1 #include <linux/bitops.h>
2 #include <linux/slab.h>
5 #include <linux/pagemap.h>
6 #include <linux/page-flags.h>
7 #include <linux/spinlock.h>
8 #include <linux/blkdev.h>
9 #include <linux/swap.h>
10 #include <linux/writeback.h>
11 #include <linux/pagevec.h>
12 #include <linux/prefetch.h>
13 #include <linux/cleancache.h>
14 #include "extent_io.h"
15 #include "extent_map.h"
17 #include "btrfs_inode.h"
19 #include "check-integrity.h"
21 #include "rcu-string.h"
23 #include "transaction.h"
25 static struct kmem_cache *extent_state_cache;
26 static struct kmem_cache *extent_buffer_cache;
27 static struct bio_set *btrfs_bioset;
29 static inline bool extent_state_in_tree(const struct extent_state *state)
31 return !RB_EMPTY_NODE(&state->rb_node);
34 #ifdef CONFIG_BTRFS_DEBUG
35 static LIST_HEAD(buffers);
36 static LIST_HEAD(states);
38 static DEFINE_SPINLOCK(leak_lock);
41 void btrfs_leak_debug_add(struct list_head *new, struct list_head *head)
45 spin_lock_irqsave(&leak_lock, flags);
47 spin_unlock_irqrestore(&leak_lock, flags);
51 void btrfs_leak_debug_del(struct list_head *entry)
55 spin_lock_irqsave(&leak_lock, flags);
57 spin_unlock_irqrestore(&leak_lock, flags);
61 void btrfs_leak_debug_check(void)
63 struct extent_state *state;
64 struct extent_buffer *eb;
66 while (!list_empty(&states)) {
67 state = list_entry(states.next, struct extent_state, leak_list);
68 pr_err("BTRFS: state leak: start %llu end %llu state %u in tree %d refs %d\n",
69 state->start, state->end, state->state,
70 extent_state_in_tree(state),
71 refcount_read(&state->refs));
72 list_del(&state->leak_list);
73 kmem_cache_free(extent_state_cache, state);
76 while (!list_empty(&buffers)) {
77 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
78 pr_err("BTRFS: buffer leak start %llu len %lu refs %d\n",
79 eb->start, eb->len, atomic_read(&eb->refs));
80 list_del(&eb->leak_list);
81 kmem_cache_free(extent_buffer_cache, eb);
85 #define btrfs_debug_check_extent_io_range(tree, start, end) \
86 __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end))
87 static inline void __btrfs_debug_check_extent_io_range(const char *caller,
88 struct extent_io_tree *tree, u64 start, u64 end)
90 if (tree->ops && tree->ops->check_extent_io_range)
91 tree->ops->check_extent_io_range(tree->private_data, caller,
95 #define btrfs_leak_debug_add(new, head) do {} while (0)
96 #define btrfs_leak_debug_del(entry) do {} while (0)
97 #define btrfs_leak_debug_check() do {} while (0)
98 #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0)
101 #define BUFFER_LRU_MAX 64
106 struct rb_node rb_node;
109 struct extent_page_data {
111 struct extent_io_tree *tree;
112 get_extent_t *get_extent;
113 unsigned long bio_flags;
115 /* tells writepage not to lock the state bits for this range
116 * it still does the unlocking
118 unsigned int extent_locked:1;
120 /* tells the submit_bio code to use REQ_SYNC */
121 unsigned int sync_io:1;
124 static void add_extent_changeset(struct extent_state *state, unsigned bits,
125 struct extent_changeset *changeset,
132 if (set && (state->state & bits) == bits)
134 if (!set && (state->state & bits) == 0)
136 changeset->bytes_changed += state->end - state->start + 1;
137 ret = ulist_add(&changeset->range_changed, state->start, state->end,
143 static noinline void flush_write_bio(void *data);
144 static inline struct btrfs_fs_info *
145 tree_fs_info(struct extent_io_tree *tree)
148 return tree->ops->tree_fs_info(tree->private_data);
152 int __init extent_io_init(void)
154 extent_state_cache = kmem_cache_create("btrfs_extent_state",
155 sizeof(struct extent_state), 0,
156 SLAB_MEM_SPREAD, NULL);
157 if (!extent_state_cache)
160 extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer",
161 sizeof(struct extent_buffer), 0,
162 SLAB_MEM_SPREAD, NULL);
163 if (!extent_buffer_cache)
164 goto free_state_cache;
166 btrfs_bioset = bioset_create(BIO_POOL_SIZE,
167 offsetof(struct btrfs_io_bio, bio));
169 goto free_buffer_cache;
171 if (bioset_integrity_create(btrfs_bioset, BIO_POOL_SIZE))
177 bioset_free(btrfs_bioset);
181 kmem_cache_destroy(extent_buffer_cache);
182 extent_buffer_cache = NULL;
185 kmem_cache_destroy(extent_state_cache);
186 extent_state_cache = NULL;
190 void extent_io_exit(void)
192 btrfs_leak_debug_check();
195 * Make sure all delayed rcu free are flushed before we
199 kmem_cache_destroy(extent_state_cache);
200 kmem_cache_destroy(extent_buffer_cache);
202 bioset_free(btrfs_bioset);
205 void extent_io_tree_init(struct extent_io_tree *tree,
208 tree->state = RB_ROOT;
210 tree->dirty_bytes = 0;
211 spin_lock_init(&tree->lock);
212 tree->private_data = private_data;
215 static struct extent_state *alloc_extent_state(gfp_t mask)
217 struct extent_state *state;
220 * The given mask might be not appropriate for the slab allocator,
221 * drop the unsupported bits
223 mask &= ~(__GFP_DMA32|__GFP_HIGHMEM);
224 state = kmem_cache_alloc(extent_state_cache, mask);
228 state->failrec = NULL;
229 RB_CLEAR_NODE(&state->rb_node);
230 btrfs_leak_debug_add(&state->leak_list, &states);
231 refcount_set(&state->refs, 1);
232 init_waitqueue_head(&state->wq);
233 trace_alloc_extent_state(state, mask, _RET_IP_);
237 void free_extent_state(struct extent_state *state)
241 if (refcount_dec_and_test(&state->refs)) {
242 WARN_ON(extent_state_in_tree(state));
243 btrfs_leak_debug_del(&state->leak_list);
244 trace_free_extent_state(state, _RET_IP_);
245 kmem_cache_free(extent_state_cache, state);
249 static struct rb_node *tree_insert(struct rb_root *root,
250 struct rb_node *search_start,
252 struct rb_node *node,
253 struct rb_node ***p_in,
254 struct rb_node **parent_in)
257 struct rb_node *parent = NULL;
258 struct tree_entry *entry;
260 if (p_in && parent_in) {
266 p = search_start ? &search_start : &root->rb_node;
269 entry = rb_entry(parent, struct tree_entry, rb_node);
271 if (offset < entry->start)
273 else if (offset > entry->end)
280 rb_link_node(node, parent, p);
281 rb_insert_color(node, root);
285 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
286 struct rb_node **prev_ret,
287 struct rb_node **next_ret,
288 struct rb_node ***p_ret,
289 struct rb_node **parent_ret)
291 struct rb_root *root = &tree->state;
292 struct rb_node **n = &root->rb_node;
293 struct rb_node *prev = NULL;
294 struct rb_node *orig_prev = NULL;
295 struct tree_entry *entry;
296 struct tree_entry *prev_entry = NULL;
300 entry = rb_entry(prev, struct tree_entry, rb_node);
303 if (offset < entry->start)
305 else if (offset > entry->end)
318 while (prev && offset > prev_entry->end) {
319 prev = rb_next(prev);
320 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
327 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
328 while (prev && offset < prev_entry->start) {
329 prev = rb_prev(prev);
330 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
337 static inline struct rb_node *
338 tree_search_for_insert(struct extent_io_tree *tree,
340 struct rb_node ***p_ret,
341 struct rb_node **parent_ret)
343 struct rb_node *prev = NULL;
346 ret = __etree_search(tree, offset, &prev, NULL, p_ret, parent_ret);
352 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
355 return tree_search_for_insert(tree, offset, NULL, NULL);
358 static void merge_cb(struct extent_io_tree *tree, struct extent_state *new,
359 struct extent_state *other)
361 if (tree->ops && tree->ops->merge_extent_hook)
362 tree->ops->merge_extent_hook(tree->private_data, new, other);
366 * utility function to look for merge candidates inside a given range.
367 * Any extents with matching state are merged together into a single
368 * extent in the tree. Extents with EXTENT_IO in their state field
369 * are not merged because the end_io handlers need to be able to do
370 * operations on them without sleeping (or doing allocations/splits).
372 * This should be called with the tree lock held.
374 static void merge_state(struct extent_io_tree *tree,
375 struct extent_state *state)
377 struct extent_state *other;
378 struct rb_node *other_node;
380 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
383 other_node = rb_prev(&state->rb_node);
385 other = rb_entry(other_node, struct extent_state, rb_node);
386 if (other->end == state->start - 1 &&
387 other->state == state->state) {
388 merge_cb(tree, state, other);
389 state->start = other->start;
390 rb_erase(&other->rb_node, &tree->state);
391 RB_CLEAR_NODE(&other->rb_node);
392 free_extent_state(other);
395 other_node = rb_next(&state->rb_node);
397 other = rb_entry(other_node, struct extent_state, rb_node);
398 if (other->start == state->end + 1 &&
399 other->state == state->state) {
400 merge_cb(tree, state, other);
401 state->end = other->end;
402 rb_erase(&other->rb_node, &tree->state);
403 RB_CLEAR_NODE(&other->rb_node);
404 free_extent_state(other);
409 static void set_state_cb(struct extent_io_tree *tree,
410 struct extent_state *state, unsigned *bits)
412 if (tree->ops && tree->ops->set_bit_hook)
413 tree->ops->set_bit_hook(tree->private_data, state, bits);
416 static void clear_state_cb(struct extent_io_tree *tree,
417 struct extent_state *state, unsigned *bits)
419 if (tree->ops && tree->ops->clear_bit_hook)
420 tree->ops->clear_bit_hook(tree->private_data, state, bits);
423 static void set_state_bits(struct extent_io_tree *tree,
424 struct extent_state *state, unsigned *bits,
425 struct extent_changeset *changeset);
428 * insert an extent_state struct into the tree. 'bits' are set on the
429 * struct before it is inserted.
431 * This may return -EEXIST if the extent is already there, in which case the
432 * state struct is freed.
434 * The tree lock is not taken internally. This is a utility function and
435 * probably isn't what you want to call (see set/clear_extent_bit).
437 static int insert_state(struct extent_io_tree *tree,
438 struct extent_state *state, u64 start, u64 end,
440 struct rb_node **parent,
441 unsigned *bits, struct extent_changeset *changeset)
443 struct rb_node *node;
446 WARN(1, KERN_ERR "BTRFS: end < start %llu %llu\n",
448 state->start = start;
451 set_state_bits(tree, state, bits, changeset);
453 node = tree_insert(&tree->state, NULL, end, &state->rb_node, p, parent);
455 struct extent_state *found;
456 found = rb_entry(node, struct extent_state, rb_node);
457 pr_err("BTRFS: found node %llu %llu on insert of %llu %llu\n",
458 found->start, found->end, start, end);
461 merge_state(tree, state);
465 static void split_cb(struct extent_io_tree *tree, struct extent_state *orig,
468 if (tree->ops && tree->ops->split_extent_hook)
469 tree->ops->split_extent_hook(tree->private_data, orig, split);
473 * split a given extent state struct in two, inserting the preallocated
474 * struct 'prealloc' as the newly created second half. 'split' indicates an
475 * offset inside 'orig' where it should be split.
478 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
479 * are two extent state structs in the tree:
480 * prealloc: [orig->start, split - 1]
481 * orig: [ split, orig->end ]
483 * The tree locks are not taken by this function. They need to be held
486 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
487 struct extent_state *prealloc, u64 split)
489 struct rb_node *node;
491 split_cb(tree, orig, split);
493 prealloc->start = orig->start;
494 prealloc->end = split - 1;
495 prealloc->state = orig->state;
498 node = tree_insert(&tree->state, &orig->rb_node, prealloc->end,
499 &prealloc->rb_node, NULL, NULL);
501 free_extent_state(prealloc);
507 static struct extent_state *next_state(struct extent_state *state)
509 struct rb_node *next = rb_next(&state->rb_node);
511 return rb_entry(next, struct extent_state, rb_node);
517 * utility function to clear some bits in an extent state struct.
518 * it will optionally wake up any one waiting on this state (wake == 1).
520 * If no bits are set on the state struct after clearing things, the
521 * struct is freed and removed from the tree
523 static struct extent_state *clear_state_bit(struct extent_io_tree *tree,
524 struct extent_state *state,
525 unsigned *bits, int wake,
526 struct extent_changeset *changeset)
528 struct extent_state *next;
529 unsigned bits_to_clear = *bits & ~EXTENT_CTLBITS;
531 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
532 u64 range = state->end - state->start + 1;
533 WARN_ON(range > tree->dirty_bytes);
534 tree->dirty_bytes -= range;
536 clear_state_cb(tree, state, bits);
537 add_extent_changeset(state, bits_to_clear, changeset, 0);
538 state->state &= ~bits_to_clear;
541 if (state->state == 0) {
542 next = next_state(state);
543 if (extent_state_in_tree(state)) {
544 rb_erase(&state->rb_node, &tree->state);
545 RB_CLEAR_NODE(&state->rb_node);
546 free_extent_state(state);
551 merge_state(tree, state);
552 next = next_state(state);
557 static struct extent_state *
558 alloc_extent_state_atomic(struct extent_state *prealloc)
561 prealloc = alloc_extent_state(GFP_ATOMIC);
566 static void extent_io_tree_panic(struct extent_io_tree *tree, int err)
568 btrfs_panic(tree_fs_info(tree), err,
569 "Locking error: Extent tree was modified by another thread while locked.");
573 * clear some bits on a range in the tree. This may require splitting
574 * or inserting elements in the tree, so the gfp mask is used to
575 * indicate which allocations or sleeping are allowed.
577 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
578 * the given range from the tree regardless of state (ie for truncate).
580 * the range [start, end] is inclusive.
582 * This takes the tree lock, and returns 0 on success and < 0 on error.
584 static int __clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
585 unsigned bits, int wake, int delete,
586 struct extent_state **cached_state,
587 gfp_t mask, struct extent_changeset *changeset)
589 struct extent_state *state;
590 struct extent_state *cached;
591 struct extent_state *prealloc = NULL;
592 struct rb_node *node;
597 btrfs_debug_check_extent_io_range(tree, start, end);
599 if (bits & EXTENT_DELALLOC)
600 bits |= EXTENT_NORESERVE;
603 bits |= ~EXTENT_CTLBITS;
604 bits |= EXTENT_FIRST_DELALLOC;
606 if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY))
609 if (!prealloc && gfpflags_allow_blocking(mask)) {
611 * Don't care for allocation failure here because we might end
612 * up not needing the pre-allocated extent state at all, which
613 * is the case if we only have in the tree extent states that
614 * cover our input range and don't cover too any other range.
615 * If we end up needing a new extent state we allocate it later.
617 prealloc = alloc_extent_state(mask);
620 spin_lock(&tree->lock);
622 cached = *cached_state;
625 *cached_state = NULL;
629 if (cached && extent_state_in_tree(cached) &&
630 cached->start <= start && cached->end > start) {
632 refcount_dec(&cached->refs);
637 free_extent_state(cached);
640 * this search will find the extents that end after
643 node = tree_search(tree, start);
646 state = rb_entry(node, struct extent_state, rb_node);
648 if (state->start > end)
650 WARN_ON(state->end < start);
651 last_end = state->end;
653 /* the state doesn't have the wanted bits, go ahead */
654 if (!(state->state & bits)) {
655 state = next_state(state);
660 * | ---- desired range ---- |
662 * | ------------- state -------------- |
664 * We need to split the extent we found, and may flip
665 * bits on second half.
667 * If the extent we found extends past our range, we
668 * just split and search again. It'll get split again
669 * the next time though.
671 * If the extent we found is inside our range, we clear
672 * the desired bit on it.
675 if (state->start < start) {
676 prealloc = alloc_extent_state_atomic(prealloc);
678 err = split_state(tree, state, prealloc, start);
680 extent_io_tree_panic(tree, err);
685 if (state->end <= end) {
686 state = clear_state_bit(tree, state, &bits, wake,
693 * | ---- desired range ---- |
695 * We need to split the extent, and clear the bit
698 if (state->start <= end && state->end > end) {
699 prealloc = alloc_extent_state_atomic(prealloc);
701 err = split_state(tree, state, prealloc, end + 1);
703 extent_io_tree_panic(tree, err);
708 clear_state_bit(tree, prealloc, &bits, wake, changeset);
714 state = clear_state_bit(tree, state, &bits, wake, changeset);
716 if (last_end == (u64)-1)
718 start = last_end + 1;
719 if (start <= end && state && !need_resched())
725 spin_unlock(&tree->lock);
726 if (gfpflags_allow_blocking(mask))
731 spin_unlock(&tree->lock);
733 free_extent_state(prealloc);
739 static void wait_on_state(struct extent_io_tree *tree,
740 struct extent_state *state)
741 __releases(tree->lock)
742 __acquires(tree->lock)
745 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
746 spin_unlock(&tree->lock);
748 spin_lock(&tree->lock);
749 finish_wait(&state->wq, &wait);
753 * waits for one or more bits to clear on a range in the state tree.
