)->io_tree;
2556 /* We always issue full-page reads, but if some block
2557 * in a page fails to read, blk_update_request() will
2558 * advance bv_offset and adjust bv_len to compensate.
2559 * Print a warning for nonzero offsets, and an error
2560 * if they don't add up to a full page. */
2561 if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) {
2562 if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE)
2564 "partial page read in btrfs with offset %u and length %u",
2565 bvec->bv_offset, bvec->bv_len);
2568 "incomplete page read in btrfs with offset %u and length %u",
2569 bvec->bv_offset, bvec->bv_len);
2572 start = page_offset(page);
2573 end = start + bvec->bv_offset + bvec->bv_len - 1;
2576 mirror = io_bio->mirror_num;
2577 if (likely(uptodate && tree->ops &&
2578 tree->ops->readpage_end_io_hook)) {
2579 ret = tree->ops->readpage_end_io_hook(io_bio, offset,
2585 clean_io_failure(inode, start, page, 0);
2588 if (likely(uptodate))
2591 if (tree->ops && tree->ops->readpage_io_failed_hook) {
2592 ret = tree->ops->readpage_io_failed_hook(page, mirror);
2593 if (!ret && !bio->bi_error)
2597 * The generic bio_readpage_error handles errors the
2598 * following way: If possible, new read requests are
2599 * created and submitted and will end up in
2600 * end_bio_extent_readpage as well (if we're lucky, not
2601 * in the !uptodate case). In that case it returns 0 and
2602 * we just go on with the next page in our bio. If it
2603 * can't handle the error it will return -EIO and we
2604 * remain responsible for that page.
2606 ret = bio_readpage_error(bio, offset, page, start, end,
2609 uptodate = !bio->bi_error;
2615 if (likely(uptodate)) {
2616 loff_t i_size = i_size_read(inode);
2617 pgoff_t end_index = i_size >> PAGE_SHIFT;
2620 /* Zero out the end if this page straddles i_size */
2621 off = i_size & (PAGE_SIZE-1);
2622 if (page->index == end_index && off)
2623 zero_user_segment(page, off, PAGE_SIZE);
2624 SetPageUptodate(page);
2626 ClearPageUptodate(page);
2632 if (unlikely(!uptodate)) {
2634 endio_readpage_release_extent(tree,
2640 endio_readpage_release_extent(tree, start,
2641 end - start + 1, 0);
2642 } else if (!extent_len) {
2643 extent_start = start;
2644 extent_len = end + 1 - start;
2645 } else if (extent_start + extent_len == start) {
2646 extent_len += end + 1 - start;
2648 endio_readpage_release_extent(tree, extent_start,
2649 extent_len, uptodate);
2650 extent_start = start;
2651 extent_len = end + 1 - start;
2656 endio_readpage_release_extent(tree, extent_start, extent_len,
2659 io_bio->end_io(io_bio, bio->bi_error);
2664 * this allocates from the btrfs_bioset. We're returning a bio right now
2665 * but you can call btrfs_io_bio for the appropriate container_of magic
2668 btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
2671 struct btrfs_io_bio *btrfs_bio;
2674 bio = bio_alloc_bioset(gfp_flags, nr_vecs, btrfs_bioset);
2676 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
2677 while (!bio && (nr_vecs /= 2)) {
2678 bio = bio_alloc_bioset(gfp_flags,
2679 nr_vecs, btrfs_bioset);
2684 bio->bi_bdev = bdev;
2685 bio->bi_iter.bi_sector = first_sector;
2686 btrfs_bio = btrfs_io_bio(bio);
2687 btrfs_bio->csum = NULL;
2688 btrfs_bio->csum_allocated = NULL;
2689 btrfs_bio->end_io = NULL;
2694 struct bio *btrfs_bio_clone(struct bio *bio, gfp_t gfp_mask)
2696 struct btrfs_io_bio *btrfs_bio;
2699 new = bio_clone_bioset(bio, gfp_mask, btrfs_bioset);
2701 btrfs_bio = btrfs_io_bio(new);
2702 btrfs_bio->csum = NULL;
2703 btrfs_bio->csum_allocated = NULL;
2704 btrfs_bio->end_io = NULL;
2709 /* this also allocates from the btrfs_bioset */
2710 struct bio *btrfs_io_bio_alloc(gfp_t gfp_mask, unsigned int nr_iovecs)
2712 struct btrfs_io_bio *btrfs_bio;
2715 bio = bio_alloc_bioset(gfp_mask, nr_iovecs, btrfs_bioset);
2717 btrfs_bio = btrfs_io_bio(bio);
2718 btrfs_bio->csum = NULL;
2719 btrfs_bio->csum_allocated = NULL;
2720 btrfs_bio->end_io = NULL;
2726 static int __must_check submit_one_bio(struct bio *bio, int mirror_num,
2727 unsigned long bio_flags)
2730 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2731 struct page *page = bvec->bv_page;
2732 struct extent_io_tree *tree = bio->bi_private;
2735 start = page_offset(page) + bvec->bv_offset;
2737 bio->bi_private = NULL;
2740 if (tree->ops && tree->ops->submit_bio_hook)
2741 ret = tree->ops->submit_bio_hook(page->mapping->host, bio,
2742 mirror_num, bio_flags, start);
2744 btrfsic_submit_bio(bio);
2750 static int merge_bio(struct extent_io_tree *tree, struct page *page,
2751 unsigned long offset, size_t size, struct bio *bio,
2752 unsigned long bio_flags)
2755 if (tree->ops && tree->ops->merge_bio_hook)
2756 ret = tree->ops->merge_bio_hook(page, offset, size, bio,
2762 static int submit_extent_page(int op, int op_flags, struct extent_io_tree *tree,
2763 struct writeback_control *wbc,
2764 struct page *page, sector_t sector,
2765 size_t size, unsigned long offset,
2766 struct block_device *bdev,
2767 struct bio **bio_ret,
2768 unsigned long max_pages,
2769 bio_end_io_t end_io_func,
2771 unsigned long prev_bio_flags,
2772 unsigned long bio_flags,
2773 bool force_bio_submit)
2778 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
2779 size_t page_size = min_t(size_t, size, PAGE_SIZE);
2781 if (bio_ret && *bio_ret) {
2784 contig = bio->bi_iter.bi_sector == sector;
2786 contig = bio_end_sector(bio) == sector;
2788 if (prev_bio_flags != bio_flags || !contig ||
2790 merge_bio(tree, page, offset, page_size, bio, bio_flags) ||
2791 bio_add_page(bio, page, page_size, offset) < page_size) {
2792 ret = submit_one_bio(bio, mirror_num, prev_bio_flags);
2800 wbc_account_io(wbc, page, page_size);
2805 bio = btrfs_bio_alloc(bdev, sector, BIO_MAX_PAGES,
2806 GFP_NOFS | __GFP_HIGH);
2810 bio_add_page(bio, page, page_size, offset);
2811 bio->bi_end_io = end_io_func;
2812 bio->bi_private = tree;
2813 bio_set_op_attrs(bio, op, op_flags);
2815 wbc_init_bio(wbc, bio);
2816 wbc_account_io(wbc, page, page_size);
2822 ret = submit_one_bio(bio, mirror_num, bio_flags);
2827 static void attach_extent_buffer_page(struct extent_buffer *eb,
2830 if (!PagePrivate(page)) {
2831 SetPagePrivate(page);
2833 set_page_private(page, (unsigned long)eb);
2835 WARN_ON(page->private != (unsigned long)eb);
2839 void set_page_extent_mapped(struct page *page)
2841 if (!PagePrivate(page)) {
2842 SetPagePrivate(page);
2844 set_page_private(page, EXTENT_PAGE_PRIVATE);
2848 static struct extent_map *
2849 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
2850 u64 start, u64 len, get_extent_t *get_extent,
2851 struct extent_map **em_cached)
2853 struct extent_map *em;
2855 if (em_cached && *em_cached) {
2857 if (extent_map_in_tree(em) && start >= em->start &&
2858 start < extent_map_end(em)) {
2859 atomic_inc(&em->refs);
2863 free_extent_map(em);
2867 em = get_extent(inode, page, pg_offset, start, len, 0);
2868 if (em_cached && !IS_ERR_OR_NULL(em)) {
2870 atomic_inc(&em->refs);
2876 * basic readpage implementation. Locked extent state structs are inserted
2877 * into the tree that are removed when the IO is done (by the end_io
2879 * XXX JDM: This needs looking at to ensure proper page locking
2880 * return 0 on success, otherwise return error
2882 static int __do_readpage(struct extent_io_tree *tree,
2884 get_extent_t *get_extent,
2885 struct extent_map **em_cached,
2886 struct bio **bio, int mirror_num,
2887 unsigned long *bio_flags, int read_flags,
2890 struct inode *inode = page->mapping->host;
2891 u64 start = page_offset(page);
2892 u64 page_end = start + PAGE_SIZE - 1;
2896 u64 last_byte = i_size_read(inode);
2900 struct extent_map *em;
2901 struct block_device *bdev;
2904 size_t pg_offset = 0;
2906 size_t disk_io_size;
2907 size_t blocksize = inode->i_sb->s_blocksize;
2908 unsigned long this_bio_flag = 0;
2910 set_page_extent_mapped(page);
2913 if (!PageUptodate(page)) {
2914 if (cleancache_get_page(page) == 0) {
2915 BUG_ON(blocksize != PAGE_SIZE);
2916 unlock_extent(tree, start, end);
2921 if (page->index == last_byte >> PAGE_SHIFT) {
2923 size_t zero_offset = last_byte & (PAGE_SIZE - 1);
2926 iosize = PAGE_SIZE - zero_offset;
2927 userpage = kmap_atomic(page);
2928 memset(userpage + zero_offset, 0, iosize);
2929 flush_dcache_page(page);
2930 kunmap_atomic(userpage);
2933 while (cur <= end) {
2934 unsigned long pnr = (last_byte >> PAGE_SHIFT) + 1;
2935 bool force_bio_submit = false;
2937 if (cur >= last_byte) {
2939 struct extent_state *cached = NULL;
2941 iosize = PAGE_SIZE - pg_offset;
2942 userpage = kmap_atomic(page);
2943 memset(userpage + pg_offset, 0, iosize);
2944 flush_dcache_page(page);
2945 kunmap_atomic(userpage);
2946 set_extent_uptodate(tree, cur, cur + iosize - 1,
2948 unlock_extent_cached(tree, cur,
2953 em = __get_extent_map(inode, page, pg_offset, cur,
2954 end - cur + 1, get_extent, em_cached);
2955 if (IS_ERR_OR_NULL(em)) {
2957 unlock_extent(tree, cur, end);
2960 extent_offset = cur - em->start;
2961 BUG_ON(extent_map_end(em) <= cur);
2964 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2965 this_bio_flag |= EXTENT_BIO_COMPRESSED;
2966 extent_set_compress_type(&this_bio_flag,
2970 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2971 cur_end = min(extent_map_end(em) - 1, end);
2972 iosize = ALIGN(iosize, blocksize);
2973 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2974 disk_io_size = em->block_len;
2975 sector = em->block_start >> 9;
2977 sector = (em->block_start + extent_offset) >> 9;
2978 disk_io_size = iosize;
2981 block_start = em->block_start;
2982 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2983 block_start = EXTENT_MAP_HOLE;
2986 * If we have a file range that points to a compressed extent
2987 * and it's followed by a consecutive file range that points to
2988 * to the same compressed extent (possibly with a different
2989 * offset and/or length, so it either points to the whole extent
2990 * or only part of it), we must make sure we do not submit a
2991 * single bio to populate the pages for the 2 ranges because
2992 * this makes the compressed extent read zero out the pages
2993 * belonging to the 2nd range. Imagine the following scenario:
2996 * [0 - 8K] [8K - 24K]
2999 * points to extent X, points to extent X,
3000 * offset 4K, length of 8K offset 0, length 16K
3002 * [extent X, compressed length = 4K uncompressed length = 16K]
3004 * If the bio to read the compressed extent covers both ranges,
3005 * it will decompress extent X into the pages belonging to the
3006 * first range and then it will stop, zeroing out the remaining
3007 * pages that belong to the other range that points to extent X.
