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 int __clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
585 unsigned bits, int wake, int delete,
586 struct extent_state **cached_state,
587 gfp_t mask, struct extent_changeset *changeset)
589 struct extent_state *state;
590 struct extent_state *cached;
591 struct extent_state *prealloc = NULL;
592 struct rb_node *node;
597 btrfs_debug_check_extent_io_range(tree, start, end);
599 if (bits & EXTENT_DELALLOC)
600 bits |= EXTENT_NORESERVE;
603 bits |= ~EXTENT_CTLBITS;
604 bits |= EXTENT_FIRST_DELALLOC;
606 if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY))
609 if (!prealloc && gfpflags_allow_blocking(mask)) {
611 * Don't care for allocation failure here because we might end
612 * up not needing the pre-allocated extent state at all, which
613 * is the case if we only have in the tree extent states that
614 * cover our input range and don't cover too any other range.
615 * If we end up needing a new extent state we allocate it later.
617 prealloc = alloc_extent_state(mask);
620 spin_lock(&tree->lock);
622 cached = *cached_state;
625 *cached_state = NULL;
629 if (cached && extent_state_in_tree(cached) &&
630 cached->start <= start && cached->end > start) {
632 refcount_dec(&cached->refs);
637 free_extent_state(cached);
640 * this search will find the extents that end after
643 node = tree_search(tree, start);
646 state = rb_entry(node, struct extent_state, rb_node);
648 if (state->start > end)
650 WARN_ON(state->end < start);
651 last_end = state->end;
653 /* the state doesn't have the wanted bits, go ahead */
654 if (!(state->state & bits)) {
655 state = next_state(state);
660 * | ---- desired range ---- |
662 * | ------------- state -------------- |
664 * We need to split the extent we found, and may flip
665 * bits on second half.
667 * If the extent we found extends past our range, we
668 * just split and search again. It'll get split again
669 * the next time though.
671 * If the extent we found is inside our range, we clear
672 * the desired bit on it.
675 if (state->start < start) {
676 prealloc = alloc_extent_state_atomic(prealloc);
678 err = split_state(tree, state, prealloc, start);
680 extent_io_tree_panic(tree, err);
685 if (state->end <= end) {
686 state = clear_state_bit(tree, state, &bits, wake,
693 * | ---- desired range ---- |
695 * We need to split the extent, and clear the bit
698 if (state->start <= end && state->end > end) {
699 prealloc = alloc_extent_state_atomic(prealloc);
701 err = split_state(tree, state, prealloc, end + 1);
703 extent_io_tree_panic(tree, err);
708 clear_state_bit(tree, prealloc, &bits, wake, changeset);
714 state = clear_state_bit(tree, state, &bits, wake, changeset);
716 if (last_end == (u64)-1)
718 start = last_end + 1;
719 if (start <= end && state && !need_resched())
725 spin_unlock(&tree->lock);
726 if (gfpflags_allow_blocking(mask))
731 spin_unlock(&tree->lock);
733 free_extent_state(prealloc);
739 static void wait_on_state(struct extent_io_tree *tree,
740 struct extent_state *state)
741 __releases(tree->lock)
742 __acquires(tree->lock)
745 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
746 spin_unlock(&tree->lock);
748 spin_lock(&tree->lock);
749 finish_wait(&state->wq, &wait);
753 * waits for one or more bits to clear on a range in the state tree.
754 * The range [start, end] is inclusive.
755 * The tree lock is taken by this function
757 static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
760 struct extent_state *state;
761 struct rb_node *node;
763 btrfs_debug_check_extent_io_range(tree, start, end);
765 spin_lock(&tree->lock);
769 * this search will find all the extents that end after
772 node = tree_search(tree, start);
777 state = rb_entry(node, struct extent_state, rb_node);
779 if (state->start > end)
782 if (state->state & bits) {
783 start = state->start;
784 refcount_inc(&state->refs);
785 wait_on_state(tree, state);
786 free_extent_state(state);
789 start = state->end + 1;
794 if (!cond_resched_lock(&tree->lock)) {
795 node = rb_next(node);
800 spin_unlock(&tree->lock);
803 static void set_state_bits(struct extent_io_tree *tree,
804 struct extent_state *state,
805 unsigned *bits, struct extent_changeset *changeset)
807 unsigned bits_to_set = *bits & ~EXTENT_CTLBITS;
809 set_state_cb(tree, state, bits);
810 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
811 u64 range = state->end - state->start + 1;
812 tree->dirty_bytes += range;
814 add_extent_changeset(state, bits_to_set, changeset, 1);
815 state->state |= bits_to_set;
818 static void cache_state_if_flags(struct extent_state *state,
819 struct extent_state **cached_ptr,
822 if (cached_ptr && !(*cached_ptr)) {
823 if (!flags || (state->state & flags)) {
825 refcount_inc(&state->refs);
830 static void cache_state(struct extent_state *state,
831 struct extent_state **cached_ptr)
833 return cache_state_if_flags(state, cached_ptr,
834 EXTENT_IOBITS | EXTENT_BOUNDARY);
838 * set some bits on a range in the tree. This may require allocations or
839 * sleeping, so the gfp mask is used to indicate what is allowed.
841 * If any of the exclusive bits are set, this will fail with -EEXIST if some
842 * part of the range already has the desired bits set. The start of the
843 * existing range is returned in failed_start in this case.
845 * [start, end] is inclusive This takes the tree lock.
848 static int __must_check
849 __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
850 unsigned bits, unsigned exclusive_bits,
851 u64 *failed_start, struct extent_state **cached_state,
852 gfp_t mask, struct extent_changeset *changeset)
854 struct extent_state *state;
855 struct extent_state *prealloc = NULL;
856 struct rb_node *node;
858 struct rb_node *parent;
863 btrfs_debug_check_extent_io_range(tree, start, end);
865 bits |= EXTENT_FIRST_DELALLOC;
867 if (!prealloc && gfpflags_allow_blocking(mask)) {
869 * Don't care for allocation failure here because we might end
870 * up not needing the pre-allocated extent state at all, which
871 * is the case if we only have in the tree extent states that
872 * cover our input range and don't cover too any other range.
873 * If we end up needing a new extent state we allocate it later.
875 prealloc = alloc_extent_state(mask);
878 spin_lock(&tree->lock);
879 if (cached_state && *cached_state) {
880 state = *cached_state;
881 if (state->start <= start && state->end > start &&
882 extent_state_in_tree(state)) {
883 node = &state->rb_node;
888 * this search will find all the extents that end after
891 node = tree_search_for_insert(tree, start, &p, &parent);
893 prealloc = alloc_extent_state_atomic(prealloc);
895 err = insert_state(tree, prealloc, start, end,
896 &p, &parent, &bits, changeset);
898 extent_io_tree_panic(tree, err);
900 cache_state(prealloc, cached_state);
904 state = rb_entry(node, struct extent_state, rb_node);
906 last_start = state->start;
907 last_end = state->end;
910 * | ---- desired range ---- |
913 * Just lock what we found and keep going
915 if (state->start == start && state->end <= end) {
916 if (state->state & exclusive_bits) {
917 *failed_start = state->start;
922 set_state_bits(tree, state, &bits, changeset);
923 cache_state(state, cached_state);
924 merge_state(tree, state);
925 if (last_end == (u64)-1)
927 start = last_end + 1;
928 state = next_state(state);
929 if (start < end && state && state->start == start &&
936 * | ---- desired range ---- |
939 * | ------------- state -------------- |
941 * We need to split the extent we found, and may flip bits on
944 * If the extent we found extends past our
945 * range, we just split and search again. It'll get split
946 * again the next time though.
948 * If the extent we found is inside our range, we set the
951 if (state->start < start) {
952 if (state->state & exclusive_bits) {
953 *failed_start = start;
958 prealloc = alloc_extent_state_atomic(prealloc);
960 err = split_state(tree, state, prealloc, start);
962 extent_io_tree_panic(tree, err);
967 if (state->end <= end) {
968 set_state_bits(tree, state, &bits, changeset);
969 cache_state(state, cached_state);
970 merge_state(tree, state);
971 if (last_end == (u64)-1)
973 start = last_end + 1;
974 state = next_state(state);
975 if (start < end && state && state->start == start &&
982 * | ---- desired range ---- |
983 * | state | or | state |
985 * There's a hole, we need to insert something in it and
986 * ignore the extent we found.
988 if (state->start > start) {
990 if (end < last_start)
993 this_end = last_start - 1;
995 prealloc = alloc_extent_state_atomic(prealloc);
999 * Avoid to free 'prealloc' if it can be merged with
1002 err = insert_state(tree, prealloc, start, this_end,
1003 NULL, NULL, &bits, changeset);
1005 extent_io_tree_panic(tree, err);
1007 cache_state(prealloc, cached_state);
1009 start = this_end + 1;
1013 * | ---- desired range ---- |
1015 * We need to split the extent, and set the bit
1018 if (state->start <= end && state->end > end) {
1019 if (state->state & exclusive_bits) {
1020 *failed_start = start;
1025 prealloc = alloc_extent_state_atomic(prealloc);
1027 err = split_state(tree, state, prealloc, end + 1);
1029 extent_io_tree_panic(tree, err);
1031 set_state_bits(tree, prealloc, &bits, changeset);
1032 cache_state(prealloc, cached_state);
1033 merge_state(tree, prealloc);
1041 spin_unlock(&tree->lock);
1042 if (gfpflags_allow_blocking(mask))
1047 spin_unlock(&tree->lock);
1049 free_extent_state(prealloc);
1055 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1056 unsigned bits, u64 * failed_start,
1057 struct extent_state **cached_state, gfp_t mask)
1059 return __set_extent_bit(tree, start, end, bits, 0, failed_start,
1060 cached_state, mask, NULL);
1065 * convert_extent_bit - convert all bits in a given range from one bit to
1067 * @tree: the io tree to search
1068 * @start: the start offset in bytes
1069 * @end: the end offset in bytes (inclusive)
1070 * @bits: the bits to set in this range
1071 * @clear_bits: the bits to clear in this range
1072 * @cached_state: state that we're going to cache
1074 * This will go through and set bits for the given range. If any states exist
1075 * already in this range they are set with the given bit and cleared of the
1076 * clear_bits. This is only meant to be used by things that are mergeable, ie
1077 * converting from say DELALLOC to DIRTY. This is not meant to be used with
1078 * boundary bits like LOCK.
1080 * All allocations are done with GFP_NOFS.
1082 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1083 unsigned bits, unsigned clear_bits,
1084 struct extent_state **cached_state)
1086 struct extent_state *state;
1087 struct extent_state *prealloc = NULL;
1088 struct rb_node *node;
1090 struct rb_node *parent;
1094 bool first_iteration = true;
1096 btrfs_debug_check_extent_io_range(tree, start, end);
1101 * Best effort, don't worry if extent state allocation fails
1102 * here for the first iteration. We might have a cached state
1103 * that matches exactly the target range, in which case no
1104 * extent state allocations are needed. We'll only know this
1105 * after locking the tree.