754 * The range [start, end] is inclusive.
755 * The tree lock is taken by this function
757 static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
760 struct extent_state *state;
761 struct rb_node *node;
763 btrfs_debug_check_extent_io_range(tree, start, end);
765 spin_lock(&tree->lock);
769 * this search will find all the extents that end after
772 node = tree_search(tree, start);
777 state = rb_entry(node, struct extent_state, rb_node);
779 if (state->start > end)
782 if (state->state & bits) {
783 start = state->start;
784 refcount_inc(&state->refs);
785 wait_on_state(tree, state);
786 free_extent_state(state);
789 start = state->end + 1;
794 if (!cond_resched_lock(&tree->lock)) {
795 node = rb_next(node);
800 spin_unlock(&tree->lock);
803 static void set_state_bits(struct extent_io_tree *tree,
804 struct extent_state *state,
805 unsigned *bits, struct extent_changeset *changeset)
807 unsigned bits_to_set = *bits & ~EXTENT_CTLBITS;
809 set_state_cb(tree, state, bits);
810 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
811 u64 range = state->end - state->start + 1;
812 tree->dirty_bytes += range;
814 add_extent_changeset(state, bits_to_set, changeset, 1);
815 state->state |= bits_to_set;
818 static void cache_state_if_flags(struct extent_state *state,
819 struct extent_state **cached_ptr,
822 if (cached_ptr && !(*cached_ptr)) {
823 if (!flags || (state->state & flags)) {
825 refcount_inc(&state->refs);
830 static void cache_state(struct extent_state *state,
831 struct extent_state **cached_ptr)
833 return cache_state_if_flags(state, cached_ptr,
834 EXTENT_IOBITS | EXTENT_BOUNDARY);
838 * set some bits on a range in the tree. This may require allocations or
839 * sleeping, so the gfp mask is used to indicate what is allowed.
841 * If any of the exclusive bits are set, this will fail with -EEXIST if some
842 * part of the range already has the desired bits set. The start of the
843 * existing range is returned in failed_start in this case.
845 * [start, end] is inclusive This takes the tree lock.
848 static int __must_check
849 __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
850 unsigned bits, unsigned exclusive_bits,
851 u64 *failed_start, struct extent_state **cached_state,
852 gfp_t mask, struct extent_changeset *changeset)
854 struct extent_state *state;
855 struct extent_state *prealloc = NULL;
856 struct rb_node *node;
858 struct rb_node *parent;
863 btrfs_debug_check_extent_io_range(tree, start, end);
865 bits |= EXTENT_FIRST_DELALLOC;
867 if (!prealloc && gfpflags_allow_blocking(mask)) {
869 * Don't care for allocation failure here because we might end
870 * up not needing the pre-allocated extent state at all, which
871 * is the case if we only have in the tree extent states that
872 * cover our input range and don't cover too any other range.
873 * If we end up needing a new extent state we allocate it later.
875 prealloc = alloc_extent_state(mask);
878 spin_lock(&tree->lock);
879 if (cached_state && *cached_state) {
880 state = *cached_state;
881 if (state->start <= start && state->end > start &&
882 extent_state_in_tree(state)) {
883 node = &state->rb_node;
888 * this search will find all the extents that end after
891 node = tree_search_for_insert(tree, start, &p, &parent);
893 prealloc = alloc_extent_state_atomic(prealloc);
895 err = insert_state(tree, prealloc, start, end,
896 &p, &parent, &bits, changeset);
898 extent_io_tree_panic(tree, err);
900 cache_state(prealloc, cached_state);
904 state = rb_entry(node, struct extent_state, rb_node);
906 last_start = state->start;
907 last_end = state->end;
910 * | ---- desired range ---- |
913 * Just lock what we found and keep going
915 if (state->start == start && state->end <= end) {
916 if (state->state & exclusive_bits) {
917 *failed_start = state->start;
922 set_state_bits(tree, state, &bits, changeset);
923 cache_state(state, cached_state);
924 merge_state(tree, state);
925 if (last_end == (u64)-1)
927 start = last_end + 1;
928 state = next_state(state);
929 if (start < end && state && state->start == start &&
936 * | ---- desired range ---- |
939 * | ------------- state -------------- |
941 * We need to split the extent we found, and may flip bits on
944 * If the extent we found extends past our
945 * range, we just split and search again. It'll get split
946 * again the next time though.
948 * If the extent we found is inside our range, we set the
951 if (state->start < start) {
952 if (state->state & exclusive_bits) {
953 *failed_start = start;
958 prealloc = alloc_extent_state_atomic(prealloc);
960 err = split_state(tree, state, prealloc, start);
962 extent_io_tree_panic(tree, err);
967 if (state->end <= end) {
968 set_state_bits(tree, state, &bits, changeset);
969 cache_state(state, cached_state);
970 merge_state(tree, state);
971 if (last_end == (u64)-1)
973 start = last_end + 1;
974 state = next_state(state);
975 if (start < end && state && state->start == start &&
982 * | ---- desired range ---- |
983 * | state | or | state |
985 * There's a hole, we need to insert something in it and
986 * ignore the extent we found.
988 if (state->start > start) {
990 if (end < last_start)
993 this_end = last_start - 1;
995 prealloc = alloc_extent_state_atomic(prealloc);
999 * Avoid to free 'prealloc' if it can be merged with
1002 err = insert_state(tree, prealloc, start, this_end,
1003 NULL, NULL, &bits, changeset);
1005 extent_io_tree_panic(tree, err);
1007 cache_state(prealloc, cached_state);
1009 start = this_end + 1;
1013 * | ---- desired range ---- |
1015 * We need to split the extent, and set the bit
1018 if (state->start <= end && state->end > end) {
1019 if (state->state & exclusive_bits) {
1020 *failed_start = start;
1025 prealloc = alloc_extent_state_atomic(prealloc);
1027 err = split_state(tree, state, prealloc, end + 1);
1029 extent_io_tree_panic(tree, err);
1031 set_state_bits(tree, prealloc, &bits, changeset);
1032 cache_state(prealloc, cached_state);
1033 merge_state(tree, prealloc);
1041 spin_unlock(&tree->lock);
1042 if (gfpflags_allow_blocking(mask))
1047 spin_unlock(&tree->lock);
1049 free_extent_state(prealloc);
1055 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1056 unsigned bits, u64 * failed_start,
1057 struct extent_state **cached_state, gfp_t mask)
1059 return __set_extent_bit(tree, start, end, bits, 0, failed_start,
1060 cached_state, mask, NULL);
1065 * convert_extent_bit - convert all bits in a given range from one bit to
1067 * @tree: the io tree to search
1068 * @start: the start offset in bytes
1069 * @end: the end offset in bytes (inclusive)
1070 * @bits: the bits to set in this range
1071 * @clear_bits: the bits to clear in this range
1072 * @cached_state: state that we're going to cache
1074 * This will go through and set bits for the given range. If any states exist
1075 * already in this range they are set with the given bit and cleared of the
1076 * clear_bits. This is only meant to be used by things that are mergeable, ie
1077 * converting from say DELALLOC to DIRTY. This is not meant to be used with
1078 * boundary bits like LOCK.
1080 * All allocations are done with GFP_NOFS.
1082 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1083 unsigned bits, unsigned clear_bits,
1084 struct extent_state **cached_state)
1086 struct extent_state *state;
1087 struct extent_state *prealloc = NULL;
1088 struct rb_node *node;
1090 struct rb_node *parent;
1094 bool first_iteration = true;
1096 btrfs_debug_check_extent_io_range(tree, start, end);
1101 * Best effort, don't worry if extent state allocation fails
1102 * here for the first iteration. We might have a cached state
1103 * that matches exactly the target range, in which case no
1104 * extent state allocations are needed. We'll only know this
1105 * after locking the tree.
1107 prealloc = alloc_extent_state(GFP_NOFS);
1108 if (!prealloc && !first_iteration)
1112 spin_lock(&tree->lock);
1113 if (cached_state && *cached_state) {
1114 state = *cached_state;
1115 if (state->start <= start && state->end > start &&
1116 extent_state_in_tree(state)) {
1117 node = &state->rb_node;
1123 * this search will find all the extents that end after
1126 node = tree_search_for_insert(tree, start, &p, &parent);
1128 prealloc = alloc_extent_state_atomic(prealloc);
1133 err = insert_state(tree, prealloc, start, end,
1134 &p, &parent, &bits, NULL);
1136 extent_io_tree_panic(tree, err);
1137 cache_state(prealloc, cached_state);
1141 state = rb_entry(node, struct extent_state, rb_node);
1143 last_start = state->start;
1144 last_end = state->end;
1147 * | ---- desired range ---- |
1150 * Just lock what we found and keep going
1152 if (state->start == start && state->end <= end) {
1153 set_state_bits(tree, state, &bits, NULL);
1154 cache_state(state, cached_state);
1155 state = clear_state_bit(tree, state, &clear_bits, 0, NULL);
1156 if (last_end == (u64)-1)
1158 start = last_end + 1;
1159 if (start < end && state && state->start == start &&
1166 * | ---- desired range ---- |
1169 * | ------------- state -------------- |
1171 * We need to split the extent we found, and may flip bits on
1174 * If the extent we found extends past our
1175 * range, we just split and search again. It'll get split
1176 * again the next time though.
1178 * If the extent we found is inside our range, we set the
1179 * desired bit on it.
1181 if (state->start < start) {
1182 prealloc = alloc_extent_state_atomic(prealloc);
1187 err = split_state(tree, state, prealloc, start);
1189 extent_io_tree_panic(tree, err);
1193 if (state->end <= end) {
1194 set_state_bits(tree, state, &bits, NULL);
1195 cache_state(state, cached_state);
1196 state = clear_state_bit(tree, state, &clear_bits, 0,
1198 if (last_end == (u64)-1)
1200 start = last_end + 1;
1201 if (start < end && state && state->start == start &&
1208 * | ---- desired range ---- |
1209 * | state | or | state |
1211 * There's a hole, we need to insert something in it and
1212 * ignore the extent we found.
1214 if (state->start > start) {
1216 if (end < last_start)
1219 this_end = last_start - 1;
1221 prealloc = alloc_extent_state_atomic(prealloc);
1228 * Avoid to free 'prealloc' if it can be merged with
1231 err = insert_state(tree, prealloc, start, this_end,
1232 NULL, NULL, &bits, NULL);
1234 extent_io_tree_panic(tree, err);
1235 cache_state(prealloc, cached_state);
1237 start = this_end + 1;
1241 * | ---- desired range ---- |
1243 * We need to split the extent, and set the bit
1246 if (state->start <= end && state->end > end) {
1247 prealloc = alloc_extent_state_atomic(prealloc);
1253 err = split_state(tree, state, prealloc, end + 1);
1255 extent_io_tree_panic(tree, err);
1257 set_state_bits(tree, prealloc, &bits, NULL);
1258 cache_state(prealloc, cached_state);
1259 clear_state_bit(tree, prealloc, &clear_bits, 0, NULL);
1267 spin_unlock(&tree->lock);
1269 first_iteration = false;
1273 spin_unlock(&tree->lock);
1275 free_extent_state(prealloc);
1280 /* wrappers around set/clear extent bit */
1281 int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1282 unsigned bits, struct extent_changeset *changeset)
1285 * We don't support EXTENT_LOCKED yet, as current changeset will
1286 * record any bits changed, so for EXTENT_LOCKED case, it will
1287 * either fail with -EEXIST or changeset will record the whole
1290 BUG_ON(bits & EXTENT_LOCKED);
1292 return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL, GFP_NOFS,
1296 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1297 unsigned bits, int wake, int delete,
1298 struct extent_state **cached, gfp_t mask)
1300 return __clear_extent_bit(tree, start, end, bits, wake, delete,
1301 cached, mask, NULL);
1304 int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1305 unsigned bits, struct extent_changeset *changeset)
1308 * Don't support EXTENT_LOCKED case, same reason as
1309 * set_record_extent_bits().
1311 BUG_ON(bits & EXTENT_LOCKED);
1313 return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, GFP_NOFS,
1318 * either insert or lock state struct between start and end use mask to tell
1319 * us if waiting is desired.
1321 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1322 struct extent_state **cached_state)
1328 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED,
1329 EXTENT_LOCKED, &failed_start,
1330 cached_state, GFP_NOFS, NULL);
1331 if (err == -EEXIST) {
1332 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1333 start = failed_start;
1336 WARN_ON(start > end);
1341 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1346 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1347 &failed_start, NULL, GFP_NOFS, NULL);
1348 if (err == -EEXIST) {
1349 if (failed_start > start)
1350 clear_extent_bit(tree, start, failed_start - 1,
1351 EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS);
1357 void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
1359 unsigned long index = start >> PAGE_SHIFT;
1360 unsigned long end_index = end >> PAGE_SHIFT;
1363 while (index <= end_index) {
1364 page = find_get_page(inode->i_mapping, index);
1365 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1366 clear_page_dirty_for_io(page);
1372 void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
1374 unsigned long index = start >> PAGE_SHIFT;
1375 unsigned long end_index = end >> PAGE_SHIFT;
1378 while (index <= end_index) {
1379 page = find_get_page(inode->i_mapping, index);
1380 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1381 __set_page_dirty_nobuffers(page);
1382 account_page_redirty(page);
1389 * helper function to set both pages and extents in the tree writeback
1391 static void set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
1393 tree->ops->set_range_writeback(tree->private_data, start, end);
1396 /* find the first state struct with 'bits' set after 'start', and
1397 * return it. tree->lock must be held. NULL will returned if
1398 * nothing was found after 'start'
1400 static struct extent_state *
1401 find_first_extent_bit_state(struct extent_io_tree *tree,
1402 u64 start, unsigned bits)
1404 struct rb_node *node;
1405 struct extent_state *state;
1408 * this search will find all the extents that end after
1411 node = tree_search(tree, start);
1416 state = rb_entry(node, struct extent_state, rb_node);
1417 if (state->end >= start && (state->state & bits))
1420 node = rb_next(node);
1429 * find the first offset in the io tree with 'bits' set. zero is
1430 * returned if we find something, and *start_ret and *end_ret are
1431 * set to reflect the state struct that was found.
1433 * If nothing was found, 1 is returned. If found something, return 0.