3008 * So here we make sure we submit 2 bios, one for the first
3009 * range and another one for the third range. Both will target
3010 * the same physical extent from disk, but we can't currently
3011 * make the compressed bio endio callback populate the pages
3012 * for both ranges because each compressed bio is tightly
3013 * coupled with a single extent map, and each range can have
3014 * an extent map with a different offset value relative to the
3015 * uncompressed data of our extent and different lengths. This
3016 * is a corner case so we prioritize correctness over
3017 * non-optimal behavior (submitting 2 bios for the same extent).
3019 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) &&
3020 prev_em_start && *prev_em_start != (u64)-1 &&
3021 *prev_em_start != em->start)
3022 force_bio_submit = true;
3025 *prev_em_start = em->start;
3027 free_extent_map(em);
3030 /* we've found a hole, just zero and go on */
3031 if (block_start == EXTENT_MAP_HOLE) {
3033 struct extent_state *cached = NULL;
3035 userpage = kmap_atomic(page);
3036 memset(userpage + pg_offset, 0, iosize);
3037 flush_dcache_page(page);
3038 kunmap_atomic(userpage);
3040 set_extent_uptodate(tree, cur, cur + iosize - 1,
3042 unlock_extent_cached(tree, cur,
3046 pg_offset += iosize;
3049 /* the get_extent function already copied into the page */
3050 if (test_range_bit(tree, cur, cur_end,
3051 EXTENT_UPTODATE, 1, NULL)) {
3052 check_page_uptodate(tree, page);
3053 unlock_extent(tree, cur, cur + iosize - 1);
3055 pg_offset += iosize;
3058 /* we have an inline extent but it didn't get marked up
3059 * to date. Error out
3061 if (block_start == EXTENT_MAP_INLINE) {
3063 unlock_extent(tree, cur, cur + iosize - 1);
3065 pg_offset += iosize;
3070 ret = submit_extent_page(REQ_OP_READ, read_flags, tree, NULL,
3071 page, sector, disk_io_size, pg_offset,
3073 end_bio_extent_readpage, mirror_num,
3079 *bio_flags = this_bio_flag;
3082 unlock_extent(tree, cur, cur + iosize - 1);
3086 pg_offset += iosize;
3090 if (!PageError(page))
3091 SetPageUptodate(page);
3097 static inline void __do_contiguous_readpages(struct extent_io_tree *tree,
3098 struct page *pages[], int nr_pages,
3100 get_extent_t *get_extent,
3101 struct extent_map **em_cached,
3102 struct bio **bio, int mirror_num,
3103 unsigned long *bio_flags,
3106 struct inode *inode;
3107 struct btrfs_ordered_extent *ordered;
3110 inode = pages[0]->mapping->host;
3112 lock_extent(tree, start, end);
3113 ordered = btrfs_lookup_ordered_range(inode, start,
3117 unlock_extent(tree, start, end);
3118 btrfs_start_ordered_extent(inode, ordered, 1);
3119 btrfs_put_ordered_extent(ordered);
3122 for (index = 0; index < nr_pages; index++) {
3123 __do_readpage(tree, pages[index], get_extent, em_cached, bio,
3124 mirror_num, bio_flags, 0, prev_em_start);
3125 put_page(pages[index]);
3129 static void __extent_readpages(struct extent_io_tree *tree,
3130 struct page *pages[],
3131 int nr_pages, get_extent_t *get_extent,
3132 struct extent_map **em_cached,
3133 struct bio **bio, int mirror_num,
3134 unsigned long *bio_flags,
3141 int first_index = 0;
3143 for (index = 0; index < nr_pages; index++) {
3144 page_start = page_offset(pages[index]);
3147 end = start + PAGE_SIZE - 1;
3148 first_index = index;
3149 } else if (end + 1 == page_start) {
3152 __do_contiguous_readpages(tree, &pages[first_index],
3153 index - first_index, start,
3154 end, get_extent, em_cached,
3155 bio, mirror_num, bio_flags,
3158 end = start + PAGE_SIZE - 1;
3159 first_index = index;
3164 __do_contiguous_readpages(tree, &pages[first_index],
3165 index - first_index, start,
3166 end, get_extent, em_cached, bio,
3167 mirror_num, bio_flags,
3171 static int __extent_read_full_page(struct extent_io_tree *tree,
3173 get_extent_t *get_extent,
3174 struct bio **bio, int mirror_num,
3175 unsigned long *bio_flags, int read_flags)
3177 struct inode *inode = page->mapping->host;
3178 struct btrfs_ordered_extent *ordered;
3179 u64 start = page_offset(page);
3180 u64 end = start + PAGE_SIZE - 1;
3184 lock_extent(tree, start, end);
3185 ordered = btrfs_lookup_ordered_range(inode, start,
3189 unlock_extent(tree, start, end);
3190 btrfs_start_ordered_extent(inode, ordered, 1);
3191 btrfs_put_ordered_extent(ordered);
3194 ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num,
3195 bio_flags, read_flags, NULL);
3199 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
3200 get_extent_t *get_extent, int mirror_num)
3202 struct bio *bio = NULL;
3203 unsigned long bio_flags = 0;
3206 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
3209 ret = submit_one_bio(bio, mirror_num, bio_flags);
3213 static void update_nr_written(struct page *page, struct writeback_control *wbc,
3214 unsigned long nr_written)
3216 wbc->nr_to_write -= nr_written;
3220 * helper for __extent_writepage, doing all of the delayed allocation setup.
3222 * This returns 1 if our fill_delalloc function did all the work required
3223 * to write the page (copy into inline extent). In this case the IO has
3224 * been started and the page is already unlocked.
3226 * This returns 0 if all went well (page still locked)
3227 * This returns < 0 if there were errors (page still locked)
3229 static noinline_for_stack int writepage_delalloc(struct inode *inode,
3230 struct page *page, struct writeback_control *wbc,
3231 struct extent_page_data *epd,
3233 unsigned long *nr_written)
3235 struct extent_io_tree *tree = epd->tree;
3236 u64 page_end = delalloc_start + PAGE_SIZE - 1;
3238 u64 delalloc_to_write = 0;
3239 u64 delalloc_end = 0;
3241 int page_started = 0;
3243 if (epd->extent_locked || !tree->ops || !tree->ops->fill_delalloc)
3246 while (delalloc_end < page_end) {
3247 nr_delalloc = find_lock_delalloc_range(inode, tree,
3251 BTRFS_MAX_EXTENT_SIZE);
3252 if (nr_delalloc == 0) {
3253 delalloc_start = delalloc_end + 1;
3256 ret = tree->ops->fill_delalloc(inode, page,
3261 /* File system has been set read-only */
3264 /* fill_delalloc should be return < 0 for error
3265 * but just in case, we use > 0 here meaning the
3266 * IO is started, so we don't want to return > 0
3267 * unless things are going well.
3269 ret = ret < 0 ? ret : -EIO;
3273 * delalloc_end is already one less than the total length, so
3274 * we don't subtract one from PAGE_SIZE
3276 delalloc_to_write += (delalloc_end - delalloc_start +
3277 PAGE_SIZE) >> PAGE_SHIFT;
3278 delalloc_start = delalloc_end + 1;
3280 if (wbc->nr_to_write < delalloc_to_write) {
3283 if (delalloc_to_write < thresh * 2)
3284 thresh = delalloc_to_write;
3285 wbc->nr_to_write = min_t(u64, delalloc_to_write,
3289 /* did the fill delalloc function already unlock and start
3294 * we've unlocked the page, so we can't update
3295 * the mapping's writeback index, just update
3298 wbc->nr_to_write -= *nr_written;
3309 * helper for __extent_writepage. This calls the writepage start hooks,
3310 * and does the loop to map the page into extents and bios.
