1 // SPDX-License-Identifier: GPL-2.0
3 #include <linux/bitops.h>
4 #include <linux/slab.h>
7 #include <linux/pagemap.h>
8 #include <linux/page-flags.h>
9 #include <linux/spinlock.h>
10 #include <linux/blkdev.h>
11 #include <linux/swap.h>
12 #include <linux/writeback.h>
13 #include <linux/pagevec.h>
14 #include <linux/prefetch.h>
15 #include <linux/cleancache.h>
16 #include "extent_io.h"
17 #include "extent_map.h"
19 #include "btrfs_inode.h"
21 #include "check-integrity.h"
23 #include "rcu-string.h"
27 static struct kmem_cache *extent_state_cache;
28 static struct kmem_cache *extent_buffer_cache;
29 static struct bio_set btrfs_bioset;
31 static inline bool extent_state_in_tree(const struct extent_state *state)
33 return !RB_EMPTY_NODE(&state->rb_node);
36 #ifdef CONFIG_BTRFS_DEBUG
37 static LIST_HEAD(buffers);
38 static LIST_HEAD(states);
40 static DEFINE_SPINLOCK(leak_lock);
43 void btrfs_leak_debug_add(struct list_head *new, struct list_head *head)
47 spin_lock_irqsave(&leak_lock, flags);
49 spin_unlock_irqrestore(&leak_lock, flags);
53 void btrfs_leak_debug_del(struct list_head *entry)
57 spin_lock_irqsave(&leak_lock, flags);
59 spin_unlock_irqrestore(&leak_lock, flags);
63 void btrfs_leak_debug_check(void)
65 struct extent_state *state;
66 struct extent_buffer *eb;
68 while (!list_empty(&states)) {
69 state = list_entry(states.next, struct extent_state, leak_list);
70 pr_err("BTRFS: state leak: start %llu end %llu state %u in tree %d refs %d\n",
71 state->start, state->end, state->state,
72 extent_state_in_tree(state),
73 refcount_read(&state->refs));
74 list_del(&state->leak_list);
75 kmem_cache_free(extent_state_cache, state);
78 while (!list_empty(&buffers)) {
79 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
80 pr_err("BTRFS: buffer leak start %llu len %lu refs %d bflags %lu\n",
81 eb->start, eb->len, atomic_read(&eb->refs), eb->bflags);
82 list_del(&eb->leak_list);
83 kmem_cache_free(extent_buffer_cache, eb);
87 #define btrfs_debug_check_extent_io_range(tree, start, end) \
88 __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end))
89 static inline void __btrfs_debug_check_extent_io_range(const char *caller,
90 struct extent_io_tree *tree, u64 start, u64 end)
92 if (tree->ops && tree->ops->check_extent_io_range)
93 tree->ops->check_extent_io_range(tree->private_data, caller,
97 #define btrfs_leak_debug_add(new, head) do {} while (0)
98 #define btrfs_leak_debug_del(entry) do {} while (0)
99 #define btrfs_leak_debug_check() do {} while (0)
100 #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0)
103 #define BUFFER_LRU_MAX 64
108 struct rb_node rb_node;
111 struct extent_page_data {
113 struct extent_io_tree *tree;
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 int 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,
141 static void flush_write_bio(struct extent_page_data *epd);
143 static inline struct btrfs_fs_info *
144 tree_fs_info(struct extent_io_tree *tree)
147 return tree->ops->tree_fs_info(tree->private_data);
151 int __init extent_io_init(void)
153 extent_state_cache = kmem_cache_create("btrfs_extent_state",
154 sizeof(struct extent_state), 0,
155 SLAB_MEM_SPREAD, NULL);
156 if (!extent_state_cache)
159 extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer",
160 sizeof(struct extent_buffer), 0,
161 SLAB_MEM_SPREAD, NULL);
162 if (!extent_buffer_cache)
163 goto free_state_cache;
165 if (bioset_init(&btrfs_bioset, BIO_POOL_SIZE,
166 offsetof(struct btrfs_io_bio, bio),
168 goto free_buffer_cache;
170 if (bioset_integrity_create(&btrfs_bioset, BIO_POOL_SIZE))
176 bioset_exit(&btrfs_bioset);
179 kmem_cache_destroy(extent_buffer_cache);
180 extent_buffer_cache = NULL;
183 kmem_cache_destroy(extent_state_cache);
184 extent_state_cache = NULL;
188 void __cold extent_io_exit(void)
190 btrfs_leak_debug_check();
193 * Make sure all delayed rcu free are flushed before we
197 kmem_cache_destroy(extent_state_cache);
198 kmem_cache_destroy(extent_buffer_cache);
199 bioset_exit(&btrfs_bioset);
202 void extent_io_tree_init(struct extent_io_tree *tree,
205 tree->state = RB_ROOT;
207 tree->dirty_bytes = 0;
208 spin_lock_init(&tree->lock);
209 tree->private_data = private_data;
212 static struct extent_state *alloc_extent_state(gfp_t mask)
214 struct extent_state *state;
217 * The given mask might be not appropriate for the slab allocator,
218 * drop the unsupported bits
220 mask &= ~(__GFP_DMA32|__GFP_HIGHMEM);
221 state = kmem_cache_alloc(extent_state_cache, mask);
225 state->failrec = NULL;
226 RB_CLEAR_NODE(&state->rb_node);
227 btrfs_leak_debug_add(&state->leak_list, &states);
228 refcount_set(&state->refs, 1);
229 init_waitqueue_head(&state->wq);
230 trace_alloc_extent_state(state, mask, _RET_IP_);
234 void free_extent_state(struct extent_state *state)
238 if (refcount_dec_and_test(&state->refs)) {
239 WARN_ON(extent_state_in_tree(state));
240 btrfs_leak_debug_del(&state->leak_list);
241 trace_free_extent_state(state, _RET_IP_);
242 kmem_cache_free(extent_state_cache, state);
246 static struct rb_node *tree_insert(struct rb_root *root,
247 struct rb_node *search_start,
249 struct rb_node *node,
250 struct rb_node ***p_in,
251 struct rb_node **parent_in)
254 struct rb_node *parent = NULL;
255 struct tree_entry *entry;
257 if (p_in && parent_in) {
263 p = search_start ? &search_start : &root->rb_node;
266 entry = rb_entry(parent, struct tree_entry, rb_node);
268 if (offset < entry->start)
270 else if (offset > entry->end)
277 rb_link_node(node, parent, p);
278 rb_insert_color(node, root);
282 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
283 struct rb_node **prev_ret,
284 struct rb_node **next_ret,
285 struct rb_node ***p_ret,
286 struct rb_node **parent_ret)
288 struct rb_root *root = &tree->state;
289 struct rb_node **n = &root->rb_node;
290 struct rb_node *prev = NULL;
291 struct rb_node *orig_prev = NULL;
292 struct tree_entry *entry;
293 struct tree_entry *prev_entry = NULL;
297 entry = rb_entry(prev, struct tree_entry, rb_node);
300 if (offset < entry->start)
302 else if (offset > entry->end)
315 while (prev && offset > prev_entry->end) {
316 prev = rb_next(prev);
317 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
324 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
325 while (prev && offset < prev_entry->start) {
326 prev = rb_prev(prev);
327 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
334 static inline struct rb_node *
335 tree_search_for_insert(struct extent_io_tree *tree,
337 struct rb_node ***p_ret,
338 struct rb_node **parent_ret)
340 struct rb_node *prev = NULL;
343 ret = __etree_search(tree, offset, &prev, NULL, p_ret, parent_ret);
349 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
352 return tree_search_for_insert(tree, offset, NULL, NULL);
355 static void merge_cb(struct extent_io_tree *tree, struct extent_state *new,
356 struct extent_state *other)
358 if (tree->ops && tree->ops->merge_extent_hook)
359 tree->ops->merge_extent_hook(tree->private_data, new, other);
363 * utility function to look for merge candidates inside a given range.
364 * Any extents with matching state are merged together into a single
365 * extent in the tree. Extents with EXTENT_IO in their state field
366 * are not merged because the end_io handlers need to be able to do
367 * operations on them without sleeping (or doing allocations/splits).
369 * This should be called with the tree lock held.
371 static void merge_state(struct extent_io_tree *tree,
372 struct extent_state *state)
374 struct extent_state *other;
375 struct rb_node *other_node;
377 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
380 other_node = rb_prev(&state->rb_node);
382 other = rb_entry(other_node, struct extent_state, rb_node);
383 if (other->end == state->start - 1 &&
384 other->state == state->state) {
385 merge_cb(tree, state, other);
386 state->start = other->start;
387 rb_erase(&other->rb_node, &tree->state);
388 RB_CLEAR_NODE(&other->rb_node);
389 free_extent_state(other);
392 other_node = rb_next(&state->rb_node);
394 other = rb_entry(other_node, struct extent_state, rb_node);
395 if (other->start == state->end + 1 &&
396 other->state == state->state) {
397 merge_cb(tree, state, other);
398 state->end = other->end;
399 rb_erase(&other->rb_node, &tree->state);
400 RB_CLEAR_NODE(&other->rb_node);
401 free_extent_state(other);
406 static void set_state_cb(struct extent_io_tree *tree,
407 struct extent_state *state, unsigned *bits)
409 if (tree->ops && tree->ops->set_bit_hook)
410 tree->ops->set_bit_hook(tree->private_data, state, bits);
413 static void clear_state_cb(struct extent_io_tree *tree,
414 struct extent_state *state, unsigned *bits)
416 if (tree->ops && tree->ops->clear_bit_hook)
417 tree->ops->clear_bit_hook(tree->private_data, state, bits);
420 static void set_state_bits(struct extent_io_tree *tree,
421 struct extent_state *state, unsigned *bits,
422 struct extent_changeset *changeset);
425 * insert an extent_state struct into the tree. 'bits' are set on the
426 * struct before it is inserted.
428 * This may return -EEXIST if the extent is already there, in which case the
429 * state struct is freed.
431 * The tree lock is not taken internally. This is a utility function and
432 * probably isn't what you want to call (see set/clear_extent_bit).
434 static int insert_state(struct extent_io_tree *tree,
435 struct extent_state *state, u64 start, u64 end,
437 struct rb_node **parent,
438 unsigned *bits, struct extent_changeset *changeset)
440 struct rb_node *node;
443 WARN(1, KERN_ERR "BTRFS: end < start %llu %llu\n",
445 state->start = start;
448 set_state_bits(tree, state, bits, changeset);
450 node = tree_insert(&tree->state, NULL, end, &state->rb_node, p, parent);
452 struct extent_state *found;
453 found = rb_entry(node, struct extent_state, rb_node);
454 pr_err("BTRFS: found node %llu %llu on insert of %llu %llu\n",
455 found->start, found->end, start, end);
458 merge_state(tree, state);
462 static void split_cb(struct extent_io_tree *tree, struct extent_state *orig,
465 if (tree->ops && tree->ops->split_extent_hook)
466 tree->ops->split_extent_hook(tree->private_data, orig, split);
470 * split a given extent state struct in two, inserting the preallocated
471 * struct 'prealloc' as the newly created second half. 'split' indicates an
472 * offset inside 'orig' where it should be split.
475 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
476 * are two extent state structs in the tree:
477 * prealloc: [orig->start, split - 1]
478 * orig: [ split, orig->end ]
480 * The tree locks are not taken by this function. They need to be held
483 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
484 struct extent_state *prealloc, u64 split)
486 struct rb_node *node;
488 split_cb(tree, orig, split);
490 prealloc->start = orig->start;
491 prealloc->end = split - 1;
492 prealloc->state = orig->state;
495 node = tree_insert(&tree->state, &orig->rb_node, prealloc->end,
496 &prealloc->rb_node, NULL, NULL);
498 free_extent_state(prealloc);
504 static struct extent_state *next_state(struct extent_state *state)
506 struct rb_node *next = rb_next(&state->rb_node);
508 return rb_entry(next, struct extent_state, rb_node);
514 * utility function to clear some bits in an extent state struct.
515 * it will optionally wake up any one waiting on this state (wake == 1).
517 * If no bits are set on the state struct after clearing things, the
518 * struct is freed and removed from the tree
520 static struct extent_state *clear_state_bit(struct extent_io_tree *tree,
521 struct extent_state *state,
522 unsigned *bits, int wake,
523 struct extent_changeset *changeset)
525 struct extent_state *next;
526 unsigned bits_to_clear = *bits & ~EXTENT_CTLBITS;
529 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
530 u64 range = state->end - state->start + 1;
531 WARN_ON(range > tree->dirty_bytes);
532 tree->dirty_bytes -= range;
534 clear_state_cb(tree, state, bits);
535 ret = add_extent_changeset(state, bits_to_clear, changeset, 0);
537 state->state &= ~bits_to_clear;
540 if (state->state == 0) {
541 next = next_state(state);
542 if (extent_state_in_tree(state)) {
543 rb_erase(&state->rb_node, &tree->state);
544 RB_CLEAR_NODE(&state->rb_node);
545 free_extent_state(state);
550 merge_state(tree, state);
551 next = next_state(state);
556 static struct extent_state *
557 alloc_extent_state_atomic(struct extent_state *prealloc)
560 prealloc = alloc_extent_state(GFP_ATOMIC);
565 static void extent_io_tree_panic(struct extent_io_tree *tree, int err)
567 btrfs_panic(tree_fs_info(tree), err,
568 "Locking error: Extent tree was modified by another thread while locked.");
572 * clear some bits on a range in the tree. This may require splitting
573 * or inserting elements in the tree, so the gfp mask is used to
574 * indicate which allocations or sleeping are allowed.
576 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
577 * the given range from the tree regardless of state (ie for truncate).
579 * the range [start, end] is inclusive.
581 * This takes the tree lock, and returns 0 on success and < 0 on error.
583 int __clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
584 unsigned bits, int wake, int delete,
585 struct extent_state **cached_state,
586 gfp_t mask, struct extent_changeset *changeset)
588 struct extent_state *state;
589 struct extent_state *cached;
590 struct extent_state *prealloc = NULL;
591 struct rb_node *node;
596 btrfs_debug_check_extent_io_range(tree, start, end);
598 if (bits & EXTENT_DELALLOC)
599 bits |= EXTENT_NORESERVE;
602 bits |= ~EXTENT_CTLBITS;
603 bits |= EXTENT_FIRST_DELALLOC;
605 if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY))
608 if (!prealloc && gfpflags_allow_blocking(mask)) {
610 * Don't care for allocation failure here because we might end
611 * up not needing the pre-allocated extent state at all, which
612 * is the case if we only have in the tree extent states that
613 * cover our input range and don't cover too any other range.
614 * If we end up needing a new extent state we allocate it later.
616 prealloc = alloc_extent_state(mask);
619 spin_lock(&tree->lock);
621 cached = *cached_state;
624 *cached_state = NULL;
628 if (cached && extent_state_in_tree(cached) &&
629 cached->start <= start && cached->end > start) {
631 refcount_dec(&cached->refs);
636 free_extent_state(cached);
639 * this search will find the extents that end after
642 node = tree_search(tree, start);
645 state = rb_entry(node, struct extent_state, rb_node);
647 if (state->start > end)
649 WARN_ON(state->end < start);
650 last_end = state->end;
652 /* the state doesn't have the wanted bits, go ahead */
653 if (!(state->state & bits)) {
654 state = next_state(state);
659 * | ---- desired range ---- |
661 * | ------------- state -------------- |
663 * We need to split the extent we found, and may flip
664 * bits on second half.
666 * If the extent we found extends past our range, we
667 * just split and search again. It'll get split again
668 * the next time though.
670 * If the extent we found is inside our range, we clear
671 * the desired bit on it.
