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"
25 static struct kmem_cache *extent_state_cache;
26 static struct kmem_cache *extent_buffer_cache;
27 static struct bio_set *btrfs_bioset;
29 static inline bool extent_state_in_tree(const struct extent_state *state)
31 return !RB_EMPTY_NODE(&state->rb_node);
34 #ifdef CONFIG_BTRFS_DEBUG
35 static LIST_HEAD(buffers);
36 static LIST_HEAD(states);
38 static DEFINE_SPINLOCK(leak_lock);
41 void btrfs_leak_debug_add(struct list_head *new, struct list_head *head)
45 spin_lock_irqsave(&leak_lock, flags);
47 spin_unlock_irqrestore(&leak_lock, flags);
51 void btrfs_leak_debug_del(struct list_head *entry)
55 spin_lock_irqsave(&leak_lock, flags);
57 spin_unlock_irqrestore(&leak_lock, flags);
61 void btrfs_leak_debug_check(void)
63 struct extent_state *state;
64 struct extent_buffer *eb;
66 while (!list_empty(&states)) {
67 state = list_entry(states.next, struct extent_state, leak_list);
68 pr_err("BTRFS: state leak: start %llu end %llu state %u in tree %d refs %d\n",
69 state->start, state->end, state->state,
70 extent_state_in_tree(state),
71 refcount_read(&state->refs));
72 list_del(&state->leak_list);
73 kmem_cache_free(extent_state_cache, state);
76 while (!list_empty(&buffers)) {
77 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
78 pr_err("BTRFS: buffer leak start %llu len %lu refs %d\n",
79 eb->start, eb->len, atomic_read(&eb->refs));
80 list_del(&eb->leak_list);
81 kmem_cache_free(extent_buffer_cache, eb);
85 #define btrfs_debug_check_extent_io_range(tree, start, end) \
86 __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end))
87 static inline void __btrfs_debug_check_extent_io_range(const char *caller,
88 struct extent_io_tree *tree, u64 start, u64 end)
90 if (tree->ops && tree->ops->check_extent_io_range)
91 tree->ops->check_extent_io_range(tree->private_data, caller,
95 #define btrfs_leak_debug_add(new, head) do {} while (0)
96 #define btrfs_leak_debug_del(entry) do {} while (0)
97 #define btrfs_leak_debug_check() do {} while (0)
98 #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0)
101 #define BUFFER_LRU_MAX 64
106 struct rb_node rb_node;
109 struct extent_page_data {
111 struct extent_io_tree *tree;
112 get_extent_t *get_extent;
114 /* tells writepage not to lock the state bits for this range
115 * it still does the unlocking
117 unsigned int extent_locked:1;
119 /* tells the submit_bio code to use REQ_SYNC */
120 unsigned int sync_io:1;
123 static void add_extent_changeset(struct extent_state *state, unsigned bits,
124 struct extent_changeset *changeset,
131 if (set && (state->state & bits) == bits)
133 if (!set && (state->state & bits) == 0)
135 changeset->bytes_changed += state->end - state->start + 1;
136 ret = ulist_add(&changeset->range_changed, state->start, state->end,
142 static noinline void flush_write_bio(void *data);
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 static 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, gfp_t mask)
1300 return __clear_extent_bit(tree, start, end, bits, wake, delete,
1301 cached, mask, 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, GFP_NOFS);
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,
1651 &cached_state, GFP_NOFS);
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 & MS_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 || !dev->writeable) {
2031 btrfs_bio_counter_dec(fs_info);
2035 bio_set_dev(bio, dev->bdev);
2036 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC;
2037 bio_add_page(bio, page, length, pg_offset);
2039 if (btrfsic_submit_bio_wait(bio)) {
2040 /* try to remap that extent elsewhere? */
2041 btrfs_bio_counter_dec(fs_info);
2043 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
2047 btrfs_info_rl_in_rcu(fs_info,
2048 "read error corrected: ino %llu off %llu (dev %s sector %llu)",
2050 rcu_str_deref(dev->name), sector);
2051 btrfs_bio_counter_dec(fs_info);
2056 int repair_eb_io_failure(struct btrfs_fs_info *fs_info,
2057 struct extent_buffer *eb, int mirror_num)
2059 u64 start = eb->start;
2060 unsigned long i, num_pages = num_extent_pages(eb->start, eb->len);
2063 if (sb_rdonly(fs_info->sb))
2066 for (i = 0; i < num_pages; i++) {
2067 struct page *p = eb->pages[i];
2069 ret = repair_io_failure(fs_info, 0, start, PAGE_SIZE, start, p,
2070 start - page_offset(p), mirror_num);
2080 * each time an IO finishes, we do a fast check in the IO failure tree
2081 * to see if we need to process or clean up an io_failure_record
2083 int clean_io_failure(struct btrfs_fs_info *fs_info,
2084 struct extent_io_tree *failure_tree,
2085 struct extent_io_tree *io_tree, u64 start,
2086 struct page *page, u64 ino, unsigned int pg_offset)
2089 struct io_failure_record *failrec;
2090 struct extent_state *state;
2095 ret = count_range_bits(failure_tree, &private, (u64)-1, 1,
2100 ret = get_state_failrec(failure_tree, start, &failrec);
2104 BUG_ON(!failrec->this_mirror);
2106 if (failrec->in_validation) {
2107 /* there was no real error, just free the record */
2108 btrfs_debug(fs_info,
2109 "clean_io_failure: freeing dummy error at %llu",
2113 if (sb_rdonly(fs_info->sb))
2116 spin_lock(&io_tree->lock);
2117 state = find_first_extent_bit_state(io_tree,
2120 spin_unlock(&io_tree->lock);
2122 if (state && state->start <= failrec->start &&
2123 state->end >= failrec->start + failrec->len - 1) {
2124 num_copies = btrfs_num_copies(fs_info, failrec->logical,
2126 if (num_copies > 1) {
2127 repair_io_failure(fs_info, ino, start, failrec->len,
2128 failrec->logical, page, pg_offset,
2129 failrec->failed_mirror);
2134 free_io_failure(failure_tree, io_tree, failrec);
2140 * Can be called when
2141 * - hold extent lock
2142 * - under ordered extent
2143 * - the inode is freeing
2145 void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start, u64 end)
2147 struct extent_io_tree *failure_tree = &inode->io_failure_tree;
2148 struct io_failure_record *failrec;
2149 struct extent_state *state, *next;
2151 if (RB_EMPTY_ROOT(&failure_tree->state))
2154 spin_lock(&failure_tree->lock);
2155 state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY);
2157 if (state->start > end)
2160 ASSERT(state->end <= end);
2162 next = next_state(state);
2164 failrec = state->failrec;
2165 free_extent_state(state);
2170 spin_unlock(&failure_tree->lock);
2173 int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end,
2174 struct io_failure_record **failrec_ret)
2176 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2177 struct io_failure_record *failrec;
2178 struct extent_map *em;
2179 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2180 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2181 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2185 ret = get_state_failrec(failure_tree, start, &failrec);
2187 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2191 failrec->start = start;
2192 failrec->len = end - start + 1;
2193 failrec->this_mirror = 0;
2194 failrec->bio_flags = 0;
2195 failrec->in_validation = 0;
2197 read_lock(&em_tree->lock);
2198 em = lookup_extent_mapping(em_tree, start, failrec->len);
2200 read_unlock(&em_tree->lock);
2205 if (em->start > start || em->start + em->len <= start) {
2206 free_extent_map(em);
2209 read_unlock(&em_tree->lock);
2215 logical = start - em->start;
2216 logical = em->block_start + logical;
2217 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2218 logical = em->block_start;
2219 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2220 extent_set_compress_type(&failrec->bio_flags,
2224 btrfs_debug(fs_info,
2225 "Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu",
2226 logical, start, failrec->len);
2228 failrec->logical = logical;
2229 free_extent_map(em);
2231 /* set the bits in the private failure tree */
2232 ret = set_extent_bits(failure_tree, start, end,
2233 EXTENT_LOCKED | EXTENT_DIRTY);
2235 ret = set_state_failrec(failure_tree, start, failrec);
2236 /* set the bits in the inode's tree */
2238 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED);
2244 btrfs_debug(fs_info,
2245 "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d",
2246 failrec->logical, failrec->start, failrec->len,
2247 failrec->in_validation);
2249 * when data can be on disk more than twice, add to failrec here
2250 * (e.g. with a list for failed_mirror) to make
2251 * clean_io_failure() clean all those errors at once.
2255 *failrec_ret = failrec;
2260 bool btrfs_check_repairable(struct inode *inode, struct bio *failed_bio,
2261 struct io_failure_record *failrec, int failed_mirror)
2263 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2266 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
2267 if (num_copies == 1) {
2269 * we only have a single copy of the data, so don't bother with
2270 * all the retry and error correction code that follows. no
2271 * matter what the error is, it is very likely to persist.
2273 btrfs_debug(fs_info,
2274 "Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
2275 num_copies, failrec->this_mirror, failed_mirror);
2280 * there are two premises:
2281 * a) deliver good data to the caller
2282 * b) correct the bad sectors on disk
2284 if (failed_bio->bi_vcnt > 1) {
2286 * to fulfill b), we need to know the exact failing sectors, as
2287 * we don't want to rewrite any more than the failed ones. thus,
2288 * we need separate read requests for the failed bio
2290 * if the following BUG_ON triggers, our validation request got
2291 * merged. we need separate requests for our algorithm to work.
2293 BUG_ON(failrec->in_validation);
2294 failrec->in_validation = 1;
2295 failrec->this_mirror = failed_mirror;
2298 * we're ready to fulfill a) and b) alongside. get a good copy
2299 * of the failed sector and if we succeed, we have setup
2300 * everything for repair_io_failure to do the rest for us.
