1 // SPDX-License-Identifier: GPL-2.0
3 #include <linux/bitops.h>
4 #include <linux/slab.h>
7 #include <linux/pagemap.h>
8 #include <linux/page-flags.h>
9 #include <linux/spinlock.h>
10 #include <linux/blkdev.h>
11 #include <linux/swap.h>
12 #include <linux/writeback.h>
13 #include <linux/pagevec.h>
14 #include <linux/prefetch.h>
15 #include <linux/cleancache.h>
16 #include "extent_io.h"
17 #include "extent_map.h"
19 #include "btrfs_inode.h"
21 #include "check-integrity.h"
23 #include "rcu-string.h"
27 static struct kmem_cache *extent_state_cache;
28 static struct kmem_cache *extent_buffer_cache;
29 static struct bio_set btrfs_bioset;
31 static inline bool extent_state_in_tree(const struct extent_state *state)
33 return !RB_EMPTY_NODE(&state->rb_node);
36 #ifdef CONFIG_BTRFS_DEBUG
37 static LIST_HEAD(buffers);
38 static LIST_HEAD(states);
40 static DEFINE_SPINLOCK(leak_lock);
43 void btrfs_leak_debug_add(struct list_head *new, struct list_head *head)
47 spin_lock_irqsave(&leak_lock, flags);
49 spin_unlock_irqrestore(&leak_lock, flags);
53 void btrfs_leak_debug_del(struct list_head *entry)
57 spin_lock_irqsave(&leak_lock, flags);
59 spin_unlock_irqrestore(&leak_lock, flags);
63 void btrfs_leak_debug_check(void)
65 struct extent_state *state;
66 struct extent_buffer *eb;
68 while (!list_empty(&states)) {
69 state = list_entry(states.next, struct extent_state, leak_list);
70 pr_err("BTRFS: state leak: start %llu end %llu state %u in tree %d refs %d\n",
71 state->start, state->end, state->state,
72 extent_state_in_tree(state),
73 refcount_read(&state->refs));
74 list_del(&state->leak_list);
75 kmem_cache_free(extent_state_cache, state);
78 while (!list_empty(&buffers)) {
79 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
80 pr_err("BTRFS: buffer leak start %llu len %lu refs %d bflags %lu\n",
81 eb->start, eb->len, atomic_read(&eb->refs), eb->bflags);
82 list_del(&eb->leak_list);
83 kmem_cache_free(extent_buffer_cache, eb);
87 #define btrfs_debug_check_extent_io_range(tree, start, end) \
88 __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end))
89 static inline void __btrfs_debug_check_extent_io_range(const char *caller,
90 struct extent_io_tree *tree, u64 start, u64 end)
92 struct inode *inode = tree->private_data;
95 if (!inode || !is_data_inode(inode))
98 isize = i_size_read(inode);
99 if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) {
100 btrfs_debug_rl(BTRFS_I(inode)->root->fs_info,
101 "%s: ino %llu isize %llu odd range [%llu,%llu]",
102 caller, btrfs_ino(BTRFS_I(inode)), isize, start, end);
106 #define btrfs_leak_debug_add(new, head) do {} while (0)
107 #define btrfs_leak_debug_del(entry) do {} while (0)
108 #define btrfs_leak_debug_check() do {} while (0)
109 #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0)
112 #define BUFFER_LRU_MAX 64
117 struct rb_node rb_node;
120 struct extent_page_data {
122 struct extent_io_tree *tree;
123 /* tells writepage not to lock the state bits for this range
124 * it still does the unlocking
126 unsigned int extent_locked:1;
128 /* tells the submit_bio code to use REQ_SYNC */
129 unsigned int sync_io:1;
132 static int add_extent_changeset(struct extent_state *state, unsigned bits,
133 struct extent_changeset *changeset,
140 if (set && (state->state & bits) == bits)
142 if (!set && (state->state & bits) == 0)
144 changeset->bytes_changed += state->end - state->start + 1;
145 ret = ulist_add(&changeset->range_changed, state->start, state->end,
150 static void flush_write_bio(struct extent_page_data *epd);
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 if (bioset_init(&btrfs_bioset, 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_exit(&btrfs_bioset);
180 kmem_cache_destroy(extent_buffer_cache);
181 extent_buffer_cache = NULL;
184 kmem_cache_destroy(extent_state_cache);
185 extent_state_cache = NULL;
189 void __cold extent_io_exit(void)
191 btrfs_leak_debug_check();
194 * Make sure all delayed rcu free are flushed before we
198 kmem_cache_destroy(extent_state_cache);
199 kmem_cache_destroy(extent_buffer_cache);
200 bioset_exit(&btrfs_bioset);
203 void extent_io_tree_init(struct extent_io_tree *tree,
206 tree->state = RB_ROOT;
208 tree->dirty_bytes = 0;
209 spin_lock_init(&tree->lock);
210 tree->private_data = private_data;
213 static struct extent_state *alloc_extent_state(gfp_t mask)
215 struct extent_state *state;
218 * The given mask might be not appropriate for the slab allocator,
219 * drop the unsupported bits
221 mask &= ~(__GFP_DMA32|__GFP_HIGHMEM);
222 state = kmem_cache_alloc(extent_state_cache, mask);
226 state->failrec = NULL;
227 RB_CLEAR_NODE(&state->rb_node);
228 btrfs_leak_debug_add(&state->leak_list, &states);
229 refcount_set(&state->refs, 1);
230 init_waitqueue_head(&state->wq);
231 trace_alloc_extent_state(state, mask, _RET_IP_);
235 void free_extent_state(struct extent_state *state)
239 if (refcount_dec_and_test(&state->refs)) {
240 WARN_ON(extent_state_in_tree(state));
241 btrfs_leak_debug_del(&state->leak_list);
242 trace_free_extent_state(state, _RET_IP_);
243 kmem_cache_free(extent_state_cache, state);
247 static struct rb_node *tree_insert(struct rb_root *root,
248 struct rb_node *search_start,
250 struct rb_node *node,
251 struct rb_node ***p_in,
252 struct rb_node **parent_in)
255 struct rb_node *parent = NULL;
256 struct tree_entry *entry;
258 if (p_in && parent_in) {
264 p = search_start ? &search_start : &root->rb_node;
267 entry = rb_entry(parent, struct tree_entry, rb_node);
269 if (offset < entry->start)
271 else if (offset > entry->end)
278 rb_link_node(node, parent, p);
279 rb_insert_color(node, root);
283 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
284 struct rb_node **prev_ret,
285 struct rb_node **next_ret,
286 struct rb_node ***p_ret,
287 struct rb_node **parent_ret)
289 struct rb_root *root = &tree->state;
290 struct rb_node **n = &root->rb_node;
291 struct rb_node *prev = NULL;
292 struct rb_node *orig_prev = NULL;
293 struct tree_entry *entry;
294 struct tree_entry *prev_entry = NULL;
298 entry = rb_entry(prev, struct tree_entry, rb_node);
301 if (offset < entry->start)
303 else if (offset > entry->end)
316 while (prev && offset > prev_entry->end) {
317 prev = rb_next(prev);
318 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
325 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
326 while (prev && offset < prev_entry->start) {
327 prev = rb_prev(prev);
328 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
335 static inline struct rb_node *
336 tree_search_for_insert(struct extent_io_tree *tree,
338 struct rb_node ***p_ret,
339 struct rb_node **parent_ret)
341 struct rb_node *prev = NULL;
344 ret = __etree_search(tree, offset, &prev, NULL, p_ret, parent_ret);
350 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
353 return tree_search_for_insert(tree, offset, NULL, NULL);
357 * utility function to look for merge candidates inside a given range.
358 * Any extents with matching state are merged together into a single
359 * extent in the tree. Extents with EXTENT_IO in their state field
360 * are not merged because the end_io handlers need to be able to do
361 * operations on them without sleeping (or doing allocations/splits).
363 * This should be called with the tree lock held.
365 static void merge_state(struct extent_io_tree *tree,
366 struct extent_state *state)
368 struct extent_state *other;
369 struct rb_node *other_node;
371 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
374 other_node = rb_prev(&state->rb_node);
376 other = rb_entry(other_node, struct extent_state, rb_node);
377 if (other->end == state->start - 1 &&
378 other->state == state->state) {
379 if (tree->private_data &&
380 is_data_inode(tree->private_data))
381 btrfs_merge_delalloc_extent(tree->private_data,
383 state->start = other->start;
384 rb_erase(&other->rb_node, &tree->state);
385 RB_CLEAR_NODE(&other->rb_node);
386 free_extent_state(other);
389 other_node = rb_next(&state->rb_node);
391 other = rb_entry(other_node, struct extent_state, rb_node);
392 if (other->start == state->end + 1 &&
393 other->state == state->state) {
394 if (tree->private_data &&
395 is_data_inode(tree->private_data))
396 btrfs_merge_delalloc_extent(tree->private_data,
398 state->end = other->end;
399 rb_erase(&other->rb_node, &tree->state);
400 RB_CLEAR_NODE(&other->rb_node);
401 free_extent_state(other);
406 static void set_state_bits(struct extent_io_tree *tree,
407 struct extent_state *state, unsigned *bits,
408 struct extent_changeset *changeset);
411 * insert an extent_state struct into the tree. 'bits' are set on the
412 * struct before it is inserted.
414 * This may return -EEXIST if the extent is already there, in which case the
415 * state struct is freed.
417 * The tree lock is not taken internally. This is a utility function and
418 * probably isn't what you want to call (see set/clear_extent_bit).
420 static int insert_state(struct extent_io_tree *tree,
421 struct extent_state *state, u64 start, u64 end,
423 struct rb_node **parent,
424 unsigned *bits, struct extent_changeset *changeset)
426 struct rb_node *node;
429 WARN(1, KERN_ERR "BTRFS: end < start %llu %llu\n",
431 state->start = start;
434 set_state_bits(tree, state, bits, changeset);
436 node = tree_insert(&tree->state, NULL, end, &state->rb_node, p, parent);
438 struct extent_state *found;
439 found = rb_entry(node, struct extent_state, rb_node);
440 pr_err("BTRFS: found node %llu %llu on insert of %llu %llu\n",
441 found->start, found->end, start, end);
444 merge_state(tree, state);
449 * split a given extent state struct in two, inserting the preallocated
450 * struct 'prealloc' as the newly created second half. 'split' indicates an
451 * offset inside 'orig' where it should be split.
454 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
455 * are two extent state structs in the tree:
456 * prealloc: [orig->start, split - 1]
457 * orig: [ split, orig->end ]
459 * The tree locks are not taken by this function. They need to be held
462 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
463 struct extent_state *prealloc, u64 split)
465 struct rb_node *node;
467 if (tree->private_data && is_data_inode(tree->private_data))
468 btrfs_split_delalloc_extent(tree->private_data, orig, split);
470 prealloc->start = orig->start;
471 prealloc->end = split - 1;
472 prealloc->state = orig->state;
475 node = tree_insert(&tree->state, &orig->rb_node, prealloc->end,
476 &prealloc->rb_node, NULL, NULL);
478 free_extent_state(prealloc);
484 static struct extent_state *next_state(struct extent_state *state)
486 struct rb_node *next = rb_next(&state->rb_node);
488 return rb_entry(next, struct extent_state, rb_node);
494 * utility function to clear some bits in an extent state struct.
495 * it will optionally wake up any one waiting on this state (wake == 1).