1435 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1436 u64 *start_ret, u64 *end_ret, unsigned bits,
1437 struct extent_state **cached_state)
1439 struct extent_state *state;
1443 spin_lock(&tree->lock);
1444 if (cached_state && *cached_state) {
1445 state = *cached_state;
1446 if (state->end == start - 1 && extent_state_in_tree(state)) {
1447 n = rb_next(&state->rb_node);
1449 state = rb_entry(n, struct extent_state,
1451 if (state->state & bits)
1455 free_extent_state(*cached_state);
1456 *cached_state = NULL;
1459 free_extent_state(*cached_state);
1460 *cached_state = NULL;
1463 state = find_first_extent_bit_state(tree, start, bits);
1466 cache_state_if_flags(state, cached_state, 0);
1467 *start_ret = state->start;
1468 *end_ret = state->end;
1472 spin_unlock(&tree->lock);
1477 * find a contiguous range of bytes in the file marked as delalloc, not
1478 * more than 'max_bytes'. start and end are used to return the range,
1480 * 1 is returned if we find something, 0 if nothing was in the tree
1482 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1483 u64 *start, u64 *end, u64 max_bytes,
1484 struct extent_state **cached_state)
1486 struct rb_node *node;
1487 struct extent_state *state;
1488 u64 cur_start = *start;
1490 u64 total_bytes = 0;
1492 spin_lock(&tree->lock);
1495 * this search will find all the extents that end after
1498 node = tree_search(tree, cur_start);
1506 state = rb_entry(node, struct extent_state, rb_node);
1507 if (found && (state->start != cur_start ||
1508 (state->state & EXTENT_BOUNDARY))) {
1511 if (!(state->state & EXTENT_DELALLOC)) {
1517 *start = state->start;
1518 *cached_state = state;
1519 refcount_inc(&state->refs);
1523 cur_start = state->end + 1;
1524 node = rb_next(node);
1525 total_bytes += state->end - state->start + 1;
1526 if (total_bytes >= max_bytes)
1532 spin_unlock(&tree->lock);
1536 static int __process_pages_contig(struct address_space *mapping,
1537 struct page *locked_page,
1538 pgoff_t start_index, pgoff_t end_index,
1539 unsigned long page_ops, pgoff_t *index_ret);
1541 static noinline void __unlock_for_delalloc(struct inode *inode,
1542 struct page *locked_page,
1545 unsigned long index = start >> PAGE_SHIFT;
1546 unsigned long end_index = end >> PAGE_SHIFT;
1548 ASSERT(locked_page);
1549 if (index == locked_page->index && end_index == index)
1552 __process_pages_contig(inode->i_mapping, locked_page, index, end_index,
1556 static noinline int lock_delalloc_pages(struct inode *inode,
1557 struct page *locked_page,
1561 unsigned long index = delalloc_start >> PAGE_SHIFT;
1562 unsigned long index_ret = index;
1563 unsigned long end_index = delalloc_end >> PAGE_SHIFT;
1566 ASSERT(locked_page);
1567 if (index == locked_page->index && index == end_index)
1570 ret = __process_pages_contig(inode->i_mapping, locked_page, index,
1571 end_index, PAGE_LOCK, &index_ret);
1573 __unlock_for_delalloc(inode, locked_page, delalloc_start,
1574 (u64)index_ret << PAGE_SHIFT);
1579 * find a contiguous range of bytes in the file marked as delalloc, not
1580 * more than 'max_bytes'. start and end are used to return the range,
1582 * 1 is returned if we find something, 0 if nothing was in the tree
1584 STATIC u64 find_lock_delalloc_range(struct inode *inode,
1585 struct extent_io_tree *tree,
1586 struct page *locked_page, u64 *start,
1587 u64 *end, u64 max_bytes)
1592 struct extent_state *cached_state = NULL;
1597 /* step one, find a bunch of delalloc bytes starting at start */
1598 delalloc_start = *start;
1600 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1601 max_bytes, &cached_state);
1602 if (!found || delalloc_end <= *start) {
1603 *start = delalloc_start;
1604 *end = delalloc_end;
1605 free_extent_state(cached_state);
1610 * start comes from the offset of locked_page. We have to lock
1611 * pages in order, so we can't process delalloc bytes before
1614 if (delalloc_start < *start)
1615 delalloc_start = *start;
1618 * make sure to limit the number of pages we try to lock down
1620 if (delalloc_end + 1 - delalloc_start > max_bytes)
1621 delalloc_end = delalloc_start + max_bytes - 1;
1623 /* step two, lock all the pages after the page that has start */
1624 ret = lock_delalloc_pages(inode, locked_page,
1625 delalloc_start, delalloc_end);
1626 if (ret == -EAGAIN) {
1627 /* some of the pages are gone, lets avoid looping by
1628 * shortening the size of the delalloc range we're searching
1630 free_extent_state(cached_state);
1631 cached_state = NULL;
1633 max_bytes = PAGE_SIZE;
1641 BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */
1643 /* step three, lock the state bits for the whole range */
1644 lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state);
1646 /* then test to make sure it is all still delalloc */
1647 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1648 EXTENT_DELALLOC, 1, cached_state);
1650 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1651 &cached_state, GFP_NOFS);
1652 __unlock_for_delalloc(inode, locked_page,
1653 delalloc_start, delalloc_end);
1657 free_extent_state(cached_state);
1658 *start = delalloc_start;
1659 *end = delalloc_end;
1664 static int __process_pages_contig(struct address_space *mapping,
1665 struct page *locked_page,
1666 pgoff_t start_index, pgoff_t end_index,
1667 unsigned long page_ops, pgoff_t *index_ret)
1669 unsigned long nr_pages = end_index - start_index + 1;
1670 unsigned long pages_locked = 0;
1671 pgoff_t index = start_index;
1672 struct page *pages[16];
1677 if (page_ops & PAGE_LOCK) {
1678 ASSERT(page_ops == PAGE_LOCK);
1679 ASSERT(index_ret && *index_ret == start_index);
1682 if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0)
1683 mapping_set_error(mapping, -EIO);
1685 while (nr_pages > 0) {
1686 ret = find_get_pages_contig(mapping, index,
1687 min_t(unsigned long,
1688 nr_pages, ARRAY_SIZE(pages)), pages);
1691 * Only if we're going to lock these pages,
1692 * can we find nothing at @index.
1694 ASSERT(page_ops & PAGE_LOCK);
1699 for (i = 0; i < ret; i++) {
1700 if (page_ops & PAGE_SET_PRIVATE2)
1701 SetPagePrivate2(pages[i]);
1703 if (pages[i] == locked_page) {
1708 if (page_ops & PAGE_CLEAR_DIRTY)
1709 clear_page_dirty_for_io(pages[i]);
1710 if (page_ops & PAGE_SET_WRITEBACK)
1711 set_page_writeback(pages[i]);
1712 if (page_ops & PAGE_SET_ERROR)
1713 SetPageError(pages[i]);
1714 if (page_ops & PAGE_END_WRITEBACK)
1715 end_page_writeback(pages[i]);
1716 if (page_ops & PAGE_UNLOCK)
1717 unlock_page(pages[i]);
1718 if (page_ops & PAGE_LOCK) {
1719 lock_page(pages[i]);
1720 if (!PageDirty(pages[i]) ||
1721 pages[i]->mapping != mapping) {
1722 unlock_page(pages[i]);
1736 if (err && index_ret)
1737 *index_ret = start_index + pages_locked - 1;
1741 void extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end,
1742 u64 delalloc_end, struct page *locked_page,
1743 unsigned clear_bits,
1744 unsigned long page_ops)
1746 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, clear_bits, 1, 0,
1749 __process_pages_contig(inode->i_mapping, locked_page,
1750 start >> PAGE_SHIFT, end >> PAGE_SHIFT,
1755 * count the number of bytes in the tree that have a given bit(s)
1756 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1757 * cached. The total number found is returned.
1759 u64 count_range_bits(struct extent_io_tree *tree,
1760 u64 *start, u64 search_end, u64 max_bytes,
1761 unsigned bits, int contig)
1763 struct rb_node *node;
1764 struct extent_state *state;
1765 u64 cur_start = *start;
1766 u64 total_bytes = 0;
1770 if (WARN_ON(search_end <= cur_start))
1773 spin_lock(&tree->lock);
1774 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1775 total_bytes = tree->dirty_bytes;
1779 * this search will find all the extents that end after
1782 node = tree_search(tree, cur_start);
1787 state = rb_entry(node, struct extent_state, rb_node);
1788 if (state->start > search_end)
1790 if (contig && found && state->start > last + 1)
1792 if (state->end >= cur_start && (state->state & bits) == bits) {
1793 total_bytes += min(search_end, state->end) + 1 -
1794 max(cur_start, state->start);
1795 if (total_bytes >= max_bytes)
1798 *start = max(cur_start, state->start);
1802 } else if (contig && found) {
1805 node = rb_next(node);
1810 spin_unlock(&tree->lock);
1815 * set the private field for a given byte offset in the tree. If there isn't
1816 * an extent_state there already, this does nothing.
1818 static noinline int set_state_failrec(struct extent_io_tree *tree, u64 start,
1819 struct io_failure_record *failrec)
1821 struct rb_node *node;
1822 struct extent_state *state;
1825 spin_lock(&tree->lock);
1827 * this search will find all the extents that end after
1830 node = tree_search(tree, start);
1835 state = rb_entry(node, struct extent_state, rb_node);
1836 if (state->start != start) {
1840 state->failrec = failrec;
1842 spin_unlock(&tree->lock);
1846 static noinline int get_state_failrec(struct extent_io_tree *tree, u64 start,
1847 struct io_failure_record **failrec)
1849 struct rb_node *node;
1850 struct extent_state *state;
1853 spin_lock(&tree->lock);
1855 * this search will find all the extents that end after
1858 node = tree_search(tree, start);
1863 state = rb_entry(node, struct extent_state, rb_node);
1864 if (state->start != start) {
1868 *failrec = state->failrec;
1870 spin_unlock(&tree->lock);
1875 * searches a range in the state tree for a given mask.
1876 * If 'filled' == 1, this returns 1 only if every extent in the tree
1877 * has the bits set. Otherwise, 1 is returned if any bit in the
1878 * range is found set.
1880 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1881 unsigned bits, int filled, struct extent_state *cached)
1883 struct extent_state *state = NULL;
1884 struct rb_node *node;
1887 spin_lock(&tree->lock);
1888 if (cached && extent_state_in_tree(cached) && cached->start <= start &&
1889 cached->end > start)
1890 node = &cached->rb_node;
1892 node = tree_search(tree, start);
1893 while (node && start <= end) {
1894 state = rb_entry(node, struct extent_state, rb_node);
1896 if (filled && state->start > start) {
1901 if (state->start > end)
1904 if (state->state & bits) {
1908 } else if (filled) {
1913 if (state->end == (u64)-1)
1916 start = state->end + 1;
1919 node = rb_next(node);
1926 spin_unlock(&tree->lock);
1931 * helper function to set a given page up to date if all the
1932 * extents in the tree for that page are up to date
1934 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
1936 u64 start = page_offset(page);
1937 u64 end = start + PAGE_SIZE - 1;
1938 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1939 SetPageUptodate(page);
1942 int free_io_failure(struct extent_io_tree *failure_tree,
1943 struct extent_io_tree *io_tree,
1944 struct io_failure_record *rec)
1949 set_state_failrec(failure_tree, rec->start, NULL);
1950 ret = clear_extent_bits(failure_tree, rec->start,
1951 rec->start + rec->len - 1,
1952 EXTENT_LOCKED | EXTENT_DIRTY);
1956 ret = clear_extent_bits(io_tree, rec->start,
1957 rec->start + rec->len - 1,
1967 * this bypasses the standard btrfs submit functions deliberately, as
1968 * the standard behavior is to write all copies in a raid setup. here we only
1969 * want to write the one bad copy. so we do the mapping for ourselves and issue
1970 * submit_bio directly.
1971 * to avoid any synchronization issues, wait for the data after writing, which
1972 * actually prevents the read that triggered the error from finishing.
1973 * currently, there can be no more than two copies of every data bit. thus,
1974 * exactly one rewrite is required.
1976 int repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start,
1977 u64 length, u64 logical, struct page *page,
1978 unsigned int pg_offset, int mirror_num)
1981 struct btrfs_device *dev;
1984 struct btrfs_bio *bbio = NULL;
1987 ASSERT(!(fs_info->sb->s_flags & MS_RDONLY));
1988 BUG_ON(!mirror_num);
1990 bio = btrfs_io_bio_alloc(1);
1991 bio->bi_iter.bi_size = 0;
1992 map_length = length;
1995 * Avoid races with device replace and make sure our bbio has devices
1996 * associated to its stripes that don't go away while we are doing the
1997 * read repair operation.
1999 btrfs_bio_counter_inc_blocked(fs_info);
2000 if (btrfs_is_parity_mirror(fs_info, logical, length, mirror_num)) {
2002 * Note that we don't use BTRFS_MAP_WRITE because it's supposed
2003 * to update all raid stripes, but here we just want to correct
2004 * bad stripe, thus BTRFS_MAP_READ is abused to only get the bad
2005 * stripe's dev and sector.
2007 ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, logical,
2008 &map_length, &bbio, 0);
2010 btrfs_bio_counter_dec(fs_info);
2014 ASSERT(bbio->mirror_num == 1);
2016 ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical,
2017 &map_length, &bbio, mirror_num);
2019 btrfs_bio_counter_dec(fs_info);
2023 BUG_ON(mirror_num != bbio->mirror_num);
2026 sector = bbio->stripes[bbio->mirror_num - 1].physical >> 9;
2027 bio->bi_iter.bi_sector = sector;
2028 dev = bbio->stripes[bbio->mirror_num - 1].dev;
2029 btrfs_put_bbio(bbio);
2030 if (!dev || !dev->bdev || !dev->writeable) {
2031 btrfs_bio_counter_dec(fs_info);
2035 bio->bi_bdev = dev->bdev;
2036 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC;
2037 bio_add_page(bio, page, length, pg_offset);
2039 if (btrfsic_submit_bio_wait(bio)) {
2040 /* try to remap that extent elsewhere? */
2041 btrfs_bio_counter_dec(fs_info);
2043 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
2047 btrfs_info_rl_in_rcu(fs_info,
2048 "read error corrected: ino %llu off %llu (dev %s sector %llu)",
2050 rcu_str_deref(dev->name), sector);
2051 btrfs_bio_counter_dec(fs_info);
2056 int repair_eb_io_failure(struct btrfs_fs_info *fs_info,
2057 struct extent_buffer *eb, int mirror_num)
2059 u64 start = eb->start;
2060 unsigned long i, num_pages = num_extent_pages(eb->start, eb->len);
2063 if (fs_info->sb->s_flags & MS_RDONLY)
2066 for (i = 0; i < num_pages; i++) {
2067 struct page *p = eb->pages[i];
2069 ret = repair_io_failure(fs_info, 0, start, PAGE_SIZE, start, p,
2070 start - page_offset(p), mirror_num);
2080 * each time an IO finishes, we do a fast check in the IO failure tree
2081 * to see if we need to process or clean up an io_failure_record
2083 int clean_io_failure(struct btrfs_fs_info *fs_info,
2084 struct extent_io_tree *failure_tree,
2085 struct extent_io_tree *io_tree, u64 start,
2086 struct page *page, u64 ino, unsigned int pg_offset)
2089 struct io_failure_record *failrec;
2090 struct extent_state *state;
2095 ret = count_range_bits(failure_tree, &private, (u64)-1, 1,
2100 ret = get_state_failrec(failure_tree, start, &failrec);
2104 BUG_ON(!failrec->this_mirror);
2106 if (failrec->in_validation) {
2107 /* there was no real error, just free the record */
2108 btrfs_debug(fs_info,
2109 "clean_io_failure: freeing dummy error at %llu",
2113 if (fs_info->sb->s_flags & MS_RDONLY)
2116 spin_lock(&io_tree->lock);
2117 state = find_first_extent_bit_state(io_tree,
2120 spin_unlock(&io_tree->lock);
2122 if (state && state->start <= failrec->start &&
2123 state->end >= failrec->start + failrec->len - 1) {
2124 num_copies = btrfs_num_copies(fs_info, failrec->logical,
2126 if (num_copies > 1) {
2127 repair_io_failure(fs_info, ino, start, failrec->len,
2128 failrec->logical, page, pg_offset,
2129 failrec->failed_mirror);
2134 free_io_failure(failure_tree, io_tree, failrec);
2140 * Can be called when
2141 * - hold extent lock
2142 * - under ordered extent
2143 * - the inode is freeing
2145 void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start, u64 end)
2147 struct extent_io_tree *failure_tree = &inode->io_failure_tree;
2148 struct io_failure_record *failrec;
2149 struct extent_state *state, *next;
2151 if (RB_EMPTY_ROOT(&failure_tree->state))
2154 spin_lock(&failure_tree->lock);
2155 state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY);
2157 if (state->start > end)
2160 ASSERT(state->end <= end);
2162 next = next_state(state);
2164 failrec = state->failrec;
2165 free_extent_state(state);
2170 spin_unlock(&failure_tree->lock);
2173 int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end,
2174 struct io_failure_record **failrec_ret)
2176 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2177 struct io_failure_record *failrec;
2178 struct extent_map *em;
2179 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2180 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2181 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2185 ret = get_state_failrec(failure_tree, start, &failrec);
2187 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2191 failrec->start = start;
2192 failrec->len = end - start + 1;
2193 failrec->this_mirror = 0;
2194 failrec->bio_flags = 0;
2195 failrec->in_validation = 0;
2197 read_lock(&em_tree->lock);
2198 em = lookup_extent_mapping(em_tree, start, failrec->len);
2200 read_unlock(&em_tree->lock);
2205 if (em->start > start || em->start + em->len <= start) {
2206 free_extent_map(em);
2209 read_unlock(&em_tree->lock);
2215 logical = start - em->start;
2216 logical = em->block_start + logical;
2217 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2218 logical = em->block_start;
2219 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2220 extent_set_compress_type(&failrec->bio_flags,
2224 btrfs_debug(fs_info,
2225 "Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu",
2226 logical, start, failrec->len);
2228 failrec->logical = logical;
2229 free_extent_map(em);
2231 /* set the bits in the private failure tree */
2232 ret = set_extent_bits(failure_tree, start, end,
2233 EXTENT_LOCKED | EXTENT_DIRTY);
2235 ret = set_state_failrec(failure_tree, start, failrec);
2236 /* set the bits in the inode's tree */
2238 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED);
2244 btrfs_debug(fs_info,
2245 "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d",
2246 failrec->logical, failrec->start, failrec->len,
2247 failrec->in_validation);
2249 * when data can be on disk more than twice, add to failrec here
2250 * (e.g. with a list for failed_mirror) to make
2251 * clean_io_failure() clean all those errors at once.
2255 *failrec_ret = failrec;
2260 int btrfs_check_repairable(struct inode *inode, struct bio *failed_bio,
2261 struct io_failure_record *failrec, int failed_mirror)
2263 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2266 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
2267 if (num_copies == 1) {
2269 * we only have a single copy of the data, so don't bother with
2270 * all the retry and error correction code that follows. no
2271 * matter what the error is, it is very likely to persist.
2273 btrfs_debug(fs_info,
2274 "Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
2275 num_copies, failrec->this_mirror, failed_mirror);
2280 * there are two premises:
2281 * a) deliver good data to the caller
2282 * b) correct the bad sectors on disk
2284 if (failed_bio->bi_vcnt > 1) {
2286 * to fulfill b), we need to know the exact failing sectors, as
2287 * we don't want to rewrite any more than the failed ones. thus,
2288 * we need separate read requests for the failed bio
2290 * if the following BUG_ON triggers, our validation request got
2291 * merged. we need separate requests for our algorithm to work.
2293 BUG_ON(failrec->in_validation);
2294 failrec->in_validation = 1;
2295 failrec->this_mirror = failed_mirror;
2298 * we're ready to fulfill a) and b) alongside. get a good copy
2299 * of the failed sector and if we succeed, we have setup
2300 * everything for repair_io_failure to do the rest for us.