2302 if (failrec->in_validation) {
2303 BUG_ON(failrec->this_mirror != failed_mirror);
2304 failrec->in_validation = 0;
2305 failrec->this_mirror = 0;
2307 failrec->failed_mirror = failed_mirror;
2308 failrec->this_mirror++;
2309 if (failrec->this_mirror == failed_mirror)
2310 failrec->this_mirror++;
2313 if (failrec->this_mirror > num_copies) {
2314 btrfs_debug(fs_info,
2315 "Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
2316 num_copies, failrec->this_mirror, failed_mirror);
2324 struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio,
2325 struct io_failure_record *failrec,
2326 struct page *page, int pg_offset, int icsum,
2327 bio_end_io_t *endio_func, void *data)
2329 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2331 struct btrfs_io_bio *btrfs_failed_bio;
2332 struct btrfs_io_bio *btrfs_bio;
2334 bio = btrfs_io_bio_alloc(1);
2335 bio->bi_end_io = endio_func;
2336 bio->bi_iter.bi_sector = failrec->logical >> 9;
2337 bio_set_dev(bio, fs_info->fs_devices->latest_bdev);
2338 bio->bi_iter.bi_size = 0;
2339 bio->bi_private = data;
2341 btrfs_failed_bio = btrfs_io_bio(failed_bio);
2342 if (btrfs_failed_bio->csum) {
2343 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
2345 btrfs_bio = btrfs_io_bio(bio);
2346 btrfs_bio->csum = btrfs_bio->csum_inline;
2348 memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + icsum,
2352 bio_add_page(bio, page, failrec->len, pg_offset);
2358 * this is a generic handler for readpage errors (default
2359 * readpage_io_failed_hook). if other copies exist, read those and write back
2360 * good data to the failed position. does not investigate in remapping the
2361 * failed extent elsewhere, hoping the device will be smart enough to do this as
2365 static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset,
2366 struct page *page, u64 start, u64 end,
2369 struct io_failure_record *failrec;
2370 struct inode *inode = page->mapping->host;
2371 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2372 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2375 blk_status_t status;
2378 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
2380 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
2384 if (!btrfs_check_repairable(inode, failed_bio, failrec,
2386 free_io_failure(failure_tree, tree, failrec);
2390 if (failed_bio->bi_vcnt > 1)
2391 read_mode |= REQ_FAILFAST_DEV;
2393 phy_offset >>= inode->i_sb->s_blocksize_bits;
2394 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
2395 start - page_offset(page),
2396 (int)phy_offset, failed_bio->bi_end_io,
2398 bio_set_op_attrs(bio, REQ_OP_READ, read_mode);
2400 btrfs_debug(btrfs_sb(inode->i_sb),
2401 "Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d",
2402 read_mode, failrec->this_mirror, failrec->in_validation);
2404 status = tree->ops->submit_bio_hook(tree->private_data, bio, failrec->this_mirror,
2405 failrec->bio_flags, 0);
2407 free_io_failure(failure_tree, tree, failrec);
2409 ret = blk_status_to_errno(status);
2415 /* lots and lots of room for performance fixes in the end_bio funcs */
2417 void end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2419 int uptodate = (err == 0);
2420 struct extent_io_tree *tree;
2423 tree = &BTRFS_I(page->mapping->host)->io_tree;
2425 if (tree->ops && tree->ops->writepage_end_io_hook)
2426 tree->ops->writepage_end_io_hook(page, start, end, NULL,
2430 ClearPageUptodate(page);
2432 ret = err < 0 ? err : -EIO;
2433 mapping_set_error(page->mapping, ret);
2438 * after a writepage IO is done, we need to:
2439 * clear the uptodate bits on error
2440 * clear the writeback bits in the extent tree for this IO
2441 * end_page_writeback if the page has no more pending IO
2443 * Scheduling is not allowed, so the extent state tree is expected
2444 * to have one and only one object corresponding to this IO.
2446 static void end_bio_extent_writepage(struct bio *bio)
2448 int error = blk_status_to_errno(bio->bi_status);
2449 struct bio_vec *bvec;
2454 ASSERT(!bio_flagged(bio, BIO_CLONED));
2455 bio_for_each_segment_all(bvec, bio, i) {
2456 struct page *page = bvec->bv_page;
2457 struct inode *inode = page->mapping->host;
2458 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2460 /* We always issue full-page reads, but if some block
2461 * in a page fails to read, blk_update_request() will
2462 * advance bv_offset and adjust bv_len to compensate.
2463 * Print a warning for nonzero offsets, and an error
2464 * if they don't add up to a full page. */
2465 if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) {
2466 if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE)
2468 "partial page write in btrfs with offset %u and length %u",
2469 bvec->bv_offset, bvec->bv_len);
2472 "incomplete page write in btrfs with offset %u and length %u",
2473 bvec->bv_offset, bvec->bv_len);
2476 start = page_offset(page);
2477 end = start + bvec->bv_offset + bvec->bv_len - 1;
2479 end_extent_writepage(page, error, start, end);
2480 end_page_writeback(page);
2487 endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len,
2490 struct extent_state *cached = NULL;
2491 u64 end = start + len - 1;
2493 if (uptodate && tree->track_uptodate)
2494 set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC);
2495 unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC);
2499 * after a readpage IO is done, we need to:
2500 * clear the uptodate bits on error
2501 * set the uptodate bits if things worked
2502 * set the page up to date if all extents in the tree are uptodate
2503 * clear the lock bit in the extent tree
2504 * unlock the page if there are no other extents locked for it
2506 * Scheduling is not allowed, so the extent state tree is expected
2507 * to have one and only one object corresponding to this IO.
2509 static void end_bio_extent_readpage(struct bio *bio)
2511 struct bio_vec *bvec;
2512 int uptodate = !bio->bi_status;
2513 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
2514 struct extent_io_tree *tree, *failure_tree;
2519 u64 extent_start = 0;
2525 ASSERT(!bio_flagged(bio, BIO_CLONED));
2526 bio_for_each_segment_all(bvec, bio, i) {
2527 struct page *page = bvec->bv_page;
2528 struct inode *inode = page->mapping->host;
2529 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2531 btrfs_debug(fs_info,
2532 "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u",
2533 (u64)bio->bi_iter.bi_sector, bio->bi_status,
2534 io_bio->mirror_num);
2535 tree = &BTRFS_I(inode)->io_tree;
2536 failure_tree = &BTRFS_I(inode)->io_failure_tree;
2538 /* We always issue full-page reads, but if some block
2539 * in a page fails to read, blk_update_request() will
2540 * advance bv_offset and adjust bv_len to compensate.
2541 * Print a warning for nonzero offsets, and an error
2542 * if they don't add up to a full page. */
2543 if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) {
2544 if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE)
2546 "partial page read in btrfs with offset %u and length %u",
2547 bvec->bv_offset, bvec->bv_len);
2550 "incomplete page read in btrfs with offset %u and length %u",
2551 bvec->bv_offset, bvec->bv_len);
2554 start = page_offset(page);
2555 end = start + bvec->bv_offset + bvec->bv_len - 1;
2558 mirror = io_bio->mirror_num;
2559 if (likely(uptodate && tree->ops)) {
2560 ret = tree->ops->readpage_end_io_hook(io_bio, offset,
2566 clean_io_failure(BTRFS_I(inode)->root->fs_info,
2567 failure_tree, tree, start,
2569 btrfs_ino(BTRFS_I(inode)), 0);
2572 if (likely(uptodate))
2576 ret = tree->ops->readpage_io_failed_hook(page, mirror);
2577 if (ret == -EAGAIN) {
2579 * Data inode's readpage_io_failed_hook() always
2582 * The generic bio_readpage_error handles errors
2583 * the following way: If possible, new read
2584 * requests are created and submitted and will
2585 * end up in end_bio_extent_readpage as well (if
2586 * we're lucky, not in the !uptodate case). In
2587 * that case it returns 0 and we just go on with
2588 * the next page in our bio. If it can't handle
2589 * the error it will return -EIO and we remain
2590 * responsible for that page.
2592 ret = bio_readpage_error(bio, offset, page,
2593 start, end, mirror);
2595 uptodate = !bio->bi_status;
2602 * metadata's readpage_io_failed_hook() always returns
2603 * -EIO and fixes nothing. -EIO is also returned if
2604 * data inode error could not be fixed.
2606 ASSERT(ret == -EIO);
2609 if (likely(uptodate)) {
2610 loff_t i_size = i_size_read(inode);
2611 pgoff_t end_index = i_size >> PAGE_SHIFT;
2614 /* Zero out the end if this page straddles i_size */
2615 off = i_size & (PAGE_SIZE-1);
2616 if (page->index == end_index && off)
2617 zero_user_segment(page, off, PAGE_SIZE);
2618 SetPageUptodate(page);
2620 ClearPageUptodate(page);
2626 if (unlikely(!uptodate)) {
2628 endio_readpage_release_extent(tree,
2634 endio_readpage_release_extent(tree, start,
2635 end - start + 1, 0);
2636 } else if (!extent_len) {
2637 extent_start = start;
2638 extent_len = end + 1 - start;
2639 } else if (extent_start + extent_len == start) {
2640 extent_len += end + 1 - start;
2642 endio_readpage_release_extent(tree, extent_start,
2643 extent_len, uptodate);
2644 extent_start = start;
2645 extent_len = end + 1 - start;
2650 endio_readpage_release_extent(tree, extent_start, extent_len,
2653 io_bio->end_io(io_bio, blk_status_to_errno(bio->bi_status));
2658 * Initialize the members up to but not including 'bio'. Use after allocating a
2659 * new bio by bio_alloc_bioset as it does not initialize the bytes outside of
2660 * 'bio' because use of __GFP_ZERO is not supported.
2662 static inline void btrfs_io_bio_init(struct btrfs_io_bio *btrfs_bio)
2664 memset(btrfs_bio, 0, offsetof(struct btrfs_io_bio, bio));
2668 * The following helpers allocate a bio. As it's backed by a bioset, it'll
2669 * never fail. We're returning a bio right now but you can call btrfs_io_bio
2670 * for the appropriate container_of magic
2672 struct bio *btrfs_bio_alloc(struct block_device *bdev, u64 first_byte)
2676 bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, btrfs_bioset);
2677 bio_set_dev(bio, bdev);
2678 bio->bi_iter.bi_sector = first_byte >> 9;
2679 btrfs_io_bio_init(btrfs_io_bio(bio));
2683 struct bio *btrfs_bio_clone(struct bio *bio)
2685 struct btrfs_io_bio *btrfs_bio;
2688 /* Bio allocation backed by a bioset does not fail */
2689 new = bio_clone_fast(bio, GFP_NOFS, btrfs_bioset);
2690 btrfs_bio = btrfs_io_bio(new);
2691 btrfs_io_bio_init(btrfs_bio);
2692 btrfs_bio->iter = bio->bi_iter;
2696 struct bio *btrfs_io_bio_alloc(unsigned int nr_iovecs)
2700 /* Bio allocation backed by a bioset does not fail */
2701 bio = bio_alloc_bioset(GFP_NOFS, nr_iovecs, btrfs_bioset);
2702 btrfs_io_bio_init(btrfs_io_bio(bio));
2706 struct bio *btrfs_bio_clone_partial(struct bio *orig, int offset, int size)
2709 struct btrfs_io_bio *btrfs_bio;
2711 /* this will never fail when it's backed by a bioset */
2712 bio = bio_clone_fast(orig, GFP_NOFS, btrfs_bioset);
2715 btrfs_bio = btrfs_io_bio(bio);
2716 btrfs_io_bio_init(btrfs_bio);
2718 bio_trim(bio, offset >> 9, size >> 9);
2719 btrfs_bio->iter = bio->bi_iter;
2723 static int __must_check submit_one_bio(struct bio *bio, int mirror_num,
2724 unsigned long bio_flags)
2726 blk_status_t ret = 0;
2727 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2728 struct page *page = bvec->bv_page;
2729 struct extent_io_tree *tree = bio->bi_private;
2732 start = page_offset(page) + bvec->bv_offset;
2734 bio->bi_private = NULL;
2738 ret = tree->ops->submit_bio_hook(tree->private_data, bio,
2739 mirror_num, bio_flags, start);
2741 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,
2949 em = __get_extent_map(inode, page, pg_offset, cur,
2950 end - cur + 1, get_extent, em_cached);
2951 if (IS_ERR_OR_NULL(em)) {
2953 unlock_extent(tree, cur, end);
2956 extent_offset = cur - em->start;
2957 BUG_ON(extent_map_end(em) <= cur);
2960 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2961 this_bio_flag |= EXTENT_BIO_COMPRESSED;
2962 extent_set_compress_type(&this_bio_flag,
2966 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2967 cur_end = min(extent_map_end(em) - 1, end);
2968 iosize = ALIGN(iosize, blocksize);
2969 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2970 disk_io_size = em->block_len;
2971 offset = em->block_start;
2973 offset = em->block_start + extent_offset;
2974 disk_io_size = iosize;
2977 block_start = em->block_start;
2978 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2979 block_start = EXTENT_MAP_HOLE;
2982 * If we have a file range that points to a compressed extent
2983 * and it's followed by a consecutive file range that points to
2984 * to the same compressed extent (possibly with a different
2985 * offset and/or length, so it either points to the whole extent
2986 * or only part of it), we must make sure we do not submit a
2987 * single bio to populate the pages for the 2 ranges because
2988 * this makes the compressed extent read zero out the pages
2989 * belonging to the 2nd range. Imagine the following scenario:
2992 * [0 - 8K] [8K - 24K]
2995 * points to extent X, points to extent X,
2996 * offset 4K, length of 8K offset 0, length 16K
2998 * [extent X, compressed length = 4K uncompressed length = 16K]
3000 * If the bio to read the compressed extent covers both ranges,
3001 * it will decompress extent X into the pages belonging to the
3002 * first range and then it will stop, zeroing out the remaining
3003 * pages that belong to the other range that points to extent X.