1107 prealloc = alloc_extent_state(GFP_NOFS);
1108 if (!prealloc && !first_iteration)
1112 spin_lock(&tree->lock);
1113 if (cached_state && *cached_state) {
1114 state = *cached_state;
1115 if (state->start <= start && state->end > start &&
1116 extent_state_in_tree(state)) {
1117 node = &state->rb_node;
1123 * this search will find all the extents that end after
1126 node = tree_search_for_insert(tree, start, &p, &parent);
1128 prealloc = alloc_extent_state_atomic(prealloc);
1133 err = insert_state(tree, prealloc, start, end,
1134 &p, &parent, &bits, NULL);
1136 extent_io_tree_panic(tree, err);
1137 cache_state(prealloc, cached_state);
1141 state = rb_entry(node, struct extent_state, rb_node);
1143 last_start = state->start;
1144 last_end = state->end;
1147 * | ---- desired range ---- |
1150 * Just lock what we found and keep going
1152 if (state->start == start && state->end <= end) {
1153 set_state_bits(tree, state, &bits, NULL);
1154 cache_state(state, cached_state);
1155 state = clear_state_bit(tree, state, &clear_bits, 0, NULL);
1156 if (last_end == (u64)-1)
1158 start = last_end + 1;
1159 if (start < end && state && state->start == start &&
1166 * | ---- desired range ---- |
1169 * | ------------- state -------------- |
1171 * We need to split the extent we found, and may flip bits on
1174 * If the extent we found extends past our
1175 * range, we just split and search again. It'll get split
1176 * again the next time though.
1178 * If the extent we found is inside our range, we set the
1179 * desired bit on it.
1181 if (state->start < start) {
1182 prealloc = alloc_extent_state_atomic(prealloc);
1187 err = split_state(tree, state, prealloc, start);
1189 extent_io_tree_panic(tree, err);
1193 if (state->end <= end) {
1194 set_state_bits(tree, state, &bits, NULL);
1195 cache_state(state, cached_state);
1196 state = clear_state_bit(tree, state, &clear_bits, 0,
1198 if (last_end == (u64)-1)
1200 start = last_end + 1;
1201 if (start < end && state && state->start == start &&
1208 * | ---- desired range ---- |
1209 * | state | or | state |
1211 * There's a hole, we need to insert something in it and
1212 * ignore the extent we found.
1214 if (state->start > start) {
1216 if (end < last_start)
1219 this_end = last_start - 1;
1221 prealloc = alloc_extent_state_atomic(prealloc);
1228 * Avoid to free 'prealloc' if it can be merged with
1231 err = insert_state(tree, prealloc, start, this_end,
1232 NULL, NULL, &bits, NULL);
1234 extent_io_tree_panic(tree, err);
1235 cache_state(prealloc, cached_state);
1237 start = this_end + 1;
1241 * | ---- desired range ---- |
1243 * We need to split the extent, and set the bit
1246 if (state->start <= end && state->end > end) {
1247 prealloc = alloc_extent_state_atomic(prealloc);
1253 err = split_state(tree, state, prealloc, end + 1);
1255 extent_io_tree_panic(tree, err);
1257 set_state_bits(tree, prealloc, &bits, NULL);
1258 cache_state(prealloc, cached_state);
1259 clear_state_bit(tree, prealloc, &clear_bits, 0, NULL);
1267 spin_unlock(&tree->lock);
1269 first_iteration = false;
1273 spin_unlock(&tree->lock);
1275 free_extent_state(prealloc);
1280 /* wrappers around set/clear extent bit */
1281 int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1282 unsigned bits, struct extent_changeset *changeset)
1285 * We don't support EXTENT_LOCKED yet, as current changeset will
1286 * record any bits changed, so for EXTENT_LOCKED case, it will
1287 * either fail with -EEXIST or changeset will record the whole
1290 BUG_ON(bits & EXTENT_LOCKED);
1292 return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL, GFP_NOFS,
1296 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1297 unsigned bits, int wake, int delete,
1298 struct extent_state **cached)
1300 return __clear_extent_bit(tree, start, end, bits, wake, delete,
1301 cached, GFP_NOFS, NULL);
1304 int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1305 unsigned bits, struct extent_changeset *changeset)
1308 * Don't support EXTENT_LOCKED case, same reason as
1309 * set_record_extent_bits().
1311 BUG_ON(bits & EXTENT_LOCKED);
1313 return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, GFP_NOFS,
1318 * either insert or lock state struct between start and end use mask to tell
1319 * us if waiting is desired.
1321 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1322 struct extent_state **cached_state)
1328 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED,
1329 EXTENT_LOCKED, &failed_start,
1330 cached_state, GFP_NOFS, NULL);
1331 if (err == -EEXIST) {
1332 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1333 start = failed_start;
1336 WARN_ON(start > end);
1341 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1346 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1347 &failed_start, NULL, GFP_NOFS, NULL);
1348 if (err == -EEXIST) {
1349 if (failed_start > start)
1350 clear_extent_bit(tree, start, failed_start - 1,
1351 EXTENT_LOCKED, 1, 0, NULL);
1357 void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
1359 unsigned long index = start >> PAGE_SHIFT;
1360 unsigned long end_index = end >> PAGE_SHIFT;
1363 while (index <= end_index) {
1364 page = find_get_page(inode->i_mapping, index);
1365 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1366 clear_page_dirty_for_io(page);
1372 void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
1374 unsigned long index = start >> PAGE_SHIFT;
1375 unsigned long end_index = end >> PAGE_SHIFT;
1378 while (index <= end_index) {
1379 page = find_get_page(inode->i_mapping, index);
1380 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1381 __set_page_dirty_nobuffers(page);
1382 account_page_redirty(page);
1389 * helper function to set both pages and extents in the tree writeback
1391 static void set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
1393 tree->ops->set_range_writeback(tree->private_data, start, end);
1396 /* find the first state struct with 'bits' set after 'start', and
1397 * return it. tree->lock must be held. NULL will returned if
1398 * nothing was found after 'start'
1400 static struct extent_state *
1401 find_first_extent_bit_state(struct extent_io_tree *tree,
1402 u64 start, unsigned bits)
1404 struct rb_node *node;
1405 struct extent_state *state;
1408 * this search will find all the extents that end after
1411 node = tree_search(tree, start);
1416 state = rb_entry(node, struct extent_state, rb_node);
1417 if (state->end >= start && (state->state & bits))
1420 node = rb_next(node);
1429 * find the first offset in the io tree with 'bits' set. zero is
1430 * returned if we find something, and *start_ret and *end_ret are
1431 * set to reflect the state struct that was found.
1433 * If nothing was found, 1 is returned. If found something, return 0.
1435 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1436 u64 *start_ret, u64 *end_ret, unsigned bits,
1437 struct extent_state **cached_state)
1439 struct extent_state *state;
1443 spin_lock(&tree->lock);
1444 if (cached_state && *cached_state) {
1445 state = *cached_state;
1446 if (state->end == start - 1 && extent_state_in_tree(state)) {
1447 n = rb_next(&state->rb_node);
1449 state = rb_entry(n, struct extent_state,
1451 if (state->state & bits)
1455 free_extent_state(*cached_state);
1456 *cached_state = NULL;
1459 free_extent_state(*cached_state);
1460 *cached_state = NULL;
1463 state = find_first_extent_bit_state(tree, start, bits);
1466 cache_state_if_flags(state, cached_state, 0);
1467 *start_ret = state->start;
1468 *end_ret = state->end;
1472 spin_unlock(&tree->lock);
1477 * find a contiguous range of bytes in the file marked as delalloc, not
1478 * more than 'max_bytes'. start and end are used to return the range,
1480 * 1 is returned if we find something, 0 if nothing was in the tree
1482 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1483 u64 *start, u64 *end, u64 max_bytes,
1484 struct extent_state **cached_state)
1486 struct rb_node *node;
1487 struct extent_state *state;
1488 u64 cur_start = *start;
1490 u64 total_bytes = 0;
1492 spin_lock(&tree->lock);
1495 * this search will find all the extents that end after
1498 node = tree_search(tree, cur_start);
1506 state = rb_entry(node, struct extent_state, rb_node);
1507 if (found && (state->start != cur_start ||
1508 (state->state & EXTENT_BOUNDARY))) {
1511 if (!(state->state & EXTENT_DELALLOC)) {
1517 *start = state->start;
1518 *cached_state = state;
1519 refcount_inc(&state->refs);
1523 cur_start = state->end + 1;
1524 node = rb_next(node);
1525 total_bytes += state->end - state->start + 1;
1526 if (total_bytes >= max_bytes)
1532 spin_unlock(&tree->lock);
1536 static int __process_pages_contig(struct address_space *mapping,
1537 struct page *locked_page,
1538 pgoff_t start_index, pgoff_t end_index,
1539 unsigned long page_ops, pgoff_t *index_ret);
1541 static noinline void __unlock_for_delalloc(struct inode *inode,
1542 struct page *locked_page,
1545 unsigned long index = start >> PAGE_SHIFT;
1546 unsigned long end_index = end >> PAGE_SHIFT;
1548 ASSERT(locked_page);
1549 if (index == locked_page->index && end_index == index)
1552 __process_pages_contig(inode->i_mapping, locked_page, index, end_index,
1556 static noinline int lock_delalloc_pages(struct inode *inode,
1557 struct page *locked_page,
1561 unsigned long index = delalloc_start >> PAGE_SHIFT;
1562 unsigned long index_ret = index;
1563 unsigned long end_index = delalloc_end >> PAGE_SHIFT;
1566 ASSERT(locked_page);
1567 if (index == locked_page->index && index == end_index)
1570 ret = __process_pages_contig(inode->i_mapping, locked_page, index,
1571 end_index, PAGE_LOCK, &index_ret);
1573 __unlock_for_delalloc(inode, locked_page, delalloc_start,
1574 (u64)index_ret << PAGE_SHIFT);
1579 * find a contiguous range of bytes in the file marked as delalloc, not
1580 * more than 'max_bytes'. start and end are used to return the range,
1582 * 1 is returned if we find something, 0 if nothing was in the tree
1584 STATIC u64 find_lock_delalloc_range(struct inode *inode,
1585 struct extent_io_tree *tree,
1586 struct page *locked_page, u64 *start,
1587 u64 *end, u64 max_bytes)
1592 struct extent_state *cached_state = NULL;
1597 /* step one, find a bunch of delalloc bytes starting at start */
1598 delalloc_start = *start;
1600 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1601 max_bytes, &cached_state);
1602 if (!found || delalloc_end <= *start) {
1603 *start = delalloc_start;
1604 *end = delalloc_end;
1605 free_extent_state(cached_state);
1610 * start comes from the offset of locked_page. We have to lock
1611 * pages in order, so we can't process delalloc bytes before
1614 if (delalloc_start < *start)
1615 delalloc_start = *start;
1618 * make sure to limit the number of pages we try to lock down
1620 if (delalloc_end + 1 - delalloc_start > max_bytes)
1621 delalloc_end = delalloc_start + max_bytes - 1;
1623 /* step two, lock all the pages after the page that has start */
1624 ret = lock_delalloc_pages(inode, locked_page,
1625 delalloc_start, delalloc_end);
1626 if (ret == -EAGAIN) {
1627 /* some of the pages are gone, lets avoid looping by
1628 * shortening the size of the delalloc range we're searching
1630 free_extent_state(cached_state);
1631 cached_state = NULL;
1633 max_bytes = PAGE_SIZE;
1641 BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */
1643 /* step three, lock the state bits for the whole range */
1644 lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state);
1646 /* then test to make sure it is all still delalloc */
1647 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1648 EXTENT_DELALLOC, 1, cached_state);
1650 unlock_extent_cached(tree, delalloc_start, delalloc_end,
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 & SB_RDONLY));
1988 BUG_ON(!mirror_num);
1990 bio = btrfs_io_bio_alloc(1);
1991 bio->bi_iter.bi_size = 0;
1992 map_length = length;
1995 * Avoid races with device replace and make sure our bbio has devices
1996 * associated to its stripes that don't go away while we are doing the
1997 * read repair operation.