3312 * We return 1 if the IO is started and the page is unlocked,
3313 * 0 if all went well (page still locked)
3314 * < 0 if there were errors (page still locked)
3316 static noinline_for_stack int __extent_writepage_io(struct inode *inode,
3318 struct writeback_control *wbc,
3319 struct extent_page_data *epd,
3321 unsigned long nr_written,
3322 int write_flags, int *nr_ret)
3324 struct extent_io_tree *tree = epd->tree;
3325 u64 start = page_offset(page);
3326 u64 page_end = start + PAGE_SIZE - 1;
3333 struct extent_state *cached_state = NULL;
3334 struct extent_map *em;
3335 struct block_device *bdev;
3336 size_t pg_offset = 0;
3342 if (tree->ops && tree->ops->writepage_start_hook) {
3343 ret = tree->ops->writepage_start_hook(page, start,
3346 /* Fixup worker will requeue */
3348 wbc->pages_skipped++;
3350 redirty_page_for_writepage(wbc, page);
3352 update_nr_written(page, wbc, nr_written);
3360 * we don't want to touch the inode after unlocking the page,
3361 * so we update the mapping writeback index now
3363 update_nr_written(page, wbc, nr_written + 1);
3366 if (i_size <= start) {
3367 if (tree->ops && tree->ops->writepage_end_io_hook)
3368 tree->ops->writepage_end_io_hook(page, start,
3373 blocksize = inode->i_sb->s_blocksize;
3375 while (cur <= end) {
3377 unsigned long max_nr;
3379 if (cur >= i_size) {
3380 if (tree->ops && tree->ops->writepage_end_io_hook)
3381 tree->ops->writepage_end_io_hook(page, cur,
3385 em = epd->get_extent(inode, page, pg_offset, cur,
3387 if (IS_ERR_OR_NULL(em)) {
3389 ret = PTR_ERR_OR_ZERO(em);
3393 extent_offset = cur - em->start;
3394 em_end = extent_map_end(em);
3395 BUG_ON(em_end <= cur);
3397 iosize = min(em_end - cur, end - cur + 1);
3398 iosize = ALIGN(iosize, blocksize);
3399 sector = (em->block_start + extent_offset) >> 9;
3401 block_start = em->block_start;
3402 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3403 free_extent_map(em);
3407 * compressed and inline extents are written through other
3410 if (compressed || block_start == EXTENT_MAP_HOLE ||
3411 block_start == EXTENT_MAP_INLINE) {
3413 * end_io notification does not happen here for
3414 * compressed extents
3416 if (!compressed && tree->ops &&
3417 tree->ops->writepage_end_io_hook)
3418 tree->ops->writepage_end_io_hook(page, cur,
3421 else if (compressed) {
3422 /* we don't want to end_page_writeback on
3423 * a compressed extent. this happens
3430 pg_offset += iosize;
3434 max_nr = (i_size >> PAGE_SHIFT) + 1;
3436 set_range_writeback(tree, cur, cur + iosize - 1);
3437 if (!PageWriteback(page)) {
3438 btrfs_err(BTRFS_I(inode)->root->fs_info,
3439 "page %lu not writeback, cur %llu end %llu",
3440 page->index, cur, end);
3443 ret = submit_extent_page(REQ_OP_WRITE, write_flags, tree, wbc,
3444 page, sector, iosize, pg_offset,
3445 bdev, &epd->bio, max_nr,
3446 end_bio_extent_writepage,
3452 pg_offset += iosize;
3460 /* drop our reference on any cached states */
3461 free_extent_state(cached_state);
3466 * the writepage semantics are similar to regular writepage. extent
3467 * records are inserted to lock ranges in the tree, and as dirty areas
3468 * are found, they are marked writeback. Then the lock bits are removed
3469 * and the end_io handler clears the writeback ranges
3471 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
3474 struct inode *inode = page->mapping->host;
3475 struct extent_page_data *epd = data;
3476 u64 start = page_offset(page);
3477 u64 page_end = start + PAGE_SIZE - 1;
3480 size_t pg_offset = 0;
3481 loff_t i_size = i_size_read(inode);
3482 unsigned long end_index = i_size >> PAGE_SHIFT;
3483 int write_flags = 0;
3484 unsigned long nr_written = 0;
3486 if (wbc->sync_mode == WB_SYNC_ALL)
3487 write_flags = WRITE_SYNC;
3489 trace___extent_writepage(page, inode, wbc);
3491 WARN_ON(!PageLocked(page));
3493 ClearPageError(page);
3495 pg_offset = i_size & (PAGE_SIZE - 1);
3496 if (page->index > end_index ||
3497 (page->index == end_index && !pg_offset)) {
3498 page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
3503 if (page->index == end_index) {
3506 userpage = kmap_atomic(page);
3507 memset(userpage + pg_offset, 0,
3508 PAGE_SIZE - pg_offset);
3509 kunmap_atomic(userpage);
3510 flush_dcache_page(page);
3515 set_page_extent_mapped(page);
3517 ret = writepage_delalloc(inode, page, wbc, epd, start, &nr_written);
3523 ret = __extent_writepage_io(inode, page, wbc, epd,
3524 i_size, nr_written, write_flags, &nr);
3530 /* make sure the mapping tag for page dirty gets cleared */
3531 set_page_writeback(page);
3532 end_page_writeback(page);
3534 if (PageError(page)) {
3535 ret = ret < 0 ? ret : -EIO;
3536 end_extent_writepage(page, ret, start, page_end);
3545 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3547 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
3548 TASK_UNINTERRUPTIBLE);
3551 static noinline_for_stack int
3552 lock_extent_buffer_for_io(struct extent_buffer *eb,
3553 struct btrfs_fs_info *fs_info,
3554 struct extent_page_data *epd)
3556 unsigned long i, num_pages;
3560 if (!btrfs_try_tree_write_lock(eb)) {
3562 flush_write_bio(epd);
3563 btrfs_tree_lock(eb);
3566 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3567 btrfs_tree_unlock(eb);
3571 flush_write_bio(epd);
3575 wait_on_extent_buffer_writeback(eb);
3576 btrfs_tree_lock(eb);
3577 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3579 btrfs_tree_unlock(eb);
3584 * We need to do this to prevent races in people who check if the eb is
3585 * under IO since we can end up having no IO bits set for a short period
3588 spin_lock(&eb->refs_lock);
3589 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3590 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3591 spin_unlock(&eb->refs_lock);
3592 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3593 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
3595 fs_info->dirty_metadata_batch);
3598 spin_unlock(&eb->refs_lock);
3601 btrfs_tree_unlock(eb);
3606 num_pages = num_extent_pages(eb->start, eb->len);
3607 for (i = 0; i < num_pages; i++) {
3608 struct page *p = eb->pages[i];
3610 if (!trylock_page(p)) {
3612 flush_write_bio(epd);
3622 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3624 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3625 smp_mb__after_atomic();
3626 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3629 static void set_btree_ioerr(struct page *page)
3631 struct extent_buffer *eb = (struct extent_buffer *)page->private;
3634 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))
3638 * If writeback for a btree extent that doesn't belong to a log tree
3639 * failed, increment the counter transaction->eb_write_errors.
3640 * We do this because while the transaction is running and before it's
3641 * committing (when we call filemap_fdata[write|wait]_range against
3642 * the btree inode), we might have
3643 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
3644 * returns an error or an error happens during writeback, when we're
3645 * committing the transaction we wouldn't know about it, since the pages
3646 * can be no longer dirty nor marked anymore for writeback (if a
3647 * subsequent modification to the extent buffer didn't happen before the
3648 * transaction commit), which makes filemap_fdata[write|wait]_range not
3649 * able to find the pages tagged with SetPageError at transaction
3650 * commit time. So if this happens we must abort the transaction,
3651 * otherwise we commit a super block with btree roots that point to
3652 * btree nodes/leafs whose content on disk is invalid - either garbage
3653 * or the content of some node/leaf from a past generation that got
3654 * cowed or deleted and is no longer valid.
3656 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
3657 * not be enough - we need to distinguish between log tree extents vs
3658 * non-log tree extents, and the next filemap_fdatawait_range() call
3659 * will catch and clear such errors in the mapping - and that call might
3660 * be from a log sync and not from a transaction commit. Also, checking
3661 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
3662 * not done and would not be reliable - the eb might have been released
3663 * from memory and reading it back again means that flag would not be
3664 * set (since it's a runtime flag, not persisted on disk).
3666 * Using the flags below in the btree inode also makes us achieve the
3667 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
3668 * writeback for all dirty pages and before filemap_fdatawait_range()
3669 * is called, the writeback for all dirty pages had already finished
3670 * with errors - because we were not using AS_EIO/AS_ENOSPC,
3671 * filemap_fdatawait_range() would return success, as it could not know
3672 * that writeback errors happened (the pages were no longer tagged for
3675 switch (eb->log_index) {
3677 set_bit(BTRFS_FS_BTREE_ERR, &eb->fs_info->flags);
3680 set_bit(BTRFS_FS_LOG1_ERR, &eb->fs_info->flags);
3683 set_bit(BTRFS_FS_LOG2_ERR, &eb->fs_info->flags);
3686 BUG(); /* unexpected, logic error */
3690 static void end_bio_extent_buffer_writepage(struct bio *bio)
3692 struct bio_vec *bvec;
3693 struct extent_buffer *eb;
3696 bio_for_each_segment_all(bvec, bio, i) {
3697 struct page *page = bvec->bv_page;
3699 eb = (struct extent_buffer *)page->private;
3701 done = atomic_dec_and_test(&eb->io_pages);
3703 if (bio->bi_error ||
3704 test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) {
3705 ClearPageUptodate(page);
3706 set_btree_ioerr(page);
3709 end_page_writeback(page);
3714 end_extent_buffer_writeback(eb);
3720 static noinline_for_stack int write_one_eb(struct extent_buffer *eb,
3721 struct btrfs_fs_info *fs_info,
3722 struct writeback_control *wbc,
3723 struct extent_page_data *epd)
3725 struct block_device *bdev = fs_info->fs_devices->latest_bdev;
3726 struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
3727 u64 offset = eb->start;
3729 unsigned long i, num_pages;
3730 unsigned long bio_flags = 0;
3731 unsigned long start, end;
3732 int write_flags = (epd->sync_io ? WRITE_SYNC : 0) | REQ_META;
3735 clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
3736 num_pages = num_extent_pages(eb->start, eb->len);
3737 atomic_set(&eb->io_pages, num_pages);
3738 if (btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID)
3739 bio_flags = EXTENT_BIO_TREE_LOG;
3741 /* set btree blocks beyond nritems with 0 to avoid stale content. */
3742 nritems = btrfs_header_nritems(eb);
3743 if (btrfs_header_level(eb) > 0) {
3744 end = btrfs_node_key_ptr_offset(nritems);
3746 memset_extent_buffer(eb, 0, end, eb->len - end);
3750 * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0
3752 start = btrfs_item_nr_offset(nritems);
3753 end = btrfs_leaf_data(eb) +
3754 leaf_data_end(fs_info->tree_root, eb);
3755 memset_extent_buffer(eb, 0, start, end - start);
3758 for (i = 0; i < num_pages; i++) {
3759 struct page *p = eb->pages[i];
3761 clear_page_dirty_for_io(p);
3762 set_page_writeback(p);
3763 ret = submit_extent_page(REQ_OP_WRITE, write_flags, tree, wbc,
3764 p, offset >> 9, PAGE_SIZE, 0, bdev,
3766 end_bio_extent_buffer_writepage,
3767 0, epd->bio_flags, bio_flags, false);
3768 epd->bio_flags = bio_flags;
3771 end_page_writeback(p);
3772 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3773 end_extent_buffer_writeback(eb);
3777 offset += PAGE_SIZE;
3778 update_nr_written(p, wbc, 1);
3782 if (unlikely(ret)) {
3783 for (; i < num_pages; i++) {
3784 struct page *p = eb->pages[i];
3785 clear_page_dirty_for_io(p);
3793 int btree_write_cache_pages(struct address_space *mapping,
3794 struct writeback_control *wbc)
3796 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
3797 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
3798 struct extent_buffer *eb, *prev_eb = NULL;
3799 struct extent_page_data epd = {
3803 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3808 int nr_to_write_done = 0;
3809 struct pagevec pvec;
3812 pgoff_t end; /* Inclusive */
3816 pagevec_init(&pvec, 0);
3817 if (wbc->range_cyclic) {
3818 index = mapping->writeback_index; /* Start from prev offset */
3821 index = wbc->range_start >> PAGE_SHIFT;
3822 end = wbc->range_end >> PAGE_SHIFT;
3825 if (wbc->sync_mode == WB_SYNC_ALL)
3826 tag = PAGECACHE_TAG_TOWRITE;
3828 tag = PAGECACHE_TAG_DIRTY;
3830 if (wbc->sync_mode == WB_SYNC_ALL)
3831 tag_pages_for_writeback(mapping, index, end);
3832 while (!done && !nr_to_write_done && (index <= end) &&
3833 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3834 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3838 for (i = 0; i < nr_pages; i++) {
3839 struct page *page = pvec.pages[i];
3841 if (!PagePrivate(page))
3844 if (!wbc->range_cyclic && page->index > end) {
3849 spin_lock(&mapping->private_lock);
3850 if (!PagePrivate(page)) {
3851 spin_unlock(&mapping->private_lock);
3855 eb = (struct extent_buffer *)page->private;
3858 * Shouldn't happen and normally this would be a BUG_ON
3859 * but no sense in crashing the users box for something
3860 * we can survive anyway.