3004 * So here we make sure we submit 2 bios, one for the first
3005 * range and another one for the third range. Both will target
3006 * the same physical extent from disk, but we can't currently
3007 * make the compressed bio endio callback populate the pages
3008 * for both ranges because each compressed bio is tightly
3009 * coupled with a single extent map, and each range can have
3010 * an extent map with a different offset value relative to the
3011 * uncompressed data of our extent and different lengths. This
3012 * is a corner case so we prioritize correctness over
3013 * non-optimal behavior (submitting 2 bios for the same extent).
3015 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) &&
3016 prev_em_start && *prev_em_start != (u64)-1 &&
3017 *prev_em_start != em->orig_start)
3018 force_bio_submit = true;
3021 *prev_em_start = em->orig_start;
3023 free_extent_map(em);
3026 /* we've found a hole, just zero and go on */
3027 if (block_start == EXTENT_MAP_HOLE) {
3029 struct extent_state *cached = NULL;
3031 userpage = kmap_atomic(page);
3032 memset(userpage + pg_offset, 0, iosize);
3033 flush_dcache_page(page);
3034 kunmap_atomic(userpage);
3036 set_extent_uptodate(tree, cur, cur + iosize - 1,
3038 unlock_extent_cached(tree, cur,
3042 pg_offset += iosize;
3045 /* the get_extent function already copied into the page */
3046 if (test_range_bit(tree, cur, cur_end,
3047 EXTENT_UPTODATE, 1, NULL)) {
3048 check_page_uptodate(tree, page);
3049 unlock_extent(tree, cur, cur + iosize - 1);
3051 pg_offset += iosize;
3054 /* we have an inline extent but it didn't get marked up
3055 * to date. Error out
3057 if (block_start == EXTENT_MAP_INLINE) {
3059 unlock_extent(tree, cur, cur + iosize - 1);
3061 pg_offset += iosize;
3065 ret = submit_extent_page(REQ_OP_READ | read_flags, tree, NULL,
3066 page, offset, disk_io_size,
3067 pg_offset, bdev, bio,
3068 end_bio_extent_readpage, mirror_num,
3074 *bio_flags = this_bio_flag;
3077 unlock_extent(tree, cur, cur + iosize - 1);
3081 pg_offset += iosize;
3085 if (!PageError(page))
3086 SetPageUptodate(page);
3092 static inline void __do_contiguous_readpages(struct extent_io_tree *tree,
3093 struct page *pages[], int nr_pages,
3095 get_extent_t *get_extent,
3096 struct extent_map **em_cached,
3097 struct bio **bio, int mirror_num,
3098 unsigned long *bio_flags,
3101 struct inode *inode;
3102 struct btrfs_ordered_extent *ordered;
3105 inode = pages[0]->mapping->host;
3107 lock_extent(tree, start, end);
3108 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
3112 unlock_extent(tree, start, end);
3113 btrfs_start_ordered_extent(inode, ordered, 1);
3114 btrfs_put_ordered_extent(ordered);
3117 for (index = 0; index < nr_pages; index++) {
3118 __do_readpage(tree, pages[index], get_extent, em_cached, bio,
3119 mirror_num, bio_flags, 0, prev_em_start);
3120 put_page(pages[index]);
3124 static void __extent_readpages(struct extent_io_tree *tree,
3125 struct page *pages[],
3126 int nr_pages, get_extent_t *get_extent,
3127 struct extent_map **em_cached,
3128 struct bio **bio, int mirror_num,
3129 unsigned long *bio_flags,
3136 int first_index = 0;
3138 for (index = 0; index < nr_pages; index++) {
3139 page_start = page_offset(pages[index]);
3142 end = start + PAGE_SIZE - 1;
3143 first_index = index;
3144 } else if (end + 1 == page_start) {
3147 __do_contiguous_readpages(tree, &pages[first_index],
3148 index - first_index, start,
3149 end, get_extent, em_cached,
3150 bio, mirror_num, bio_flags,
3153 end = start + PAGE_SIZE - 1;
3154 first_index = index;
3159 __do_contiguous_readpages(tree, &pages[first_index],
3160 index - first_index, start,
3161 end, get_extent, em_cached, bio,
3162 mirror_num, bio_flags,
3166 static int __extent_read_full_page(struct extent_io_tree *tree,
3168 get_extent_t *get_extent,
3169 struct bio **bio, int mirror_num,
3170 unsigned long *bio_flags,
3171 unsigned int read_flags)
3173 struct inode *inode = page->mapping->host;
3174 struct btrfs_ordered_extent *ordered;
3175 u64 start = page_offset(page);
3176 u64 end = start + PAGE_SIZE - 1;
3180 lock_extent(tree, start, end);
3181 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
3185 unlock_extent(tree, start, end);
3186 btrfs_start_ordered_extent(inode, ordered, 1);
3187 btrfs_put_ordered_extent(ordered);
3190 ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num,
3191 bio_flags, read_flags, NULL);
3195 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
3196 get_extent_t *get_extent, int mirror_num)
3198 struct bio *bio = NULL;
3199 unsigned long bio_flags = 0;
3202 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
3205 ret = submit_one_bio(bio, mirror_num, bio_flags);
3209 static void update_nr_written(struct writeback_control *wbc,
3210 unsigned long nr_written)
3212 wbc->nr_to_write -= nr_written;
3216 * helper for __extent_writepage, doing all of the delayed allocation setup.
3218 * This returns 1 if our fill_delalloc function did all the work required
3219 * to write the page (copy into inline extent). In this case the IO has
3220 * been started and the page is already unlocked.
3222 * This returns 0 if all went well (page still locked)
3223 * This returns < 0 if there were errors (page still locked)
3225 static noinline_for_stack int writepage_delalloc(struct inode *inode,
3226 struct page *page, struct writeback_control *wbc,
3227 struct extent_page_data *epd,
3229 unsigned long *nr_written)
3231 struct extent_io_tree *tree = epd->tree;
3232 u64 page_end = delalloc_start + PAGE_SIZE - 1;
3234 u64 delalloc_to_write = 0;
3235 u64 delalloc_end = 0;
3237 int page_started = 0;
3239 if (epd->extent_locked || !tree->ops || !tree->ops->fill_delalloc)
3242 while (delalloc_end < page_end) {
3243 nr_delalloc = find_lock_delalloc_range(inode, tree,
3247 BTRFS_MAX_EXTENT_SIZE);
3248 if (nr_delalloc == 0) {
3249 delalloc_start = delalloc_end + 1;
3252 ret = tree->ops->fill_delalloc(inode, page,
3257 /* File system has been set read-only */
3260 /* fill_delalloc should be return < 0 for error
3261 * but just in case, we use > 0 here meaning the
3262 * IO is started, so we don't want to return > 0
3263 * unless things are going well.
3265 ret = ret < 0 ? ret : -EIO;
3269 * delalloc_end is already one less than the total length, so
3270 * we don't subtract one from PAGE_SIZE
3272 delalloc_to_write += (delalloc_end - delalloc_start +
3273 PAGE_SIZE) >> PAGE_SHIFT;
3274 delalloc_start = delalloc_end + 1;
3276 if (wbc->nr_to_write < delalloc_to_write) {
3279 if (delalloc_to_write < thresh * 2)
3280 thresh = delalloc_to_write;
3281 wbc->nr_to_write = min_t(u64, delalloc_to_write,
3285 /* did the fill delalloc function already unlock and start
3290 * we've unlocked the page, so we can't update
3291 * the mapping's writeback index, just update
3294 wbc->nr_to_write -= *nr_written;
3305 * helper for __extent_writepage. This calls the writepage start hooks,
3306 * and does the loop to map the page into extents and bios.
3308 * We return 1 if the IO is started and the page is unlocked,
3309 * 0 if all went well (page still locked)
3310 * < 0 if there were errors (page still locked)
3312 static noinline_for_stack int __extent_writepage_io(struct inode *inode,
3314 struct writeback_control *wbc,
3315 struct extent_page_data *epd,
3317 unsigned long nr_written,
3318 unsigned int write_flags, int *nr_ret)
3320 struct extent_io_tree *tree = epd->tree;
3321 u64 start = page_offset(page);
3322 u64 page_end = start + PAGE_SIZE - 1;
3328 struct extent_map *em;
3329 struct block_device *bdev;
3330 size_t pg_offset = 0;
3336 if (tree->ops && tree->ops->writepage_start_hook) {
3337 ret = tree->ops->writepage_start_hook(page, start,
3340 /* Fixup worker will requeue */
3342 wbc->pages_skipped++;
3344 redirty_page_for_writepage(wbc, page);
3346 update_nr_written(wbc, nr_written);
3353 * we don't want to touch the inode after unlocking the page,
3354 * so we update the mapping writeback index now
3356 update_nr_written(wbc, nr_written + 1);
3359 if (i_size <= start) {
3360 if (tree->ops && tree->ops->writepage_end_io_hook)
3361 tree->ops->writepage_end_io_hook(page, start,
3366 blocksize = inode->i_sb->s_blocksize;
3368 while (cur <= end) {
3372 if (cur >= i_size) {
3373 if (tree->ops && tree->ops->writepage_end_io_hook)
3374 tree->ops->writepage_end_io_hook(page, cur,
3378 em = epd->get_extent(BTRFS_I(inode), page, pg_offset, cur,
3380 if (IS_ERR_OR_NULL(em)) {
3382 ret = PTR_ERR_OR_ZERO(em);
3386 extent_offset = cur - em->start;
3387 em_end = extent_map_end(em);
3388 BUG_ON(em_end <= cur);
3390 iosize = min(em_end - cur, end - cur + 1);
3391 iosize = ALIGN(iosize, blocksize);
3392 offset = em->block_start + extent_offset;
3394 block_start = em->block_start;
3395 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3396 free_extent_map(em);
3400 * compressed and inline extents are written through other
3403 if (compressed || block_start == EXTENT_MAP_HOLE ||
3404 block_start == EXTENT_MAP_INLINE) {
3406 * end_io notification does not happen here for
3407 * compressed extents
3409 if (!compressed && tree->ops &&
3410 tree->ops->writepage_end_io_hook)
3411 tree->ops->writepage_end_io_hook(page, cur,
3414 else if (compressed) {
3415 /* we don't want to end_page_writeback on
3416 * a compressed extent. this happens
3423 pg_offset += iosize;
3427 set_range_writeback(tree, cur, cur + iosize - 1);
3428 if (!PageWriteback(page)) {
3429 btrfs_err(BTRFS_I(inode)->root->fs_info,
3430 "page %lu not writeback, cur %llu end %llu",
3431 page->index, cur, end);
3434 ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc,
3435 page, offset, iosize, pg_offset,
3437 end_bio_extent_writepage,
3441 if (PageWriteback(page))
3442 end_page_writeback(page);
3446 pg_offset += iosize;
3455 * the writepage semantics are similar to regular writepage. extent
3456 * records are inserted to lock ranges in the tree, and as dirty areas
3457 * are found, they are marked writeback. Then the lock bits are removed
3458 * and the end_io handler clears the writeback ranges
3460 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
3463 struct inode *inode = page->mapping->host;
3464 struct extent_page_data *epd = data;
3465 u64 start = page_offset(page);
3466 u64 page_end = start + PAGE_SIZE - 1;
3469 size_t pg_offset = 0;
3470 loff_t i_size = i_size_read(inode);
3471 unsigned long end_index = i_size >> PAGE_SHIFT;
3472 unsigned int write_flags = 0;
3473 unsigned long nr_written = 0;
3475 write_flags = wbc_to_write_flags(wbc);
3477 trace___extent_writepage(page, inode, wbc);
3479 WARN_ON(!PageLocked(page));
3481 ClearPageError(page);
3483 pg_offset = i_size & (PAGE_SIZE - 1);
3484 if (page->index > end_index ||
3485 (page->index == end_index && !pg_offset)) {
3486 page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
3491 if (page->index == end_index) {
3494 userpage = kmap_atomic(page);
3495 memset(userpage + pg_offset, 0,
3496 PAGE_SIZE - pg_offset);
3497 kunmap_atomic(userpage);
3498 flush_dcache_page(page);
3503 set_page_extent_mapped(page);
3505 ret = writepage_delalloc(inode, page, wbc, epd, start, &nr_written);
3511 ret = __extent_writepage_io(inode, page, wbc, epd,
3512 i_size, nr_written, write_flags, &nr);
3518 /* make sure the mapping tag for page dirty gets cleared */
3519 set_page_writeback(page);
3520 end_page_writeback(page);
3522 if (PageError(page)) {
3523 ret = ret < 0 ? ret : -EIO;
3524 end_extent_writepage(page, ret, start, page_end);
3533 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3535 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
3536 TASK_UNINTERRUPTIBLE);
3539 static noinline_for_stack int
3540 lock_extent_buffer_for_io(struct extent_buffer *eb,
3541 struct btrfs_fs_info *fs_info,
3542 struct extent_page_data *epd)
3544 unsigned long i, num_pages;
3548 if (!btrfs_try_tree_write_lock(eb)) {
3550 flush_write_bio(epd);
3551 btrfs_tree_lock(eb);
3554 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3555 btrfs_tree_unlock(eb);
3559 flush_write_bio(epd);
3563 wait_on_extent_buffer_writeback(eb);
3564 btrfs_tree_lock(eb);
3565 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3567 btrfs_tree_unlock(eb);
3572 * We need to do this to prevent races in people who check if the eb is
3573 * under IO since we can end up having no IO bits set for a short period
3576 spin_lock(&eb->refs_lock);
3577 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3578 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3579 spin_unlock(&eb->refs_lock);
3580 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3581 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
3583 fs_info->dirty_metadata_batch);
3586 spin_unlock(&eb->refs_lock);
3589 btrfs_tree_unlock(eb);
3594 num_pages = num_extent_pages(eb->start, eb->len);
3595 for (i = 0; i < num_pages; i++) {
3596 struct page *p = eb->pages[i];
3598 if (!trylock_page(p)) {
3600 flush_write_bio(epd);
3610 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3612 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3613 smp_mb__after_atomic();
3614 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3617 static void set_btree_ioerr(struct page *page)
3619 struct extent_buffer *eb = (struct extent_buffer *)page->private;
3622 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))
3626 * If writeback for a btree extent that doesn't belong to a log tree
3627 * failed, increment the counter transaction->eb_write_errors.