497 * If no bits are set on the state struct after clearing things, the
498 * struct is freed and removed from the tree
500 static struct extent_state *clear_state_bit(struct extent_io_tree *tree,
501 struct extent_state *state,
502 unsigned *bits, int wake,
503 struct extent_changeset *changeset)
505 struct extent_state *next;
506 unsigned bits_to_clear = *bits & ~EXTENT_CTLBITS;
509 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
510 u64 range = state->end - state->start + 1;
511 WARN_ON(range > tree->dirty_bytes);
512 tree->dirty_bytes -= range;
515 if (tree->private_data && is_data_inode(tree->private_data))
516 btrfs_clear_delalloc_extent(tree->private_data, state, bits);
518 ret = add_extent_changeset(state, bits_to_clear, changeset, 0);
520 state->state &= ~bits_to_clear;
523 if (state->state == 0) {
524 next = next_state(state);
525 if (extent_state_in_tree(state)) {
526 rb_erase(&state->rb_node, &tree->state);
527 RB_CLEAR_NODE(&state->rb_node);
528 free_extent_state(state);
533 merge_state(tree, state);
534 next = next_state(state);
539 static struct extent_state *
540 alloc_extent_state_atomic(struct extent_state *prealloc)
543 prealloc = alloc_extent_state(GFP_ATOMIC);
548 static void extent_io_tree_panic(struct extent_io_tree *tree, int err)
550 struct inode *inode = tree->private_data;
552 btrfs_panic(btrfs_sb(inode->i_sb), err,
553 "locking error: extent tree was modified by another thread while locked");
557 * clear some bits on a range in the tree. This may require splitting
558 * or inserting elements in the tree, so the gfp mask is used to
559 * indicate which allocations or sleeping are allowed.
561 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
562 * the given range from the tree regardless of state (ie for truncate).
564 * the range [start, end] is inclusive.
566 * This takes the tree lock, and returns 0 on success and < 0 on error.
568 int __clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
569 unsigned bits, int wake, int delete,
570 struct extent_state **cached_state,
571 gfp_t mask, struct extent_changeset *changeset)
573 struct extent_state *state;
574 struct extent_state *cached;
575 struct extent_state *prealloc = NULL;
576 struct rb_node *node;
581 btrfs_debug_check_extent_io_range(tree, start, end);
583 if (bits & EXTENT_DELALLOC)
584 bits |= EXTENT_NORESERVE;
587 bits |= ~EXTENT_CTLBITS;
588 bits |= EXTENT_FIRST_DELALLOC;
590 if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY))
593 if (!prealloc && gfpflags_allow_blocking(mask)) {
595 * Don't care for allocation failure here because we might end
596 * up not needing the pre-allocated extent state at all, which
597 * is the case if we only have in the tree extent states that
598 * cover our input range and don't cover too any other range.
599 * If we end up needing a new extent state we allocate it later.
601 prealloc = alloc_extent_state(mask);
604 spin_lock(&tree->lock);
606 cached = *cached_state;
609 *cached_state = NULL;
613 if (cached && extent_state_in_tree(cached) &&
614 cached->start <= start && cached->end > start) {
616 refcount_dec(&cached->refs);
621 free_extent_state(cached);
624 * this search will find the extents that end after
627 node = tree_search(tree, start);
630 state = rb_entry(node, struct extent_state, rb_node);
632 if (state->start > end)
634 WARN_ON(state->end < start);
635 last_end = state->end;
637 /* the state doesn't have the wanted bits, go ahead */
638 if (!(state->state & bits)) {
639 state = next_state(state);
644 * | ---- desired range ---- |
646 * | ------------- state -------------- |
648 * We need to split the extent we found, and may flip
649 * bits on second half.
651 * If the extent we found extends past our range, we
652 * just split and search again. It'll get split again
653 * the next time though.
655 * If the extent we found is inside our range, we clear
656 * the desired bit on it.
659 if (state->start < start) {
660 prealloc = alloc_extent_state_atomic(prealloc);
662 err = split_state(tree, state, prealloc, start);
664 extent_io_tree_panic(tree, err);
669 if (state->end <= end) {
670 state = clear_state_bit(tree, state, &bits, wake,
677 * | ---- desired range ---- |
679 * We need to split the extent, and clear the bit
682 if (state->start <= end && state->end > end) {
683 prealloc = alloc_extent_state_atomic(prealloc);
685 err = split_state(tree, state, prealloc, end + 1);
687 extent_io_tree_panic(tree, err);
692 clear_state_bit(tree, prealloc, &bits, wake, changeset);
698 state = clear_state_bit(tree, state, &bits, wake, changeset);
700 if (last_end == (u64)-1)
702 start = last_end + 1;
703 if (start <= end && state && !need_resched())
709 spin_unlock(&tree->lock);
710 if (gfpflags_allow_blocking(mask))
715 spin_unlock(&tree->lock);
717 free_extent_state(prealloc);
723 static void wait_on_state(struct extent_io_tree *tree,
724 struct extent_state *state)
725 __releases(tree->lock)
726 __acquires(tree->lock)
729 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
730 spin_unlock(&tree->lock);
732 spin_lock(&tree->lock);
733 finish_wait(&state->wq, &wait);
737 * waits for one or more bits to clear on a range in the state tree.
738 * The range [start, end] is inclusive.
739 * The tree lock is taken by this function
741 static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
744 struct extent_state *state;
745 struct rb_node *node;
747 btrfs_debug_check_extent_io_range(tree, start, end);
749 spin_lock(&tree->lock);
753 * this search will find all the extents that end after
756 node = tree_search(tree, start);
761 state = rb_entry(node, struct extent_state, rb_node);
763 if (state->start > end)
766 if (state->state & bits) {
767 start = state->start;
768 refcount_inc(&state->refs);
769 wait_on_state(tree, state);
770 free_extent_state(state);
773 start = state->end + 1;
778 if (!cond_resched_lock(&tree->lock)) {
779 node = rb_next(node);
784 spin_unlock(&tree->lock);
787 static void set_state_bits(struct extent_io_tree *tree,
788 struct extent_state *state,
789 unsigned *bits, struct extent_changeset *changeset)
791 unsigned bits_to_set = *bits & ~EXTENT_CTLBITS;
794 if (tree->private_data && is_data_inode(tree->private_data))
795 btrfs_set_delalloc_extent(tree->private_data, state, bits);
797 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
798 u64 range = state->end - state->start + 1;
799 tree->dirty_bytes += range;
801 ret = add_extent_changeset(state, bits_to_set, changeset, 1);
803 state->state |= bits_to_set;
806 static void cache_state_if_flags(struct extent_state *state,
807 struct extent_state **cached_ptr,
810 if (cached_ptr && !(*cached_ptr)) {
811 if (!flags || (state->state & flags)) {
813 refcount_inc(&state->refs);
818 static void cache_state(struct extent_state *state,
819 struct extent_state **cached_ptr)
821 return cache_state_if_flags(state, cached_ptr,
822 EXTENT_IOBITS | EXTENT_BOUNDARY);
826 * set some bits on a range in the tree. This may require allocations or
827 * sleeping, so the gfp mask is used to indicate what is allowed.
829 * If any of the exclusive bits are set, this will fail with -EEXIST if some
830 * part of the range already has the desired bits set. The start of the
831 * existing range is returned in failed_start in this case.
833 * [start, end] is inclusive This takes the tree lock.
836 static int __must_check
837 __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
838 unsigned bits, unsigned exclusive_bits,
839 u64 *failed_start, struct extent_state **cached_state,
840 gfp_t mask, struct extent_changeset *changeset)
842 struct extent_state *state;
843 struct extent_state *prealloc = NULL;
844 struct rb_node *node;
846 struct rb_node *parent;
851 btrfs_debug_check_extent_io_range(tree, start, end);
853 bits |= EXTENT_FIRST_DELALLOC;
855 if (!prealloc && gfpflags_allow_blocking(mask)) {
857 * Don't care for allocation failure here because we might end
858 * up not needing the pre-allocated extent state at all, which
859 * is the case if we only have in the tree extent states that
860 * cover our input range and don't cover too any other range.
861 * If we end up needing a new extent state we allocate it later.
863 prealloc = alloc_extent_state(mask);
866 spin_lock(&tree->lock);
867 if (cached_state && *cached_state) {
868 state = *cached_state;
869 if (state->start <= start && state->end > start &&
870 extent_state_in_tree(state)) {
871 node = &state->rb_node;
876 * this search will find all the extents that end after
879 node = tree_search_for_insert(tree, start, &p, &parent);
881 prealloc = alloc_extent_state_atomic(prealloc);
883 err = insert_state(tree, prealloc, start, end,
884 &p, &parent, &bits, changeset);
886 extent_io_tree_panic(tree, err);
888 cache_state(prealloc, cached_state);
892 state = rb_entry(node, struct extent_state, rb_node);
894 last_start = state->start;
895 last_end = state->end;
898 * | ---- desired range ---- |
901 * Just lock what we found and keep going
903 if (state->start == start && state->end <= end) {
904 if (state->state & exclusive_bits) {
905 *failed_start = state->start;
910 set_state_bits(tree, state, &bits, changeset);
911 cache_state(state, cached_state);
912 merge_state(tree, state);
913 if (last_end == (u64)-1)
915 start = last_end + 1;
916 state = next_state(state);
917 if (start < end && state && state->start == start &&
924 * | ---- desired range ---- |
927 * | ------------- state -------------- |
929 * We need to split the extent we found, and may flip bits on
932 * If the extent we found extends past our
933 * range, we just split and search again. It'll get split
934 * again the next time though.
936 * If the extent we found is inside our range, we set the
939 if (state->start < start) {
940 if (state->state & exclusive_bits) {
941 *failed_start = start;
946 prealloc = alloc_extent_state_atomic(prealloc);
948 err = split_state(tree, state, prealloc, start);
950 extent_io_tree_panic(tree, err);
955 if (state->end <= end) {
956 set_state_bits(tree, state, &bits, changeset);
957 cache_state(state, cached_state);
958 merge_state(tree, state);
959 if (last_end == (u64)-1)
961 start = last_end + 1;
962 state = next_state(state);
963 if (start < end && state && state->start == start &&
970 * | ---- desired range ---- |
971 * | state | or | state |
973 * There's a hole, we need to insert something in it and
974 * ignore the extent we found.
976 if (state->start > start) {
978 if (end < last_start)
981 this_end = last_start - 1;
983 prealloc = alloc_extent_state_atomic(prealloc);
987 * Avoid to free 'prealloc' if it can be merged with
990 err = insert_state(tree, prealloc, start, this_end,
991 NULL, NULL, &bits, changeset);
993 extent_io_tree_panic(tree, err);
995 cache_state(prealloc, cached_state);
997 start = this_end + 1;
1001 * | ---- desired range ---- |
1003 * We need to split the extent, and set the bit
1006 if (state->start <= end && state->end > end) {
1007 if (state->state & exclusive_bits) {
1008 *failed_start = start;
1013 prealloc = alloc_extent_state_atomic(prealloc);
1015 err = split_state(tree, state, prealloc, end + 1);
1017 extent_io_tree_panic(tree, err);
1019 set_state_bits(tree, prealloc, &bits, changeset);
1020 cache_state(prealloc, cached_state);
1021 merge_state(tree, prealloc);
1029 spin_unlock(&tree->lock);
1030 if (gfpflags_allow_blocking(mask))
1035 spin_unlock(&tree->lock);
1037 free_extent_state(prealloc);
1043 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1044 unsigned bits, u64 * failed_start,
1045 struct extent_state **cached_state, gfp_t mask)
1047 return __set_extent_bit(tree, start, end, bits, 0, failed_start,
1048 cached_state, mask, NULL);
1053 * convert_extent_bit - convert all bits in a given range from one bit to
1055 * @tree: the io tree to search
1056 * @start: the start offset in bytes
1057 * @end: the end offset in bytes (inclusive)
1058 * @bits: the bits to set in this range
1059 * @clear_bits: the bits to clear in this range
1060 * @cached_state: state that we're going to cache
1062 * This will go through and set bits for the given range. If any states exist
1063 * already in this range they are set with the given bit and cleared of the
1064 * clear_bits. This is only meant to be used by things that are mergeable, ie
1065 * converting from say DELALLOC to DIRTY. This is not meant to be used with
1066 * boundary bits like LOCK.