2302 if (failrec->in_validation) {
2303 BUG_ON(failrec->this_mirror != failed_mirror);
2304 failrec->in_validation = 0;
2305 failrec->this_mirror = 0;
2307 failrec->failed_mirror = failed_mirror;
2308 failrec->this_mirror++;
2309 if (failrec->this_mirror == failed_mirror)
2310 failrec->this_mirror++;
2313 if (failrec->this_mirror > num_copies) {
2314 btrfs_debug(fs_info,
2315 "Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
2316 num_copies, failrec->this_mirror, failed_mirror);
2324 struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio,
2325 struct io_failure_record *failrec,
2326 struct page *page, int pg_offset, int icsum,
2327 bio_end_io_t *endio_func, void *data)
2329 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2331 struct btrfs_io_bio *btrfs_failed_bio;
2332 struct btrfs_io_bio *btrfs_bio;
2334 bio = btrfs_io_bio_alloc(1);
2335 bio->bi_end_io = endio_func;
2336 bio->bi_iter.bi_sector = failrec->logical >> 9;
2337 bio->bi_bdev = fs_info->fs_devices->latest_bdev;
2338 bio->bi_iter.bi_size = 0;
2339 bio->bi_private = data;
2341 btrfs_failed_bio = btrfs_io_bio(failed_bio);
2342 if (btrfs_failed_bio->csum) {
2343 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
2345 btrfs_bio = btrfs_io_bio(bio);
2346 btrfs_bio->csum = btrfs_bio->csum_inline;
2348 memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + icsum,
2352 bio_add_page(bio, page, failrec->len, pg_offset);
2358 * this is a generic handler for readpage errors (default
2359 * readpage_io_failed_hook). if other copies exist, read those and write back
2360 * good data to the failed position. does not investigate in remapping the
2361 * failed extent elsewhere, hoping the device will be smart enough to do this as
2365 static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset,
2366 struct page *page, u64 start, u64 end,
2369 struct io_failure_record *failrec;
2370 struct inode *inode = page->mapping->host;
2371 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2372 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2377 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
2379 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
2383 ret = btrfs_check_repairable(inode, failed_bio, failrec, failed_mirror);
2385 free_io_failure(failure_tree, tree, failrec);
2389 if (failed_bio->bi_vcnt > 1)
2390 read_mode |= REQ_FAILFAST_DEV;
2392 phy_offset >>= inode->i_sb->s_blocksize_bits;
2393 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
2394 start - page_offset(page),
2395 (int)phy_offset, failed_bio->bi_end_io,
2397 bio_set_op_attrs(bio, REQ_OP_READ, read_mode);
2399 btrfs_debug(btrfs_sb(inode->i_sb),
2400 "Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d",
2401 read_mode, failrec->this_mirror, failrec->in_validation);
2403 ret = tree->ops->submit_bio_hook(tree->private_data, bio, failrec->this_mirror,
2404 failrec->bio_flags, 0);
2406 free_io_failure(failure_tree, tree, failrec);
2413 /* lots and lots of room for performance fixes in the end_bio funcs */
2415 void end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2417 int uptodate = (err == 0);
2418 struct extent_io_tree *tree;
2421 tree = &BTRFS_I(page->mapping->host)->io_tree;
2423 if (tree->ops && tree->ops->writepage_end_io_hook)
2424 tree->ops->writepage_end_io_hook(page, start, end, NULL,
2428 ClearPageUptodate(page);
2430 ret = err < 0 ? err : -EIO;
2431 mapping_set_error(page->mapping, ret);
2436 * after a writepage IO is done, we need to:
2437 * clear the uptodate bits on error
2438 * clear the writeback bits in the extent tree for this IO
2439 * end_page_writeback if the page has no more pending IO
2441 * Scheduling is not allowed, so the extent state tree is expected
2442 * to have one and only one object corresponding to this IO.
2444 static void end_bio_extent_writepage(struct bio *bio)
2446 struct bio_vec *bvec;
2451 ASSERT(!bio_flagged(bio, BIO_CLONED));
2452 bio_for_each_segment_all(bvec, bio, i) {
2453 struct page *page = bvec->bv_page;
2454 struct inode *inode = page->mapping->host;
2455 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2457 /* We always issue full-page reads, but if some block
2458 * in a page fails to read, blk_update_request() will
2459 * advance bv_offset and adjust bv_len to compensate.
2460 * Print a warning for nonzero offsets, and an error
2461 * if they don't add up to a full page. */
2462 if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) {
2463 if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE)
2465 "partial page write in btrfs with offset %u and length %u",
2466 bvec->bv_offset, bvec->bv_len);
2469 "incomplete page write in btrfs with offset %u and length %u",
2470 bvec->bv_offset, bvec->bv_len);
2473 start = page_offset(page);
2474 end = start + bvec->bv_offset + bvec->bv_len - 1;
2476 end_extent_writepage(page, bio->bi_error, start, end);
2477 end_page_writeback(page);
2484 endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len,
2487 struct extent_state *cached = NULL;
2488 u64 end = start + len - 1;
2490 if (uptodate && tree->track_uptodate)
2491 set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC);
2492 unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC);
2496 * after a readpage IO is done, we need to:
2497 * clear the uptodate bits on error
2498 * set the uptodate bits if things worked
2499 * set the page up to date if all extents in the tree are uptodate
2500 * clear the lock bit in the extent tree
2501 * unlock the page if there are no other extents locked for it
2503 * Scheduling is not allowed, so the extent state tree is expected
2504 * to have one and only one object corresponding to this IO.
2506 static void end_bio_extent_readpage(struct bio *bio)
2508 struct bio_vec *bvec;
2509 int uptodate = !bio->bi_error;
2510 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
2511 struct extent_io_tree *tree, *failure_tree;
2516 u64 extent_start = 0;
2522 ASSERT(!bio_flagged(bio, BIO_CLONED));
2523 bio_for_each_segment_all(bvec, bio, i) {
2524 struct page *page = bvec->bv_page;
2525 struct inode *inode = page->mapping->host;
2526 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2528 btrfs_debug(fs_info,
2529 "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u",
2530 (u64)bio->bi_iter.bi_sector, bio->bi_error,
2531 io_bio->mirror_num);
2532 tree = &BTRFS_I(inode)->io_tree;
2533 failure_tree = &BTRFS_I(inode)->io_failure_tree;
2535 /* We always issue full-page reads, but if some block
2536 * in a page fails to read, blk_update_request() will
2537 * advance bv_offset and adjust bv_len to compensate.
2538 * Print a warning for nonzero offsets, and an error
2539 * if they don't add up to a full page. */
2540 if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) {
2541 if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE)
2543 "partial page read in btrfs with offset %u and length %u",
2544 bvec->bv_offset, bvec->bv_len);
2547 "incomplete page read in btrfs with offset %u and length %u",
2548 bvec->bv_offset, bvec->bv_len);
2551 start = page_offset(page);
2552 end = start + bvec->bv_offset + bvec->bv_len - 1;
2555 mirror = io_bio->mirror_num;
2556 if (likely(uptodate && tree->ops)) {
2557 ret = tree->ops->readpage_end_io_hook(io_bio, offset,
2563 clean_io_failure(BTRFS_I(inode)->root->fs_info,
2564 failure_tree, tree, start,
2566 btrfs_ino(BTRFS_I(inode)), 0);
2569 if (likely(uptodate))
2573 ret = tree->ops->readpage_io_failed_hook(page, mirror);
2574 if (ret == -EAGAIN) {
2576 * Data inode's readpage_io_failed_hook() always
2579 * The generic bio_readpage_error handles errors
2580 * the following way: If possible, new read
2581 * requests are created and submitted and will
2582 * end up in end_bio_extent_readpage as well (if
2583 * we're lucky, not in the !uptodate case). In
2584 * that case it returns 0 and we just go on with
2585 * the next page in our bio. If it can't handle
2586 * the error it will return -EIO and we remain
2587 * responsible for that page.
2589 ret = bio_readpage_error(bio, offset, page,
2590 start, end, mirror);
2592 uptodate = !bio->bi_error;
2599 * metadata's readpage_io_failed_hook() always returns
2600 * -EIO and fixes nothing. -EIO is also returned if
2601 * data inode error could not be fixed.
2603 ASSERT(ret == -EIO);
2606 if (likely(uptodate)) {
2607 loff_t i_size = i_size_read(inode);
2608 pgoff_t end_index = i_size >> PAGE_SHIFT;
2611 /* Zero out the end if this page straddles i_size */
2612 off = i_size & (PAGE_SIZE-1);
2613 if (page->index == end_index && off)
2614 zero_user_segment(page, off, PAGE_SIZE);
2615 SetPageUptodate(page);
2617 ClearPageUptodate(page);
2623 if (unlikely(!uptodate)) {
2625 endio_readpage_release_extent(tree,
2631 endio_readpage_release_extent(tree, start,
2632 end - start + 1, 0);
2633 } else if (!extent_len) {
2634 extent_start = start;
2635 extent_len = end + 1 - start;
2636 } else if (extent_start + extent_len == start) {
2637 extent_len += end + 1 - start;
2639 endio_readpage_release_extent(tree, extent_start,
2640 extent_len, uptodate);
2641 extent_start = start;
2642 extent_len = end + 1 - start;
2647 endio_readpage_release_extent(tree, extent_start, extent_len,
2650 io_bio->end_io(io_bio, bio->bi_error);
2655 * Initialize the members up to but not including 'bio'. Use after allocating a
2656 * new bio by bio_alloc_bioset as it does not initialize the bytes outside of
2657 * 'bio' because use of __GFP_ZERO is not supported.
2659 static inline void btrfs_io_bio_init(struct btrfs_io_bio *btrfs_bio)
2661 memset(btrfs_bio, 0, offsetof(struct btrfs_io_bio, bio));
2665 * The following helpers allocate a bio. As it's backed by a bioset, it'll
2666 * never fail. We're returning a bio right now but you can call btrfs_io_bio
2667 * for the appropriate container_of magic
2669 struct bio *btrfs_bio_alloc(struct block_device *bdev, u64 first_byte)
2673 bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, btrfs_bioset);
2674 bio->bi_bdev = bdev;
2675 bio->bi_iter.bi_sector = first_byte >> 9;
2676 btrfs_io_bio_init(btrfs_io_bio(bio));
2680 struct bio *btrfs_bio_clone(struct bio *bio)
2682 struct btrfs_io_bio *btrfs_bio;
2685 /* Bio allocation backed by a bioset does not fail */
2686 new = bio_clone_fast(bio, GFP_NOFS, btrfs_bioset);
2687 btrfs_bio = btrfs_io_bio(new);
2688 btrfs_io_bio_init(btrfs_bio);
2689 btrfs_bio->iter = bio->bi_iter;
2693 struct bio *btrfs_io_bio_alloc(unsigned int nr_iovecs)
2697 /* Bio allocation backed by a bioset does not fail */
2698 bio = bio_alloc_bioset(GFP_NOFS, nr_iovecs, btrfs_bioset);
2699 btrfs_io_bio_init(btrfs_io_bio(bio));
2703 struct bio *btrfs_bio_clone_partial(struct bio *orig, int offset, int size)
2706 struct btrfs_io_bio *btrfs_bio;
2708 /* this will never fail when it's backed by a bioset */
2709 bio = bio_clone_fast(orig, GFP_NOFS, btrfs_bioset);
2712 btrfs_bio = btrfs_io_bio(bio);
2713 btrfs_io_bio_init(btrfs_bio);
2715 bio_trim(bio, offset >> 9, size >> 9);
2716 btrfs_bio->iter = bio->bi_iter;
2720 static int __must_check submit_one_bio(struct bio *bio, int mirror_num,
2721 unsigned long bio_flags)
2724 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2725 struct page *page = bvec->bv_page;
2726 struct extent_io_tree *tree = bio->bi_private;
2729 start = page_offset(page) + bvec->bv_offset;
2731 bio->bi_private = NULL;
2735 ret = tree->ops->submit_bio_hook(tree->private_data, bio,
2736 mirror_num, bio_flags, start);
2738 btrfsic_submit_bio(bio);
2744 static int merge_bio(struct extent_io_tree *tree, struct page *page,
2745 unsigned long offset, size_t size, struct bio *bio,
2746 unsigned long bio_flags)
2750 ret = tree->ops->merge_bio_hook(page, offset, size, bio,
2756 static int submit_extent_page(int op, int op_flags, struct extent_io_tree *tree,
2757 struct writeback_control *wbc,
2758 struct page *page, sector_t sector,
2759 size_t size, unsigned long offset,
2760 struct block_device *bdev,
2761 struct bio **bio_ret,
2762 bio_end_io_t end_io_func,
2764 unsigned long prev_bio_flags,
2765 unsigned long bio_flags,
2766 bool force_bio_submit)
2771 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
2772 size_t page_size = min_t(size_t, size, PAGE_SIZE);
2774 if (bio_ret && *bio_ret) {
2777 contig = bio->bi_iter.bi_sector == sector;
2779 contig = bio_end_sector(bio) == sector;
2781 if (prev_bio_flags != bio_flags || !contig ||
2783 merge_bio(tree, page, offset, page_size, bio, bio_flags) ||
2784 bio_add_page(bio, page, page_size, offset) < page_size) {
2785 ret = submit_one_bio(bio, mirror_num, prev_bio_flags);
2793 wbc_account_io(wbc, page, page_size);
2798 bio = btrfs_bio_alloc(bdev, sector << 9);
2799 bio_add_page(bio, page, page_size, offset);
2800 bio->bi_end_io = end_io_func;
2801 bio->bi_private = tree;
2802 bio_set_op_attrs(bio, op, op_flags);
2804 wbc_init_bio(wbc, bio);
2805 wbc_account_io(wbc, page, page_size);
2811 ret = submit_one_bio(bio, mirror_num, bio_flags);
2816 static void attach_extent_buffer_page(struct extent_buffer *eb,
2819 if (!PagePrivate(page)) {
2820 SetPagePrivate(page);
2822 set_page_private(page, (unsigned long)eb);
2824 WARN_ON(page->private != (unsigned long)eb);
2828 void set_page_extent_mapped(struct page *page)
2830 if (!PagePrivate(page)) {
2831 SetPagePrivate(page);
2833 set_page_private(page, EXTENT_PAGE_PRIVATE);
2837 static struct extent_map *
2838 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
2839 u64 start, u64 len, get_extent_t *get_extent,
2840 struct extent_map **em_cached)
2842 struct extent_map *em;
2844 if (em_cached && *em_cached) {
2846 if (extent_map_in_tree(em) && start >= em->start &&
2847 start < extent_map_end(em)) {
2848 refcount_inc(&em->refs);
2852 free_extent_map(em);
2856 em = get_extent(BTRFS_I(inode), page, pg_offset, start, len, 0);
2857 if (em_cached && !IS_ERR_OR_NULL(em)) {
2859 refcount_inc(&em->refs);
2865 * basic readpage implementation. Locked extent state structs are inserted
2866 * into the tree that are removed when the IO is done (by the end_io
2868 * XXX JDM: This needs looking at to ensure proper page locking
2869 * return 0 on success, otherwise return error
2871 static int __do_readpage(struct extent_io_tree *tree,
2873 get_extent_t *get_extent,
2874 struct extent_map **em_cached,
2875 struct bio **bio, int mirror_num,
2876 unsigned long *bio_flags, int read_flags,
2879 struct inode *inode = page->mapping->host;
2880 u64 start = page_offset(page);
2881 u64 page_end = start + PAGE_SIZE - 1;
2885 u64 last_byte = i_size_read(inode);
2889 struct extent_map *em;
2890 struct block_device *bdev;
2893 size_t pg_offset = 0;
2895 size_t disk_io_size;
2896 size_t blocksize = inode->i_sb->s_blocksize;
2897 unsigned long this_bio_flag = 0;
2899 set_page_extent_mapped(page);
2902 if (!PageUptodate(page)) {
2903 if (cleancache_get_page(page) == 0) {
2904 BUG_ON(blocksize != PAGE_SIZE);
2905 unlock_extent(tree, start, end);
2910 if (page->index == last_byte >> PAGE_SHIFT) {
2912 size_t zero_offset = last_byte & (PAGE_SIZE - 1);
2915 iosize = PAGE_SIZE - zero_offset;
2916 userpage = kmap_atomic(page);
2917 memset(userpage + zero_offset, 0, iosize);
2918 flush_dcache_page(page);
2919 kunmap_atomic(userpage);
2922 while (cur <= end) {
2923 bool force_bio_submit = false;
2925 if (cur >= last_byte) {
2927 struct extent_state *cached = NULL;
2929 iosize = PAGE_SIZE - pg_offset;
2930 userpage = kmap_atomic(page);
2931 memset(userpage + pg_offset, 0, iosize);
2932 flush_dcache_page(page);
2933 kunmap_atomic(userpage);
2934 set_extent_uptodate(tree, cur, cur + iosize - 1,
2936 unlock_extent_cached(tree, cur,
2941 em = __get_extent_map(inode, page, pg_offset, cur,
2942 end - cur + 1, get_extent, em_cached);
2943 if (IS_ERR_OR_NULL(em)) {
2945 unlock_extent(tree, cur, end);
2948 extent_offset = cur - em->start;
2949 BUG_ON(extent_map_end(em) <= cur);
2952 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2953 this_bio_flag |= EXTENT_BIO_COMPRESSED;
2954 extent_set_compress_type(&this_bio_flag,
2958 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2959 cur_end = min(extent_map_end(em) - 1, end);
2960 iosize = ALIGN(iosize, blocksize);
2961 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2962 disk_io_size = em->block_len;
2963 sector = em->block_start >> 9;
2965 sector = (em->block_start + extent_offset) >> 9;
2966 disk_io_size = iosize;
2969 block_start = em->block_start;
2970 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2971 block_start = EXTENT_MAP_HOLE;
2974 * If we have a file range that points to a compressed extent
2975 * and it's followed by a consecutive file range that points to
2976 * to the same compressed extent (possibly with a different
2977 * offset and/or length, so it either points to the whole extent
2978 * or only part of it), we must make sure we do not submit a
2979 * single bio to populate the pages for the 2 ranges because
2980 * this makes the compressed extent read zero out the pages
2981 * belonging to the 2nd range. Imagine the following scenario:
2984 * [0 - 8K] [8K - 24K]
2987 * points to extent X, points to extent X,
2988 * offset 4K, length of 8K offset 0, length 16K
2990 * [extent X, compressed length = 4K uncompressed length = 16K]
2992 * If the bio to read the compressed extent covers both ranges,
2993 * it will decompress extent X into the pages belonging to the
2994 * first range and then it will stop, zeroing out the remaining
2995 * pages that belong to the other range that points to extent X.