3863 spin_unlock(&mapping->private_lock);
3867 if (eb == prev_eb) {
3868 spin_unlock(&mapping->private_lock);
3872 ret = atomic_inc_not_zero(&eb->refs);
3873 spin_unlock(&mapping->private_lock);
3878 ret = lock_extent_buffer_for_io(eb, fs_info, &epd);
3880 free_extent_buffer(eb);
3882 } else if (ret < 0) {
3884 free_extent_buffer(eb);
3888 ret = write_one_eb(eb, fs_info, wbc, &epd);
3891 free_extent_buffer(eb);
3894 free_extent_buffer(eb);
3897 * the filesystem may choose to bump up nr_to_write.
3898 * We have to make sure to honor the new nr_to_write
3901 nr_to_write_done = wbc->nr_to_write <= 0;
3903 pagevec_release(&pvec);
3906 if (!scanned && !done) {
3908 * We hit the last page and there is more work to be done: wrap
3909 * back to the start of the file
3915 flush_write_bio(&epd);
3920 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3921 * @mapping: address space structure to write
3922 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3923 * @writepage: function called for each page
3924 * @data: data passed to writepage function
3926 * If a page is already under I/O, write_cache_pages() skips it, even
3927 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3928 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3929 * and msync() need to guarantee that all the data which was dirty at the time
3930 * the call was made get new I/O started against them. If wbc->sync_mode is
3931 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3932 * existing IO to complete.
3934 static int extent_write_cache_pages(struct extent_io_tree *tree,
3935 struct address_space *mapping,
3936 struct writeback_control *wbc,
3937 writepage_t writepage, void *data,
3938 void (*flush_fn)(void *))
3940 struct inode *inode = mapping->host;
3943 int nr_to_write_done = 0;
3944 struct pagevec pvec;
3947 pgoff_t end; /* Inclusive */
3949 int range_whole = 0;
3954 * We have to hold onto the inode so that ordered extents can do their
3955 * work when the IO finishes. The alternative to this is failing to add
3956 * an ordered extent if the igrab() fails there and that is a huge pain
3957 * to deal with, so instead just hold onto the inode throughout the
3958 * writepages operation. If it fails here we are freeing up the inode
3959 * anyway and we'd rather not waste our time writing out stuff that is
3960 * going to be truncated anyway.
3965 pagevec_init(&pvec, 0);
3966 if (wbc->range_cyclic) {
3967 index = mapping->writeback_index; /* Start from prev offset */
3970 index = wbc->range_start >> PAGE_SHIFT;
3971 end = wbc->range_end >> PAGE_SHIFT;
3972 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
3976 if (wbc->sync_mode == WB_SYNC_ALL)
3977 tag = PAGECACHE_TAG_TOWRITE;
3979 tag = PAGECACHE_TAG_DIRTY;
3981 if (wbc->sync_mode == WB_SYNC_ALL)
3982 tag_pages_for_writeback(mapping, index, end);
3984 while (!done && !nr_to_write_done && (index <= end) &&
3985 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3986 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3990 for (i = 0; i < nr_pages; i++) {
3991 struct page *page = pvec.pages[i];
3993 done_index = page->index;
3995 * At this point we hold neither mapping->tree_lock nor
3996 * lock on the page itself: the page may be truncated or
3997 * invalidated (changing page->mapping to NULL), or even
3998 * swizzled back from swapper_space to tmpfs file
4001 if (!trylock_page(page)) {
4006 if (unlikely(page->mapping != mapping)) {
4011 if (!wbc->range_cyclic && page->index > end) {
4017 if (wbc->sync_mode != WB_SYNC_NONE) {
4018 if (PageWriteback(page))
4020 wait_on_page_writeback(page);
4023 if (PageWriteback(page) ||
4024 !clear_page_dirty_for_io(page)) {
4029 ret = (*writepage)(page, wbc, data);
4031 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
4037 * done_index is set past this page,
4038 * so media errors will not choke
4039 * background writeout for the entire
4040 * file. This has consequences for
4041 * range_cyclic semantics (ie. it may
4042 * not be suitable for data integrity
4045 done_index = page->index + 1;
4051 * the filesystem may choose to bump up nr_to_write.
4052 * We have to make sure to honor the new nr_to_write
4055 nr_to_write_done = wbc->nr_to_write <= 0;
4057 pagevec_release(&pvec);
4060 if (!scanned && !done) {
4062 * We hit the last page and there is more work to be done: wrap
4063 * back to the start of the file
4069 * If we're looping we could run into a page that is locked by a
4070 * writer and that writer could be waiting on writeback for a
4071 * page in our current bio, and thus deadlock, so flush the
4074 flush_write_bio(data);
4078 if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole))
4079 mapping->writeback_index = done_index;
4081 btrfs_add_delayed_iput(inode);
4085 static void flush_epd_write_bio(struct extent_page_data *epd)
4090 bio_set_op_attrs(epd->bio, REQ_OP_WRITE,
4091 epd->sync_io ? WRITE_SYNC : 0);
4093 ret = submit_one_bio(epd->bio, 0, epd->bio_flags);
4094 BUG_ON(ret < 0); /* -ENOMEM */
4099 static noinline void flush_write_bio(void *data)
4101 struct extent_page_data *epd = data;
4102 flush_epd_write_bio(epd);
4105 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
4106 get_extent_t *get_extent,
4107 struct writeback_control *wbc)
4110 struct extent_page_data epd = {
4113 .get_extent = get_extent,
4115 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4119 ret = __extent_writepage(page, wbc, &epd);
4121 flush_epd_write_bio(&epd);
4125 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
4126 u64 start, u64 end, get_extent_t *get_extent,
4130 struct address_space *mapping = inode->i_mapping;
4132 unsigned long nr_pages = (end - start + PAGE_SIZE) >>
4135 struct extent_page_data epd = {
4138 .get_extent = get_extent,
4140 .sync_io = mode == WB_SYNC_ALL,
4143 struct writeback_control wbc_writepages = {
4145 .nr_to_write = nr_pages * 2,
4146 .range_start = start,
4147 .range_end = end + 1,
4150 while (start <= end) {
4151 page = find_get_page(mapping, start >> PAGE_SHIFT);
4152 if (clear_page_dirty_for_io(page))
4153 ret = __extent_writepage(page, &wbc_writepages, &epd);
4155 if (tree->ops && tree->ops->writepage_end_io_hook)
4156 tree->ops->writepage_end_io_hook(page, start,
4157 start + PAGE_SIZE - 1,
4165 flush_epd_write_bio(&epd);
4169 int extent_writepages(struct extent_io_tree *tree,
4170 struct address_space *mapping,
4171 get_extent_t *get_extent,
4172 struct writeback_control *wbc)
4175 struct extent_page_data epd = {
4178 .get_extent = get_extent,
4180 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4184 ret = extent_write_cache_pages(tree, mapping, wbc,
4185 __extent_writepage, &epd,
4187 flush_epd_write_bio(&epd);
4191 int extent_readpages(struct extent_io_tree *tree,
4192 struct address_space *mapping,
4193 struct list_head *pages, unsigned nr_pages,
4194 get_extent_t get_extent)
4196 struct bio *bio = NULL;
4198 unsigned long bio_flags = 0;
4199 struct page *pagepool[16];
4201 struct extent_map *em_cached = NULL;
4203 u64 prev_em_start = (u64)-1;
4205 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
4206 page = list_entry(pages->prev, struct page, lru);
4208 prefetchw(&page->flags);
4209 list_del(&page->lru);
4210 if (add_to_page_cache_lru(page, mapping,
4212 readahead_gfp_mask(mapping))) {
4217 pagepool[nr++] = page;
4218 if (nr < ARRAY_SIZE(pagepool))
4220 __extent_readpages(tree, pagepool, nr, get_extent, &em_cached,
4221 &bio, 0, &bio_flags, &prev_em_start);
4225 __extent_readpages(tree, pagepool, nr, get_extent, &em_cached,
4226 &bio, 0, &bio_flags, &prev_em_start);
4229 free_extent_map(em_cached);
4231 BUG_ON(!list_empty(pages));
4233 return submit_one_bio(bio, 0, bio_flags);
4238 * basic invalidatepage code, this waits on any locked or writeback
4239 * ranges corresponding to the page, and then deletes any extent state
4240 * records from the tree
4242 int extent_invalidatepage(struct extent_io_tree *tree,
4243 struct page *page, unsigned long offset)
4245 struct extent_state *cached_state = NULL;
4246 u64 start = page_offset(page);
4247 u64 end = start + PAGE_SIZE - 1;
4248 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
4250 start += ALIGN(offset, blocksize);
4254 lock_extent_bits(tree, start, end, &cached_state);
4255 wait_on_page_writeback(page);
4256 clear_extent_bit(tree, start, end,
4257 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4258 EXTENT_DO_ACCOUNTING,
4259 1, 1, &cached_state, GFP_NOFS);
4264 * a helper for releasepage, this tests for areas of the page that
4265 * are locked or under IO and drops the related state bits if it is safe
4268 static int try_release_extent_state(struct extent_map_tree *map,
4269 struct extent_io_tree *tree,
4270 struct page *page, gfp_t mask)
4272 u64 start = page_offset(page);
4273 u64 end = start + PAGE_SIZE - 1;
4276 if (test_range_bit(tree, start, end,
4277 EXTENT_IOBITS, 0, NULL))
4280 if ((mask & GFP_NOFS) == GFP_NOFS)
4283 * at this point we can safely clear everything except the
4284 * locked bit and the nodatasum bit
4286 ret = clear_extent_bit(tree, start, end,
4287 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
4290 /* if clear_extent_bit failed for enomem reasons,
4291 * we can't allow the release to continue.