1068 * All allocations are done with GFP_NOFS.
1070 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1071 unsigned bits, unsigned clear_bits,
1072 struct extent_state **cached_state)
1074 struct extent_state *state;
1075 struct extent_state *prealloc = NULL;
1076 struct rb_node *node;
1078 struct rb_node *parent;
1082 bool first_iteration = true;
1084 btrfs_debug_check_extent_io_range(tree, start, end);
1089 * Best effort, don't worry if extent state allocation fails
1090 * here for the first iteration. We might have a cached state
1091 * that matches exactly the target range, in which case no
1092 * extent state allocations are needed. We'll only know this
1093 * after locking the tree.
1095 prealloc = alloc_extent_state(GFP_NOFS);
1096 if (!prealloc && !first_iteration)
1100 spin_lock(&tree->lock);
1101 if (cached_state && *cached_state) {
1102 state = *cached_state;
1103 if (state->start <= start && state->end > start &&
1104 extent_state_in_tree(state)) {
1105 node = &state->rb_node;
1111 * this search will find all the extents that end after
1114 node = tree_search_for_insert(tree, start, &p, &parent);
1116 prealloc = alloc_extent_state_atomic(prealloc);
1121 err = insert_state(tree, prealloc, start, end,
1122 &p, &parent, &bits, NULL);
1124 extent_io_tree_panic(tree, err);
1125 cache_state(prealloc, cached_state);
1129 state = rb_entry(node, struct extent_state, rb_node);
1131 last_start = state->start;
1132 last_end = state->end;
1135 * | ---- desired range ---- |
1138 * Just lock what we found and keep going
1140 if (state->start == start && state->end <= end) {
1141 set_state_bits(tree, state, &bits, NULL);
1142 cache_state(state, cached_state);
1143 state = clear_state_bit(tree, state, &clear_bits, 0, NULL);
1144 if (last_end == (u64)-1)
1146 start = last_end + 1;
1147 if (start < end && state && state->start == start &&
1154 * | ---- desired range ---- |
1157 * | ------------- state -------------- |
1159 * We need to split the extent we found, and may flip bits on
1162 * If the extent we found extends past our
1163 * range, we just split and search again. It'll get split
1164 * again the next time though.
1166 * If the extent we found is inside our range, we set the
1167 * desired bit on it.
1169 if (state->start < start) {
1170 prealloc = alloc_extent_state_atomic(prealloc);
1175 err = split_state(tree, state, prealloc, start);
1177 extent_io_tree_panic(tree, err);
1181 if (state->end <= end) {
1182 set_state_bits(tree, state, &bits, NULL);
1183 cache_state(state, cached_state);
1184 state = clear_state_bit(tree, state, &clear_bits, 0,
1186 if (last_end == (u64)-1)
1188 start = last_end + 1;
1189 if (start < end && state && state->start == start &&
1196 * | ---- desired range ---- |
1197 * | state | or | state |
1199 * There's a hole, we need to insert something in it and
1200 * ignore the extent we found.
1202 if (state->start > start) {
1204 if (end < last_start)
1207 this_end = last_start - 1;
1209 prealloc = alloc_extent_state_atomic(prealloc);
1216 * Avoid to free 'prealloc' if it can be merged with
1219 err = insert_state(tree, prealloc, start, this_end,
1220 NULL, NULL, &bits, NULL);
1222 extent_io_tree_panic(tree, err);
1223 cache_state(prealloc, cached_state);
1225 start = this_end + 1;
1229 * | ---- desired range ---- |
1231 * We need to split the extent, and set the bit
1234 if (state->start <= end && state->end > end) {
1235 prealloc = alloc_extent_state_atomic(prealloc);
1241 err = split_state(tree, state, prealloc, end + 1);
1243 extent_io_tree_panic(tree, err);
1245 set_state_bits(tree, prealloc, &bits, NULL);
1246 cache_state(prealloc, cached_state);
1247 clear_state_bit(tree, prealloc, &clear_bits, 0, NULL);
1255 spin_unlock(&tree->lock);
1257 first_iteration = false;
1261 spin_unlock(&tree->lock);
1263 free_extent_state(prealloc);
1268 /* wrappers around set/clear extent bit */
1269 int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1270 unsigned bits, struct extent_changeset *changeset)
1273 * We don't support EXTENT_LOCKED yet, as current changeset will
1274 * record any bits changed, so for EXTENT_LOCKED case, it will
1275 * either fail with -EEXIST or changeset will record the whole
1278 BUG_ON(bits & EXTENT_LOCKED);
1280 return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL, GFP_NOFS,
1284 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1285 unsigned bits, int wake, int delete,
1286 struct extent_state **cached)
1288 return __clear_extent_bit(tree, start, end, bits, wake, delete,
1289 cached, GFP_NOFS, NULL);
1292 int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1293 unsigned bits, struct extent_changeset *changeset)
1296 * Don't support EXTENT_LOCKED case, same reason as
1297 * set_record_extent_bits().
1299 BUG_ON(bits & EXTENT_LOCKED);
1301 return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, GFP_NOFS,
1306 * either insert or lock state struct between start and end use mask to tell
1307 * us if waiting is desired.
1309 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1310 struct extent_state **cached_state)
1316 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED,
1317 EXTENT_LOCKED, &failed_start,
1318 cached_state, GFP_NOFS, NULL);
1319 if (err == -EEXIST) {
1320 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1321 start = failed_start;
1324 WARN_ON(start > end);
1329 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1334 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1335 &failed_start, NULL, GFP_NOFS, NULL);
1336 if (err == -EEXIST) {
1337 if (failed_start > start)
1338 clear_extent_bit(tree, start, failed_start - 1,
1339 EXTENT_LOCKED, 1, 0, NULL);
1345 void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
1347 unsigned long index = start >> PAGE_SHIFT;
1348 unsigned long end_index = end >> PAGE_SHIFT;
1351 while (index <= end_index) {
1352 page = find_get_page(inode->i_mapping, index);
1353 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1354 clear_page_dirty_for_io(page);
1360 void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
1362 unsigned long index = start >> PAGE_SHIFT;
1363 unsigned long end_index = end >> PAGE_SHIFT;
1366 while (index <= end_index) {
1367 page = find_get_page(inode->i_mapping, index);
1368 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1369 __set_page_dirty_nobuffers(page);
1370 account_page_redirty(page);
1376 /* find the first state struct with 'bits' set after 'start', and
1377 * return it. tree->lock must be held. NULL will returned if
1378 * nothing was found after 'start'
1380 static struct extent_state *
1381 find_first_extent_bit_state(struct extent_io_tree *tree,
1382 u64 start, unsigned bits)
1384 struct rb_node *node;
1385 struct extent_state *state;
1388 * this search will find all the extents that end after
1391 node = tree_search(tree, start);
1396 state = rb_entry(node, struct extent_state, rb_node);
1397 if (state->end >= start && (state->state & bits))
1400 node = rb_next(node);
1409 * find the first offset in the io tree with 'bits' set. zero is
1410 * returned if we find something, and *start_ret and *end_ret are
1411 * set to reflect the state struct that was found.
1413 * If nothing was found, 1 is returned. If found something, return 0.
1415 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1416 u64 *start_ret, u64 *end_ret, unsigned bits,
1417 struct extent_state **cached_state)
1419 struct extent_state *state;
1422 spin_lock(&tree->lock);
1423 if (cached_state && *cached_state) {
1424 state = *cached_state;
1425 if (state->end == start - 1 && extent_state_in_tree(state)) {
1426 while ((state = next_state(state)) != NULL) {
1427 if (state->state & bits)
1430 free_extent_state(*cached_state);
1431 *cached_state = NULL;
1434 free_extent_state(*cached_state);
1435 *cached_state = NULL;
1438 state = find_first_extent_bit_state(tree, start, bits);
1441 cache_state_if_flags(state, cached_state, 0);
1442 *start_ret = state->start;
1443 *end_ret = state->end;
1447 spin_unlock(&tree->lock);
1452 * find a contiguous range of bytes in the file marked as delalloc, not
1453 * more than 'max_bytes'. start and end are used to return the range,
1455 * 1 is returned if we find something, 0 if nothing was in the tree
1457 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1458 u64 *start, u64 *end, u64 max_bytes,
1459 struct extent_state **cached_state)
1461 struct rb_node *node;
1462 struct extent_state *state;
1463 u64 cur_start = *start;
1465 u64 total_bytes = 0;
1467 spin_lock(&tree->lock);
1470 * this search will find all the extents that end after
1473 node = tree_search(tree, cur_start);
1481 state = rb_entry(node, struct extent_state, rb_node);
1482 if (found && (state->start != cur_start ||
1483 (state->state & EXTENT_BOUNDARY))) {
1486 if (!(state->state & EXTENT_DELALLOC)) {
1492 *start = state->start;
1493 *cached_state = state;
1494 refcount_inc(&state->refs);
1498 cur_start = state->end + 1;
1499 node = rb_next(node);
1500 total_bytes += state->end - state->start + 1;
1501 if (total_bytes >= max_bytes)
1507 spin_unlock(&tree->lock);
1511 static int __process_pages_contig(struct address_space *mapping,
1512 struct page *locked_page,
1513 pgoff_t start_index, pgoff_t end_index,
1514 unsigned long page_ops, pgoff_t *index_ret);
1516 static noinline void __unlock_for_delalloc(struct inode *inode,
1517 struct page *locked_page,
1520 unsigned long index = start >> PAGE_SHIFT;
1521 unsigned long end_index = end >> PAGE_SHIFT;
1523 ASSERT(locked_page);
1524 if (index == locked_page->index && end_index == index)
1527 __process_pages_contig(inode->i_mapping, locked_page, index, end_index,
1531 static noinline int lock_delalloc_pages(struct inode *inode,
1532 struct page *locked_page,
1536 unsigned long index = delalloc_start >> PAGE_SHIFT;
1537 unsigned long index_ret = index;
1538 unsigned long end_index = delalloc_end >> PAGE_SHIFT;
1541 ASSERT(locked_page);
1542 if (index == locked_page->index && index == end_index)
1545 ret = __process_pages_contig(inode->i_mapping, locked_page, index,
1546 end_index, PAGE_LOCK, &index_ret);
1548 __unlock_for_delalloc(inode, locked_page, delalloc_start,
1549 (u64)index_ret << PAGE_SHIFT);
1554 * find a contiguous range of bytes in the file marked as delalloc, not
1555 * more than 'max_bytes'. start and end are used to return the range,
1557 * 1 is returned if we find something, 0 if nothing was in the tree
1559 static noinline_for_stack u64 find_lock_delalloc_range(struct inode *inode,
1560 struct extent_io_tree *tree,
1561 struct page *locked_page, u64 *start,
1562 u64 *end, u64 max_bytes)
1567 struct extent_state *cached_state = NULL;
1572 /* step one, find a bunch of delalloc bytes starting at start */
1573 delalloc_start = *start;
1575 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1576 max_bytes, &cached_state);
1577 if (!found || delalloc_end <= *start) {
1578 *start = delalloc_start;
1579 *end = delalloc_end;
1580 free_extent_state(cached_state);
1585 * start comes from the offset of locked_page. We have to lock
1586 * pages in order, so we can't process delalloc bytes before
1589 if (delalloc_start < *start)
1590 delalloc_start = *start;
1593 * make sure to limit the number of pages we try to lock down
1595 if (delalloc_end + 1 - delalloc_start > max_bytes)
1596 delalloc_end = delalloc_start + max_bytes - 1;
1598 /* step two, lock all the pages after the page that has start */
1599 ret = lock_delalloc_pages(inode, locked_page,
1600 delalloc_start, delalloc_end);
1601 if (ret == -EAGAIN) {
1602 /* some of the pages are gone, lets avoid looping by
1603 * shortening the size of the delalloc range we're searching
1605 free_extent_state(cached_state);
1606 cached_state = NULL;
1608 max_bytes = PAGE_SIZE;
1616 BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */
1618 /* step three, lock the state bits for the whole range */
1619 lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state);
1621 /* then test to make sure it is all still delalloc */
1622 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1623 EXTENT_DELALLOC, 1, cached_state);
1625 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1627 __unlock_for_delalloc(inode, locked_page,
1628 delalloc_start, delalloc_end);
1632 free_extent_state(cached_state);
1633 *start = delalloc_start;
1634 *end = delalloc_end;
1639 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1640 u64 btrfs_find_lock_delalloc_range(struct inode *inode,
1641 struct extent_io_tree *tree,
1642 struct page *locked_page, u64 *start,
1643 u64 *end, u64 max_bytes)
1645 return find_lock_delalloc_range(inode, tree, locked_page, start, end,
1650 static int __process_pages_contig(struct address_space *mapping,
1651 struct page *locked_page,
1652 pgoff_t start_index, pgoff_t end_index,
1653 unsigned long page_ops, pgoff_t *index_ret)
1655 unsigned long nr_pages = end_index - start_index + 1;
1656 unsigned long pages_locked = 0;
1657 pgoff_t index = start_index;
1658 struct page *pages[16];
1663 if (page_ops & PAGE_LOCK) {
1664 ASSERT(page_ops == PAGE_LOCK);
1665 ASSERT(index_ret && *index_ret == start_index);
1668 if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0)
1669 mapping_set_error(mapping, -EIO);
1671 while (nr_pages > 0) {
1672 ret = find_get_pages_contig(mapping, index,
1673 min_t(unsigned long,
1674 nr_pages, ARRAY_SIZE(pages)), pages);
1677 * Only if we're going to lock these pages,
1678 * can we find nothing at @index.