674 if (state->start < start) {
675 prealloc = alloc_extent_state_atomic(prealloc);
677 err = split_state(tree, state, prealloc, start);
679 extent_io_tree_panic(tree, err);
684 if (state->end <= end) {
685 state = clear_state_bit(tree, state, &bits, wake,
692 * | ---- desired range ---- |
694 * We need to split the extent, and clear the bit
697 if (state->start <= end && state->end > end) {
698 prealloc = alloc_extent_state_atomic(prealloc);
700 err = split_state(tree, state, prealloc, end + 1);
702 extent_io_tree_panic(tree, err);
707 clear_state_bit(tree, prealloc, &bits, wake, changeset);
713 state = clear_state_bit(tree, state, &bits, wake, changeset);
715 if (last_end == (u64)-1)
717 start = last_end + 1;
718 if (start <= end && state && !need_resched())
724 spin_unlock(&tree->lock);
725 if (gfpflags_allow_blocking(mask))
730 spin_unlock(&tree->lock);
732 free_extent_state(prealloc);
738 static void wait_on_state(struct extent_io_tree *tree,
739 struct extent_state *state)
740 __releases(tree->lock)
741 __acquires(tree->lock)
744 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
745 spin_unlock(&tree->lock);
747 spin_lock(&tree->lock);
748 finish_wait(&state->wq, &wait);
752 * waits for one or more bits to clear on a range in the state tree.
753 * The range [start, end] is inclusive.
754 * The tree lock is taken by this function
756 static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
759 struct extent_state *state;
760 struct rb_node *node;
762 btrfs_debug_check_extent_io_range(tree, start, end);
764 spin_lock(&tree->lock);
768 * this search will find all the extents that end after
771 node = tree_search(tree, start);
776 state = rb_entry(node, struct extent_state, rb_node);
778 if (state->start > end)
781 if (state->state & bits) {
782 start = state->start;
783 refcount_inc(&state->refs);
784 wait_on_state(tree, state);
785 free_extent_state(state);
788 start = state->end + 1;
793 if (!cond_resched_lock(&tree->lock)) {
794 node = rb_next(node);
799 spin_unlock(&tree->lock);
802 static void set_state_bits(struct extent_io_tree *tree,
803 struct extent_state *state,
804 unsigned *bits, struct extent_changeset *changeset)
806 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 ret = add_extent_changeset(state, bits_to_set, changeset, 1);
816 state->state |= bits_to_set;
819 static void cache_state_if_flags(struct extent_state *state,
820 struct extent_state **cached_ptr,
823 if (cached_ptr && !(*cached_ptr)) {
824 if (!flags || (state->state & flags)) {
826 refcount_inc(&state->refs);
831 static void cache_state(struct extent_state *state,
832 struct extent_state **cached_ptr)
834 return cache_state_if_flags(state, cached_ptr,
835 EXTENT_IOBITS | EXTENT_BOUNDARY);
839 * set some bits on a range in the tree. This may require allocations or
840 * sleeping, so the gfp mask is used to indicate what is allowed.
842 * If any of the exclusive bits are set, this will fail with -EEXIST if some
843 * part of the range already has the desired bits set. The start of the
844 * existing range is returned in failed_start in this case.
846 * [start, end] is inclusive This takes the tree lock.
849 static int __must_check
850 __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
851 unsigned bits, unsigned exclusive_bits,
852 u64 *failed_start, struct extent_state **cached_state,
853 gfp_t mask, struct extent_changeset *changeset)
855 struct extent_state *state;
856 struct extent_state *prealloc = NULL;
857 struct rb_node *node;
859 struct rb_node *parent;
864 btrfs_debug_check_extent_io_range(tree, start, end);
866 bits |= EXTENT_FIRST_DELALLOC;
868 if (!prealloc && gfpflags_allow_blocking(mask)) {
870 * Don't care for allocation failure here because we might end
871 * up not needing the pre-allocated extent state at all, which
872 * is the case if we only have in the tree extent states that
873 * cover our input range and don't cover too any other range.
874 * If we end up needing a new extent state we allocate it later.
876 prealloc = alloc_extent_state(mask);
879 spin_lock(&tree->lock);
880 if (cached_state && *cached_state) {
881 state = *cached_state;
882 if (state->start <= start && state->end > start &&
883 extent_state_in_tree(state)) {
884 node = &state->rb_node;
889 * this search will find all the extents that end after
892 node = tree_search_for_insert(tree, start, &p, &parent);
894 prealloc = alloc_extent_state_atomic(prealloc);
896 err = insert_state(tree, prealloc, start, end,
897 &p, &parent, &bits, changeset);
899 extent_io_tree_panic(tree, err);
901 cache_state(prealloc, cached_state);
905 state = rb_entry(node, struct extent_state, rb_node);
907 last_start = state->start;
908 last_end = state->end;
911 * | ---- desired range ---- |
914 * Just lock what we found and keep going
916 if (state->start == start && state->end <= end) {
917 if (state->state & exclusive_bits) {
918 *failed_start = state->start;
923 set_state_bits(tree, state, &bits, changeset);
924 cache_state(state, cached_state);
925 merge_state(tree, state);
926 if (last_end == (u64)-1)
928 start = last_end + 1;
929 state = next_state(state);
930 if (start < end && state && state->start == start &&
937 * | ---- desired range ---- |
940 * | ------------- state -------------- |
942 * We need to split the extent we found, and may flip bits on
945 * If the extent we found extends past our
946 * range, we just split and search again. It'll get split
947 * again the next time though.
949 * If the extent we found is inside our range, we set the
952 if (state->start < start) {
953 if (state->state & exclusive_bits) {
954 *failed_start = start;
959 prealloc = alloc_extent_state_atomic(prealloc);
961 err = split_state(tree, state, prealloc, start);
963 extent_io_tree_panic(tree, err);
968 if (state->end <= end) {
969 set_state_bits(tree, state, &bits, changeset);
970 cache_state(state, cached_state);
971 merge_state(tree, state);
972 if (last_end == (u64)-1)
974 start = last_end + 1;
975 state = next_state(state);
976 if (start < end && state && state->start == start &&
983 * | ---- desired range ---- |
984 * | state | or | state |
986 * There's a hole, we need to insert something in it and
987 * ignore the extent we found.
989 if (state->start > start) {
991 if (end < last_start)
994 this_end = last_start - 1;
996 prealloc = alloc_extent_state_atomic(prealloc);
1000 * Avoid to free 'prealloc' if it can be merged with
1003 err = insert_state(tree, prealloc, start, this_end,
1004 NULL, NULL, &bits, changeset);
1006 extent_io_tree_panic(tree, err);
1008 cache_state(prealloc, cached_state);
1010 start = this_end + 1;
1014 * | ---- desired range ---- |
1016 * We need to split the extent, and set the bit
1019 if (state->start <= end && state->end > end) {
1020 if (state->state & exclusive_bits) {
1021 *failed_start = start;
1026 prealloc = alloc_extent_state_atomic(prealloc);
1028 err = split_state(tree, state, prealloc, end + 1);
1030 extent_io_tree_panic(tree, err);
1032 set_state_bits(tree, prealloc, &bits, changeset);
1033 cache_state(prealloc, cached_state);
1034 merge_state(tree, prealloc);
1042 spin_unlock(&tree->lock);
1043 if (gfpflags_allow_blocking(mask))
1048 spin_unlock(&tree->lock);
1050 free_extent_state(prealloc);
1056 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1057 unsigned bits, u64 * failed_start,
1058 struct extent_state **cached_state, gfp_t mask)
1060 return __set_extent_bit(tree, start, end, bits, 0, failed_start,
1061 cached_state, mask, NULL);
1066 * convert_extent_bit - convert all bits in a given range from one bit to
1068 * @tree: the io tree to search
1069 * @start: the start offset in bytes
1070 * @end: the end offset in bytes (inclusive)
1071 * @bits: the bits to set in this range
1072 * @clear_bits: the bits to clear in this range
1073 * @cached_state: state that we're going to cache
1075 * This will go through and set bits for the given range. If any states exist
1076 * already in this range they are set with the given bit and cleared of the
1077 * clear_bits. This is only meant to be used by things that are mergeable, ie
1078 * converting from say DELALLOC to DIRTY. This is not meant to be used with
1079 * boundary bits like LOCK.
1081 * All allocations are done with GFP_NOFS.
1083 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1084 unsigned bits, unsigned clear_bits,
1085 struct extent_state **cached_state)
1087 struct extent_state *state;
1088 struct extent_state *prealloc = NULL;
1089 struct rb_node *node;
1091 struct rb_node *parent;
1095 bool first_iteration = true;
1097 btrfs_debug_check_extent_io_range(tree, start, end);
1102 * Best effort, don't worry if extent state allocation fails
1103 * here for the first iteration. We might have a cached state
1104 * that matches exactly the target range, in which case no
1105 * extent state allocations are needed. We'll only know this
1106 * after locking the tree.
1108 prealloc = alloc_extent_state(GFP_NOFS);
1109 if (!prealloc && !first_iteration)
1113 spin_lock(&tree->lock);
1114 if (cached_state && *cached_state) {
1115 state = *cached_state;
1116 if (state->start <= start && state->end > start &&
1117 extent_state_in_tree(state)) {
1118 node = &state->rb_node;
1124 * this search will find all the extents that end after
1127 node = tree_search_for_insert(tree, start, &p, &parent);
1129 prealloc = alloc_extent_state_atomic(prealloc);
1134 err = insert_state(tree, prealloc, start, end,
1135 &p, &parent, &bits, NULL);
1137 extent_io_tree_panic(tree, err);
1138 cache_state(prealloc, cached_state);
1142 state = rb_entry(node, struct extent_state, rb_node);
1144 last_start = state->start;
1145 last_end = state->end;
1148 * | ---- desired range ---- |
1151 * Just lock what we found and keep going
1153 if (state->start == start && state->end <= end) {
1154 set_state_bits(tree, state, &bits, NULL);
1155 cache_state(state, cached_state);
1156 state = clear_state_bit(tree, state, &clear_bits, 0, NULL);
1157 if (last_end == (u64)-1)
1159 start = last_end + 1;
1160 if (start < end && state && state->start == start &&
1167 * | ---- desired range ---- |
1170 * | ------------- state -------------- |
1172 * We need to split the extent we found, and may flip bits on
1175 * If the extent we found extends past our
1176 * range, we just split and search again. It'll get split
1177 * again the next time though.
1179 * If the extent we found is inside our range, we set the
1180 * desired bit on it.
1182 if (state->start < start) {
1183 prealloc = alloc_extent_state_atomic(prealloc);
1188 err = split_state(tree, state, prealloc, start);
1190 extent_io_tree_panic(tree, err);
1194 if (state->end <= end) {
1195 set_state_bits(tree, state, &bits, NULL);
1196 cache_state(state, cached_state);
1197 state = clear_state_bit(tree, state, &clear_bits, 0,
1199 if (last_end == (u64)-1)
1201 start = last_end + 1;
1202 if (start < end && state && state->start == start &&
1209 * | ---- desired range ---- |
1210 * | state | or | state |
1212 * There's a hole, we need to insert something in it and
1213 * ignore the extent we found.
1215 if (state->start > start) {
1217 if (end < last_start)
1220 this_end = last_start - 1;
1222 prealloc = alloc_extent_state_atomic(prealloc);
1229 * Avoid to free 'prealloc' if it can be merged with
1232 err = insert_state(tree, prealloc, start, this_end,
1233 NULL, NULL, &bits, NULL);
1235 extent_io_tree_panic(tree, err);
1236 cache_state(prealloc, cached_state);
1238 start = this_end + 1;
1242 * | ---- desired range ---- |
1244 * We need to split the extent, and set the bit
1247 if (state->start <= end && state->end > end) {
1248 prealloc = alloc_extent_state_atomic(prealloc);
1254 err = split_state(tree, state, prealloc, end + 1);
1256 extent_io_tree_panic(tree, err);
1258 set_state_bits(tree, prealloc, &bits, NULL);
1259 cache_state(prealloc, cached_state);
1260 clear_state_bit(tree, prealloc, &clear_bits, 0, NULL);
1268 spin_unlock(&tree->lock);
1270 first_iteration = false;
1274 spin_unlock(&tree->lock);
1276 free_extent_state(prealloc);
1281 /* wrappers around set/clear extent bit */
1282 int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1283 unsigned bits, struct extent_changeset *changeset)
1286 * We don't support EXTENT_LOCKED yet, as current changeset will
1287 * record any bits changed, so for EXTENT_LOCKED case, it will
1288 * either fail with -EEXIST or changeset will record the whole
1291 BUG_ON(bits & EXTENT_LOCKED);
1293 return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL, GFP_NOFS,
1297 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1298 unsigned bits, int wake, int delete,
1299 struct extent_state **cached)
1301 return __clear_extent_bit(tree, start, end, bits, wake, delete,
1302 cached, GFP_NOFS, NULL);
1305 int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1306 unsigned bits, struct extent_changeset *changeset)
1309 * Don't support EXTENT_LOCKED case, same reason as
1310 * set_record_extent_bits().
1312 BUG_ON(bits & EXTENT_LOCKED);
1314 return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, GFP_NOFS,
1319 * either insert or lock state struct between start and end use mask to tell
1320 * us if waiting is desired.
1322 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1323 struct extent_state **cached_state)
1329 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED,
1330 EXTENT_LOCKED, &failed_start,
1331 cached_state, GFP_NOFS, NULL);
1332 if (err == -EEXIST) {
1333 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1334 start = failed_start;
1337 WARN_ON(start > end);
1342 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1347 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1348 &failed_start, NULL, GFP_NOFS, NULL);
1349 if (err == -EEXIST) {
1350 if (failed_start > start)
1351 clear_extent_bit(tree, start, failed_start - 1,
1352 EXTENT_LOCKED, 1, 0, NULL);
1358 void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
1360 unsigned long index = start >> PAGE_SHIFT;
1361 unsigned long end_index = end >> PAGE_SHIFT;
1364 while (index <= end_index) {
1365 page = find_get_page(inode->i_mapping, index);
1366 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1367 clear_page_dirty_for_io(page);
1373 void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
1375 unsigned long index = start >> PAGE_SHIFT;
1376 unsigned long end_index = end >> PAGE_SHIFT;
1379 while (index <= end_index) {
1380 page = find_get_page(inode->i_mapping, index);
1381 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1382 __set_page_dirty_nobuffers(page);
1383 account_page_redirty(page);
1390 * helper function to set both pages and extents in the tree writeback
1392 static void set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
1394 tree->ops->set_range_writeback(tree->private_data, start, end);
1397 /* find the first state struct with 'bits' set after 'start', and
1398 * return it. tree->lock must be held. NULL will returned if
1399 * nothing was found after 'start'
1401 static struct extent_state *
1402 find_first_extent_bit_state(struct extent_io_tree *tree,
1403 u64 start, unsigned bits)
1405 struct rb_node *node;
1406 struct extent_state *state;
1409 * this search will find all the extents that end after
1412 node = tree_search(tree, start);
1417 state = rb_entry(node, struct extent_state, rb_node);
1418 if (state->end >= start && (state->state & bits))
1421 node = rb_next(node);
1430 * find the first offset in the io tree with 'bits' set. zero is
1431 * returned if we find something, and *start_ret and *end_ret are
1432 * set to reflect the state struct that was found.
1434 * If nothing was found, 1 is returned. If found something, return 0.