2996 * So here we make sure we submit 2 bios, one for the first
2997 * range and another one for the third range. Both will target
2998 * the same physical extent from disk, but we can't currently
2999 * make the compressed bio endio callback populate the pages
3000 * for both ranges because each compressed bio is tightly
3001 * coupled with a single extent map, and each range can have
3002 * an extent map with a different offset value relative to the
3003 * uncompressed data of our extent and different lengths. This
3004 * is a corner case so we prioritize correctness over
3005 * non-optimal behavior (submitting 2 bios for the same extent).
3007 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) &&
3008 prev_em_start && *prev_em_start != (u64)-1 &&
3009 *prev_em_start != em->orig_start)
3010 force_bio_submit = true;
3013 *prev_em_start = em->orig_start;
3015 free_extent_map(em);
3018 /* we've found a hole, just zero and go on */
3019 if (block_start == EXTENT_MAP_HOLE) {
3021 struct extent_state *cached = NULL;
3023 userpage = kmap_atomic(page);
3024 memset(userpage + pg_offset, 0, iosize);
3025 flush_dcache_page(page);
3026 kunmap_atomic(userpage);
3028 set_extent_uptodate(tree, cur, cur + iosize - 1,
3030 unlock_extent_cached(tree, cur,
3034 pg_offset += iosize;
3037 /* the get_extent function already copied into the page */
3038 if (test_range_bit(tree, cur, cur_end,
3039 EXTENT_UPTODATE, 1, NULL)) {
3040 check_page_uptodate(tree, page);
3041 unlock_extent(tree, cur, cur + iosize - 1);
3043 pg_offset += iosize;
3046 /* we have an inline extent but it didn't get marked up
3047 * to date. Error out
3049 if (block_start == EXTENT_MAP_INLINE) {
3051 unlock_extent(tree, cur, cur + iosize - 1);
3053 pg_offset += iosize;
3057 ret = submit_extent_page(REQ_OP_READ, read_flags, tree, NULL,
3058 page, sector, disk_io_size, pg_offset,
3060 end_bio_extent_readpage, mirror_num,
3066 *bio_flags = this_bio_flag;
3069 unlock_extent(tree, cur, cur + iosize - 1);
3073 pg_offset += iosize;
3077 if (!PageError(page))
3078 SetPageUptodate(page);
3084 static inline void __do_contiguous_readpages(struct extent_io_tree *tree,
3085 struct page *pages[], int nr_pages,
3087 get_extent_t *get_extent,
3088 struct extent_map **em_cached,
3089 struct bio **bio, int mirror_num,
3090 unsigned long *bio_flags,
3093 struct inode *inode;
3094 struct btrfs_ordered_extent *ordered;
3097 inode = pages[0]->mapping->host;
3099 lock_extent(tree, start, end);
3100 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
3104 unlock_extent(tree, start, end);
3105 btrfs_start_ordered_extent(inode, ordered, 1);
3106 btrfs_put_ordered_extent(ordered);
3109 for (index = 0; index < nr_pages; index++) {
3110 __do_readpage(tree, pages[index], get_extent, em_cached, bio,
3111 mirror_num, bio_flags, 0, prev_em_start);
3112 put_page(pages[index]);
3116 static void __extent_readpages(struct extent_io_tree *tree,
3117 struct page *pages[],
3118 int nr_pages, get_extent_t *get_extent,
3119 struct extent_map **em_cached,
3120 struct bio **bio, int mirror_num,
3121 unsigned long *bio_flags,
3128 int first_index = 0;
3130 for (index = 0; index < nr_pages; index++) {
3131 page_start = page_offset(pages[index]);
3134 end = start + PAGE_SIZE - 1;
3135 first_index = index;
3136 } else if (end + 1 == page_start) {
3139 __do_contiguous_readpages(tree, &pages[first_index],
3140 index - first_index, start,
3141 end, get_extent, em_cached,
3142 bio, mirror_num, bio_flags,
3145 end = start + PAGE_SIZE - 1;
3146 first_index = index;
3151 __do_contiguous_readpages(tree, &pages[first_index],
3152 index - first_index, start,
3153 end, get_extent, em_cached, bio,
3154 mirror_num, bio_flags,
3158 static int __extent_read_full_page(struct extent_io_tree *tree,
3160 get_extent_t *get_extent,
3161 struct bio **bio, int mirror_num,
3162 unsigned long *bio_flags, int read_flags)
3164 struct inode *inode = page->mapping->host;
3165 struct btrfs_ordered_extent *ordered;
3166 u64 start = page_offset(page);
3167 u64 end = start + PAGE_SIZE - 1;
3171 lock_extent(tree, start, end);
3172 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
3176 unlock_extent(tree, start, end);
3177 btrfs_start_ordered_extent(inode, ordered, 1);
3178 btrfs_put_ordered_extent(ordered);
3181 ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num,
3182 bio_flags, read_flags, NULL);
3186 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
3187 get_extent_t *get_extent, int mirror_num)
3189 struct bio *bio = NULL;
3190 unsigned long bio_flags = 0;
3193 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
3196 ret = submit_one_bio(bio, mirror_num, bio_flags);
3200 static void update_nr_written(struct writeback_control *wbc,
3201 unsigned long nr_written)
3203 wbc->nr_to_write -= nr_written;
3207 * helper for __extent_writepage, doing all of the delayed allocation setup.
3209 * This returns 1 if our fill_delalloc function did all the work required
3210 * to write the page (copy into inline extent). In this case the IO has
3211 * been started and the page is already unlocked.
3213 * This returns 0 if all went well (page still locked)
3214 * This returns < 0 if there were errors (page still locked)
3216 static noinline_for_stack int writepage_delalloc(struct inode *inode,
3217 struct page *page, struct writeback_control *wbc,
3218 struct extent_page_data *epd,
3220 unsigned long *nr_written)
3222 struct extent_io_tree *tree = epd->tree;
3223 u64 page_end = delalloc_start + PAGE_SIZE - 1;
3225 u64 delalloc_to_write = 0;
3226 u64 delalloc_end = 0;
3228 int page_started = 0;
3230 if (epd->extent_locked || !tree->ops || !tree->ops->fill_delalloc)
3233 while (delalloc_end < page_end) {
3234 nr_delalloc = find_lock_delalloc_range(inode, tree,
3238 BTRFS_MAX_EXTENT_SIZE);
3239 if (nr_delalloc == 0) {
3240 delalloc_start = delalloc_end + 1;
3243 ret = tree->ops->fill_delalloc(inode, page,
3248 /* File system has been set read-only */
3251 /* fill_delalloc should be return < 0 for error
3252 * but just in case, we use > 0 here meaning the
3253 * IO is started, so we don't want to return > 0
3254 * unless things are going well.
3256 ret = ret < 0 ? ret : -EIO;
3260 * delalloc_end is already one less than the total length, so
3261 * we don't subtract one from PAGE_SIZE
3263 delalloc_to_write += (delalloc_end - delalloc_start +
3264 PAGE_SIZE) >> PAGE_SHIFT;
3265 delalloc_start = delalloc_end + 1;
3267 if (wbc->nr_to_write < delalloc_to_write) {
3270 if (delalloc_to_write < thresh * 2)
3271 thresh = delalloc_to_write;
3272 wbc->nr_to_write = min_t(u64, delalloc_to_write,
3276 /* did the fill delalloc function already unlock and start
3281 * we've unlocked the page, so we can't update
3282 * the mapping's writeback index, just update
3285 wbc->nr_to_write -= *nr_written;
3296 * helper for __extent_writepage. This calls the writepage start hooks,
3297 * and does the loop to map the page into extents and bios.
3299 * We return 1 if the IO is started and the page is unlocked,
3300 * 0 if all went well (page still locked)
3301 * < 0 if there were errors (page still locked)
3303 static noinline_for_stack int __extent_writepage_io(struct inode *inode,
3305 struct writeback_control *wbc,
3306 struct extent_page_data *epd,
3308 unsigned long nr_written,
3309 int write_flags, int *nr_ret)
3311 struct extent_io_tree *tree = epd->tree;
3312 u64 start = page_offset(page);
3313 u64 page_end = start + PAGE_SIZE - 1;
3320 struct extent_map *em;
3321 struct block_device *bdev;
3322 size_t pg_offset = 0;
3328 if (tree->ops && tree->ops->writepage_start_hook) {
3329 ret = tree->ops->writepage_start_hook(page, start,
3332 /* Fixup worker will requeue */
3334 wbc->pages_skipped++;
3336 redirty_page_for_writepage(wbc, page);
3338 update_nr_written(wbc, nr_written);
3345 * we don't want to touch the inode after unlocking the page,
3346 * so we update the mapping writeback index now
3348 update_nr_written(wbc, nr_written + 1);
3351 if (i_size <= start) {
3352 if (tree->ops && tree->ops->writepage_end_io_hook)
3353 tree->ops->writepage_end_io_hook(page, start,
3358 blocksize = inode->i_sb->s_blocksize;
3360 while (cur <= end) {
3363 if (cur >= i_size) {
3364 if (tree->ops && tree->ops->writepage_end_io_hook)
3365 tree->ops->writepage_end_io_hook(page, cur,
3369 em = epd->get_extent(BTRFS_I(inode), page, pg_offset, cur,
3371 if (IS_ERR_OR_NULL(em)) {
3373 ret = PTR_ERR_OR_ZERO(em);
3377 extent_offset = cur - em->start;
3378 em_end = extent_map_end(em);
3379 BUG_ON(em_end <= cur);
3381 iosize = min(em_end - cur, end - cur + 1);
3382 iosize = ALIGN(iosize, blocksize);
3383 sector = (em->block_start + extent_offset) >> 9;
3385 block_start = em->block_start;
3386 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3387 free_extent_map(em);
3391 * compressed and inline extents are written through other
3394 if (compressed || block_start == EXTENT_MAP_HOLE ||
3395 block_start == EXTENT_MAP_INLINE) {
3397 * end_io notification does not happen here for
3398 * compressed extents
3400 if (!compressed && tree->ops &&
3401 tree->ops->writepage_end_io_hook)
3402 tree->ops->writepage_end_io_hook(page, cur,
3405 else if (compressed) {
3406 /* we don't want to end_page_writeback on
3407 * a compressed extent. this happens
3414 pg_offset += iosize;
3418 set_range_writeback(tree, cur, cur + iosize - 1);
3419 if (!PageWriteback(page)) {
3420 btrfs_err(BTRFS_I(inode)->root->fs_info,
3421 "page %lu not writeback, cur %llu end %llu",
3422 page->index, cur, end);
3425 ret = submit_extent_page(REQ_OP_WRITE, write_flags, tree, wbc,
3426 page, sector, iosize, pg_offset,
3428 end_bio_extent_writepage,
3432 if (PageWriteback(page))
3433 end_page_writeback(page);
3437 pg_offset += iosize;
3446 * the writepage semantics are similar to regular writepage. extent
3447 * records are inserted to lock ranges in the tree, and as dirty areas
3448 * are found, they are marked writeback. Then the lock bits are removed
3449 * and the end_io handler clears the writeback ranges
3451 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
3454 struct inode *inode = page->mapping->host;
3455 struct extent_page_data *epd = data;
3456 u64 start = page_offset(page);
3457 u64 page_end = start + PAGE_SIZE - 1;
3460 size_t pg_offset = 0;
3461 loff_t i_size = i_size_read(inode);
3462 unsigned long end_index = i_size >> PAGE_SHIFT;
3463 int write_flags = 0;
3464 unsigned long nr_written = 0;
3466 if (wbc->sync_mode == WB_SYNC_ALL)
3467 write_flags = REQ_SYNC;
3469 trace___extent_writepage(page, inode, wbc);
3471 WARN_ON(!PageLocked(page));
3473 ClearPageError(page);
3475 pg_offset = i_size & (PAGE_SIZE - 1);
3476 if (page->index > end_index ||
3477 (page->index == end_index && !pg_offset)) {
3478 page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
3483 if (page->index == end_index) {
3486 userpage = kmap_atomic(page);
3487 memset(userpage + pg_offset, 0,
3488 PAGE_SIZE - pg_offset);
3489 kunmap_atomic(userpage);
3490 flush_dcache_page(page);
3495 set_page_extent_mapped(page);
3497 ret = writepage_delalloc(inode, page, wbc, epd, start, &nr_written);
3503 ret = __extent_writepage_io(inode, page, wbc, epd,
3504 i_size, nr_written, write_flags, &nr);
3510 /* make sure the mapping tag for page dirty gets cleared */
3511 set_page_writeback(page);
3512 end_page_writeback(page);
3514 if (PageError(page)) {
3515 ret = ret < 0 ? ret : -EIO;
3516 end_extent_writepage(page, ret, start, page_end);
3525 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3527 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
3528 TASK_UNINTERRUPTIBLE);
3531 static noinline_for_stack int
3532 lock_extent_buffer_for_io(struct extent_buffer *eb,
3533 struct btrfs_fs_info *fs_info,
3534 struct extent_page_data *epd)
3536 unsigned long i, num_pages;
3540 if (!btrfs_try_tree_write_lock(eb)) {
3542 flush_write_bio(epd);
3543 btrfs_tree_lock(eb);
3546 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3547 btrfs_tree_unlock(eb);
3551 flush_write_bio(epd);
3555 wait_on_extent_buffer_writeback(eb);
3556 btrfs_tree_lock(eb);
3557 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3559 btrfs_tree_unlock(eb);
3564 * We need to do this to prevent races in people who check if the eb is
3565 * under IO since we can end up having no IO bits set for a short period
3568 spin_lock(&eb->refs_lock);
3569 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3570 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3571 spin_unlock(&eb->refs_lock);
3572 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3573 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
3575 fs_info->dirty_metadata_batch);
3578 spin_unlock(&eb->refs_lock);
3581 btrfs_tree_unlock(eb);
3586 num_pages = num_extent_pages(eb->start, eb->len);
3587 for (i = 0; i < num_pages; i++) {
3588 struct page *p = eb->pages[i];
3590 if (!trylock_page(p)) {
3592 flush_write_bio(epd);
3602 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3604 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3605 smp_mb__after_atomic();
3606 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3609 static void set_btree_ioerr(struct page *page)
3611 struct extent_buffer *eb = (struct extent_buffer *)page->private;
3614 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))
3618 * If writeback for a btree extent that doesn't belong to a log tree
3619 * failed, increment the counter transaction->eb_write_errors.
3620 * We do this because while the transaction is running and before it's
3621 * committing (when we call filemap_fdata[write|wait]_range against
3622 * the btree inode), we might have
3623 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
3624 * returns an error or an error happens during writeback, when we're
3625 * committing the transaction we wouldn't know about it, since the pages
3626 * can be no longer dirty nor marked anymore for writeback (if a
3627 * subsequent modification to the extent buffer didn't happen before the
3628 * transaction commit), which makes filemap_fdata[write|wait]_range not
3629 * able to find the pages tagged with SetPageError at transaction
3630 * commit time. So if this happens we must abort the transaction,
3631 * otherwise we commit a super block with btree roots that point to
3632 * btree nodes/leafs whose content on disk is invalid - either garbage
3633 * or the content of some node/leaf from a past generation that got
3634 * cowed or deleted and is no longer valid.
3636 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
3637 * not be enough - we need to distinguish between log tree extents vs
3638 * non-log tree extents, and the next filemap_fdatawait_range() call
3639 * will catch and clear such errors in the mapping - and that call might
3640 * be from a log sync and not from a transaction commit. Also, checking
3641 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
3642 * not done and would not be reliable - the eb might have been released
3643 * from memory and reading it back again means that flag would not be
3644 * set (since it's a runtime flag, not persisted on disk).
