1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/bitops.h>
3 #include <linux/slab.h>
6 #include <linux/pagemap.h>
7 #include <linux/page-flags.h>
8 #include <linux/spinlock.h>
9 #include <linux/blkdev.h>
10 #include <linux/swap.h>
11 #include <linux/writeback.h>
12 #include <linux/pagevec.h>
13 #include <linux/prefetch.h>
14 #include <linux/cleancache.h>
15 #include "extent_io.h"
16 #include "extent_map.h"
18 #include "btrfs_inode.h"
20 #include "check-integrity.h"
22 #include "rcu-string.h"
26 static struct kmem_cache *extent_state_cache;
27 static struct kmem_cache *extent_buffer_cache;
28 static struct bio_set *btrfs_bioset;
30 static inline bool extent_state_in_tree(const struct extent_state *state)
32 return !RB_EMPTY_NODE(&state->rb_node);
35 #ifdef CONFIG_BTRFS_DEBUG
36 static LIST_HEAD(buffers);
37 static LIST_HEAD(states);
39 static DEFINE_SPINLOCK(leak_lock);
42 void btrfs_leak_debug_add(struct list_head *new, struct list_head *head)
46 spin_lock_irqsave(&leak_lock, flags);
48 spin_unlock_irqrestore(&leak_lock, flags);
52 void btrfs_leak_debug_del(struct list_head *entry)
56 spin_lock_irqsave(&leak_lock, flags);
58 spin_unlock_irqrestore(&leak_lock, flags);
62 void btrfs_leak_debug_check(void)
64 struct extent_state *state;
65 struct extent_buffer *eb;
67 while (!list_empty(&states)) {
68 state = list_entry(states.next, struct extent_state, leak_list);
69 pr_err("BTRFS: state leak: start %llu end %llu state %u in tree %d refs %d\n",
70 state->start, state->end, state->state,
71 extent_state_in_tree(state),
72 refcount_read(&state->refs));
73 list_del(&state->leak_list);
74 kmem_cache_free(extent_state_cache, state);
77 while (!list_empty(&buffers)) {
78 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
79 pr_err("BTRFS: buffer leak start %llu len %lu refs %d\n",
80 eb->start, eb->len, atomic_read(&eb->refs));
81 list_del(&eb->leak_list);
82 kmem_cache_free(extent_buffer_cache, eb);
86 #define btrfs_debug_check_extent_io_range(tree, start, end) \
87 __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end))
88 static inline void __btrfs_debug_check_extent_io_range(const char *caller,
89 struct extent_io_tree *tree, u64 start, u64 end)
91 if (tree->ops && tree->ops->check_extent_io_range)
92 tree->ops->check_extent_io_range(tree->private_data, caller,
96 #define btrfs_leak_debug_add(new, head) do {} while (0)
97 #define btrfs_leak_debug_del(entry) do {} while (0)
98 #define btrfs_leak_debug_check() do {} while (0)
99 #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0)
102 #define BUFFER_LRU_MAX 64
107 struct rb_node rb_node;
110 struct extent_page_data {
112 struct extent_io_tree *tree;
113 /* tells writepage not to lock the state bits for this range
114 * it still does the unlocking
116 unsigned int extent_locked:1;
118 /* tells the submit_bio code to use REQ_SYNC */
119 unsigned int sync_io:1;
122 static void add_extent_changeset(struct extent_state *state, unsigned bits,
123 struct extent_changeset *changeset,
130 if (set && (state->state & bits) == bits)
132 if (!set && (state->state & bits) == 0)
134 changeset->bytes_changed += state->end - state->start + 1;
135 ret = ulist_add(&changeset->range_changed, state->start, state->end,
141 static void flush_write_bio(struct extent_page_data *epd);
143 static inline struct btrfs_fs_info *
144 tree_fs_info(struct extent_io_tree *tree)
147 return tree->ops->tree_fs_info(tree->private_data);
151 int __init extent_io_init(void)
153 extent_state_cache = kmem_cache_create("btrfs_extent_state",
154 sizeof(struct extent_state), 0,
155 SLAB_MEM_SPREAD, NULL);
156 if (!extent_state_cache)
159 extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer",
160 sizeof(struct extent_buffer), 0,
161 SLAB_MEM_SPREAD, NULL);
162 if (!extent_buffer_cache)
163 goto free_state_cache;
165 btrfs_bioset = bioset_create(BIO_POOL_SIZE,
166 offsetof(struct btrfs_io_bio, bio),
169 goto free_buffer_cache;
171 if (bioset_integrity_create(btrfs_bioset, BIO_POOL_SIZE))
177 bioset_free(btrfs_bioset);
181 kmem_cache_destroy(extent_buffer_cache);
182 extent_buffer_cache = NULL;
185 kmem_cache_destroy(extent_state_cache);
186 extent_state_cache = NULL;
190 void extent_io_exit(void)
192 btrfs_leak_debug_check();
195 * Make sure all delayed rcu free are flushed before we
199 kmem_cache_destroy(extent_state_cache);
200 kmem_cache_destroy(extent_buffer_cache);
202 bioset_free(btrfs_bioset);
205 void extent_io_tree_init(struct extent_io_tree *tree,
208 tree->state = RB_ROOT;
210 tree->dirty_bytes = 0;
211 spin_lock_init(&tree->lock);
212 tree->private_data = private_data;
215 static struct extent_state *alloc_extent_state(gfp_t mask)
217 struct extent_state *state;
220 * The given mask might be not appropriate for the slab allocator,
221 * drop the unsupported bits
223 mask &= ~(__GFP_DMA32|__GFP_HIGHMEM);
224 state = kmem_cache_alloc(extent_state_cache, mask);
228 state->failrec = NULL;
229 RB_CLEAR_NODE(&state->rb_node);
230 btrfs_leak_debug_add(&state->leak_list, &states);
231 refcount_set(&state->refs, 1);
232 init_waitqueue_head(&state->wq);
233 trace_alloc_extent_state(state, mask, _RET_IP_);
237 void free_extent_state(struct extent_state *state)
241 if (refcount_dec_and_test(&state->refs)) {
242 WARN_ON(extent_state_in_tree(state));
243 btrfs_leak_debug_del(&state->leak_list);
244 trace_free_extent_state(state, _RET_IP_);
245 kmem_cache_free(extent_state_cache, state);
249 static struct rb_node *tree_insert(struct rb_root *root,
250 struct rb_node *search_start,
252 struct rb_node *node,
253 struct rb_node ***p_in,
254 struct rb_node **parent_in)
257 struct rb_node *parent = NULL;
258 struct tree_entry *entry;
260 if (p_in && parent_in) {
266 p = search_start ? &search_start : &root->rb_node;
269 entry = rb_entry(parent, struct tree_entry, rb_node);
271 if (offset < entry->start)
273 else if (offset > entry->end)
280 rb_link_node(node, parent, p);
281 rb_insert_color(node, root);
285 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
286 struct rb_node **prev_ret,
287 struct rb_node **next_ret,
288 struct rb_node ***p_ret,
289 struct rb_node **parent_ret)
291 struct rb_root *root = &tree->state;
292 struct rb_node **n = &root->rb_node;
293 struct rb_node *prev = NULL;
294 struct rb_node *orig_prev = NULL;
295 struct tree_entry *entry;
296 struct tree_entry *prev_entry = NULL;
300 entry = rb_entry(prev, struct tree_entry, rb_node);
303 if (offset < entry->start)
305 else if (offset > entry->end)
318 while (prev && offset > prev_entry->end) {
319 prev = rb_next(prev);
320 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
327 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
328 while (prev && offset < prev_entry->start) {
329 prev = rb_prev(prev);
330 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
337 static inline struct rb_node *
338 tree_search_for_insert(struct extent_io_tree *tree,
340 struct rb_node ***p_ret,
341 struct rb_node **parent_ret)
343 struct rb_node *prev = NULL;
346 ret = __etree_search(tree, offset, &prev, NULL, p_ret, parent_ret);
352 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
355 return tree_search_for_insert(tree, offset, NULL, NULL);
358 static void merge_cb(struct extent_io_tree *tree, struct extent_state *new,
359 struct extent_state *other)
361 if (tree->ops && tree->ops->merge_extent_hook)
362 tree->ops->merge_extent_hook(tree->private_data, new, other);
366 * utility function to look for merge candidates inside a given range.
367 * Any extents with matching state are merged together into a single
368 * extent in the tree. Extents with EXTENT_IO in their state field
369 * are not merged because the end_io handlers need to be able to do
370 * operations on them without sleeping (or doing allocations/splits).
372 * This should be called with the tree lock held.
374 static void merge_state(struct extent_io_tree *tree,
375 struct extent_state *state)
377 struct extent_state *other;
378 struct rb_node *other_node;
380 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
383 other_node = rb_prev(&state->rb_node);
385 other = rb_entry(other_node, struct extent_state, rb_node);
386 if (other->end == state->start - 1 &&
387 other->state == state->state) {
388 merge_cb(tree, state, other);
389 state->start = other->start;
390 rb_erase(&other->rb_node, &tree->state);
391 RB_CLEAR_NODE(&other->rb_node);
392 free_extent_state(other);
395 other_node = rb_next(&state->rb_node);
397 other = rb_entry(other_node, struct extent_state, rb_node);
398 if (other->start == state->end + 1 &&
399 other->state == state->state) {
400 merge_cb(tree, state, other);
401 state->end = other->end;
402 rb_erase(&other->rb_node, &tree->state);
403 RB_CLEAR_NODE(&other->rb_node);
404 free_extent_state(other);
409 static void set_state_cb(struct extent_io_tree *tree,
410 struct extent_state *state, unsigned *bits)
412 if (tree->ops && tree->ops->set_bit_hook)
413 tree->ops->set_bit_hook(tree->private_data, state, bits);
416 static void clear_state_cb(struct extent_io_tree *tree,
417 struct extent_state *state, unsigned *bits)
419 if (tree->ops && tree->ops->clear_bit_hook)
420 tree->ops->clear_bit_hook(tree->private_data, state, bits);
423 static void set_state_bits(struct extent_io_tree *tree,
424 struct extent_state *state, unsigned *bits,
425 struct extent_changeset *changeset);
428 * insert an extent_state struct into the tree. 'bits' are set on the
429 * struct before it is inserted.
431 * This may return -EEXIST if the extent is already there, in which case the
432 * state struct is freed.
434 * The tree lock is not taken internally. This is a utility function and
435 * probably isn't what you want to call (see set/clear_extent_bit).
437 static int insert_state(struct extent_io_tree *tree,
438 struct extent_state *state, u64 start, u64 end,
440 struct rb_node **parent,
441 unsigned *bits, struct extent_changeset *changeset)
443 struct rb_node *node;
446 WARN(1, KERN_ERR "BTRFS: end < start %llu %llu\n",
448 state->start = start;
451 set_state_bits(tree, state, bits, changeset);
453 node = tree_insert(&tree->state, NULL, end, &state->rb_node, p, parent);
455 struct extent_state *found;
456 found = rb_entry(node, struct extent_state, rb_node);
457 pr_err("BTRFS: found node %llu %llu on insert of %llu %llu\n",
458 found->start, found->end, start, end);
461 merge_state(tree, state);
465 static void split_cb(struct extent_io_tree *tree, struct extent_state *orig,
468 if (tree->ops && tree->ops->split_extent_hook)
469 tree->ops->split_extent_hook(tree->private_data, orig, split);
473 * split a given extent state struct in two, inserting the preallocated
474 * struct 'prealloc' as the newly created second half. 'split' indicates an
475 * offset inside 'orig' where it should be split.
478 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
479 * are two extent state structs in the tree:
480 * prealloc: [orig->start, split - 1]
481 * orig: [ split, orig->end ]
483 * The tree locks are not taken by this function. They need to be held
486 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
487 struct extent_state *prealloc, u64 split)
489 struct rb_node *node;
491 split_cb(tree, orig, split);
493 prealloc->start = orig->start;
494 prealloc->end = split - 1;
495 prealloc->state = orig->state;
498 node = tree_insert(&tree->state, &orig->rb_node, prealloc->end,
499 &prealloc->rb_node, NULL, NULL);
501 free_extent_state(prealloc);
507 static struct extent_state *next_state(struct extent_state *state)
509 struct rb_node *next = rb_next(&state->rb_node);
511 return rb_entry(next, struct extent_state, rb_node);
517 * utility function to clear some bits in an extent state struct.
518 * it will optionally wake up any one waiting on this state (wake == 1).
520 * If no bits are set on the state struct after clearing things, the
521 * struct is freed and removed from the tree
523 static struct extent_state *clear_state_bit(struct extent_io_tree *tree,
524 struct extent_state *state,
525 unsigned *bits, int wake,
526 struct extent_changeset *changeset)
528 struct extent_state *next;
529 unsigned bits_to_clear = *bits & ~EXTENT_CTLBITS;
531 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
532 u64 range = state->end - state->start + 1;
533 WARN_ON(range > tree->dirty_bytes);
534 tree->dirty_bytes -= range;
536 clear_state_cb(tree, state, bits);
537 add_extent_changeset(state, bits_to_clear, changeset, 0);
538 state->state &= ~bits_to_clear;
541 if (state->state == 0) {
542 next = next_state(state);
543 if (extent_state_in_tree(state)) {
544 rb_erase(&state->rb_node, &tree->state);
545 RB_CLEAR_NODE(&state->rb_node);
546 free_extent_state(state);
551 merge_state(tree, state);
552 next = next_state(state);
557 static struct extent_state *
558 alloc_extent_state_atomic(struct extent_state *prealloc)
561 prealloc = alloc_extent_state(GFP_ATOMIC);
566 static void extent_io_tree_panic(struct extent_io_tree *tree, int err)
568 btrfs_panic(tree_fs_info(tree), err,
569 "Locking error: Extent tree was modified by another thread while locked.");
573 * clear some bits on a range in the tree. This may require splitting
574 * or inserting elements in the tree, so the gfp mask is used to
575 * indicate which allocations or sleeping are allowed.
577 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
578 * the given range from the tree regardless of state (ie for truncate).
580 * the range [start, end] is inclusive.
582 * This takes the tree lock, and returns 0 on success and < 0 on error.
584 int __clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
585 unsigned bits, int wake, int delete,
586 struct extent_state **cached_state,
587 gfp_t mask, struct extent_changeset *changeset)
589 struct extent_state *state;
590 struct extent_state *cached;
591 struct extent_state *prealloc = NULL;
592 struct rb_node *node;
597 btrfs_debug_check_extent_io_range(tree, start, end);
599 if (bits & EXTENT_DELALLOC)
600 bits |= EXTENT_NORESERVE;
603 bits |= ~EXTENT_CTLBITS;
604 bits |= EXTENT_FIRST_DELALLOC;
606 if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY))
609 if (!prealloc && gfpflags_allow_blocking(mask)) {
611 * Don't care for allocation failure here because we might end
612 * up not needing the pre-allocated extent state at all, which
613 * is the case if we only have in the tree extent states that
614 * cover our input range and don't cover too any other range.
615 * If we end up needing a new extent state we allocate it later.
617 prealloc = alloc_extent_state(mask);
620 spin_lock(&tree->lock);
622 cached = *cached_state;
625 *cached_state = NULL;
629 if (cached && extent_state_in_tree(cached) &&
630 cached->start <= start && cached->end > start) {
632 refcount_dec(&cached->refs);
637 free_extent_state(cached);
640 * this search will find the extents that end after
643 node = tree_search(tree, start);
646 state = rb_entry(node, struct extent_state, rb_node);
648 if (state->start > end)
650 WARN_ON(state->end < start);
651 last_end = state->end;
653 /* the state doesn't have the wanted bits, go ahead */
654 if (!(state->state & bits)) {
655 state = next_state(state);
660 * | ---- desired range ---- |
662 * | ------------- state -------------- |
664 * We need to split the extent we found, and may flip
665 * bits on second half.
667 * If the extent we found extends past our range, we
668 * just split and search again. It'll get split again
669 * the next time though.
671 * If the extent we found is inside our range, we clear
672 * the desired bit on it.
675 if (state->start < start) {
676 prealloc = alloc_extent_state_atomic(prealloc);
678 err = split_state(tree, state, prealloc, start);
680 extent_io_tree_panic(tree, err);
685 if (state->end <= end) {
686 state = clear_state_bit(tree, state, &bits, wake,
693 * | ---- desired range ---- |
695 * We need to split the extent, and clear the bit
698 if (state->start <= end && state->end > end) {
699 prealloc = alloc_extent_state_atomic(prealloc);
701 err = split_state(tree, state, prealloc, end + 1);
703 extent_io_tree_panic(tree, err);
708 clear_state_bit(tree, prealloc, &bits, wake, changeset);
714 state = clear_state_bit(tree, state, &bits, wake, changeset);
716 if (last_end == (u64)-1)
718 start = last_end + 1;
719 if (start <= end && state && !need_resched())
725 spin_unlock(&tree->lock);
726 if (gfpflags_allow_blocking(mask))
731 spin_unlock(&tree->lock);
733 free_extent_state(prealloc);
739 static void wait_on_state(struct extent_io_tree *tree,
740 struct extent_state *state)
741 __releases(tree->lock)
742 __acquires(tree->lock)
745 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
746 spin_unlock(&tree->lock);
748 spin_lock(&tree->lock);
749 finish_wait(&state->wq, &wait);
753 * waits for one or more bits to clear on a range in the state tree.