1436 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1437 u64 *start_ret, u64 *end_ret, unsigned bits,
1438 struct extent_state **cached_state)
1440 struct extent_state *state;
1444 spin_lock(&tree->lock);
1445 if (cached_state && *cached_state) {
1446 state = *cached_state;
1447 if (state->end == start - 1 && extent_state_in_tree(state)) {
1448 n = rb_next(&state->rb_node);
1450 state = rb_entry(n, struct extent_state,
1452 if (state->state & bits)
1456 free_extent_state(*cached_state);
1457 *cached_state = NULL;
1460 free_extent_state(*cached_state);
1461 *cached_state = NULL;
1464 state = find_first_extent_bit_state(tree, start, bits);
1467 cache_state_if_flags(state, cached_state, 0);
1468 *start_ret = state->start;
1469 *end_ret = state->end;
1473 spin_unlock(&tree->lock);
1478 * find a contiguous range of bytes in the file marked as delalloc, not
1479 * more than 'max_bytes'. start and end are used to return the range,
1481 * 1 is returned if we find something, 0 if nothing was in the tree
1483 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1484 u64 *start, u64 *end, u64 max_bytes,
1485 struct extent_state **cached_state)
1487 struct rb_node *node;
1488 struct extent_state *state;
1489 u64 cur_start = *start;
1491 u64 total_bytes = 0;
1493 spin_lock(&tree->lock);
1496 * this search will find all the extents that end after
1499 node = tree_search(tree, cur_start);
1507 state = rb_entry(node, struct extent_state, rb_node);
1508 if (found && (state->start != cur_start ||
1509 (state->state & EXTENT_BOUNDARY))) {
1512 if (!(state->state & EXTENT_DELALLOC)) {
1518 *start = state->start;
1519 *cached_state = state;
1520 refcount_inc(&state->refs);
1524 cur_start = state->end + 1;
1525 node = rb_next(node);
1526 total_bytes += state->end - state->start + 1;
1527 if (total_bytes >= max_bytes)
1533 spin_unlock(&tree->lock);
1537 static int __process_pages_contig(struct address_space *mapping,
1538 struct page *locked_page,
1539 pgoff_t start_index, pgoff_t end_index,
1540 unsigned long page_ops, pgoff_t *index_ret);
1542 static noinline void __unlock_for_delalloc(struct inode *inode,
1543 struct page *locked_page,
1546 unsigned long index = start >> PAGE_SHIFT;
1547 unsigned long end_index = end >> PAGE_SHIFT;
1549 ASSERT(locked_page);
1550 if (index == locked_page->index && end_index == index)
1553 __process_pages_contig(inode->i_mapping, locked_page, index, end_index,
1557 static noinline int lock_delalloc_pages(struct inode *inode,
1558 struct page *locked_page,
1562 unsigned long index = delalloc_start >> PAGE_SHIFT;
1563 unsigned long index_ret = index;
1564 unsigned long end_index = delalloc_end >> PAGE_SHIFT;
1567 ASSERT(locked_page);
1568 if (index == locked_page->index && index == end_index)
1571 ret = __process_pages_contig(inode->i_mapping, locked_page, index,
1572 end_index, PAGE_LOCK, &index_ret);
1574 __unlock_for_delalloc(inode, locked_page, delalloc_start,
1575 (u64)index_ret << PAGE_SHIFT);
1580 * find a contiguous range of bytes in the file marked as delalloc, not
1581 * more than 'max_bytes'. start and end are used to return the range,
1583 * 1 is returned if we find something, 0 if nothing was in the tree
1585 STATIC u64 find_lock_delalloc_range(struct inode *inode,
1586 struct extent_io_tree *tree,
1587 struct page *locked_page, u64 *start,
1588 u64 *end, u64 max_bytes)
1593 struct extent_state *cached_state = NULL;
1598 /* step one, find a bunch of delalloc bytes starting at start */
1599 delalloc_start = *start;
1601 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1602 max_bytes, &cached_state);
1603 if (!found || delalloc_end <= *start) {
1604 *start = delalloc_start;
1605 *end = delalloc_end;
1606 free_extent_state(cached_state);
1611 * start comes from the offset of locked_page. We have to lock
1612 * pages in order, so we can't process delalloc bytes before
1615 if (delalloc_start < *start)
1616 delalloc_start = *start;
1619 * make sure to limit the number of pages we try to lock down
1621 if (delalloc_end + 1 - delalloc_start > max_bytes)
1622 delalloc_end = delalloc_start + max_bytes - 1;
1624 /* step two, lock all the pages after the page that has start */
1625 ret = lock_delalloc_pages(inode, locked_page,
1626 delalloc_start, delalloc_end);
1627 if (ret == -EAGAIN) {
1628 /* some of the pages are gone, lets avoid looping by
1629 * shortening the size of the delalloc range we're searching
1631 free_extent_state(cached_state);
1632 cached_state = NULL;
1634 max_bytes = PAGE_SIZE;
1642 BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */
1644 /* step three, lock the state bits for the whole range */
1645 lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state);
1647 /* then test to make sure it is all still delalloc */
1648 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1649 EXTENT_DELALLOC, 1, cached_state);
1651 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1653 __unlock_for_delalloc(inode, locked_page,
1654 delalloc_start, delalloc_end);
1658 free_extent_state(cached_state);
1659 *start = delalloc_start;
1660 *end = delalloc_end;
1665 static int __process_pages_contig(struct address_space *mapping,
1666 struct page *locked_page,
1667 pgoff_t start_index, pgoff_t end_index,
1668 unsigned long page_ops, pgoff_t *index_ret)
1670 unsigned long nr_pages = end_index - start_index + 1;
1671 unsigned long pages_locked = 0;
1672 pgoff_t index = start_index;
1673 struct page *pages[16];
1678 if (page_ops & PAGE_LOCK) {
1679 ASSERT(page_ops == PAGE_LOCK);
1680 ASSERT(index_ret && *index_ret == start_index);
1683 if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0)
1684 mapping_set_error(mapping, -EIO);
1686 while (nr_pages > 0) {
1687 ret = find_get_pages_contig(mapping, index,
1688 min_t(unsigned long,
1689 nr_pages, ARRAY_SIZE(pages)), pages);
1692 * Only if we're going to lock these pages,
1693 * can we find nothing at @index.
1695 ASSERT(page_ops & PAGE_LOCK);
1700 for (i = 0; i < ret; i++) {
1701 if (page_ops & PAGE_SET_PRIVATE2)
1702 SetPagePrivate2(pages[i]);
1704 if (pages[i] == locked_page) {
1709 if (page_ops & PAGE_CLEAR_DIRTY)
1710 clear_page_dirty_for_io(pages[i]);
1711 if (page_ops & PAGE_SET_WRITEBACK)
1712 set_page_writeback(pages[i]);
1713 if (page_ops & PAGE_SET_ERROR)
1714 SetPageError(pages[i]);
1715 if (page_ops & PAGE_END_WRITEBACK)
1716 end_page_writeback(pages[i]);
1717 if (page_ops & PAGE_UNLOCK)
1718 unlock_page(pages[i]);
1719 if (page_ops & PAGE_LOCK) {
1720 lock_page(pages[i]);
1721 if (!PageDirty(pages[i]) ||
1722 pages[i]->mapping != mapping) {
1723 unlock_page(pages[i]);
1737 if (err && index_ret)
1738 *index_ret = start_index + pages_locked - 1;
1742 void extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end,
1743 u64 delalloc_end, struct page *locked_page,
1744 unsigned clear_bits,
1745 unsigned long page_ops)
1747 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, clear_bits, 1, 0,
1750 __process_pages_contig(inode->i_mapping, locked_page,
1751 start >> PAGE_SHIFT, end >> PAGE_SHIFT,
1756 * count the number of bytes in the tree that have a given bit(s)
1757 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1758 * cached. The total number found is returned.
1760 u64 count_range_bits(struct extent_io_tree *tree,
1761 u64 *start, u64 search_end, u64 max_bytes,
1762 unsigned bits, int contig)
1764 struct rb_node *node;
1765 struct extent_state *state;
1766 u64 cur_start = *start;
1767 u64 total_bytes = 0;
1771 if (WARN_ON(search_end <= cur_start))
1774 spin_lock(&tree->lock);
1775 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1776 total_bytes = tree->dirty_bytes;
1780 * this search will find all the extents that end after
1783 node = tree_search(tree, cur_start);
1788 state = rb_entry(node, struct extent_state, rb_node);
1789 if (state->start > search_end)
1791 if (contig && found && state->start > last + 1)
1793 if (state->end >= cur_start && (state->state & bits) == bits) {
1794 total_bytes += min(search_end, state->end) + 1 -
1795 max(cur_start, state->start);
1796 if (total_bytes >= max_bytes)
1799 *start = max(cur_start, state->start);
1803 } else if (contig && found) {
1806 node = rb_next(node);
1811 spin_unlock(&tree->lock);
1816 * set the private field for a given byte offset in the tree. If there isn't
1817 * an extent_state there already, this does nothing.
1819 static noinline int set_state_failrec(struct extent_io_tree *tree, u64 start,
1820 struct io_failure_record *failrec)
1822 struct rb_node *node;
1823 struct extent_state *state;
1826 spin_lock(&tree->lock);
1828 * this search will find all the extents that end after
1831 node = tree_search(tree, start);
1836 state = rb_entry(node, struct extent_state, rb_node);
1837 if (state->start != start) {
1841 state->failrec = failrec;
1843 spin_unlock(&tree->lock);
1847 static noinline int get_state_failrec(struct extent_io_tree *tree, u64 start,
1848 struct io_failure_record **failrec)
1850 struct rb_node *node;
1851 struct extent_state *state;
1854 spin_lock(&tree->lock);
1856 * this search will find all the extents that end after
1859 node = tree_search(tree, start);
1864 state = rb_entry(node, struct extent_state, rb_node);
1865 if (state->start != start) {
1869 *failrec = state->failrec;
1871 spin_unlock(&tree->lock);
1876 * searches a range in the state tree for a given mask.
1877 * If 'filled' == 1, this returns 1 only if every extent in the tree
1878 * has the bits set. Otherwise, 1 is returned if any bit in the
1879 * range is found set.
1881 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1882 unsigned bits, int filled, struct extent_state *cached)
1884 struct extent_state *state = NULL;
1885 struct rb_node *node;
1888 spin_lock(&tree->lock);
1889 if (cached && extent_state_in_tree(cached) && cached->start <= start &&
1890 cached->end > start)
1891 node = &cached->rb_node;
1893 node = tree_search(tree, start);
1894 while (node && start <= end) {
1895 state = rb_entry(node, struct extent_state, rb_node);
1897 if (filled && state->start > start) {
1902 if (state->start > end)
1905 if (state->state & bits) {
1909 } else if (filled) {
1914 if (state->end == (u64)-1)
1917 start = state->end + 1;
1920 node = rb_next(node);
1927 spin_unlock(&tree->lock);
1932 * helper function to set a given page up to date if all the
1933 * extents in the tree for that page are up to date
1935 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
1937 u64 start = page_offset(page);
1938 u64 end = start + PAGE_SIZE - 1;
1939 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1940 SetPageUptodate(page);
1943 int free_io_failure(struct extent_io_tree *failure_tree,
1944 struct extent_io_tree *io_tree,
1945 struct io_failure_record *rec)
1950 set_state_failrec(failure_tree, rec->start, NULL);
1951 ret = clear_extent_bits(failure_tree, rec->start,
1952 rec->start + rec->len - 1,
1953 EXTENT_LOCKED | EXTENT_DIRTY);
1957 ret = clear_extent_bits(io_tree, rec->start,
1958 rec->start + rec->len - 1,
1968 * this bypasses the standard btrfs submit functions deliberately, as
1969 * the standard behavior is to write all copies in a raid setup. here we only
1970 * want to write the one bad copy. so we do the mapping for ourselves and issue
1971 * submit_bio directly.
1972 * to avoid any synchronization issues, wait for the data after writing, which
1973 * actually prevents the read that triggered the error from finishing.
1974 * currently, there can be no more than two copies of every data bit. thus,
1975 * exactly one rewrite is required.
1977 int repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start,
1978 u64 length, u64 logical, struct page *page,
1979 unsigned int pg_offset, int mirror_num)
1982 struct btrfs_device *dev;
1985 struct btrfs_bio *bbio = NULL;
1988 ASSERT(!(fs_info->sb->s_flags & SB_RDONLY));
1989 BUG_ON(!mirror_num);
1991 bio = btrfs_io_bio_alloc(1);
1992 bio->bi_iter.bi_size = 0;
1993 map_length = length;
1996 * Avoid races with device replace and make sure our bbio has devices
1997 * associated to its stripes that don't go away while we are doing the
1998 * read repair operation.
2000 btrfs_bio_counter_inc_blocked(fs_info);
2001 if (btrfs_is_parity_mirror(fs_info, logical, length)) {
2003 * Note that we don't use BTRFS_MAP_WRITE because it's supposed
2004 * to update all raid stripes, but here we just want to correct
2005 * bad stripe, thus BTRFS_MAP_READ is abused to only get the bad
2006 * stripe's dev and sector.
2008 ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, logical,
2009 &map_length, &bbio, 0);
2011 btrfs_bio_counter_dec(fs_info);
2015 ASSERT(bbio->mirror_num == 1);
2017 ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical,
2018 &map_length, &bbio, mirror_num);
2020 btrfs_bio_counter_dec(fs_info);
2024 BUG_ON(mirror_num != bbio->mirror_num);
2027 sector = bbio->stripes[bbio->mirror_num - 1].physical >> 9;
2028 bio->bi_iter.bi_sector = sector;
2029 dev = bbio->stripes[bbio->mirror_num - 1].dev;
2030 btrfs_put_bbio(bbio);
2031 if (!dev || !dev->bdev ||
2032 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) {
2033 btrfs_bio_counter_dec(fs_info);
2037 bio_set_dev(bio, dev->bdev);
2038 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC;
2039 bio_add_page(bio, page, length, pg_offset);
2041 if (btrfsic_submit_bio_wait(bio)) {
2042 /* try to remap that extent elsewhere? */
2043 btrfs_bio_counter_dec(fs_info);
2045 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
2049 btrfs_info_rl_in_rcu(fs_info,
2050 "read error corrected: ino %llu off %llu (dev %s sector %llu)",
2052 rcu_str_deref(dev->name), sector);
2053 btrfs_bio_counter_dec(fs_info);
2058 int repair_eb_io_failure(struct btrfs_fs_info *fs_info,
2059 struct extent_buffer *eb, int mirror_num)
2061 u64 start = eb->start;
2062 unsigned long i, num_pages = num_extent_pages(eb->start, eb->len);
2065 if (sb_rdonly(fs_info->sb))
2068 for (i = 0; i < num_pages; i++) {
2069 struct page *p = eb->pages[i];
2071 ret = repair_io_failure(fs_info, 0, start, PAGE_SIZE, start, p,
2072 start - page_offset(p), mirror_num);
2082 * each time an IO finishes, we do a fast check in the IO failure tree
2083 * to see if we need to process or clean up an io_failure_record
2085 int clean_io_failure(struct btrfs_fs_info *fs_info,
2086 struct extent_io_tree *failure_tree,
2087 struct extent_io_tree *io_tree, u64 start,
2088 struct page *page, u64 ino, unsigned int pg_offset)
2091 struct io_failure_record *failrec;
2092 struct extent_state *state;
2097 ret = count_range_bits(failure_tree, &private, (u64)-1, 1,
2102 ret = get_state_failrec(failure_tree, start, &failrec);
2106 BUG_ON(!failrec->this_mirror);
2108 if (failrec->in_validation) {
2109 /* there was no real error, just free the record */
2110 btrfs_debug(fs_info,
2111 "clean_io_failure: freeing dummy error at %llu",
2115 if (sb_rdonly(fs_info->sb))
2118 spin_lock(&io_tree->lock);
2119 state = find_first_extent_bit_state(io_tree,
2122 spin_unlock(&io_tree->lock);
2124 if (state && state->start <= failrec->start &&
2125 state->end >= failrec->start + failrec->len - 1) {
2126 num_copies = btrfs_num_copies(fs_info, failrec->logical,
2128 if (num_copies > 1) {
2129 repair_io_failure(fs_info, ino, start, failrec->len,
2130 failrec->logical, page, pg_offset,
2131 failrec->failed_mirror);
2136 free_io_failure(failure_tree, io_tree, failrec);
2142 * Can be called when
2143 * - hold extent lock
2144 * - under ordered extent
2145 * - the inode is freeing
2147 void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start, u64 end)
2149 struct extent_io_tree *failure_tree = &inode->io_failure_tree;
2150 struct io_failure_record *failrec;
2151 struct extent_state *state, *next;
2153 if (RB_EMPTY_ROOT(&failure_tree->state))
2156 spin_lock(&failure_tree->lock);
2157 state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY);
2159 if (state->start > end)
2162 ASSERT(state->end <= end);
2164 next = next_state(state);
2166 failrec = state->failrec;
2167 free_extent_state(state);
2172 spin_unlock(&failure_tree->lock);
2175 int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end,
2176 struct io_failure_record **failrec_ret)
2178 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2179 struct io_failure_record *failrec;
2180 struct extent_map *em;
2181 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2182 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2183 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2187 ret = get_state_failrec(failure_tree, start, &failrec);
2189 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2193 failrec->start = start;
2194 failrec->len = end - start + 1;
2195 failrec->this_mirror = 0;
2196 failrec->bio_flags = 0;
2197 failrec->in_validation = 0;
2199 read_lock(&em_tree->lock);
2200 em = lookup_extent_mapping(em_tree, start, failrec->len);
2202 read_unlock(&em_tree->lock);
2207 if (em->start > start || em->start + em->len <= start) {
2208 free_extent_map(em);
2211 read_unlock(&em_tree->lock);
2217 logical = start - em->start;
2218 logical = em->block_start + logical;
2219 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2220 logical = em->block_start;
2221 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2222 extent_set_compress_type(&failrec->bio_flags,
2226 btrfs_debug(fs_info,
2227 "Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu",
2228 logical, start, failrec->len);
2230 failrec->logical = logical;
2231 free_extent_map(em);
2233 /* set the bits in the private failure tree */
2234 ret = set_extent_bits(failure_tree, start, end,
2235 EXTENT_LOCKED | EXTENT_DIRTY);
2237 ret = set_state_failrec(failure_tree, start, failrec);
2238 /* set the bits in the inode's tree */
2240 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED);
2246 btrfs_debug(fs_info,
2247 "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d",
2248 failrec->logical, failrec->start, failrec->len,
2249 failrec->in_validation);
2251 * when data can be on disk more than twice, add to failrec here
2252 * (e.g. with a list for failed_mirror) to make
2253 * clean_io_failure() clean all those errors at once.