1999 btrfs_bio_counter_inc_blocked(fs_info);
2000 if (btrfs_is_parity_mirror(fs_info, logical, length)) {
2002 * Note that we don't use BTRFS_MAP_WRITE because it's supposed
2003 * to update all raid stripes, but here we just want to correct
2004 * bad stripe, thus BTRFS_MAP_READ is abused to only get the bad
2005 * stripe's dev and sector.
2007 ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, logical,
2008 &map_length, &bbio, 0);
2010 btrfs_bio_counter_dec(fs_info);
2014 ASSERT(bbio->mirror_num == 1);
2016 ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical,
2017 &map_length, &bbio, mirror_num);
2019 btrfs_bio_counter_dec(fs_info);
2023 BUG_ON(mirror_num != bbio->mirror_num);
2026 sector = bbio->stripes[bbio->mirror_num - 1].physical >> 9;
2027 bio->bi_iter.bi_sector = sector;
2028 dev = bbio->stripes[bbio->mirror_num - 1].dev;
2029 btrfs_put_bbio(bbio);
2030 if (!dev || !dev->bdev || !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,
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 0, 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, unsigned long *bio_flags,
3135 int first_index = 0;
3137 for (index = 0; index < nr_pages; index++) {
3138 page_start = page_offset(pages[index]);
3141 end = start + PAGE_SIZE - 1;
3142 first_index = index;
3143 } else if (end + 1 == page_start) {
3146 __do_contiguous_readpages(tree, &pages[first_index],
3147 index - first_index, start,
3148 end, get_extent, em_cached,
3152 end = start + PAGE_SIZE - 1;
3153 first_index = index;
3158 __do_contiguous_readpages(tree, &pages[first_index],
3159 index - first_index, start,
3160 end, get_extent, em_cached, bio,
3161 bio_flags, prev_em_start);
3164 static int __extent_read_full_page(struct extent_io_tree *tree,
3166 get_extent_t *get_extent,
3167 struct bio **bio, int mirror_num,
3168 unsigned long *bio_flags,
3169 unsigned int read_flags)
3171 struct inode *inode = page->mapping->host;
3172 struct btrfs_ordered_extent *ordered;
3173 u64 start = page_offset(page);
3174 u64 end = start + PAGE_SIZE - 1;
3178 lock_extent(tree, start, end);
3179 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
3183 unlock_extent(tree, start, end);
3184 btrfs_start_ordered_extent(inode, ordered, 1);
3185 btrfs_put_ordered_extent(ordered);
3188 ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num,
3189 bio_flags, read_flags, NULL);
3193 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
3194 get_extent_t *get_extent, int mirror_num)
3196 struct bio *bio = NULL;
3197 unsigned long bio_flags = 0;
3200 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
3203 ret = submit_one_bio(bio, mirror_num, bio_flags);
3207 static void update_nr_written(struct writeback_control *wbc,
3208 unsigned long nr_written)
3210 wbc->nr_to_write -= nr_written;
3214 * helper for __extent_writepage, doing all of the delayed allocation setup.
3216 * This returns 1 if our fill_delalloc function did all the work required
3217 * to write the page (copy into inline extent). In this case the IO has
3218 * been started and the page is already unlocked.
3220 * This returns 0 if all went well (page still locked)
3221 * This returns < 0 if there were errors (page still locked)
3223 static noinline_for_stack int writepage_delalloc(struct inode *inode,
3224 struct page *page, struct writeback_control *wbc,
3225 struct extent_page_data *epd,
3227 unsigned long *nr_written)
3229 struct extent_io_tree *tree = epd->tree;
3230 u64 page_end = delalloc_start + PAGE_SIZE - 1;
3232 u64 delalloc_to_write = 0;
3233 u64 delalloc_end = 0;
3235 int page_started = 0;
3237 if (epd->extent_locked || !tree->ops || !tree->ops->fill_delalloc)
3240 while (delalloc_end < page_end) {
3241 nr_delalloc = find_lock_delalloc_range(inode, tree,
3245 BTRFS_MAX_EXTENT_SIZE);
3246 if (nr_delalloc == 0) {
3247 delalloc_start = delalloc_end + 1;
3250 ret = tree->ops->fill_delalloc(inode, page,
3255 /* File system has been set read-only */
3258 /* fill_delalloc should be return < 0 for error
3259 * but just in case, we use > 0 here meaning the
3260 * IO is started, so we don't want to return > 0
3261 * unless things are going well.
3263 ret = ret < 0 ? ret : -EIO;
3267 * delalloc_end is already one less than the total length, so
3268 * we don't subtract one from PAGE_SIZE
3270 delalloc_to_write += (delalloc_end - delalloc_start +
3271 PAGE_SIZE) >> PAGE_SHIFT;
3272 delalloc_start = delalloc_end + 1;
3274 if (wbc->nr_to_write < delalloc_to_write) {
3277 if (delalloc_to_write < thresh * 2)
3278 thresh = delalloc_to_write;
3279 wbc->nr_to_write = min_t(u64, delalloc_to_write,
3283 /* did the fill delalloc function already unlock and start
3288 * we've unlocked the page, so we can't update
3289 * the mapping's writeback index, just update
3292 wbc->nr_to_write -= *nr_written;
3303 * helper for __extent_writepage. This calls the writepage start hooks,
3304 * and does the loop to map the page into extents and bios.
3306 * We return 1 if the IO is started and the page is unlocked,
3307 * 0 if all went well (page still locked)
3308 * < 0 if there were errors (page still locked)
3310 static noinline_for_stack int __extent_writepage_io(struct inode *inode,
3312 struct writeback_control *wbc,
3313 struct extent_page_data *epd,
3315 unsigned long nr_written,
3316 unsigned int write_flags, int *nr_ret)
3318 struct extent_io_tree *tree = epd->tree;
3319 u64 start = page_offset(page);
3320 u64 page_end = start + PAGE_SIZE - 1;
3326 struct extent_map *em;
3327 struct block_device *bdev;
3328 size_t pg_offset = 0;
3334 if (tree->ops && tree->ops->writepage_start_hook) {
3335 ret = tree->ops->writepage_start_hook(page, start,
3338 /* Fixup worker will requeue */
3340 wbc->pages_skipped++;
3342 redirty_page_for_writepage(wbc, page);
3344 update_nr_written(wbc, nr_written);
3351 * we don't want to touch the inode after unlocking the page,
3352 * so we update the mapping writeback index now
3354 update_nr_written(wbc, nr_written + 1);
3357 if (i_size <= start) {
3358 if (tree->ops && tree->ops->writepage_end_io_hook)
3359 tree->ops->writepage_end_io_hook(page, start,
3364 blocksize = inode->i_sb->s_blocksize;
3366 while (cur <= end) {
3370 if (cur >= i_size) {
3371 if (tree->ops && tree->ops->writepage_end_io_hook)
3372 tree->ops->writepage_end_io_hook(page, cur,
3376 em = epd->get_extent(BTRFS_I(inode), page, pg_offset, cur,
3378 if (IS_ERR_OR_NULL(em)) {
3380 ret = PTR_ERR_OR_ZERO(em);
3384 extent_offset = cur - em->start;
3385 em_end = extent_map_end(em);
3386 BUG_ON(em_end <= cur);
3388 iosize = min(em_end - cur, end - cur + 1);
3389 iosize = ALIGN(iosize, blocksize);
3390 offset = em->block_start + extent_offset;
3392 block_start = em->block_start;
3393 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3394 free_extent_map(em);
3398 * compressed and inline extents are written through other
3401 if (compressed || block_start == EXTENT_MAP_HOLE ||
3402 block_start == EXTENT_MAP_INLINE) {
3404 * end_io notification does not happen here for
3405 * compressed extents
3407 if (!compressed && tree->ops &&
3408 tree->ops->writepage_end_io_hook)
3409 tree->ops->writepage_end_io_hook(page, cur,
3412 else if (compressed) {
3413 /* we don't want to end_page_writeback on
3414 * a compressed extent. this happens
3421 pg_offset += iosize;
3425 set_range_writeback(tree, cur, cur + iosize - 1);
3426 if (!PageWriteback(page)) {
3427 btrfs_err(BTRFS_I(inode)->root->fs_info,
3428 "page %lu not writeback, cur %llu end %llu",
3429 page->index, cur, end);
3432 ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc,
3433 page, offset, iosize, pg_offset,
3435 end_bio_extent_writepage,
3439 if (PageWriteback(page))
3440 end_page_writeback(page);
3444 pg_offset += iosize;
3453 * the writepage semantics are similar to regular writepage. extent
3454 * records are inserted to lock ranges in the tree, and as dirty areas
3455 * are found, they are marked writeback. Then the lock bits are removed
3456 * and the end_io handler clears the writeback ranges
3458 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
3461 struct inode *inode = page->mapping->host;
3462 struct extent_page_data *epd = data;
3463 u64 start = page_offset(page);
3464 u64 page_end = start + PAGE_SIZE - 1;
3467 size_t pg_offset = 0;
3468 loff_t i_size = i_size_read(inode);
3469 unsigned long end_index = i_size >> PAGE_SHIFT;
3470 unsigned int write_flags = 0;
3471 unsigned long nr_written = 0;
3473 write_flags = wbc_to_write_flags(wbc);
3475 trace___extent_writepage(page, inode, wbc);
3477 WARN_ON(!PageLocked(page));
3479 ClearPageError(page);
3481 pg_offset = i_size & (PAGE_SIZE - 1);
3482 if (page->index > end_index ||
3483 (page->index == end_index && !pg_offset)) {
3484 page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
3489 if (page->index == end_index) {
3492 userpage = kmap_atomic(page);
3493 memset(userpage + pg_offset, 0,
3494 PAGE_SIZE - pg_offset);
3495 kunmap_atomic(userpage);
3496 flush_dcache_page(page);
3501 set_page_extent_mapped(page);
3503 ret = writepage_delalloc(inode, page, wbc, epd, start, &nr_written);
3509 ret = __extent_writepage_io(inode, page, wbc, epd,
3510 i_size, nr_written, write_flags, &nr);
3516 /* make sure the mapping tag for page dirty gets cleared */
3517 set_page_writeback(page);
3518 end_page_writeback(page);
3520 if (PageError(page)) {
3521 ret = ret < 0 ? ret : -EIO;
3522 end_extent_writepage(page, ret, start, page_end);
3531 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3533 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
3534 TASK_UNINTERRUPTIBLE);
3537 static noinline_for_stack int
3538 lock_extent_buffer_for_io(struct extent_buffer *eb,
3539 struct btrfs_fs_info *fs_info,
3540 struct extent_page_data *epd)
3542 unsigned long i, num_pages;
3546 if (!btrfs_try_tree_write_lock(eb)) {
3548 flush_write_bio(epd);
3549 btrfs_tree_lock(eb);
3552 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3553 btrfs_tree_unlock(eb);
3557 flush_write_bio(epd);
3561 wait_on_extent_buffer_writeback(eb);
3562 btrfs_tree_lock(eb);
3563 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3565 btrfs_tree_unlock(eb);
3570 * We need to do this to prevent races in people who check if the eb is
3571 * under IO since we can end up having no IO bits set for a short period
3574 spin_lock(&eb->refs_lock);
3575 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3576 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3577 spin_unlock(&eb->refs_lock);
3578 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3579 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
3581 fs_info->dirty_metadata_batch);
3584 spin_unlock(&eb->refs_lock);
3587 btrfs_tree_unlock(eb);
3592 num_pages = num_extent_pages(eb->start, eb->len);
3593 for (i = 0; i < num_pages; i++) {
3594 struct page *p = eb->pages[i];
3596 if (!trylock_page(p)) {
3598 flush_write_bio(epd);
3608 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3610 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3611 smp_mb__after_atomic();
3612 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3615 static void set_btree_ioerr(struct page *page)
3617 struct extent_buffer *eb = (struct extent_buffer *)page->private;
3620 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))
3624 * If writeback for a btree extent that doesn't belong to a log tree
3625 * failed, increment the counter transaction->eb_write_errors.