754 * The range [start, end] is inclusive.
755 * The tree lock is taken by this function
757 static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
760 struct extent_state *state;
761 struct rb_node *node;
763 btrfs_debug_check_extent_io_range(tree, start, end);
765 spin_lock(&tree->lock);
769 * this search will find all the extents that end after
772 node = tree_search(tree, start);
777 state = rb_entry(node, struct extent_state, rb_node);
779 if (state->start > end)
782 if (state->state & bits) {
783 start = state->start;
784 refcount_inc(&state->refs);
785 wait_on_state(tree, state);
786 free_extent_state(state);
789 start = state->end + 1;
794 if (!cond_resched_lock(&tree->lock)) {
795 node = rb_next(node);
800 spin_unlock(&tree->lock);
803 static void set_state_bits(struct extent_io_tree *tree,
804 struct extent_state *state,
805 unsigned *bits, struct extent_changeset *changeset)
807 unsigned bits_to_set = *bits & ~EXTENT_CTLBITS;
809 set_state_cb(tree, state, bits);
810 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
811 u64 range = state->end - state->start + 1;
812 tree->dirty_bytes += range;
814 add_extent_changeset(state, bits_to_set, changeset, 1);
815 state->state |= bits_to_set;
818 static void cache_state_if_flags(struct extent_state *state,
819 struct extent_state **cached_ptr,
822 if (cached_ptr && !(*cached_ptr)) {
823 if (!flags || (state->state & flags)) {
825 refcount_inc(&state->refs);
830 static void cache_state(struct extent_state *state,
831 struct extent_state **cached_ptr)
833 return cache_state_if_flags(state, cached_ptr,
834 EXTENT_IOBITS | EXTENT_BOUNDARY);
838 * set some bits on a range in the tree. This may require allocations or
839 * sleeping, so the gfp mask is used to indicate what is allowed.
841 * If any of the exclusive bits are set, this will fail with -EEXIST if some
842 * part of the range already has the desired bits set. The start of the
843 * existing range is returned in failed_start in this case.
845 * [start, end] is inclusive This takes the tree lock.
848 static int __must_check
849 __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
850 unsigned bits, unsigned exclusive_bits,
851 u64 *failed_start, struct extent_state **cached_state,
852 gfp_t mask, struct extent_changeset *changeset)
854 struct extent_state *state;
855 struct extent_state *prealloc = NULL;
856 struct rb_node *node;
858 struct rb_node *parent;
863 btrfs_debug_check_extent_io_range(tree, start, end);
865 bits |= EXTENT_FIRST_DELALLOC;
867 if (!prealloc && gfpflags_allow_blocking(mask)) {
869 * Don't care for allocation failure here because we might end
870 * up not needing the pre-allocated extent state at all, which
871 * is the case if we only have in the tree extent states that
872 * cover our input range and don't cover too any other range.
873 * If we end up needing a new extent state we allocate it later.
875 prealloc = alloc_extent_state(mask);
878 spin_lock(&tree->lock);
879 if (cached_state && *cached_state) {
880 state = *cached_state;
881 if (state->start <= start && state->end > start &&
882 extent_state_in_tree(state)) {
883 node = &state->rb_node;
888 * this search will find all the extents that end after
891 node = tree_search_for_insert(tree, start, &p, &parent);
893 prealloc = alloc_extent_state_atomic(prealloc);
895 err = insert_state(tree, prealloc, start, end,
896 &p, &parent, &bits, changeset);
898 extent_io_tree_panic(tree, err);
900 cache_state(prealloc, cached_state);
904 state = rb_entry(node, struct extent_state, rb_node);
906 last_start = state->start;
907 last_end = state->end;
910 * | ---- desired range ---- |
913 * Just lock what we found and keep going
915 if (state->start == start && state->end <= end) {
916 if (state->state & exclusive_bits) {
917 *failed_start = state->start;
922 set_state_bits(tree, state, &bits, changeset);
923 cache_state(state, cached_state);
924 merge_state(tree, state);
925 if (last_end == (u64)-1)
927 start = last_end + 1;
928 state = next_state(state);
929 if (start < end && state && state->start == start &&
936 * | ---- desired range ---- |
939 * | ------------- state -------------- |
941 * We need to split the extent we found, and may flip bits on
944 * If the extent we found extends past our
945 * range, we just split and search again. It'll get split
946 * again the next time though.
948 * If the extent we found is inside our range, we set the
951 if (state->start < start) {
952 if (state->state & exclusive_bits) {
953 *failed_start = start;
958 prealloc = alloc_extent_state_atomic(prealloc);
960 err = split_state(tree, state, prealloc, start);
962 extent_io_tree_panic(tree, err);
967 if (state->end <= end) {
968 set_state_bits(tree, state, &bits, changeset);
969 cache_state(state, cached_state);
970 merge_state(tree, state);
971 if (last_end == (u64)-1)
973 start = last_end + 1;
974 state = next_state(state);
975 if (start < end && state && state->start == start &&
982 * | ---- desired range ---- |
983 * | state | or | state |
985 * There's a hole, we need to insert something in it and
986 * ignore the extent we found.
988 if (state->start > start) {
990 if (end < last_start)
993 this_end = last_start - 1;
995 prealloc = alloc_extent_state_atomic(prealloc);
999 * Avoid to free 'prealloc' if it can be merged with
1002 err = insert_state(tree, prealloc, start, this_end,
1003 NULL, NULL, &bits, changeset);
1005 extent_io_tree_panic(tree, err);
1007 cache_state(prealloc, cached_state);
1009 start = this_end + 1;
1013 * | ---- desired range ---- |
1015 * We need to split the extent, and set the bit
1018 if (state->start <= end && state->end > end) {
1019 if (state->state & exclusive_bits) {
1020 *failed_start = start;
1025 prealloc = alloc_extent_state_atomic(prealloc);
1027 err = split_state(tree, state, prealloc, end + 1);
1029 extent_io_tree_panic(tree, err);
1031 set_state_bits(tree, prealloc, &bits, changeset);
1032 cache_state(prealloc, cached_state);
1033 merge_state(tree, prealloc);
1041 spin_unlock(&tree->lock);
1042 if (gfpflags_allow_blocking(mask))
1047 spin_unlock(&tree->lock);
1049 free_extent_state(prealloc);
1055 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1056 unsigned bits, u64 * failed_start,
1057 struct extent_state **cached_state, gfp_t mask)
1059 return __set_extent_bit(tree, start, end, bits, 0, failed_start,
1060 cached_state, mask, NULL);
1065 * convert_extent_bit - convert all bits in a given range from one bit to
1067 * @tree: the io tree to search
1068 * @start: the start offset in bytes
1069 * @end: the end offset in bytes (inclusive)
1070 * @bits: the bits to set in this range
1071 * @clear_bits: the bits to clear in this range
1072 * @cached_state: state that we're going to cache
1074 * This will go through and set bits for the given range. If any states exist
1075 * already in this range they are set with the given bit and cleared of the
1076 * clear_bits. This is only meant to be used by things that are mergeable, ie
1077 * converting from say DELALLOC to DIRTY. This is not meant to be used with
1078 * boundary bits like LOCK.
1080 * All allocations are done with GFP_NOFS.
1082 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1083 unsigned bits, unsigned clear_bits,
1084 struct extent_state **cached_state)
1086 struct extent_state *state;
1087 struct extent_state *prealloc = NULL;
1088 struct rb_node *node;
1090 struct rb_node *parent;
1094 bool first_iteration = true;
1096 btrfs_debug_check_extent_io_range(tree, start, end);
1101 * Best effort, don't worry if extent state allocation fails
1102 * here for the first iteration. We might have a cached state
1103 * that matches exactly the target range, in which case no
1104 * extent state allocations are needed. We'll only know this
1105 * after locking the tree.
1107 prealloc = alloc_extent_state(GFP_NOFS);
1108 if (!prealloc && !first_iteration)
1112 spin_lock(&tree->lock);
1113 if (cached_state && *cached_state) {
1114 state = *cached_state;
1115 if (state->start <= start && state->end > start &&
1116 extent_state_in_tree(state)) {
1117 node = &state->rb_node;
1123 * this search will find all the extents that end after
1126 node = tree_search_for_insert(tree, start, &p, &parent);
1128 prealloc = alloc_extent_state_atomic(prealloc);
1133 err = insert_state(tree, prealloc, start, end,
1134 &p, &parent, &bits, NULL);
1136 extent_io_tree_panic(tree, err);
1137 cache_state(prealloc, cached_state);
1141 state = rb_entry(node, struct extent_state, rb_node);
1143 last_start = state->start;
1144 last_end = state->end;
1147 * | ---- desired range ---- |
1150 * Just lock what we found and keep going
1152 if (state->start == start && state->end <= end) {
1153 set_state_bits(tree, state, &bits, NULL);
1154 cache_state(state, cached_state);
1155 state = clear_state_bit(tree, state, &clear_bits, 0, NULL);
1156 if (last_end == (u64)-1)
1158 start = last_end + 1;
1159 if (start < end && state && state->start == start &&
1166 * | ---- desired range ---- |
1169 * | ------------- state -------------- |
1171 * We need to split the extent we found, and may flip bits on
1174 * If the extent we found extends past our
1175 * range, we just split and search again. It'll get split
1176 * again the next time though.
1178 * If the extent we found is inside our range, we set the
1179 * desired bit on it.
1181 if (state->start < start) {
1182 prealloc = alloc_extent_state_atomic(prealloc);
1187 err = split_state(tree, state, prealloc, start);
1189 extent_io_tree_panic(tree, err);
1193 if (state->end <= end) {
1194 set_state_bits(tree, state, &bits, NULL);
1195 cache_state(state, cached_state);
1196 state = clear_state_bit(tree, state, &clear_bits, 0,
1198 if (last_end == (u64)-1)
1200 start = last_end + 1;
1201 if (start < end && state && state->start == start &&
1208 * | ---- desired range ---- |
1209 * | state | or | state |
1211 * There's a hole, we need to insert something in it and
1212 * ignore the extent we found.
1214 if (state->start > start) {
1216 if (end < last_start)
1219 this_end = last_start - 1;
1221 prealloc = alloc_extent_state_atomic(prealloc);
1228 * Avoid to free 'prealloc' if it can be merged with
1231 err = insert_state(tree, prealloc, start, this_end,
1232 NULL, NULL, &bits, NULL);
1234 extent_io_tree_panic(tree, err);
1235 cache_state(prealloc, cached_state);
1237 start = this_end + 1;
1241 * | ---- desired range ---- |
1243 * We need to split the extent, and set the bit
1246 if (state->start <= end && state->end > end) {
1247 prealloc = alloc_extent_state_atomic(prealloc);
1253 err = split_state(tree, state, prealloc, end + 1);
1255 extent_io_tree_panic(tree, err);
1257 set_state_bits(tree, prealloc, &bits, NULL);
1258 cache_state(prealloc, cached_state);
1259 clear_state_bit(tree, prealloc, &clear_bits, 0, NULL);
1267 spin_unlock(&tree->lock);
1269 first_iteration = false;
1273 spin_unlock(&tree->lock);
1275 free_extent_state(prealloc);
1280 /* wrappers around set/clear extent bit */
1281 int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1282 unsigned bits, struct extent_changeset *changeset)
1285 * We don't support EXTENT_LOCKED yet, as current changeset will
1286 * record any bits changed, so for EXTENT_LOCKED case, it will
1287 * either fail with -EEXIST or changeset will record the whole
1290 BUG_ON(bits & EXTENT_LOCKED);
1292 return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL, GFP_NOFS,
1296 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1297 unsigned bits, int wake, int delete,
1298 struct extent_state **cached)
1300 return __clear_extent_bit(tree, start, end, bits, wake, delete,
1301 cached, GFP_NOFS, NULL);
1304 int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1305 unsigned bits, struct extent_changeset *changeset)
1308 * Don't support EXTENT_LOCKED case, same reason as
1309 * set_record_extent_bits().
1311 BUG_ON(bits & EXTENT_LOCKED);
1313 return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, GFP_NOFS,
1318 * either insert or lock state struct between start and end use mask to tell
1319 * us if waiting is desired.
1321 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1322 struct extent_state **cached_state)
1328 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED,
1329 EXTENT_LOCKED, &failed_start,
1330 cached_state, GFP_NOFS, NULL);
1331 if (err == -EEXIST) {
1332 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1333 start = failed_start;
1336 WARN_ON(start > end);
1341 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1346 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1347 &failed_start, NULL, GFP_NOFS, NULL);
1348 if (err == -EEXIST) {
1349 if (failed_start > start)
1350 clear_extent_bit(tree, start, failed_start - 1,
1351 EXTENT_LOCKED, 1, 0, NULL);
1357 void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
1359 unsigned long index = start >> PAGE_SHIFT;
1360 unsigned long end_index = end >> PAGE_SHIFT;
1363 while (index <= end_index) {
1364 page = find_get_page(inode->i_mapping, index);
1365 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1366 clear_page_dirty_for_io(page);
1372 void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
1374 unsigned long index = start >> PAGE_SHIFT;
1375 unsigned long end_index = end >> PAGE_SHIFT;
1378 while (index <= end_index) {
1379 page = find_get_page(inode->i_mapping, index);
1380 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1381 __set_page_dirty_nobuffers(page);
1382 account_page_redirty(page);
1389 * helper function to set both pages and extents in the tree writeback
1391 static void set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
1393 tree->ops->set_range_writeback(tree->private_data, start, end);
1396 /* find the first state struct with 'bits' set after 'start', and
1397 * return it. tree->lock must be held. NULL will returned if
1398 * nothing was found after 'start'
1400 static struct extent_state *
1401 find_first_extent_bit_state(struct extent_io_tree *tree,
1402 u64 start, unsigned bits)
1404 struct rb_node *node;
1405 struct extent_state *state;
1408 * this search will find all the extents that end after
1411 node = tree_search(tree, start);
1416 state = rb_entry(node, struct extent_state, rb_node);
1417 if (state->end >= start && (state->state & bits))
1420 node = rb_next(node);
1429 * find the first offset in the io tree with 'bits' set. zero is
1430 * returned if we find something, and *start_ret and *end_ret are
1431 * set to reflect the state struct that was found.
1433 * If nothing was found, 1 is returned. If found something, return 0.