2257 *failrec_ret = failrec;
2262 bool btrfs_check_repairable(struct inode *inode, unsigned failed_bio_pages,
2263 struct io_failure_record *failrec, int failed_mirror)
2265 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2268 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
2269 if (num_copies == 1) {
2271 * we only have a single copy of the data, so don't bother with
2272 * all the retry and error correction code that follows. no
2273 * matter what the error is, it is very likely to persist.
2275 btrfs_debug(fs_info,
2276 "Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
2277 num_copies, failrec->this_mirror, failed_mirror);
2282 * there are two premises:
2283 * a) deliver good data to the caller
2284 * b) correct the bad sectors on disk
2286 if (failed_bio_pages > 1) {
2288 * to fulfill b), we need to know the exact failing sectors, as
2289 * we don't want to rewrite any more than the failed ones. thus,
2290 * we need separate read requests for the failed bio
2292 * if the following BUG_ON triggers, our validation request got
2293 * merged. we need separate requests for our algorithm to work.
2295 BUG_ON(failrec->in_validation);
2296 failrec->in_validation = 1;
2297 failrec->this_mirror = failed_mirror;
2300 * we're ready to fulfill a) and b) alongside. get a good copy
2301 * of the failed sector and if we succeed, we have setup
2302 * everything for repair_io_failure to do the rest for us.
2304 if (failrec->in_validation) {
2305 BUG_ON(failrec->this_mirror != failed_mirror);
2306 failrec->in_validation = 0;
2307 failrec->this_mirror = 0;
2309 failrec->failed_mirror = failed_mirror;
2310 failrec->this_mirror++;
2311 if (failrec->this_mirror == failed_mirror)
2312 failrec->this_mirror++;
2315 if (failrec->this_mirror > num_copies) {
2316 btrfs_debug(fs_info,
2317 "Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
2318 num_copies, failrec->this_mirror, failed_mirror);
2326 struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio,
2327 struct io_failure_record *failrec,
2328 struct page *page, int pg_offset, int icsum,
2329 bio_end_io_t *endio_func, void *data)
2331 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2333 struct btrfs_io_bio *btrfs_failed_bio;
2334 struct btrfs_io_bio *btrfs_bio;
2336 bio = btrfs_io_bio_alloc(1);
2337 bio->bi_end_io = endio_func;
2338 bio->bi_iter.bi_sector = failrec->logical >> 9;
2339 bio_set_dev(bio, fs_info->fs_devices->latest_bdev);
2340 bio->bi_iter.bi_size = 0;
2341 bio->bi_private = data;
2343 btrfs_failed_bio = btrfs_io_bio(failed_bio);
2344 if (btrfs_failed_bio->csum) {
2345 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
2347 btrfs_bio = btrfs_io_bio(bio);
2348 btrfs_bio->csum = btrfs_bio->csum_inline;
2350 memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + icsum,
2354 bio_add_page(bio, page, failrec->len, pg_offset);
2360 * this is a generic handler for readpage errors (default
2361 * readpage_io_failed_hook). if other copies exist, read those and write back
2362 * good data to the failed position. does not investigate in remapping the
2363 * failed extent elsewhere, hoping the device will be smart enough to do this as
2367 static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset,
2368 struct page *page, u64 start, u64 end,
2371 struct io_failure_record *failrec;
2372 struct inode *inode = page->mapping->host;
2373 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2374 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2377 blk_status_t status;
2379 unsigned failed_bio_pages = bio_pages_all(failed_bio);
2381 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
2383 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
2387 if (!btrfs_check_repairable(inode, failed_bio_pages, failrec,
2389 free_io_failure(failure_tree, tree, failrec);
2393 if (failed_bio_pages > 1)
2394 read_mode |= REQ_FAILFAST_DEV;
2396 phy_offset >>= inode->i_sb->s_blocksize_bits;
2397 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
2398 start - page_offset(page),
2399 (int)phy_offset, failed_bio->bi_end_io,
2401 bio_set_op_attrs(bio, REQ_OP_READ, read_mode);
2403 btrfs_debug(btrfs_sb(inode->i_sb),
2404 "Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d",
2405 read_mode, failrec->this_mirror, failrec->in_validation);
2407 status = tree->ops->submit_bio_hook(tree->private_data, bio, failrec->this_mirror,
2408 failrec->bio_flags, 0);
2410 free_io_failure(failure_tree, tree, failrec);
2412 ret = blk_status_to_errno(status);
2418 /* lots and lots of room for performance fixes in the end_bio funcs */
2420 void end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2422 int uptodate = (err == 0);
2423 struct extent_io_tree *tree;
2426 tree = &BTRFS_I(page->mapping->host)->io_tree;
2428 if (tree->ops && tree->ops->writepage_end_io_hook)
2429 tree->ops->writepage_end_io_hook(page, start, end, NULL,
2433 ClearPageUptodate(page);
2435 ret = err < 0 ? err : -EIO;
2436 mapping_set_error(page->mapping, ret);
2441 * after a writepage IO is done, we need to:
2442 * clear the uptodate bits on error
2443 * clear the writeback bits in the extent tree for this IO
2444 * end_page_writeback if the page has no more pending IO
2446 * Scheduling is not allowed, so the extent state tree is expected
2447 * to have one and only one object corresponding to this IO.
2449 static void end_bio_extent_writepage(struct bio *bio)
2451 int error = blk_status_to_errno(bio->bi_status);
2452 struct bio_vec *bvec;
2457 ASSERT(!bio_flagged(bio, BIO_CLONED));
2458 bio_for_each_segment_all(bvec, bio, i) {
2459 struct page *page = bvec->bv_page;
2460 struct inode *inode = page->mapping->host;
2461 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2463 /* We always issue full-page reads, but if some block
2464 * in a page fails to read, blk_update_request() will
2465 * advance bv_offset and adjust bv_len to compensate.
2466 * Print a warning for nonzero offsets, and an error
2467 * if they don't add up to a full page. */
2468 if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) {
2469 if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE)
2471 "partial page write in btrfs with offset %u and length %u",
2472 bvec->bv_offset, bvec->bv_len);
2475 "incomplete page write in btrfs with offset %u and length %u",
2476 bvec->bv_offset, bvec->bv_len);
2479 start = page_offset(page);
2480 end = start + bvec->bv_offset + bvec->bv_len - 1;
2482 end_extent_writepage(page, error, start, end);
2483 end_page_writeback(page);
2490 endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len,
2493 struct extent_state *cached = NULL;
2494 u64 end = start + len - 1;
2496 if (uptodate && tree->track_uptodate)
2497 set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC);
2498 unlock_extent_cached_atomic(tree, start, end, &cached);
2502 * after a readpage IO is done, we need to:
2503 * clear the uptodate bits on error
2504 * set the uptodate bits if things worked
2505 * set the page up to date if all extents in the tree are uptodate
2506 * clear the lock bit in the extent tree
2507 * unlock the page if there are no other extents locked for it
2509 * Scheduling is not allowed, so the extent state tree is expected
2510 * to have one and only one object corresponding to this IO.
2512 static void end_bio_extent_readpage(struct bio *bio)
2514 struct bio_vec *bvec;
2515 int uptodate = !bio->bi_status;
2516 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
2517 struct extent_io_tree *tree, *failure_tree;
2522 u64 extent_start = 0;
2528 ASSERT(!bio_flagged(bio, BIO_CLONED));
2529 bio_for_each_segment_all(bvec, bio, i) {
2530 struct page *page = bvec->bv_page;
2531 struct inode *inode = page->mapping->host;
2532 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2534 btrfs_debug(fs_info,
2535 "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u",
2536 (u64)bio->bi_iter.bi_sector, bio->bi_status,
2537 io_bio->mirror_num);
2538 tree = &BTRFS_I(inode)->io_tree;
2539 failure_tree = &BTRFS_I(inode)->io_failure_tree;
2541 /* We always issue full-page reads, but if some block
2542 * in a page fails to read, blk_update_request() will
2543 * advance bv_offset and adjust bv_len to compensate.
2544 * Print a warning for nonzero offsets, and an error
2545 * if they don't add up to a full page. */
2546 if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) {
2547 if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE)
2549 "partial page read in btrfs with offset %u and length %u",
2550 bvec->bv_offset, bvec->bv_len);
2553 "incomplete page read in btrfs with offset %u and length %u",
2554 bvec->bv_offset, bvec->bv_len);
2557 start = page_offset(page);
2558 end = start + bvec->bv_offset + bvec->bv_len - 1;
2561 mirror = io_bio->mirror_num;
2562 if (likely(uptodate && tree->ops)) {
2563 ret = tree->ops->readpage_end_io_hook(io_bio, offset,
2569 clean_io_failure(BTRFS_I(inode)->root->fs_info,
2570 failure_tree, tree, start,
2572 btrfs_ino(BTRFS_I(inode)), 0);
2575 if (likely(uptodate))
2579 ret = tree->ops->readpage_io_failed_hook(page, mirror);
2580 if (ret == -EAGAIN) {
2582 * Data inode's readpage_io_failed_hook() always
2585 * The generic bio_readpage_error handles errors
2586 * the following way: If possible, new read
2587 * requests are created and submitted and will
2588 * end up in end_bio_extent_readpage as well (if
2589 * we're lucky, not in the !uptodate case). In
2590 * that case it returns 0 and we just go on with
2591 * the next page in our bio. If it can't handle
2592 * the error it will return -EIO and we remain
2593 * responsible for that page.
2595 ret = bio_readpage_error(bio, offset, page,
2596 start, end, mirror);
2598 uptodate = !bio->bi_status;
2605 * metadata's readpage_io_failed_hook() always returns
2606 * -EIO and fixes nothing. -EIO is also returned if
2607 * data inode error could not be fixed.
2609 ASSERT(ret == -EIO);
2612 if (likely(uptodate)) {
2613 loff_t i_size = i_size_read(inode);
2614 pgoff_t end_index = i_size >> PAGE_SHIFT;
2617 /* Zero out the end if this page straddles i_size */
2618 off = i_size & (PAGE_SIZE-1);
2619 if (page->index == end_index && off)
2620 zero_user_segment(page, off, PAGE_SIZE);
2621 SetPageUptodate(page);
2623 ClearPageUptodate(page);
2629 if (unlikely(!uptodate)) {
2631 endio_readpage_release_extent(tree,
2637 endio_readpage_release_extent(tree, start,
2638 end - start + 1, 0);
2639 } else if (!extent_len) {
2640 extent_start = start;
2641 extent_len = end + 1 - start;
2642 } else if (extent_start + extent_len == start) {
2643 extent_len += end + 1 - start;
2645 endio_readpage_release_extent(tree, extent_start,
2646 extent_len, uptodate);
2647 extent_start = start;
2648 extent_len = end + 1 - start;
2653 endio_readpage_release_extent(tree, extent_start, extent_len,
2656 io_bio->end_io(io_bio, blk_status_to_errno(bio->bi_status));
2661 * Initialize the members up to but not including 'bio'. Use after allocating a
2662 * new bio by bio_alloc_bioset as it does not initialize the bytes outside of
2663 * 'bio' because use of __GFP_ZERO is not supported.