3626 * We do this because while the transaction is running and before it's
3627 * committing (when we call filemap_fdata[write|wait]_range against
3628 * the btree inode), we might have
3629 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
3630 * returns an error or an error happens during writeback, when we're
3631 * committing the transaction we wouldn't know about it, since the pages
3632 * can be no longer dirty nor marked anymore for writeback (if a
3633 * subsequent modification to the extent buffer didn't happen before the
3634 * transaction commit), which makes filemap_fdata[write|wait]_range not
3635 * able to find the pages tagged with SetPageError at transaction
3636 * commit time. So if this happens we must abort the transaction,
3637 * otherwise we commit a super block with btree roots that point to
3638 * btree nodes/leafs whose content on disk is invalid - either garbage
3639 * or the content of some node/leaf from a past generation that got
3640 * cowed or deleted and is no longer valid.
3642 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
3643 * not be enough - we need to distinguish between log tree extents vs
3644 * non-log tree extents, and the next filemap_fdatawait_range() call
3645 * will catch and clear such errors in the mapping - and that call might
3646 * be from a log sync and not from a transaction commit. Also, checking
3647 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
3648 * not done and would not be reliable - the eb might have been released
3649 * from memory and reading it back again means that flag would not be
3650 * set (since it's a runtime flag, not persisted on disk).
3652 * Using the flags below in the btree inode also makes us achieve the
3653 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
3654 * writeback for all dirty pages and before filemap_fdatawait_range()
3655 * is called, the writeback for all dirty pages had already finished
3656 * with errors - because we were not using AS_EIO/AS_ENOSPC,
3657 * filemap_fdatawait_range() would return success, as it could not know
3658 * that writeback errors happened (the pages were no longer tagged for
3661 switch (eb->log_index) {
3663 set_bit(BTRFS_FS_BTREE_ERR, &eb->fs_info->flags);
3666 set_bit(BTRFS_FS_LOG1_ERR, &eb->fs_info->flags);
3669 set_bit(BTRFS_FS_LOG2_ERR, &eb->fs_info->flags);
3672 BUG(); /* unexpected, logic error */
3676 static void end_bio_extent_buffer_writepage(struct bio *bio)
3678 struct bio_vec *bvec;
3679 struct extent_buffer *eb;
3682 ASSERT(!bio_flagged(bio, BIO_CLONED));
3683 bio_for_each_segment_all(bvec, bio, i) {
3684 struct page *page = bvec->bv_page;
3686 eb = (struct extent_buffer *)page->private;
3688 done = atomic_dec_and_test(&eb->io_pages);
3690 if (bio->bi_status ||
3691 test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) {
3692 ClearPageUptodate(page);
3693 set_btree_ioerr(page);
3696 end_page_writeback(page);
3701 end_extent_buffer_writeback(eb);
3707 static noinline_for_stack int write_one_eb(struct extent_buffer *eb,
3708 struct btrfs_fs_info *fs_info,
3709 struct writeback_control *wbc,
3710 struct extent_page_data *epd)
3712 struct block_device *bdev = fs_info->fs_devices->latest_bdev;
3713 struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
3714 u64 offset = eb->start;
3716 unsigned long i, num_pages;
3717 unsigned long start, end;
3718 unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META;
3721 clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
3722 num_pages = num_extent_pages(eb->start, eb->len);
3723 atomic_set(&eb->io_pages, num_pages);
3725 /* set btree blocks beyond nritems with 0 to avoid stale content. */
3726 nritems = btrfs_header_nritems(eb);
3727 if (btrfs_header_level(eb) > 0) {
3728 end = btrfs_node_key_ptr_offset(nritems);
3730 memzero_extent_buffer(eb, end, eb->len - end);
3734 * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0
3736 start = btrfs_item_nr_offset(nritems);
3737 end = BTRFS_LEAF_DATA_OFFSET + leaf_data_end(fs_info, eb);
3738 memzero_extent_buffer(eb, start, end - start);
3741 for (i = 0; i < num_pages; i++) {
3742 struct page *p = eb->pages[i];
3744 clear_page_dirty_for_io(p);
3745 set_page_writeback(p);
3746 ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc,
3747 p, offset, PAGE_SIZE, 0, bdev,
3749 end_bio_extent_buffer_writepage,
3753 if (PageWriteback(p))
3754 end_page_writeback(p);
3755 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3756 end_extent_buffer_writeback(eb);
3760 offset += PAGE_SIZE;
3761 update_nr_written(wbc, 1);
3765 if (unlikely(ret)) {
3766 for (; i < num_pages; i++) {
3767 struct page *p = eb->pages[i];
3768 clear_page_dirty_for_io(p);
3776 int btree_write_cache_pages(struct address_space *mapping,
3777 struct writeback_control *wbc)
3779 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
3780 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
3781 struct extent_buffer *eb, *prev_eb = NULL;
3782 struct extent_page_data epd = {
3786 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3790 int nr_to_write_done = 0;
3791 struct pagevec pvec;
3794 pgoff_t end; /* Inclusive */
3798 pagevec_init(&pvec);
3799 if (wbc->range_cyclic) {
3800 index = mapping->writeback_index; /* Start from prev offset */
3803 index = wbc->range_start >> PAGE_SHIFT;
3804 end = wbc->range_end >> PAGE_SHIFT;
3807 if (wbc->sync_mode == WB_SYNC_ALL)
3808 tag = PAGECACHE_TAG_TOWRITE;
3810 tag = PAGECACHE_TAG_DIRTY;
3812 if (wbc->sync_mode == WB_SYNC_ALL)
3813 tag_pages_for_writeback(mapping, index, end);
3814 while (!done && !nr_to_write_done && (index <= end) &&
3815 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
3820 for (i = 0; i < nr_pages; i++) {
3821 struct page *page = pvec.pages[i];
3823 if (!PagePrivate(page))
3826 spin_lock(&mapping->private_lock);
3827 if (!PagePrivate(page)) {
3828 spin_unlock(&mapping->private_lock);
3832 eb = (struct extent_buffer *)page->private;
3835 * Shouldn't happen and normally this would be a BUG_ON
3836 * but no sense in crashing the users box for something
3837 * we can survive anyway.
3840 spin_unlock(&mapping->private_lock);
3844 if (eb == prev_eb) {
3845 spin_unlock(&mapping->private_lock);
3849 ret = atomic_inc_not_zero(&eb->refs);
3850 spin_unlock(&mapping->private_lock);
3855 ret = lock_extent_buffer_for_io(eb, fs_info, &epd);
3857 free_extent_buffer(eb);
3861 ret = write_one_eb(eb, fs_info, wbc, &epd);
3864 free_extent_buffer(eb);
3867 free_extent_buffer(eb);
3870 * the filesystem may choose to bump up nr_to_write.
3871 * We have to make sure to honor the new nr_to_write
3874 nr_to_write_done = wbc->nr_to_write <= 0;
3876 pagevec_release(&pvec);
3879 if (!scanned && !done) {
3881 * We hit the last page and there is more work to be done: wrap
3882 * back to the start of the file
3888 flush_write_bio(&epd);
3893 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3894 * @mapping: address space structure to write
3895 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3896 * @writepage: function called for each page
3897 * @data: data passed to writepage function
3899 * If a page is already under I/O, write_cache_pages() skips it, even
3900 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3901 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3902 * and msync() need to guarantee that all the data which was dirty at the time
3903 * the call was made get new I/O started against them. If wbc->sync_mode is
3904 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3905 * existing IO to complete.
3907 static int extent_write_cache_pages(struct address_space *mapping,
3908 struct writeback_control *wbc,
3909 writepage_t writepage, void *data,
3910 void (*flush_fn)(void *))
3912 struct inode *inode = mapping->host;
3915 int nr_to_write_done = 0;
3916 struct pagevec pvec;
3919 pgoff_t end; /* Inclusive */
3921 int range_whole = 0;
3926 * We have to hold onto the inode so that ordered extents can do their
3927 * work when the IO finishes. The alternative to this is failing to add
3928 * an ordered extent if the igrab() fails there and that is a huge pain
3929 * to deal with, so instead just hold onto the inode throughout the
3930 * writepages operation. If it fails here we are freeing up the inode
3931 * anyway and we'd rather not waste our time writing out stuff that is
3932 * going to be truncated anyway.
3937 pagevec_init(&pvec);
3938 if (wbc->range_cyclic) {
3939 index = mapping->writeback_index; /* Start from prev offset */
3942 index = wbc->range_start >> PAGE_SHIFT;
3943 end = wbc->range_end >> PAGE_SHIFT;
3944 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
3948 if (wbc->sync_mode == WB_SYNC_ALL)
3949 tag = PAGECACHE_TAG_TOWRITE;
3951 tag = PAGECACHE_TAG_DIRTY;
3953 if (wbc->sync_mode == WB_SYNC_ALL)
3954 tag_pages_for_writeback(mapping, index, end);
3956 while (!done && !nr_to_write_done && (index <= end) &&
3957 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping,
3958 &index, end, tag))) {
3962 for (i = 0; i < nr_pages; i++) {
3963 struct page *page = pvec.pages[i];
3965 done_index = page->index;
3967 * At this point we hold neither mapping->tree_lock nor
3968 * lock on the page itself: the page may be truncated or
3969 * invalidated (changing page->mapping to NULL), or even
3970 * swizzled back from swapper_space to tmpfs file
3973 if (!trylock_page(page)) {
3978 if (unlikely(page->mapping != mapping)) {
3983 if (wbc->sync_mode != WB_SYNC_NONE) {
3984 if (PageWriteback(page))
3986 wait_on_page_writeback(page);
3989 if (PageWriteback(page) ||
3990 !clear_page_dirty_for_io(page)) {
3995 ret = (*writepage)(page, wbc, data);
3997 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
4003 * done_index is set past this page,
4004 * so media errors will not choke
4005 * background writeout for the entire
4006 * file. This has consequences for
4007 * range_cyclic semantics (ie. it may
4008 * not be suitable for data integrity
4011 done_index = page->index + 1;
4017 * the filesystem may choose to bump up nr_to_write.