3646 * Using the flags below in the btree inode also makes us achieve the
3647 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
3648 * writeback for all dirty pages and before filemap_fdatawait_range()
3649 * is called, the writeback for all dirty pages had already finished
3650 * with errors - because we were not using AS_EIO/AS_ENOSPC,
3651 * filemap_fdatawait_range() would return success, as it could not know
3652 * that writeback errors happened (the pages were no longer tagged for
3655 switch (eb->log_index) {
3657 set_bit(BTRFS_FS_BTREE_ERR, &eb->fs_info->flags);
3660 set_bit(BTRFS_FS_LOG1_ERR, &eb->fs_info->flags);
3663 set_bit(BTRFS_FS_LOG2_ERR, &eb->fs_info->flags);
3666 BUG(); /* unexpected, logic error */
3670 static void end_bio_extent_buffer_writepage(struct bio *bio)
3672 struct bio_vec *bvec;
3673 struct extent_buffer *eb;
3676 ASSERT(!bio_flagged(bio, BIO_CLONED));
3677 bio_for_each_segment_all(bvec, bio, i) {
3678 struct page *page = bvec->bv_page;
3680 eb = (struct extent_buffer *)page->private;
3682 done = atomic_dec_and_test(&eb->io_pages);
3684 if (bio->bi_error ||
3685 test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) {
3686 ClearPageUptodate(page);
3687 set_btree_ioerr(page);
3690 end_page_writeback(page);
3695 end_extent_buffer_writeback(eb);
3701 static noinline_for_stack int write_one_eb(struct extent_buffer *eb,
3702 struct btrfs_fs_info *fs_info,
3703 struct writeback_control *wbc,
3704 struct extent_page_data *epd)
3706 struct block_device *bdev = fs_info->fs_devices->latest_bdev;
3707 struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
3708 u64 offset = eb->start;
3710 unsigned long i, num_pages;
3711 unsigned long bio_flags = 0;
3712 unsigned long start, end;
3713 int write_flags = (epd->sync_io ? REQ_SYNC : 0) | REQ_META;
3716 clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
3717 num_pages = num_extent_pages(eb->start, eb->len);
3718 atomic_set(&eb->io_pages, num_pages);
3719 if (btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID)
3720 bio_flags = EXTENT_BIO_TREE_LOG;
3722 /* set btree blocks beyond nritems with 0 to avoid stale content. */
3723 nritems = btrfs_header_nritems(eb);
3724 if (btrfs_header_level(eb) > 0) {
3725 end = btrfs_node_key_ptr_offset(nritems);
3727 memzero_extent_buffer(eb, end, eb->len - end);
3731 * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0
3733 start = btrfs_item_nr_offset(nritems);
3734 end = BTRFS_LEAF_DATA_OFFSET + leaf_data_end(fs_info, eb);
3735 memzero_extent_buffer(eb, start, end - start);
3738 for (i = 0; i < num_pages; i++) {
3739 struct page *p = eb->pages[i];
3741 clear_page_dirty_for_io(p);
3742 set_page_writeback(p);
3743 ret = submit_extent_page(REQ_OP_WRITE, write_flags, tree, wbc,
3744 p, offset >> 9, PAGE_SIZE, 0, bdev,
3746 end_bio_extent_buffer_writepage,
3747 0, epd->bio_flags, bio_flags, false);
3748 epd->bio_flags = bio_flags;
3751 if (PageWriteback(p))
3752 end_page_writeback(p);
3753 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3754 end_extent_buffer_writeback(eb);
3758 offset += PAGE_SIZE;
3759 update_nr_written(wbc, 1);
3763 if (unlikely(ret)) {
3764 for (; i < num_pages; i++) {
3765 struct page *p = eb->pages[i];
3766 clear_page_dirty_for_io(p);
3774 int btree_write_cache_pages(struct address_space *mapping,
3775 struct writeback_control *wbc)
3777 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
3778 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
3779 struct extent_buffer *eb, *prev_eb = NULL;
3780 struct extent_page_data epd = {
3784 .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, 0);
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_tag(&pvec, mapping, &index, tag,
3815 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3819 for (i = 0; i < nr_pages; i++) {
3820 struct page *page = pvec.pages[i];
3822 if (!PagePrivate(page))
3825 if (!wbc->range_cyclic && page->index > end) {
3830 spin_lock(&mapping->private_lock);
3831 if (!PagePrivate(page)) {
3832 spin_unlock(&mapping->private_lock);
3836 eb = (struct extent_buffer *)page->private;
3839 * Shouldn't happen and normally this would be a BUG_ON
3840 * but no sense in crashing the users box for something
3841 * we can survive anyway.
3844 spin_unlock(&mapping->private_lock);
3848 if (eb == prev_eb) {
3849 spin_unlock(&mapping->private_lock);
3853 ret = atomic_inc_not_zero(&eb->refs);
3854 spin_unlock(&mapping->private_lock);
3859 ret = lock_extent_buffer_for_io(eb, fs_info, &epd);
3861 free_extent_buffer(eb);
3865 ret = write_one_eb(eb, fs_info, wbc, &epd);
3868 free_extent_buffer(eb);
3871 free_extent_buffer(eb);
3874 * the filesystem may choose to bump up nr_to_write.
3875 * We have to make sure to honor the new nr_to_write
3878 nr_to_write_done = wbc->nr_to_write <= 0;
3880 pagevec_release(&pvec);
3883 if (!scanned && !done) {
3885 * We hit the last page and there is more work to be done: wrap
3886 * back to the start of the file
3892 flush_write_bio(&epd);
3897 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3898 * @mapping: address space structure to write
3899 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3900 * @writepage: function called for each page
3901 * @data: data passed to writepage function
3903 * If a page is already under I/O, write_cache_pages() skips it, even
3904 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3905 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3906 * and msync() need to guarantee that all the data which was dirty at the time
3907 * the call was made get new I/O started against them. If wbc->sync_mode is
3908 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3909 * existing IO to complete.
3911 static int extent_write_cache_pages(struct address_space *mapping,
3912 struct writeback_control *wbc,
3913 writepage_t writepage, void *data,
3914 void (*flush_fn)(void *))
3916 struct inode *inode = mapping->host;
3919 int nr_to_write_done = 0;
3920 struct pagevec pvec;
3923 pgoff_t end; /* Inclusive */
3925 int range_whole = 0;
3930 * We have to hold onto the inode so that ordered extents can do their
3931 * work when the IO finishes. The alternative to this is failing to add
3932 * an ordered extent if the igrab() fails there and that is a huge pain
3933 * to deal with, so instead just hold onto the inode throughout the
3934 * writepages operation. If it fails here we are freeing up the inode
3935 * anyway and we'd rather not waste our time writing out stuff that is
3936 * going to be truncated anyway.
3941 pagevec_init(&pvec, 0);
3942 if (wbc->range_cyclic) {
3943 index = mapping->writeback_index; /* Start from prev offset */
3946 index = wbc->range_start >> PAGE_SHIFT;
3947 end = wbc->range_end >> PAGE_SHIFT;
3948 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
3952 if (wbc->sync_mode == WB_SYNC_ALL)
3953 tag = PAGECACHE_TAG_TOWRITE;
3955 tag = PAGECACHE_TAG_DIRTY;
3957 if (wbc->sync_mode == WB_SYNC_ALL)
3958 tag_pages_for_writeback(mapping, index, end);
3960 while (!done && !nr_to_write_done && (index <= end) &&
3961 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3962 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3966 for (i = 0; i < nr_pages; i++) {
3967 struct page *page = pvec.pages[i];
3969 done_index = page->index;
3971 * At this point we hold neither mapping->tree_lock nor
3972 * lock on the page itself: the page may be truncated or
3973 * invalidated (changing page->mapping to NULL), or even
3974 * swizzled back from swapper_space to tmpfs file
3977 if (!trylock_page(page)) {
3982 if (unlikely(page->mapping != mapping)) {
3987 if (!wbc->range_cyclic && page->index > end) {
3993 if (wbc->sync_mode != WB_SYNC_NONE) {
3994 if (PageWriteback(page))
3996 wait_on_page_writeback(page);
3999 if (PageWriteback(page) ||
4000 !clear_page_dirty_for_io(page)) {
4005 ret = (*writepage)(page, wbc, data);
4007 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
4013 * done_index is set past this page,
4014 * so media errors will not choke
4015 * background writeout for the entire
4016 * file. This has consequences for
4017 * range_cyclic semantics (ie. it may
4018 * not be suitable for data integrity
4021 done_index = page->index + 1;
4027 * the filesystem may choose to bump up nr_to_write.
4028 * We have to make sure to honor the new nr_to_write
4031 nr_to_write_done = wbc->nr_to_write <= 0;
4033 pagevec_release(&pvec);
4036 if (!scanned && !done) {
4038 * We hit the last page and there is more work to be done: wrap
4039 * back to the start of the file
4046 if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole))
4047 mapping->writeback_index = done_index;
4049 btrfs_add_delayed_iput(inode);
4053 static void flush_epd_write_bio(struct extent_page_data *epd)
4058 bio_set_op_attrs(epd->bio, REQ_OP_WRITE,
4059 epd->sync_io ? REQ_SYNC : 0);
4061 ret = submit_one_bio(epd->bio, 0, epd->bio_flags);
4062 BUG_ON(ret < 0); /* -ENOMEM */
4067 static noinline void flush_write_bio(void *data)
4069 struct extent_page_data *epd = data;
4070 flush_epd_write_bio(epd);
4073 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
4074 get_extent_t *get_extent,
4075 struct writeback_control *wbc)
4078 struct extent_page_data epd = {
4081 .get_extent = get_extent,
4083 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4087 ret = __extent_writepage(page, wbc, &epd);
4089 flush_epd_write_bio(&epd);
4093 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
4094 u64 start, u64 end, get_extent_t *get_extent,
4098 struct address_space *mapping = inode->i_mapping;
4100 unsigned long nr_pages = (end - start + PAGE_SIZE) >>
4103 struct extent_page_data epd = {
4106 .get_extent = get_extent,
4108 .sync_io = mode == WB_SYNC_ALL,
4111 struct writeback_control wbc_writepages = {
4113 .nr_to_write = nr_pages * 2,
4114 .range_start = start,
4115 .range_end = end + 1,
4118 while (start <= end) {
4119 page = find_get_page(mapping, start >> PAGE_SHIFT);
4120 if (clear_page_dirty_for_io(page))
4121 ret = __extent_writepage(page, &wbc_writepages, &epd);
4123 if (tree->ops && tree->ops->writepage_end_io_hook)
4124 tree->ops->writepage_end_io_hook(page, start,
4125 start + PAGE_SIZE - 1,
4133 flush_epd_write_bio(&epd);
4137 int extent_writepages(struct extent_io_tree *tree,
4138 struct address_space *mapping,
4139 get_extent_t *get_extent,
4140 struct writeback_control *wbc)
4143 struct extent_page_data epd = {
4146 .get_extent = get_extent,
4148 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4152 ret = extent_write_cache_pages(mapping, wbc, __extent_writepage, &epd,
4154 flush_epd_write_bio(&epd);
4158 int extent_readpages(struct extent_io_tree *tree,
4159 struct address_space *mapping,
4160 struct list_head *pages, unsigned nr_pages,
4161 get_extent_t get_extent)
4163 struct bio *bio = NULL;
4165 unsigned long bio_flags = 0;
4166 struct page *pagepool[16];
4168 struct extent_map *em_cached = NULL;
4170 u64 prev_em_start = (u64)-1;
4172 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
4173 page = list_entry(pages->prev, struct page, lru);
4175 prefetchw(&page->flags);
4176 list_del(&page->lru);
4177 if (add_to_page_cache_lru(page, mapping,
4179 readahead_gfp_mask(mapping))) {
4184 pagepool[nr++] = page;
4185 if (nr < ARRAY_SIZE(pagepool))
4187 __extent_readpages(tree, pagepool, nr, get_extent, &em_cached,
4188 &bio, 0, &bio_flags, &prev_em_start);
4192 __extent_readpages(tree, pagepool, nr, get_extent, &em_cached,
4193 &bio, 0, &bio_flags, &prev_em_start);
4196 free_extent_map(em_cached);
4198 BUG_ON(!list_empty(pages));
4200 return submit_one_bio(bio, 0, bio_flags);
4205 * basic invalidatepage code, this waits on any locked or writeback
4206 * ranges corresponding to the page, and then deletes any extent state
4207 * records from the tree
4209 int extent_invalidatepage(struct extent_io_tree *tree,
4210 struct page *page, unsigned long offset)
4212 struct extent_state *cached_state = NULL;
4213 u64 start = page_offset(page);
4214 u64 end = start + PAGE_SIZE - 1;
4215 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
4217 start += ALIGN(offset, blocksize);
4221 lock_extent_bits(tree, start, end, &cached_state);
4222 wait_on_page_writeback(page);
4223 clear_extent_bit(tree, start, end,
4224 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4225 EXTENT_DO_ACCOUNTING,
4226 1, 1, &cached_state, GFP_NOFS);
4231 * a helper for releasepage, this tests for areas of the page that
4232 * are locked or under IO and drops the related state bits if it is safe
4235 static int try_release_extent_state(struct extent_map_tree *map,
4236 struct extent_io_tree *tree,
4237 struct page *page, gfp_t mask)
4239 u64 start = page_offset(page);
4240 u64 end = start + PAGE_SIZE - 1;
4243 if (test_range_bit(tree, start, end,
4244 EXTENT_IOBITS, 0, NULL))
4248 * at this point we can safely clear everything except the
4249 * locked bit and the nodatasum bit
4251 ret = clear_extent_bit(tree, start, end,
4252 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
4255 /* if clear_extent_bit failed for enomem reasons,
4256 * we can't allow the release to continue.
4267 * a helper for releasepage. As long as there are no locked extents
4268 * in the range corresponding to the page, both state records and extent
4269 * map records are removed
4271 int try_release_extent_mapping(struct extent_map_tree *map,
4272 struct extent_io_tree *tree, struct page *page,
4275 struct extent_map *em;
4276 u64 start = page_offset(page);
4277 u64 end = start + PAGE_SIZE - 1;
4279 if (gfpflags_allow_blocking(mask) &&
4280 page->mapping->host->i_size > SZ_16M) {
4282 while (start <= end) {
4283 len = end - start + 1;
4284 write_lock(&map->lock);
4285 em = lookup_extent_mapping(map, start, len);
4287 write_unlock(&map->lock);
4290 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
4291 em->start != start) {
4292 write_unlock(&map->lock);
4293 free_extent_map(em);
4296 if (!test_range_bit(tree, em->start,
4297 extent_map_end(em) - 1,
4298 EXTENT_LOCKED | EXTENT_WRITEBACK,
4300 remove_extent_mapping(map, em);
4301 /* once for the rb tree */
4302 free_extent_map(em);
4304 start = extent_map_end(em);
4305 write_unlock(&map->lock);
4308 free_extent_map(em);
4311 return try_release_extent_state(map, tree, page, mask);
4315 * helper function for fiemap, which doesn't want to see any holes.
4316 * This maps until we find something past 'last'
4318 static struct extent_map *get_extent_skip_holes(struct inode *inode,
4321 get_extent_t *get_extent)
4323 u64 sectorsize = btrfs_inode_sectorsize(inode);
4324 struct extent_map *em;
4331 len = last - offset;
4334 len = ALIGN(len, sectorsize);
4335 em = get_extent(BTRFS_I(inode), NULL, 0, offset, len, 0);
4336 if (IS_ERR_OR_NULL(em))
4339 /* if this isn't a hole return it */
4340 if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) &&
4341 em->block_start != EXTENT_MAP_HOLE) {
4345 /* this is a hole, advance to the next extent */
4346 offset = extent_map_end(em);
4347 free_extent_map(em);
4355 * To cache previous fiemap extent
4357 * Will be used for merging fiemap extent
4359 struct fiemap_cache {
4368 * Helper to submit fiemap extent.
4370 * Will try to merge current fiemap extent specified by @offset, @phys,
4371 * @len and @flags with cached one.
4372 * And only when we fails to merge, cached one will be submitted as
4375 * Return value is the same as fiemap_fill_next_extent().
4377 static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
4378 struct fiemap_cache *cache,
4379 u64 offset, u64 phys, u64 len, u32 flags)
4387 * Sanity check, extent_fiemap() should have ensured that new
4388 * fiemap extent won't overlap with cahced one.
4391 * NOTE: Physical address can overlap, due to compression
4393 if (cache->offset + cache->len > offset) {
4399 * Only merges fiemap extents if
4400 * 1) Their logical addresses are continuous
4402 * 2) Their physical addresses are continuous
4403 * So truly compressed (physical size smaller than logical size)
4404 * extents won't get merged with each other
4406 * 3) Share same flags except FIEMAP_EXTENT_LAST
4407 * So regular extent won't get merged with prealloc extent
4409 if (cache->offset + cache->len == offset &&
4410 cache->phys + cache->len == phys &&
4411 (cache->flags & ~FIEMAP_EXTENT_LAST) ==
4412 (flags & ~FIEMAP_EXTENT_LAST)) {
4414 cache->flags |= flags;
4415 goto try_submit_last;
4418 /* Not mergeable, need to submit cached one */
4419 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
4420 cache->len, cache->flags);
4421 cache->cached = false;
4425 cache->cached = true;
4426 cache->offset = offset;
4429 cache->flags = flags;
4431 if (cache->flags & FIEMAP_EXTENT_LAST) {
4432 ret = fiemap_fill_next_extent(fieinfo, cache->offset,
4433 cache->phys, cache->len, cache->flags);
4434 cache->cached = false;
4440 * Emit last fiemap cache
4442 * The last fiemap cache may still be cached in the following case:
4444 * |<- Fiemap range ->|
4445 * |<------------ First extent ----------->|
4447 * In this case, the first extent range will be cached but not emitted.
4448 * So we must emit it before ending extent_fiemap().