3628 * We do this because while the transaction is running and before it's
3629 * committing (when we call filemap_fdata[write|wait]_range against
3630 * the btree inode), we might have
3631 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
3632 * returns an error or an error happens during writeback, when we're
3633 * committing the transaction we wouldn't know about it, since the pages
3634 * can be no longer dirty nor marked anymore for writeback (if a
3635 * subsequent modification to the extent buffer didn't happen before the
3636 * transaction commit), which makes filemap_fdata[write|wait]_range not
3637 * able to find the pages tagged with SetPageError at transaction
3638 * commit time. So if this happens we must abort the transaction,
3639 * otherwise we commit a super block with btree roots that point to
3640 * btree nodes/leafs whose content on disk is invalid - either garbage
3641 * or the content of some node/leaf from a past generation that got
3642 * cowed or deleted and is no longer valid.
3644 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
3645 * not be enough - we need to distinguish between log tree extents vs
3646 * non-log tree extents, and the next filemap_fdatawait_range() call
3647 * will catch and clear such errors in the mapping - and that call might
3648 * be from a log sync and not from a transaction commit. Also, checking
3649 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
3650 * not done and would not be reliable - the eb might have been released
3651 * from memory and reading it back again means that flag would not be
3652 * set (since it's a runtime flag, not persisted on disk).
3654 * Using the flags below in the btree inode also makes us achieve the
3655 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
3656 * writeback for all dirty pages and before filemap_fdatawait_range()
3657 * is called, the writeback for all dirty pages had already finished
3658 * with errors - because we were not using AS_EIO/AS_ENOSPC,
3659 * filemap_fdatawait_range() would return success, as it could not know
3660 * that writeback errors happened (the pages were no longer tagged for
3663 switch (eb->log_index) {
3665 set_bit(BTRFS_FS_BTREE_ERR, &eb->fs_info->flags);
3668 set_bit(BTRFS_FS_LOG1_ERR, &eb->fs_info->flags);
3671 set_bit(BTRFS_FS_LOG2_ERR, &eb->fs_info->flags);
3674 BUG(); /* unexpected, logic error */
3678 static void end_bio_extent_buffer_writepage(struct bio *bio)
3680 struct bio_vec *bvec;
3681 struct extent_buffer *eb;
3684 ASSERT(!bio_flagged(bio, BIO_CLONED));
3685 bio_for_each_segment_all(bvec, bio, i) {
3686 struct page *page = bvec->bv_page;
3688 eb = (struct extent_buffer *)page->private;
3690 done = atomic_dec_and_test(&eb->io_pages);
3692 if (bio->bi_status ||
3693 test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) {
3694 ClearPageUptodate(page);
3695 set_btree_ioerr(page);
3698 end_page_writeback(page);
3703 end_extent_buffer_writeback(eb);
3709 static noinline_for_stack int write_one_eb(struct extent_buffer *eb,
3710 struct btrfs_fs_info *fs_info,
3711 struct writeback_control *wbc,
3712 struct extent_page_data *epd)
3714 struct block_device *bdev = fs_info->fs_devices->latest_bdev;
3715 struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
3716 u64 offset = eb->start;
3718 unsigned long i, num_pages;
3719 unsigned long start, end;
3720 unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META;
3723 clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
3724 num_pages = num_extent_pages(eb->start, eb->len);
3725 atomic_set(&eb->io_pages, num_pages);
3727 /* set btree blocks beyond nritems with 0 to avoid stale content. */
3728 nritems = btrfs_header_nritems(eb);
3729 if (btrfs_header_level(eb) > 0) {
3730 end = btrfs_node_key_ptr_offset(nritems);
3732 memzero_extent_buffer(eb, end, eb->len - end);
3736 * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0
3738 start = btrfs_item_nr_offset(nritems);
3739 end = BTRFS_LEAF_DATA_OFFSET + leaf_data_end(fs_info, eb);
3740 memzero_extent_buffer(eb, start, end - start);
3743 for (i = 0; i < num_pages; i++) {
3744 struct page *p = eb->pages[i];
3746 clear_page_dirty_for_io(p);
3747 set_page_writeback(p);
3748 ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc,
3749 p, offset, PAGE_SIZE, 0, bdev,
3751 end_bio_extent_buffer_writepage,
3755 if (PageWriteback(p))
3756 end_page_writeback(p);
3757 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3758 end_extent_buffer_writeback(eb);
3762 offset += PAGE_SIZE;
3763 update_nr_written(wbc, 1);
3767 if (unlikely(ret)) {
3768 for (; i < num_pages; i++) {
3769 struct page *p = eb->pages[i];
3770 clear_page_dirty_for_io(p);
3778 int btree_write_cache_pages(struct address_space *mapping,
3779 struct writeback_control *wbc)
3781 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
3782 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
3783 struct extent_buffer *eb, *prev_eb = NULL;
3784 struct extent_page_data epd = {
3788 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3792 int nr_to_write_done = 0;
3793 struct pagevec pvec;
3796 pgoff_t end; /* Inclusive */
3800 pagevec_init(&pvec, 0);
3801 if (wbc->range_cyclic) {
3802 index = mapping->writeback_index; /* Start from prev offset */
3805 index = wbc->range_start >> PAGE_SHIFT;
3806 end = wbc->range_end >> PAGE_SHIFT;
3809 if (wbc->sync_mode == WB_SYNC_ALL)
3810 tag = PAGECACHE_TAG_TOWRITE;
3812 tag = PAGECACHE_TAG_DIRTY;
3814 if (wbc->sync_mode == WB_SYNC_ALL)
3815 tag_pages_for_writeback(mapping, index, end);
3816 while (!done && !nr_to_write_done && (index <= end) &&
3817 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3818 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3822 for (i = 0; i < nr_pages; i++) {
3823 struct page *page = pvec.pages[i];
3825 if (!PagePrivate(page))
3828 if (!wbc->range_cyclic && page->index > end) {
3833 spin_lock(&mapping->private_lock);
3834 if (!PagePrivate(page)) {
3835 spin_unlock(&mapping->private_lock);
3839 eb = (struct extent_buffer *)page->private;
3842 * Shouldn't happen and normally this would be a BUG_ON
3843 * but no sense in crashing the users box for something
3844 * we can survive anyway.
3847 spin_unlock(&mapping->private_lock);
3851 if (eb == prev_eb) {
3852 spin_unlock(&mapping->private_lock);
3856 ret = atomic_inc_not_zero(&eb->refs);
3857 spin_unlock(&mapping->private_lock);
3862 ret = lock_extent_buffer_for_io(eb, fs_info, &epd);
3864 free_extent_buffer(eb);
3868 ret = write_one_eb(eb, fs_info, wbc, &epd);
3871 free_extent_buffer(eb);
3874 free_extent_buffer(eb);
3877 * the filesystem may choose to bump up nr_to_write.
3878 * We have to make sure to honor the new nr_to_write
3881 nr_to_write_done = wbc->nr_to_write <= 0;
3883 pagevec_release(&pvec);
3886 if (!scanned && !done) {
3888 * We hit the last page and there is more work to be done: wrap
3889 * back to the start of the file
3895 flush_write_bio(&epd);
3900 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3901 * @mapping: address space structure to write
3902 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3903 * @writepage: function called for each page
3904 * @data: data passed to writepage function
3906 * If a page is already under I/O, write_cache_pages() skips it, even
3907 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3908 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3909 * and msync() need to guarantee that all the data which was dirty at the time
3910 * the call was made get new I/O started against them. If wbc->sync_mode is
3911 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3912 * existing IO to complete.
3914 static int extent_write_cache_pages(struct address_space *mapping,
3915 struct writeback_control *wbc,
3916 writepage_t writepage, void *data,
3917 void (*flush_fn)(void *))
3919 struct inode *inode = mapping->host;
3922 int nr_to_write_done = 0;
3923 struct pagevec pvec;
3926 pgoff_t end; /* Inclusive */
3928 int range_whole = 0;
3933 * We have to hold onto the inode so that ordered extents can do their
3934 * work when the IO finishes. The alternative to this is failing to add
3935 * an ordered extent if the igrab() fails there and that is a huge pain
3936 * to deal with, so instead just hold onto the inode throughout the
3937 * writepages operation. If it fails here we are freeing up the inode
3938 * anyway and we'd rather not waste our time writing out stuff that is
3939 * going to be truncated anyway.