4018 * We have to make sure to honor the new nr_to_write
4021 nr_to_write_done = wbc->nr_to_write <= 0;
4023 pagevec_release(&pvec);
4026 if (!scanned && !done) {
4028 * We hit the last page and there is more work to be done: wrap
4029 * back to the start of the file
4036 if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole))
4037 mapping->writeback_index = done_index;
4039 btrfs_add_delayed_iput(inode);
4043 static void flush_epd_write_bio(struct extent_page_data *epd)
4048 ret = submit_one_bio(epd->bio, 0, 0);
4049 BUG_ON(ret < 0); /* -ENOMEM */
4054 static noinline void flush_write_bio(void *data)
4056 struct extent_page_data *epd = data;
4057 flush_epd_write_bio(epd);
4060 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
4061 struct writeback_control *wbc)
4064 struct extent_page_data epd = {
4067 .get_extent = btrfs_get_extent,
4069 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4072 ret = __extent_writepage(page, wbc, &epd);
4074 flush_epd_write_bio(&epd);
4078 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
4079 u64 start, u64 end, int mode)
4082 struct address_space *mapping = inode->i_mapping;
4084 unsigned long nr_pages = (end - start + PAGE_SIZE) >>
4087 struct extent_page_data epd = {
4090 .get_extent = btrfs_get_extent,
4092 .sync_io = mode == WB_SYNC_ALL,
4094 struct writeback_control wbc_writepages = {
4096 .nr_to_write = nr_pages * 2,
4097 .range_start = start,
4098 .range_end = end + 1,
4101 while (start <= end) {
4102 page = find_get_page(mapping, start >> PAGE_SHIFT);
4103 if (clear_page_dirty_for_io(page))
4104 ret = __extent_writepage(page, &wbc_writepages, &epd);
4106 if (tree->ops && tree->ops->writepage_end_io_hook)
4107 tree->ops->writepage_end_io_hook(page, start,
4108 start + PAGE_SIZE - 1,
4116 flush_epd_write_bio(&epd);
4120 int extent_writepages(struct extent_io_tree *tree,
4121 struct address_space *mapping,
4122 struct writeback_control *wbc)
4125 struct extent_page_data epd = {
4128 .get_extent = btrfs_get_extent,
4130 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4133 ret = extent_write_cache_pages(mapping, wbc, __extent_writepage, &epd,
4135 flush_epd_write_bio(&epd);
4139 int extent_readpages(struct extent_io_tree *tree,
4140 struct address_space *mapping,
4141 struct list_head *pages, unsigned nr_pages,
4142 get_extent_t get_extent)
4144 struct bio *bio = NULL;
4146 unsigned long bio_flags = 0;
4147 struct page *pagepool[16];
4149 struct extent_map *em_cached = NULL;
4151 u64 prev_em_start = (u64)-1;
4153 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
4154 page = list_entry(pages->prev, struct page, lru);
4156 prefetchw(&page->flags);
4157 list_del(&page->lru);
4158 if (add_to_page_cache_lru(page, mapping,
4160 readahead_gfp_mask(mapping))) {
4165 pagepool[nr++] = page;
4166 if (nr < ARRAY_SIZE(pagepool))
4168 __extent_readpages(tree, pagepool, nr, get_extent, &em_cached,
4169 &bio, &bio_flags, &prev_em_start);
4173 __extent_readpages(tree, pagepool, nr, get_extent, &em_cached,
4174 &bio, &bio_flags, &prev_em_start);
4177 free_extent_map(em_cached);
4179 BUG_ON(!list_empty(pages));
4181 return submit_one_bio(bio, 0, bio_flags);
4186 * basic invalidatepage code, this waits on any locked or writeback
4187 * ranges corresponding to the page, and then deletes any extent state
4188 * records from the tree
4190 int extent_invalidatepage(struct extent_io_tree *tree,
4191 struct page *page, unsigned long offset)
4193 struct extent_state *cached_state = NULL;
4194 u64 start = page_offset(page);
4195 u64 end = start + PAGE_SIZE - 1;
4196 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
4198 start += ALIGN(offset, blocksize);
4202 lock_extent_bits(tree, start, end, &cached_state);
4203 wait_on_page_writeback(page);
4204 clear_extent_bit(tree, start, end,
4205 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4206 EXTENT_DO_ACCOUNTING,
4207 1, 1, &cached_state);
4212 * a helper for releasepage, this tests for areas of the page that
4213 * are locked or under IO and drops the related state bits if it is safe
4216 static int try_release_extent_state(struct extent_map_tree *map,
4217 struct extent_io_tree *tree,
4218 struct page *page, gfp_t mask)
4220 u64 start = page_offset(page);
4221 u64 end = start + PAGE_SIZE - 1;
4224 if (test_range_bit(tree, start, end,
4225 EXTENT_IOBITS, 0, NULL))
4229 * at this point we can safely clear everything except the
4230 * locked bit and the nodatasum bit
4232 ret = __clear_extent_bit(tree, start, end,
4233 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
4234 0, 0, NULL, mask, NULL);
4236 /* if clear_extent_bit failed for enomem reasons,
4237 * we can't allow the release to continue.
4248 * a helper for releasepage. As long as there are no locked extents
4249 * in the range corresponding to the page, both state records and extent
4250 * map records are removed
4252 int try_release_extent_mapping(struct extent_map_tree *map,
4253 struct extent_io_tree *tree, struct page *page,
4256 struct extent_map *em;
4257 u64 start = page_offset(page);
4258 u64 end = start + PAGE_SIZE - 1;
4260 if (gfpflags_allow_blocking(mask) &&
4261 page->mapping->host->i_size > SZ_16M) {
4263 while (start <= end) {
4264 len = end - start + 1;
4265 write_lock(&map->lock);
4266 em = lookup_extent_mapping(map, start, len);
4268 write_unlock(&map->lock);
4271 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
4272 em->start != start) {
4273 write_unlock(&map->lock);
4274 free_extent_map(em);
4277 if (!test_range_bit(tree, em->start,
4278 extent_map_end(em) - 1,
4279 EXTENT_LOCKED | EXTENT_WRITEBACK,
4281 remove_extent_mapping(map, em);
4282 /* once for the rb tree */
4283 free_extent_map(em);
4285 start = extent_map_end(em);
4286 write_unlock(&map->lock);
4289 free_extent_map(em);
4292 return try_release_extent_state(map, tree, page, mask);
4296 * helper function for fiemap, which doesn't want to see any holes.
4297 * This maps until we find something past 'last'
4299 static struct extent_map *get_extent_skip_holes(struct inode *inode,
4302 get_extent_t *get_extent)
4304 u64 sectorsize = btrfs_inode_sectorsize(inode);
4305 struct extent_map *em;
4312 len = last - offset;
4315 len = ALIGN(len, sectorsize);
4316 em = get_extent(BTRFS_I(inode), NULL, 0, offset, len, 0);
4317 if (IS_ERR_OR_NULL(em))
4320 /* if this isn't a hole return it */
4321 if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) &&
4322 em->block_start != EXTENT_MAP_HOLE) {
4326 /* this is a hole, advance to the next extent */
4327 offset = extent_map_end(em);
4328 free_extent_map(em);
4336 * To cache previous fiemap extent
4338 * Will be used for merging fiemap extent
4340 struct fiemap_cache {
4349 * Helper to submit fiemap extent.
4351 * Will try to merge current fiemap extent specified by @offset, @phys,
4352 * @len and @flags with cached one.
4353 * And only when we fails to merge, cached one will be submitted as
4356 * Return value is the same as fiemap_fill_next_extent().
4358 static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
4359 struct fiemap_cache *cache,
4360 u64 offset, u64 phys, u64 len, u32 flags)
4368 * Sanity check, extent_fiemap() should have ensured that new
4369 * fiemap extent won't overlap with cahced one.
4372 * NOTE: Physical address can overlap, due to compression
4374 if (cache->offset + cache->len > offset) {
4380 * Only merges fiemap extents if
4381 * 1) Their logical addresses are continuous
4383 * 2) Their physical addresses are continuous
4384 * So truly compressed (physical size smaller than logical size)
4385 * extents won't get merged with each other
4387 * 3) Share same flags except FIEMAP_EXTENT_LAST
4388 * So regular extent won't get merged with prealloc extent
4390 if (cache->offset + cache->len == offset &&
4391 cache->phys + cache->len == phys &&
4392 (cache->flags & ~FIEMAP_EXTENT_LAST) ==
4393 (flags & ~FIEMAP_EXTENT_LAST)) {
4395 cache->flags |= flags;
4396 goto try_submit_last;
4399 /* Not mergeable, need to submit cached one */
4400 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
4401 cache->len, cache->flags);
4402 cache->cached = false;
4406 cache->cached = true;
4407 cache->offset = offset;
4410 cache->flags = flags;
4412 if (cache->flags & FIEMAP_EXTENT_LAST) {
4413 ret = fiemap_fill_next_extent(fieinfo, cache->offset,
4414 cache->phys, cache->len, cache->flags);
4415 cache->cached = false;
4421 * Emit last fiemap cache
4423 * The last fiemap cache may still be cached in the following case:
4425 * |<- Fiemap range ->|
4426 * |<------------ First extent ----------->|
4428 * In this case, the first extent range will be cached but not emitted.
4429 * So we must emit it before ending extent_fiemap().
4431 static int emit_last_fiemap_cache(struct btrfs_fs_info *fs_info,
4432 struct fiemap_extent_info *fieinfo,
4433 struct fiemap_cache *cache)
4440 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
4441 cache->len, cache->flags);
4442 cache->cached = false;
4448 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4449 __u64 start, __u64 len, get_extent_t *get_extent)
4453 u64 max = start + len;
4457 u64 last_for_get_extent = 0;
4459 u64 isize = i_size_read(inode);
4460 struct btrfs_key found_key;
4461 struct extent_map *em = NULL;
4462 struct extent_state *cached_state = NULL;
4463 struct btrfs_path *path;
4464 struct btrfs_root *root = BTRFS_I(inode)->root;
4465 struct fiemap_cache cache = { 0 };
4474 path = btrfs_alloc_path();
4477 path->leave_spinning = 1;
4479 start = round_down(start, btrfs_inode_sectorsize(inode));
4480 len = round_up(max, btrfs_inode_sectorsize(inode)) - start;
4483 * lookup the last file extent. We're not using i_size here
4484 * because there might be preallocation past i_size
4486 ret = btrfs_lookup_file_extent(NULL, root, path,
4487 btrfs_ino(BTRFS_I(inode)), -1, 0);
4489 btrfs_free_path(path);
4498 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
4499 found_type = found_key.type;
4501 /* No extents, but there might be delalloc bits */
4502 if (found_key.objectid != btrfs_ino(BTRFS_I(inode)) ||
4503 found_type != BTRFS_EXTENT_DATA_KEY) {
4504 /* have to trust i_size as the end */
4506 last_for_get_extent = isize;
4509 * remember the start of the last extent. There are a
4510 * bunch of different factors that go into the length of the
4511 * extent, so its much less complex to remember where it started
4513 last = found_key.offset;
4514 last_for_get_extent = last + 1;
4516 btrfs_release_path(path);
4519 * we might have some extents allocated but more delalloc past those
4520 * extents. so, we trust isize unless the start of the last extent is
4525 last_for_get_extent = isize;
4528 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4531 em = get_extent_skip_holes(inode, start, last_for_get_extent,
4541 u64 offset_in_extent = 0;
4543 /* break if the extent we found is outside the range */
4544 if (em->start >= max || extent_map_end(em) < off)
4548 * get_extent may return an extent that starts before our
4549 * requested range. We have to make sure the ranges
4550 * we return to fiemap always move forward and don't
4551 * overlap, so adjust the offsets here
4553 em_start = max(em->start, off);
4556 * record the offset from the start of the extent
4557 * for adjusting the disk offset below. Only do this if the
4558 * extent isn't compressed since our in ram offset may be past
4559 * what we have actually allocated on disk.