4302 * a helper for releasepage. As long as there are no locked extents
4303 * in the range corresponding to the page, both state records and extent
4304 * map records are removed
4306 int try_release_extent_mapping(struct extent_map_tree *map,
4307 struct extent_io_tree *tree, struct page *page,
4310 struct extent_map *em;
4311 u64 start = page_offset(page);
4312 u64 end = start + PAGE_SIZE - 1;
4313 struct btrfs_inode *btrfs_inode = BTRFS_I(page->mapping->host);
4315 if (gfpflags_allow_blocking(mask) &&
4316 page->mapping->host->i_size > SZ_16M) {
4318 while (start <= end) {
4319 len = end - start + 1;
4320 write_lock(&map->lock);
4321 em = lookup_extent_mapping(map, start, len);
4323 write_unlock(&map->lock);
4326 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
4327 em->start != start) {
4328 write_unlock(&map->lock);
4329 free_extent_map(em);
4332 if (!test_range_bit(tree, em->start,
4333 extent_map_end(em) - 1,
4334 EXTENT_LOCKED | EXTENT_WRITEBACK,
4336 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4337 &btrfs_inode->runtime_flags);
4338 remove_extent_mapping(map, em);
4339 /* once for the rb tree */
4340 free_extent_map(em);
4342 start = extent_map_end(em);
4343 write_unlock(&map->lock);
4346 free_extent_map(em);
4348 cond_resched(); /* Allow large-extent preemption. */
4351 return try_release_extent_state(map, tree, page, mask);
4355 * helper function for fiemap, which doesn't want to see any holes.
4356 * This maps until we find something past 'last'
4358 static struct extent_map *get_extent_skip_holes(struct inode *inode,
4361 get_extent_t *get_extent)
4363 u64 sectorsize = BTRFS_I(inode)->root->sectorsize;
4364 struct extent_map *em;
4371 len = last - offset;
4374 len = ALIGN(len, sectorsize);
4375 em = get_extent(inode, NULL, 0, offset, len, 0);
4376 if (IS_ERR_OR_NULL(em))
4379 /* if this isn't a hole return it */
4380 if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) &&
4381 em->block_start != EXTENT_MAP_HOLE) {
4385 /* this is a hole, advance to the next extent */
4386 offset = extent_map_end(em);
4387 free_extent_map(em);
4395 * To cache previous fiemap extent
4397 * Will be used for merging fiemap extent
4399 struct fiemap_cache {
4408 * Helper to submit fiemap extent.
4410 * Will try to merge current fiemap extent specified by @offset, @phys,
4411 * @len and @flags with cached one.
4412 * And only when we fails to merge, cached one will be submitted as
4415 * Return value is the same as fiemap_fill_next_extent().
4417 static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
4418 struct fiemap_cache *cache,
4419 u64 offset, u64 phys, u64 len, u32 flags)
4427 * Sanity check, extent_fiemap() should have ensured that new
4428 * fiemap extent won't overlap with cahced one.
4431 * NOTE: Physical address can overlap, due to compression
4433 if (cache->offset + cache->len > offset) {
4439 * Only merges fiemap extents if
4440 * 1) Their logical addresses are continuous
4442 * 2) Their physical addresses are continuous
4443 * So truly compressed (physical size smaller than logical size)
4444 * extents won't get merged with each other
4446 * 3) Share same flags except FIEMAP_EXTENT_LAST
4447 * So regular extent won't get merged with prealloc extent
4449 if (cache->offset + cache->len == offset &&
4450 cache->phys + cache->len == phys &&
4451 (cache->flags & ~FIEMAP_EXTENT_LAST) ==
4452 (flags & ~FIEMAP_EXTENT_LAST)) {
4454 cache->flags |= flags;
4455 goto try_submit_last;
4458 /* Not mergeable, need to submit cached one */
4459 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
4460 cache->len, cache->flags);
4461 cache->cached = false;
4465 cache->cached = true;
4466 cache->offset = offset;
4469 cache->flags = flags;
4471 if (cache->flags & FIEMAP_EXTENT_LAST) {
4472 ret = fiemap_fill_next_extent(fieinfo, cache->offset,
4473 cache->phys, cache->len, cache->flags);
4474 cache->cached = false;
4480 * Emit last fiemap cache
4482 * The last fiemap cache may still be cached in the following case:
4484 * |<- Fiemap range ->|
4485 * |<------------ First extent ----------->|
4487 * In this case, the first extent range will be cached but not emitted.
4488 * So we must emit it before ending extent_fiemap().
4490 static int emit_last_fiemap_cache(struct btrfs_fs_info *fs_info,
4491 struct fiemap_extent_info *fieinfo,
4492 struct fiemap_cache *cache)
4499 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
4500 cache->len, cache->flags);
4501 cache->cached = false;
4507 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4508 __u64 start, __u64 len, get_extent_t *get_extent)
4512 u64 max = start + len;
4516 u64 last_for_get_extent = 0;
4518 u64 isize = i_size_read(inode);
4519 struct btrfs_key found_key;
4520 struct extent_map *em = NULL;
4521 struct extent_state *cached_state = NULL;
4522 struct btrfs_path *path;
4523 struct btrfs_root *root = BTRFS_I(inode)->root;
4524 struct fiemap_cache cache = { 0 };
4533 path = btrfs_alloc_path();
4536 path->leave_spinning = 1;
4538 start = round_down(start, BTRFS_I(inode)->root->sectorsize);
4539 len = round_up(max, BTRFS_I(inode)->root->sectorsize) - start;
4542 * lookup the last file extent. We're not using i_size here
4543 * because there might be preallocation past i_size
4545 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode), -1,
4548 btrfs_free_path(path);
4557 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
4558 found_type = found_key.type;
4560 /* No extents, but there might be delalloc bits */
4561 if (found_key.objectid != btrfs_ino(inode) ||
4562 found_type != BTRFS_EXTENT_DATA_KEY) {
4563 /* have to trust i_size as the end */
4565 last_for_get_extent = isize;
4568 * remember the start of the last extent. There are a
4569 * bunch of different factors that go into the length of the
4570 * extent, so its much less complex to remember where it started
4572 last = found_key.offset;
4573 last_for_get_extent = last + 1;
4575 btrfs_release_path(path);
4578 * we might have some extents allocated but more delalloc past those
4579 * extents. so, we trust isize unless the start of the last extent is
4584 last_for_get_extent = isize;
4587 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4590 em = get_extent_skip_holes(inode, start, last_for_get_extent,
4600 u64 offset_in_extent = 0;
4602 /* break if the extent we found is outside the range */
4603 if (em->start >= max || extent_map_end(em) < off)
4607 * get_extent may return an extent that starts before our
4608 * requested range. We have to make sure the ranges
4609 * we return to fiemap always move forward and don't
4610 * overlap, so adjust the offsets here
4612 em_start = max(em->start, off);
4615 * record the offset from the start of the extent
4616 * for adjusting the disk offset below. Only do this if the
4617 * extent isn't compressed since our in ram offset may be past
4618 * what we have actually allocated on disk.
4620 if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4621 offset_in_extent = em_start - em->start;
4622 em_end = extent_map_end(em);
4623 em_len = em_end - em_start;
4628 * bump off for our next call to get_extent
4630 off = extent_map_end(em);
4634 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
4636 flags |= FIEMAP_EXTENT_LAST;
4637 } else if (em->block_start == EXTENT_MAP_INLINE) {
4638 flags |= (FIEMAP_EXTENT_DATA_INLINE |
4639 FIEMAP_EXTENT_NOT_ALIGNED);
4640 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
4641 flags |= (FIEMAP_EXTENT_DELALLOC |
4642 FIEMAP_EXTENT_UNKNOWN);
4643 } else if (fieinfo->fi_extents_max) {
4644 struct btrfs_trans_handle *trans;
4646 u64 bytenr = em->block_start -
4647 (em->start - em->orig_start);
4649 disko = em->block_start + offset_in_extent;
4652 * We need a trans handle to get delayed refs
4654 trans = btrfs_join_transaction(root);
4656 * It's OK if we can't start a trans we can still check
4663 * As btrfs supports shared space, this information
4664 * can be exported to userspace tools via
4665 * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0
4666 * then we're just getting a count and we can skip the
4669 ret = btrfs_check_shared(trans, root->fs_info,
4671 btrfs_ino(inode), bytenr);
4673 btrfs_end_transaction(trans, root);
4677 flags |= FIEMAP_EXTENT_SHARED;
4680 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4681 flags |= FIEMAP_EXTENT_ENCODED;
4682 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4683 flags |= FIEMAP_EXTENT_UNWRITTEN;
4685 free_extent_map(em);
4687 if ((em_start >= last) || em_len == (u64)-1 ||
4688 (last == (u64)-1 && isize <= em_end)) {
4689 flags |= FIEMAP_EXTENT_LAST;
4693 /* now scan forward to see if this is really the last extent. */
4694 em = get_extent_skip_holes(inode, off, last_for_get_extent,
4701 flags |= FIEMAP_EXTENT_LAST;
4704 ret = emit_fiemap_extent(fieinfo, &cache, em_start, disko,
4714 ret = emit_last_fiemap_cache(root->fs_info, fieinfo, &cache);
4715 free_extent_map(em);
4717 btrfs_free_path(path);
4718 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4719 &cached_state, GFP_NOFS);
4723 static void __free_extent_buffer(struct extent_buffer *eb)
4725 btrfs_leak_debug_del(&eb->leak_list);
4726 kmem_cache_free(extent_buffer_cache, eb);
4729 int extent_buffer_under_io(struct extent_buffer *eb)
4731 return (atomic_read(&eb->io_pages) ||
4732 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4733 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4737 * Helper for releasing extent buffer page.