3944 pagevec_init(&pvec, 0);
3945 if (wbc->range_cyclic) {
3946 index = mapping->writeback_index; /* Start from prev offset */
3949 index = wbc->range_start >> PAGE_SHIFT;
3950 end = wbc->range_end >> PAGE_SHIFT;
3951 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
3955 if (wbc->sync_mode == WB_SYNC_ALL)
3956 tag = PAGECACHE_TAG_TOWRITE;
3958 tag = PAGECACHE_TAG_DIRTY;
3960 if (wbc->sync_mode == WB_SYNC_ALL)
3961 tag_pages_for_writeback(mapping, index, end);
3963 while (!done && !nr_to_write_done && (index <= end) &&
3964 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3965 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3969 for (i = 0; i < nr_pages; i++) {
3970 struct page *page = pvec.pages[i];
3972 done_index = page->index;
3974 * At this point we hold neither mapping->tree_lock nor
3975 * lock on the page itself: the page may be truncated or
3976 * invalidated (changing page->mapping to NULL), or even
3977 * swizzled back from swapper_space to tmpfs file
3980 if (!trylock_page(page)) {
3985 if (unlikely(page->mapping != mapping)) {
3990 if (!wbc->range_cyclic && page->index > end) {
3996 if (wbc->sync_mode != WB_SYNC_NONE) {
3997 if (PageWriteback(page))
3999 wait_on_page_writeback(page);
4002 if (PageWriteback(page) ||
4003 !clear_page_dirty_for_io(page)) {
4008 ret = (*writepage)(page, wbc, data);
4010 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
4016 * done_index is set past this page,
4017 * so media errors will not choke
4018 * background writeout for the entire
4019 * file. This has consequences for
4020 * range_cyclic semantics (ie. it may
4021 * not be suitable for data integrity
4024 done_index = page->index + 1;
4030 * the filesystem may choose to bump up nr_to_write.
4031 * We have to make sure to honor the new nr_to_write
4034 nr_to_write_done = wbc->nr_to_write <= 0;
4036 pagevec_release(&pvec);
4039 if (!scanned && !done) {
4041 * We hit the last page and there is more work to be done: wrap
4042 * back to the start of the file
4049 if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole))
4050 mapping->writeback_index = done_index;
4052 btrfs_add_delayed_iput(inode);
4056 static void flush_epd_write_bio(struct extent_page_data *epd)
4061 ret = submit_one_bio(epd->bio, 0, 0);
4062 BUG_ON(ret < 0); /* -ENOMEM */
4067 static noinline void flush_write_bio(void *data)
4069 struct extent_page_data *epd = data;
4070 flush_epd_write_bio(epd);
4073 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
4074 get_extent_t *get_extent,
4075 struct writeback_control *wbc)
4078 struct extent_page_data epd = {
4081 .get_extent = get_extent,
4083 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4086 ret = __extent_writepage(page, wbc, &epd);
4088 flush_epd_write_bio(&epd);
4092 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
4093 u64 start, u64 end, get_extent_t *get_extent,
4097 struct address_space *mapping = inode->i_mapping;
4099 unsigned long nr_pages = (end - start + PAGE_SIZE) >>
4102 struct extent_page_data epd = {
4105 .get_extent = get_extent,
4107 .sync_io = mode == WB_SYNC_ALL,
4109 struct writeback_control wbc_writepages = {
4111 .nr_to_write = nr_pages * 2,
4112 .range_start = start,
4113 .range_end = end + 1,
4116 while (start <= end) {
4117 page = find_get_page(mapping, start >> PAGE_SHIFT);
4118 if (clear_page_dirty_for_io(page))
4119 ret = __extent_writepage(page, &wbc_writepages, &epd);
4121 if (tree->ops && tree->ops->writepage_end_io_hook)
4122 tree->ops->writepage_end_io_hook(page, start,
4123 start + PAGE_SIZE - 1,
4131 flush_epd_write_bio(&epd);
4135 int extent_writepages(struct extent_io_tree *tree,
4136 struct address_space *mapping,
4137 get_extent_t *get_extent,
4138 struct writeback_control *wbc)
4141 struct extent_page_data epd = {
4144 .get_extent = get_extent,
4146 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4149 ret = extent_write_cache_pages(mapping, wbc, __extent_writepage, &epd,
4151 flush_epd_write_bio(&epd);
4155 int extent_readpages(struct extent_io_tree *tree,
4156 struct address_space *mapping,
4157 struct list_head *pages, unsigned nr_pages,
4158 get_extent_t get_extent)
4160 struct bio *bio = NULL;
4162 unsigned long bio_flags = 0;
4163 struct page *pagepool[16];
4165 struct extent_map *em_cached = NULL;
4167 u64 prev_em_start = (u64)-1;
4169 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
4170 page = list_entry(pages->prev, struct page, lru);
4172 prefetchw(&page->flags);
4173 list_del(&page->lru);
4174 if (add_to_page_cache_lru(page, mapping,
4176 readahead_gfp_mask(mapping))) {
4181 pagepool[nr++] = page;
4182 if (nr < ARRAY_SIZE(pagepool))
4184 __extent_readpages(tree, pagepool, nr, get_extent, &em_cached,
4185 &bio, 0, &bio_flags, &prev_em_start);
4189 __extent_readpages(tree, pagepool, nr, get_extent, &em_cached,
4190 &bio, 0, &bio_flags, &prev_em_start);
4193 free_extent_map(em_cached);
4195 BUG_ON(!list_empty(pages));
4197 return submit_one_bio(bio, 0, bio_flags);
4202 * basic invalidatepage code, this waits on any locked or writeback
4203 * ranges corresponding to the page, and then deletes any extent state
4204 * records from the tree
4206 int extent_invalidatepage(struct extent_io_tree *tree,
4207 struct page *page, unsigned long offset)
4209 struct extent_state *cached_state = NULL;
4210 u64 start = page_offset(page);
4211 u64 end = start + PAGE_SIZE - 1;
4212 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
4214 start += ALIGN(offset, blocksize);
4218 lock_extent_bits(tree, start, end, &cached_state);
4219 wait_on_page_writeback(page);
4220 clear_extent_bit(tree, start, end,
4221 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4222 EXTENT_DO_ACCOUNTING,
4223 1, 1, &cached_state, GFP_NOFS);
4228 * a helper for releasepage, this tests for areas of the page that
4229 * are locked or under IO and drops the related state bits if it is safe
4232 static int try_release_extent_state(struct extent_map_tree *map,
4233 struct extent_io_tree *tree,
4234 struct page *page, gfp_t mask)
4236 u64 start = page_offset(page);
4237 u64 end = start + PAGE_SIZE - 1;
4240 if (test_range_bit(tree, start, end,
4241 EXTENT_IOBITS, 0, NULL))
4245 * at this point we can safely clear everything except the
4246 * locked bit and the nodatasum bit
4248 ret = clear_extent_bit(tree, start, end,
4249 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
4252 /* if clear_extent_bit failed for enomem reasons,
4253 * we can't allow the release to continue.
4264 * a helper for releasepage. As long as there are no locked extents
4265 * in the range corresponding to the page, both state records and extent
4266 * map records are removed
4268 int try_release_extent_mapping(struct extent_map_tree *map,
4269 struct extent_io_tree *tree, struct page *page,
4272 struct extent_map *em;
4273 u64 start = page_offset(page);
4274 u64 end = start + PAGE_SIZE - 1;
4276 if (gfpflags_allow_blocking(mask) &&
4277 page->mapping->host->i_size > SZ_16M) {
4279 while (start <= end) {
4280 len = end - start + 1;
4281 write_lock(&map->lock);
4282 em = lookup_extent_mapping(map, start, len);
4284 write_unlock(&map->lock);
4287 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
4288 em->start != start) {
4289 write_unlock(&map->lock);
4290 free_extent_map(em);
4293 if (!test_range_bit(tree, em->start,
4294 extent_map_end(em) - 1,
4295 EXTENT_LOCKED | EXTENT_WRITEBACK,
4297 remove_extent_mapping(map, em);
4298 /* once for the rb tree */
4299 free_extent_map(em);
4301 start = extent_map_end(em);
4302 write_unlock(&map->lock);
4305 free_extent_map(em);
4308 return try_release_extent_state(map, tree, page, mask);
4312 * helper function for fiemap, which doesn't want to see any holes.
4313 * This maps until we find something past 'last'
4315 static struct extent_map *get_extent_skip_holes(struct inode *inode,
4318 get_extent_t *get_extent)
4320 u64 sectorsize = btrfs_inode_sectorsize(inode);
4321 struct extent_map *em;
4328 len = last - offset;
4331 len = ALIGN(len, sectorsize);
4332 em = get_extent(BTRFS_I(inode), NULL, 0, offset, len, 0);
4333 if (IS_ERR_OR_NULL(em))
4336 /* if this isn't a hole return it */
4337 if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) &&
4338 em->block_start != EXTENT_MAP_HOLE) {
4342 /* this is a hole, advance to the next extent */
4343 offset = extent_map_end(em);
4344 free_extent_map(em);
4352 * To cache previous fiemap extent
4354 * Will be used for merging fiemap extent
4356 struct fiemap_cache {
4365 * Helper to submit fiemap extent.
4367 * Will try to merge current fiemap extent specified by @offset, @phys,
4368 * @len and @flags with cached one.
4369 * And only when we fails to merge, cached one will be submitted as
4372 * Return value is the same as fiemap_fill_next_extent().
4374 static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
4375 struct fiemap_cache *cache,
4376 u64 offset, u64 phys, u64 len, u32 flags)
4384 * Sanity check, extent_fiemap() should have ensured that new
4385 * fiemap extent won't overlap with cahced one.
4388 * NOTE: Physical address can overlap, due to compression
4390 if (cache->offset + cache->len > offset) {
4396 * Only merges fiemap extents if
4397 * 1) Their logical addresses are continuous
4399 * 2) Their physical addresses are continuous
4400 * So truly compressed (physical size smaller than logical size)
4401 * extents won't get merged with each other
4403 * 3) Share same flags except FIEMAP_EXTENT_LAST
4404 * So regular extent won't get merged with prealloc extent
4406 if (cache->offset + cache->len == offset &&
4407 cache->phys + cache->len == phys &&
4408 (cache->flags & ~FIEMAP_EXTENT_LAST) ==
4409 (flags & ~FIEMAP_EXTENT_LAST)) {
4411 cache->flags |= flags;
4412 goto try_submit_last;
4415 /* Not mergeable, need to submit cached one */
4416 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
4417 cache->len, cache->flags);
4418 cache->cached = false;
4422 cache->cached = true;
4423 cache->offset = offset;
4426 cache->flags = flags;
4428 if (cache->flags & FIEMAP_EXTENT_LAST) {
4429 ret = fiemap_fill_next_extent(fieinfo, cache->offset,
4430 cache->phys, cache->len, cache->flags);
4431 cache->cached = false;
4437 * Emit last fiemap cache
4439 * The last fiemap cache may still be cached in the following case:
4441 * |<- Fiemap range ->|
4442 * |<------------ First extent ----------->|
4444 * In this case, the first extent range will be cached but not emitted.
4445 * So we must emit it before ending extent_fiemap().