1680 ASSERT(page_ops & PAGE_LOCK);
1685 for (i = 0; i < ret; i++) {
1686 if (page_ops & PAGE_SET_PRIVATE2)
1687 SetPagePrivate2(pages[i]);
1689 if (pages[i] == locked_page) {
1694 if (page_ops & PAGE_CLEAR_DIRTY)
1695 clear_page_dirty_for_io(pages[i]);
1696 if (page_ops & PAGE_SET_WRITEBACK)
1697 set_page_writeback(pages[i]);
1698 if (page_ops & PAGE_SET_ERROR)
1699 SetPageError(pages[i]);
1700 if (page_ops & PAGE_END_WRITEBACK)
1701 end_page_writeback(pages[i]);
1702 if (page_ops & PAGE_UNLOCK)
1703 unlock_page(pages[i]);
1704 if (page_ops & PAGE_LOCK) {
1705 lock_page(pages[i]);
1706 if (!PageDirty(pages[i]) ||
1707 pages[i]->mapping != mapping) {
1708 unlock_page(pages[i]);
1722 if (err && index_ret)
1723 *index_ret = start_index + pages_locked - 1;
1727 void extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end,
1728 u64 delalloc_end, struct page *locked_page,
1729 unsigned clear_bits,
1730 unsigned long page_ops)
1732 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, clear_bits, 1, 0,
1735 __process_pages_contig(inode->i_mapping, locked_page,
1736 start >> PAGE_SHIFT, end >> PAGE_SHIFT,
1741 * count the number of bytes in the tree that have a given bit(s)
1742 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1743 * cached. The total number found is returned.
1745 u64 count_range_bits(struct extent_io_tree *tree,
1746 u64 *start, u64 search_end, u64 max_bytes,
1747 unsigned bits, int contig)
1749 struct rb_node *node;
1750 struct extent_state *state;
1751 u64 cur_start = *start;
1752 u64 total_bytes = 0;
1756 if (WARN_ON(search_end <= cur_start))
1759 spin_lock(&tree->lock);
1760 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1761 total_bytes = tree->dirty_bytes;
1765 * this search will find all the extents that end after
1768 node = tree_search(tree, cur_start);
1773 state = rb_entry(node, struct extent_state, rb_node);
1774 if (state->start > search_end)
1776 if (contig && found && state->start > last + 1)
1778 if (state->end >= cur_start && (state->state & bits) == bits) {
1779 total_bytes += min(search_end, state->end) + 1 -
1780 max(cur_start, state->start);
1781 if (total_bytes >= max_bytes)
1784 *start = max(cur_start, state->start);
1788 } else if (contig && found) {
1791 node = rb_next(node);
1796 spin_unlock(&tree->lock);
1801 * set the private field for a given byte offset in the tree. If there isn't
1802 * an extent_state there already, this does nothing.
1804 static noinline int set_state_failrec(struct extent_io_tree *tree, u64 start,
1805 struct io_failure_record *failrec)
1807 struct rb_node *node;
1808 struct extent_state *state;
1811 spin_lock(&tree->lock);
1813 * this search will find all the extents that end after
1816 node = tree_search(tree, start);
1821 state = rb_entry(node, struct extent_state, rb_node);
1822 if (state->start != start) {
1826 state->failrec = failrec;
1828 spin_unlock(&tree->lock);
1832 static noinline int get_state_failrec(struct extent_io_tree *tree, u64 start,
1833 struct io_failure_record **failrec)
1835 struct rb_node *node;
1836 struct extent_state *state;
1839 spin_lock(&tree->lock);
1841 * this search will find all the extents that end after
1844 node = tree_search(tree, start);
1849 state = rb_entry(node, struct extent_state, rb_node);
1850 if (state->start != start) {
1854 *failrec = state->failrec;
1856 spin_unlock(&tree->lock);
1861 * searches a range in the state tree for a given mask.
1862 * If 'filled' == 1, this returns 1 only if every extent in the tree
1863 * has the bits set. Otherwise, 1 is returned if any bit in the
1864 * range is found set.
1866 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1867 unsigned bits, int filled, struct extent_state *cached)
1869 struct extent_state *state = NULL;
1870 struct rb_node *node;
1873 spin_lock(&tree->lock);
1874 if (cached && extent_state_in_tree(cached) && cached->start <= start &&
1875 cached->end > start)
1876 node = &cached->rb_node;
1878 node = tree_search(tree, start);
1879 while (node && start <= end) {
1880 state = rb_entry(node, struct extent_state, rb_node);
1882 if (filled && state->start > start) {
1887 if (state->start > end)
1890 if (state->state & bits) {
1894 } else if (filled) {
1899 if (state->end == (u64)-1)
1902 start = state->end + 1;
1905 node = rb_next(node);
1912 spin_unlock(&tree->lock);
1917 * helper function to set a given page up to date if all the
1918 * extents in the tree for that page are up to date
1920 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
1922 u64 start = page_offset(page);
1923 u64 end = start + PAGE_SIZE - 1;
1924 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1925 SetPageUptodate(page);
1928 int free_io_failure(struct extent_io_tree *failure_tree,
1929 struct extent_io_tree *io_tree,
1930 struct io_failure_record *rec)
1935 set_state_failrec(failure_tree, rec->start, NULL);
1936 ret = clear_extent_bits(failure_tree, rec->start,
1937 rec->start + rec->len - 1,
1938 EXTENT_LOCKED | EXTENT_DIRTY);
1942 ret = clear_extent_bits(io_tree, rec->start,
1943 rec->start + rec->len - 1,
1953 * this bypasses the standard btrfs submit functions deliberately, as
1954 * the standard behavior is to write all copies in a raid setup. here we only
1955 * want to write the one bad copy. so we do the mapping for ourselves and issue
1956 * submit_bio directly.
1957 * to avoid any synchronization issues, wait for the data after writing, which
1958 * actually prevents the read that triggered the error from finishing.
1959 * currently, there can be no more than two copies of every data bit. thus,
1960 * exactly one rewrite is required.
1962 int repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start,
1963 u64 length, u64 logical, struct page *page,
1964 unsigned int pg_offset, int mirror_num)
1967 struct btrfs_device *dev;
1970 struct btrfs_bio *bbio = NULL;
1973 ASSERT(!(fs_info->sb->s_flags & SB_RDONLY));
1974 BUG_ON(!mirror_num);
1976 bio = btrfs_io_bio_alloc(1);
1977 bio->bi_iter.bi_size = 0;
1978 map_length = length;
1981 * Avoid races with device replace and make sure our bbio has devices
1982 * associated to its stripes that don't go away while we are doing the
1983 * read repair operation.
1985 btrfs_bio_counter_inc_blocked(fs_info);
1986 if (btrfs_is_parity_mirror(fs_info, logical, length)) {
1988 * Note that we don't use BTRFS_MAP_WRITE because it's supposed
1989 * to update all raid stripes, but here we just want to correct
1990 * bad stripe, thus BTRFS_MAP_READ is abused to only get the bad
1991 * stripe's dev and sector.