1435 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1436 u64 *start_ret, u64 *end_ret, unsigned bits,
1437 struct extent_state **cached_state)
1439 struct extent_state *state;
1443 spin_lock(&tree->lock);
1444 if (cached_state && *cached_state) {
1445 state = *cached_state;
1446 if (state->end == start - 1 && extent_state_in_tree(state)) {
1447 n = rb_next(&state->rb_node);
1449 state = rb_entry(n, struct extent_state,
1451 if (state->state & bits)
1455 free_extent_state(*cached_state);
1456 *cached_state = NULL;
1459 free_extent_state(*cached_state);
1460 *cached_state = NULL;
1463 state = find_first_extent_bit_state(tree, start, bits);
1466 cache_state_if_flags(state, cached_state, 0);
1467 *start_ret = state->start;
1468 *end_ret = state->end;
1472 spin_unlock(&tree->lock);
1477 * find a contiguous range of bytes in the file marked as delalloc, not
1478 * more than 'max_bytes'. start and end are used to return the range,
1480 * 1 is returned if we find something, 0 if nothing was in the tree
1482 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1483 u64 *start, u64 *end, u64 max_bytes,
1484 struct extent_state **cached_state)
1486 struct rb_node *node;
1487 struct extent_state *state;
1488 u64 cur_start = *start;
1490 u64 total_bytes = 0;
1492 spin_lock(&tree->lock);
1495 * this search will find all the extents that end after
1498 node = tree_search(tree, cur_start);
1506 state = rb_entry(node, struct extent_state, rb_node);
1507 if (found && (state->start != cur_start ||
1508 (state->state & EXTENT_BOUNDARY))) {
1511 if (!(state->state & EXTENT_DELALLOC)) {
1517 *start = state->start;
1518 *cached_state = state;
1519 refcount_inc(&state->refs);
1523 cur_start = state->end + 1;
1524 node = rb_next(node);
1525 total_bytes += state->end - state->start + 1;
1526 if (total_bytes >= max_bytes)
1532 spin_unlock(&tree->lock);
1536 static int __process_pages_contig(struct address_space *mapping,
1537 struct page *locked_page,
1538 pgoff_t start_index, pgoff_t end_index,
1539 unsigned long page_ops, pgoff_t *index_ret);
1541 static noinline void __unlock_for_delalloc(struct inode *inode,
1542 struct page *locked_page,
1545 unsigned long index = start >> PAGE_SHIFT;
1546 unsigned long end_index = end >> PAGE_SHIFT;
1548 ASSERT(locked_page);
1549 if (index == locked_page->index && end_index == index)
1552 __process_pages_contig(inode->i_mapping, locked_page, index, end_index,
1556 static noinline int lock_delalloc_pages(struct inode *inode,
1557 struct page *locked_page,
1561 unsigned long index = delalloc_start >> PAGE_SHIFT;
1562 unsigned long index_ret = index;
1563 unsigned long end_index = delalloc_end >> PAGE_SHIFT;
1566 ASSERT(locked_page);
1567 if (index == locked_page->index && index == end_index)
1570 ret = __process_pages_contig(inode->i_mapping, locked_page, index,
1571 end_index, PAGE_LOCK, &index_ret);
1573 __unlock_for_delalloc(inode, locked_page, delalloc_start,
1574 (u64)index_ret << PAGE_SHIFT);
1579 * find a contiguous range of bytes in the file marked as delalloc, not
1580 * more than 'max_bytes'. start and end are used to return the range,
1582 * 1 is returned if we find something, 0 if nothing was in the tree
1584 STATIC u64 find_lock_delalloc_range(struct inode *inode,
1585 struct extent_io_tree *tree,
1586 struct page *locked_page, u64 *start,
1587 u64 *end, u64 max_bytes)
1592 struct extent_state *cached_state = NULL;
1597 /* step one, find a bunch of delalloc bytes starting at start */
1598 delalloc_start = *start;
1600 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1601 max_bytes, &cached_state);
1602 if (!found || delalloc_end <= *start) {
1603 *start = delalloc_start;
1604 *end = delalloc_end;
1605 free_extent_state(cached_state);
1610 * start comes from the offset of locked_page. We have to lock
1611 * pages in order, so we can't process delalloc bytes before
1614 if (delalloc_start < *start)
1615 delalloc_start = *start;
1618 * make sure to limit the number of pages we try to lock down
1620 if (delalloc_end + 1 - delalloc_start > max_bytes)
1621 delalloc_end = delalloc_start + max_bytes - 1;
1623 /* step two, lock all the pages after the page that has start */
1624 ret = lock_delalloc_pages(inode, locked_page,
1625 delalloc_start, delalloc_end);
1626 if (ret == -EAGAIN) {
1627 /* some of the pages are gone, lets avoid looping by
1628 * shortening the size of the delalloc range we're searching
1630 free_extent_state(cached_state);
1631 cached_state = NULL;
1633 max_bytes = PAGE_SIZE;
1641 BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */
1643 /* step three, lock the state bits for the whole range */
1644 lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state);
1646 /* then test to make sure it is all still delalloc */
1647 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1648 EXTENT_DELALLOC, 1, cached_state);
1650 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1652 __unlock_for_delalloc(inode, locked_page,
1653 delalloc_start, delalloc_end);
1657 free_extent_state(cached_state);
1658 *start = delalloc_start;
1659 *end = delalloc_end;
1664 static int __process_pages_contig(struct address_space *mapping,
1665 struct page *locked_page,
1666 pgoff_t start_index, pgoff_t end_index,
1667 unsigned long page_ops, pgoff_t *index_ret)
1669 unsigned long nr_pages = end_index - start_index + 1;
1670 unsigned long pages_locked = 0;
1671 pgoff_t index = start_index;
1672 struct page *pages[16];
1677 if (page_ops & PAGE_LOCK) {
1678 ASSERT(page_ops == PAGE_LOCK);
1679 ASSERT(index_ret && *index_ret == start_index);
1682 if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0)
1683 mapping_set_error(mapping, -EIO);
1685 while (nr_pages > 0) {
1686 ret = find_get_pages_contig(mapping, index,
1687 min_t(unsigned long,
1688 nr_pages, ARRAY_SIZE(pages)), pages);
1691 * Only if we're going to lock these pages,
1692 * can we find nothing at @index.
1694 ASSERT(page_ops & PAGE_LOCK);
1699 for (i = 0; i < ret; i++) {
1700 if (page_ops & PAGE_SET_PRIVATE2)
1701 SetPagePrivate2(pages[i]);
1703 if (pages[i] == locked_page) {
1708 if (page_ops & PAGE_CLEAR_DIRTY)
1709 clear_page_dirty_for_io(pages[i]);
1710 if (page_ops & PAGE_SET_WRITEBACK)
1711 set_page_writeback(pages[i]);
1712 if (page_ops & PAGE_SET_ERROR)
1713 SetPageError(pages[i]);
1714 if (page_ops & PAGE_END_WRITEBACK)
1715 end_page_writeback(pages[i]);
1716 if (page_ops & PAGE_UNLOCK)
1717 unlock_page(pages[i]);
1718 if (page_ops & PAGE_LOCK) {
1719 lock_page(pages[i]);
1720 if (!PageDirty(pages[i]) ||
1721 pages[i]->mapping != mapping) {
1722 unlock_page(pages[i]);
1736 if (err && index_ret)
1737 *index_ret = start_index + pages_locked - 1;
1741 void extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end,
1742 u64 delalloc_end, struct page *locked_page,
1743 unsigned clear_bits,
1744 unsigned long page_ops)
1746 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, clear_bits, 1, 0,
1749 __process_pages_contig(inode->i_mapping, locked_page,
1750 start >> PAGE_SHIFT, end >> PAGE_SHIFT,
1755 * count the number of bytes in the tree that have a given bit(s)
1756 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1757 * cached. The total number found is returned.
1759 u64 count_range_bits(struct extent_io_tree *tree,
1760 u64 *start, u64 search_end, u64 max_bytes,
1761 unsigned bits, int contig)
1763 struct rb_node *node;
1764 struct extent_state *state;
1765 u64 cur_start = *start;
1766 u64 total_bytes = 0;
1770 if (WARN_ON(search_end <= cur_start))
1773 spin_lock(&tree->lock);
1774 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1775 total_bytes = tree->dirty_bytes;
1779 * this search will find all the extents that end after
1782 node = tree_search(tree, cur_start);
1787 state = rb_entry(node, struct extent_state, rb_node);
1788 if (state->start > search_end)
1790 if (contig && found && state->start > last + 1)
1792 if (state->end >= cur_start && (state->state & bits) == bits) {
1793 total_bytes += min(search_end, state->end) + 1 -
1794 max(cur_start, state->start);
1795 if (total_bytes >= max_bytes)
1798 *start = max(cur_start, state->start);
1802 } else if (contig && found) {
1805 node = rb_next(node);
1810 spin_unlock(&tree->lock);
1815 * set the private field for a given byte offset in the tree. If there isn't
1816 * an extent_state there already, this does nothing.
1818 static noinline int set_state_failrec(struct extent_io_tree *tree, u64 start,
1819 struct io_failure_record *failrec)
1821 struct rb_node *node;
1822 struct extent_state *state;
1825 spin_lock(&tree->lock);
1827 * this search will find all the extents that end after
1830 node = tree_search(tree, start);
1835 state = rb_entry(node, struct extent_state, rb_node);
1836 if (state->start != start) {
1840 state->failrec = failrec;
1842 spin_unlock(&tree->lock);
1846 static noinline int get_state_failrec(struct extent_io_tree *tree, u64 start,
1847 struct io_failure_record **failrec)
1849 struct rb_node *node;
1850 struct extent_state *state;
1853 spin_lock(&tree->lock);
1855 * this search will find all the extents that end after
1858 node = tree_search(tree, start);
1863 state = rb_entry(node, struct extent_state, rb_node);
1864 if (state->start != start) {
1868 *failrec = state->failrec;
1870 spin_unlock(&tree->lock);
1875 * searches a range in the state tree for a given mask.
1876 * If 'filled' == 1, this returns 1 only if every extent in the tree
1877 * has the bits set. Otherwise, 1 is returned if any bit in the
1878 * range is found set.
1880 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1881 unsigned bits, int filled, struct extent_state *cached)
1883 struct extent_state *state = NULL;
1884 struct rb_node *node;
1887 spin_lock(&tree->lock);
1888 if (cached && extent_state_in_tree(cached) && cached->start <= start &&
1889 cached->end > start)
1890 node = &cached->rb_node;
1892 node = tree_search(tree, start);
1893 while (node && start <= end) {
1894 state = rb_entry(node, struct extent_state, rb_node);
1896 if (filled && state->start > start) {
1901 if (state->start > end)
1904 if (state->state & bits) {
1908 } else if (filled) {
1913 if (state->end == (u64)-1)
1916 start = state->end + 1;
1919 node = rb_next(node);
1926 spin_unlock(&tree->lock);
1931 * helper function to set a given page up to date if all the
1932 * extents in the tree for that page are up to date
1934 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
1936 u64 start = page_offset(page);
1937 u64 end = start + PAGE_SIZE - 1;
1938 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1939 SetPageUptodate(page);
1942 int free_io_failure(struct extent_io_tree *failure_tree,
1943 struct extent_io_tree *io_tree,
1944 struct io_failure_record *rec)
1949 set_state_failrec(failure_tree, rec->start, NULL);
1950 ret = clear_extent_bits(failure_tree, rec->start,
1951 rec->start + rec->len - 1,
1952 EXTENT_LOCKED | EXTENT_DIRTY);
1956 ret = clear_extent_bits(io_tree, rec->start,
1957 rec->start + rec->len - 1,
1967 * this bypasses the standard btrfs submit functions deliberately, as
1968 * the standard behavior is to write all copies in a raid setup. here we only
1969 * want to write the one bad copy. so we do the mapping for ourselves and issue
1970 * submit_bio directly.
1971 * to avoid any synchronization issues, wait for the data after writing, which
1972 * actually prevents the read that triggered the error from finishing.
1973 * currently, there can be no more than two copies of every data bit. thus,
1974 * exactly one rewrite is required.
1976 int repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start,
1977 u64 length, u64 logical, struct page *page,
1978 unsigned int pg_offset, int mirror_num)
1981 struct btrfs_device *dev;
1984 struct btrfs_bio *bbio = NULL;
1987 ASSERT(!(fs_info->sb->s_flags & SB_RDONLY));
1988 BUG_ON(!mirror_num);
1990 bio = btrfs_io_bio_alloc(1);
1991 bio->bi_iter.bi_size = 0;
1992 map_length = length;
1995 * Avoid races with device replace and make sure our bbio has devices
1996 * associated to its stripes that don't go away while we are doing the
1997 * read repair operation.
1999 btrfs_bio_counter_inc_blocked(fs_info);
2000 if (btrfs_is_parity_mirror(fs_info, logical, length)) {
2002 * Note that we don't use BTRFS_MAP_WRITE because it's supposed
2003 * to update all raid stripes, but here we just want to correct
2004 * bad stripe, thus BTRFS_MAP_READ is abused to only get the bad
2005 * stripe's dev and sector.
2007 ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, logical,
2008 &map_length, &bbio, 0);
2010 btrfs_bio_counter_dec(fs_info);
2014 ASSERT(bbio->mirror_num == 1);
2016 ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical,
2017 &map_length, &bbio, mirror_num);
2019 btrfs_bio_counter_dec(fs_info);
2023 BUG_ON(mirror_num != bbio->mirror_num);
2026 sector = bbio->stripes[bbio->mirror_num - 1].physical >> 9;
2027 bio->bi_iter.bi_sector = sector;
2028 dev = bbio->stripes[bbio->mirror_num - 1].dev;
2029 btrfs_put_bbio(bbio);
2030 if (!dev || !dev->bdev ||
2031 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) {
2032 btrfs_bio_counter_dec(fs_info);
2036 bio_set_dev(bio, dev->bdev);
2037 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC;
2038 bio_add_page(bio, page, length, pg_offset);
2040 if (btrfsic_submit_bio_wait(bio)) {
2041 /* try to remap that extent elsewhere? */
2042 btrfs_bio_counter_dec(fs_info);
2044 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
2048 btrfs_info_rl_in_rcu(fs_info,
2049 "read error corrected: ino %llu off %llu (dev %s sector %llu)",
2051 rcu_str_deref(dev->name), sector);
2052 btrfs_bio_counter_dec(fs_info);
2057 int repair_eb_io_failure(struct btrfs_fs_info *fs_info,
2058 struct extent_buffer *eb, int mirror_num)
2060 u64 start = eb->start;
2061 unsigned long i, num_pages = num_extent_pages(eb->start, eb->len);
2064 if (sb_rdonly(fs_info->sb))
2067 for (i = 0; i < num_pages; i++) {
2068 struct page *p = eb->pages[i];
2070 ret = repair_io_failure(fs_info, 0, start, PAGE_SIZE, start, p,
2071 start - page_offset(p), mirror_num);
2081 * each time an IO finishes, we do a fast check in the IO failure tree
2082 * to see if we need to process or clean up an io_failure_record
2084 int clean_io_failure(struct btrfs_fs_info *fs_info,
2085 struct extent_io_tree *failure_tree,
2086 struct extent_io_tree *io_tree, u64 start,
2087 struct page *page, u64 ino, unsigned int pg_offset)
2090 struct io_failure_record *failrec;
2091 struct extent_state *state;
2096 ret = count_range_bits(failure_tree, &private, (u64)-1, 1,
2101 ret = get_state_failrec(failure_tree, start, &failrec);
2105 BUG_ON(!failrec->this_mirror);
2107 if (failrec->in_validation) {
2108 /* there was no real error, just free the record */
2109 btrfs_debug(fs_info,
2110 "clean_io_failure: freeing dummy error at %llu",
2114 if (sb_rdonly(fs_info->sb))
2117 spin_lock(&io_tree->lock);
2118 state = find_first_extent_bit_state(io_tree,
2121 spin_unlock(&io_tree->lock);
2123 if (state && state->start <= failrec->start &&
2124 state->end >= failrec->start + failrec->len - 1) {
2125 num_copies = btrfs_num_copies(fs_info, failrec->logical,
2127 if (num_copies > 1) {
2128 repair_io_failure(fs_info, ino, start, failrec->len,
2129 failrec->logical, page, pg_offset,
2130 failrec->failed_mirror);
2135 free_io_failure(failure_tree, io_tree, failrec);
2141 * Can be called when
2142 * - hold extent lock
2143 * - under ordered extent
2144 * - the inode is freeing
2146 void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start, u64 end)
2148 struct extent_io_tree *failure_tree = &inode->io_failure_tree;
2149 struct io_failure_record *failrec;
2150 struct extent_state *state, *next;
2152 if (RB_EMPTY_ROOT(&failure_tree->state))
2155 spin_lock(&failure_tree->lock);
2156 state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY);
2158 if (state->start > end)
2161 ASSERT(state->end <= end);
2163 next = next_state(state);
2165 failrec = state->failrec;
2166 free_extent_state(state);
2171 spin_unlock(&failure_tree->lock);
2174 int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end,
2175 struct io_failure_record **failrec_ret)
2177 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2178 struct io_failure_record *failrec;
2179 struct extent_map *em;
2180 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2181 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2182 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2186 ret = get_state_failrec(failure_tree, start, &failrec);
2188 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2192 failrec->start = start;
2193 failrec->len = end - start + 1;
2194 failrec->this_mirror = 0;
2195 failrec->bio_flags = 0;
2196 failrec->in_validation = 0;
2198 read_lock(&em_tree->lock);
2199 em = lookup_extent_mapping(em_tree, start, failrec->len);
2201 read_unlock(&em_tree->lock);
2206 if (em->start > start || em->start + em->len <= start) {
2207 free_extent_map(em);
2210 read_unlock(&em_tree->lock);
2216 logical = start - em->start;
2217 logical = em->block_start + logical;
2218 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2219 logical = em->block_start;
2220 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2221 extent_set_compress_type(&failrec->bio_flags,
2225 btrfs_debug(fs_info,
2226 "Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu",
2227 logical, start, failrec->len);
2229 failrec->logical = logical;
2230 free_extent_map(em);
2232 /* set the bits in the private failure tree */
2233 ret = set_extent_bits(failure_tree, start, end,
2234 EXTENT_LOCKED | EXTENT_DIRTY);
2236 ret = set_state_failrec(failure_tree, start, failrec);
2237 /* set the bits in the inode's tree */
2239 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED);
2245 btrfs_debug(fs_info,
2246 "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d",
2247 failrec->logical, failrec->start, failrec->len,
2248 failrec->in_validation);
2250 * when data can be on disk more than twice, add to failrec here
2251 * (e.g. with a list for failed_mirror) to make
2252 * clean_io_failure() clean all those errors at once.
2256 *failrec_ret = failrec;
2261 bool btrfs_check_repairable(struct inode *inode, unsigned failed_bio_pages,
2262 struct io_failure_record *failrec, int failed_mirror)
2264 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2267 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
2268 if (num_copies == 1) {
2270 * we only have a single copy of the data, so don't bother with
2271 * all the retry and error correction code that follows. no
2272 * matter what the error is, it is very likely to persist.
2274 btrfs_debug(fs_info,
2275 "Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
2276 num_copies, failrec->this_mirror, failed_mirror);
2281 * there are two premises:
2282 * a) deliver good data to the caller
2283 * b) correct the bad sectors on disk
2285 if (failed_bio_pages > 1) {
2287 * to fulfill b), we need to know the exact failing sectors, as
2288 * we don't want to rewrite any more than the failed ones. thus,
2289 * we need separate read requests for the failed bio
2291 * if the following BUG_ON triggers, our validation request got
2292 * merged. we need separate requests for our algorithm to work.
2294 BUG_ON(failrec->in_validation);
2295 failrec->in_validation = 1;
2296 failrec->this_mirror = failed_mirror;
2299 * we're ready to fulfill a) and b) alongside. get a good copy
2300 * of the failed sector and if we succeed, we have setup
2301 * everything for repair_io_failure to do the rest for us.