2665 static inline void btrfs_io_bio_init(struct btrfs_io_bio *btrfs_bio)
2667 memset(btrfs_bio, 0, offsetof(struct btrfs_io_bio, bio));
2671 * The following helpers allocate a bio. As it's backed by a bioset, it'll
2672 * never fail. We're returning a bio right now but you can call btrfs_io_bio
2673 * for the appropriate container_of magic
2675 struct bio *btrfs_bio_alloc(struct block_device *bdev, u64 first_byte)
2679 bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, &btrfs_bioset);
2680 bio_set_dev(bio, bdev);
2681 bio->bi_iter.bi_sector = first_byte >> 9;
2682 btrfs_io_bio_init(btrfs_io_bio(bio));
2686 struct bio *btrfs_bio_clone(struct bio *bio)
2688 struct btrfs_io_bio *btrfs_bio;
2691 /* Bio allocation backed by a bioset does not fail */
2692 new = bio_clone_fast(bio, GFP_NOFS, &btrfs_bioset);
2693 btrfs_bio = btrfs_io_bio(new);
2694 btrfs_io_bio_init(btrfs_bio);
2695 btrfs_bio->iter = bio->bi_iter;
2699 struct bio *btrfs_io_bio_alloc(unsigned int nr_iovecs)
2703 /* Bio allocation backed by a bioset does not fail */
2704 bio = bio_alloc_bioset(GFP_NOFS, nr_iovecs, &btrfs_bioset);
2705 btrfs_io_bio_init(btrfs_io_bio(bio));
2709 struct bio *btrfs_bio_clone_partial(struct bio *orig, int offset, int size)
2712 struct btrfs_io_bio *btrfs_bio;
2714 /* this will never fail when it's backed by a bioset */
2715 bio = bio_clone_fast(orig, GFP_NOFS, &btrfs_bioset);
2718 btrfs_bio = btrfs_io_bio(bio);
2719 btrfs_io_bio_init(btrfs_bio);
2721 bio_trim(bio, offset >> 9, size >> 9);
2722 btrfs_bio->iter = bio->bi_iter;
2726 static int __must_check submit_one_bio(struct bio *bio, int mirror_num,
2727 unsigned long bio_flags)
2729 blk_status_t ret = 0;
2730 struct bio_vec *bvec = bio_last_bvec_all(bio);
2731 struct page *page = bvec->bv_page;
2732 struct extent_io_tree *tree = bio->bi_private;
2735 start = page_offset(page) + bvec->bv_offset;
2737 bio->bi_private = NULL;
2740 ret = tree->ops->submit_bio_hook(tree->private_data, bio,
2741 mirror_num, bio_flags, start);
2743 btrfsic_submit_bio(bio);
2745 return blk_status_to_errno(ret);
2749 * @opf: bio REQ_OP_* and REQ_* flags as one value
2750 * @tree: tree so we can call our merge_bio hook
2751 * @wbc: optional writeback control for io accounting
2752 * @page: page to add to the bio
2753 * @pg_offset: offset of the new bio or to check whether we are adding
2754 * a contiguous page to the previous one
2755 * @size: portion of page that we want to write
2756 * @offset: starting offset in the page
2757 * @bdev: attach newly created bios to this bdev
2758 * @bio_ret: must be valid pointer, newly allocated bio will be stored there
2759 * @end_io_func: end_io callback for new bio
2760 * @mirror_num: desired mirror to read/write
2761 * @prev_bio_flags: flags of previous bio to see if we can merge the current one
2762 * @bio_flags: flags of the current bio to see if we can merge them
2764 static int submit_extent_page(unsigned int opf, struct extent_io_tree *tree,
2765 struct writeback_control *wbc,
2766 struct page *page, u64 offset,
2767 size_t size, unsigned long pg_offset,
2768 struct block_device *bdev,
2769 struct bio **bio_ret,
2770 bio_end_io_t end_io_func,
2772 unsigned long prev_bio_flags,
2773 unsigned long bio_flags,
2774 bool force_bio_submit)
2778 size_t page_size = min_t(size_t, size, PAGE_SIZE);
2779 sector_t sector = offset >> 9;
2785 bool can_merge = true;
2788 if (prev_bio_flags & EXTENT_BIO_COMPRESSED)
2789 contig = bio->bi_iter.bi_sector == sector;
2791 contig = bio_end_sector(bio) == sector;
2793 if (tree->ops && tree->ops->merge_bio_hook(page, offset,
2794 page_size, bio, bio_flags))
2797 if (prev_bio_flags != bio_flags || !contig || !can_merge ||
2799 bio_add_page(bio, page, page_size, pg_offset) < page_size) {
2800 ret = submit_one_bio(bio, mirror_num, prev_bio_flags);
2808 wbc_account_io(wbc, page, page_size);
2813 bio = btrfs_bio_alloc(bdev, offset);
2814 bio_add_page(bio, page, page_size, pg_offset);
2815 bio->bi_end_io = end_io_func;
2816 bio->bi_private = tree;
2817 bio->bi_write_hint = page->mapping->host->i_write_hint;
2820 wbc_init_bio(wbc, bio);
2821 wbc_account_io(wbc, page, page_size);
2829 static void attach_extent_buffer_page(struct extent_buffer *eb,
2832 if (!PagePrivate(page)) {
2833 SetPagePrivate(page);
2835 set_page_private(page, (unsigned long)eb);
2837 WARN_ON(page->private != (unsigned long)eb);
2841 void set_page_extent_mapped(struct page *page)
2843 if (!PagePrivate(page)) {
2844 SetPagePrivate(page);
2846 set_page_private(page, EXTENT_PAGE_PRIVATE);
2850 static struct extent_map *
2851 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
2852 u64 start, u64 len, get_extent_t *get_extent,
2853 struct extent_map **em_cached)
2855 struct extent_map *em;
2857 if (em_cached && *em_cached) {
2859 if (extent_map_in_tree(em) && start >= em->start &&
2860 start < extent_map_end(em)) {
2861 refcount_inc(&em->refs);
2865 free_extent_map(em);
2869 em = get_extent(BTRFS_I(inode), page, pg_offset, start, len, 0);
2870 if (em_cached && !IS_ERR_OR_NULL(em)) {
2872 refcount_inc(&em->refs);
2878 * basic readpage implementation. Locked extent state structs are inserted
2879 * into the tree that are removed when the IO is done (by the end_io
2881 * XXX JDM: This needs looking at to ensure proper page locking
2882 * return 0 on success, otherwise return error
2884 static int __do_readpage(struct extent_io_tree *tree,
2886 get_extent_t *get_extent,
2887 struct extent_map **em_cached,
2888 struct bio **bio, int mirror_num,
2889 unsigned long *bio_flags, unsigned int read_flags,
2892 struct inode *inode = page->mapping->host;
2893 u64 start = page_offset(page);
2894 const u64 end = start + PAGE_SIZE - 1;
2897 u64 last_byte = i_size_read(inode);
2900 struct extent_map *em;
2901 struct block_device *bdev;
2904 size_t pg_offset = 0;
2906 size_t disk_io_size;
2907 size_t blocksize = inode->i_sb->s_blocksize;
2908 unsigned long this_bio_flag = 0;
2910 set_page_extent_mapped(page);
2912 if (!PageUptodate(page)) {
2913 if (cleancache_get_page(page) == 0) {
2914 BUG_ON(blocksize != PAGE_SIZE);
2915 unlock_extent(tree, start, end);
2920 if (page->index == last_byte >> PAGE_SHIFT) {
2922 size_t zero_offset = last_byte & (PAGE_SIZE - 1);
2925 iosize = PAGE_SIZE - zero_offset;
2926 userpage = kmap_atomic(page);
2927 memset(userpage + zero_offset, 0, iosize);
2928 flush_dcache_page(page);
2929 kunmap_atomic(userpage);
2932 while (cur <= end) {
2933 bool force_bio_submit = false;
2936 if (cur >= last_byte) {
2938 struct extent_state *cached = NULL;
2940 iosize = PAGE_SIZE - pg_offset;
2941 userpage = kmap_atomic(page);
2942 memset(userpage + pg_offset, 0, iosize);
2943 flush_dcache_page(page);
2944 kunmap_atomic(userpage);
2945 set_extent_uptodate(tree, cur, cur + iosize - 1,
2947 unlock_extent_cached(tree, cur,
2948 cur + iosize - 1, &cached);
2951 em = __get_extent_map(inode, page, pg_offset, cur,
2952 end - cur + 1, get_extent, em_cached);
2953 if (IS_ERR_OR_NULL(em)) {
2955 unlock_extent(tree, cur, end);
2958 extent_offset = cur - em->start;
2959 BUG_ON(extent_map_end(em) <= cur);
2962 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2963 this_bio_flag |= EXTENT_BIO_COMPRESSED;
2964 extent_set_compress_type(&this_bio_flag,
2968 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2969 cur_end = min(extent_map_end(em) - 1, end);
2970 iosize = ALIGN(iosize, blocksize);
2971 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2972 disk_io_size = em->block_len;
2973 offset = em->block_start;
2975 offset = em->block_start + extent_offset;
2976 disk_io_size = iosize;
2979 block_start = em->block_start;
2980 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2981 block_start = EXTENT_MAP_HOLE;
2984 * If we have a file range that points to a compressed extent
2985 * and it's followed by a consecutive file range that points to
2986 * to the same compressed extent (possibly with a different
2987 * offset and/or length, so it either points to the whole extent
2988 * or only part of it), we must make sure we do not submit a
2989 * single bio to populate the pages for the 2 ranges because
2990 * this makes the compressed extent read zero out the pages
2991 * belonging to the 2nd range. Imagine the following scenario:
2994 * [0 - 8K] [8K - 24K]
2997 * points to extent X, points to extent X,
2998 * offset 4K, length of 8K offset 0, length 16K
3000 * [extent X, compressed length = 4K uncompressed length = 16K]
3002 * If the bio to read the compressed extent covers both ranges,
3003 * it will decompress extent X into the pages belonging to the
3004 * first range and then it will stop, zeroing out the remaining
3005 * pages that belong to the other range that points to extent X.
3006 * So here we make sure we submit 2 bios, one for the first
3007 * range and another one for the third range. Both will target
3008 * the same physical extent from disk, but we can't currently
3009 * make the compressed bio endio callback populate the pages
3010 * for both ranges because each compressed bio is tightly
3011 * coupled with a single extent map, and each range can have
3012 * an extent map with a different offset value relative to the
3013 * uncompressed data of our extent and different lengths. This
3014 * is a corner case so we prioritize correctness over
3015 * non-optimal behavior (submitting 2 bios for the same extent).
3017 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) &&
3018 prev_em_start && *prev_em_start != (u64)-1 &&
3019 *prev_em_start != em->orig_start)
3020 force_bio_submit = true;
3023 *prev_em_start = em->orig_start;
3025 free_extent_map(em);
3028 /* we've found a hole, just zero and go on */
3029 if (block_start == EXTENT_MAP_HOLE) {
3031 struct extent_state *cached = NULL;
3033 userpage = kmap_atomic(page);
3034 memset(userpage + pg_offset, 0, iosize);
3035 flush_dcache_page(page);
3036 kunmap_atomic(userpage);
3038 set_extent_uptodate(tree, cur, cur + iosize - 1,
3040 unlock_extent_cached(tree, cur,
3041 cur + iosize - 1, &cached);
3043 pg_offset += iosize;
3046 /* the get_extent function already copied into the page */
3047 if (test_range_bit(tree, cur, cur_end,
3048 EXTENT_UPTODATE, 1, NULL)) {
3049 check_page_uptodate(tree, page);
3050 unlock_extent(tree, cur, cur + iosize - 1);
3052 pg_offset += iosize;
3055 /* we have an inline extent but it didn't get marked up
3056 * to date. Error out
3058 if (block_start == EXTENT_MAP_INLINE) {
3060 unlock_extent(tree, cur, cur + iosize - 1);
3062 pg_offset += iosize;
3066 ret = submit_extent_page(REQ_OP_READ | read_flags, tree, NULL,
3067 page, offset, disk_io_size,
3068 pg_offset, bdev, bio,
3069 end_bio_extent_readpage, mirror_num,
3075 *bio_flags = this_bio_flag;
3078 unlock_extent(tree, cur, cur + iosize - 1);
3082 pg_offset += iosize;
3086 if (!PageError(page))
3087 SetPageUptodate(page);
3093 static inline void __do_contiguous_readpages(struct extent_io_tree *tree,
3094 struct page *pages[], int nr_pages,
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], btrfs_get_extent, em_cached,
3119 bio, 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[],
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,
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, 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 = btrfs_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,
3459 struct extent_page_data *epd)
3461 struct inode *inode = page->mapping->host;
3462 u64 start = page_offset(page);
3463 u64 page_end = start + PAGE_SIZE - 1;
3466 size_t pg_offset = 0;
3467 loff_t i_size = i_size_read(inode);
3468 unsigned long end_index = i_size >> PAGE_SHIFT;
3469 unsigned int write_flags = 0;
3470 unsigned long nr_written = 0;
3472 write_flags = wbc_to_write_flags(wbc);
3474 trace___extent_writepage(page, inode, wbc);
3476 WARN_ON(!PageLocked(page));
3478 ClearPageError(page);
3480 pg_offset = i_size & (PAGE_SIZE - 1);
3481 if (page->index > end_index ||
3482 (page->index == end_index && !pg_offset)) {
3483 page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
3488 if (page->index == end_index) {
3491 userpage = kmap_atomic(page);
3492 memset(userpage + pg_offset, 0,
3493 PAGE_SIZE - pg_offset);
3494 kunmap_atomic(userpage);
3495 flush_dcache_page(page);
3500 set_page_extent_mapped(page);
3502 ret = writepage_delalloc(inode, page, wbc, epd, start, &nr_written);
3508 ret = __extent_writepage_io(inode, page, wbc, epd,
3509 i_size, nr_written, write_flags, &nr);
3515 /* make sure the mapping tag for page dirty gets cleared */
3516 set_page_writeback(page);
3517 end_page_writeback(page);
3519 if (PageError(page)) {
3520 ret = ret < 0 ? ret : -EIO;
3521 end_extent_writepage(page, ret, start, page_end);
3530 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3532 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
3533 TASK_UNINTERRUPTIBLE);
3536 static noinline_for_stack int
3537 lock_extent_buffer_for_io(struct extent_buffer *eb,
3538 struct btrfs_fs_info *fs_info,
3539 struct extent_page_data *epd)
3541 unsigned long i, num_pages;
3545 if (!btrfs_try_tree_write_lock(eb)) {
3547 flush_write_bio(epd);
3548 btrfs_tree_lock(eb);
3551 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3552 btrfs_tree_unlock(eb);
3556 flush_write_bio(epd);
3560 wait_on_extent_buffer_writeback(eb);
3561 btrfs_tree_lock(eb);
3562 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3564 btrfs_tree_unlock(eb);
3569 * We need to do this to prevent races in people who check if the eb is
3570 * under IO since we can end up having no IO bits set for a short period
3573 spin_lock(&eb->refs_lock);
3574 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3575 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3576 spin_unlock(&eb->refs_lock);
3577 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3578 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
3580 fs_info->dirty_metadata_batch);
3583 spin_unlock(&eb->refs_lock);
3586 btrfs_tree_unlock(eb);
3591 num_pages = num_extent_pages(eb->start, eb->len);
3592 for (i = 0; i < num_pages; i++) {
3593 struct page *p = eb->pages[i];
3595 if (!trylock_page(p)) {
3597 flush_write_bio(epd);
3607 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3609 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3610 smp_mb__after_atomic();
3611 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3614 static void set_btree_ioerr(struct page *page)
3616 struct extent_buffer *eb = (struct extent_buffer *)page->private;
3619 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))
3623 * If writeback for a btree extent that doesn't belong to a log tree
3624 * failed, increment the counter transaction->eb_write_errors.
3625 * We do this because while the transaction is running and before it's
3626 * committing (when we call filemap_fdata[write|wait]_range against
3627 * the btree inode), we might have
3628 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
3629 * returns an error or an error happens during writeback, when we're
3630 * committing the transaction we wouldn't know about it, since the pages
3631 * can be no longer dirty nor marked anymore for writeback (if a
3632 * subsequent modification to the extent buffer didn't happen before the
3633 * transaction commit), which makes filemap_fdata[write|wait]_range not
3634 * able to find the pages tagged with SetPageError at transaction
3635 * commit time. So if this happens we must abort the transaction,
3636 * otherwise we commit a super block with btree roots that point to
3637 * btree nodes/leafs whose content on disk is invalid - either garbage
3638 * or the content of some node/leaf from a past generation that got
3639 * cowed or deleted and is no longer valid.
3641 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
3642 * not be enough - we need to distinguish between log tree extents vs
3643 * non-log tree extents, and the next filemap_fdatawait_range() call
3644 * will catch and clear such errors in the mapping - and that call might
3645 * be from a log sync and not from a transaction commit. Also, checking
3646 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
3647 * not done and would not be reliable - the eb might have been released
3648 * from memory and reading it back again means that flag would not be
3649 * set (since it's a runtime flag, not persisted on disk).
3651 * Using the flags below in the btree inode also makes us achieve the
3652 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
3653 * writeback for all dirty pages and before filemap_fdatawait_range()
3654 * is called, the writeback for all dirty pages had already finished
3655 * with errors - because we were not using AS_EIO/AS_ENOSPC,
3656 * filemap_fdatawait_range() would return success, as it could not know
3657 * that writeback errors happened (the pages were no longer tagged for
3660 switch (eb->log_index) {
3662 set_bit(BTRFS_FS_BTREE_ERR, &eb->fs_info->flags);
3665 set_bit(BTRFS_FS_LOG1_ERR, &eb->fs_info->flags);
3668 set_bit(BTRFS_FS_LOG2_ERR, &eb->fs_info->flags);
3671 BUG(); /* unexpected, logic error */
3675 static void end_bio_extent_buffer_writepage(struct bio *bio)
3677 struct bio_vec *bvec;
3678 struct extent_buffer *eb;
3681 ASSERT(!bio_flagged(bio, BIO_CLONED));
3682 bio_for_each_segment_all(bvec, bio, i) {
3683 struct page *page = bvec->bv_page;
3685 eb = (struct extent_buffer *)page->private;
3687 done = atomic_dec_and_test(&eb->io_pages);
3689 if (bio->bi_status ||
3690 test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) {
3691 ClearPageUptodate(page);
3692 set_btree_ioerr(page);
3695 end_page_writeback(page);
3700 end_extent_buffer_writeback(eb);
3706 static noinline_for_stack int write_one_eb(struct extent_buffer *eb,
3707 struct btrfs_fs_info *fs_info,
3708 struct writeback_control *wbc,
3709 struct extent_page_data *epd)
3711 struct block_device *bdev = fs_info->fs_devices->latest_bdev;
3712 struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
3713 u64 offset = eb->start;
3715 unsigned long i, num_pages;
3716 unsigned long start, end;
3717 unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META;
3720 clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
3721 num_pages = num_extent_pages(eb->start, eb->len);
3722 atomic_set(&eb->io_pages, num_pages);
3724 /* set btree blocks beyond nritems with 0 to avoid stale content. */
3725 nritems = btrfs_header_nritems(eb);
3726 if (btrfs_header_level(eb) > 0) {
3727 end = btrfs_node_key_ptr_offset(nritems);
3729 memzero_extent_buffer(eb, end, eb->len - end);
3733 * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0
3735 start = btrfs_item_nr_offset(nritems);
3736 end = BTRFS_LEAF_DATA_OFFSET + leaf_data_end(fs_info, eb);
3737 memzero_extent_buffer(eb, start, end - start);
3740 for (i = 0; i < num_pages; i++) {
3741 struct page *p = eb->pages[i];
3743 clear_page_dirty_for_io(p);
3744 set_page_writeback(p);
3745 ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc,
3746 p, offset, PAGE_SIZE, 0, bdev,
3748 end_bio_extent_buffer_writepage,
3752 if (PageWriteback(p))
3753 end_page_writeback(p);
3754 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3755 end_extent_buffer_writeback(eb);
3759 offset += PAGE_SIZE;
3760 update_nr_written(wbc, 1);
3764 if (unlikely(ret)) {
3765 for (; i < num_pages; i++) {
3766 struct page *p = eb->pages[i];
3767 clear_page_dirty_for_io(p);
3775 int btree_write_cache_pages(struct address_space *mapping,
3776 struct writeback_control *wbc)
3778 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
3779 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
3780 struct extent_buffer *eb, *prev_eb = NULL;
3781 struct extent_page_data epd = {
3785 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3789 int nr_to_write_done = 0;
3790 struct pagevec pvec;
3793 pgoff_t end; /* Inclusive */
3797 pagevec_init(&pvec);
3798 if (wbc->range_cyclic) {
3799 index = mapping->writeback_index; /* Start from prev offset */
3802 index = wbc->range_start >> PAGE_SHIFT;
3803 end = wbc->range_end >> PAGE_SHIFT;
3806 if (wbc->sync_mode == WB_SYNC_ALL)
3807 tag = PAGECACHE_TAG_TOWRITE;
3809 tag = PAGECACHE_TAG_DIRTY;
3811 if (wbc->sync_mode == WB_SYNC_ALL)
3812 tag_pages_for_writeback(mapping, index, end);
3813 while (!done && !nr_to_write_done && (index <= end) &&
3814 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
3819 for (i = 0; i < nr_pages; i++) {
3820 struct page *page = pvec.pages[i];
3822 if (!PagePrivate(page))
3825 spin_lock(&mapping->private_lock);
3826 if (!PagePrivate(page)) {
3827 spin_unlock(&mapping->private_lock);
3831 eb = (struct extent_buffer *)page->private;
3834 * Shouldn't happen and normally this would be a BUG_ON
3835 * but no sense in crashing the users box for something
3836 * we can survive anyway.