1993 ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, logical,
1994 &map_length, &bbio, 0);
1996 btrfs_bio_counter_dec(fs_info);
2000 ASSERT(bbio->mirror_num == 1);
2002 ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical,
2003 &map_length, &bbio, mirror_num);
2005 btrfs_bio_counter_dec(fs_info);
2009 BUG_ON(mirror_num != bbio->mirror_num);
2012 sector = bbio->stripes[bbio->mirror_num - 1].physical >> 9;
2013 bio->bi_iter.bi_sector = sector;
2014 dev = bbio->stripes[bbio->mirror_num - 1].dev;
2015 btrfs_put_bbio(bbio);
2016 if (!dev || !dev->bdev ||
2017 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) {
2018 btrfs_bio_counter_dec(fs_info);
2022 bio_set_dev(bio, dev->bdev);
2023 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC;
2024 bio_add_page(bio, page, length, pg_offset);
2026 if (btrfsic_submit_bio_wait(bio)) {
2027 /* try to remap that extent elsewhere? */
2028 btrfs_bio_counter_dec(fs_info);
2030 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
2034 btrfs_info_rl_in_rcu(fs_info,
2035 "read error corrected: ino %llu off %llu (dev %s sector %llu)",
2037 rcu_str_deref(dev->name), sector);
2038 btrfs_bio_counter_dec(fs_info);
2043 int repair_eb_io_failure(struct btrfs_fs_info *fs_info,
2044 struct extent_buffer *eb, int mirror_num)
2046 u64 start = eb->start;
2047 int i, num_pages = num_extent_pages(eb);
2050 if (sb_rdonly(fs_info->sb))
2053 for (i = 0; i < num_pages; i++) {
2054 struct page *p = eb->pages[i];
2056 ret = repair_io_failure(fs_info, 0, start, PAGE_SIZE, start, p,
2057 start - page_offset(p), mirror_num);
2067 * each time an IO finishes, we do a fast check in the IO failure tree
2068 * to see if we need to process or clean up an io_failure_record
2070 int clean_io_failure(struct btrfs_fs_info *fs_info,
2071 struct extent_io_tree *failure_tree,
2072 struct extent_io_tree *io_tree, u64 start,
2073 struct page *page, u64 ino, unsigned int pg_offset)
2076 struct io_failure_record *failrec;
2077 struct extent_state *state;
2082 ret = count_range_bits(failure_tree, &private, (u64)-1, 1,
2087 ret = get_state_failrec(failure_tree, start, &failrec);
2091 BUG_ON(!failrec->this_mirror);
2093 if (failrec->in_validation) {
2094 /* there was no real error, just free the record */
2095 btrfs_debug(fs_info,
2096 "clean_io_failure: freeing dummy error at %llu",
2100 if (sb_rdonly(fs_info->sb))
2103 spin_lock(&io_tree->lock);
2104 state = find_first_extent_bit_state(io_tree,
2107 spin_unlock(&io_tree->lock);
2109 if (state && state->start <= failrec->start &&
2110 state->end >= failrec->start + failrec->len - 1) {
2111 num_copies = btrfs_num_copies(fs_info, failrec->logical,
2113 if (num_copies > 1) {
2114 repair_io_failure(fs_info, ino, start, failrec->len,
2115 failrec->logical, page, pg_offset,
2116 failrec->failed_mirror);
2121 free_io_failure(failure_tree, io_tree, failrec);
2127 * Can be called when
2128 * - hold extent lock
2129 * - under ordered extent
2130 * - the inode is freeing
2132 void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start, u64 end)
2134 struct extent_io_tree *failure_tree = &inode->io_failure_tree;
2135 struct io_failure_record *failrec;
2136 struct extent_state *state, *next;
2138 if (RB_EMPTY_ROOT(&failure_tree->state))
2141 spin_lock(&failure_tree->lock);
2142 state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY);
2144 if (state->start > end)
2147 ASSERT(state->end <= end);
2149 next = next_state(state);
2151 failrec = state->failrec;
2152 free_extent_state(state);
2157 spin_unlock(&failure_tree->lock);
2160 int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end,
2161 struct io_failure_record **failrec_ret)
2163 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2164 struct io_failure_record *failrec;
2165 struct extent_map *em;
2166 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2167 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2168 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2172 ret = get_state_failrec(failure_tree, start, &failrec);
2174 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2178 failrec->start = start;
2179 failrec->len = end - start + 1;
2180 failrec->this_mirror = 0;
2181 failrec->bio_flags = 0;
2182 failrec->in_validation = 0;
2184 read_lock(&em_tree->lock);
2185 em = lookup_extent_mapping(em_tree, start, failrec->len);
2187 read_unlock(&em_tree->lock);
2192 if (em->start > start || em->start + em->len <= start) {
2193 free_extent_map(em);
2196 read_unlock(&em_tree->lock);
2202 logical = start - em->start;
2203 logical = em->block_start + logical;
2204 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2205 logical = em->block_start;
2206 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2207 extent_set_compress_type(&failrec->bio_flags,
2211 btrfs_debug(fs_info,
2212 "Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu",
2213 logical, start, failrec->len);
2215 failrec->logical = logical;
2216 free_extent_map(em);
2218 /* set the bits in the private failure tree */
2219 ret = set_extent_bits(failure_tree, start, end,
2220 EXTENT_LOCKED | EXTENT_DIRTY);
2222 ret = set_state_failrec(failure_tree, start, failrec);
2223 /* set the bits in the inode's tree */
2225 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED);
2231 btrfs_debug(fs_info,
2232 "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d",
2233 failrec->logical, failrec->start, failrec->len,
2234 failrec->in_validation);
2236 * when data can be on disk more than twice, add to failrec here
2237 * (e.g. with a list for failed_mirror) to make
2238 * clean_io_failure() clean all those errors at once.
2242 *failrec_ret = failrec;
2247 bool btrfs_check_repairable(struct inode *inode, unsigned failed_bio_pages,
2248 struct io_failure_record *failrec, int failed_mirror)
2250 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2253 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
2254 if (num_copies == 1) {
2256 * we only have a single copy of the data, so don't bother with
2257 * all the retry and error correction code that follows. no
2258 * matter what the error is, it is very likely to persist.
2260 btrfs_debug(fs_info,
2261 "Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
2262 num_copies, failrec->this_mirror, failed_mirror);
2267 * there are two premises:
2268 * a) deliver good data to the caller
2269 * b) correct the bad sectors on disk
2271 if (failed_bio_pages > 1) {
2273 * to fulfill b), we need to know the exact failing sectors, as
2274 * we don't want to rewrite any more than the failed ones. thus,
2275 * we need separate read requests for the failed bio
2277 * if the following BUG_ON triggers, our validation request got
2278 * merged. we need separate requests for our algorithm to work.
2280 BUG_ON(failrec->in_validation);
2281 failrec->in_validation = 1;
2282 failrec->this_mirror = failed_mirror;
2285 * we're ready to fulfill a) and b) alongside. get a good copy
2286 * of the failed sector and if we succeed, we have setup
2287 * everything for repair_io_failure to do the rest for us.
2289 if (failrec->in_validation) {
2290 BUG_ON(failrec->this_mirror != failed_mirror);
2291 failrec->in_validation = 0;
2292 failrec->this_mirror = 0;
2294 failrec->failed_mirror = failed_mirror;
2295 failrec->this_mirror++;
2296 if (failrec->this_mirror == failed_mirror)
2297 failrec->this_mirror++;
2300 if (failrec->this_mirror > num_copies) {
2301 btrfs_debug(fs_info,
2302 "Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
2303 num_copies, failrec->this_mirror, failed_mirror);
2311 struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio,
2312 struct io_failure_record *failrec,
2313 struct page *page, int pg_offset, int icsum,
2314 bio_end_io_t *endio_func, void *data)
2316 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2318 struct btrfs_io_bio *btrfs_failed_bio;
2319 struct btrfs_io_bio *btrfs_bio;
2321 bio = btrfs_io_bio_alloc(1);
2322 bio->bi_end_io = endio_func;
2323 bio->bi_iter.bi_sector = failrec->logical >> 9;
2324 bio_set_dev(bio, fs_info->fs_devices->latest_bdev);
2325 bio->bi_iter.bi_size = 0;
2326 bio->bi_private = data;
2328 btrfs_failed_bio = btrfs_io_bio(failed_bio);
2329 if (btrfs_failed_bio->csum) {
2330 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
2332 btrfs_bio = btrfs_io_bio(bio);
2333 btrfs_bio->csum = btrfs_bio->csum_inline;
2335 memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + icsum,
2339 bio_add_page(bio, page, failrec->len, pg_offset);
2345 * this is a generic handler for readpage errors (default
2346 * readpage_io_failed_hook). if other copies exist, read those and write back
2347 * good data to the failed position. does not investigate in remapping the
2348 * failed extent elsewhere, hoping the device will be smart enough to do this as
2352 static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset,
2353 struct page *page, u64 start, u64 end,
2356 struct io_failure_record *failrec;
2357 struct inode *inode = page->mapping->host;
2358 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2359 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2362 blk_status_t status;
2364 unsigned failed_bio_pages = bio_pages_all(failed_bio);
2366 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
2368 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
2372 if (!btrfs_check_repairable(inode, failed_bio_pages, failrec,
2374 free_io_failure(failure_tree, tree, failrec);
2378 if (failed_bio_pages > 1)
2379 read_mode |= REQ_FAILFAST_DEV;
2381 phy_offset >>= inode->i_sb->s_blocksize_bits;
2382 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
2383 start - page_offset(page),
2384 (int)phy_offset, failed_bio->bi_end_io,
2386 bio->bi_opf = REQ_OP_READ | read_mode;
2388 btrfs_debug(btrfs_sb(inode->i_sb),
2389 "Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d",
2390 read_mode, failrec->this_mirror, failrec->in_validation);
2392 status = tree->ops->submit_bio_hook(tree->private_data, bio, failrec->this_mirror,
2393 failrec->bio_flags, 0);
2395 free_io_failure(failure_tree, tree, failrec);
2397 ret = blk_status_to_errno(status);
2403 /* lots and lots of room for performance fixes in the end_bio funcs */
2405 void end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2407 int uptodate = (err == 0);
2410 btrfs_writepage_endio_finish_ordered(page, start, end, NULL, uptodate);
2413 ClearPageUptodate(page);
2415 ret = err < 0 ? err : -EIO;
2416 mapping_set_error(page->mapping, ret);
2421 * after a writepage IO is done, we need to:
2422 * clear the uptodate bits on error
2423 * clear the writeback bits in the extent tree for this IO
2424 * end_page_writeback if the page has no more pending IO
2426 * Scheduling is not allowed, so the extent state tree is expected
2427 * to have one and only one object corresponding to this IO.
2429 static void end_bio_extent_writepage(struct bio *bio)
2431 int error = blk_status_to_errno(bio->bi_status);
2432 struct bio_vec *bvec;
2437 ASSERT(!bio_flagged(bio, BIO_CLONED));
2438 bio_for_each_segment_all(bvec, bio, i) {
2439 struct page *page = bvec->bv_page;
2440 struct inode *inode = page->mapping->host;
2441 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2443 /* We always issue full-page reads, but if some block
2444 * in a page fails to read, blk_update_request() will
2445 * advance bv_offset and adjust bv_len to compensate.
2446 * Print a warning for nonzero offsets, and an error
2447 * if they don't add up to a full page. */
2448 if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) {
2449 if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE)
2451 "partial page write in btrfs with offset %u and length %u",
2452 bvec->bv_offset, bvec->bv_len);
2455 "incomplete page write in btrfs with offset %u and length %u",
2456 bvec->bv_offset, bvec->bv_len);
2459 start = page_offset(page);
2460 end = start + bvec->bv_offset + bvec->bv_len - 1;
2462 end_extent_writepage(page, error, start, end);
2463 end_page_writeback(page);
2470 endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len,
2473 struct extent_state *cached = NULL;
2474 u64 end = start + len - 1;
2476 if (uptodate && tree->track_uptodate)
2477 set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC);
2478 unlock_extent_cached_atomic(tree, start, end, &cached);
2482 * after a readpage IO is done, we need to:
2483 * clear the uptodate bits on error
2484 * set the uptodate bits if things worked
2485 * set the page up to date if all extents in the tree are uptodate
2486 * clear the lock bit in the extent tree
2487 * unlock the page if there are no other extents locked for it
2489 * Scheduling is not allowed, so the extent state tree is expected
2490 * to have one and only one object corresponding to this IO.
2492 static void end_bio_extent_readpage(struct bio *bio)
2494 struct bio_vec *bvec;
2495 int uptodate = !bio->bi_status;
2496 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
2497 struct extent_io_tree *tree, *failure_tree;
2502 u64 extent_start = 0;
2508 ASSERT(!bio_flagged(bio, BIO_CLONED));
2509 bio_for_each_segment_all(bvec, bio, i) {
2510 struct page *page = bvec->bv_page;
2511 struct inode *inode = page->mapping->host;
2512 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2514 btrfs_debug(fs_info,
2515 "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u",
2516 (u64)bio->bi_iter.bi_sector, bio->bi_status,
2517 io_bio->mirror_num);
2518 tree = &BTRFS_I(inode)->io_tree;
2519 failure_tree = &BTRFS_I(inode)->io_failure_tree;
2521 /* We always issue full-page reads, but if some block
2522 * in a page fails to read, blk_update_request() will
2523 * advance bv_offset and adjust bv_len to compensate.