2303 if (failrec->in_validation) {
2304 BUG_ON(failrec->this_mirror != failed_mirror);
2305 failrec->in_validation = 0;
2306 failrec->this_mirror = 0;
2308 failrec->failed_mirror = failed_mirror;
2309 failrec->this_mirror++;
2310 if (failrec->this_mirror == failed_mirror)
2311 failrec->this_mirror++;
2314 if (failrec->this_mirror > num_copies) {
2315 btrfs_debug(fs_info,
2316 "Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
2317 num_copies, failrec->this_mirror, failed_mirror);
2325 struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio,
2326 struct io_failure_record *failrec,
2327 struct page *page, int pg_offset, int icsum,
2328 bio_end_io_t *endio_func, void *data)
2330 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2332 struct btrfs_io_bio *btrfs_failed_bio;
2333 struct btrfs_io_bio *btrfs_bio;
2335 bio = btrfs_io_bio_alloc(1);
2336 bio->bi_end_io = endio_func;
2337 bio->bi_iter.bi_sector = failrec->logical >> 9;
2338 bio_set_dev(bio, fs_info->fs_devices->latest_bdev);
2339 bio->bi_iter.bi_size = 0;
2340 bio->bi_private = data;
2342 btrfs_failed_bio = btrfs_io_bio(failed_bio);
2343 if (btrfs_failed_bio->csum) {
2344 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
2346 btrfs_bio = btrfs_io_bio(bio);
2347 btrfs_bio->csum = btrfs_bio->csum_inline;
2349 memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + icsum,
2353 bio_add_page(bio, page, failrec->len, pg_offset);
2359 * this is a generic handler for readpage errors (default
2360 * readpage_io_failed_hook). if other copies exist, read those and write back
2361 * good data to the failed position. does not investigate in remapping the
2362 * failed extent elsewhere, hoping the device will be smart enough to do this as
2366 static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset,
2367 struct page *page, u64 start, u64 end,
2370 struct io_failure_record *failrec;
2371 struct inode *inode = page->mapping->host;
2372 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2373 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2376 blk_status_t status;
2378 unsigned failed_bio_pages = bio_pages_all(failed_bio);
2380 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
2382 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
2386 if (!btrfs_check_repairable(inode, failed_bio_pages, failrec,
2388 free_io_failure(failure_tree, tree, failrec);
2392 if (failed_bio_pages > 1)
2393 read_mode |= REQ_FAILFAST_DEV;
2395 phy_offset >>= inode->i_sb->s_blocksize_bits;
2396 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
2397 start - page_offset(page),
2398 (int)phy_offset, failed_bio->bi_end_io,
2400 bio_set_op_attrs(bio, REQ_OP_READ, read_mode);
2402 btrfs_debug(btrfs_sb(inode->i_sb),
2403 "Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d",
2404 read_mode, failrec->this_mirror, failrec->in_validation);
2406 status = tree->ops->submit_bio_hook(tree->private_data, bio, failrec->this_mirror,
2407 failrec->bio_flags, 0);
2409 free_io_failure(failure_tree, tree, failrec);
2411 ret = blk_status_to_errno(status);
2417 /* lots and lots of room for performance fixes in the end_bio funcs */
2419 void end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2421 int uptodate = (err == 0);
2422 struct extent_io_tree *tree;
2425 tree = &BTRFS_I(page->mapping->host)->io_tree;
2427 if (tree->ops && tree->ops->writepage_end_io_hook)
2428 tree->ops->writepage_end_io_hook(page, start, end, NULL,
2432 ClearPageUptodate(page);
2434 ret = err < 0 ? err : -EIO;
2435 mapping_set_error(page->mapping, ret);
2440 * after a writepage IO is done, we need to:
2441 * clear the uptodate bits on error
2442 * clear the writeback bits in the extent tree for this IO
2443 * end_page_writeback if the page has no more pending IO
2445 * Scheduling is not allowed, so the extent state tree is expected
2446 * to have one and only one object corresponding to this IO.
2448 static void end_bio_extent_writepage(struct bio *bio)
2450 int error = blk_status_to_errno(bio->bi_status);
2451 struct bio_vec *bvec;
2456 ASSERT(!bio_flagged(bio, BIO_CLONED));
2457 bio_for_each_segment_all(bvec, bio, i) {
2458 struct page *page = bvec->bv_page;
2459 struct inode *inode = page->mapping->host;
2460 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2462 /* We always issue full-page reads, but if some block
2463 * in a page fails to read, blk_update_request() will
2464 * advance bv_offset and adjust bv_len to compensate.
2465 * Print a warning for nonzero offsets, and an error
2466 * if they don't add up to a full page. */
2467 if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) {
2468 if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE)
2470 "partial page write in btrfs with offset %u and length %u",
2471 bvec->bv_offset, bvec->bv_len);
2474 "incomplete page write in btrfs with offset %u and length %u",
2475 bvec->bv_offset, bvec->bv_len);
2478 start = page_offset(page);
2479 end = start + bvec->bv_offset + bvec->bv_len - 1;
2481 end_extent_writepage(page, error, start, end);
2482 end_page_writeback(page);
2489 endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len,
2492 struct extent_state *cached = NULL;
2493 u64 end = start + len - 1;
2495 if (uptodate && tree->track_uptodate)
2496 set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC);
2497 unlock_extent_cached_atomic(tree, start, end, &cached);
2501 * after a readpage IO is done, we need to:
2502 * clear the uptodate bits on error
2503 * set the uptodate bits if things worked
2504 * set the page up to date if all extents in the tree are uptodate
2505 * clear the lock bit in the extent tree
2506 * unlock the page if there are no other extents locked for it
2508 * Scheduling is not allowed, so the extent state tree is expected
2509 * to have one and only one object corresponding to this IO.
2511 static void end_bio_extent_readpage(struct bio *bio)
2513 struct bio_vec *bvec;
2514 int uptodate = !bio->bi_status;
2515 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
2516 struct extent_io_tree *tree, *failure_tree;
2521 u64 extent_start = 0;
2527 ASSERT(!bio_flagged(bio, BIO_CLONED));
2528 bio_for_each_segment_all(bvec, bio, i) {
2529 struct page *page = bvec->bv_page;
2530 struct inode *inode = page->mapping->host;
2531 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2533 btrfs_debug(fs_info,
2534 "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u",
2535 (u64)bio->bi_iter.bi_sector, bio->bi_status,
2536 io_bio->mirror_num);
2537 tree = &BTRFS_I(inode)->io_tree;
2538 failure_tree = &BTRFS_I(inode)->io_failure_tree;
2540 /* We always issue full-page reads, but if some block
2541 * in a page fails to read, blk_update_request() will
2542 * advance bv_offset and adjust bv_len to compensate.
2543 * Print a warning for nonzero offsets, and an error
2544 * if they don't add up to a full page. */
2545 if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) {
2546 if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE)
2548 "partial page read in btrfs with offset %u and length %u",
2549 bvec->bv_offset, bvec->bv_len);
2552 "incomplete page read in btrfs with offset %u and length %u",
2553 bvec->bv_offset, bvec->bv_len);
2556 start = page_offset(page);
2557 end = start + bvec->bv_offset + bvec->bv_len - 1;
2560 mirror = io_bio->mirror_num;
2561 if (likely(uptodate && tree->ops)) {
2562 ret = tree->ops->readpage_end_io_hook(io_bio, offset,
2568 clean_io_failure(BTRFS_I(inode)->root->fs_info,
2569 failure_tree, tree, start,
2571 btrfs_ino(BTRFS_I(inode)), 0);
2574 if (likely(uptodate))
2578 ret = tree->ops->readpage_io_failed_hook(page, mirror);
2579 if (ret == -EAGAIN) {
2581 * Data inode's readpage_io_failed_hook() always
2584 * The generic bio_readpage_error handles errors
2585 * the following way: If possible, new read
2586 * requests are created and submitted and will
2587 * end up in end_bio_extent_readpage as well (if
2588 * we're lucky, not in the !uptodate case). In
2589 * that case it returns 0 and we just go on with
2590 * the next page in our bio. If it can't handle
2591 * the error it will return -EIO and we remain
2592 * responsible for that page.
2594 ret = bio_readpage_error(bio, offset, page,
2595 start, end, mirror);
2597 uptodate = !bio->bi_status;
2604 * metadata's readpage_io_failed_hook() always returns
2605 * -EIO and fixes nothing. -EIO is also returned if
2606 * data inode error could not be fixed.
2608 ASSERT(ret == -EIO);
2611 if (likely(uptodate)) {
2612 loff_t i_size = i_size_read(inode);
2613 pgoff_t end_index = i_size >> PAGE_SHIFT;
2616 /* Zero out the end if this page straddles i_size */
2617 off = i_size & (PAGE_SIZE-1);
2618 if (page->index == end_index && off)
2619 zero_user_segment(page, off, PAGE_SIZE);
2620 SetPageUptodate(page);
2622 ClearPageUptodate(page);
2628 if (unlikely(!uptodate)) {
2630 endio_readpage_release_extent(tree,
2636 endio_readpage_release_extent(tree, start,
2637 end - start + 1, 0);
2638 } else if (!extent_len) {
2639 extent_start = start;
2640 extent_len = end + 1 - start;
2641 } else if (extent_start + extent_len == start) {
2642 extent_len += end + 1 - start;
2644 endio_readpage_release_extent(tree, extent_start,
2645 extent_len, uptodate);
2646 extent_start = start;
2647 extent_len = end + 1 - start;
2652 endio_readpage_release_extent(tree, extent_start, extent_len,
2655 io_bio->end_io(io_bio, blk_status_to_errno(bio->bi_status));
2660 * Initialize the members up to but not including 'bio'. Use after allocating a
2661 * new bio by bio_alloc_bioset as it does not initialize the bytes outside of
2662 * 'bio' because use of __GFP_ZERO is not supported.
2664 static inline void btrfs_io_bio_init(struct btrfs_io_bio *btrfs_bio)
2666 memset(btrfs_bio, 0, offsetof(struct btrfs_io_bio, bio));
2670 * The following helpers allocate a bio. As it's backed by a bioset, it'll
2671 * never fail. We're returning a bio right now but you can call btrfs_io_bio
2672 * for the appropriate container_of magic
2674 struct bio *btrfs_bio_alloc(struct block_device *bdev, u64 first_byte)
2678 bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, btrfs_bioset);
2679 bio_set_dev(bio, bdev);
2680 bio->bi_iter.bi_sector = first_byte >> 9;
2681 btrfs_io_bio_init(btrfs_io_bio(bio));
2685 struct bio *btrfs_bio_clone(struct bio *bio)
2687 struct btrfs_io_bio *btrfs_bio;
2690 /* Bio allocation backed by a bioset does not fail */
2691 new = bio_clone_fast(bio, GFP_NOFS, btrfs_bioset);
2692 btrfs_bio = btrfs_io_bio(new);
2693 btrfs_io_bio_init(btrfs_bio);
2694 btrfs_bio->iter = bio->bi_iter;
2698 struct bio *btrfs_io_bio_alloc(unsigned int nr_iovecs)
2702 /* Bio allocation backed by a bioset does not fail */
2703 bio = bio_alloc_bioset(GFP_NOFS, nr_iovecs, btrfs_bioset);
2704 btrfs_io_bio_init(btrfs_io_bio(bio));
2708 struct bio *btrfs_bio_clone_partial(struct bio *orig, int offset, int size)
2711 struct btrfs_io_bio *btrfs_bio;
2713 /* this will never fail when it's backed by a bioset */
2714 bio = bio_clone_fast(orig, GFP_NOFS, btrfs_bioset);
2717 btrfs_bio = btrfs_io_bio(bio);
2718 btrfs_io_bio_init(btrfs_bio);
2720 bio_trim(bio, offset >> 9, size >> 9);
2721 btrfs_bio->iter = bio->bi_iter;
2725 static int __must_check submit_one_bio(struct bio *bio, int mirror_num,
2726 unsigned long bio_flags)
2728 blk_status_t ret = 0;
2729 struct bio_vec *bvec = bio_last_bvec_all(bio);
2730 struct page *page = bvec->bv_page;
2731 struct extent_io_tree *tree = bio->bi_private;
2734 start = page_offset(page) + bvec->bv_offset;
2736 bio->bi_private = NULL;
2739 ret = tree->ops->submit_bio_hook(tree->private_data, bio,
2740 mirror_num, bio_flags, start);
2742 btrfsic_submit_bio(bio);
2744 return blk_status_to_errno(ret);
2747 static int merge_bio(struct extent_io_tree *tree, struct page *page,
2748 unsigned long offset, size_t size, struct bio *bio,
2749 unsigned long bio_flags)
2753 ret = tree->ops->merge_bio_hook(page, offset, size, bio,
2760 * @opf: bio REQ_OP_* and REQ_* flags as one value
2762 static int submit_extent_page(unsigned int opf, struct extent_io_tree *tree,
2763 struct writeback_control *wbc,
2764 struct page *page, u64 offset,
2765 size_t size, unsigned long pg_offset,
2766 struct block_device *bdev,
2767 struct bio **bio_ret,
2768 bio_end_io_t end_io_func,
2770 unsigned long prev_bio_flags,
2771 unsigned long bio_flags,
2772 bool force_bio_submit)
2777 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
2778 size_t page_size = min_t(size_t, size, PAGE_SIZE);
2779 sector_t sector = offset >> 9;
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, pg_offset, page_size, bio, bio_flags) ||
2791 bio_add_page(bio, page, page_size, pg_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, offset);
2806 bio_add_page(bio, page, page_size, pg_offset);
2807 bio->bi_end_io = end_io_func;
2808 bio->bi_private = tree;
2809 bio->bi_write_hint = page->mapping->host->i_write_hint;
2812 wbc_init_bio(wbc, bio);
2813 wbc_account_io(wbc, page, page_size);
2819 ret = submit_one_bio(bio, mirror_num, bio_flags);
2824 static void attach_extent_buffer_page(struct extent_buffer *eb,
2827 if (!PagePrivate(page)) {
2828 SetPagePrivate(page);
2830 set_page_private(page, (unsigned long)eb);
2832 WARN_ON(page->private != (unsigned long)eb);
2836 void set_page_extent_mapped(struct page *page)
2838 if (!PagePrivate(page)) {
2839 SetPagePrivate(page);
2841 set_page_private(page, EXTENT_PAGE_PRIVATE);
2845 static struct extent_map *
2846 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
2847 u64 start, u64 len, get_extent_t *get_extent,
2848 struct extent_map **em_cached)
2850 struct extent_map *em;
2852 if (em_cached && *em_cached) {
2854 if (extent_map_in_tree(em) && start >= em->start &&
2855 start < extent_map_end(em)) {
2856 refcount_inc(&em->refs);
2860 free_extent_map(em);
2864 em = get_extent(BTRFS_I(inode), page, pg_offset, start, len, 0);
2865 if (em_cached && !IS_ERR_OR_NULL(em)) {
2867 refcount_inc(&em->refs);
2873 * basic readpage implementation. Locked extent state structs are inserted
2874 * into the tree that are removed when the IO is done (by the end_io
2876 * XXX JDM: This needs looking at to ensure proper page locking
2877 * return 0 on success, otherwise return error
2879 static int __do_readpage(struct extent_io_tree *tree,
2881 get_extent_t *get_extent,
2882 struct extent_map **em_cached,
2883 struct bio **bio, int mirror_num,
2884 unsigned long *bio_flags, unsigned int read_flags,
2887 struct inode *inode = page->mapping->host;
2888 u64 start = page_offset(page);
2889 u64 page_end = start + PAGE_SIZE - 1;
2893 u64 last_byte = i_size_read(inode);
2896 struct extent_map *em;
2897 struct block_device *bdev;
2900 size_t pg_offset = 0;
2902 size_t disk_io_size;
2903 size_t blocksize = inode->i_sb->s_blocksize;
2904 unsigned long this_bio_flag = 0;
2906 set_page_extent_mapped(page);
2909 if (!PageUptodate(page)) {
2910 if (cleancache_get_page(page) == 0) {
2911 BUG_ON(blocksize != PAGE_SIZE);
2912 unlock_extent(tree, start, end);
2917 if (page->index == last_byte >> PAGE_SHIFT) {
2919 size_t zero_offset = last_byte & (PAGE_SIZE - 1);
2922 iosize = PAGE_SIZE - zero_offset;
2923 userpage = kmap_atomic(page);
2924 memset(userpage + zero_offset, 0, iosize);
2925 flush_dcache_page(page);
2926 kunmap_atomic(userpage);
2929 while (cur <= end) {
2930 bool force_bio_submit = false;
2933 if (cur >= last_byte) {
2935 struct extent_state *cached = NULL;
2937 iosize = PAGE_SIZE - pg_offset;
2938 userpage = kmap_atomic(page);
2939 memset(userpage + pg_offset, 0, iosize);
2940 flush_dcache_page(page);
2941 kunmap_atomic(userpage);
2942 set_extent_uptodate(tree, cur, cur + iosize - 1,
2944 unlock_extent_cached(tree, cur,
2945 cur + iosize - 1, &cached);
2948 em = __get_extent_map(inode, page, pg_offset, cur,
2949 end - cur + 1, get_extent, em_cached);
2950 if (IS_ERR_OR_NULL(em)) {
2952 unlock_extent(tree, cur, end);
2955 extent_offset = cur - em->start;
2956 BUG_ON(extent_map_end(em) <= cur);
2959 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2960 this_bio_flag |= EXTENT_BIO_COMPRESSED;
2961 extent_set_compress_type(&this_bio_flag,
2965 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2966 cur_end = min(extent_map_end(em) - 1, end);
2967 iosize = ALIGN(iosize, blocksize);
2968 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2969 disk_io_size = em->block_len;
2970 offset = em->block_start;
2972 offset = em->block_start + extent_offset;
2973 disk_io_size = iosize;
2976 block_start = em->block_start;
2977 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2978 block_start = EXTENT_MAP_HOLE;
2981 * If we have a file range that points to a compressed extent
2982 * and it's followed by a consecutive file range that points to
2983 * to the same compressed extent (possibly with a different
2984 * offset and/or length, so it either points to the whole extent
2985 * or only part of it), we must make sure we do not submit a
2986 * single bio to populate the pages for the 2 ranges because
2987 * this makes the compressed extent read zero out the pages
2988 * belonging to the 2nd range. Imagine the following scenario:
2991 * [0 - 8K] [8K - 24K]
2994 * points to extent X, points to extent X,
2995 * offset 4K, length of 8K offset 0, length 16K
2997 * [extent X, compressed length = 4K uncompressed length = 16K]
2999 * If the bio to read the compressed extent covers both ranges,
3000 * it will decompress extent X into the pages belonging to the
3001 * first range and then it will stop, zeroing out the remaining
3002 * pages that belong to the other range that points to extent X.