4561 if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4562 offset_in_extent = em_start - em->start;
4563 em_end = extent_map_end(em);
4564 em_len = em_end - em_start;
4569 * bump off for our next call to get_extent
4571 off = extent_map_end(em);
4575 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
4577 flags |= FIEMAP_EXTENT_LAST;
4578 } else if (em->block_start == EXTENT_MAP_INLINE) {
4579 flags |= (FIEMAP_EXTENT_DATA_INLINE |
4580 FIEMAP_EXTENT_NOT_ALIGNED);
4581 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
4582 flags |= (FIEMAP_EXTENT_DELALLOC |
4583 FIEMAP_EXTENT_UNKNOWN);
4584 } else if (fieinfo->fi_extents_max) {
4585 u64 bytenr = em->block_start -
4586 (em->start - em->orig_start);
4588 disko = em->block_start + offset_in_extent;
4591 * As btrfs supports shared space, this information
4592 * can be exported to userspace tools via
4593 * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0
4594 * then we're just getting a count and we can skip the
4597 ret = btrfs_check_shared(root,
4598 btrfs_ino(BTRFS_I(inode)),
4603 flags |= FIEMAP_EXTENT_SHARED;
4606 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4607 flags |= FIEMAP_EXTENT_ENCODED;
4608 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4609 flags |= FIEMAP_EXTENT_UNWRITTEN;
4611 free_extent_map(em);
4613 if ((em_start >= last) || em_len == (u64)-1 ||
4614 (last == (u64)-1 && isize <= em_end)) {
4615 flags |= FIEMAP_EXTENT_LAST;
4619 /* now scan forward to see if this is really the last extent. */
4620 em = get_extent_skip_holes(inode, off, last_for_get_extent,
4627 flags |= FIEMAP_EXTENT_LAST;
4630 ret = emit_fiemap_extent(fieinfo, &cache, em_start, disko,
4640 ret = emit_last_fiemap_cache(root->fs_info, fieinfo, &cache);
4641 free_extent_map(em);
4643 btrfs_free_path(path);
4644 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4645 &cached_state, GFP_NOFS);
4649 static void __free_extent_buffer(struct extent_buffer *eb)
4651 btrfs_leak_debug_del(&eb->leak_list);
4652 kmem_cache_free(extent_buffer_cache, eb);
4655 int extent_buffer_under_io(struct extent_buffer *eb)
4657 return (atomic_read(&eb->io_pages) ||
4658 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4659 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4663 * Helper for releasing extent buffer page.
4665 static void btrfs_release_extent_buffer_page(struct extent_buffer *eb)
4667 unsigned long index;
4669 int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4671 BUG_ON(extent_buffer_under_io(eb));
4673 index = num_extent_pages(eb->start, eb->len);
4679 page = eb->pages[index];
4683 spin_lock(&page->mapping->private_lock);
4685 * We do this since we'll remove the pages after we've
4686 * removed the eb from the radix tree, so we could race
4687 * and have this page now attached to the new eb. So
4688 * only clear page_private if it's still connected to
4691 if (PagePrivate(page) &&
4692 page->private == (unsigned long)eb) {
4693 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4694 BUG_ON(PageDirty(page));
4695 BUG_ON(PageWriteback(page));
4697 * We need to make sure we haven't be attached
4700 ClearPagePrivate(page);
4701 set_page_private(page, 0);
4702 /* One for the page private */
4707 spin_unlock(&page->mapping->private_lock);
4709 /* One for when we allocated the page */
4711 } while (index != 0);
4715 * Helper for releasing the extent buffer.
4717 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4719 btrfs_release_extent_buffer_page(eb);
4720 __free_extent_buffer(eb);
4723 static struct extent_buffer *
4724 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
4727 struct extent_buffer *eb = NULL;
4729 eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL);
4732 eb->fs_info = fs_info;
4734 rwlock_init(&eb->lock);
4735 atomic_set(&eb->write_locks, 0);
4736 atomic_set(&eb->read_locks, 0);
4737 atomic_set(&eb->blocking_readers, 0);
4738 atomic_set(&eb->blocking_writers, 0);
4739 atomic_set(&eb->spinning_readers, 0);
4740 atomic_set(&eb->spinning_writers, 0);
4741 eb->lock_nested = 0;
4742 init_waitqueue_head(&eb->write_lock_wq);
4743 init_waitqueue_head(&eb->read_lock_wq);
4745 btrfs_leak_debug_add(&eb->leak_list, &buffers);
4747 spin_lock_init(&eb->refs_lock);
4748 atomic_set(&eb->refs, 1);
4749 atomic_set(&eb->io_pages, 0);
4752 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4754 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4755 > MAX_INLINE_EXTENT_BUFFER_SIZE);
4756 BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE);
4761 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src)
4765 struct extent_buffer *new;
4766 unsigned long num_pages = num_extent_pages(src->start, src->len);
4768 new = __alloc_extent_buffer(src->fs_info, src->start, src->len);
4772 for (i = 0; i < num_pages; i++) {
4773 p = alloc_page(GFP_NOFS);
4775 btrfs_release_extent_buffer(new);
4778 attach_extent_buffer_page(new, p);
4779 WARN_ON(PageDirty(p));
4782 copy_page(page_address(p), page_address(src->pages[i]));
4785 set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
4786 set_bit(EXTENT_BUFFER_DUMMY, &new->bflags);
4791 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4792 u64 start, unsigned long len)
4794 struct extent_buffer *eb;
4795 unsigned long num_pages;
4798 num_pages = num_extent_pages(start, len);
4800 eb = __alloc_extent_buffer(fs_info, start, len);
4804 for (i = 0; i < num_pages; i++) {
4805 eb->pages[i] = alloc_page(GFP_NOFS);
4809 set_extent_buffer_uptodate(eb);
4810 btrfs_set_header_nritems(eb, 0);
4811 set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4816 __free_page(eb->pages[i - 1]);
4817 __free_extent_buffer(eb);
4821 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4824 return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize);
4827 static void check_buffer_tree_ref(struct extent_buffer *eb)
4830 /* the ref bit is tricky. We have to make sure it is set
4831 * if we have the buffer dirty. Otherwise the
4832 * code to free a buffer can end up dropping a dirty
4835 * Once the ref bit is set, it won't go away while the
4836 * buffer is dirty or in writeback, and it also won't
4837 * go away while we have the reference count on the
4840 * We can't just set the ref bit without bumping the
4841 * ref on the eb because free_extent_buffer might
4842 * see the ref bit and try to clear it. If this happens
4843 * free_extent_buffer might end up dropping our original
4844 * ref by mistake and freeing the page before we are able
4845 * to add one more ref.
4847 * So bump the ref count first, then set the bit. If someone
4848 * beat us to it, drop the ref we added.
4850 refs = atomic_read(&eb->refs);
4851 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4854 spin_lock(&eb->refs_lock);
4855 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4856 atomic_inc(&eb->refs);
4857 spin_unlock(&eb->refs_lock);
4860 static void mark_extent_buffer_accessed(struct extent_buffer *eb,
4861 struct page *accessed)
4863 unsigned long num_pages, i;
4865 check_buffer_tree_ref(eb);
4867 num_pages = num_extent_pages(eb->start, eb->len);
4868 for (i = 0; i < num_pages; i++) {
4869 struct page *p = eb->pages[i];
4872 mark_page_accessed(p);
4876 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
4879 struct extent_buffer *eb;
4882 eb = radix_tree_lookup(&fs_info->buffer_radix,
4883 start >> PAGE_SHIFT);
4884 if (eb && atomic_inc_not_zero(&eb->refs)) {
4887 * Lock our eb's refs_lock to avoid races with
4888 * free_extent_buffer. When we get our eb it might be flagged
4889 * with EXTENT_BUFFER_STALE and another task running
4890 * free_extent_buffer might have seen that flag set,
4891 * eb->refs == 2, that the buffer isn't under IO (dirty and
4892 * writeback flags not set) and it's still in the tree (flag
4893 * EXTENT_BUFFER_TREE_REF set), therefore being in the process
4894 * of decrementing the extent buffer's reference count twice.
4895 * So here we could race and increment the eb's reference count,
4896 * clear its stale flag, mark it as dirty and drop our reference
4897 * before the other task finishes executing free_extent_buffer,
4898 * which would later result in an attempt to free an extent
4899 * buffer that is dirty.
4901 if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) {
4902 spin_lock(&eb->refs_lock);
4903 spin_unlock(&eb->refs_lock);
4905 mark_extent_buffer_accessed(eb, NULL);
4913 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4914 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
4917 struct extent_buffer *eb, *exists = NULL;
4920 eb = find_extent_buffer(fs_info, start);
4923 eb = alloc_dummy_extent_buffer(fs_info, start);
4926 eb->fs_info = fs_info;
4928 ret = radix_tree_preload(GFP_NOFS);
4931 spin_lock(&fs_info->buffer_lock);
4932 ret = radix_tree_insert(&fs_info->buffer_radix,
4933 start >> PAGE_SHIFT, eb);
4934 spin_unlock(&fs_info->buffer_lock);
4935 radix_tree_preload_end();
4936 if (ret == -EEXIST) {
4937 exists = find_extent_buffer(fs_info, start);
4943 check_buffer_tree_ref(eb);
4944 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
4947 * We will free dummy extent buffer's if they come into
4948 * free_extent_buffer with a ref count of 2, but if we are using this we
4949 * want the buffers to stay in memory until we're done with them, so
4950 * bump the ref count again.
4952 atomic_inc(&eb->refs);
4955 btrfs_release_extent_buffer(eb);
4960 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
4963 unsigned long len = fs_info->nodesize;
4964 unsigned long num_pages = num_extent_pages(start, len);
4966 unsigned long index = start >> PAGE_SHIFT;
4967 struct extent_buffer *eb;
4968 struct extent_buffer *exists = NULL;
4970 struct address_space *mapping = fs_info->btree_inode->i_mapping;
4974 if (!IS_ALIGNED(start, fs_info->sectorsize)) {
4975 btrfs_err(fs_info, "bad tree block start %llu", start);
4976 return ERR_PTR(-EINVAL);
4979 eb = find_extent_buffer(fs_info, start);
4983 eb = __alloc_extent_buffer(fs_info, start, len);
4985 return ERR_PTR(-ENOMEM);
4987 for (i = 0; i < num_pages; i++, index++) {
4988 p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL);
4990 exists = ERR_PTR(-ENOMEM);
4994 spin_lock(&mapping->private_lock);
4995 if (PagePrivate(p)) {
4997 * We could have already allocated an eb for this page
4998 * and attached one so lets see if we can get a ref on
4999 * the existing eb, and if we can we know it's good and
5000 * we can just return that one, else we know we can just
5001 * overwrite page->private.