2524 * Print a warning for nonzero offsets, and an error
2525 * if they don't add up to a full page. */
2526 if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) {
2527 if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE)
2529 "partial page read in btrfs with offset %u and length %u",
2530 bvec->bv_offset, bvec->bv_len);
2533 "incomplete page read in btrfs with offset %u and length %u",
2534 bvec->bv_offset, bvec->bv_len);
2537 start = page_offset(page);
2538 end = start + bvec->bv_offset + bvec->bv_len - 1;
2541 mirror = io_bio->mirror_num;
2542 if (likely(uptodate && tree->ops)) {
2543 ret = tree->ops->readpage_end_io_hook(io_bio, offset,
2549 clean_io_failure(BTRFS_I(inode)->root->fs_info,
2550 failure_tree, tree, start,
2552 btrfs_ino(BTRFS_I(inode)), 0);
2555 if (likely(uptodate))
2559 ret = tree->ops->readpage_io_failed_hook(page, mirror);
2560 if (ret == -EAGAIN) {
2562 * Data inode's readpage_io_failed_hook() always
2565 * The generic bio_readpage_error handles errors
2566 * the following way: If possible, new read
2567 * requests are created and submitted and will
2568 * end up in end_bio_extent_readpage as well (if
2569 * we're lucky, not in the !uptodate case). In
2570 * that case it returns 0 and we just go on with
2571 * the next page in our bio. If it can't handle
2572 * the error it will return -EIO and we remain
2573 * responsible for that page.
2575 ret = bio_readpage_error(bio, offset, page,
2576 start, end, mirror);
2578 uptodate = !bio->bi_status;
2585 * metadata's readpage_io_failed_hook() always returns
2586 * -EIO and fixes nothing. -EIO is also returned if
2587 * data inode error could not be fixed.
2589 ASSERT(ret == -EIO);
2592 if (likely(uptodate)) {
2593 loff_t i_size = i_size_read(inode);
2594 pgoff_t end_index = i_size >> PAGE_SHIFT;
2597 /* Zero out the end if this page straddles i_size */
2598 off = i_size & (PAGE_SIZE-1);
2599 if (page->index == end_index && off)
2600 zero_user_segment(page, off, PAGE_SIZE);
2601 SetPageUptodate(page);
2603 ClearPageUptodate(page);
2609 if (unlikely(!uptodate)) {
2611 endio_readpage_release_extent(tree,
2617 endio_readpage_release_extent(tree, start,
2618 end - start + 1, 0);
2619 } else if (!extent_len) {
2620 extent_start = start;
2621 extent_len = end + 1 - start;
2622 } else if (extent_start + extent_len == start) {
2623 extent_len += end + 1 - start;
2625 endio_readpage_release_extent(tree, extent_start,
2626 extent_len, uptodate);
2627 extent_start = start;
2628 extent_len = end + 1 - start;
2633 endio_readpage_release_extent(tree, extent_start, extent_len,
2636 io_bio->end_io(io_bio, blk_status_to_errno(bio->bi_status));
2641 * Initialize the members up to but not including 'bio'. Use after allocating a
2642 * new bio by bio_alloc_bioset as it does not initialize the bytes outside of
2643 * 'bio' because use of __GFP_ZERO is not supported.
2645 static inline void btrfs_io_bio_init(struct btrfs_io_bio *btrfs_bio)
2647 memset(btrfs_bio, 0, offsetof(struct btrfs_io_bio, bio));
2651 * The following helpers allocate a bio. As it's backed by a bioset, it'll
2652 * never fail. We're returning a bio right now but you can call btrfs_io_bio
2653 * for the appropriate container_of magic
2655 struct bio *btrfs_bio_alloc(struct block_device *bdev, u64 first_byte)
2659 bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, &btrfs_bioset);
2660 bio_set_dev(bio, bdev);
2661 bio->bi_iter.bi_sector = first_byte >> 9;
2662 btrfs_io_bio_init(btrfs_io_bio(bio));
2666 struct bio *btrfs_bio_clone(struct bio *bio)
2668 struct btrfs_io_bio *btrfs_bio;
2671 /* Bio allocation backed by a bioset does not fail */
2672 new = bio_clone_fast(bio, GFP_NOFS, &btrfs_bioset);
2673 btrfs_bio = btrfs_io_bio(new);
2674 btrfs_io_bio_init(btrfs_bio);
2675 btrfs_bio->iter = bio->bi_iter;
2679 struct bio *btrfs_io_bio_alloc(unsigned int nr_iovecs)
2683 /* Bio allocation backed by a bioset does not fail */
2684 bio = bio_alloc_bioset(GFP_NOFS, nr_iovecs, &btrfs_bioset);
2685 btrfs_io_bio_init(btrfs_io_bio(bio));
2689 struct bio *btrfs_bio_clone_partial(struct bio *orig, int offset, int size)
2692 struct btrfs_io_bio *btrfs_bio;
2694 /* this will never fail when it's backed by a bioset */
2695 bio = bio_clone_fast(orig, GFP_NOFS, &btrfs_bioset);
2698 btrfs_bio = btrfs_io_bio(bio);
2699 btrfs_io_bio_init(btrfs_bio);
2701 bio_trim(bio, offset >> 9, size >> 9);
2702 btrfs_bio->iter = bio->bi_iter;
2706 static int __must_check submit_one_bio(struct bio *bio, int mirror_num,
2707 unsigned long bio_flags)
2709 blk_status_t ret = 0;
2710 struct bio_vec *bvec = bio_last_bvec_all(bio);
2711 struct page *page = bvec->bv_page;
2712 struct extent_io_tree *tree = bio->bi_private;
2715 start = page_offset(page) + bvec->bv_offset;
2717 bio->bi_private = NULL;
2720 ret = tree->ops->submit_bio_hook(tree->private_data, bio,
2721 mirror_num, bio_flags, start);
2723 btrfsic_submit_bio(bio);
2725 return blk_status_to_errno(ret);
2729 * @opf: bio REQ_OP_* and REQ_* flags as one value
2730 * @tree: tree so we can call our merge_bio hook
2731 * @wbc: optional writeback control for io accounting
2732 * @page: page to add to the bio
2733 * @pg_offset: offset of the new bio or to check whether we are adding
2734 * a contiguous page to the previous one
2735 * @size: portion of page that we want to write
2736 * @offset: starting offset in the page
2737 * @bdev: attach newly created bios to this bdev
2738 * @bio_ret: must be valid pointer, newly allocated bio will be stored there
2739 * @end_io_func: end_io callback for new bio
2740 * @mirror_num: desired mirror to read/write
2741 * @prev_bio_flags: flags of previous bio to see if we can merge the current one
2742 * @bio_flags: flags of the current bio to see if we can merge them
2744 static int submit_extent_page(unsigned int opf, struct extent_io_tree *tree,
2745 struct writeback_control *wbc,
2746 struct page *page, u64 offset,
2747 size_t size, unsigned long pg_offset,
2748 struct block_device *bdev,
2749 struct bio **bio_ret,
2750 bio_end_io_t end_io_func,
2752 unsigned long prev_bio_flags,
2753 unsigned long bio_flags,
2754 bool force_bio_submit)
2758 size_t page_size = min_t(size_t, size, PAGE_SIZE);
2759 sector_t sector = offset >> 9;
2765 bool can_merge = true;
2768 if (prev_bio_flags & EXTENT_BIO_COMPRESSED)
2769 contig = bio->bi_iter.bi_sector == sector;
2771 contig = bio_end_sector(bio) == sector;
2773 if (tree->ops && btrfs_merge_bio_hook(page, offset, page_size,
2777 if (prev_bio_flags != bio_flags || !contig || !can_merge ||
2779 bio_add_page(bio, page, page_size, pg_offset) < page_size) {
2780 ret = submit_one_bio(bio, mirror_num, prev_bio_flags);
2788 wbc_account_io(wbc, page, page_size);
2793 bio = btrfs_bio_alloc(bdev, offset);
2794 bio_add_page(bio, page, page_size, pg_offset);
2795 bio->bi_end_io = end_io_func;
2796 bio->bi_private = tree;
2797 bio->bi_write_hint = page->mapping->host->i_write_hint;
2800 wbc_init_bio(wbc, bio);
2801 wbc_account_io(wbc, page, page_size);
2809 static void attach_extent_buffer_page(struct extent_buffer *eb,
2812 if (!PagePrivate(page)) {
2813 SetPagePrivate(page);
2815 set_page_private(page, (unsigned long)eb);
2817 WARN_ON(page->private != (unsigned long)eb);
2821 void set_page_extent_mapped(struct page *page)
2823 if (!PagePrivate(page)) {
2824 SetPagePrivate(page);
2826 set_page_private(page, EXTENT_PAGE_PRIVATE);
2830 static struct extent_map *
2831 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
2832 u64 start, u64 len, get_extent_t *get_extent,
2833 struct extent_map **em_cached)
2835 struct extent_map *em;
2837 if (em_cached && *em_cached) {
2839 if (extent_map_in_tree(em) && start >= em->start &&
2840 start < extent_map_end(em)) {
2841 refcount_inc(&em->refs);
2845 free_extent_map(em);
2849 em = get_extent(BTRFS_I(inode), page, pg_offset, start, len, 0);
2850 if (em_cached && !IS_ERR_OR_NULL(em)) {
2852 refcount_inc(&em->refs);
2858 * basic readpage implementation. Locked extent state structs are inserted
2859 * into the tree that are removed when the IO is done (by the end_io
2861 * XXX JDM: This needs looking at to ensure proper page locking
2862 * return 0 on success, otherwise return error
2864 static int __do_readpage(struct extent_io_tree *tree,
2866 get_extent_t *get_extent,
2867 struct extent_map **em_cached,
2868 struct bio **bio, int mirror_num,
2869 unsigned long *bio_flags, unsigned int read_flags,
2872 struct inode *inode = page->mapping->host;
2873 u64 start = page_offset(page);
2874 const u64 end = start + PAGE_SIZE - 1;
2877 u64 last_byte = i_size_read(inode);
2880 struct extent_map *em;
2881 struct block_device *bdev;
2884 size_t pg_offset = 0;
2886 size_t disk_io_size;
2887 size_t blocksize = inode->i_sb->s_blocksize;
2888 unsigned long this_bio_flag = 0;
2890 set_page_extent_mapped(page);
2892 if (!PageUptodate(page)) {
2893 if (cleancache_get_page(page) == 0) {
2894 BUG_ON(blocksize != PAGE_SIZE);
2895 unlock_extent(tree, start, end);
2900 if (page->index == last_byte >> PAGE_SHIFT) {
2902 size_t zero_offset = last_byte & (PAGE_SIZE - 1);
2905 iosize = PAGE_SIZE - zero_offset;
2906 userpage = kmap_atomic(page);
2907 memset(userpage + zero_offset, 0, iosize);
2908 flush_dcache_page(page);
2909 kunmap_atomic(userpage);
2912 while (cur <= end) {
2913 bool force_bio_submit = false;
2916 if (cur >= last_byte) {
2918 struct extent_state *cached = NULL;
2920 iosize = PAGE_SIZE - pg_offset;
2921 userpage = kmap_atomic(page);
2922 memset(userpage + pg_offset, 0, iosize);
2923 flush_dcache_page(page);
2924 kunmap_atomic(userpage);
2925 set_extent_uptodate(tree, cur, cur + iosize - 1,
2927 unlock_extent_cached(tree, cur,
2928 cur + iosize - 1, &cached);
2931 em = __get_extent_map(inode, page, pg_offset, cur,
2932 end - cur + 1, get_extent, em_cached);
2933 if (IS_ERR_OR_NULL(em)) {
2935 unlock_extent(tree, cur, end);
2938 extent_offset = cur - em->start;
2939 BUG_ON(extent_map_end(em) <= cur);
2942 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2943 this_bio_flag |= EXTENT_BIO_COMPRESSED;
2944 extent_set_compress_type(&this_bio_flag,
2948 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2949 cur_end = min(extent_map_end(em) - 1, end);
2950 iosize = ALIGN(iosize, blocksize);
2951 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2952 disk_io_size = em->block_len;
2953 offset = em->block_start;
2955 offset = em->block_start + extent_offset;
2956 disk_io_size = iosize;
2959 block_start = em->block_start;
2960 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2961 block_start = EXTENT_MAP_HOLE;
2964 * If we have a file range that points to a compressed extent
2965 * and it's followed by a consecutive file range that points to
2966 * to the same compressed extent (possibly with a different
2967 * offset and/or length, so it either points to the whole extent
2968 * or only part of it), we must make sure we do not submit a
2969 * single bio to populate the pages for the 2 ranges because
2970 * this makes the compressed extent read zero out the pages
2971 * belonging to the 2nd range. Imagine the following scenario:
2974 * [0 - 8K] [8K - 24K]
2977 * points to extent X, points to extent X,
2978 * offset 4K, length of 8K offset 0, length 16K
2980 * [extent X, compressed length = 4K uncompressed length = 16K]
2982 * If the bio to read the compressed extent covers both ranges,
2983 * it will decompress extent X into the pages belonging to the
2984 * first range and then it will stop, zeroing out the remaining
2985 * pages that belong to the other range that points to extent X.