4739 static void btrfs_release_extent_buffer_page(struct extent_buffer *eb)
4741 unsigned long index;
4743 int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4745 BUG_ON(extent_buffer_under_io(eb));
4747 index = num_extent_pages(eb->start, eb->len);
4753 page = eb->pages[index];
4757 spin_lock(&page->mapping->private_lock);
4759 * We do this since we'll remove the pages after we've
4760 * removed the eb from the radix tree, so we could race
4761 * and have this page now attached to the new eb. So
4762 * only clear page_private if it's still connected to
4765 if (PagePrivate(page) &&
4766 page->private == (unsigned long)eb) {
4767 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4768 BUG_ON(PageDirty(page));
4769 BUG_ON(PageWriteback(page));
4771 * We need to make sure we haven't be attached
4774 ClearPagePrivate(page);
4775 set_page_private(page, 0);
4776 /* One for the page private */
4781 spin_unlock(&page->mapping->private_lock);
4783 /* One for when we allocated the page */
4785 } while (index != 0);
4789 * Helper for releasing the extent buffer.
4791 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4793 btrfs_release_extent_buffer_page(eb);
4794 __free_extent_buffer(eb);
4797 static struct extent_buffer *
4798 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
4801 struct extent_buffer *eb = NULL;
4803 eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL);
4806 eb->fs_info = fs_info;
4808 rwlock_init(&eb->lock);
4809 atomic_set(&eb->write_locks, 0);
4810 atomic_set(&eb->read_locks, 0);
4811 atomic_set(&eb->blocking_readers, 0);
4812 atomic_set(&eb->blocking_writers, 0);
4813 atomic_set(&eb->spinning_readers, 0);
4814 atomic_set(&eb->spinning_writers, 0);
4815 eb->lock_nested = 0;
4816 init_waitqueue_head(&eb->write_lock_wq);
4817 init_waitqueue_head(&eb->read_lock_wq);
4819 btrfs_leak_debug_add(&eb->leak_list, &buffers);
4821 spin_lock_init(&eb->refs_lock);
4822 atomic_set(&eb->refs, 1);
4823 atomic_set(&eb->io_pages, 0);
4826 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4828 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4829 > MAX_INLINE_EXTENT_BUFFER_SIZE);
4830 BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE);
4835 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src)
4839 struct extent_buffer *new;
4840 unsigned long num_pages = num_extent_pages(src->start, src->len);
4842 new = __alloc_extent_buffer(src->fs_info, src->start, src->len);
4846 for (i = 0; i < num_pages; i++) {
4847 p = alloc_page(GFP_NOFS);
4849 btrfs_release_extent_buffer(new);
4852 attach_extent_buffer_page(new, p);
4853 WARN_ON(PageDirty(p));
4858 copy_extent_buffer(new, src, 0, 0, src->len);
4859 set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
4860 set_bit(EXTENT_BUFFER_DUMMY, &new->bflags);
4865 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4866 u64 start, unsigned long len)
4868 struct extent_buffer *eb;
4869 unsigned long num_pages;
4872 num_pages = num_extent_pages(start, len);
4874 eb = __alloc_extent_buffer(fs_info, start, len);
4878 for (i = 0; i < num_pages; i++) {
4879 eb->pages[i] = alloc_page(GFP_NOFS);
4883 set_extent_buffer_uptodate(eb);
4884 btrfs_set_header_nritems(eb, 0);
4885 set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4890 __free_page(eb->pages[i - 1]);
4891 __free_extent_buffer(eb);
4895 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4896 u64 start, u32 nodesize)
4902 * Called only from tests that don't always have a fs_info
4907 len = fs_info->tree_root->nodesize;
4910 return __alloc_dummy_extent_buffer(fs_info, start, len);
4913 static void check_buffer_tree_ref(struct extent_buffer *eb)
4917 * The TREE_REF bit is first set when the extent_buffer is added
4918 * to the radix tree. It is also reset, if unset, when a new reference
4919 * is created by find_extent_buffer.
4921 * It is only cleared in two cases: freeing the last non-tree
4922 * reference to the extent_buffer when its STALE bit is set or
4923 * calling releasepage when the tree reference is the only reference.
4925 * In both cases, care is taken to ensure that the extent_buffer's
4926 * pages are not under io. However, releasepage can be concurrently
4927 * called with creating new references, which is prone to race
4928 * conditions between the calls to check_buffer_tree_ref in those
4929 * codepaths and clearing TREE_REF in try_release_extent_buffer.
4931 * The actual lifetime of the extent_buffer in the radix tree is
4932 * adequately protected by the refcount, but the TREE_REF bit and
4933 * its corresponding reference are not. To protect against this
4934 * class of races, we call check_buffer_tree_ref from the codepaths
4935 * which trigger io after they set eb->io_pages. Note that once io is
4936 * initiated, TREE_REF can no longer be cleared, so that is the
4937 * moment at which any such race is best fixed.
4939 refs = atomic_read(&eb->refs);
4940 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4943 spin_lock(&eb->refs_lock);
4944 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4945 atomic_inc(&eb->refs);
4946 spin_unlock(&eb->refs_lock);
4949 static void mark_extent_buffer_accessed(struct extent_buffer *eb,
4950 struct page *accessed)
4952 unsigned long num_pages, i;
4954 check_buffer_tree_ref(eb);
4956 num_pages = num_extent_pages(eb->start, eb->len);
4957 for (i = 0; i < num_pages; i++) {
4958 struct page *p = eb->pages[i];
4961 mark_page_accessed(p);
4965 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
4968 struct extent_buffer *eb;
4971 eb = radix_tree_lookup(&fs_info->buffer_radix,
4972 start >> PAGE_SHIFT);
4973 if (eb && atomic_inc_not_zero(&eb->refs)) {
4976 * Lock our eb's refs_lock to avoid races with
4977 * free_extent_buffer. When we get our eb it might be flagged
4978 * with EXTENT_BUFFER_STALE and another task running
4979 * free_extent_buffer might have seen that flag set,
4980 * eb->refs == 2, that the buffer isn't under IO (dirty and
4981 * writeback flags not set) and it's still in the tree (flag
4982 * EXTENT_BUFFER_TREE_REF set), therefore being in the process
4983 * of decrementing the extent buffer's reference count twice.
4984 * So here we could race and increment the eb's reference count,
4985 * clear its stale flag, mark it as dirty and drop our reference
4986 * before the other task finishes executing free_extent_buffer,
4987 * which would later result in an attempt to free an extent
4988 * buffer that is dirty.
4990 if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) {
4991 spin_lock(&eb->refs_lock);
4992 spin_unlock(&eb->refs_lock);
4994 mark_extent_buffer_accessed(eb, NULL);
5002 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5003 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
5004 u64 start, u32 nodesize)
5006 struct extent_buffer *eb, *exists = NULL;
5009 eb = find_extent_buffer(fs_info, start);
5012 eb = alloc_dummy_extent_buffer(fs_info, start, nodesize);
5014 return ERR_PTR(-ENOMEM);
5015 eb->fs_info = fs_info;
5017 ret = radix_tree_preload(GFP_NOFS);
5019 exists = ERR_PTR(ret);
5022 spin_lock(&fs_info->buffer_lock);
5023 ret = radix_tree_insert(&fs_info->buffer_radix,
5024 start >> PAGE_SHIFT, eb);
5025 spin_unlock(&fs_info->buffer_lock);
5026 radix_tree_preload_end();
5027 if (ret == -EEXIST) {
5028 exists = find_extent_buffer(fs_info, start);
5034 check_buffer_tree_ref(eb);
5035 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
5038 * We will free dummy extent buffer's if they come into
5039 * free_extent_buffer with a ref count of 2, but if we are using this we
5040 * want the buffers to stay in memory until we're done with them, so
5041 * bump the ref count again.
5043 atomic_inc(&eb->refs);
5046 btrfs_release_extent_buffer(eb);
5051 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
5054 unsigned long len = fs_info->tree_root->nodesize;
5055 unsigned long num_pages = num_extent_pages(start, len);
5057 unsigned long index = start >> PAGE_SHIFT;
5058 struct extent_buffer *eb;
5059 struct extent_buffer *exists = NULL;
5061 struct address_space *mapping = fs_info->btree_inode->i_mapping;
5065 if (!IS_ALIGNED(start, fs_info->tree_root->sectorsize)) {
5066 btrfs_err(fs_info, "bad tree block start %llu", start);
5067 return ERR_PTR(-EINVAL);
5070 eb = find_extent_buffer(fs_info, start);
5074 eb = __alloc_extent_buffer(fs_info, start, len);
5076 return ERR_PTR(-ENOMEM);
5078 for (i = 0; i < num_pages; i++, index++) {
5079 p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL);
5081 exists = ERR_PTR(-ENOMEM);
5085 spin_lock(&mapping->private_lock);
5086 if (PagePrivate(p)) {
5088 * We could have already allocated an eb for this page
5089 * and attached one so lets see if we can get a ref on
5090 * the existing eb, and if we can we know it's good and
5091 * we can just return that one, else we know we can just
5092 * overwrite page->private.
5094 exists = (struct extent_buffer *)p->private;
5095 if (atomic_inc_not_zero(&exists->refs)) {
5096 spin_unlock(&mapping->private_lock);
5099 mark_extent_buffer_accessed(exists, p);
5105 * Do this so attach doesn't complain and we need to
5106 * drop the ref the old guy had.