3839 spin_unlock(&mapping->private_lock);
3843 if (eb == prev_eb) {
3844 spin_unlock(&mapping->private_lock);
3848 ret = atomic_inc_not_zero(&eb->refs);
3849 spin_unlock(&mapping->private_lock);
3854 ret = lock_extent_buffer_for_io(eb, fs_info, &epd);
3856 free_extent_buffer(eb);
3860 ret = write_one_eb(eb, fs_info, wbc, &epd);
3863 free_extent_buffer(eb);
3866 free_extent_buffer(eb);
3869 * the filesystem may choose to bump up nr_to_write.
3870 * We have to make sure to honor the new nr_to_write
3873 nr_to_write_done = wbc->nr_to_write <= 0;
3875 pagevec_release(&pvec);
3878 if (!scanned && !done) {
3880 * We hit the last page and there is more work to be done: wrap
3881 * back to the start of the file
3887 flush_write_bio(&epd);
3892 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3893 * @mapping: address space structure to write
3894 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3895 * @data: data passed to __extent_writepage function
3897 * If a page is already under I/O, write_cache_pages() skips it, even
3898 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3899 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3900 * and msync() need to guarantee that all the data which was dirty at the time
3901 * the call was made get new I/O started against them. If wbc->sync_mode is
3902 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3903 * existing IO to complete.
3905 static int extent_write_cache_pages(struct address_space *mapping,
3906 struct writeback_control *wbc,
3907 struct extent_page_data *epd)
3909 struct inode *inode = mapping->host;
3912 int nr_to_write_done = 0;
3913 struct pagevec pvec;
3916 pgoff_t end; /* Inclusive */
3918 int range_whole = 0;
3923 * We have to hold onto the inode so that ordered extents can do their
3924 * work when the IO finishes. The alternative to this is failing to add
3925 * an ordered extent if the igrab() fails there and that is a huge pain
3926 * to deal with, so instead just hold onto the inode throughout the
3927 * writepages operation. If it fails here we are freeing up the inode
3928 * anyway and we'd rather not waste our time writing out stuff that is
3929 * going to be truncated anyway.
3934 pagevec_init(&pvec);
3935 if (wbc->range_cyclic) {
3936 index = mapping->writeback_index; /* Start from prev offset */
3939 index = wbc->range_start >> PAGE_SHIFT;
3940 end = wbc->range_end >> PAGE_SHIFT;
3941 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
3945 if (wbc->sync_mode == WB_SYNC_ALL)
3946 tag = PAGECACHE_TAG_TOWRITE;
3948 tag = PAGECACHE_TAG_DIRTY;
3950 if (wbc->sync_mode == WB_SYNC_ALL)
3951 tag_pages_for_writeback(mapping, index, end);
3953 while (!done && !nr_to_write_done && (index <= end) &&
3954 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping,
3955 &index, end, tag))) {
3959 for (i = 0; i < nr_pages; i++) {
3960 struct page *page = pvec.pages[i];
3962 done_index = page->index;
3964 * At this point we hold neither the i_pages lock nor
3965 * the page lock: the page may be truncated or
3966 * invalidated (changing page->mapping to NULL),
3967 * or even swizzled back from swapper_space to
3968 * tmpfs file mapping
3970 if (!trylock_page(page)) {
3971 flush_write_bio(epd);
3975 if (unlikely(page->mapping != mapping)) {
3980 if (wbc->sync_mode != WB_SYNC_NONE) {
3981 if (PageWriteback(page))
3982 flush_write_bio(epd);
3983 wait_on_page_writeback(page);
3986 if (PageWriteback(page) ||
3987 !clear_page_dirty_for_io(page)) {
3992 ret = __extent_writepage(page, wbc, epd);
3994 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
4000 * done_index is set past this page,
4001 * so media errors will not choke
4002 * background writeout for the entire
4003 * file. This has consequences for
4004 * range_cyclic semantics (ie. it may
4005 * not be suitable for data integrity
4008 done_index = page->index + 1;
4014 * the filesystem may choose to bump up nr_to_write.
4015 * We have to make sure to honor the new nr_to_write
4018 nr_to_write_done = wbc->nr_to_write <= 0;
4020 pagevec_release(&pvec);
4023 if (!scanned && !done) {
4025 * We hit the last page and there is more work to be done: wrap
4026 * back to the start of the file
4033 if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole))
4034 mapping->writeback_index = done_index;
4036 btrfs_add_delayed_iput(inode);
4040 static void flush_write_bio(struct extent_page_data *epd)
4045 ret = submit_one_bio(epd->bio, 0, 0);
4046 BUG_ON(ret < 0); /* -ENOMEM */
4051 int extent_write_full_page(struct page *page, struct writeback_control *wbc)
4054 struct extent_page_data epd = {
4056 .tree = &BTRFS_I(page->mapping->host)->io_tree,
4058 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4061 ret = __extent_writepage(page, wbc, &epd);
4063 flush_write_bio(&epd);
4067 int extent_write_locked_range(struct inode *inode, u64 start, u64 end,
4071 struct address_space *mapping = inode->i_mapping;
4072 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
4074 unsigned long nr_pages = (end - start + PAGE_SIZE) >>
4077 struct extent_page_data epd = {
4081 .sync_io = mode == WB_SYNC_ALL,
4083 struct writeback_control wbc_writepages = {
4085 .nr_to_write = nr_pages * 2,
4086 .range_start = start,
4087 .range_end = end + 1,
4090 while (start <= end) {
4091 page = find_get_page(mapping, start >> PAGE_SHIFT);
4092 if (clear_page_dirty_for_io(page))
4093 ret = __extent_writepage(page, &wbc_writepages, &epd);
4095 if (tree->ops && tree->ops->writepage_end_io_hook)
4096 tree->ops->writepage_end_io_hook(page, start,
4097 start + PAGE_SIZE - 1,
4105 flush_write_bio(&epd);
4109 int extent_writepages(struct address_space *mapping,
4110 struct writeback_control *wbc)
4113 struct extent_page_data epd = {
4115 .tree = &BTRFS_I(mapping->host)->io_tree,
4117 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4120 ret = extent_write_cache_pages(mapping, wbc, &epd);
4121 flush_write_bio(&epd);
4125 int extent_readpages(struct address_space *mapping, struct list_head *pages,
4128 struct bio *bio = NULL;
4130 unsigned long bio_flags = 0;
4131 struct page *pagepool[16];
4133 struct extent_map *em_cached = NULL;
4134 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
4136 u64 prev_em_start = (u64)-1;
4138 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
4139 page = list_entry(pages->prev, struct page, lru);
4141 prefetchw(&page->flags);
4142 list_del(&page->lru);
4143 if (add_to_page_cache_lru(page, mapping,
4145 readahead_gfp_mask(mapping))) {
4150 pagepool[nr++] = page;
4151 if (nr < ARRAY_SIZE(pagepool))
4153 __extent_readpages(tree, pagepool, nr, &em_cached, &bio,
4154 &bio_flags, &prev_em_start);
4158 __extent_readpages(tree, pagepool, nr, &em_cached, &bio,
4159 &bio_flags, &prev_em_start);
4162 free_extent_map(em_cached);
4164 BUG_ON(!list_empty(pages));
4166 return submit_one_bio(bio, 0, bio_flags);
4171 * basic invalidatepage code, this waits on any locked or writeback
4172 * ranges corresponding to the page, and then deletes any extent state
4173 * records from the tree
4175 int extent_invalidatepage(struct extent_io_tree *tree,
4176 struct page *page, unsigned long offset)
4178 struct extent_state *cached_state = NULL;
4179 u64 start = page_offset(page);
4180 u64 end = start + PAGE_SIZE - 1;
4181 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
4183 start += ALIGN(offset, blocksize);
4187 lock_extent_bits(tree, start, end, &cached_state);
4188 wait_on_page_writeback(page);
4189 clear_extent_bit(tree, start, end,
4190 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4191 EXTENT_DO_ACCOUNTING,
4192 1, 1, &cached_state);
4197 * a helper for releasepage, this tests for areas of the page that
4198 * are locked or under IO and drops the related state bits if it is safe
4201 static int try_release_extent_state(struct extent_io_tree *tree,
4202 struct page *page, gfp_t mask)
4204 u64 start = page_offset(page);
4205 u64 end = start + PAGE_SIZE - 1;
4208 if (test_range_bit(tree, start, end,
4209 EXTENT_IOBITS, 0, NULL))
4213 * at this point we can safely clear everything except the
4214 * locked bit and the nodatasum bit
4216 ret = __clear_extent_bit(tree, start, end,
4217 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
4218 0, 0, NULL, mask, NULL);
4220 /* if clear_extent_bit failed for enomem reasons,
4221 * we can't allow the release to continue.
4232 * a helper for releasepage. As long as there are no locked extents
4233 * in the range corresponding to the page, both state records and extent
4234 * map records are removed
4236 int try_release_extent_mapping(struct page *page, gfp_t mask)
4238 struct extent_map *em;
4239 u64 start = page_offset(page);
4240 u64 end = start + PAGE_SIZE - 1;
4241 struct extent_io_tree *tree = &BTRFS_I(page->mapping->host)->io_tree;
4242 struct extent_map_tree *map = &BTRFS_I(page->mapping->host)->extent_tree;
4244 if (gfpflags_allow_blocking(mask) &&
4245 page->mapping->host->i_size > SZ_16M) {
4247 while (start <= end) {
4248 len = end - start + 1;
4249 write_lock(&map->lock);
4250 em = lookup_extent_mapping(map, start, len);
4252 write_unlock(&map->lock);
4255 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
4256 em->start != start) {
4257 write_unlock(&map->lock);
4258 free_extent_map(em);
4261 if (!test_range_bit(tree, em->start,
4262 extent_map_end(em) - 1,
4263 EXTENT_LOCKED | EXTENT_WRITEBACK,
4265 remove_extent_mapping(map, em);
4266 /* once for the rb tree */
4267 free_extent_map(em);
4269 start = extent_map_end(em);
4270 write_unlock(&map->lock);
4273 free_extent_map(em);
4276 return try_release_extent_state(tree, page, mask);
4280 * helper function for fiemap, which doesn't want to see any holes.
4281 * This maps until we find something past 'last'
4283 static struct extent_map *get_extent_skip_holes(struct inode *inode,
4284 u64 offset, u64 last)
4286 u64 sectorsize = btrfs_inode_sectorsize(inode);
4287 struct extent_map *em;
4294 len = last - offset;
4297 len = ALIGN(len, sectorsize);
4298 em = btrfs_get_extent_fiemap(BTRFS_I(inode), NULL, 0, offset,
4300 if (IS_ERR_OR_NULL(em))
4303 /* if this isn't a hole return it */
4304 if (em->block_start != EXTENT_MAP_HOLE)
4307 /* this is a hole, advance to the next extent */
4308 offset = extent_map_end(em);
4309 free_extent_map(em);
4317 * To cache previous fiemap extent
4319 * Will be used for merging fiemap extent
4321 struct fiemap_cache {
4330 * Helper to submit fiemap extent.
4332 * Will try to merge current fiemap extent specified by @offset, @phys,
4333 * @len and @flags with cached one.
4334 * And only when we fails to merge, cached one will be submitted as
4337 * Return value is the same as fiemap_fill_next_extent().
4339 static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
4340 struct fiemap_cache *cache,
4341 u64 offset, u64 phys, u64 len, u32 flags)
4349 * Sanity check, extent_fiemap() should have ensured that new
4350 * fiemap extent won't overlap with cahced one.
4353 * NOTE: Physical address can overlap, due to compression
4355 if (cache->offset + cache->len > offset) {
4361 * Only merges fiemap extents if
4362 * 1) Their logical addresses are continuous
4364 * 2) Their physical addresses are continuous
4365 * So truly compressed (physical size smaller than logical size)
4366 * extents won't get merged with each other
4368 * 3) Share same flags except FIEMAP_EXTENT_LAST
4369 * So regular extent won't get merged with prealloc extent
4371 if (cache->offset + cache->len == offset &&
4372 cache->phys + cache->len == phys &&
4373 (cache->flags & ~FIEMAP_EXTENT_LAST) ==
4374 (flags & ~FIEMAP_EXTENT_LAST)) {
4376 cache->flags |= flags;
4377 goto try_submit_last;
4380 /* Not mergeable, need to submit cached one */
4381 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
4382 cache->len, cache->flags);
4383 cache->cached = false;
4387 cache->cached = true;
4388 cache->offset = offset;
4391 cache->flags = flags;
4393 if (cache->flags & FIEMAP_EXTENT_LAST) {
4394 ret = fiemap_fill_next_extent(fieinfo, cache->offset,
4395 cache->phys, cache->len, cache->flags);
4396 cache->cached = false;
4402 * Emit last fiemap cache
4404 * The last fiemap cache may still be cached in the following case:
4406 * |<- Fiemap range ->|
4407 * |<------------ First extent ----------->|
4409 * In this case, the first extent range will be cached but not emitted.
4410 * So we must emit it before ending extent_fiemap().
4412 static int emit_last_fiemap_cache(struct btrfs_fs_info *fs_info,
4413 struct fiemap_extent_info *fieinfo,
4414 struct fiemap_cache *cache)
4421 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
4422 cache->len, cache->flags);
4423 cache->cached = false;
4429 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4430 __u64 start, __u64 len)
4434 u64 max = start + len;
4438 u64 last_for_get_extent = 0;
4440 u64 isize = i_size_read(inode);
4441 struct btrfs_key found_key;
4442 struct extent_map *em = NULL;
4443 struct extent_state *cached_state = NULL;
4444 struct btrfs_path *path;
4445 struct btrfs_root *root = BTRFS_I(inode)->root;
4446 struct fiemap_cache cache = { 0 };
4455 path = btrfs_alloc_path();
4458 path->leave_spinning = 1;
4460 start = round_down(start, btrfs_inode_sectorsize(inode));
4461 len = round_up(max, btrfs_inode_sectorsize(inode)) - start;
4464 * lookup the last file extent. We're not using i_size here
4465 * because there might be preallocation past i_size
4467 ret = btrfs_lookup_file_extent(NULL, root, path,
4468 btrfs_ino(BTRFS_I(inode)), -1, 0);
4470 btrfs_free_path(path);
4479 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
4480 found_type = found_key.type;
4482 /* No extents, but there might be delalloc bits */
4483 if (found_key.objectid != btrfs_ino(BTRFS_I(inode)) ||
4484 found_type != BTRFS_EXTENT_DATA_KEY) {
4485 /* have to trust i_size as the end */
4487 last_for_get_extent = isize;
4490 * remember the start of the last extent. There are a
4491 * bunch of different factors that go into the length of the
4492 * extent, so its much less complex to remember where it started
4494 last = found_key.offset;
4495 last_for_get_extent = last + 1;
4497 btrfs_release_path(path);
4500 * we might have some extents allocated but more delalloc past those
4501 * extents. so, we trust isize unless the start of the last extent is
4506 last_for_get_extent = isize;
4509 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4512 em = get_extent_skip_holes(inode, start, last_for_get_extent);
4521 u64 offset_in_extent = 0;
4523 /* break if the extent we found is outside the range */
4524 if (em->start >= max || extent_map_end(em) < off)
4528 * get_extent may return an extent that starts before our
4529 * requested range. We have to make sure the ranges
4530 * we return to fiemap always move forward and don't
4531 * overlap, so adjust the offsets here
4533 em_start = max(em->start, off);
4536 * record the offset from the start of the extent
4537 * for adjusting the disk offset below. Only do this if the
4538 * extent isn't compressed since our in ram offset may be past
4539 * what we have actually allocated on disk.