2986 * So here we make sure we submit 2 bios, one for the first
2987 * range and another one for the third range. Both will target
2988 * the same physical extent from disk, but we can't currently
2989 * make the compressed bio endio callback populate the pages
2990 * for both ranges because each compressed bio is tightly
2991 * coupled with a single extent map, and each range can have
2992 * an extent map with a different offset value relative to the
2993 * uncompressed data of our extent and different lengths. This
2994 * is a corner case so we prioritize correctness over
2995 * non-optimal behavior (submitting 2 bios for the same extent).
2997 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) &&
2998 prev_em_start && *prev_em_start != (u64)-1 &&
2999 *prev_em_start != em->orig_start)
3000 force_bio_submit = true;
3003 *prev_em_start = em->orig_start;
3005 free_extent_map(em);
3008 /* we've found a hole, just zero and go on */
3009 if (block_start == EXTENT_MAP_HOLE) {
3011 struct extent_state *cached = NULL;
3013 userpage = kmap_atomic(page);
3014 memset(userpage + pg_offset, 0, iosize);
3015 flush_dcache_page(page);
3016 kunmap_atomic(userpage);
3018 set_extent_uptodate(tree, cur, cur + iosize - 1,
3020 unlock_extent_cached(tree, cur,
3021 cur + iosize - 1, &cached);
3023 pg_offset += iosize;
3026 /* the get_extent function already copied into the page */
3027 if (test_range_bit(tree, cur, cur_end,
3028 EXTENT_UPTODATE, 1, NULL)) {
3029 check_page_uptodate(tree, page);
3030 unlock_extent(tree, cur, cur + iosize - 1);
3032 pg_offset += iosize;
3035 /* we have an inline extent but it didn't get marked up
3036 * to date. Error out
3038 if (block_start == EXTENT_MAP_INLINE) {
3040 unlock_extent(tree, cur, cur + iosize - 1);
3042 pg_offset += iosize;
3046 ret = submit_extent_page(REQ_OP_READ | read_flags, tree, NULL,
3047 page, offset, disk_io_size,
3048 pg_offset, bdev, bio,
3049 end_bio_extent_readpage, mirror_num,
3055 *bio_flags = this_bio_flag;
3058 unlock_extent(tree, cur, cur + iosize - 1);
3062 pg_offset += iosize;
3066 if (!PageError(page))
3067 SetPageUptodate(page);
3073 static inline void __do_contiguous_readpages(struct extent_io_tree *tree,
3074 struct page *pages[], int nr_pages,
3076 struct extent_map **em_cached,
3078 unsigned long *bio_flags,
3081 struct inode *inode;
3082 struct btrfs_ordered_extent *ordered;
3085 inode = pages[0]->mapping->host;
3087 lock_extent(tree, start, end);
3088 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
3092 unlock_extent(tree, start, end);
3093 btrfs_start_ordered_extent(inode, ordered, 1);
3094 btrfs_put_ordered_extent(ordered);
3097 for (index = 0; index < nr_pages; index++) {
3098 __do_readpage(tree, pages[index], btrfs_get_extent, em_cached,
3099 bio, 0, bio_flags, REQ_RAHEAD, prev_em_start);
3100 put_page(pages[index]);
3104 static void __extent_readpages(struct extent_io_tree *tree,
3105 struct page *pages[],
3107 struct extent_map **em_cached,
3108 struct bio **bio, unsigned long *bio_flags,
3115 int first_index = 0;
3117 for (index = 0; index < nr_pages; index++) {
3118 page_start = page_offset(pages[index]);
3121 end = start + PAGE_SIZE - 1;
3122 first_index = index;
3123 } else if (end + 1 == page_start) {
3126 __do_contiguous_readpages(tree, &pages[first_index],
3127 index - first_index, start,
3132 end = start + PAGE_SIZE - 1;
3133 first_index = index;
3138 __do_contiguous_readpages(tree, &pages[first_index],
3139 index - first_index, start,
3140 end, em_cached, bio,
3141 bio_flags, prev_em_start);
3144 static int __extent_read_full_page(struct extent_io_tree *tree,
3146 get_extent_t *get_extent,
3147 struct bio **bio, int mirror_num,
3148 unsigned long *bio_flags,
3149 unsigned int read_flags)
3151 struct inode *inode = page->mapping->host;
3152 struct btrfs_ordered_extent *ordered;
3153 u64 start = page_offset(page);
3154 u64 end = start + PAGE_SIZE - 1;
3158 lock_extent(tree, start, end);
3159 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
3163 unlock_extent(tree, start, end);
3164 btrfs_start_ordered_extent(inode, ordered, 1);
3165 btrfs_put_ordered_extent(ordered);
3168 ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num,
3169 bio_flags, read_flags, NULL);
3173 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
3174 get_extent_t *get_extent, int mirror_num)
3176 struct bio *bio = NULL;
3177 unsigned long bio_flags = 0;
3180 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
3183 ret = submit_one_bio(bio, mirror_num, bio_flags);
3187 static void update_nr_written(struct writeback_control *wbc,
3188 unsigned long nr_written)
3190 wbc->nr_to_write -= nr_written;
3194 * helper for __extent_writepage, doing all of the delayed allocation setup.
3196 * This returns 1 if btrfs_run_delalloc_range function did all the work required
3197 * to write the page (copy into inline extent). In this case the IO has
3198 * been started and the page is already unlocked.
3200 * This returns 0 if all went well (page still locked)
3201 * This returns < 0 if there were errors (page still locked)
3203 static noinline_for_stack int writepage_delalloc(struct inode *inode,
3204 struct page *page, struct writeback_control *wbc,
3205 struct extent_page_data *epd,
3207 unsigned long *nr_written)
3209 struct extent_io_tree *tree = epd->tree;
3210 u64 page_end = delalloc_start + PAGE_SIZE - 1;
3212 u64 delalloc_to_write = 0;
3213 u64 delalloc_end = 0;
3215 int page_started = 0;
3218 while (delalloc_end < page_end) {
3219 nr_delalloc = find_lock_delalloc_range(inode, tree,
3223 BTRFS_MAX_EXTENT_SIZE);
3224 if (nr_delalloc == 0) {
3225 delalloc_start = delalloc_end + 1;
3228 ret = btrfs_run_delalloc_range(inode, page, delalloc_start,
3229 delalloc_end, &page_started, nr_written, wbc);
3230 /* File system has been set read-only */
3234 * btrfs_run_delalloc_range should return < 0 for error
3235 * but just in case, we use > 0 here meaning the IO is
3236 * started, so we don't want to return > 0 unless
3237 * things are going well.
3239 ret = ret < 0 ? ret : -EIO;
3243 * delalloc_end is already one less than the total length, so
3244 * we don't subtract one from PAGE_SIZE
3246 delalloc_to_write += (delalloc_end - delalloc_start +
3247 PAGE_SIZE) >> PAGE_SHIFT;
3248 delalloc_start = delalloc_end + 1;
3250 if (wbc->nr_to_write < delalloc_to_write) {
3253 if (delalloc_to_write < thresh * 2)
3254 thresh = delalloc_to_write;
3255 wbc->nr_to_write = min_t(u64, delalloc_to_write,
3259 /* did the fill delalloc function already unlock and start
3264 * we've unlocked the page, so we can't update
3265 * the mapping's writeback index, just update
3268 wbc->nr_to_write -= *nr_written;
3279 * helper for __extent_writepage. This calls the writepage start hooks,
3280 * and does the loop to map the page into extents and bios.
3282 * We return 1 if the IO is started and the page is unlocked,
3283 * 0 if all went well (page still locked)
3284 * < 0 if there were errors (page still locked)
3286 static noinline_for_stack int __extent_writepage_io(struct inode *inode,
3288 struct writeback_control *wbc,
3289 struct extent_page_data *epd,
3291 unsigned long nr_written,
3292 unsigned int write_flags, int *nr_ret)
3294 struct extent_io_tree *tree = epd->tree;
3295 u64 start = page_offset(page);
3296 u64 page_end = start + PAGE_SIZE - 1;
3302 struct extent_map *em;
3303 struct block_device *bdev;
3304 size_t pg_offset = 0;
3310 ret = btrfs_writepage_cow_fixup(page, start, page_end);
3312 /* Fixup worker will requeue */
3314 wbc->pages_skipped++;
3316 redirty_page_for_writepage(wbc, page);
3318 update_nr_written(wbc, nr_written);
3324 * we don't want to touch the inode after unlocking the page,
3325 * so we update the mapping writeback index now
3327 update_nr_written(wbc, nr_written + 1);
3330 if (i_size <= start) {
3331 btrfs_writepage_endio_finish_ordered(page, start, page_end,
3336 blocksize = inode->i_sb->s_blocksize;
3338 while (cur <= end) {
3342 if (cur >= i_size) {
3343 btrfs_writepage_endio_finish_ordered(page, cur,
3347 em = btrfs_get_extent(BTRFS_I(inode), page, pg_offset, cur,
3349 if (IS_ERR_OR_NULL(em)) {
3351 ret = PTR_ERR_OR_ZERO(em);
3355 extent_offset = cur - em->start;
3356 em_end = extent_map_end(em);
3357 BUG_ON(em_end <= cur);
3359 iosize = min(em_end - cur, end - cur + 1);
3360 iosize = ALIGN(iosize, blocksize);
3361 offset = em->block_start + extent_offset;
3363 block_start = em->block_start;
3364 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3365 free_extent_map(em);
3369 * compressed and inline extents are written through other
3372 if (compressed || block_start == EXTENT_MAP_HOLE ||
3373 block_start == EXTENT_MAP_INLINE) {
3375 * end_io notification does not happen here for
3376 * compressed extents
3379 btrfs_writepage_endio_finish_ordered(page, cur,
3382 else if (compressed) {
3383 /* we don't want to end_page_writeback on
3384 * a compressed extent. this happens
3391 pg_offset += iosize;
3395 btrfs_set_range_writeback(tree, cur, cur + iosize - 1);
3396 if (!PageWriteback(page)) {
3397 btrfs_err(BTRFS_I(inode)->root->fs_info,
3398 "page %lu not writeback, cur %llu end %llu",
3399 page->index, cur, end);
3402 ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc,
3403 page, offset, iosize, pg_offset,
3405 end_bio_extent_writepage,
3409 if (PageWriteback(page))
3410 end_page_writeback(page);
3414 pg_offset += iosize;
3423 * the writepage semantics are similar to regular writepage. extent
3424 * records are inserted to lock ranges in the tree, and as dirty areas
3425 * are found, they are marked writeback. Then the lock bits are removed
3426 * and the end_io handler clears the writeback ranges
3428 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
3429 struct extent_page_data *epd)
3431 struct inode *inode = page->mapping->host;
3432 u64 start = page_offset(page);
3433 u64 page_end = start + PAGE_SIZE - 1;
3436 size_t pg_offset = 0;
3437 loff_t i_size = i_size_read(inode);
3438 unsigned long end_index = i_size >> PAGE_SHIFT;
3439 unsigned int write_flags = 0;
3440 unsigned long nr_written = 0;
3442 write_flags = wbc_to_write_flags(wbc);
3444 trace___extent_writepage(page, inode, wbc);
3446 WARN_ON(!PageLocked(page));
3448 ClearPageError(page);
3450 pg_offset = i_size & (PAGE_SIZE - 1);
3451 if (page->index > end_index ||
3452 (page->index == end_index && !pg_offset)) {
3453 page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
3458 if (page->index == end_index) {
3461 userpage = kmap_atomic(page);
3462 memset(userpage + pg_offset, 0,
3463 PAGE_SIZE - pg_offset);
3464 kunmap_atomic(userpage);
3465 flush_dcache_page(page);
3470 set_page_extent_mapped(page);
3472 if (!epd->extent_locked) {
3473 ret = writepage_delalloc(inode, page, wbc, epd, start,
3481 ret = __extent_writepage_io(inode, page, wbc, epd,
3482 i_size, nr_written, write_flags, &nr);
3488 /* make sure the mapping tag for page dirty gets cleared */
3489 set_page_writeback(page);
3490 end_page_writeback(page);
3492 if (PageError(page)) {
3493 ret = ret < 0 ? ret : -EIO;
3494 end_extent_writepage(page, ret, start, page_end);
3503 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3505 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
3506 TASK_UNINTERRUPTIBLE);
3509 static noinline_for_stack int
3510 lock_extent_buffer_for_io(struct extent_buffer *eb,
3511 struct btrfs_fs_info *fs_info,
3512 struct extent_page_data *epd)
3518 if (!btrfs_try_tree_write_lock(eb)) {
3520 flush_write_bio(epd);
3521 btrfs_tree_lock(eb);
3524 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3525 btrfs_tree_unlock(eb);
3529 flush_write_bio(epd);
3533 wait_on_extent_buffer_writeback(eb);
3534 btrfs_tree_lock(eb);
3535 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3537 btrfs_tree_unlock(eb);
3542 * We need to do this to prevent races in people who check if the eb is
3543 * under IO since we can end up having no IO bits set for a short period
3546 spin_lock(&eb->refs_lock);
3547 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3548 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3549 spin_unlock(&eb->refs_lock);
3550 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3551 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
3553 fs_info->dirty_metadata_batch);
3556 spin_unlock(&eb->refs_lock);
3559 btrfs_tree_unlock(eb);
3564 num_pages = num_extent_pages(eb);
3565 for (i = 0; i < num_pages; i++) {
3566 struct page *p = eb->pages[i];
3568 if (!trylock_page(p)) {
3570 flush_write_bio(epd);
3580 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3582 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3583 smp_mb__after_atomic();
3584 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3587 static void set_btree_ioerr(struct page *page)
3589 struct extent_buffer *eb = (struct extent_buffer *)page->private;
3592 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))
3596 * If writeback for a btree extent that doesn't belong to a log tree
3597 * failed, increment the counter transaction->eb_write_errors.