4447 static int emit_last_fiemap_cache(struct btrfs_fs_info *fs_info,
4448 struct fiemap_extent_info *fieinfo,
4449 struct fiemap_cache *cache)
4456 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
4457 cache->len, cache->flags);
4458 cache->cached = false;
4464 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4465 __u64 start, __u64 len, get_extent_t *get_extent)
4469 u64 max = start + len;
4473 u64 last_for_get_extent = 0;
4475 u64 isize = i_size_read(inode);
4476 struct btrfs_key found_key;
4477 struct extent_map *em = NULL;
4478 struct extent_state *cached_state = NULL;
4479 struct btrfs_path *path;
4480 struct btrfs_root *root = BTRFS_I(inode)->root;
4481 struct fiemap_cache cache = { 0 };
4490 path = btrfs_alloc_path();
4493 path->leave_spinning = 1;
4495 start = round_down(start, btrfs_inode_sectorsize(inode));
4496 len = round_up(max, btrfs_inode_sectorsize(inode)) - start;
4499 * lookup the last file extent. We're not using i_size here
4500 * because there might be preallocation past i_size
4502 ret = btrfs_lookup_file_extent(NULL, root, path,
4503 btrfs_ino(BTRFS_I(inode)), -1, 0);
4505 btrfs_free_path(path);
4514 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
4515 found_type = found_key.type;
4517 /* No extents, but there might be delalloc bits */
4518 if (found_key.objectid != btrfs_ino(BTRFS_I(inode)) ||
4519 found_type != BTRFS_EXTENT_DATA_KEY) {
4520 /* have to trust i_size as the end */
4522 last_for_get_extent = isize;
4525 * remember the start of the last extent. There are a
4526 * bunch of different factors that go into the length of the
4527 * extent, so its much less complex to remember where it started
4529 last = found_key.offset;
4530 last_for_get_extent = last + 1;
4532 btrfs_release_path(path);
4535 * we might have some extents allocated but more delalloc past those
4536 * extents. so, we trust isize unless the start of the last extent is
4541 last_for_get_extent = isize;
4544 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4547 em = get_extent_skip_holes(inode, start, last_for_get_extent,
4557 u64 offset_in_extent = 0;
4559 /* break if the extent we found is outside the range */
4560 if (em->start >= max || extent_map_end(em) < off)
4564 * get_extent may return an extent that starts before our
4565 * requested range. We have to make sure the ranges
4566 * we return to fiemap always move forward and don't
4567 * overlap, so adjust the offsets here
4569 em_start = max(em->start, off);
4572 * record the offset from the start of the extent
4573 * for adjusting the disk offset below. Only do this if the
4574 * extent isn't compressed since our in ram offset may be past
4575 * what we have actually allocated on disk.
4577 if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4578 offset_in_extent = em_start - em->start;
4579 em_end = extent_map_end(em);
4580 em_len = em_end - em_start;
4585 * bump off for our next call to get_extent
4587 off = extent_map_end(em);
4591 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
4593 flags |= FIEMAP_EXTENT_LAST;
4594 } else if (em->block_start == EXTENT_MAP_INLINE) {
4595 flags |= (FIEMAP_EXTENT_DATA_INLINE |
4596 FIEMAP_EXTENT_NOT_ALIGNED);
4597 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
4598 flags |= (FIEMAP_EXTENT_DELALLOC |
4599 FIEMAP_EXTENT_UNKNOWN);
4600 } else if (fieinfo->fi_extents_max) {
4601 u64 bytenr = em->block_start -
4602 (em->start - em->orig_start);
4604 disko = em->block_start + offset_in_extent;
4607 * As btrfs supports shared space, this information
4608 * can be exported to userspace tools via
4609 * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0
4610 * then we're just getting a count and we can skip the
4613 ret = btrfs_check_shared(root,
4614 btrfs_ino(BTRFS_I(inode)),
4619 flags |= FIEMAP_EXTENT_SHARED;
4622 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4623 flags |= FIEMAP_EXTENT_ENCODED;
4624 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4625 flags |= FIEMAP_EXTENT_UNWRITTEN;
4627 free_extent_map(em);
4629 if ((em_start >= last) || em_len == (u64)-1 ||
4630 (last == (u64)-1 && isize <= em_end)) {
4631 flags |= FIEMAP_EXTENT_LAST;
4635 /* now scan forward to see if this is really the last extent. */
4636 em = get_extent_skip_holes(inode, off, last_for_get_extent,
4643 flags |= FIEMAP_EXTENT_LAST;
4646 ret = emit_fiemap_extent(fieinfo, &cache, em_start, disko,
4656 ret = emit_last_fiemap_cache(root->fs_info, fieinfo, &cache);
4657 free_extent_map(em);
4659 btrfs_free_path(path);
4660 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4661 &cached_state, GFP_NOFS);
4665 static void __free_extent_buffer(struct extent_buffer *eb)
4667 btrfs_leak_debug_del(&eb->leak_list);
4668 kmem_cache_free(extent_buffer_cache, eb);
4671 int extent_buffer_under_io(struct extent_buffer *eb)
4673 return (atomic_read(&eb->io_pages) ||
4674 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4675 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4679 * Helper for releasing extent buffer page.
4681 static void btrfs_release_extent_buffer_page(struct extent_buffer *eb)
4683 unsigned long index;
4685 int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4687 BUG_ON(extent_buffer_under_io(eb));
4689 index = num_extent_pages(eb->start, eb->len);
4695 page = eb->pages[index];
4699 spin_lock(&page->mapping->private_lock);
4701 * We do this since we'll remove the pages after we've
4702 * removed the eb from the radix tree, so we could race
4703 * and have this page now attached to the new eb. So
4704 * only clear page_private if it's still connected to
4707 if (PagePrivate(page) &&
4708 page->private == (unsigned long)eb) {
4709 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4710 BUG_ON(PageDirty(page));
4711 BUG_ON(PageWriteback(page));
4713 * We need to make sure we haven't be attached
4716 ClearPagePrivate(page);
4717 set_page_private(page, 0);
4718 /* One for the page private */
4723 spin_unlock(&page->mapping->private_lock);
4725 /* One for when we allocated the page */
4727 } while (index != 0);
4731 * Helper for releasing the extent buffer.
4733 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4735 btrfs_release_extent_buffer_page(eb);
4736 __free_extent_buffer(eb);
4739 static struct extent_buffer *
4740 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
4743 struct extent_buffer *eb = NULL;
4745 eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL);
4748 eb->fs_info = fs_info;
4750 rwlock_init(&eb->lock);
4751 atomic_set(&eb->write_locks, 0);
4752 atomic_set(&eb->read_locks, 0);
4753 atomic_set(&eb->blocking_readers, 0);
4754 atomic_set(&eb->blocking_writers, 0);
4755 atomic_set(&eb->spinning_readers, 0);
4756 atomic_set(&eb->spinning_writers, 0);
4757 eb->lock_nested = 0;
4758 init_waitqueue_head(&eb->write_lock_wq);
4759 init_waitqueue_head(&eb->read_lock_wq);
4761 btrfs_leak_debug_add(&eb->leak_list, &buffers);
4763 spin_lock_init(&eb->refs_lock);
4764 atomic_set(&eb->refs, 1);
4765 atomic_set(&eb->io_pages, 0);
4768 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4770 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4771 > MAX_INLINE_EXTENT_BUFFER_SIZE);
4772 BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE);
4777 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src)
4781 struct extent_buffer *new;
4782 unsigned long num_pages = num_extent_pages(src->start, src->len);
4784 new = __alloc_extent_buffer(src->fs_info, src->start, src->len);
4788 for (i = 0; i < num_pages; i++) {
4789 p = alloc_page(GFP_NOFS);
4791 btrfs_release_extent_buffer(new);
4794 attach_extent_buffer_page(new, p);
4795 WARN_ON(PageDirty(p));
4798 copy_page(page_address(p), page_address(src->pages[i]));
4801 set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
4802 set_bit(EXTENT_BUFFER_DUMMY, &new->bflags);
4807 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4808 u64 start, unsigned long len)
4810 struct extent_buffer *eb;
4811 unsigned long num_pages;
4814 num_pages = num_extent_pages(start, len);
4816 eb = __alloc_extent_buffer(fs_info, start, len);
4820 for (i = 0; i < num_pages; i++) {
4821 eb->pages[i] = alloc_page(GFP_NOFS);
4825 set_extent_buffer_uptodate(eb);
4826 btrfs_set_header_nritems(eb, 0);
4827 set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4832 __free_page(eb->pages[i - 1]);
4833 __free_extent_buffer(eb);
4837 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4840 return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize);
4843 static void check_buffer_tree_ref(struct extent_buffer *eb)
4846 /* the ref bit is tricky. We have to make sure it is set
4847 * if we have the buffer dirty. Otherwise the
4848 * code to free a buffer can end up dropping a dirty
4851 * Once the ref bit is set, it won't go away while the
4852 * buffer is dirty or in writeback, and it also won't
4853 * go away while we have the reference count on the
4856 * We can't just set the ref bit without bumping the
4857 * ref on the eb because free_extent_buffer might
4858 * see the ref bit and try to clear it. If this happens
4859 * free_extent_buffer might end up dropping our original
4860 * ref by mistake and freeing the page before we are able
4861 * to add one more ref.
4863 * So bump the ref count first, then set the bit. If someone
4864 * beat us to it, drop the ref we added.
4866 refs = atomic_read(&eb->refs);
4867 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4870 spin_lock(&eb->refs_lock);
4871 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4872 atomic_inc(&eb->refs);
4873 spin_unlock(&eb->refs_lock);
4876 static void mark_extent_buffer_accessed(struct extent_buffer *eb,
4877 struct page *accessed)
4879 unsigned long num_pages, i;
4881 check_buffer_tree_ref(eb);
4883 num_pages = num_extent_pages(eb->start, eb->len);
4884 for (i = 0; i < num_pages; i++) {
4885 struct page *p = eb->pages[i];
4888 mark_page_accessed(p);
4892 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
4895 struct extent_buffer *eb;
4898 eb = radix_tree_lookup(&fs_info->buffer_radix,
4899 start >> PAGE_SHIFT);
4900 if (eb && atomic_inc_not_zero(&eb->refs)) {
4903 * Lock our eb's refs_lock to avoid races with
4904 * free_extent_buffer. When we get our eb it might be flagged
4905 * with EXTENT_BUFFER_STALE and another task running
4906 * free_extent_buffer might have seen that flag set,
4907 * eb->refs == 2, that the buffer isn't under IO (dirty and
4908 * writeback flags not set) and it's still in the tree (flag
4909 * EXTENT_BUFFER_TREE_REF set), therefore being in the process
4910 * of decrementing the extent buffer's reference count twice.
4911 * So here we could race and increment the eb's reference count,
4912 * clear its stale flag, mark it as dirty and drop our reference
4913 * before the other task finishes executing free_extent_buffer,
4914 * which would later result in an attempt to free an extent
4915 * buffer that is dirty.
4917 if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) {
4918 spin_lock(&eb->refs_lock);
4919 spin_unlock(&eb->refs_lock);
4921 mark_extent_buffer_accessed(eb, NULL);
4929 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4930 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
4933 struct extent_buffer *eb, *exists = NULL;
4936 eb = find_extent_buffer(fs_info, start);
4939 eb = alloc_dummy_extent_buffer(fs_info, start);
4942 eb->fs_info = fs_info;
4944 ret = radix_tree_preload(GFP_NOFS);
4947 spin_lock(&fs_info->buffer_lock);
4948 ret = radix_tree_insert(&fs_info->buffer_radix,
4949 start >> PAGE_SHIFT, eb);
4950 spin_unlock(&fs_info->buffer_lock);
4951 radix_tree_preload_end();
4952 if (ret == -EEXIST) {
4953 exists = find_extent_buffer(fs_info, start);
4959 check_buffer_tree_ref(eb);
4960 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
4963 * We will free dummy extent buffer's if they come into
4964 * free_extent_buffer with a ref count of 2, but if we are using this we
4965 * want the buffers to stay in memory until we're done with them, so
4966 * bump the ref count again.