3003 * So here we make sure we submit 2 bios, one for the first
3004 * range and another one for the third range. Both will target
3005 * the same physical extent from disk, but we can't currently
3006 * make the compressed bio endio callback populate the pages
3007 * for both ranges because each compressed bio is tightly
3008 * coupled with a single extent map, and each range can have
3009 * an extent map with a different offset value relative to the
3010 * uncompressed data of our extent and different lengths. This
3011 * is a corner case so we prioritize correctness over
3012 * non-optimal behavior (submitting 2 bios for the same extent).
3014 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) &&
3015 prev_em_start && *prev_em_start != (u64)-1 &&
3016 *prev_em_start != em->orig_start)
3017 force_bio_submit = true;
3020 *prev_em_start = em->orig_start;
3022 free_extent_map(em);
3025 /* we've found a hole, just zero and go on */
3026 if (block_start == EXTENT_MAP_HOLE) {
3028 struct extent_state *cached = NULL;
3030 userpage = kmap_atomic(page);
3031 memset(userpage + pg_offset, 0, iosize);
3032 flush_dcache_page(page);
3033 kunmap_atomic(userpage);
3035 set_extent_uptodate(tree, cur, cur + iosize - 1,
3037 unlock_extent_cached(tree, cur,
3038 cur + iosize - 1, &cached);
3040 pg_offset += iosize;
3043 /* the get_extent function already copied into the page */
3044 if (test_range_bit(tree, cur, cur_end,
3045 EXTENT_UPTODATE, 1, NULL)) {
3046 check_page_uptodate(tree, page);
3047 unlock_extent(tree, cur, cur + iosize - 1);
3049 pg_offset += iosize;
3052 /* we have an inline extent but it didn't get marked up
3053 * to date. Error out
3055 if (block_start == EXTENT_MAP_INLINE) {
3057 unlock_extent(tree, cur, cur + iosize - 1);
3059 pg_offset += iosize;
3063 ret = submit_extent_page(REQ_OP_READ | read_flags, tree, NULL,
3064 page, offset, disk_io_size,
3065 pg_offset, bdev, bio,
3066 end_bio_extent_readpage, mirror_num,
3072 *bio_flags = this_bio_flag;
3075 unlock_extent(tree, cur, cur + iosize - 1);
3079 pg_offset += iosize;
3083 if (!PageError(page))
3084 SetPageUptodate(page);
3090 static inline void __do_contiguous_readpages(struct extent_io_tree *tree,
3091 struct page *pages[], int nr_pages,
3093 struct extent_map **em_cached,
3095 unsigned long *bio_flags,
3098 struct inode *inode;
3099 struct btrfs_ordered_extent *ordered;
3102 inode = pages[0]->mapping->host;
3104 lock_extent(tree, start, end);
3105 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
3109 unlock_extent(tree, start, end);
3110 btrfs_start_ordered_extent(inode, ordered, 1);
3111 btrfs_put_ordered_extent(ordered);
3114 for (index = 0; index < nr_pages; index++) {
3115 __do_readpage(tree, pages[index], btrfs_get_extent, em_cached,
3116 bio, 0, bio_flags, 0, prev_em_start);
3117 put_page(pages[index]);
3121 static void __extent_readpages(struct extent_io_tree *tree,
3122 struct page *pages[],
3124 struct extent_map **em_cached,
3125 struct bio **bio, unsigned long *bio_flags,
3132 int first_index = 0;
3134 for (index = 0; index < nr_pages; index++) {
3135 page_start = page_offset(pages[index]);
3138 end = start + PAGE_SIZE - 1;
3139 first_index = index;
3140 } else if (end + 1 == page_start) {
3143 __do_contiguous_readpages(tree, &pages[first_index],
3144 index - first_index, start,
3149 end = start + PAGE_SIZE - 1;
3150 first_index = index;
3155 __do_contiguous_readpages(tree, &pages[first_index],
3156 index - first_index, start,
3157 end, em_cached, bio,
3158 bio_flags, prev_em_start);
3161 static int __extent_read_full_page(struct extent_io_tree *tree,
3163 get_extent_t *get_extent,
3164 struct bio **bio, int mirror_num,
3165 unsigned long *bio_flags,
3166 unsigned int read_flags)
3168 struct inode *inode = page->mapping->host;
3169 struct btrfs_ordered_extent *ordered;
3170 u64 start = page_offset(page);
3171 u64 end = start + PAGE_SIZE - 1;
3175 lock_extent(tree, start, end);
3176 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
3180 unlock_extent(tree, start, end);
3181 btrfs_start_ordered_extent(inode, ordered, 1);
3182 btrfs_put_ordered_extent(ordered);
3185 ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num,
3186 bio_flags, read_flags, NULL);
3190 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
3191 get_extent_t *get_extent, int mirror_num)
3193 struct bio *bio = NULL;
3194 unsigned long bio_flags = 0;
3197 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
3200 ret = submit_one_bio(bio, mirror_num, bio_flags);
3204 static void update_nr_written(struct writeback_control *wbc,
3205 unsigned long nr_written)
3207 wbc->nr_to_write -= nr_written;
3211 * helper for __extent_writepage, doing all of the delayed allocation setup.
3213 * This returns 1 if our fill_delalloc function did all the work required
3214 * to write the page (copy into inline extent). In this case the IO has
3215 * been started and the page is already unlocked.
3217 * This returns 0 if all went well (page still locked)
3218 * This returns < 0 if there were errors (page still locked)
3220 static noinline_for_stack int writepage_delalloc(struct inode *inode,
3221 struct page *page, struct writeback_control *wbc,
3222 struct extent_page_data *epd,
3224 unsigned long *nr_written)
3226 struct extent_io_tree *tree = epd->tree;
3227 u64 page_end = delalloc_start + PAGE_SIZE - 1;
3229 u64 delalloc_to_write = 0;
3230 u64 delalloc_end = 0;
3232 int page_started = 0;
3234 if (epd->extent_locked || !tree->ops || !tree->ops->fill_delalloc)
3237 while (delalloc_end < page_end) {
3238 nr_delalloc = find_lock_delalloc_range(inode, tree,
3242 BTRFS_MAX_EXTENT_SIZE);
3243 if (nr_delalloc == 0) {
3244 delalloc_start = delalloc_end + 1;
3247 ret = tree->ops->fill_delalloc(inode, page,
3252 /* File system has been set read-only */
3255 /* fill_delalloc should be return < 0 for error
3256 * but just in case, we use > 0 here meaning the
3257 * IO is started, so we don't want to return > 0
3258 * unless things are going well.
3260 ret = ret < 0 ? ret : -EIO;
3264 * delalloc_end is already one less than the total length, so
3265 * we don't subtract one from PAGE_SIZE
3267 delalloc_to_write += (delalloc_end - delalloc_start +
3268 PAGE_SIZE) >> PAGE_SHIFT;
3269 delalloc_start = delalloc_end + 1;
3271 if (wbc->nr_to_write < delalloc_to_write) {
3274 if (delalloc_to_write < thresh * 2)
3275 thresh = delalloc_to_write;
3276 wbc->nr_to_write = min_t(u64, delalloc_to_write,
3280 /* did the fill delalloc function already unlock and start
3285 * we've unlocked the page, so we can't update
3286 * the mapping's writeback index, just update
3289 wbc->nr_to_write -= *nr_written;
3300 * helper for __extent_writepage. This calls the writepage start hooks,
3301 * and does the loop to map the page into extents and bios.
3303 * We return 1 if the IO is started and the page is unlocked,
3304 * 0 if all went well (page still locked)
3305 * < 0 if there were errors (page still locked)
3307 static noinline_for_stack int __extent_writepage_io(struct inode *inode,
3309 struct writeback_control *wbc,
3310 struct extent_page_data *epd,
3312 unsigned long nr_written,
3313 unsigned int write_flags, int *nr_ret)
3315 struct extent_io_tree *tree = epd->tree;
3316 u64 start = page_offset(page);
3317 u64 page_end = start + PAGE_SIZE - 1;
3323 struct extent_map *em;
3324 struct block_device *bdev;
3325 size_t pg_offset = 0;
3331 if (tree->ops && tree->ops->writepage_start_hook) {
3332 ret = tree->ops->writepage_start_hook(page, start,
3335 /* Fixup worker will requeue */
3337 wbc->pages_skipped++;
3339 redirty_page_for_writepage(wbc, page);
3341 update_nr_written(wbc, nr_written);
3348 * we don't want to touch the inode after unlocking the page,
3349 * so we update the mapping writeback index now
3351 update_nr_written(wbc, nr_written + 1);
3354 if (i_size <= start) {
3355 if (tree->ops && tree->ops->writepage_end_io_hook)
3356 tree->ops->writepage_end_io_hook(page, start,
3361 blocksize = inode->i_sb->s_blocksize;
3363 while (cur <= end) {
3367 if (cur >= i_size) {
3368 if (tree->ops && tree->ops->writepage_end_io_hook)
3369 tree->ops->writepage_end_io_hook(page, cur,
3373 em = btrfs_get_extent(BTRFS_I(inode), page, pg_offset, cur,
3375 if (IS_ERR_OR_NULL(em)) {
3377 ret = PTR_ERR_OR_ZERO(em);
3381 extent_offset = cur - em->start;
3382 em_end = extent_map_end(em);
3383 BUG_ON(em_end <= cur);
3385 iosize = min(em_end - cur, end - cur + 1);
3386 iosize = ALIGN(iosize, blocksize);
3387 offset = em->block_start + extent_offset;
3389 block_start = em->block_start;
3390 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3391 free_extent_map(em);
3395 * compressed and inline extents are written through other
3398 if (compressed || block_start == EXTENT_MAP_HOLE ||
3399 block_start == EXTENT_MAP_INLINE) {
3401 * end_io notification does not happen here for
3402 * compressed extents
3404 if (!compressed && tree->ops &&
3405 tree->ops->writepage_end_io_hook)
3406 tree->ops->writepage_end_io_hook(page, cur,
3409 else if (compressed) {
3410 /* we don't want to end_page_writeback on
3411 * a compressed extent. this happens
3418 pg_offset += iosize;
3422 set_range_writeback(tree, cur, cur + iosize - 1);
3423 if (!PageWriteback(page)) {
3424 btrfs_err(BTRFS_I(inode)->root->fs_info,
3425 "page %lu not writeback, cur %llu end %llu",
3426 page->index, cur, end);
3429 ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc,
3430 page, offset, iosize, pg_offset,
3432 end_bio_extent_writepage,
3436 if (PageWriteback(page))
3437 end_page_writeback(page);
3441 pg_offset += iosize;
3450 * the writepage semantics are similar to regular writepage. extent
3451 * records are inserted to lock ranges in the tree, and as dirty areas
3452 * are found, they are marked writeback. Then the lock bits are removed
3453 * and the end_io handler clears the writeback ranges
3455 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
3456 struct extent_page_data *epd)
3458 struct inode *inode = page->mapping->host;
3459 u64 start = page_offset(page);
3460 u64 page_end = start + PAGE_SIZE - 1;
3463 size_t pg_offset = 0;
3464 loff_t i_size = i_size_read(inode);
3465 unsigned long end_index = i_size >> PAGE_SHIFT;
3466 unsigned int write_flags = 0;
3467 unsigned long nr_written = 0;
3469 write_flags = wbc_to_write_flags(wbc);
3471 trace___extent_writepage(page, inode, wbc);
3473 WARN_ON(!PageLocked(page));
3475 ClearPageError(page);
3477 pg_offset = i_size & (PAGE_SIZE - 1);
3478 if (page->index > end_index ||
3479 (page->index == end_index && !pg_offset)) {
3480 page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
3485 if (page->index == end_index) {
3488 userpage = kmap_atomic(page);
3489 memset(userpage + pg_offset, 0,
3490 PAGE_SIZE - pg_offset);
3491 kunmap_atomic(userpage);
3492 flush_dcache_page(page);
3497 set_page_extent_mapped(page);
3499 ret = writepage_delalloc(inode, page, wbc, epd, start, &nr_written);
3505 ret = __extent_writepage_io(inode, page, wbc, epd,
3506 i_size, nr_written, write_flags, &nr);
3512 /* make sure the mapping tag for page dirty gets cleared */
3513 set_page_writeback(page);
3514 end_page_writeback(page);
3516 if (PageError(page)) {
3517 ret = ret < 0 ? ret : -EIO;
3518 end_extent_writepage(page, ret, start, page_end);
3527 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3529 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
3530 TASK_UNINTERRUPTIBLE);
3533 static noinline_for_stack int
3534 lock_extent_buffer_for_io(struct extent_buffer *eb,
3535 struct btrfs_fs_info *fs_info,
3536 struct extent_page_data *epd)
3538 unsigned long i, num_pages;
3542 if (!btrfs_try_tree_write_lock(eb)) {
3544 flush_write_bio(epd);
3545 btrfs_tree_lock(eb);
3548 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3549 btrfs_tree_unlock(eb);
3553 flush_write_bio(epd);
3557 wait_on_extent_buffer_writeback(eb);
3558 btrfs_tree_lock(eb);
3559 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3561 btrfs_tree_unlock(eb);
3566 * We need to do this to prevent races in people who check if the eb is
3567 * under IO since we can end up having no IO bits set for a short period
3570 spin_lock(&eb->refs_lock);
3571 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3572 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3573 spin_unlock(&eb->refs_lock);
3574 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3575 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
3577 fs_info->dirty_metadata_batch);
3580 spin_unlock(&eb->refs_lock);
3583 btrfs_tree_unlock(eb);
3588 num_pages = num_extent_pages(eb->start, eb->len);
3589 for (i = 0; i < num_pages; i++) {
3590 struct page *p = eb->pages[i];
3592 if (!trylock_page(p)) {
3594 flush_write_bio(epd);
3604 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3606 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3607 smp_mb__after_atomic();
3608 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3611 static void set_btree_ioerr(struct page *page)
3613 struct extent_buffer *eb = (struct extent_buffer *)page->private;
3616 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))
3620 * If writeback for a btree extent that doesn't belong to a log tree
3621 * failed, increment the counter transaction->eb_write_errors.
3622 * We do this because while the transaction is running and before it's
3623 * committing (when we call filemap_fdata[write|wait]_range against
3624 * the btree inode), we might have
3625 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
3626 * returns an error or an error happens during writeback, when we're
3627 * committing the transaction we wouldn't know about it, since the pages
3628 * can be no longer dirty nor marked anymore for writeback (if a
3629 * subsequent modification to the extent buffer didn't happen before the
3630 * transaction commit), which makes filemap_fdata[write|wait]_range not
3631 * able to find the pages tagged with SetPageError at transaction
3632 * commit time. So if this happens we must abort the transaction,
3633 * otherwise we commit a super block with btree roots that point to
3634 * btree nodes/leafs whose content on disk is invalid - either garbage
3635 * or the content of some node/leaf from a past generation that got
3636 * cowed or deleted and is no longer valid.
3638 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
3639 * not be enough - we need to distinguish between log tree extents vs
3640 * non-log tree extents, and the next filemap_fdatawait_range() call
3641 * will catch and clear such errors in the mapping - and that call might
3642 * be from a log sync and not from a transaction commit. Also, checking
3643 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
3644 * not done and would not be reliable - the eb might have been released
3645 * from memory and reading it back again means that flag would not be
3646 * set (since it's a runtime flag, not persisted on disk).