4541 if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4542 offset_in_extent = em_start - em->start;
4543 em_end = extent_map_end(em);
4544 em_len = em_end - em_start;
4545 disko = em->block_start + offset_in_extent;
4549 * bump off for our next call to get_extent
4551 off = extent_map_end(em);
4555 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
4557 flags |= FIEMAP_EXTENT_LAST;
4558 } else if (em->block_start == EXTENT_MAP_INLINE) {
4559 flags |= (FIEMAP_EXTENT_DATA_INLINE |
4560 FIEMAP_EXTENT_NOT_ALIGNED);
4561 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
4562 flags |= (FIEMAP_EXTENT_DELALLOC |
4563 FIEMAP_EXTENT_UNKNOWN);
4564 } else if (fieinfo->fi_extents_max) {
4565 u64 bytenr = em->block_start -
4566 (em->start - em->orig_start);
4569 * As btrfs supports shared space, this information
4570 * can be exported to userspace tools via
4571 * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0
4572 * then we're just getting a count and we can skip the
4575 ret = btrfs_check_shared(root,
4576 btrfs_ino(BTRFS_I(inode)),
4581 flags |= FIEMAP_EXTENT_SHARED;
4584 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4585 flags |= FIEMAP_EXTENT_ENCODED;
4586 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4587 flags |= FIEMAP_EXTENT_UNWRITTEN;
4589 free_extent_map(em);
4591 if ((em_start >= last) || em_len == (u64)-1 ||
4592 (last == (u64)-1 && isize <= em_end)) {
4593 flags |= FIEMAP_EXTENT_LAST;
4597 /* now scan forward to see if this is really the last extent. */
4598 em = get_extent_skip_holes(inode, off, last_for_get_extent);
4604 flags |= FIEMAP_EXTENT_LAST;
4607 ret = emit_fiemap_extent(fieinfo, &cache, em_start, disko,
4617 ret = emit_last_fiemap_cache(root->fs_info, fieinfo, &cache);
4618 free_extent_map(em);
4620 btrfs_free_path(path);
4621 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4626 static void __free_extent_buffer(struct extent_buffer *eb)
4628 btrfs_leak_debug_del(&eb->leak_list);
4629 kmem_cache_free(extent_buffer_cache, eb);
4632 int extent_buffer_under_io(struct extent_buffer *eb)
4634 return (atomic_read(&eb->io_pages) ||
4635 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4636 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4640 * Helper for releasing extent buffer page.
4642 static void btrfs_release_extent_buffer_page(struct extent_buffer *eb)
4644 unsigned long index;
4646 int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4648 BUG_ON(extent_buffer_under_io(eb));
4650 index = num_extent_pages(eb->start, eb->len);
4656 page = eb->pages[index];
4660 spin_lock(&page->mapping->private_lock);
4662 * We do this since we'll remove the pages after we've
4663 * removed the eb from the radix tree, so we could race
4664 * and have this page now attached to the new eb. So
4665 * only clear page_private if it's still connected to
4668 if (PagePrivate(page) &&
4669 page->private == (unsigned long)eb) {
4670 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4671 BUG_ON(PageDirty(page));
4672 BUG_ON(PageWriteback(page));
4674 * We need to make sure we haven't be attached
4677 ClearPagePrivate(page);
4678 set_page_private(page, 0);
4679 /* One for the page private */
4684 spin_unlock(&page->mapping->private_lock);
4686 /* One for when we allocated the page */
4688 } while (index != 0);
4692 * Helper for releasing the extent buffer.
4694 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4696 btrfs_release_extent_buffer_page(eb);
4697 __free_extent_buffer(eb);
4700 static struct extent_buffer *
4701 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
4704 struct extent_buffer *eb = NULL;
4706 eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL);
4709 eb->fs_info = fs_info;
4711 rwlock_init(&eb->lock);
4712 atomic_set(&eb->write_locks, 0);
4713 atomic_set(&eb->read_locks, 0);
4714 atomic_set(&eb->blocking_readers, 0);
4715 atomic_set(&eb->blocking_writers, 0);
4716 atomic_set(&eb->spinning_readers, 0);
4717 atomic_set(&eb->spinning_writers, 0);
4718 eb->lock_nested = 0;
4719 init_waitqueue_head(&eb->write_lock_wq);
4720 init_waitqueue_head(&eb->read_lock_wq);
4722 btrfs_leak_debug_add(&eb->leak_list, &buffers);
4724 spin_lock_init(&eb->refs_lock);
4725 atomic_set(&eb->refs, 1);
4726 atomic_set(&eb->io_pages, 0);
4729 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4731 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4732 > MAX_INLINE_EXTENT_BUFFER_SIZE);
4733 BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE);
4738 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src)
4742 struct extent_buffer *new;
4743 unsigned long num_pages = num_extent_pages(src->start, src->len);
4745 new = __alloc_extent_buffer(src->fs_info, src->start, src->len);
4749 for (i = 0; i < num_pages; i++) {
4750 p = alloc_page(GFP_NOFS);
4752 btrfs_release_extent_buffer(new);
4755 attach_extent_buffer_page(new, p);
4756 WARN_ON(PageDirty(p));
4759 copy_page(page_address(p), page_address(src->pages[i]));
4762 set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
4763 set_bit(EXTENT_BUFFER_DUMMY, &new->bflags);
4768 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4769 u64 start, unsigned long len)
4771 struct extent_buffer *eb;
4772 unsigned long num_pages;
4775 num_pages = num_extent_pages(start, len);
4777 eb = __alloc_extent_buffer(fs_info, start, len);
4781 for (i = 0; i < num_pages; i++) {
4782 eb->pages[i] = alloc_page(GFP_NOFS);
4786 set_extent_buffer_uptodate(eb);
4787 btrfs_set_header_nritems(eb, 0);
4788 set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4793 __free_page(eb->pages[i - 1]);
4794 __free_extent_buffer(eb);
4798 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4801 return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize);
4804 static void check_buffer_tree_ref(struct extent_buffer *eb)
4807 /* the ref bit is tricky. We have to make sure it is set
4808 * if we have the buffer dirty. Otherwise the
4809 * code to free a buffer can end up dropping a dirty
4812 * Once the ref bit is set, it won't go away while the
4813 * buffer is dirty or in writeback, and it also won't
4814 * go away while we have the reference count on the
4817 * We can't just set the ref bit without bumping the
4818 * ref on the eb because free_extent_buffer might
4819 * see the ref bit and try to clear it. If this happens
4820 * free_extent_buffer might end up dropping our original
4821 * ref by mistake and freeing the page before we are able
4822 * to add one more ref.
4824 * So bump the ref count first, then set the bit. If someone
4825 * beat us to it, drop the ref we added.
4827 refs = atomic_read(&eb->refs);
4828 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4831 spin_lock(&eb->refs_lock);
4832 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4833 atomic_inc(&eb->refs);
4834 spin_unlock(&eb->refs_lock);
4837 static void mark_extent_buffer_accessed(struct extent_buffer *eb,
4838 struct page *accessed)
4840 unsigned long num_pages, i;
4842 check_buffer_tree_ref(eb);
4844 num_pages = num_extent_pages(eb->start, eb->len);
4845 for (i = 0; i < num_pages; i++) {
4846 struct page *p = eb->pages[i];
4849 mark_page_accessed(p);
4853 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
4856 struct extent_buffer *eb;
4859 eb = radix_tree_lookup(&fs_info->buffer_radix,
4860 start >> PAGE_SHIFT);
4861 if (eb && atomic_inc_not_zero(&eb->refs)) {
4864 * Lock our eb's refs_lock to avoid races with
4865 * free_extent_buffer. When we get our eb it might be flagged
4866 * with EXTENT_BUFFER_STALE and another task running
4867 * free_extent_buffer might have seen that flag set,
4868 * eb->refs == 2, that the buffer isn't under IO (dirty and
4869 * writeback flags not set) and it's still in the tree (flag
4870 * EXTENT_BUFFER_TREE_REF set), therefore being in the process
4871 * of decrementing the extent buffer's reference count twice.
4872 * So here we could race and increment the eb's reference count,
4873 * clear its stale flag, mark it as dirty and drop our reference
4874 * before the other task finishes executing free_extent_buffer,
4875 * which would later result in an attempt to free an extent
4876 * buffer that is dirty.
4878 if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) {
4879 spin_lock(&eb->refs_lock);
4880 spin_unlock(&eb->refs_lock);
4882 mark_extent_buffer_accessed(eb, NULL);
4890 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4891 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
4894 struct extent_buffer *eb, *exists = NULL;
4897 eb = find_extent_buffer(fs_info, start);
4900 eb = alloc_dummy_extent_buffer(fs_info, start);
4903 eb->fs_info = fs_info;
4905 ret = radix_tree_preload(GFP_NOFS);
4908 spin_lock(&fs_info->buffer_lock);
4909 ret = radix_tree_insert(&fs_info->buffer_radix,
4910 start >> PAGE_SHIFT, eb);
4911 spin_unlock(&fs_info->buffer_lock);
4912 radix_tree_preload_end();
4913 if (ret == -EEXIST) {
4914 exists = find_extent_buffer(fs_info, start);
4920 check_buffer_tree_ref(eb);
4921 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
4924 * We will free dummy extent buffer's if they come into
4925 * free_extent_buffer with a ref count of 2, but if we are using this we
4926 * want the buffers to stay in memory until we're done with them, so
4927 * bump the ref count again.
4929 atomic_inc(&eb->refs);
4932 btrfs_release_extent_buffer(eb);
4937 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
4940 unsigned long len = fs_info->nodesize;
4941 unsigned long num_pages = num_extent_pages(start, len);
4943 unsigned long index = start >> PAGE_SHIFT;
4944 struct extent_buffer *eb;
4945 struct extent_buffer *exists = NULL;
4947 struct address_space *mapping = fs_info->btree_inode->i_mapping;
4951 if (!IS_ALIGNED(start, fs_info->sectorsize)) {
4952 btrfs_err(fs_info, "bad tree block start %llu", start);
4953 return ERR_PTR(-EINVAL);
4956 eb = find_extent_buffer(fs_info, start);
4960 eb = __alloc_extent_buffer(fs_info, start, len);
4962 return ERR_PTR(-ENOMEM);
4964 for (i = 0; i < num_pages; i++, index++) {
4965 p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL);
4967 exists = ERR_PTR(-ENOMEM);
4971 spin_lock(&mapping->private_lock);
4972 if (PagePrivate(p)) {
4974 * We could have already allocated an eb for this page
4975 * and attached one so lets see if we can get a ref on
4976 * the existing eb, and if we can we know it's good and
4977 * we can just return that one, else we know we can just
4978 * overwrite page->private.
4980 exists = (struct extent_buffer *)p->private;
4981 if (atomic_inc_not_zero(&exists->refs)) {
4982 spin_unlock(&mapping->private_lock);
4985 mark_extent_buffer_accessed(exists, p);
4991 * Do this so attach doesn't complain and we need to
4992 * drop the ref the old guy had.
4994 ClearPagePrivate(p);
4995 WARN_ON(PageDirty(p));
4998 attach_extent_buffer_page(eb, p);
4999 spin_unlock(&mapping->private_lock);
5000 WARN_ON(PageDirty(p));
5002 if (!PageUptodate(p))
5006 * see below about how we avoid a nasty race with release page
5007 * and why we unlock later
5011 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5013 ret = radix_tree_preload(GFP_NOFS);
5015 exists = ERR_PTR(ret);
5019 spin_lock(&fs_info->buffer_lock);
5020 ret = radix_tree_insert(&fs_info->buffer_radix,
5021 start >> PAGE_SHIFT, eb);
5022 spin_unlock(&fs_info->buffer_lock);
5023 radix_tree_preload_end();
5024 if (ret == -EEXIST) {
5025 exists = find_extent_buffer(fs_info, start);
5031 /* add one reference for the tree */
5032 check_buffer_tree_ref(eb);
5033 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
5036 * there is a race where release page may have
5037 * tried to find this extent buffer in the radix
5038 * but failed. It will tell the VM it is safe to
5039 * reclaim the, and it will clear the page private bit.
5040 * We must make sure to set the page private bit properly
5041 * after the extent buffer is in the radix tree so
5042 * it doesn't get lost
5044 SetPageChecked(eb->pages[0]);
5045 for (i = 1; i < num_pages; i++) {
5047 ClearPageChecked(p);
5050 unlock_page(eb->pages[0]);
5054 WARN_ON(!atomic_dec_and_test(&eb->refs));
5055 for (i = 0; i < num_pages; i++) {
5057 unlock_page(eb->pages[i]);
5060 btrfs_release_extent_buffer(eb);
5064 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
5066 struct extent_buffer *eb =
5067 container_of(head, struct extent_buffer, rcu_head);
5069 __free_extent_buffer(eb);
5072 /* Expects to have eb->eb_lock already held */
5073 static int release_extent_buffer(struct extent_buffer *eb)
5075 WARN_ON(atomic_read(&eb->refs) == 0);
5076 if (atomic_dec_and_test(&eb->refs)) {
5077 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
5078 struct btrfs_fs_info *fs_info = eb->fs_info;
5080 spin_unlock(&eb->refs_lock);
5082 spin_lock(&fs_info->buffer_lock);
5083 radix_tree_delete(&fs_info->buffer_radix,
5084 eb->start >> PAGE_SHIFT);
5085 spin_unlock(&fs_info->buffer_lock);
5087 spin_unlock(&eb->refs_lock);
5090 /* Should be safe to release our pages at this point */
5091 btrfs_release_extent_buffer_page(eb);
5092 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5093 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))) {
5094 __free_extent_buffer(eb);
5098 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
5101 spin_unlock(&eb->refs_lock);
5106 void free_extent_buffer(struct extent_buffer *eb)
5114 refs = atomic_read(&eb->refs);
5117 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
5122 spin_lock(&eb->refs_lock);
5123 if (atomic_read(&eb->refs) == 2 &&
5124 test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))
5125 atomic_dec(&eb->refs);
5127 if (atomic_read(&eb->refs) == 2 &&
5128 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
5129 !extent_buffer_under_io(eb) &&
5130 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5131 atomic_dec(&eb->refs);
5134 * I know this is terrible, but it's temporary until we stop tracking
5135 * the uptodate bits and such for the extent buffers.