4968 atomic_inc(&eb->refs);
4971 btrfs_release_extent_buffer(eb);
4976 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
4979 unsigned long len = fs_info->nodesize;
4980 unsigned long num_pages = num_extent_pages(start, len);
4982 unsigned long index = start >> PAGE_SHIFT;
4983 struct extent_buffer *eb;
4984 struct extent_buffer *exists = NULL;
4986 struct address_space *mapping = fs_info->btree_inode->i_mapping;
4990 if (!IS_ALIGNED(start, fs_info->sectorsize)) {
4991 btrfs_err(fs_info, "bad tree block start %llu", start);
4992 return ERR_PTR(-EINVAL);
4995 eb = find_extent_buffer(fs_info, start);
4999 eb = __alloc_extent_buffer(fs_info, start, len);
5001 return ERR_PTR(-ENOMEM);
5003 for (i = 0; i < num_pages; i++, index++) {
5004 p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL);
5006 exists = ERR_PTR(-ENOMEM);
5010 spin_lock(&mapping->private_lock);
5011 if (PagePrivate(p)) {
5013 * We could have already allocated an eb for this page
5014 * and attached one so lets see if we can get a ref on
5015 * the existing eb, and if we can we know it's good and
5016 * we can just return that one, else we know we can just
5017 * overwrite page->private.
5019 exists = (struct extent_buffer *)p->private;
5020 if (atomic_inc_not_zero(&exists->refs)) {
5021 spin_unlock(&mapping->private_lock);
5024 mark_extent_buffer_accessed(exists, p);
5030 * Do this so attach doesn't complain and we need to
5031 * drop the ref the old guy had.
5033 ClearPagePrivate(p);
5034 WARN_ON(PageDirty(p));
5037 attach_extent_buffer_page(eb, p);
5038 spin_unlock(&mapping->private_lock);
5039 WARN_ON(PageDirty(p));
5041 if (!PageUptodate(p))
5045 * see below about how we avoid a nasty race with release page
5046 * and why we unlock later
5050 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5052 ret = radix_tree_preload(GFP_NOFS);
5054 exists = ERR_PTR(ret);
5058 spin_lock(&fs_info->buffer_lock);
5059 ret = radix_tree_insert(&fs_info->buffer_radix,
5060 start >> PAGE_SHIFT, eb);
5061 spin_unlock(&fs_info->buffer_lock);
5062 radix_tree_preload_end();
5063 if (ret == -EEXIST) {
5064 exists = find_extent_buffer(fs_info, start);
5070 /* add one reference for the tree */
5071 check_buffer_tree_ref(eb);
5072 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
5075 * there is a race where release page may have
5076 * tried to find this extent buffer in the radix
5077 * but failed. It will tell the VM it is safe to
5078 * reclaim the, and it will clear the page private bit.
5079 * We must make sure to set the page private bit properly
5080 * after the extent buffer is in the radix tree so
5081 * it doesn't get lost
5083 SetPageChecked(eb->pages[0]);
5084 for (i = 1; i < num_pages; i++) {
5086 ClearPageChecked(p);
5089 unlock_page(eb->pages[0]);
5093 WARN_ON(!atomic_dec_and_test(&eb->refs));
5094 for (i = 0; i < num_pages; i++) {
5096 unlock_page(eb->pages[i]);
5099 btrfs_release_extent_buffer(eb);
5103 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
5105 struct extent_buffer *eb =
5106 container_of(head, struct extent_buffer, rcu_head);
5108 __free_extent_buffer(eb);
5111 /* Expects to have eb->eb_lock already held */
5112 static int release_extent_buffer(struct extent_buffer *eb)
5114 WARN_ON(atomic_read(&eb->refs) == 0);
5115 if (atomic_dec_and_test(&eb->refs)) {
5116 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
5117 struct btrfs_fs_info *fs_info = eb->fs_info;
5119 spin_unlock(&eb->refs_lock);
5121 spin_lock(&fs_info->buffer_lock);
5122 radix_tree_delete(&fs_info->buffer_radix,
5123 eb->start >> PAGE_SHIFT);
5124 spin_unlock(&fs_info->buffer_lock);
5126 spin_unlock(&eb->refs_lock);
5129 /* Should be safe to release our pages at this point */
5130 btrfs_release_extent_buffer_page(eb);
5131 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5132 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))) {
5133 __free_extent_buffer(eb);
5137 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
5140 spin_unlock(&eb->refs_lock);
5145 void free_extent_buffer(struct extent_buffer *eb)
5153 refs = atomic_read(&eb->refs);
5156 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
5161 spin_lock(&eb->refs_lock);
5162 if (atomic_read(&eb->refs) == 2 &&
5163 test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))
5164 atomic_dec(&eb->refs);
5166 if (atomic_read(&eb->refs) == 2 &&
5167 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
5168 !extent_buffer_under_io(eb) &&
5169 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5170 atomic_dec(&eb->refs);
5173 * I know this is terrible, but it's temporary until we stop tracking
5174 * the uptodate bits and such for the extent buffers.
5176 release_extent_buffer(eb);
5179 void free_extent_buffer_stale(struct extent_buffer *eb)
5184 spin_lock(&eb->refs_lock);
5185 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
5187 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
5188 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5189 atomic_dec(&eb->refs);
5190 release_extent_buffer(eb);
5193 void clear_extent_buffer_dirty(struct extent_buffer *eb)
5196 unsigned long num_pages;
5199 num_pages = num_extent_pages(eb->start, eb->len);
5201 for (i = 0; i < num_pages; i++) {
5202 page = eb->pages[i];
5203 if (!PageDirty(page))
5207 WARN_ON(!PagePrivate(page));
5209 clear_page_dirty_for_io(page);
5210 spin_lock_irq(&page->mapping->tree_lock);
5211 if (!PageDirty(page)) {
5212 radix_tree_tag_clear(&page->mapping->page_tree,
5214 PAGECACHE_TAG_DIRTY);
5216 spin_unlock_irq(&page->mapping->tree_lock);
5217 ClearPageError(page);
5220 WARN_ON(atomic_read(&eb->refs) == 0);
5223 int set_extent_buffer_dirty(struct extent_buffer *eb)
5226 unsigned long num_pages;
5229 check_buffer_tree_ref(eb);
5231 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
5233 num_pages = num_extent_pages(eb->start, eb->len);
5234 WARN_ON(atomic_read(&eb->refs) == 0);
5235 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
5237 for (i = 0; i < num_pages; i++)
5238 set_page_dirty(eb->pages[i]);
5242 void clear_extent_buffer_uptodate(struct extent_buffer *eb)
5246 unsigned long num_pages;
5248 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5249 num_pages = num_extent_pages(eb->start, eb->len);
5250 for (i = 0; i < num_pages; i++) {
5251 page = eb->pages[i];
5253 ClearPageUptodate(page);
5257 void set_extent_buffer_uptodate(struct extent_buffer *eb)
5261 unsigned long num_pages;
5263 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5264 num_pages = num_extent_pages(eb->start, eb->len);
5265 for (i = 0; i < num_pages; i++) {
5266 page = eb->pages[i];
5267 SetPageUptodate(page);
5271 int extent_buffer_uptodate(struct extent_buffer *eb)
5273 return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5276 int read_extent_buffer_pages(struct extent_io_tree *tree,
5277 struct extent_buffer *eb, int wait,
5278 get_extent_t *get_extent, int mirror_num)
5284 int locked_pages = 0;
5285 int all_uptodate = 1;
5286 unsigned long num_pages;
5287 unsigned long num_reads = 0;
5288 struct bio *bio = NULL;
5289 unsigned long bio_flags = 0;
5291 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
5294 num_pages = num_extent_pages(eb->start, eb->len);
5295 for (i = 0; i < num_pages; i++) {
5296 page = eb->pages[i];
5297 if (wait == WAIT_NONE) {
5298 if (!trylock_page(page))
5306 * We need to firstly lock all pages to make sure that
5307 * the uptodate bit of our pages won't be affected by
5308 * clear_extent_buffer_uptodate().
5310 for (i = 0; i < num_pages; i++) {
5311 page = eb->pages[i];
5312 if (!PageUptodate(page)) {
5319 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5323 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
5324 eb->read_mirror = 0;
5325 atomic_set(&eb->io_pages, num_reads);
5326 for (i = 0; i < num_pages; i++) {
5327 page = eb->pages[i];
5329 if (!PageUptodate(page)) {
5331 atomic_dec(&eb->io_pages);
5336 ClearPageError(page);
5337 err = __extent_read_full_page(tree, page,
5339 mirror_num, &bio_flags,
5344 * We use &bio in above __extent_read_full_page,
5345 * so we ensure that if it returns error, the
5346 * current page fails to add itself to bio and
5347 * it's been unlocked.
5349 * We must dec io_pages by ourselves.
5351 atomic_dec(&eb->io_pages);
5359 err = submit_one_bio(bio, mirror_num, bio_flags);
5364 if (ret || wait != WAIT_COMPLETE)
5367 for (i = 0; i < num_pages; i++) {
5368 page = eb->pages[i];
5369 wait_on_page_locked(page);
5370 if (!PageUptodate(page))
5377 while (locked_pages > 0) {
5379 page = eb->pages[locked_pages];
5385 void read_extent_buffer(const struct extent_buffer *eb, void *dstv,
5386 unsigned long start, unsigned long len)
5392 char *dst = (char *)dstv;
5393 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5394 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5396 if (start + len > eb->len) {
5397 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n",
5398 eb->start, eb->len, start, len);
5399 memset(dst, 0, len);
5403 offset = (start_offset + start) & (PAGE_SIZE - 1);
5406 page = eb->pages[i];
5408 cur = min(len, (PAGE_SIZE - offset));
5409 kaddr = page_address(page);
5410 memcpy(dst, kaddr + offset, cur);
5419 int read_extent_buffer_to_user(const struct extent_buffer *eb,
5421 unsigned long start, unsigned long len)
5427 char __user *dst = (char __user *)dstv;
5428 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5429 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5432 WARN_ON(start > eb->len);
5433 WARN_ON(start + len > eb->start + eb->len);
5435 offset = (start_offset + start) & (PAGE_SIZE - 1);
5438 page = eb->pages[i];
5440 cur = min(len, (PAGE_SIZE - offset));
5441 kaddr = page_address(page);
5442 if (copy_to_user(dst, kaddr + offset, cur)) {
5457 * return 0 if the item is found within a page.
5458 * return 1 if the item spans two pages.
5459 * return -EINVAL otherwise.