5003 exists = (struct extent_buffer *)p->private;
5004 if (atomic_inc_not_zero(&exists->refs)) {
5005 spin_unlock(&mapping->private_lock);
5008 mark_extent_buffer_accessed(exists, p);
5014 * Do this so attach doesn't complain and we need to
5015 * drop the ref the old guy had.
5017 ClearPagePrivate(p);
5018 WARN_ON(PageDirty(p));
5021 attach_extent_buffer_page(eb, p);
5022 spin_unlock(&mapping->private_lock);
5023 WARN_ON(PageDirty(p));
5025 if (!PageUptodate(p))
5029 * see below about how we avoid a nasty race with release page
5030 * and why we unlock later
5034 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5036 ret = radix_tree_preload(GFP_NOFS);
5038 exists = ERR_PTR(ret);
5042 spin_lock(&fs_info->buffer_lock);
5043 ret = radix_tree_insert(&fs_info->buffer_radix,
5044 start >> PAGE_SHIFT, eb);
5045 spin_unlock(&fs_info->buffer_lock);
5046 radix_tree_preload_end();
5047 if (ret == -EEXIST) {
5048 exists = find_extent_buffer(fs_info, start);
5054 /* add one reference for the tree */
5055 check_buffer_tree_ref(eb);
5056 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
5059 * there is a race where release page may have
5060 * tried to find this extent buffer in the radix
5061 * but failed. It will tell the VM it is safe to
5062 * reclaim the, and it will clear the page private bit.
5063 * We must make sure to set the page private bit properly
5064 * after the extent buffer is in the radix tree so
5065 * it doesn't get lost
5067 SetPageChecked(eb->pages[0]);
5068 for (i = 1; i < num_pages; i++) {
5070 ClearPageChecked(p);
5073 unlock_page(eb->pages[0]);
5077 WARN_ON(!atomic_dec_and_test(&eb->refs));
5078 for (i = 0; i < num_pages; i++) {
5080 unlock_page(eb->pages[i]);
5083 btrfs_release_extent_buffer(eb);
5087 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
5089 struct extent_buffer *eb =
5090 container_of(head, struct extent_buffer, rcu_head);
5092 __free_extent_buffer(eb);
5095 /* Expects to have eb->eb_lock already held */
5096 static int release_extent_buffer(struct extent_buffer *eb)
5098 WARN_ON(atomic_read(&eb->refs) == 0);
5099 if (atomic_dec_and_test(&eb->refs)) {
5100 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
5101 struct btrfs_fs_info *fs_info = eb->fs_info;
5103 spin_unlock(&eb->refs_lock);
5105 spin_lock(&fs_info->buffer_lock);
5106 radix_tree_delete(&fs_info->buffer_radix,
5107 eb->start >> PAGE_SHIFT);
5108 spin_unlock(&fs_info->buffer_lock);
5110 spin_unlock(&eb->refs_lock);
5113 /* Should be safe to release our pages at this point */
5114 btrfs_release_extent_buffer_page(eb);
5115 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5116 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))) {
5117 __free_extent_buffer(eb);
5121 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
5124 spin_unlock(&eb->refs_lock);
5129 void free_extent_buffer(struct extent_buffer *eb)
5137 refs = atomic_read(&eb->refs);
5140 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
5145 spin_lock(&eb->refs_lock);
5146 if (atomic_read(&eb->refs) == 2 &&
5147 test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))
5148 atomic_dec(&eb->refs);
5150 if (atomic_read(&eb->refs) == 2 &&
5151 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
5152 !extent_buffer_under_io(eb) &&
5153 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5154 atomic_dec(&eb->refs);
5157 * I know this is terrible, but it's temporary until we stop tracking
5158 * the uptodate bits and such for the extent buffers.
5160 release_extent_buffer(eb);
5163 void free_extent_buffer_stale(struct extent_buffer *eb)
5168 spin_lock(&eb->refs_lock);
5169 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
5171 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
5172 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5173 atomic_dec(&eb->refs);
5174 release_extent_buffer(eb);
5177 void clear_extent_buffer_dirty(struct extent_buffer *eb)
5180 unsigned long num_pages;
5183 num_pages = num_extent_pages(eb->start, eb->len);
5185 for (i = 0; i < num_pages; i++) {
5186 page = eb->pages[i];
5187 if (!PageDirty(page))
5191 WARN_ON(!PagePrivate(page));
5193 clear_page_dirty_for_io(page);
5194 spin_lock_irq(&page->mapping->tree_lock);
5195 if (!PageDirty(page)) {
5196 radix_tree_tag_clear(&page->mapping->page_tree,
5198 PAGECACHE_TAG_DIRTY);
5200 spin_unlock_irq(&page->mapping->tree_lock);
5201 ClearPageError(page);
5204 WARN_ON(atomic_read(&eb->refs) == 0);
5207 int set_extent_buffer_dirty(struct extent_buffer *eb)
5210 unsigned long num_pages;
5213 check_buffer_tree_ref(eb);
5215 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
5217 num_pages = num_extent_pages(eb->start, eb->len);
5218 WARN_ON(atomic_read(&eb->refs) == 0);
5219 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
5221 for (i = 0; i < num_pages; i++)
5222 set_page_dirty(eb->pages[i]);
5226 void clear_extent_buffer_uptodate(struct extent_buffer *eb)
5230 unsigned long num_pages;
5232 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5233 num_pages = num_extent_pages(eb->start, eb->len);
5234 for (i = 0; i < num_pages; i++) {
5235 page = eb->pages[i];
5237 ClearPageUptodate(page);
5241 void set_extent_buffer_uptodate(struct extent_buffer *eb)
5245 unsigned long num_pages;
5247 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5248 num_pages = num_extent_pages(eb->start, eb->len);
5249 for (i = 0; i < num_pages; i++) {
5250 page = eb->pages[i];
5251 SetPageUptodate(page);
5255 int extent_buffer_uptodate(struct extent_buffer *eb)
5257 return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5260 int read_extent_buffer_pages(struct extent_io_tree *tree,
5261 struct extent_buffer *eb, int wait,
5262 get_extent_t *get_extent, int mirror_num)
5268 int locked_pages = 0;
5269 int all_uptodate = 1;
5270 unsigned long num_pages;
5271 unsigned long num_reads = 0;
5272 struct bio *bio = NULL;
5273 unsigned long bio_flags = 0;
5275 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
5278 num_pages = num_extent_pages(eb->start, eb->len);
5279 for (i = 0; i < num_pages; i++) {
5280 page = eb->pages[i];
5281 if (wait == WAIT_NONE) {
5282 if (!trylock_page(page))
5290 * We need to firstly lock all pages to make sure that
5291 * the uptodate bit of our pages won't be affected by
5292 * clear_extent_buffer_uptodate().
5294 for (i = 0; i < num_pages; i++) {
5295 page = eb->pages[i];
5296 if (!PageUptodate(page)) {
5303 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5307 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
5308 eb->read_mirror = 0;
5309 atomic_set(&eb->io_pages, num_reads);
5310 for (i = 0; i < num_pages; i++) {
5311 page = eb->pages[i];
5313 if (!PageUptodate(page)) {
5315 atomic_dec(&eb->io_pages);
5320 ClearPageError(page);
5321 err = __extent_read_full_page(tree, page,
5323 mirror_num, &bio_flags,
5328 * We use &bio in above __extent_read_full_page,
5329 * so we ensure that if it returns error, the
5330 * current page fails to add itself to bio and
5331 * it's been unlocked.
5333 * We must dec io_pages by ourselves.
5335 atomic_dec(&eb->io_pages);
5343 err = submit_one_bio(bio, mirror_num, bio_flags);
5348 if (ret || wait != WAIT_COMPLETE)
5351 for (i = 0; i < num_pages; i++) {
5352 page = eb->pages[i];
5353 wait_on_page_locked(page);
5354 if (!PageUptodate(page))
5361 while (locked_pages > 0) {
5363 page = eb->pages[locked_pages];
5369 void read_extent_buffer(const struct extent_buffer *eb, void *dstv,
5370 unsigned long start, unsigned long len)
5376 char *dst = (char *)dstv;
5377 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5378 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5380 if (start + len > eb->len) {
5381 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n",
5382 eb->start, eb->len, start, len);
5383 memset(dst, 0, len);
5387 offset = (start_offset + start) & (PAGE_SIZE - 1);
5390 page = eb->pages[i];
5392 cur = min(len, (PAGE_SIZE - offset));
5393 kaddr = page_address(page);
5394 memcpy(dst, kaddr + offset, cur);
5403 int read_extent_buffer_to_user(const struct extent_buffer *eb,
5405 unsigned long start, unsigned long len)
5411 char __user *dst = (char __user *)dstv;
5412 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5413 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5416 WARN_ON(start > eb->len);
5417 WARN_ON(start + len > eb->start + eb->len);
5419 offset = (start_offset + start) & (PAGE_SIZE - 1);
5422 page = eb->pages[i];
5424 cur = min(len, (PAGE_SIZE - offset));
5425 kaddr = page_address(page);
5426 if (copy_to_user(dst, kaddr + offset, cur)) {
5441 * return 0 if the item is found within a page.
5442 * return 1 if the item spans two pages.
5443 * return -EINVAL otherwise.