5108 ClearPagePrivate(p);
5109 WARN_ON(PageDirty(p));
5112 attach_extent_buffer_page(eb, p);
5113 spin_unlock(&mapping->private_lock);
5114 WARN_ON(PageDirty(p));
5116 if (!PageUptodate(p))
5120 * see below about how we avoid a nasty race with release page
5121 * and why we unlock later
5125 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5127 ret = radix_tree_preload(GFP_NOFS);
5129 exists = ERR_PTR(ret);
5133 spin_lock(&fs_info->buffer_lock);
5134 ret = radix_tree_insert(&fs_info->buffer_radix,
5135 start >> PAGE_SHIFT, eb);
5136 spin_unlock(&fs_info->buffer_lock);
5137 radix_tree_preload_end();
5138 if (ret == -EEXIST) {
5139 exists = find_extent_buffer(fs_info, start);
5145 /* add one reference for the tree */
5146 check_buffer_tree_ref(eb);
5147 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
5150 * there is a race where release page may have
5151 * tried to find this extent buffer in the radix
5152 * but failed. It will tell the VM it is safe to
5153 * reclaim the, and it will clear the page private bit.
5154 * We must make sure to set the page private bit properly
5155 * after the extent buffer is in the radix tree so
5156 * it doesn't get lost
5158 SetPageChecked(eb->pages[0]);
5159 for (i = 1; i < num_pages; i++) {
5161 ClearPageChecked(p);
5164 unlock_page(eb->pages[0]);
5168 WARN_ON(!atomic_dec_and_test(&eb->refs));
5169 for (i = 0; i < num_pages; i++) {
5171 unlock_page(eb->pages[i]);
5174 btrfs_release_extent_buffer(eb);
5178 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
5180 struct extent_buffer *eb =
5181 container_of(head, struct extent_buffer, rcu_head);
5183 __free_extent_buffer(eb);
5186 /* Expects to have eb->eb_lock already held */
5187 static int release_extent_buffer(struct extent_buffer *eb)
5189 WARN_ON(atomic_read(&eb->refs) == 0);
5190 if (atomic_dec_and_test(&eb->refs)) {
5191 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
5192 struct btrfs_fs_info *fs_info = eb->fs_info;
5194 spin_unlock(&eb->refs_lock);
5196 spin_lock(&fs_info->buffer_lock);
5197 radix_tree_delete(&fs_info->buffer_radix,
5198 eb->start >> PAGE_SHIFT);
5199 spin_unlock(&fs_info->buffer_lock);
5201 spin_unlock(&eb->refs_lock);
5204 /* Should be safe to release our pages at this point */
5205 btrfs_release_extent_buffer_page(eb);
5206 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5207 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))) {
5208 __free_extent_buffer(eb);
5212 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
5215 spin_unlock(&eb->refs_lock);
5220 void free_extent_buffer(struct extent_buffer *eb)
5228 refs = atomic_read(&eb->refs);
5231 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
5236 spin_lock(&eb->refs_lock);
5237 if (atomic_read(&eb->refs) == 2 &&
5238 test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))
5239 atomic_dec(&eb->refs);
5241 if (atomic_read(&eb->refs) == 2 &&
5242 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
5243 !extent_buffer_under_io(eb) &&
5244 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5245 atomic_dec(&eb->refs);
5248 * I know this is terrible, but it's temporary until we stop tracking
5249 * the uptodate bits and such for the extent buffers.
5251 release_extent_buffer(eb);
5254 void free_extent_buffer_stale(struct extent_buffer *eb)
5259 spin_lock(&eb->refs_lock);
5260 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
5262 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
5263 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5264 atomic_dec(&eb->refs);
5265 release_extent_buffer(eb);
5268 void clear_extent_buffer_dirty(struct extent_buffer *eb)
5271 unsigned long num_pages;
5274 num_pages = num_extent_pages(eb->start, eb->len);
5276 for (i = 0; i < num_pages; i++) {
5277 page = eb->pages[i];
5278 if (!PageDirty(page))
5282 WARN_ON(!PagePrivate(page));
5284 clear_page_dirty_for_io(page);
5285 spin_lock_irq(&page->mapping->tree_lock);
5286 if (!PageDirty(page)) {
5287 radix_tree_tag_clear(&page->mapping->page_tree,
5289 PAGECACHE_TAG_DIRTY);
5291 spin_unlock_irq(&page->mapping->tree_lock);
5292 ClearPageError(page);
5295 WARN_ON(atomic_read(&eb->refs) == 0);
5298 int set_extent_buffer_dirty(struct extent_buffer *eb)
5301 unsigned long num_pages;
5304 check_buffer_tree_ref(eb);
5306 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
5308 num_pages = num_extent_pages(eb->start, eb->len);
5309 WARN_ON(atomic_read(&eb->refs) == 0);
5310 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
5312 for (i = 0; i < num_pages; i++)
5313 set_page_dirty(eb->pages[i]);
5317 void clear_extent_buffer_uptodate(struct extent_buffer *eb)
5321 unsigned long num_pages;
5323 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5324 num_pages = num_extent_pages(eb->start, eb->len);
5325 for (i = 0; i < num_pages; i++) {
5326 page = eb->pages[i];
5328 ClearPageUptodate(page);
5332 void set_extent_buffer_uptodate(struct extent_buffer *eb)
5336 unsigned long num_pages;
5338 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5339 num_pages = num_extent_pages(eb->start, eb->len);
5340 for (i = 0; i < num_pages; i++) {
5341 page = eb->pages[i];
5342 SetPageUptodate(page);
5346 int extent_buffer_uptodate(struct extent_buffer *eb)
5348 return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5351 int read_extent_buffer_pages(struct extent_io_tree *tree,
5352 struct extent_buffer *eb, int wait,
5353 get_extent_t *get_extent, int mirror_num)
5359 int locked_pages = 0;
5360 int all_uptodate = 1;
5361 unsigned long num_pages;
5362 unsigned long num_reads = 0;
5363 struct bio *bio = NULL;
5364 unsigned long bio_flags = 0;
5366 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
5369 num_pages = num_extent_pages(eb->start, eb->len);
5370 for (i = 0; i < num_pages; i++) {
5371 page = eb->pages[i];
5372 if (wait == WAIT_NONE) {
5373 if (!trylock_page(page))
5381 * We need to firstly lock all pages to make sure that
5382 * the uptodate bit of our pages won't be affected by
5383 * clear_extent_buffer_uptodate().
5385 for (i = 0; i < num_pages; i++) {
5386 page = eb->pages[i];
5387 if (!PageUptodate(page)) {
5394 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5398 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
5399 eb->read_mirror = 0;
5400 atomic_set(&eb->io_pages, num_reads);
5402 * It is possible for releasepage to clear the TREE_REF bit before we
5403 * set io_pages. See check_buffer_tree_ref for a more detailed comment.
5405 check_buffer_tree_ref(eb);
5406 for (i = 0; i < num_pages; i++) {
5407 page = eb->pages[i];
5409 if (!PageUptodate(page)) {
5411 atomic_dec(&eb->io_pages);
5416 ClearPageError(page);
5417 err = __extent_read_full_page(tree, page,
5419 mirror_num, &bio_flags,
5424 * We use &bio in above __extent_read_full_page,
5425 * so we ensure that if it returns error, the
5426 * current page fails to add itself to bio and
5427 * it's been unlocked.
5429 * We must dec io_pages by ourselves.
5431 atomic_dec(&eb->io_pages);
5439 err = submit_one_bio(bio, mirror_num, bio_flags);
5444 if (ret || wait != WAIT_COMPLETE)
5447 for (i = 0; i < num_pages; i++) {
5448 page = eb->pages[i];
5449 wait_on_page_locked(page);
5450 if (!PageUptodate(page))
5457 while (locked_pages > 0) {
5459 page = eb->pages[locked_pages];
5465 void read_extent_buffer(const struct extent_buffer *eb, void *dstv,
5466 unsigned long start, unsigned long len)
5472 char *dst = (char *)dstv;
5473 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5474 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5476 WARN_ON(start > eb->len);
5477 WARN_ON(start + len > eb->start + eb->len);
5479 offset = (start_offset + start) & (PAGE_SIZE - 1);
5482 page = eb->pages[i];
5484 cur = min(len, (PAGE_SIZE - offset));
5485 kaddr = page_address(page);
5486 memcpy(dst, kaddr + offset, cur);
5495 int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb,
5497 unsigned long start, unsigned long len)
5503 char __user *dst = (char __user *)dstv;
5504 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5505 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5508 WARN_ON(start > eb->len);
5509 WARN_ON(start + len > eb->start + eb->len);
5511 offset = (start_offset + start) & (PAGE_SIZE - 1);
5514 page = eb->pages[i];
5516 cur = min(len, (PAGE_SIZE - offset));
5517 kaddr = page_address(page);
5518 if (probe_user_write(dst, kaddr + offset, cur)) {
5533 * return 0 if the item is found within a page.
5534 * return 1 if the item spans two pages.
5535 * return -EINVAL otherwise.