5461 int map_private_extent_buffer(const struct extent_buffer *eb,
5462 unsigned long start, unsigned long min_len,
5463 char **map, unsigned long *map_start,
5464 unsigned long *map_len)
5466 size_t offset = start & (PAGE_SIZE - 1);
5469 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5470 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5471 unsigned long end_i = (start_offset + start + min_len - 1) >>
5474 if (start + min_len > eb->len) {
5475 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n",
5476 eb->start, eb->len, start, min_len);
5484 offset = start_offset;
5488 *map_start = ((u64)i << PAGE_SHIFT) - start_offset;
5492 kaddr = page_address(p);
5493 *map = kaddr + offset;
5494 *map_len = PAGE_SIZE - offset;
5498 int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,
5499 unsigned long start, unsigned long len)
5505 char *ptr = (char *)ptrv;
5506 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5507 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5510 WARN_ON(start > eb->len);
5511 WARN_ON(start + len > eb->start + eb->len);
5513 offset = (start_offset + start) & (PAGE_SIZE - 1);
5516 page = eb->pages[i];
5518 cur = min(len, (PAGE_SIZE - offset));
5520 kaddr = page_address(page);
5521 ret = memcmp(ptr, kaddr + offset, cur);
5533 void write_extent_buffer_chunk_tree_uuid(struct extent_buffer *eb,
5538 WARN_ON(!PageUptodate(eb->pages[0]));
5539 kaddr = page_address(eb->pages[0]);
5540 memcpy(kaddr + offsetof(struct btrfs_header, chunk_tree_uuid), srcv,
5544 void write_extent_buffer_fsid(struct extent_buffer *eb, const void *srcv)
5548 WARN_ON(!PageUptodate(eb->pages[0]));
5549 kaddr = page_address(eb->pages[0]);
5550 memcpy(kaddr + offsetof(struct btrfs_header, fsid), srcv,
5554 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
5555 unsigned long start, unsigned long len)
5561 char *src = (char *)srcv;
5562 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5563 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5565 WARN_ON(start > eb->len);
5566 WARN_ON(start + len > eb->start + eb->len);
5568 offset = (start_offset + start) & (PAGE_SIZE - 1);
5571 page = eb->pages[i];
5572 WARN_ON(!PageUptodate(page));
5574 cur = min(len, PAGE_SIZE - offset);
5575 kaddr = page_address(page);
5576 memcpy(kaddr + offset, src, cur);
5585 void memzero_extent_buffer(struct extent_buffer *eb, unsigned long start,
5592 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5593 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5595 WARN_ON(start > eb->len);
5596 WARN_ON(start + len > eb->start + eb->len);
5598 offset = (start_offset + start) & (PAGE_SIZE - 1);
5601 page = eb->pages[i];
5602 WARN_ON(!PageUptodate(page));
5604 cur = min(len, PAGE_SIZE - offset);
5605 kaddr = page_address(page);
5606 memset(kaddr + offset, 0, cur);
5614 void copy_extent_buffer_full(struct extent_buffer *dst,
5615 struct extent_buffer *src)
5620 ASSERT(dst->len == src->len);
5622 num_pages = num_extent_pages(dst->start, dst->len);
5623 for (i = 0; i < num_pages; i++)
5624 copy_page(page_address(dst->pages[i]),
5625 page_address(src->pages[i]));
5628 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
5629 unsigned long dst_offset, unsigned long src_offset,
5632 u64 dst_len = dst->len;
5637 size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1);
5638 unsigned long i = (start_offset + dst_offset) >> PAGE_SHIFT;
5640 WARN_ON(src->len != dst_len);
5642 offset = (start_offset + dst_offset) &
5646 page = dst->pages[i];
5647 WARN_ON(!PageUptodate(page));
5649 cur = min(len, (unsigned long)(PAGE_SIZE - offset));
5651 kaddr = page_address(page);
5652 read_extent_buffer(src, kaddr + offset, src_offset, cur);
5661 void le_bitmap_set(u8 *map, unsigned int start, int len)
5663 u8 *p = map + BIT_BYTE(start);
5664 const unsigned int size = start + len;
5665 int bits_to_set = BITS_PER_BYTE - (start % BITS_PER_BYTE);
5666 u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(start);
5668 while (len - bits_to_set >= 0) {
5671 bits_to_set = BITS_PER_BYTE;
5676 mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
5681 void le_bitmap_clear(u8 *map, unsigned int start, int len)
5683 u8 *p = map + BIT_BYTE(start);
5684 const unsigned int size = start + len;
5685 int bits_to_clear = BITS_PER_BYTE - (start % BITS_PER_BYTE);
5686 u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(start);
5688 while (len - bits_to_clear >= 0) {
5689 *p &= ~mask_to_clear;
5690 len -= bits_to_clear;
5691 bits_to_clear = BITS_PER_BYTE;
5696 mask_to_clear &= BITMAP_LAST_BYTE_MASK(size);
5697 *p &= ~mask_to_clear;
5702 * eb_bitmap_offset() - calculate the page and offset of the byte containing the
5704 * @eb: the extent buffer
5705 * @start: offset of the bitmap item in the extent buffer
5707 * @page_index: return index of the page in the extent buffer that contains the
5709 * @page_offset: return offset into the page given by page_index
5711 * This helper hides the ugliness of finding the byte in an extent buffer which
5712 * contains a given bit.
5714 static inline void eb_bitmap_offset(struct extent_buffer *eb,
5715 unsigned long start, unsigned long nr,
5716 unsigned long *page_index,
5717 size_t *page_offset)
5719 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5720 size_t byte_offset = BIT_BYTE(nr);
5724 * The byte we want is the offset of the extent buffer + the offset of
5725 * the bitmap item in the extent buffer + the offset of the byte in the
5728 offset = start_offset + start + byte_offset;
5730 *page_index = offset >> PAGE_SHIFT;
5731 *page_offset = offset & (PAGE_SIZE - 1);
5735 * extent_buffer_test_bit - determine whether a bit in a bitmap item is set
5736 * @eb: the extent buffer
5737 * @start: offset of the bitmap item in the extent buffer
5738 * @nr: bit number to test
5740 int extent_buffer_test_bit(struct extent_buffer *eb, unsigned long start,
5748 eb_bitmap_offset(eb, start, nr, &i, &offset);
5749 page = eb->pages[i];
5750 WARN_ON(!PageUptodate(page));
5751 kaddr = page_address(page);
5752 return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1)));
5756 * extent_buffer_bitmap_set - set an area of a bitmap
5757 * @eb: the extent buffer
5758 * @start: offset of the bitmap item in the extent buffer
5759 * @pos: bit number of the first bit
5760 * @len: number of bits to set
5762 void extent_buffer_bitmap_set(struct extent_buffer *eb, unsigned long start,
5763 unsigned long pos, unsigned long len)
5769 const unsigned int size = pos + len;
5770 int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5771 u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos);
5773 eb_bitmap_offset(eb, start, pos, &i, &offset);
5774 page = eb->pages[i];
5775 WARN_ON(!PageUptodate(page));
5776 kaddr = page_address(page);
5778 while (len >= bits_to_set) {
5779 kaddr[offset] |= mask_to_set;
5781 bits_to_set = BITS_PER_BYTE;
5783 if (++offset >= PAGE_SIZE && len > 0) {
5785 page = eb->pages[++i];
5786 WARN_ON(!PageUptodate(page));
5787 kaddr = page_address(page);
5791 mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
5792 kaddr[offset] |= mask_to_set;
5798 * extent_buffer_bitmap_clear - clear an area of a bitmap
5799 * @eb: the extent buffer
5800 * @start: offset of the bitmap item in the extent buffer
5801 * @pos: bit number of the first bit
5802 * @len: number of bits to clear
5804 void extent_buffer_bitmap_clear(struct extent_buffer *eb, unsigned long start,
5805 unsigned long pos, unsigned long len)
5811 const unsigned int size = pos + len;
5812 int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5813 u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos);
5815 eb_bitmap_offset(eb, start, pos, &i, &offset);
5816 page = eb->pages[i];
5817 WARN_ON(!PageUptodate(page));
5818 kaddr = page_address(page);
5820 while (len >= bits_to_clear) {
5821 kaddr[offset] &= ~mask_to_clear;
5822 len -= bits_to_clear;
5823 bits_to_clear = BITS_PER_BYTE;
5825 if (++offset >= PAGE_SIZE && len > 0) {
5827 page = eb->pages[++i];
5828 WARN_ON(!PageUptodate(page));
5829 kaddr = page_address(page);
5833 mask_to_clear &= BITMAP_LAST_BYTE_MASK(size);
5834 kaddr[offset] &= ~mask_to_clear;
5838 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
5840 unsigned long distance = (src > dst) ? src - dst : dst - src;
5841 return distance < len;
5844 static void copy_pages(struct page *dst_page, struct page *src_page,
5845 unsigned long dst_off, unsigned long src_off,
5848 char *dst_kaddr = page_address(dst_page);
5850 int must_memmove = 0;
5852 if (dst_page != src_page) {
5853 src_kaddr = page_address(src_page);
5855 src_kaddr = dst_kaddr;
5856 if (areas_overlap(src_off, dst_off, len))
5861 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
5863 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
5866 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5867 unsigned long src_offset, unsigned long len)
5869 struct btrfs_fs_info *fs_info = dst->fs_info;
5871 size_t dst_off_in_page;
5872 size_t src_off_in_page;
5873 size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1);
5874 unsigned long dst_i;
5875 unsigned long src_i;
5877 if (src_offset + len > dst->len) {
5879 "memmove bogus src_offset %lu move len %lu dst len %lu",
5880 src_offset, len, dst->len);
5883 if (dst_offset + len > dst->len) {
5885 "memmove bogus dst_offset %lu move len %lu dst len %lu",
5886 dst_offset, len, dst->len);
5891 dst_off_in_page = (start_offset + dst_offset) &
5893 src_off_in_page = (start_offset + src_offset) &
5896 dst_i = (start_offset + dst_offset) >> PAGE_SHIFT;
5897 src_i = (start_offset + src_offset) >> PAGE_SHIFT;
5899 cur = min(len, (unsigned long)(PAGE_SIZE -
5901 cur = min_t(unsigned long, cur,
5902 (unsigned long)(PAGE_SIZE - dst_off_in_page));
5904 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5905 dst_off_in_page, src_off_in_page, cur);
5913 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5914 unsigned long src_offset, unsigned long len)
5916 struct btrfs_fs_info *fs_info = dst->fs_info;
5918 size_t dst_off_in_page;
5919 size_t src_off_in_page;
5920 unsigned long dst_end = dst_offset + len - 1;
5921 unsigned long src_end = src_offset + len - 1;
5922 size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1);
5923 unsigned long dst_i;
5924 unsigned long src_i;
5926 if (src_offset + len > dst->len) {
5928 "memmove bogus src_offset %lu move len %lu len %lu",
5929 src_offset, len, dst->len);
5932 if (dst_offset + len > dst->len) {
5934 "memmove bogus dst_offset %lu move len %lu len %lu",
5935 dst_offset, len, dst->len);
5938 if (dst_offset < src_offset) {
5939 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
5943 dst_i = (start_offset + dst_end) >> PAGE_SHIFT;
5944 src_i = (start_offset + src_end) >> PAGE_SHIFT;
5946 dst_off_in_page = (start_offset + dst_end) &
5948 src_off_in_page = (start_offset + src_end) &
5951 cur = min_t(unsigned long, len, src_off_in_page + 1);
5952 cur = min(cur, dst_off_in_page + 1);
5953 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5954 dst_off_in_page - cur + 1,
5955 src_off_in_page - cur + 1, cur);
5963 int try_release_extent_buffer(struct page *page)
5965 struct extent_buffer *eb;
5968 * We need to make sure nobody is attaching this page to an eb right
5971 spin_lock(&page->mapping->private_lock);
5972 if (!PagePrivate(page)) {
5973 spin_unlock(&page->mapping->private_lock);
5977 eb = (struct extent_buffer *)page->private;
5981 * This is a little awful but should be ok, we need to make sure that
5982 * the eb doesn't disappear out from under us while we're looking at
5985 spin_lock(&eb->refs_lock);
5986 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
5987 spin_unlock(&eb->refs_lock);
5988 spin_unlock(&page->mapping->private_lock);
5991 spin_unlock(&page->mapping->private_lock);
5994 * If tree ref isn't set then we know the ref on this eb is a real ref,
5995 * so just return, this page will likely be freed soon anyway.
5997 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
5998 spin_unlock(&eb->refs_lock);
6002 return release_extent_buffer(eb);