3648 * Using the flags below in the btree inode also makes us achieve the
3649 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
3650 * writeback for all dirty pages and before filemap_fdatawait_range()
3651 * is called, the writeback for all dirty pages had already finished
3652 * with errors - because we were not using AS_EIO/AS_ENOSPC,
3653 * filemap_fdatawait_range() would return success, as it could not know
3654 * that writeback errors happened (the pages were no longer tagged for
3657 switch (eb->log_index) {
3659 set_bit(BTRFS_FS_BTREE_ERR, &eb->fs_info->flags);
3662 set_bit(BTRFS_FS_LOG1_ERR, &eb->fs_info->flags);
3665 set_bit(BTRFS_FS_LOG2_ERR, &eb->fs_info->flags);
3668 BUG(); /* unexpected, logic error */
3672 static void end_bio_extent_buffer_writepage(struct bio *bio)
3674 struct bio_vec *bvec;
3675 struct extent_buffer *eb;
3678 ASSERT(!bio_flagged(bio, BIO_CLONED));
3679 bio_for_each_segment_all(bvec, bio, i) {
3680 struct page *page = bvec->bv_page;
3682 eb = (struct extent_buffer *)page->private;
3684 done = atomic_dec_and_test(&eb->io_pages);
3686 if (bio->bi_status ||
3687 test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) {
3688 ClearPageUptodate(page);
3689 set_btree_ioerr(page);
3692 end_page_writeback(page);
3697 end_extent_buffer_writeback(eb);
3703 static noinline_for_stack int write_one_eb(struct extent_buffer *eb,
3704 struct btrfs_fs_info *fs_info,
3705 struct writeback_control *wbc,
3706 struct extent_page_data *epd)
3708 struct block_device *bdev = fs_info->fs_devices->latest_bdev;
3709 struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
3710 u64 offset = eb->start;
3712 unsigned long i, num_pages;
3713 unsigned long start, end;
3714 unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META;
3717 clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
3718 num_pages = num_extent_pages(eb->start, eb->len);
3719 atomic_set(&eb->io_pages, num_pages);
3721 /* set btree blocks beyond nritems with 0 to avoid stale content. */
3722 nritems = btrfs_header_nritems(eb);
3723 if (btrfs_header_level(eb) > 0) {
3724 end = btrfs_node_key_ptr_offset(nritems);
3726 memzero_extent_buffer(eb, end, eb->len - end);
3730 * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0
3732 start = btrfs_item_nr_offset(nritems);
3733 end = BTRFS_LEAF_DATA_OFFSET + leaf_data_end(fs_info, eb);
3734 memzero_extent_buffer(eb, start, end - start);
3737 for (i = 0; i < num_pages; i++) {
3738 struct page *p = eb->pages[i];
3740 clear_page_dirty_for_io(p);
3741 set_page_writeback(p);
3742 ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc,
3743 p, offset, PAGE_SIZE, 0, bdev,
3745 end_bio_extent_buffer_writepage,
3749 if (PageWriteback(p))
3750 end_page_writeback(p);
3751 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3752 end_extent_buffer_writeback(eb);
3756 offset += PAGE_SIZE;
3757 update_nr_written(wbc, 1);
3761 if (unlikely(ret)) {
3762 for (; i < num_pages; i++) {
3763 struct page *p = eb->pages[i];
3764 clear_page_dirty_for_io(p);
3772 int btree_write_cache_pages(struct address_space *mapping,
3773 struct writeback_control *wbc)
3775 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
3776 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
3777 struct extent_buffer *eb, *prev_eb = NULL;
3778 struct extent_page_data epd = {
3782 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3786 int nr_to_write_done = 0;
3787 struct pagevec pvec;
3790 pgoff_t end; /* Inclusive */
3794 pagevec_init(&pvec);
3795 if (wbc->range_cyclic) {
3796 index = mapping->writeback_index; /* Start from prev offset */
3799 index = wbc->range_start >> PAGE_SHIFT;
3800 end = wbc->range_end >> PAGE_SHIFT;
3803 if (wbc->sync_mode == WB_SYNC_ALL)
3804 tag = PAGECACHE_TAG_TOWRITE;
3806 tag = PAGECACHE_TAG_DIRTY;
3808 if (wbc->sync_mode == WB_SYNC_ALL)
3809 tag_pages_for_writeback(mapping, index, end);
3810 while (!done && !nr_to_write_done && (index <= end) &&
3811 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
3816 for (i = 0; i < nr_pages; i++) {
3817 struct page *page = pvec.pages[i];
3819 if (!PagePrivate(page))
3822 spin_lock(&mapping->private_lock);
3823 if (!PagePrivate(page)) {
3824 spin_unlock(&mapping->private_lock);
3828 eb = (struct extent_buffer *)page->private;
3831 * Shouldn't happen and normally this would be a BUG_ON
3832 * but no sense in crashing the users box for something
3833 * we can survive anyway.
3836 spin_unlock(&mapping->private_lock);
3840 if (eb == prev_eb) {
3841 spin_unlock(&mapping->private_lock);
3845 ret = atomic_inc_not_zero(&eb->refs);
3846 spin_unlock(&mapping->private_lock);
3851 ret = lock_extent_buffer_for_io(eb, fs_info, &epd);
3853 free_extent_buffer(eb);
3857 ret = write_one_eb(eb, fs_info, wbc, &epd);
3860 free_extent_buffer(eb);
3863 free_extent_buffer(eb);
3866 * the filesystem may choose to bump up nr_to_write.
3867 * We have to make sure to honor the new nr_to_write
3870 nr_to_write_done = wbc->nr_to_write <= 0;
3872 pagevec_release(&pvec);
3875 if (!scanned && !done) {
3877 * We hit the last page and there is more work to be done: wrap
3878 * back to the start of the file
3884 flush_write_bio(&epd);
3889 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3890 * @mapping: address space structure to write
3891 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3892 * @data: data passed to __extent_writepage function
3894 * If a page is already under I/O, write_cache_pages() skips it, even
3895 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3896 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3897 * and msync() need to guarantee that all the data which was dirty at the time
3898 * the call was made get new I/O started against them. If wbc->sync_mode is
3899 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3900 * existing IO to complete.
3902 static int extent_write_cache_pages(struct address_space *mapping,
3903 struct writeback_control *wbc,
3904 struct extent_page_data *epd)
3906 struct inode *inode = mapping->host;
3909 int nr_to_write_done = 0;
3910 struct pagevec pvec;
3913 pgoff_t end; /* Inclusive */
3915 int range_whole = 0;
3920 * We have to hold onto the inode so that ordered extents can do their
3921 * work when the IO finishes. The alternative to this is failing to add
3922 * an ordered extent if the igrab() fails there and that is a huge pain
3923 * to deal with, so instead just hold onto the inode throughout the
3924 * writepages operation. If it fails here we are freeing up the inode
3925 * anyway and we'd rather not waste our time writing out stuff that is
3926 * going to be truncated anyway.
3931 pagevec_init(&pvec);
3932 if (wbc->range_cyclic) {
3933 index = mapping->writeback_index; /* Start from prev offset */
3936 index = wbc->range_start >> PAGE_SHIFT;
3937 end = wbc->range_end >> PAGE_SHIFT;
3938 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
3942 if (wbc->sync_mode == WB_SYNC_ALL)
3943 tag = PAGECACHE_TAG_TOWRITE;
3945 tag = PAGECACHE_TAG_DIRTY;
3947 if (wbc->sync_mode == WB_SYNC_ALL)
3948 tag_pages_for_writeback(mapping, index, end);
3950 while (!done && !nr_to_write_done && (index <= end) &&
3951 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping,
3952 &index, end, tag))) {
3956 for (i = 0; i < nr_pages; i++) {
3957 struct page *page = pvec.pages[i];
3959 done_index = page->index;
3961 * At this point we hold neither mapping->tree_lock nor
3962 * lock on the page itself: the page may be truncated or
3963 * invalidated (changing page->mapping to NULL), or even
3964 * swizzled back from swapper_space to tmpfs file
3967 if (!trylock_page(page)) {
3968 flush_write_bio(epd);
3972 if (unlikely(page->mapping != mapping)) {
3977 if (wbc->sync_mode != WB_SYNC_NONE) {
3978 if (PageWriteback(page))
3979 flush_write_bio(epd);
3980 wait_on_page_writeback(page);
3983 if (PageWriteback(page) ||
3984 !clear_page_dirty_for_io(page)) {
3989 ret = __extent_writepage(page, wbc, epd);
3991 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
3997 * done_index is set past this page,
3998 * so media errors will not choke
3999 * background writeout for the entire
4000 * file. This has consequences for
4001 * range_cyclic semantics (ie. it may
4002 * not be suitable for data integrity
4005 done_index = page->index + 1;
4011 * the filesystem may choose to bump up nr_to_write.
4012 * We have to make sure to honor the new nr_to_write
4015 nr_to_write_done = wbc->nr_to_write <= 0;
4017 pagevec_release(&pvec);
4020 if (!scanned && !done) {
4022 * We hit the last page and there is more work to be done: wrap
4023 * back to the start of the file
4030 if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole))
4031 mapping->writeback_index = done_index;
4033 btrfs_add_delayed_iput(inode);
4037 static void flush_write_bio(struct extent_page_data *epd)
4042 ret = submit_one_bio(epd->bio, 0, 0);
4043 BUG_ON(ret < 0); /* -ENOMEM */
4048 int extent_write_full_page(struct page *page, struct writeback_control *wbc)
4051 struct extent_page_data epd = {
4053 .tree = &BTRFS_I(page->mapping->host)->io_tree,
4055 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4058 ret = __extent_writepage(page, wbc, &epd);
4060 flush_write_bio(&epd);
4064 int extent_write_locked_range(struct inode *inode, u64 start, u64 end,
4068 struct address_space *mapping = inode->i_mapping;
4069 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
4071 unsigned long nr_pages = (end - start + PAGE_SIZE) >>
4074 struct extent_page_data epd = {
4078 .sync_io = mode == WB_SYNC_ALL,
4080 struct writeback_control wbc_writepages = {
4082 .nr_to_write = nr_pages * 2,
4083 .range_start = start,
4084 .range_end = end + 1,
4087 while (start <= end) {
4088 page = find_get_page(mapping, start >> PAGE_SHIFT);
4089 if (clear_page_dirty_for_io(page))
4090 ret = __extent_writepage(page, &wbc_writepages, &epd);
4092 if (tree->ops && tree->ops->writepage_end_io_hook)
4093 tree->ops->writepage_end_io_hook(page, start,
4094 start + PAGE_SIZE - 1,
4102 flush_write_bio(&epd);
4106 int extent_writepages(struct extent_io_tree *tree,
4107 struct address_space *mapping,
4108 struct writeback_control *wbc)
4111 struct extent_page_data epd = {
4115 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4118 ret = extent_write_cache_pages(mapping, wbc, &epd);
4119 flush_write_bio(&epd);
4123 int extent_readpages(struct extent_io_tree *tree,
4124 struct address_space *mapping,
4125 struct list_head *pages, unsigned nr_pages)
4127 struct bio *bio = NULL;
4129 unsigned long bio_flags = 0;
4130 struct page *pagepool[16];
4132 struct extent_map *em_cached = NULL;
4134 u64 prev_em_start = (u64)-1;
4136 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
4137 page = list_entry(pages->prev, struct page, lru);
4139 prefetchw(&page->flags);
4140 list_del(&page->lru);
4141 if (add_to_page_cache_lru(page, mapping,
4143 readahead_gfp_mask(mapping))) {
4148 pagepool[nr++] = page;
4149 if (nr < ARRAY_SIZE(pagepool))
4151 __extent_readpages(tree, pagepool, nr, &em_cached, &bio,
4152 &bio_flags, &prev_em_start);
4156 __extent_readpages(tree, pagepool, nr, &em_cached, &bio,
4157 &bio_flags, &prev_em_start);
4160 free_extent_map(em_cached);
4162 BUG_ON(!list_empty(pages));
4164 return submit_one_bio(bio, 0, bio_flags);
4169 * basic invalidatepage code, this waits on any locked or writeback
4170 * ranges corresponding to the page, and then deletes any extent state
4171 * records from the tree
4173 int extent_invalidatepage(struct extent_io_tree *tree,
4174 struct page *page, unsigned long offset)
4176 struct extent_state *cached_state = NULL;
4177 u64 start = page_offset(page);
4178 u64 end = start + PAGE_SIZE - 1;
4179 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
4181 start += ALIGN(offset, blocksize);
4185 lock_extent_bits(tree, start, end, &cached_state);
4186 wait_on_page_writeback(page);
4187 clear_extent_bit(tree, start, end,
4188 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4189 EXTENT_DO_ACCOUNTING,
4190 1, 1, &cached_state);
4195 * a helper for releasepage, this tests for areas of the page that
4196 * are locked or under IO and drops the related state bits if it is safe
4199 static int try_release_extent_state(struct extent_map_tree *map,
4200 struct extent_io_tree *tree,
4201 struct page *page, gfp_t mask)
4203 u64 start = page_offset(page);
4204 u64 end = start + PAGE_SIZE - 1;
4207 if (test_range_bit(tree, start, end,
4208 EXTENT_IOBITS, 0, NULL))
4212 * at this point we can safely clear everything except the
4213 * locked bit and the nodatasum bit
4215 ret = __clear_extent_bit(tree, start, end,
4216 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
4217 0, 0, NULL, mask, NULL);
4219 /* if clear_extent_bit failed for enomem reasons,
4220 * we can't allow the release to continue.
4231 * a helper for releasepage. As long as there are no locked extents
4232 * in the range corresponding to the page, both state records and extent
4233 * map records are removed
4235 int try_release_extent_mapping(struct extent_map_tree *map,
4236 struct extent_io_tree *tree, struct page *page,
4239 struct extent_map *em;
4240 u64 start = page_offset(page);
4241 u64 end = start + PAGE_SIZE - 1;
4243 if (gfpflags_allow_blocking(mask) &&
4244 page->mapping->host->i_size > SZ_16M) {
4246 while (start <= end) {
4247 len = end - start + 1;
4248 write_lock(&map->lock);
4249 em = lookup_extent_mapping(map, start, len);
4251 write_unlock(&map->lock);
4254 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
4255 em->start != start) {
4256 write_unlock(&map->lock);
4257 free_extent_map(em);
4260 if (!test_range_bit(tree, em->start,
4261 extent_map_end(em) - 1,
4262 EXTENT_LOCKED | EXTENT_WRITEBACK,
4264 remove_extent_mapping(map, em);
4265 /* once for the rb tree */
4266 free_extent_map(em);
4268 start = extent_map_end(em);
4269 write_unlock(&map->lock);
4272 free_extent_map(em);
4275 return try_release_extent_state(map, tree, page, mask);
4279 * helper function for fiemap, which doesn't want to see any holes.
4280 * This maps until we find something past 'last'
4282 static struct extent_map *get_extent_skip_holes(struct inode *inode,
4283 u64 offset, u64 last)
4285 u64 sectorsize = btrfs_inode_sectorsize(inode);
4286 struct extent_map *em;
4293 len = last - offset;
4296 len = ALIGN(len, sectorsize);
4297 em = btrfs_get_extent_fiemap(BTRFS_I(inode), NULL, 0, offset,
4299 if (IS_ERR_OR_NULL(em))
4302 /* if this isn't a hole return it */
4303 if (em->block_start != EXTENT_MAP_HOLE)
4306 /* this is a hole, advance to the next extent */
4307 offset = extent_map_end(em);
4308 free_extent_map(em);
4316 * To cache previous fiemap extent
4318 * Will be used for merging fiemap extent
4320 struct fiemap_cache {
4329 * Helper to submit fiemap extent.
4331 * Will try to merge current fiemap extent specified by @offset, @phys,
4332 * @len and @flags with cached one.
4333 * And only when we fails to merge, cached one will be submitted as
4336 * Return value is the same as fiemap_fill_next_extent().
4338 static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
4339 struct fiemap_cache *cache,
4340 u64 offset, u64 phys, u64 len, u32 flags)
4348 * Sanity check, extent_fiemap() should have ensured that new
4349 * fiemap extent won't overlap with cahced one.
4352 * NOTE: Physical address can overlap, due to compression
4354 if (cache->offset + cache->len > offset) {
4360 * Only merges fiemap extents if
4361 * 1) Their logical addresses are continuous
4363 * 2) Their physical addresses are continuous
4364 * So truly compressed (physical size smaller than logical size)
4365 * extents won't get merged with each other
4367 * 3) Share same flags except FIEMAP_EXTENT_LAST
4368 * So regular extent won't get merged with prealloc extent
4370 if (cache->offset + cache->len == offset &&
4371 cache->phys + cache->len == phys &&
4372 (cache->flags & ~FIEMAP_EXTENT_LAST) ==
4373 (flags & ~FIEMAP_EXTENT_LAST)) {
4375 cache->flags |= flags;
4376 goto try_submit_last;
4379 /* Not mergeable, need to submit cached one */
4380 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
4381 cache->len, cache->flags);
4382 cache->cached = false;
4386 cache->cached = true;
4387 cache->offset = offset;
4390 cache->flags = flags;
4392 if (cache->flags & FIEMAP_EXTENT_LAST) {
4393 ret = fiemap_fill_next_extent(fieinfo, cache->offset,
4394 cache->phys, cache->len, cache->flags);
4395 cache->cached = false;
4401 * Emit last fiemap cache
4403 * The last fiemap cache may still be cached in the following case:
4405 * |<- Fiemap range ->|
4406 * |<------------ First extent ----------->|
4408 * In this case, the first extent range will be cached but not emitted.
4409 * So we must emit it before ending extent_fiemap().