5537 int map_private_extent_buffer(const struct extent_buffer *eb,
5538 unsigned long start, unsigned long min_len,
5539 char **map, unsigned long *map_start,
5540 unsigned long *map_len)
5542 size_t offset = start & (PAGE_SIZE - 1);
5545 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5546 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5547 unsigned long end_i = (start_offset + start + min_len - 1) >>
5554 offset = start_offset;
5558 *map_start = ((u64)i << PAGE_SHIFT) - start_offset;
5561 if (start + min_len > eb->len) {
5562 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n",
5563 eb->start, eb->len, start, min_len);
5568 kaddr = page_address(p);
5569 *map = kaddr + offset;
5570 *map_len = PAGE_SIZE - offset;
5574 int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,
5575 unsigned long start, unsigned long len)
5581 char *ptr = (char *)ptrv;
5582 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5583 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5586 WARN_ON(start > eb->len);
5587 WARN_ON(start + len > eb->start + eb->len);
5589 offset = (start_offset + start) & (PAGE_SIZE - 1);
5592 page = eb->pages[i];
5594 cur = min(len, (PAGE_SIZE - offset));
5596 kaddr = page_address(page);
5597 ret = memcmp(ptr, kaddr + offset, cur);
5609 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
5610 unsigned long start, unsigned long len)
5616 char *src = (char *)srcv;
5617 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5618 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5620 WARN_ON(start > eb->len);
5621 WARN_ON(start + len > eb->start + eb->len);
5623 offset = (start_offset + start) & (PAGE_SIZE - 1);
5626 page = eb->pages[i];
5627 WARN_ON(!PageUptodate(page));
5629 cur = min(len, PAGE_SIZE - offset);
5630 kaddr = page_address(page);
5631 memcpy(kaddr + offset, src, cur);
5640 void memset_extent_buffer(struct extent_buffer *eb, char c,
5641 unsigned long start, unsigned long len)
5647 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5648 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5650 WARN_ON(start > eb->len);
5651 WARN_ON(start + len > eb->start + eb->len);
5653 offset = (start_offset + start) & (PAGE_SIZE - 1);
5656 page = eb->pages[i];
5657 WARN_ON(!PageUptodate(page));
5659 cur = min(len, PAGE_SIZE - offset);
5660 kaddr = page_address(page);
5661 memset(kaddr + offset, c, cur);
5669 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
5670 unsigned long dst_offset, unsigned long src_offset,
5673 u64 dst_len = dst->len;
5678 size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1);
5679 unsigned long i = (start_offset + dst_offset) >> PAGE_SHIFT;
5681 WARN_ON(src->len != dst_len);
5683 offset = (start_offset + dst_offset) &
5687 page = dst->pages[i];
5688 WARN_ON(!PageUptodate(page));
5690 cur = min(len, (unsigned long)(PAGE_SIZE - offset));
5692 kaddr = page_address(page);
5693 read_extent_buffer(src, kaddr + offset, src_offset, cur);
5702 void le_bitmap_set(u8 *map, unsigned int start, int len)
5704 u8 *p = map + BIT_BYTE(start);
5705 const unsigned int size = start + len;
5706 int bits_to_set = BITS_PER_BYTE - (start % BITS_PER_BYTE);
5707 u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(start);
5709 while (len - bits_to_set >= 0) {
5712 bits_to_set = BITS_PER_BYTE;
5717 mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
5722 void le_bitmap_clear(u8 *map, unsigned int start, int len)
5724 u8 *p = map + BIT_BYTE(start);
5725 const unsigned int size = start + len;
5726 int bits_to_clear = BITS_PER_BYTE - (start % BITS_PER_BYTE);
5727 u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(start);
5729 while (len - bits_to_clear >= 0) {
5730 *p &= ~mask_to_clear;
5731 len -= bits_to_clear;
5732 bits_to_clear = BITS_PER_BYTE;
5737 mask_to_clear &= BITMAP_LAST_BYTE_MASK(size);
5738 *p &= ~mask_to_clear;
5743 * eb_bitmap_offset() - calculate the page and offset of the byte containing the
5745 * @eb: the extent buffer
5746 * @start: offset of the bitmap item in the extent buffer
5748 * @page_index: return index of the page in the extent buffer that contains the
5750 * @page_offset: return offset into the page given by page_index
5752 * This helper hides the ugliness of finding the byte in an extent buffer which
5753 * contains a given bit.
5755 static inline void eb_bitmap_offset(struct extent_buffer *eb,
5756 unsigned long start, unsigned long nr,
5757 unsigned long *page_index,
5758 size_t *page_offset)
5760 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5761 size_t byte_offset = BIT_BYTE(nr);
5765 * The byte we want is the offset of the extent buffer + the offset of
5766 * the bitmap item in the extent buffer + the offset of the byte in the
5769 offset = start_offset + start + byte_offset;
5771 *page_index = offset >> PAGE_SHIFT;
5772 *page_offset = offset & (PAGE_SIZE - 1);
5776 * extent_buffer_test_bit - determine whether a bit in a bitmap item is set
5777 * @eb: the extent buffer
5778 * @start: offset of the bitmap item in the extent buffer
5779 * @nr: bit number to test
5781 int extent_buffer_test_bit(struct extent_buffer *eb, unsigned long start,
5789 eb_bitmap_offset(eb, start, nr, &i, &offset);
5790 page = eb->pages[i];
5791 WARN_ON(!PageUptodate(page));
5792 kaddr = page_address(page);
5793 return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1)));
5797 * extent_buffer_bitmap_set - set an area of a bitmap
5798 * @eb: the extent buffer
5799 * @start: offset of the bitmap item in the extent buffer
5800 * @pos: bit number of the first bit
5801 * @len: number of bits to set
5803 void extent_buffer_bitmap_set(struct extent_buffer *eb, unsigned long start,
5804 unsigned long pos, unsigned long len)
5810 const unsigned int size = pos + len;
5811 int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5812 u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos);
5814 eb_bitmap_offset(eb, start, pos, &i, &offset);
5815 page = eb->pages[i];
5816 WARN_ON(!PageUptodate(page));
5817 kaddr = page_address(page);
5819 while (len >= bits_to_set) {
5820 kaddr[offset] |= mask_to_set;
5822 bits_to_set = BITS_PER_BYTE;
5824 if (++offset >= PAGE_SIZE && len > 0) {
5826 page = eb->pages[++i];
5827 WARN_ON(!PageUptodate(page));
5828 kaddr = page_address(page);
5832 mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
5833 kaddr[offset] |= mask_to_set;
5839 * extent_buffer_bitmap_clear - clear an area of a bitmap
5840 * @eb: the extent buffer
5841 * @start: offset of the bitmap item in the extent buffer
5842 * @pos: bit number of the first bit
5843 * @len: number of bits to clear
5845 void extent_buffer_bitmap_clear(struct extent_buffer *eb, unsigned long start,
5846 unsigned long pos, unsigned long len)
5852 const unsigned int size = pos + len;
5853 int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5854 u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos);
5856 eb_bitmap_offset(eb, start, pos, &i, &offset);
5857 page = eb->pages[i];
5858 WARN_ON(!PageUptodate(page));
5859 kaddr = page_address(page);
5861 while (len >= bits_to_clear) {
5862 kaddr[offset] &= ~mask_to_clear;
5863 len -= bits_to_clear;
5864 bits_to_clear = BITS_PER_BYTE;
5866 if (++offset >= PAGE_SIZE && len > 0) {
5868 page = eb->pages[++i];
5869 WARN_ON(!PageUptodate(page));
5870 kaddr = page_address(page);
5874 mask_to_clear &= BITMAP_LAST_BYTE_MASK(size);
5875 kaddr[offset] &= ~mask_to_clear;
5879 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
5881 unsigned long distance = (src > dst) ? src - dst : dst - src;
5882 return distance < len;
5885 static void copy_pages(struct page *dst_page, struct page *src_page,
5886 unsigned long dst_off, unsigned long src_off,
5889 char *dst_kaddr = page_address(dst_page);
5891 int must_memmove = 0;
5893 if (dst_page != src_page) {
5894 src_kaddr = page_address(src_page);
5896 src_kaddr = dst_kaddr;
5897 if (areas_overlap(src_off, dst_off, len))
5902 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
5904 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
5907 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5908 unsigned long src_offset, unsigned long len)
5911 size_t dst_off_in_page;
5912 size_t src_off_in_page;
5913 size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1);
5914 unsigned long dst_i;
5915 unsigned long src_i;
5917 if (src_offset + len > dst->len) {
5918 btrfs_err(dst->fs_info,
5919 "memmove bogus src_offset %lu move len %lu dst len %lu",
5920 src_offset, len, dst->len);
5923 if (dst_offset + len > dst->len) {
5924 btrfs_err(dst->fs_info,
5925 "memmove bogus dst_offset %lu move len %lu dst len %lu",
5926 dst_offset, len, dst->len);
5931 dst_off_in_page = (start_offset + dst_offset) &
5933 src_off_in_page = (start_offset + src_offset) &
5936 dst_i = (start_offset + dst_offset) >> PAGE_SHIFT;
5937 src_i = (start_offset + src_offset) >> PAGE_SHIFT;
5939 cur = min(len, (unsigned long)(PAGE_SIZE -
5941 cur = min_t(unsigned long, cur,
5942 (unsigned long)(PAGE_SIZE - dst_off_in_page));
5944 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5945 dst_off_in_page, src_off_in_page, cur);
5953 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5954 unsigned long src_offset, unsigned long len)
5957 size_t dst_off_in_page;
5958 size_t src_off_in_page;
5959 unsigned long dst_end = dst_offset + len - 1;
5960 unsigned long src_end = src_offset + len - 1;
5961 size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1);
5962 unsigned long dst_i;
5963 unsigned long src_i;
5965 if (src_offset + len > dst->len) {
5966 btrfs_err(dst->fs_info,
5967 "memmove bogus src_offset %lu move len %lu len %lu",
5968 src_offset, len, dst->len);
5971 if (dst_offset + len > dst->len) {
5972 btrfs_err(dst->fs_info,
5973 "memmove bogus dst_offset %lu move len %lu len %lu",
5974 dst_offset, len, dst->len);
5977 if (dst_offset < src_offset) {
5978 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
5982 dst_i = (start_offset + dst_end) >> PAGE_SHIFT;
5983 src_i = (start_offset + src_end) >> PAGE_SHIFT;
5985 dst_off_in_page = (start_offset + dst_end) &
5987 src_off_in_page = (start_offset + src_end) &
5990 cur = min_t(unsigned long, len, src_off_in_page + 1);
5991 cur = min(cur, dst_off_in_page + 1);
5992 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5993 dst_off_in_page - cur + 1,
5994 src_off_in_page - cur + 1, cur);
6002 int try_release_extent_buffer(struct page *page)
6004 struct extent_buffer *eb;
6007 * We need to make sure nobody is attaching this page to an eb right
6010 spin_lock(&page->mapping->private_lock);
6011 if (!PagePrivate(page)) {
6012 spin_unlock(&page->mapping->private_lock);
6016 eb = (struct extent_buffer *)page->private;
6020 * This is a little awful but should be ok, we need to make sure that
6021 * the eb doesn't disappear out from under us while we're looking at
6024 spin_lock(&eb->refs_lock);
6025 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
6026 spin_unlock(&eb->refs_lock);
6027 spin_unlock(&page->mapping->private_lock);
6030 spin_unlock(&page->mapping->private_lock);
6033 * If tree ref isn't set then we know the ref on this eb is a real ref,
6034 * so just return, this page will likely be freed soon anyway.
6036 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
6037 spin_unlock(&eb->refs_lock);
6041 return release_extent_buffer(eb);