5445 int map_private_extent_buffer(const struct extent_buffer *eb,
5446 unsigned long start, unsigned long min_len,
5447 char **map, unsigned long *map_start,
5448 unsigned long *map_len)
5450 size_t offset = start & (PAGE_SIZE - 1);
5453 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5454 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5455 unsigned long end_i = (start_offset + start + min_len - 1) >>
5458 if (start + min_len > eb->len) {
5459 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n",
5460 eb->start, eb->len, start, min_len);
5468 offset = start_offset;
5472 *map_start = ((u64)i << PAGE_SHIFT) - start_offset;
5476 kaddr = page_address(p);
5477 *map = kaddr + offset;
5478 *map_len = PAGE_SIZE - offset;
5482 int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,
5483 unsigned long start, unsigned long len)
5489 char *ptr = (char *)ptrv;
5490 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5491 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5494 WARN_ON(start > eb->len);
5495 WARN_ON(start + len > eb->start + eb->len);
5497 offset = (start_offset + start) & (PAGE_SIZE - 1);
5500 page = eb->pages[i];
5502 cur = min(len, (PAGE_SIZE - offset));
5504 kaddr = page_address(page);
5505 ret = memcmp(ptr, kaddr + offset, cur);
5517 void write_extent_buffer_chunk_tree_uuid(struct extent_buffer *eb,
5522 WARN_ON(!PageUptodate(eb->pages[0]));
5523 kaddr = page_address(eb->pages[0]);
5524 memcpy(kaddr + offsetof(struct btrfs_header, chunk_tree_uuid), srcv,
5528 void write_extent_buffer_fsid(struct extent_buffer *eb, const void *srcv)
5532 WARN_ON(!PageUptodate(eb->pages[0]));
5533 kaddr = page_address(eb->pages[0]);
5534 memcpy(kaddr + offsetof(struct btrfs_header, fsid), srcv,
5538 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
5539 unsigned long start, unsigned long len)
5545 char *src = (char *)srcv;
5546 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5547 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5549 WARN_ON(start > eb->len);
5550 WARN_ON(start + len > eb->start + eb->len);
5552 offset = (start_offset + start) & (PAGE_SIZE - 1);
5555 page = eb->pages[i];
5556 WARN_ON(!PageUptodate(page));
5558 cur = min(len, PAGE_SIZE - offset);
5559 kaddr = page_address(page);
5560 memcpy(kaddr + offset, src, cur);
5569 void memzero_extent_buffer(struct extent_buffer *eb, unsigned long start,
5576 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5577 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5579 WARN_ON(start > eb->len);
5580 WARN_ON(start + len > eb->start + eb->len);
5582 offset = (start_offset + start) & (PAGE_SIZE - 1);
5585 page = eb->pages[i];
5586 WARN_ON(!PageUptodate(page));
5588 cur = min(len, PAGE_SIZE - offset);
5589 kaddr = page_address(page);
5590 memset(kaddr + offset, 0, cur);
5598 void copy_extent_buffer_full(struct extent_buffer *dst,
5599 struct extent_buffer *src)
5604 ASSERT(dst->len == src->len);
5606 num_pages = num_extent_pages(dst->start, dst->len);
5607 for (i = 0; i < num_pages; i++)
5608 copy_page(page_address(dst->pages[i]),
5609 page_address(src->pages[i]));
5612 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
5613 unsigned long dst_offset, unsigned long src_offset,
5616 u64 dst_len = dst->len;
5621 size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1);
5622 unsigned long i = (start_offset + dst_offset) >> PAGE_SHIFT;
5624 WARN_ON(src->len != dst_len);
5626 offset = (start_offset + dst_offset) &
5630 page = dst->pages[i];
5631 WARN_ON(!PageUptodate(page));
5633 cur = min(len, (unsigned long)(PAGE_SIZE - offset));
5635 kaddr = page_address(page);
5636 read_extent_buffer(src, kaddr + offset, src_offset, cur);
5645 void le_bitmap_set(u8 *map, unsigned int start, int len)
5647 u8 *p = map + BIT_BYTE(start);
5648 const unsigned int size = start + len;
5649 int bits_to_set = BITS_PER_BYTE - (start % BITS_PER_BYTE);
5650 u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(start);
5652 while (len - bits_to_set >= 0) {
5655 bits_to_set = BITS_PER_BYTE;
5660 mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
5665 void le_bitmap_clear(u8 *map, unsigned int start, int len)
5667 u8 *p = map + BIT_BYTE(start);
5668 const unsigned int size = start + len;
5669 int bits_to_clear = BITS_PER_BYTE - (start % BITS_PER_BYTE);
5670 u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(start);
5672 while (len - bits_to_clear >= 0) {
5673 *p &= ~mask_to_clear;
5674 len -= bits_to_clear;
5675 bits_to_clear = BITS_PER_BYTE;
5680 mask_to_clear &= BITMAP_LAST_BYTE_MASK(size);
5681 *p &= ~mask_to_clear;
5686 * eb_bitmap_offset() - calculate the page and offset of the byte containing the
5688 * @eb: the extent buffer
5689 * @start: offset of the bitmap item in the extent buffer
5691 * @page_index: return index of the page in the extent buffer that contains the
5693 * @page_offset: return offset into the page given by page_index
5695 * This helper hides the ugliness of finding the byte in an extent buffer which
5696 * contains a given bit.
5698 static inline void eb_bitmap_offset(struct extent_buffer *eb,
5699 unsigned long start, unsigned long nr,
5700 unsigned long *page_index,
5701 size_t *page_offset)
5703 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5704 size_t byte_offset = BIT_BYTE(nr);
5708 * The byte we want is the offset of the extent buffer + the offset of
5709 * the bitmap item in the extent buffer + the offset of the byte in the
5712 offset = start_offset + start + byte_offset;
5714 *page_index = offset >> PAGE_SHIFT;
5715 *page_offset = offset & (PAGE_SIZE - 1);
5719 * extent_buffer_test_bit - determine whether a bit in a bitmap item is set
5720 * @eb: the extent buffer
5721 * @start: offset of the bitmap item in the extent buffer
5722 * @nr: bit number to test
5724 int extent_buffer_test_bit(struct extent_buffer *eb, unsigned long start,
5732 eb_bitmap_offset(eb, start, nr, &i, &offset);
5733 page = eb->pages[i];
5734 WARN_ON(!PageUptodate(page));
5735 kaddr = page_address(page);
5736 return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1)));
5740 * extent_buffer_bitmap_set - set an area of a bitmap
5741 * @eb: the extent buffer
5742 * @start: offset of the bitmap item in the extent buffer
5743 * @pos: bit number of the first bit
5744 * @len: number of bits to set
5746 void extent_buffer_bitmap_set(struct extent_buffer *eb, unsigned long start,
5747 unsigned long pos, unsigned long len)
5753 const unsigned int size = pos + len;
5754 int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5755 u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos);
5757 eb_bitmap_offset(eb, start, pos, &i, &offset);
5758 page = eb->pages[i];
5759 WARN_ON(!PageUptodate(page));
5760 kaddr = page_address(page);
5762 while (len >= bits_to_set) {
5763 kaddr[offset] |= mask_to_set;
5765 bits_to_set = BITS_PER_BYTE;
5767 if (++offset >= PAGE_SIZE && len > 0) {
5769 page = eb->pages[++i];
5770 WARN_ON(!PageUptodate(page));
5771 kaddr = page_address(page);
5775 mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
5776 kaddr[offset] |= mask_to_set;
5782 * extent_buffer_bitmap_clear - clear an area of a bitmap
5783 * @eb: the extent buffer
5784 * @start: offset of the bitmap item in the extent buffer
5785 * @pos: bit number of the first bit
5786 * @len: number of bits to clear
5788 void extent_buffer_bitmap_clear(struct extent_buffer *eb, unsigned long start,
5789 unsigned long pos, unsigned long len)
5795 const unsigned int size = pos + len;
5796 int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5797 u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos);
5799 eb_bitmap_offset(eb, start, pos, &i, &offset);
5800 page = eb->pages[i];
5801 WARN_ON(!PageUptodate(page));
5802 kaddr = page_address(page);
5804 while (len >= bits_to_clear) {
5805 kaddr[offset] &= ~mask_to_clear;
5806 len -= bits_to_clear;
5807 bits_to_clear = BITS_PER_BYTE;
5809 if (++offset >= PAGE_SIZE && len > 0) {
5811 page = eb->pages[++i];
5812 WARN_ON(!PageUptodate(page));
5813 kaddr = page_address(page);
5817 mask_to_clear &= BITMAP_LAST_BYTE_MASK(size);
5818 kaddr[offset] &= ~mask_to_clear;
5822 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
5824 unsigned long distance = (src > dst) ? src - dst : dst - src;
5825 return distance < len;
5828 static void copy_pages(struct page *dst_page, struct page *src_page,
5829 unsigned long dst_off, unsigned long src_off,
5832 char *dst_kaddr = page_address(dst_page);
5834 int must_memmove = 0;
5836 if (dst_page != src_page) {
5837 src_kaddr = page_address(src_page);
5839 src_kaddr = dst_kaddr;
5840 if (areas_overlap(src_off, dst_off, len))
5845 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
5847 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
5850 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5851 unsigned long src_offset, unsigned long len)
5853 struct btrfs_fs_info *fs_info = dst->fs_info;
5855 size_t dst_off_in_page;
5856 size_t src_off_in_page;
5857 size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1);
5858 unsigned long dst_i;
5859 unsigned long src_i;
5861 if (src_offset + len > dst->len) {
5863 "memmove bogus src_offset %lu move len %lu dst len %lu",
5864 src_offset, len, dst->len);
5867 if (dst_offset + len > dst->len) {
5869 "memmove bogus dst_offset %lu move len %lu dst len %lu",
5870 dst_offset, len, dst->len);
5875 dst_off_in_page = (start_offset + dst_offset) &
5877 src_off_in_page = (start_offset + src_offset) &
5880 dst_i = (start_offset + dst_offset) >> PAGE_SHIFT;
5881 src_i = (start_offset + src_offset) >> PAGE_SHIFT;
5883 cur = min(len, (unsigned long)(PAGE_SIZE -
5885 cur = min_t(unsigned long, cur,
5886 (unsigned long)(PAGE_SIZE - dst_off_in_page));
5888 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5889 dst_off_in_page, src_off_in_page, cur);
5897 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5898 unsigned long src_offset, unsigned long len)
5900 struct btrfs_fs_info *fs_info = dst->fs_info;
5902 size_t dst_off_in_page;
5903 size_t src_off_in_page;
5904 unsigned long dst_end = dst_offset + len - 1;
5905 unsigned long src_end = src_offset + len - 1;
5906 size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1);
5907 unsigned long dst_i;
5908 unsigned long src_i;
5910 if (src_offset + len > dst->len) {
5912 "memmove bogus src_offset %lu move len %lu len %lu",
5913 src_offset, len, dst->len);
5916 if (dst_offset + len > dst->len) {
5918 "memmove bogus dst_offset %lu move len %lu len %lu",
5919 dst_offset, len, dst->len);
5922 if (dst_offset < src_offset) {
5923 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
5927 dst_i = (start_offset + dst_end) >> PAGE_SHIFT;
5928 src_i = (start_offset + src_end) >> PAGE_SHIFT;
5930 dst_off_in_page = (start_offset + dst_end) &
5932 src_off_in_page = (start_offset + src_end) &
5935 cur = min_t(unsigned long, len, src_off_in_page + 1);
5936 cur = min(cur, dst_off_in_page + 1);
5937 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5938 dst_off_in_page - cur + 1,
5939 src_off_in_page - cur + 1, cur);
5947 int try_release_extent_buffer(struct page *page)
5949 struct extent_buffer *eb;
5952 * We need to make sure nobody is attaching this page to an eb right
5955 spin_lock(&page->mapping->private_lock);
5956 if (!PagePrivate(page)) {
5957 spin_unlock(&page->mapping->private_lock);
5961 eb = (struct extent_buffer *)page->private;
5965 * This is a little awful but should be ok, we need to make sure that
5966 * the eb doesn't disappear out from under us while we're looking at
5969 spin_lock(&eb->refs_lock);
5970 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
5971 spin_unlock(&eb->refs_lock);
5972 spin_unlock(&page->mapping->private_lock);
5975 spin_unlock(&page->mapping->private_lock);
5978 * If tree ref isn't set then we know the ref on this eb is a real ref,
5979 * so just return, this page will likely be freed soon anyway.
5981 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
5982 spin_unlock(&eb->refs_lock);
5986 return release_extent_buffer(eb);