4411 static int emit_last_fiemap_cache(struct btrfs_fs_info *fs_info,
4412 struct fiemap_extent_info *fieinfo,
4413 struct fiemap_cache *cache)
4420 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
4421 cache->len, cache->flags);
4422 cache->cached = false;
4428 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4429 __u64 start, __u64 len)
4433 u64 max = start + len;
4437 u64 last_for_get_extent = 0;
4439 u64 isize = i_size_read(inode);
4440 struct btrfs_key found_key;
4441 struct extent_map *em = NULL;
4442 struct extent_state *cached_state = NULL;
4443 struct btrfs_path *path;
4444 struct btrfs_root *root = BTRFS_I(inode)->root;
4445 struct fiemap_cache cache = { 0 };
4454 path = btrfs_alloc_path();
4457 path->leave_spinning = 1;
4459 start = round_down(start, btrfs_inode_sectorsize(inode));
4460 len = round_up(max, btrfs_inode_sectorsize(inode)) - start;
4463 * lookup the last file extent. We're not using i_size here
4464 * because there might be preallocation past i_size
4466 ret = btrfs_lookup_file_extent(NULL, root, path,
4467 btrfs_ino(BTRFS_I(inode)), -1, 0);
4469 btrfs_free_path(path);
4478 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
4479 found_type = found_key.type;
4481 /* No extents, but there might be delalloc bits */
4482 if (found_key.objectid != btrfs_ino(BTRFS_I(inode)) ||
4483 found_type != BTRFS_EXTENT_DATA_KEY) {
4484 /* have to trust i_size as the end */
4486 last_for_get_extent = isize;
4489 * remember the start of the last extent. There are a
4490 * bunch of different factors that go into the length of the
4491 * extent, so its much less complex to remember where it started
4493 last = found_key.offset;
4494 last_for_get_extent = last + 1;
4496 btrfs_release_path(path);
4499 * we might have some extents allocated but more delalloc past those
4500 * extents. so, we trust isize unless the start of the last extent is
4505 last_for_get_extent = isize;
4508 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4511 em = get_extent_skip_holes(inode, start, last_for_get_extent);
4520 u64 offset_in_extent = 0;
4522 /* break if the extent we found is outside the range */
4523 if (em->start >= max || extent_map_end(em) < off)
4527 * get_extent may return an extent that starts before our
4528 * requested range. We have to make sure the ranges
4529 * we return to fiemap always move forward and don't
4530 * overlap, so adjust the offsets here
4532 em_start = max(em->start, off);
4535 * record the offset from the start of the extent
4536 * for adjusting the disk offset below. Only do this if the
4537 * extent isn't compressed since our in ram offset may be past
4538 * what we have actually allocated on disk.
4540 if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4541 offset_in_extent = em_start - em->start;
4542 em_end = extent_map_end(em);
4543 em_len = em_end - em_start;
4548 * bump off for our next call to get_extent
4550 off = extent_map_end(em);
4554 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
4556 flags |= FIEMAP_EXTENT_LAST;
4557 } else if (em->block_start == EXTENT_MAP_INLINE) {
4558 flags |= (FIEMAP_EXTENT_DATA_INLINE |
4559 FIEMAP_EXTENT_NOT_ALIGNED);
4560 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
4561 flags |= (FIEMAP_EXTENT_DELALLOC |
4562 FIEMAP_EXTENT_UNKNOWN);
4563 } else if (fieinfo->fi_extents_max) {
4564 u64 bytenr = em->block_start -
4565 (em->start - em->orig_start);
4567 disko = em->block_start + offset_in_extent;
4570 * As btrfs supports shared space, this information
4571 * can be exported to userspace tools via
4572 * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0
4573 * then we're just getting a count and we can skip the
4576 ret = btrfs_check_shared(root,
4577 btrfs_ino(BTRFS_I(inode)),
4582 flags |= FIEMAP_EXTENT_SHARED;
4585 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4586 flags |= FIEMAP_EXTENT_ENCODED;
4587 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4588 flags |= FIEMAP_EXTENT_UNWRITTEN;
4590 free_extent_map(em);
4592 if ((em_start >= last) || em_len == (u64)-1 ||
4593 (last == (u64)-1 && isize <= em_end)) {
4594 flags |= FIEMAP_EXTENT_LAST;
4598 /* now scan forward to see if this is really the last extent. */
4599 em = get_extent_skip_holes(inode, off, last_for_get_extent);
4605 flags |= FIEMAP_EXTENT_LAST;
4608 ret = emit_fiemap_extent(fieinfo, &cache, em_start, disko,
4618 ret = emit_last_fiemap_cache(root->fs_info, fieinfo, &cache);
4619 free_extent_map(em);
4621 btrfs_free_path(path);
4622 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4627 static void __free_extent_buffer(struct extent_buffer *eb)
4629 btrfs_leak_debug_del(&eb->leak_list);
4630 kmem_cache_free(extent_buffer_cache, eb);
4633 int extent_buffer_under_io(struct extent_buffer *eb)
4635 return (atomic_read(&eb->io_pages) ||
4636 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4637 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4641 * Helper for releasing extent buffer page.
4643 static void btrfs_release_extent_buffer_page(struct extent_buffer *eb)
4645 unsigned long index;
4647 int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4649 BUG_ON(extent_buffer_under_io(eb));
4651 index = num_extent_pages(eb->start, eb->len);
4657 page = eb->pages[index];
4661 spin_lock(&page->mapping->private_lock);
4663 * We do this since we'll remove the pages after we've
4664 * removed the eb from the radix tree, so we could race
4665 * and have this page now attached to the new eb. So
4666 * only clear page_private if it's still connected to
4669 if (PagePrivate(page) &&
4670 page->private == (unsigned long)eb) {
4671 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4672 BUG_ON(PageDirty(page));
4673 BUG_ON(PageWriteback(page));
4675 * We need to make sure we haven't be attached
4678 ClearPagePrivate(page);
4679 set_page_private(page, 0);
4680 /* One for the page private */
4685 spin_unlock(&page->mapping->private_lock);
4687 /* One for when we allocated the page */
4689 } while (index != 0);
4693 * Helper for releasing the extent buffer.
4695 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4697 btrfs_release_extent_buffer_page(eb);
4698 __free_extent_buffer(eb);
4701 static struct extent_buffer *
4702 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
4705 struct extent_buffer *eb = NULL;
4707 eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL);
4710 eb->fs_info = fs_info;
4712 rwlock_init(&eb->lock);
4713 atomic_set(&eb->write_locks, 0);
4714 atomic_set(&eb->read_locks, 0);
4715 atomic_set(&eb->blocking_readers, 0);
4716 atomic_set(&eb->blocking_writers, 0);
4717 atomic_set(&eb->spinning_readers, 0);
4718 atomic_set(&eb->spinning_writers, 0);
4719 eb->lock_nested = 0;
4720 init_waitqueue_head(&eb->write_lock_wq);
4721 init_waitqueue_head(&eb->read_lock_wq);
4723 btrfs_leak_debug_add(&eb->leak_list, &buffers);
4725 spin_lock_init(&eb->refs_lock);
4726 atomic_set(&eb->refs, 1);
4727 atomic_set(&eb->io_pages, 0);
4730 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4732 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4733 > MAX_INLINE_EXTENT_BUFFER_SIZE);
4734 BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE);
4739 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src)
4743 struct extent_buffer *new;
4744 unsigned long num_pages = num_extent_pages(src->start, src->len);
4746 new = __alloc_extent_buffer(src->fs_info, src->start, src->len);
4750 for (i = 0; i < num_pages; i++) {
4751 p = alloc_page(GFP_NOFS);
4753 btrfs_release_extent_buffer(new);
4756 attach_extent_buffer_page(new, p);
4757 WARN_ON(PageDirty(p));
4760 copy_page(page_address(p), page_address(src->pages[i]));
4763 set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
4764 set_bit(EXTENT_BUFFER_DUMMY, &new->bflags);
4769 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4770 u64 start, unsigned long len)
4772 struct extent_buffer *eb;
4773 unsigned long num_pages;
4776 num_pages = num_extent_pages(start, len);
4778 eb = __alloc_extent_buffer(fs_info, start, len);
4782 for (i = 0; i < num_pages; i++) {
4783 eb->pages[i] = alloc_page(GFP_NOFS);
4787 set_extent_buffer_uptodate(eb);
4788 btrfs_set_header_nritems(eb, 0);
4789 set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4794 __free_page(eb->pages[i - 1]);
4795 __free_extent_buffer(eb);
4799 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4802 return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize);
4805 static void check_buffer_tree_ref(struct extent_buffer *eb)
4808 /* the ref bit is tricky. We have to make sure it is set
4809 * if we have the buffer dirty. Otherwise the
4810 * code to free a buffer can end up dropping a dirty
4813 * Once the ref bit is set, it won't go away while the
4814 * buffer is dirty or in writeback, and it also won't
4815 * go away while we have the reference count on the
4818 * We can't just set the ref bit without bumping the
4819 * ref on the eb because free_extent_buffer might
4820 * see the ref bit and try to clear it. If this happens
4821 * free_extent_buffer might end up dropping our original
4822 * ref by mistake and freeing the page before we are able
4823 * to add one more ref.
4825 * So bump the ref count first, then set the bit. If someone
4826 * beat us to it, drop the ref we added.
4828 refs = atomic_read(&eb->refs);
4829 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4832 spin_lock(&eb->refs_lock);
4833 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4834 atomic_inc(&eb->refs);
4835 spin_unlock(&eb->refs_lock);
4838 static void mark_extent_buffer_accessed(struct extent_buffer *eb,
4839 struct page *accessed)
4841 unsigned long num_pages, i;
4843 check_buffer_tree_ref(eb);
4845 num_pages = num_extent_pages(eb->start, eb->len);
4846 for (i = 0; i < num_pages; i++) {
4847 struct page *p = eb->pages[i];
4850 mark_page_accessed(p);
4854 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
4857 struct extent_buffer *eb;
4860 eb = radix_tree_lookup(&fs_info->buffer_radix,
4861 start >> PAGE_SHIFT);
4862 if (eb && atomic_inc_not_zero(&eb->refs)) {
4865 * Lock our eb's refs_lock to avoid races with
4866 * free_extent_buffer. When we get our eb it might be flagged
4867 * with EXTENT_BUFFER_STALE and another task running
4868 * free_extent_buffer might have seen that flag set,
4869 * eb->refs == 2, that the buffer isn't under IO (dirty and
4870 * writeback flags not set) and it's still in the tree (flag
4871 * EXTENT_BUFFER_TREE_REF set), therefore being in the process
4872 * of decrementing the extent buffer's reference count twice.
4873 * So here we could race and increment the eb's reference count,
4874 * clear its stale flag, mark it as dirty and drop our reference
4875 * before the other task finishes executing free_extent_buffer,
4876 * which would later result in an attempt to free an extent
4877 * buffer that is dirty.
4879 if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) {
4880 spin_lock(&eb->refs_lock);
4881 spin_unlock(&eb->refs_lock);
4883 mark_extent_buffer_accessed(eb, NULL);
4891 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4892 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
4895 struct extent_buffer *eb, *exists = NULL;
4898 eb = find_extent_buffer(fs_info, start);
4901 eb = alloc_dummy_extent_buffer(fs_info, start);
4904 eb->fs_info = fs_info;
4906 ret = radix_tree_preload(GFP_NOFS);
4909 spin_lock(&fs_info->buffer_lock);
4910 ret = radix_tree_insert(&fs_info->buffer_radix,
4911 start >> PAGE_SHIFT, eb);
4912 spin_unlock(&fs_info->buffer_lock);
4913 radix_tree_preload_end();
4914 if (ret == -EEXIST) {
4915 exists = find_extent_buffer(fs_info, start);
4921 check_buffer_tree_ref(eb);
4922 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
4925 * We will free dummy extent buffer's if they come into
4926 * free_extent_buffer with a ref count of 2, but if we are using this we
4927 * want the buffers to stay in memory until we're done with them, so
4928 * bump the ref count again.
4930 atomic_inc(&eb->refs);
4933 btrfs_release_extent_buffer(eb);
4938 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
4941 unsigned long len = fs_info->nodesize;
4942 unsigned long num_pages = num_extent_pages(start, len);
4944 unsigned long index = start >> PAGE_SHIFT;
4945 struct extent_buffer *eb;
4946 struct extent_buffer *exists = NULL;
4948 struct address_space *mapping = fs_info->btree_inode->i_mapping;
4952 if (!IS_ALIGNED(start, fs_info->sectorsize)) {
4953 btrfs_err(fs_info, "bad tree block start %llu", start);
4954 return ERR_PTR(-EINVAL);
4957 eb = find_extent_buffer(fs_info, start);
4961 eb = __alloc_extent_buffer(fs_info, start, len);
4963 return ERR_PTR(-ENOMEM);
4965 for (i = 0; i < num_pages; i++, index++) {
4966 p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL);
4968 exists = ERR_PTR(-ENOMEM);
4972 spin_lock(&mapping->private_lock);
4973 if (PagePrivate(p)) {
4975 * We could have already allocated an eb for this page
4976 * and attached one so lets see if we can get a ref on
4977 * the existing eb, and if we can we know it's good and
4978 * we can just return that one, else we know we can just
4979 * overwrite page->private.
4981 exists = (struct extent_buffer *)p->private;
4982 if (atomic_inc_not_zero(&exists->refs)) {
4983 spin_unlock(&mapping->private_lock);
4986 mark_extent_buffer_accessed(exists, p);
4992 * Do this so attach doesn't complain and we need to
4993 * drop the ref the old guy had.
4995 ClearPagePrivate(p);
4996 WARN_ON(PageDirty(p));
4999 attach_extent_buffer_page(eb, p);
5000 spin_unlock(&mapping->private_lock);
5001 WARN_ON(PageDirty(p));
5003 if (!PageUptodate(p))
5007 * see below about how we avoid a nasty race with release page
5008 * and why we unlock later
5012 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5014 ret = radix_tree_preload(GFP_NOFS);
5016 exists = ERR_PTR(ret);
5020 spin_lock(&fs_info->buffer_lock);
5021 ret = radix_tree_insert(&fs_info->buffer_radix,
5022 start >> PAGE_SHIFT, eb);
5023 spin_unlock(&fs_info->buffer_lock);
5024 radix_tree_preload_end();
5025 if (ret == -EEXIST) {
5026 exists = find_extent_buffer(fs_info, start);
5032 /* add one reference for the tree */
5033 check_buffer_tree_ref(eb);
5034 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
5037 * there is a race where release page may have
5038 * tried to find this extent buffer in the radix
5039 * but failed. It will tell the VM it is safe to
5040 * reclaim the, and it will clear the page private bit.
5041 * We must make sure to set the page private bit properly
5042 * after the extent buffer is in the radix tree so
5043 * it doesn't get lost
5045 SetPageChecked(eb->pages[0]);
5046 for (i = 1; i < num_pages; i++) {
5048 ClearPageChecked(p);
5051 unlock_page(eb->pages[0]);
5055 WARN_ON(!atomic_dec_and_test(&eb->refs));
5056 for (i = 0; i < num_pages; i++) {
5058 unlock_page(eb->pages[i]);
5061 btrfs_release_extent_buffer(eb);
5065 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
5067 struct extent_buffer *eb =
5068 container_of(head, struct extent_buffer, rcu_head);
5070 __free_extent_buffer(eb);
5073 /* Expects to have eb->eb_lock already held */
5074 static int release_extent_buffer(struct extent_buffer *eb)
5076 WARN_ON(atomic_read(&eb->refs) == 0);
5077 if (atomic_dec_and_test(&eb->refs)) {
5078 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
5079 struct btrfs_fs_info *fs_info = eb->fs_info;
5081 spin_unlock(&eb->refs_lock);
5083 spin_lock(&fs_info->buffer_lock);
5084 radix_tree_delete(&fs_info->buffer_radix,
5085 eb->start >> PAGE_SHIFT);
5086 spin_unlock(&fs_info->buffer_lock);
5088 spin_unlock(&eb->refs_lock);
5091 /* Should be safe to release our pages at this point */
5092 btrfs_release_extent_buffer_page(eb);
5093 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5094 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))) {
5095 __free_extent_buffer(eb);
5099 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
5102 spin_unlock(&eb->refs_lock);
5107 void free_extent_buffer(struct extent_buffer *eb)
5115 refs = atomic_read(&eb->refs);
5118 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
5123 spin_lock(&eb->refs_lock);
5124 if (atomic_read(&eb->refs) == 2 &&
5125 test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))
5126 atomic_dec(&eb->refs);
5128 if (atomic_read(&eb->refs) == 2 &&
5129 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
5130 !extent_buffer_under_io(eb) &&
5131 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5132 atomic_dec(&eb->refs);
5135 * I know this is terrible, but it's temporary until we stop tracking
5136 * the uptodate bits and such for the extent buffers.