4450 static int emit_last_fiemap_cache(struct btrfs_fs_info *fs_info,
4451 struct fiemap_extent_info *fieinfo,
4452 struct fiemap_cache *cache)
4459 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
4460 cache->len, cache->flags);
4461 cache->cached = false;
4467 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4468 __u64 start, __u64 len, get_extent_t *get_extent)
4472 u64 max = start + len;
4476 u64 last_for_get_extent = 0;
4478 u64 isize = i_size_read(inode);
4479 struct btrfs_key found_key;
4480 struct extent_map *em = NULL;
4481 struct extent_state *cached_state = NULL;
4482 struct btrfs_path *path;
4483 struct btrfs_root *root = BTRFS_I(inode)->root;
4484 struct fiemap_cache cache = { 0 };
4493 path = btrfs_alloc_path();
4496 path->leave_spinning = 1;
4498 start = round_down(start, btrfs_inode_sectorsize(inode));
4499 len = round_up(max, btrfs_inode_sectorsize(inode)) - start;
4502 * lookup the last file extent. We're not using i_size here
4503 * because there might be preallocation past i_size
4505 ret = btrfs_lookup_file_extent(NULL, root, path,
4506 btrfs_ino(BTRFS_I(inode)), -1, 0);
4508 btrfs_free_path(path);
4517 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
4518 found_type = found_key.type;
4520 /* No extents, but there might be delalloc bits */
4521 if (found_key.objectid != btrfs_ino(BTRFS_I(inode)) ||
4522 found_type != BTRFS_EXTENT_DATA_KEY) {
4523 /* have to trust i_size as the end */
4525 last_for_get_extent = isize;
4528 * remember the start of the last extent. There are a
4529 * bunch of different factors that go into the length of the
4530 * extent, so its much less complex to remember where it started
4532 last = found_key.offset;
4533 last_for_get_extent = last + 1;
4535 btrfs_release_path(path);
4538 * we might have some extents allocated but more delalloc past those
4539 * extents. so, we trust isize unless the start of the last extent is
4544 last_for_get_extent = isize;
4547 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4550 em = get_extent_skip_holes(inode, start, last_for_get_extent,
4560 u64 offset_in_extent = 0;
4562 /* break if the extent we found is outside the range */
4563 if (em->start >= max || extent_map_end(em) < off)
4567 * get_extent may return an extent that starts before our
4568 * requested range. We have to make sure the ranges
4569 * we return to fiemap always move forward and don't
4570 * overlap, so adjust the offsets here
4572 em_start = max(em->start, off);
4575 * record the offset from the start of the extent
4576 * for adjusting the disk offset below. Only do this if the
4577 * extent isn't compressed since our in ram offset may be past
4578 * what we have actually allocated on disk.
4580 if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4581 offset_in_extent = em_start - em->start;
4582 em_end = extent_map_end(em);
4583 em_len = em_end - em_start;
4588 * bump off for our next call to get_extent
4590 off = extent_map_end(em);
4594 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
4596 flags |= FIEMAP_EXTENT_LAST;
4597 } else if (em->block_start == EXTENT_MAP_INLINE) {
4598 flags |= (FIEMAP_EXTENT_DATA_INLINE |
4599 FIEMAP_EXTENT_NOT_ALIGNED);
4600 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
4601 flags |= (FIEMAP_EXTENT_DELALLOC |
4602 FIEMAP_EXTENT_UNKNOWN);
4603 } else if (fieinfo->fi_extents_max) {
4604 struct btrfs_trans_handle *trans;
4606 u64 bytenr = em->block_start -
4607 (em->start - em->orig_start);
4609 disko = em->block_start + offset_in_extent;
4612 * We need a trans handle to get delayed refs
4614 trans = btrfs_join_transaction(root);
4616 * It's OK if we can't start a trans we can still check
4623 * As btrfs supports shared space, this information
4624 * can be exported to userspace tools via
4625 * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0
4626 * then we're just getting a count and we can skip the
4629 ret = btrfs_check_shared(trans, root->fs_info,
4631 btrfs_ino(BTRFS_I(inode)), bytenr);
4633 btrfs_end_transaction(trans);
4637 flags |= FIEMAP_EXTENT_SHARED;
4640 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4641 flags |= FIEMAP_EXTENT_ENCODED;
4642 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4643 flags |= FIEMAP_EXTENT_UNWRITTEN;
4645 free_extent_map(em);
4647 if ((em_start >= last) || em_len == (u64)-1 ||
4648 (last == (u64)-1 && isize <= em_end)) {
4649 flags |= FIEMAP_EXTENT_LAST;
4653 /* now scan forward to see if this is really the last extent. */
4654 em = get_extent_skip_holes(inode, off, last_for_get_extent,
4661 flags |= FIEMAP_EXTENT_LAST;
4664 ret = emit_fiemap_extent(fieinfo, &cache, em_start, disko,
4674 ret = emit_last_fiemap_cache(root->fs_info, fieinfo, &cache);
4675 free_extent_map(em);
4677 btrfs_free_path(path);
4678 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4679 &cached_state, GFP_NOFS);
4683 static void __free_extent_buffer(struct extent_buffer *eb)
4685 btrfs_leak_debug_del(&eb->leak_list);
4686 kmem_cache_free(extent_buffer_cache, eb);
4689 int extent_buffer_under_io(struct extent_buffer *eb)
4691 return (atomic_read(&eb->io_pages) ||
4692 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4693 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4697 * Helper for releasing extent buffer page.
4699 static void btrfs_release_extent_buffer_page(struct extent_buffer *eb)
4701 unsigned long index;
4703 int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4705 BUG_ON(extent_buffer_under_io(eb));
4707 index = num_extent_pages(eb->start, eb->len);
4713 page = eb->pages[index];
4717 spin_lock(&page->mapping->private_lock);
4719 * We do this since we'll remove the pages after we've
4720 * removed the eb from the radix tree, so we could race
4721 * and have this page now attached to the new eb. So
4722 * only clear page_private if it's still connected to
4725 if (PagePrivate(page) &&
4726 page->private == (unsigned long)eb) {
4727 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4728 BUG_ON(PageDirty(page));
4729 BUG_ON(PageWriteback(page));
4731 * We need to make sure we haven't be attached
4734 ClearPagePrivate(page);
4735 set_page_private(page, 0);
4736 /* One for the page private */
4741 spin_unlock(&page->mapping->private_lock);
4743 /* One for when we allocated the page */
4745 } while (index != 0);
4749 * Helper for releasing the extent buffer.
4751 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4753 btrfs_release_extent_buffer_page(eb);
4754 __free_extent_buffer(eb);
4757 static struct extent_buffer *
4758 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
4761 struct extent_buffer *eb = NULL;
4763 eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL);
4766 eb->fs_info = fs_info;
4768 rwlock_init(&eb->lock);
4769 atomic_set(&eb->write_locks, 0);
4770 atomic_set(&eb->read_locks, 0);
4771 atomic_set(&eb->blocking_readers, 0);
4772 atomic_set(&eb->blocking_writers, 0);
4773 atomic_set(&eb->spinning_readers, 0);
4774 atomic_set(&eb->spinning_writers, 0);
4775 eb->lock_nested = 0;
4776 init_waitqueue_head(&eb->write_lock_wq);
4777 init_waitqueue_head(&eb->read_lock_wq);
4779 btrfs_leak_debug_add(&eb->leak_list, &buffers);
4781 spin_lock_init(&eb->refs_lock);
4782 atomic_set(&eb->refs, 1);
4783 atomic_set(&eb->io_pages, 0);
4786 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4788 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4789 > MAX_INLINE_EXTENT_BUFFER_SIZE);
4790 BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE);
4795 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src)
4799 struct extent_buffer *new;
4800 unsigned long num_pages = num_extent_pages(src->start, src->len);
4802 new = __alloc_extent_buffer(src->fs_info, src->start, src->len);
4806 for (i = 0; i < num_pages; i++) {
4807 p = alloc_page(GFP_NOFS);
4809 btrfs_release_extent_buffer(new);
4812 attach_extent_buffer_page(new, p);
4813 WARN_ON(PageDirty(p));
4816 copy_page(page_address(p), page_address(src->pages[i]));
4819 set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
4820 set_bit(EXTENT_BUFFER_DUMMY, &new->bflags);
4825 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4826 u64 start, unsigned long len)
4828 struct extent_buffer *eb;
4829 unsigned long num_pages;
4832 num_pages = num_extent_pages(start, len);
4834 eb = __alloc_extent_buffer(fs_info, start, len);
4838 for (i = 0; i < num_pages; i++) {
4839 eb->pages[i] = alloc_page(GFP_NOFS);
4843 set_extent_buffer_uptodate(eb);
4844 btrfs_set_header_nritems(eb, 0);
4845 set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4850 __free_page(eb->pages[i - 1]);
4851 __free_extent_buffer(eb);
4855 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4858 return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize);
4861 static void check_buffer_tree_ref(struct extent_buffer *eb)
4864 /* the ref bit is tricky. We have to make sure it is set
4865 * if we have the buffer dirty. Otherwise the
4866 * code to free a buffer can end up dropping a dirty
4869 * Once the ref bit is set, it won't go away while the
4870 * buffer is dirty or in writeback, and it also won't
4871 * go away while we have the reference count on the
4874 * We can't just set the ref bit without bumping the
4875 * ref on the eb because free_extent_buffer might
4876 * see the ref bit and try to clear it. If this happens
4877 * free_extent_buffer might end up dropping our original
4878 * ref by mistake and freeing the page before we are able
4879 * to add one more ref.
4881 * So bump the ref count first, then set the bit. If someone
4882 * beat us to it, drop the ref we added.
4884 refs = atomic_read(&eb->refs);
4885 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4888 spin_lock(&eb->refs_lock);
4889 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4890 atomic_inc(&eb->refs);
4891 spin_unlock(&eb->refs_lock);
4894 static void mark_extent_buffer_accessed(struct extent_buffer *eb,
4895 struct page *accessed)
4897 unsigned long num_pages, i;
4899 check_buffer_tree_ref(eb);
4901 num_pages = num_extent_pages(eb->start, eb->len);
4902 for (i = 0; i < num_pages; i++) {
4903 struct page *p = eb->pages[i];
4906 mark_page_accessed(p);
4910 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
4913 struct extent_buffer *eb;
4916 eb = radix_tree_lookup(&fs_info->buffer_radix,
4917 start >> PAGE_SHIFT);
4918 if (eb && atomic_inc_not_zero(&eb->refs)) {
4921 * Lock our eb's refs_lock to avoid races with
4922 * free_extent_buffer. When we get our eb it might be flagged
4923 * with EXTENT_BUFFER_STALE and another task running
4924 * free_extent_buffer might have seen that flag set,
4925 * eb->refs == 2, that the buffer isn't under IO (dirty and
4926 * writeback flags not set) and it's still in the tree (flag
4927 * EXTENT_BUFFER_TREE_REF set), therefore being in the process
4928 * of decrementing the extent buffer's reference count twice.
4929 * So here we could race and increment the eb's reference count,
4930 * clear its stale flag, mark it as dirty and drop our reference
4931 * before the other task finishes executing free_extent_buffer,
4932 * which would later result in an attempt to free an extent
4933 * buffer that is dirty.
4935 if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) {
4936 spin_lock(&eb->refs_lock);
4937 spin_unlock(&eb->refs_lock);
4939 mark_extent_buffer_accessed(eb, NULL);
4947 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4948 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
4951 struct extent_buffer *eb, *exists = NULL;
4954 eb = find_extent_buffer(fs_info, start);
4957 eb = alloc_dummy_extent_buffer(fs_info, start);
4960 eb->fs_info = fs_info;
4962 ret = radix_tree_preload(GFP_NOFS);
4965 spin_lock(&fs_info->buffer_lock);
4966 ret = radix_tree_insert(&fs_info->buffer_radix,
4967 start >> PAGE_SHIFT, eb);
4968 spin_unlock(&fs_info->buffer_lock);
4969 radix_tree_preload_end();
4970 if (ret == -EEXIST) {
4971 exists = find_extent_buffer(fs_info, start);
4977 check_buffer_tree_ref(eb);
4978 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
4981 * We will free dummy extent buffer's if they come into
4982 * free_extent_buffer with a ref count of 2, but if we are using this we
4983 * want the buffers to stay in memory until we're done with them, so
4984 * bump the ref count again.
4986 atomic_inc(&eb->refs);
4989 btrfs_release_extent_buffer(eb);
4994 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
4997 unsigned long len = fs_info->nodesize;
4998 unsigned long num_pages = num_extent_pages(start, len);
5000 unsigned long index = start >> PAGE_SHIFT;
5001 struct extent_buffer *eb;
5002 struct extent_buffer *exists = NULL;
5004 struct address_space *mapping = fs_info->btree_inode->i_mapping;
5008 if (!IS_ALIGNED(start, fs_info->sectorsize)) {
5009 btrfs_err(fs_info, "bad tree block start %llu", start);
5010 return ERR_PTR(-EINVAL);
5013 eb = find_extent_buffer(fs_info, start);
5017 eb = __alloc_extent_buffer(fs_info, start, len);
5019 return ERR_PTR(-ENOMEM);
5021 for (i = 0; i < num_pages; i++, index++) {
5022 p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL);
5024 exists = ERR_PTR(-ENOMEM);
5028 spin_lock(&mapping->private_lock);
5029 if (PagePrivate(p)) {
5031 * We could have already allocated an eb for this page
5032 * and attached one so lets see if we can get a ref on
5033 * the existing eb, and if we can we know it's good and
5034 * we can just return that one, else we know we can just
5035 * overwrite page->private.
5037 exists = (struct extent_buffer *)p->private;
5038 if (atomic_inc_not_zero(&exists->refs)) {
5039 spin_unlock(&mapping->private_lock);
5042 mark_extent_buffer_accessed(exists, p);
5048 * Do this so attach doesn't complain and we need to
5049 * drop the ref the old guy had.
5051 ClearPagePrivate(p);
5052 WARN_ON(PageDirty(p));
5055 attach_extent_buffer_page(eb, p);
5056 spin_unlock(&mapping->private_lock);
5057 WARN_ON(PageDirty(p));
5059 if (!PageUptodate(p))
5063 * see below about how we avoid a nasty race with release page
5064 * and why we unlock later
5068 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5070 ret = radix_tree_preload(GFP_NOFS);
5072 exists = ERR_PTR(ret);
5076 spin_lock(&fs_info->buffer_lock);
5077 ret = radix_tree_insert(&fs_info->buffer_radix,
5078 start >> PAGE_SHIFT, eb);
5079 spin_unlock(&fs_info->buffer_lock);
5080 radix_tree_preload_end();
5081 if (ret == -EEXIST) {
5082 exists = find_extent_buffer(fs_info, start);
5088 /* add one reference for the tree */
5089 check_buffer_tree_ref(eb);
5090 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
5093 * there is a race where release page may have
5094 * tried to find this extent buffer in the radix
5095 * but failed. It will tell the VM it is safe to
5096 * reclaim the, and it will clear the page private bit.
5097 * We must make sure to set the page private bit properly
5098 * after the extent buffer is in the radix tree so
5099 * it doesn't get lost
5101 SetPageChecked(eb->pages[0]);
5102 for (i = 1; i < num_pages; i++) {
5104 ClearPageChecked(p);
5107 unlock_page(eb->pages[0]);
5111 WARN_ON(!atomic_dec_and_test(&eb->refs));
5112 for (i = 0; i < num_pages; i++) {
5114 unlock_page(eb->pages[i]);
5117 btrfs_release_extent_buffer(eb);
5121 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
5123 struct extent_buffer *eb =
5124 container_of(head, struct extent_buffer, rcu_head);
5126 __free_extent_buffer(eb);
5129 /* Expects to have eb->eb_lock already held */
5130 static int release_extent_buffer(struct extent_buffer *eb)
5132 WARN_ON(atomic_read(&eb->refs) == 0);
5133 if (atomic_dec_and_test(&eb->refs)) {
5134 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
5135 struct btrfs_fs_info *fs_info = eb->fs_info;
5137 spin_unlock(&eb->refs_lock);
5139 spin_lock(&fs_info->buffer_lock);
5140 radix_tree_delete(&fs_info->buffer_radix,
5141 eb->start >> PAGE_SHIFT);
5142 spin_unlock(&fs_info->buffer_lock);
5144 spin_unlock(&eb->refs_lock);
5147 /* Should be safe to release our pages at this point */
5148 btrfs_release_extent_buffer_page(eb);
5149 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5150 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))) {
5151 __free_extent_buffer(eb);
5155 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
5158 spin_unlock(&eb->refs_lock);
5163 void free_extent_buffer(struct extent_buffer *eb)
5171 refs = atomic_read(&eb->refs);
5174 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
5179 spin_lock(&eb->refs_lock);
5180 if (atomic_read(&eb->refs) == 2 &&
5181 test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))
5182 atomic_dec(&eb->refs);
5184 if (atomic_read(&eb->refs) == 2 &&
5185 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
5186 !extent_buffer_under_io(eb) &&
5187 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5188 atomic_dec(&eb->refs);
5191 * I know this is terrible, but it's temporary until we stop tracking
5192 * the uptodate bits and such for the extent buffers.