5137 release_extent_buffer(eb);
5140 void free_extent_buffer_stale(struct extent_buffer *eb)
5145 spin_lock(&eb->refs_lock);
5146 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
5148 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
5149 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5150 atomic_dec(&eb->refs);
5151 release_extent_buffer(eb);
5154 void clear_extent_buffer_dirty(struct extent_buffer *eb)
5157 unsigned long num_pages;
5160 num_pages = num_extent_pages(eb->start, eb->len);
5162 for (i = 0; i < num_pages; i++) {
5163 page = eb->pages[i];
5164 if (!PageDirty(page))
5168 WARN_ON(!PagePrivate(page));
5170 clear_page_dirty_for_io(page);
5171 xa_lock_irq(&page->mapping->i_pages);
5172 if (!PageDirty(page)) {
5173 radix_tree_tag_clear(&page->mapping->i_pages,
5175 PAGECACHE_TAG_DIRTY);
5177 xa_unlock_irq(&page->mapping->i_pages);
5178 ClearPageError(page);
5181 WARN_ON(atomic_read(&eb->refs) == 0);
5184 int set_extent_buffer_dirty(struct extent_buffer *eb)
5187 unsigned long num_pages;
5190 check_buffer_tree_ref(eb);
5192 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
5194 num_pages = num_extent_pages(eb->start, eb->len);
5195 WARN_ON(atomic_read(&eb->refs) == 0);
5196 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
5198 for (i = 0; i < num_pages; i++)
5199 set_page_dirty(eb->pages[i]);
5203 void clear_extent_buffer_uptodate(struct extent_buffer *eb)
5207 unsigned long num_pages;
5209 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5210 num_pages = num_extent_pages(eb->start, eb->len);
5211 for (i = 0; i < num_pages; i++) {
5212 page = eb->pages[i];
5214 ClearPageUptodate(page);
5218 void set_extent_buffer_uptodate(struct extent_buffer *eb)
5222 unsigned long num_pages;
5224 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5225 num_pages = num_extent_pages(eb->start, eb->len);
5226 for (i = 0; i < num_pages; i++) {
5227 page = eb->pages[i];
5228 SetPageUptodate(page);
5232 int read_extent_buffer_pages(struct extent_io_tree *tree,
5233 struct extent_buffer *eb, int wait, int mirror_num)
5239 int locked_pages = 0;
5240 int all_uptodate = 1;
5241 unsigned long num_pages;
5242 unsigned long num_reads = 0;
5243 struct bio *bio = NULL;
5244 unsigned long bio_flags = 0;
5246 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
5249 num_pages = num_extent_pages(eb->start, eb->len);
5250 for (i = 0; i < num_pages; i++) {
5251 page = eb->pages[i];
5252 if (wait == WAIT_NONE) {
5253 if (!trylock_page(page))
5261 * We need to firstly lock all pages to make sure that
5262 * the uptodate bit of our pages won't be affected by
5263 * clear_extent_buffer_uptodate().
5265 for (i = 0; i < num_pages; i++) {
5266 page = eb->pages[i];
5267 if (!PageUptodate(page)) {
5274 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5278 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
5279 eb->read_mirror = 0;
5280 atomic_set(&eb->io_pages, num_reads);
5281 for (i = 0; i < num_pages; i++) {
5282 page = eb->pages[i];
5284 if (!PageUptodate(page)) {
5286 atomic_dec(&eb->io_pages);
5291 ClearPageError(page);
5292 err = __extent_read_full_page(tree, page,
5293 btree_get_extent, &bio,
5294 mirror_num, &bio_flags,
5299 * We use &bio in above __extent_read_full_page,
5300 * so we ensure that if it returns error, the
5301 * current page fails to add itself to bio and
5302 * it's been unlocked.
5304 * We must dec io_pages by ourselves.
5306 atomic_dec(&eb->io_pages);
5314 err = submit_one_bio(bio, mirror_num, bio_flags);
5319 if (ret || wait != WAIT_COMPLETE)
5322 for (i = 0; i < num_pages; i++) {
5323 page = eb->pages[i];
5324 wait_on_page_locked(page);
5325 if (!PageUptodate(page))
5332 while (locked_pages > 0) {
5334 page = eb->pages[locked_pages];
5340 void read_extent_buffer(const struct extent_buffer *eb, void *dstv,
5341 unsigned long start, unsigned long len)
5347 char *dst = (char *)dstv;
5348 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5349 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5351 if (start + len > eb->len) {
5352 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n",
5353 eb->start, eb->len, start, len);
5354 memset(dst, 0, len);
5358 offset = (start_offset + start) & (PAGE_SIZE - 1);
5361 page = eb->pages[i];
5363 cur = min(len, (PAGE_SIZE - offset));
5364 kaddr = page_address(page);
5365 memcpy(dst, kaddr + offset, cur);
5374 int read_extent_buffer_to_user(const struct extent_buffer *eb,
5376 unsigned long start, unsigned long len)
5382 char __user *dst = (char __user *)dstv;
5383 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5384 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5387 WARN_ON(start > eb->len);
5388 WARN_ON(start + len > eb->start + eb->len);
5390 offset = (start_offset + start) & (PAGE_SIZE - 1);
5393 page = eb->pages[i];
5395 cur = min(len, (PAGE_SIZE - offset));
5396 kaddr = page_address(page);
5397 if (copy_to_user(dst, kaddr + offset, cur)) {
5412 * return 0 if the item is found within a page.
5413 * return 1 if the item spans two pages.
5414 * return -EINVAL otherwise.
5416 int map_private_extent_buffer(const struct extent_buffer *eb,
5417 unsigned long start, unsigned long min_len,
5418 char **map, unsigned long *map_start,
5419 unsigned long *map_len)
5421 size_t offset = start & (PAGE_SIZE - 1);
5424 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5425 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5426 unsigned long end_i = (start_offset + start + min_len - 1) >>
5429 if (start + min_len > eb->len) {
5430 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n",
5431 eb->start, eb->len, start, min_len);
5439 offset = start_offset;
5443 *map_start = ((u64)i << PAGE_SHIFT) - start_offset;
5447 kaddr = page_address(p);
5448 *map = kaddr + offset;
5449 *map_len = PAGE_SIZE - offset;
5453 int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,
5454 unsigned long start, unsigned long len)
5460 char *ptr = (char *)ptrv;
5461 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5462 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5465 WARN_ON(start > eb->len);
5466 WARN_ON(start + len > eb->start + eb->len);
5468 offset = (start_offset + start) & (PAGE_SIZE - 1);
5471 page = eb->pages[i];
5473 cur = min(len, (PAGE_SIZE - offset));
5475 kaddr = page_address(page);
5476 ret = memcmp(ptr, kaddr + offset, cur);
5488 void write_extent_buffer_chunk_tree_uuid(struct extent_buffer *eb,
5493 WARN_ON(!PageUptodate(eb->pages[0]));
5494 kaddr = page_address(eb->pages[0]);
5495 memcpy(kaddr + offsetof(struct btrfs_header, chunk_tree_uuid), srcv,
5499 void write_extent_buffer_fsid(struct extent_buffer *eb, const void *srcv)
5503 WARN_ON(!PageUptodate(eb->pages[0]));
5504 kaddr = page_address(eb->pages[0]);
5505 memcpy(kaddr + offsetof(struct btrfs_header, fsid), srcv,
5509 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
5510 unsigned long start, unsigned long len)
5516 char *src = (char *)srcv;
5517 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5518 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5520 WARN_ON(start > eb->len);
5521 WARN_ON(start + len > eb->start + eb->len);
5523 offset = (start_offset + start) & (PAGE_SIZE - 1);
5526 page = eb->pages[i];
5527 WARN_ON(!PageUptodate(page));
5529 cur = min(len, PAGE_SIZE - offset);
5530 kaddr = page_address(page);
5531 memcpy(kaddr + offset, src, cur);
5540 void memzero_extent_buffer(struct extent_buffer *eb, unsigned long start,
5547 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5548 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5550 WARN_ON(start > eb->len);
5551 WARN_ON(start + len > eb->start + eb->len);
5553 offset = (start_offset + start) & (PAGE_SIZE - 1);
5556 page = eb->pages[i];
5557 WARN_ON(!PageUptodate(page));
5559 cur = min(len, PAGE_SIZE - offset);
5560 kaddr = page_address(page);
5561 memset(kaddr + offset, 0, cur);
5569 void copy_extent_buffer_full(struct extent_buffer *dst,
5570 struct extent_buffer *src)
5575 ASSERT(dst->len == src->len);
5577 num_pages = num_extent_pages(dst->start, dst->len);
5578 for (i = 0; i < num_pages; i++)
5579 copy_page(page_address(dst->pages[i]),
5580 page_address(src->pages[i]));
5583 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
5584 unsigned long dst_offset, unsigned long src_offset,
5587 u64 dst_len = dst->len;
5592 size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1);
5593 unsigned long i = (start_offset + dst_offset) >> PAGE_SHIFT;
5595 WARN_ON(src->len != dst_len);
5597 offset = (start_offset + dst_offset) &
5601 page = dst->pages[i];
5602 WARN_ON(!PageUptodate(page));
5604 cur = min(len, (unsigned long)(PAGE_SIZE - offset));
5606 kaddr = page_address(page);
5607 read_extent_buffer(src, kaddr + offset, src_offset, cur);
5617 * eb_bitmap_offset() - calculate the page and offset of the byte containing the
5619 * @eb: the extent buffer
5620 * @start: offset of the bitmap item in the extent buffer
5622 * @page_index: return index of the page in the extent buffer that contains the
5624 * @page_offset: return offset into the page given by page_index
5626 * This helper hides the ugliness of finding the byte in an extent buffer which
5627 * contains a given bit.
5629 static inline void eb_bitmap_offset(struct extent_buffer *eb,
5630 unsigned long start, unsigned long nr,
5631 unsigned long *page_index,
5632 size_t *page_offset)
5634 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5635 size_t byte_offset = BIT_BYTE(nr);
5639 * The byte we want is the offset of the extent buffer + the offset of
5640 * the bitmap item in the extent buffer + the offset of the byte in the
5643 offset = start_offset + start + byte_offset;
5645 *page_index = offset >> PAGE_SHIFT;
5646 *page_offset = offset & (PAGE_SIZE - 1);
5650 * extent_buffer_test_bit - determine whether a bit in a bitmap item is set
5651 * @eb: the extent buffer
5652 * @start: offset of the bitmap item in the extent buffer
5653 * @nr: bit number to test
5655 int extent_buffer_test_bit(struct extent_buffer *eb, unsigned long start,
5663 eb_bitmap_offset(eb, start, nr, &i, &offset);
5664 page = eb->pages[i];
5665 WARN_ON(!PageUptodate(page));
5666 kaddr = page_address(page);
5667 return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1)));
5671 * extent_buffer_bitmap_set - set an area of a bitmap
5672 * @eb: the extent buffer
5673 * @start: offset of the bitmap item in the extent buffer
5674 * @pos: bit number of the first bit
5675 * @len: number of bits to set
5677 void extent_buffer_bitmap_set(struct extent_buffer *eb, unsigned long start,
5678 unsigned long pos, unsigned long len)
5684 const unsigned int size = pos + len;
5685 int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5686 u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos);
5688 eb_bitmap_offset(eb, start, pos, &i, &offset);
5689 page = eb->pages[i];
5690 WARN_ON(!PageUptodate(page));
5691 kaddr = page_address(page);
5693 while (len >= bits_to_set) {
5694 kaddr[offset] |= mask_to_set;
5696 bits_to_set = BITS_PER_BYTE;
5698 if (++offset >= PAGE_SIZE && len > 0) {
5700 page = eb->pages[++i];
5701 WARN_ON(!PageUptodate(page));
5702 kaddr = page_address(page);
5706 mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
5707 kaddr[offset] |= mask_to_set;
5713 * extent_buffer_bitmap_clear - clear an area of a bitmap
5714 * @eb: the extent buffer
5715 * @start: offset of the bitmap item in the extent buffer
5716 * @pos: bit number of the first bit
5717 * @len: number of bits to clear
5719 void extent_buffer_bitmap_clear(struct extent_buffer *eb, unsigned long start,
5720 unsigned long pos, unsigned long len)
5726 const unsigned int size = pos + len;
5727 int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5728 u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos);
5730 eb_bitmap_offset(eb, start, pos, &i, &offset);
5731 page = eb->pages[i];
5732 WARN_ON(!PageUptodate(page));
5733 kaddr = page_address(page);
5735 while (len >= bits_to_clear) {
5736 kaddr[offset] &= ~mask_to_clear;
5737 len -= bits_to_clear;
5738 bits_to_clear = BITS_PER_BYTE;
5740 if (++offset >= PAGE_SIZE && len > 0) {
5742 page = eb->pages[++i];
5743 WARN_ON(!PageUptodate(page));
5744 kaddr = page_address(page);
5748 mask_to_clear &= BITMAP_LAST_BYTE_MASK(size);
5749 kaddr[offset] &= ~mask_to_clear;
5753 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
5755 unsigned long distance = (src > dst) ? src - dst : dst - src;
5756 return distance < len;
5759 static void copy_pages(struct page *dst_page, struct page *src_page,
5760 unsigned long dst_off, unsigned long src_off,
5763 char *dst_kaddr = page_address(dst_page);
5765 int must_memmove = 0;
5767 if (dst_page != src_page) {
5768 src_kaddr = page_address(src_page);
5770 src_kaddr = dst_kaddr;
5771 if (areas_overlap(src_off, dst_off, len))
5776 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
5778 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
5781 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5782 unsigned long src_offset, unsigned long len)
5784 struct btrfs_fs_info *fs_info = dst->fs_info;
5786 size_t dst_off_in_page;
5787 size_t src_off_in_page;
5788 size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1);
5789 unsigned long dst_i;
5790 unsigned long src_i;
5792 if (src_offset + len > dst->len) {
5794 "memmove bogus src_offset %lu move len %lu dst len %lu",
5795 src_offset, len, dst->len);
5798 if (dst_offset + len > dst->len) {
5800 "memmove bogus dst_offset %lu move len %lu dst len %lu",
5801 dst_offset, len, dst->len);
5806 dst_off_in_page = (start_offset + dst_offset) &
5808 src_off_in_page = (start_offset + src_offset) &
5811 dst_i = (start_offset + dst_offset) >> PAGE_SHIFT;
5812 src_i = (start_offset + src_offset) >> PAGE_SHIFT;
5814 cur = min(len, (unsigned long)(PAGE_SIZE -
5816 cur = min_t(unsigned long, cur,
5817 (unsigned long)(PAGE_SIZE - dst_off_in_page));
5819 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5820 dst_off_in_page, src_off_in_page, cur);
5828 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5829 unsigned long src_offset, unsigned long len)
5831 struct btrfs_fs_info *fs_info = dst->fs_info;
5833 size_t dst_off_in_page;
5834 size_t src_off_in_page;
5835 unsigned long dst_end = dst_offset + len - 1;
5836 unsigned long src_end = src_offset + len - 1;
5837 size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1);
5838 unsigned long dst_i;
5839 unsigned long src_i;
5841 if (src_offset + len > dst->len) {
5843 "memmove bogus src_offset %lu move len %lu len %lu",
5844 src_offset, len, dst->len);
5847 if (dst_offset + len > dst->len) {
5849 "memmove bogus dst_offset %lu move len %lu len %lu",
5850 dst_offset, len, dst->len);
5853 if (dst_offset < src_offset) {
5854 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
5858 dst_i = (start_offset + dst_end) >> PAGE_SHIFT;
5859 src_i = (start_offset + src_end) >> PAGE_SHIFT;
5861 dst_off_in_page = (start_offset + dst_end) &
5863 src_off_in_page = (start_offset + src_end) &
5866 cur = min_t(unsigned long, len, src_off_in_page + 1);
5867 cur = min(cur, dst_off_in_page + 1);
5868 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5869 dst_off_in_page - cur + 1,
5870 src_off_in_page - cur + 1, cur);
5878 int try_release_extent_buffer(struct page *page)
5880 struct extent_buffer *eb;
5883 * We need to make sure nobody is attaching this page to an eb right
5886 spin_lock(&page->mapping->private_lock);
5887 if (!PagePrivate(page)) {
5888 spin_unlock(&page->mapping->private_lock);
5892 eb = (struct extent_buffer *)page->private;
5896 * This is a little awful but should be ok, we need to make sure that
5897 * the eb doesn't disappear out from under us while we're looking at
5900 spin_lock(&eb->refs_lock);
5901 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
5902 spin_unlock(&eb->refs_lock);
5903 spin_unlock(&page->mapping->private_lock);
5906 spin_unlock(&page->mapping->private_lock);
5909 * If tree ref isn't set then we know the ref on this eb is a real ref,
5910 * so just return, this page will likely be freed soon anyway.
5912 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
5913 spin_unlock(&eb->refs_lock);
5917 return release_extent_buffer(eb);