5138 release_extent_buffer(eb);
5141 void free_extent_buffer_stale(struct extent_buffer *eb)
5146 spin_lock(&eb->refs_lock);
5147 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
5149 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
5150 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5151 atomic_dec(&eb->refs);
5152 release_extent_buffer(eb);
5155 void clear_extent_buffer_dirty(struct extent_buffer *eb)
5158 unsigned long num_pages;
5161 num_pages = num_extent_pages(eb->start, eb->len);
5163 for (i = 0; i < num_pages; i++) {
5164 page = eb->pages[i];
5165 if (!PageDirty(page))
5169 WARN_ON(!PagePrivate(page));
5171 clear_page_dirty_for_io(page);
5172 spin_lock_irq(&page->mapping->tree_lock);
5173 if (!PageDirty(page)) {
5174 radix_tree_tag_clear(&page->mapping->page_tree,
5176 PAGECACHE_TAG_DIRTY);
5178 spin_unlock_irq(&page->mapping->tree_lock);
5179 ClearPageError(page);
5182 WARN_ON(atomic_read(&eb->refs) == 0);
5185 int set_extent_buffer_dirty(struct extent_buffer *eb)
5188 unsigned long num_pages;
5191 check_buffer_tree_ref(eb);
5193 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
5195 num_pages = num_extent_pages(eb->start, eb->len);
5196 WARN_ON(atomic_read(&eb->refs) == 0);
5197 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
5199 for (i = 0; i < num_pages; i++)
5200 set_page_dirty(eb->pages[i]);
5204 void clear_extent_buffer_uptodate(struct extent_buffer *eb)
5208 unsigned long num_pages;
5210 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5211 num_pages = num_extent_pages(eb->start, eb->len);
5212 for (i = 0; i < num_pages; i++) {
5213 page = eb->pages[i];
5215 ClearPageUptodate(page);
5219 void set_extent_buffer_uptodate(struct extent_buffer *eb)
5223 unsigned long num_pages;
5225 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5226 num_pages = num_extent_pages(eb->start, eb->len);
5227 for (i = 0; i < num_pages; i++) {
5228 page = eb->pages[i];
5229 SetPageUptodate(page);
5233 int extent_buffer_uptodate(struct extent_buffer *eb)
5235 return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5238 int read_extent_buffer_pages(struct extent_io_tree *tree,
5239 struct extent_buffer *eb, int wait, int mirror_num)
5245 int locked_pages = 0;
5246 int all_uptodate = 1;
5247 unsigned long num_pages;
5248 unsigned long num_reads = 0;
5249 struct bio *bio = NULL;
5250 unsigned long bio_flags = 0;
5252 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
5255 num_pages = num_extent_pages(eb->start, eb->len);
5256 for (i = 0; i < num_pages; i++) {
5257 page = eb->pages[i];
5258 if (wait == WAIT_NONE) {
5259 if (!trylock_page(page))
5267 * We need to firstly lock all pages to make sure that
5268 * the uptodate bit of our pages won't be affected by
5269 * clear_extent_buffer_uptodate().
5271 for (i = 0; i < num_pages; i++) {
5272 page = eb->pages[i];
5273 if (!PageUptodate(page)) {
5280 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5284 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
5285 eb->read_mirror = 0;
5286 atomic_set(&eb->io_pages, num_reads);
5287 for (i = 0; i < num_pages; i++) {
5288 page = eb->pages[i];
5290 if (!PageUptodate(page)) {
5292 atomic_dec(&eb->io_pages);
5297 ClearPageError(page);
5298 err = __extent_read_full_page(tree, page,
5299 btree_get_extent, &bio,
5300 mirror_num, &bio_flags,
5305 * We use &bio in above __extent_read_full_page,
5306 * so we ensure that if it returns error, the
5307 * current page fails to add itself to bio and
5308 * it's been unlocked.
5310 * We must dec io_pages by ourselves.
5312 atomic_dec(&eb->io_pages);
5320 err = submit_one_bio(bio, mirror_num, bio_flags);
5325 if (ret || wait != WAIT_COMPLETE)
5328 for (i = 0; i < num_pages; i++) {
5329 page = eb->pages[i];
5330 wait_on_page_locked(page);
5331 if (!PageUptodate(page))
5338 while (locked_pages > 0) {
5340 page = eb->pages[locked_pages];
5346 void read_extent_buffer(const struct extent_buffer *eb, void *dstv,
5347 unsigned long start, unsigned long len)
5353 char *dst = (char *)dstv;
5354 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5355 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5357 if (start + len > eb->len) {
5358 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n",
5359 eb->start, eb->len, start, len);
5360 memset(dst, 0, len);
5364 offset = (start_offset + start) & (PAGE_SIZE - 1);
5367 page = eb->pages[i];
5369 cur = min(len, (PAGE_SIZE - offset));
5370 kaddr = page_address(page);
5371 memcpy(dst, kaddr + offset, cur);
5380 int read_extent_buffer_to_user(const struct extent_buffer *eb,
5382 unsigned long start, unsigned long len)
5388 char __user *dst = (char __user *)dstv;
5389 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5390 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5393 WARN_ON(start > eb->len);
5394 WARN_ON(start + len > eb->start + eb->len);
5396 offset = (start_offset + start) & (PAGE_SIZE - 1);
5399 page = eb->pages[i];
5401 cur = min(len, (PAGE_SIZE - offset));
5402 kaddr = page_address(page);
5403 if (copy_to_user(dst, kaddr + offset, cur)) {
5418 * return 0 if the item is found within a page.
5419 * return 1 if the item spans two pages.
5420 * return -EINVAL otherwise.
5422 int map_private_extent_buffer(const struct extent_buffer *eb,
5423 unsigned long start, unsigned long min_len,
5424 char **map, unsigned long *map_start,
5425 unsigned long *map_len)
5427 size_t offset = start & (PAGE_SIZE - 1);
5430 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5431 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5432 unsigned long end_i = (start_offset + start + min_len - 1) >>
5435 if (start + min_len > eb->len) {
5436 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n",
5437 eb->start, eb->len, start, min_len);
5445 offset = start_offset;
5449 *map_start = ((u64)i << PAGE_SHIFT) - start_offset;
5453 kaddr = page_address(p);
5454 *map = kaddr + offset;
5455 *map_len = PAGE_SIZE - offset;
5459 int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,
5460 unsigned long start, unsigned long len)
5466 char *ptr = (char *)ptrv;
5467 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5468 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5471 WARN_ON(start > eb->len);
5472 WARN_ON(start + len > eb->start + eb->len);
5474 offset = (start_offset + start) & (PAGE_SIZE - 1);
5477 page = eb->pages[i];
5479 cur = min(len, (PAGE_SIZE - offset));
5481 kaddr = page_address(page);
5482 ret = memcmp(ptr, kaddr + offset, cur);
5494 void write_extent_buffer_chunk_tree_uuid(struct extent_buffer *eb,
5499 WARN_ON(!PageUptodate(eb->pages[0]));
5500 kaddr = page_address(eb->pages[0]);
5501 memcpy(kaddr + offsetof(struct btrfs_header, chunk_tree_uuid), srcv,
5505 void write_extent_buffer_fsid(struct extent_buffer *eb, const void *srcv)
5509 WARN_ON(!PageUptodate(eb->pages[0]));
5510 kaddr = page_address(eb->pages[0]);
5511 memcpy(kaddr + offsetof(struct btrfs_header, fsid), srcv,
5515 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
5516 unsigned long start, unsigned long len)
5522 char *src = (char *)srcv;
5523 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5524 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5526 WARN_ON(start > eb->len);
5527 WARN_ON(start + len > eb->start + eb->len);
5529 offset = (start_offset + start) & (PAGE_SIZE - 1);
5532 page = eb->pages[i];
5533 WARN_ON(!PageUptodate(page));
5535 cur = min(len, PAGE_SIZE - offset);
5536 kaddr = page_address(page);
5537 memcpy(kaddr + offset, src, cur);
5546 void memzero_extent_buffer(struct extent_buffer *eb, unsigned long start,
5553 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5554 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5556 WARN_ON(start > eb->len);
5557 WARN_ON(start + len > eb->start + eb->len);
5559 offset = (start_offset + start) & (PAGE_SIZE - 1);
5562 page = eb->pages[i];
5563 WARN_ON(!PageUptodate(page));
5565 cur = min(len, PAGE_SIZE - offset);
5566 kaddr = page_address(page);
5567 memset(kaddr + offset, 0, cur);
5575 void copy_extent_buffer_full(struct extent_buffer *dst,
5576 struct extent_buffer *src)
5581 ASSERT(dst->len == src->len);
5583 num_pages = num_extent_pages(dst->start, dst->len);
5584 for (i = 0; i < num_pages; i++)
5585 copy_page(page_address(dst->pages[i]),
5586 page_address(src->pages[i]));
5589 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
5590 unsigned long dst_offset, unsigned long src_offset,
5593 u64 dst_len = dst->len;
5598 size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1);
5599 unsigned long i = (start_offset + dst_offset) >> PAGE_SHIFT;
5601 WARN_ON(src->len != dst_len);
5603 offset = (start_offset + dst_offset) &
5607 page = dst->pages[i];
5608 WARN_ON(!PageUptodate(page));
5610 cur = min(len, (unsigned long)(PAGE_SIZE - offset));
5612 kaddr = page_address(page);
5613 read_extent_buffer(src, kaddr + offset, src_offset, cur);
5622 void le_bitmap_set(u8 *map, unsigned int start, int len)
5624 u8 *p = map + BIT_BYTE(start);
5625 const unsigned int size = start + len;
5626 int bits_to_set = BITS_PER_BYTE - (start % BITS_PER_BYTE);
5627 u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(start);
5629 while (len - bits_to_set >= 0) {
5632 bits_to_set = BITS_PER_BYTE;
5637 mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
5642 void le_bitmap_clear(u8 *map, unsigned int start, int len)
5644 u8 *p = map + BIT_BYTE(start);
5645 const unsigned int size = start + len;
5646 int bits_to_clear = BITS_PER_BYTE - (start % BITS_PER_BYTE);
5647 u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(start);
5649 while (len - bits_to_clear >= 0) {
5650 *p &= ~mask_to_clear;
5651 len -= bits_to_clear;
5652 bits_to_clear = BITS_PER_BYTE;
5657 mask_to_clear &= BITMAP_LAST_BYTE_MASK(size);
5658 *p &= ~mask_to_clear;
5663 * eb_bitmap_offset() - calculate the page and offset of the byte containing the
5665 * @eb: the extent buffer
5666 * @start: offset of the bitmap item in the extent buffer
5668 * @page_index: return index of the page in the extent buffer that contains the
5670 * @page_offset: return offset into the page given by page_index
5672 * This helper hides the ugliness of finding the byte in an extent buffer which
5673 * contains a given bit.
5675 static inline void eb_bitmap_offset(struct extent_buffer *eb,
5676 unsigned long start, unsigned long nr,
5677 unsigned long *page_index,
5678 size_t *page_offset)
5680 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5681 size_t byte_offset = BIT_BYTE(nr);
5685 * The byte we want is the offset of the extent buffer + the offset of
5686 * the bitmap item in the extent buffer + the offset of the byte in the
5689 offset = start_offset + start + byte_offset;
5691 *page_index = offset >> PAGE_SHIFT;
5692 *page_offset = offset & (PAGE_SIZE - 1);
5696 * extent_buffer_test_bit - determine whether a bit in a bitmap item is set
5697 * @eb: the extent buffer
5698 * @start: offset of the bitmap item in the extent buffer
5699 * @nr: bit number to test
5701 int extent_buffer_test_bit(struct extent_buffer *eb, unsigned long start,
5709 eb_bitmap_offset(eb, start, nr, &i, &offset);
5710 page = eb->pages[i];
5711 WARN_ON(!PageUptodate(page));
5712 kaddr = page_address(page);
5713 return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1)));
5717 * extent_buffer_bitmap_set - set an area of a bitmap
5718 * @eb: the extent buffer
5719 * @start: offset of the bitmap item in the extent buffer
5720 * @pos: bit number of the first bit
5721 * @len: number of bits to set
5723 void extent_buffer_bitmap_set(struct extent_buffer *eb, unsigned long start,
5724 unsigned long pos, unsigned long len)
5730 const unsigned int size = pos + len;
5731 int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5732 u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos);
5734 eb_bitmap_offset(eb, start, pos, &i, &offset);
5735 page = eb->pages[i];
5736 WARN_ON(!PageUptodate(page));
5737 kaddr = page_address(page);
5739 while (len >= bits_to_set) {
5740 kaddr[offset] |= mask_to_set;
5742 bits_to_set = BITS_PER_BYTE;
5744 if (++offset >= PAGE_SIZE && len > 0) {
5746 page = eb->pages[++i];
5747 WARN_ON(!PageUptodate(page));
5748 kaddr = page_address(page);
5752 mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
5753 kaddr[offset] |= mask_to_set;
5759 * extent_buffer_bitmap_clear - clear an area of a bitmap
5760 * @eb: the extent buffer
5761 * @start: offset of the bitmap item in the extent buffer
5762 * @pos: bit number of the first bit
5763 * @len: number of bits to clear
5765 void extent_buffer_bitmap_clear(struct extent_buffer *eb, unsigned long start,
5766 unsigned long pos, unsigned long len)
5772 const unsigned int size = pos + len;
5773 int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5774 u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos);
5776 eb_bitmap_offset(eb, start, pos, &i, &offset);
5777 page = eb->pages[i];
5778 WARN_ON(!PageUptodate(page));
5779 kaddr = page_address(page);
5781 while (len >= bits_to_clear) {
5782 kaddr[offset] &= ~mask_to_clear;
5783 len -= bits_to_clear;
5784 bits_to_clear = BITS_PER_BYTE;
5786 if (++offset >= PAGE_SIZE && len > 0) {
5788 page = eb->pages[++i];
5789 WARN_ON(!PageUptodate(page));
5790 kaddr = page_address(page);
5794 mask_to_clear &= BITMAP_LAST_BYTE_MASK(size);
5795 kaddr[offset] &= ~mask_to_clear;
5799 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
5801 unsigned long distance = (src > dst) ? src - dst : dst - src;
5802 return distance < len;
5805 static void copy_pages(struct page *dst_page, struct page *src_page,
5806 unsigned long dst_off, unsigned long src_off,
5809 char *dst_kaddr = page_address(dst_page);
5811 int must_memmove = 0;
5813 if (dst_page != src_page) {
5814 src_kaddr = page_address(src_page);
5816 src_kaddr = dst_kaddr;
5817 if (areas_overlap(src_off, dst_off, len))
5822 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
5824 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
5827 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5828 unsigned long src_offset, unsigned long len)
5830 struct btrfs_fs_info *fs_info = dst->fs_info;
5832 size_t dst_off_in_page;
5833 size_t src_off_in_page;
5834 size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1);
5835 unsigned long dst_i;
5836 unsigned long src_i;
5838 if (src_offset + len > dst->len) {
5840 "memmove bogus src_offset %lu move len %lu dst len %lu",
5841 src_offset, len, dst->len);
5844 if (dst_offset + len > dst->len) {
5846 "memmove bogus dst_offset %lu move len %lu dst len %lu",
5847 dst_offset, len, dst->len);
5852 dst_off_in_page = (start_offset + dst_offset) &
5854 src_off_in_page = (start_offset + src_offset) &
5857 dst_i = (start_offset + dst_offset) >> PAGE_SHIFT;
5858 src_i = (start_offset + src_offset) >> PAGE_SHIFT;
5860 cur = min(len, (unsigned long)(PAGE_SIZE -
5862 cur = min_t(unsigned long, cur,
5863 (unsigned long)(PAGE_SIZE - dst_off_in_page));
5865 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5866 dst_off_in_page, src_off_in_page, cur);
5874 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5875 unsigned long src_offset, unsigned long len)
5877 struct btrfs_fs_info *fs_info = dst->fs_info;
5879 size_t dst_off_in_page;
5880 size_t src_off_in_page;
5881 unsigned long dst_end = dst_offset + len - 1;
5882 unsigned long src_end = src_offset + len - 1;
5883 size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1);
5884 unsigned long dst_i;
5885 unsigned long src_i;
5887 if (src_offset + len > dst->len) {
5889 "memmove bogus src_offset %lu move len %lu len %lu",
5890 src_offset, len, dst->len);
5893 if (dst_offset + len > dst->len) {
5895 "memmove bogus dst_offset %lu move len %lu len %lu",
5896 dst_offset, len, dst->len);
5899 if (dst_offset < src_offset) {
5900 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
5904 dst_i = (start_offset + dst_end) >> PAGE_SHIFT;
5905 src_i = (start_offset + src_end) >> PAGE_SHIFT;
5907 dst_off_in_page = (start_offset + dst_end) &
5909 src_off_in_page = (start_offset + src_end) &
5912 cur = min_t(unsigned long, len, src_off_in_page + 1);
5913 cur = min(cur, dst_off_in_page + 1);
5914 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5915 dst_off_in_page - cur + 1,
5916 src_off_in_page - cur + 1, cur);
5924 int try_release_extent_buffer(struct page *page)
5926 struct extent_buffer *eb;
5929 * We need to make sure nobody is attaching this page to an eb right
5932 spin_lock(&page->mapping->private_lock);
5933 if (!PagePrivate(page)) {
5934 spin_unlock(&page->mapping->private_lock);
5938 eb = (struct extent_buffer *)page->private;
5942 * This is a little awful but should be ok, we need to make sure that
5943 * the eb doesn't disappear out from under us while we're looking at
5946 spin_lock(&eb->refs_lock);
5947 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
5948 spin_unlock(&eb->refs_lock);
5949 spin_unlock(&page->mapping->private_lock);
5952 spin_unlock(&page->mapping->private_lock);
5955 * If tree ref isn't set then we know the ref on this eb is a real ref,
5956 * so just return, this page will likely be freed soon anyway.
5958 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
5959 spin_unlock(&eb->refs_lock);
5963 return release_extent_buffer(eb);