5194 release_extent_buffer(eb);
5197 void free_extent_buffer_stale(struct extent_buffer *eb)
5202 spin_lock(&eb->refs_lock);
5203 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
5205 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
5206 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5207 atomic_dec(&eb->refs);
5208 release_extent_buffer(eb);
5211 void clear_extent_buffer_dirty(struct extent_buffer *eb)
5214 unsigned long num_pages;
5217 num_pages = num_extent_pages(eb->start, eb->len);
5219 for (i = 0; i < num_pages; i++) {
5220 page = eb->pages[i];
5221 if (!PageDirty(page))
5225 WARN_ON(!PagePrivate(page));
5227 clear_page_dirty_for_io(page);
5228 spin_lock_irq(&page->mapping->tree_lock);
5229 if (!PageDirty(page)) {
5230 radix_tree_tag_clear(&page->mapping->page_tree,
5232 PAGECACHE_TAG_DIRTY);
5234 spin_unlock_irq(&page->mapping->tree_lock);
5235 ClearPageError(page);
5238 WARN_ON(atomic_read(&eb->refs) == 0);
5241 int set_extent_buffer_dirty(struct extent_buffer *eb)
5244 unsigned long num_pages;
5247 check_buffer_tree_ref(eb);
5249 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
5251 num_pages = num_extent_pages(eb->start, eb->len);
5252 WARN_ON(atomic_read(&eb->refs) == 0);
5253 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
5255 for (i = 0; i < num_pages; i++)
5256 set_page_dirty(eb->pages[i]);
5260 void clear_extent_buffer_uptodate(struct extent_buffer *eb)
5264 unsigned long num_pages;
5266 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5267 num_pages = num_extent_pages(eb->start, eb->len);
5268 for (i = 0; i < num_pages; i++) {
5269 page = eb->pages[i];
5271 ClearPageUptodate(page);
5275 void set_extent_buffer_uptodate(struct extent_buffer *eb)
5279 unsigned long num_pages;
5281 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5282 num_pages = num_extent_pages(eb->start, eb->len);
5283 for (i = 0; i < num_pages; i++) {
5284 page = eb->pages[i];
5285 SetPageUptodate(page);
5289 int extent_buffer_uptodate(struct extent_buffer *eb)
5291 return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5294 int read_extent_buffer_pages(struct extent_io_tree *tree,
5295 struct extent_buffer *eb, int wait,
5296 get_extent_t *get_extent, int mirror_num)
5302 int locked_pages = 0;
5303 int all_uptodate = 1;
5304 unsigned long num_pages;
5305 unsigned long num_reads = 0;
5306 struct bio *bio = NULL;
5307 unsigned long bio_flags = 0;
5309 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
5312 num_pages = num_extent_pages(eb->start, eb->len);
5313 for (i = 0; i < num_pages; i++) {
5314 page = eb->pages[i];
5315 if (wait == WAIT_NONE) {
5316 if (!trylock_page(page))
5324 * We need to firstly lock all pages to make sure that
5325 * the uptodate bit of our pages won't be affected by
5326 * clear_extent_buffer_uptodate().
5328 for (i = 0; i < num_pages; i++) {
5329 page = eb->pages[i];
5330 if (!PageUptodate(page)) {
5337 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5341 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
5342 eb->read_mirror = 0;
5343 atomic_set(&eb->io_pages, num_reads);
5344 for (i = 0; i < num_pages; i++) {
5345 page = eb->pages[i];
5347 if (!PageUptodate(page)) {
5349 atomic_dec(&eb->io_pages);
5354 ClearPageError(page);
5355 err = __extent_read_full_page(tree, page,
5357 mirror_num, &bio_flags,
5362 * We use &bio in above __extent_read_full_page,
5363 * so we ensure that if it returns error, the
5364 * current page fails to add itself to bio and
5365 * it's been unlocked.
5367 * We must dec io_pages by ourselves.
5369 atomic_dec(&eb->io_pages);
5377 err = submit_one_bio(bio, mirror_num, bio_flags);
5382 if (ret || wait != WAIT_COMPLETE)
5385 for (i = 0; i < num_pages; i++) {
5386 page = eb->pages[i];
5387 wait_on_page_locked(page);
5388 if (!PageUptodate(page))
5395 while (locked_pages > 0) {
5397 page = eb->pages[locked_pages];
5403 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
5404 unsigned long start,
5411 char *dst = (char *)dstv;
5412 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5413 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5415 WARN_ON(start > eb->len);
5416 WARN_ON(start + len > eb->start + eb->len);
5418 offset = (start_offset + start) & (PAGE_SIZE - 1);
5421 page = eb->pages[i];
5423 cur = min(len, (PAGE_SIZE - offset));
5424 kaddr = page_address(page);
5425 memcpy(dst, kaddr + offset, cur);
5434 int read_extent_buffer_to_user(struct extent_buffer *eb, void __user *dstv,
5435 unsigned long start,
5442 char __user *dst = (char __user *)dstv;
5443 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5444 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5447 WARN_ON(start > eb->len);
5448 WARN_ON(start + len > eb->start + eb->len);
5450 offset = (start_offset + start) & (PAGE_SIZE - 1);
5453 page = eb->pages[i];
5455 cur = min(len, (PAGE_SIZE - offset));
5456 kaddr = page_address(page);
5457 if (copy_to_user(dst, kaddr + offset, cur)) {
5472 * return 0 if the item is found within a page.
5473 * return 1 if the item spans two pages.
5474 * return -EINVAL otherwise.
5476 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
5477 unsigned long min_len, char **map,
5478 unsigned long *map_start,
5479 unsigned long *map_len)
5481 size_t offset = start & (PAGE_SIZE - 1);
5484 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5485 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5486 unsigned long end_i = (start_offset + start + min_len - 1) >>
5493 offset = start_offset;
5497 *map_start = ((u64)i << PAGE_SHIFT) - start_offset;
5500 if (start + min_len > eb->len) {
5501 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n",
5502 eb->start, eb->len, start, min_len);
5507 kaddr = page_address(p);
5508 *map = kaddr + offset;
5509 *map_len = PAGE_SIZE - offset;
5513 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
5514 unsigned long start,
5521 char *ptr = (char *)ptrv;
5522 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5523 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5526 WARN_ON(start > eb->len);
5527 WARN_ON(start + len > eb->start + eb->len);
5529 offset = (start_offset + start) & (PAGE_SIZE - 1);
5532 page = eb->pages[i];
5534 cur = min(len, (PAGE_SIZE - offset));
5536 kaddr = page_address(page);
5537 ret = memcmp(ptr, kaddr + offset, cur);
5549 void write_extent_buffer_chunk_tree_uuid(struct extent_buffer *eb,
5554 WARN_ON(!PageUptodate(eb->pages[0]));
5555 kaddr = page_address(eb->pages[0]);
5556 memcpy(kaddr + offsetof(struct btrfs_header, chunk_tree_uuid), srcv,
5560 void write_extent_buffer_fsid(struct extent_buffer *eb, const void *srcv)
5564 WARN_ON(!PageUptodate(eb->pages[0]));
5565 kaddr = page_address(eb->pages[0]);
5566 memcpy(kaddr + offsetof(struct btrfs_header, fsid), srcv,
5570 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
5571 unsigned long start, unsigned long len)
5577 char *src = (char *)srcv;
5578 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5579 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5581 WARN_ON(start > eb->len);
5582 WARN_ON(start + len > eb->start + eb->len);
5584 offset = (start_offset + start) & (PAGE_SIZE - 1);
5587 page = eb->pages[i];
5588 WARN_ON(!PageUptodate(page));
5590 cur = min(len, PAGE_SIZE - offset);
5591 kaddr = page_address(page);
5592 memcpy(kaddr + offset, src, cur);
5601 void memzero_extent_buffer(struct extent_buffer *eb, unsigned long start,
5608 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5609 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5611 WARN_ON(start > eb->len);
5612 WARN_ON(start + len > eb->start + eb->len);
5614 offset = (start_offset + start) & (PAGE_SIZE - 1);
5617 page = eb->pages[i];
5618 WARN_ON(!PageUptodate(page));
5620 cur = min(len, PAGE_SIZE - offset);
5621 kaddr = page_address(page);
5622 memset(kaddr + offset, 0, cur);
5630 void copy_extent_buffer_full(struct extent_buffer *dst,
5631 struct extent_buffer *src)
5636 ASSERT(dst->len == src->len);
5638 num_pages = num_extent_pages(dst->start, dst->len);
5639 for (i = 0; i < num_pages; i++)
5640 copy_page(page_address(dst->pages[i]),
5641 page_address(src->pages[i]));
5644 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
5645 unsigned long dst_offset, unsigned long src_offset,
5648 u64 dst_len = dst->len;
5653 size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1);
5654 unsigned long i = (start_offset + dst_offset) >> PAGE_SHIFT;
5656 WARN_ON(src->len != dst_len);
5658 offset = (start_offset + dst_offset) &
5662 page = dst->pages[i];
5663 WARN_ON(!PageUptodate(page));
5665 cur = min(len, (unsigned long)(PAGE_SIZE - offset));
5667 kaddr = page_address(page);
5668 read_extent_buffer(src, kaddr + offset, src_offset, cur);
5677 void le_bitmap_set(u8 *map, unsigned int start, int len)
5679 u8 *p = map + BIT_BYTE(start);
5680 const unsigned int size = start + len;
5681 int bits_to_set = BITS_PER_BYTE - (start % BITS_PER_BYTE);
5682 u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(start);
5684 while (len - bits_to_set >= 0) {
5687 bits_to_set = BITS_PER_BYTE;
5692 mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
5697 void le_bitmap_clear(u8 *map, unsigned int start, int len)
5699 u8 *p = map + BIT_BYTE(start);
5700 const unsigned int size = start + len;
5701 int bits_to_clear = BITS_PER_BYTE - (start % BITS_PER_BYTE);
5702 u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(start);
5704 while (len - bits_to_clear >= 0) {
5705 *p &= ~mask_to_clear;
5706 len -= bits_to_clear;
5707 bits_to_clear = BITS_PER_BYTE;
5712 mask_to_clear &= BITMAP_LAST_BYTE_MASK(size);
5713 *p &= ~mask_to_clear;
5718 * eb_bitmap_offset() - calculate the page and offset of the byte containing the
5720 * @eb: the extent buffer
5721 * @start: offset of the bitmap item in the extent buffer
5723 * @page_index: return index of the page in the extent buffer that contains the
5725 * @page_offset: return offset into the page given by page_index
5727 * This helper hides the ugliness of finding the byte in an extent buffer which
5728 * contains a given bit.
5730 static inline void eb_bitmap_offset(struct extent_buffer *eb,
5731 unsigned long start, unsigned long nr,
5732 unsigned long *page_index,
5733 size_t *page_offset)
5735 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5736 size_t byte_offset = BIT_BYTE(nr);
5740 * The byte we want is the offset of the extent buffer + the offset of
5741 * the bitmap item in the extent buffer + the offset of the byte in the
5744 offset = start_offset + start + byte_offset;
5746 *page_index = offset >> PAGE_SHIFT;
5747 *page_offset = offset & (PAGE_SIZE - 1);
5751 * extent_buffer_test_bit - determine whether a bit in a bitmap item is set
5752 * @eb: the extent buffer
5753 * @start: offset of the bitmap item in the extent buffer
5754 * @nr: bit number to test
5756 int extent_buffer_test_bit(struct extent_buffer *eb, unsigned long start,
5764 eb_bitmap_offset(eb, start, nr, &i, &offset);
5765 page = eb->pages[i];
5766 WARN_ON(!PageUptodate(page));
5767 kaddr = page_address(page);
5768 return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1)));
5772 * extent_buffer_bitmap_set - set an area of a bitmap
5773 * @eb: the extent buffer
5774 * @start: offset of the bitmap item in the extent buffer
5775 * @pos: bit number of the first bit
5776 * @len: number of bits to set
5778 void extent_buffer_bitmap_set(struct extent_buffer *eb, unsigned long start,
5779 unsigned long pos, unsigned long len)
5785 const unsigned int size = pos + len;
5786 int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5787 u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos);
5789 eb_bitmap_offset(eb, start, pos, &i, &offset);
5790 page = eb->pages[i];
5791 WARN_ON(!PageUptodate(page));
5792 kaddr = page_address(page);
5794 while (len >= bits_to_set) {
5795 kaddr[offset] |= mask_to_set;
5797 bits_to_set = BITS_PER_BYTE;
5799 if (++offset >= PAGE_SIZE && len > 0) {
5801 page = eb->pages[++i];
5802 WARN_ON(!PageUptodate(page));
5803 kaddr = page_address(page);
5807 mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
5808 kaddr[offset] |= mask_to_set;
5814 * extent_buffer_bitmap_clear - clear an area of a bitmap
5815 * @eb: the extent buffer
5816 * @start: offset of the bitmap item in the extent buffer
5817 * @pos: bit number of the first bit
5818 * @len: number of bits to clear
5820 void extent_buffer_bitmap_clear(struct extent_buffer *eb, unsigned long start,
5821 unsigned long pos, unsigned long len)
5827 const unsigned int size = pos + len;
5828 int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5829 u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos);
5831 eb_bitmap_offset(eb, start, pos, &i, &offset);
5832 page = eb->pages[i];
5833 WARN_ON(!PageUptodate(page));
5834 kaddr = page_address(page);
5836 while (len >= bits_to_clear) {
5837 kaddr[offset] &= ~mask_to_clear;
5838 len -= bits_to_clear;
5839 bits_to_clear = BITS_PER_BYTE;
5841 if (++offset >= PAGE_SIZE && len > 0) {
5843 page = eb->pages[++i];
5844 WARN_ON(!PageUptodate(page));
5845 kaddr = page_address(page);
5849 mask_to_clear &= BITMAP_LAST_BYTE_MASK(size);
5850 kaddr[offset] &= ~mask_to_clear;
5854 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
5856 unsigned long distance = (src > dst) ? src - dst : dst - src;
5857 return distance < len;
5860 static void copy_pages(struct page *dst_page, struct page *src_page,
5861 unsigned long dst_off, unsigned long src_off,
5864 char *dst_kaddr = page_address(dst_page);
5866 int must_memmove = 0;
5868 if (dst_page != src_page) {
5869 src_kaddr = page_address(src_page);
5871 src_kaddr = dst_kaddr;
5872 if (areas_overlap(src_off, dst_off, len))
5877 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
5879 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
5882 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5883 unsigned long src_offset, unsigned long len)
5885 struct btrfs_fs_info *fs_info = dst->fs_info;
5887 size_t dst_off_in_page;
5888 size_t src_off_in_page;
5889 size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1);
5890 unsigned long dst_i;
5891 unsigned long src_i;
5893 if (src_offset + len > dst->len) {
5895 "memmove bogus src_offset %lu move len %lu dst len %lu",
5896 src_offset, len, dst->len);
5899 if (dst_offset + len > dst->len) {
5901 "memmove bogus dst_offset %lu move len %lu dst len %lu",
5902 dst_offset, len, dst->len);
5907 dst_off_in_page = (start_offset + dst_offset) &
5909 src_off_in_page = (start_offset + src_offset) &
5912 dst_i = (start_offset + dst_offset) >> PAGE_SHIFT;
5913 src_i = (start_offset + src_offset) >> PAGE_SHIFT;
5915 cur = min(len, (unsigned long)(PAGE_SIZE -
5917 cur = min_t(unsigned long, cur,
5918 (unsigned long)(PAGE_SIZE - dst_off_in_page));
5920 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5921 dst_off_in_page, src_off_in_page, cur);
5929 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5930 unsigned long src_offset, unsigned long len)
5932 struct btrfs_fs_info *fs_info = dst->fs_info;
5934 size_t dst_off_in_page;
5935 size_t src_off_in_page;
5936 unsigned long dst_end = dst_offset + len - 1;
5937 unsigned long src_end = src_offset + len - 1;
5938 size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1);
5939 unsigned long dst_i;
5940 unsigned long src_i;
5942 if (src_offset + len > dst->len) {
5944 "memmove bogus src_offset %lu move len %lu len %lu",
5945 src_offset, len, dst->len);
5948 if (dst_offset + len > dst->len) {
5950 "memmove bogus dst_offset %lu move len %lu len %lu",
5951 dst_offset, len, dst->len);
5954 if (dst_offset < src_offset) {
5955 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
5959 dst_i = (start_offset + dst_end) >> PAGE_SHIFT;
5960 src_i = (start_offset + src_end) >> PAGE_SHIFT;
5962 dst_off_in_page = (start_offset + dst_end) &
5964 src_off_in_page = (start_offset + src_end) &
5967 cur = min_t(unsigned long, len, src_off_in_page + 1);
5968 cur = min(cur, dst_off_in_page + 1);
5969 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5970 dst_off_in_page - cur + 1,
5971 src_off_in_page - cur + 1, cur);
5979 int try_release_extent_buffer(struct page *page)
5981 struct extent_buffer *eb;
5984 * We need to make sure nobody is attaching this page to an eb right
5987 spin_lock(&page->mapping->private_lock);
5988 if (!PagePrivate(page)) {
5989 spin_unlock(&page->mapping->private_lock);
5993 eb = (struct extent_buffer *)page->private;
5997 * This is a little awful but should be ok, we need to make sure that
5998 * the eb doesn't disappear out from under us while we're looking at
6001 spin_lock(&eb->refs_lock);
6002 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
6003 spin_unlock(&eb->refs_lock);
6004 spin_unlock(&page->mapping->private_lock);
6007 spin_unlock(&page->mapping->private_lock);
6010 * If tree ref isn't set then we know the ref on this eb is a real ref,
6011 * so just return, this page will likely be freed soon anyway.
6013 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
6014 spin_unlock(&eb->refs_lock);
6018 return release_extent_buffer(eb);