3598 * We do this because while the transaction is running and before it's
3599 * committing (when we call filemap_fdata[write|wait]_range against
3600 * the btree inode), we might have
3601 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
3602 * returns an error or an error happens during writeback, when we're
3603 * committing the transaction we wouldn't know about it, since the pages
3604 * can be no longer dirty nor marked anymore for writeback (if a
3605 * subsequent modification to the extent buffer didn't happen before the
3606 * transaction commit), which makes filemap_fdata[write|wait]_range not
3607 * able to find the pages tagged with SetPageError at transaction
3608 * commit time. So if this happens we must abort the transaction,
3609 * otherwise we commit a super block with btree roots that point to
3610 * btree nodes/leafs whose content on disk is invalid - either garbage
3611 * or the content of some node/leaf from a past generation that got
3612 * cowed or deleted and is no longer valid.
3614 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
3615 * not be enough - we need to distinguish between log tree extents vs
3616 * non-log tree extents, and the next filemap_fdatawait_range() call
3617 * will catch and clear such errors in the mapping - and that call might
3618 * be from a log sync and not from a transaction commit. Also, checking
3619 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
3620 * not done and would not be reliable - the eb might have been released
3621 * from memory and reading it back again means that flag would not be
3622 * set (since it's a runtime flag, not persisted on disk).
3624 * Using the flags below in the btree inode also makes us achieve the
3625 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
3626 * writeback for all dirty pages and before filemap_fdatawait_range()
3627 * is called, the writeback for all dirty pages had already finished
3628 * with errors - because we were not using AS_EIO/AS_ENOSPC,
3629 * filemap_fdatawait_range() would return success, as it could not know
3630 * that writeback errors happened (the pages were no longer tagged for
3633 switch (eb->log_index) {
3635 set_bit(BTRFS_FS_BTREE_ERR, &eb->fs_info->flags);
3638 set_bit(BTRFS_FS_LOG1_ERR, &eb->fs_info->flags);
3641 set_bit(BTRFS_FS_LOG2_ERR, &eb->fs_info->flags);
3644 BUG(); /* unexpected, logic error */
3648 static void end_bio_extent_buffer_writepage(struct bio *bio)
3650 struct bio_vec *bvec;
3651 struct extent_buffer *eb;
3654 ASSERT(!bio_flagged(bio, BIO_CLONED));
3655 bio_for_each_segment_all(bvec, bio, i) {
3656 struct page *page = bvec->bv_page;
3658 eb = (struct extent_buffer *)page->private;
3660 done = atomic_dec_and_test(&eb->io_pages);
3662 if (bio->bi_status ||
3663 test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) {
3664 ClearPageUptodate(page);
3665 set_btree_ioerr(page);
3668 end_page_writeback(page);
3673 end_extent_buffer_writeback(eb);
3679 static noinline_for_stack int write_one_eb(struct extent_buffer *eb,
3680 struct btrfs_fs_info *fs_info,
3681 struct writeback_control *wbc,
3682 struct extent_page_data *epd)
3684 struct block_device *bdev = fs_info->fs_devices->latest_bdev;
3685 struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
3686 u64 offset = eb->start;
3689 unsigned long start, end;
3690 unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META;
3693 clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
3694 num_pages = num_extent_pages(eb);
3695 atomic_set(&eb->io_pages, num_pages);
3697 /* set btree blocks beyond nritems with 0 to avoid stale content. */
3698 nritems = btrfs_header_nritems(eb);
3699 if (btrfs_header_level(eb) > 0) {
3700 end = btrfs_node_key_ptr_offset(nritems);
3702 memzero_extent_buffer(eb, end, eb->len - end);
3706 * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0
3708 start = btrfs_item_nr_offset(nritems);
3709 end = BTRFS_LEAF_DATA_OFFSET + leaf_data_end(fs_info, eb);
3710 memzero_extent_buffer(eb, start, end - start);
3713 for (i = 0; i < num_pages; i++) {
3714 struct page *p = eb->pages[i];
3716 clear_page_dirty_for_io(p);
3717 set_page_writeback(p);
3718 ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc,
3719 p, offset, PAGE_SIZE, 0, bdev,
3721 end_bio_extent_buffer_writepage,
3725 if (PageWriteback(p))
3726 end_page_writeback(p);
3727 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3728 end_extent_buffer_writeback(eb);
3732 offset += PAGE_SIZE;
3733 update_nr_written(wbc, 1);
3737 if (unlikely(ret)) {
3738 for (; i < num_pages; i++) {
3739 struct page *p = eb->pages[i];
3740 clear_page_dirty_for_io(p);
3748 int btree_write_cache_pages(struct address_space *mapping,
3749 struct writeback_control *wbc)
3751 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
3752 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
3753 struct extent_buffer *eb, *prev_eb = NULL;
3754 struct extent_page_data epd = {
3758 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3762 int nr_to_write_done = 0;
3763 struct pagevec pvec;
3766 pgoff_t end; /* Inclusive */
3770 pagevec_init(&pvec);
3771 if (wbc->range_cyclic) {
3772 index = mapping->writeback_index; /* Start from prev offset */
3775 index = wbc->range_start >> PAGE_SHIFT;
3776 end = wbc->range_end >> PAGE_SHIFT;
3779 if (wbc->sync_mode == WB_SYNC_ALL)
3780 tag = PAGECACHE_TAG_TOWRITE;
3782 tag = PAGECACHE_TAG_DIRTY;
3784 if (wbc->sync_mode == WB_SYNC_ALL)
3785 tag_pages_for_writeback(mapping, index, end);
3786 while (!done && !nr_to_write_done && (index <= end) &&
3787 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
3792 for (i = 0; i < nr_pages; i++) {
3793 struct page *page = pvec.pages[i];
3795 if (!PagePrivate(page))
3798 spin_lock(&mapping->private_lock);
3799 if (!PagePrivate(page)) {
3800 spin_unlock(&mapping->private_lock);
3804 eb = (struct extent_buffer *)page->private;
3807 * Shouldn't happen and normally this would be a BUG_ON
3808 * but no sense in crashing the users box for something
3809 * we can survive anyway.
3812 spin_unlock(&mapping->private_lock);
3816 if (eb == prev_eb) {
3817 spin_unlock(&mapping->private_lock);
3821 ret = atomic_inc_not_zero(&eb->refs);
3822 spin_unlock(&mapping->private_lock);
3827 ret = lock_extent_buffer_for_io(eb, fs_info, &epd);
3829 free_extent_buffer(eb);
3833 ret = write_one_eb(eb, fs_info, wbc, &epd);
3836 free_extent_buffer(eb);
3839 free_extent_buffer(eb);
3842 * the filesystem may choose to bump up nr_to_write.
3843 * We have to make sure to honor the new nr_to_write
3846 nr_to_write_done = wbc->nr_to_write <= 0;
3848 pagevec_release(&pvec);
3851 if (!scanned && !done) {
3853 * We hit the last page and there is more work to be done: wrap
3854 * back to the start of the file
3860 flush_write_bio(&epd);
3865 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3866 * @mapping: address space structure to write
3867 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3868 * @data: data passed to __extent_writepage function
3870 * If a page is already under I/O, write_cache_pages() skips it, even
3871 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3872 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3873 * and msync() need to guarantee that all the data which was dirty at the time
3874 * the call was made get new I/O started against them. If wbc->sync_mode is
3875 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3876 * existing IO to complete.
3878 static int extent_write_cache_pages(struct address_space *mapping,
3879 struct writeback_control *wbc,
3880 struct extent_page_data *epd)
3882 struct inode *inode = mapping->host;
3885 int nr_to_write_done = 0;
3886 struct pagevec pvec;
3889 pgoff_t end; /* Inclusive */
3891 int range_whole = 0;
3896 * We have to hold onto the inode so that ordered extents can do their
3897 * work when the IO finishes. The alternative to this is failing to add
3898 * an ordered extent if the igrab() fails there and that is a huge pain
3899 * to deal with, so instead just hold onto the inode throughout the
3900 * writepages operation. If it fails here we are freeing up the inode
3901 * anyway and we'd rather not waste our time writing out stuff that is
3902 * going to be truncated anyway.
3907 pagevec_init(&pvec);
3908 if (wbc->range_cyclic) {
3909 index = mapping->writeback_index; /* Start from prev offset */
3912 index = wbc->range_start >> PAGE_SHIFT;
3913 end = wbc->range_end >> PAGE_SHIFT;
3914 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
3918 if (wbc->sync_mode == WB_SYNC_ALL)
3919 tag = PAGECACHE_TAG_TOWRITE;
3921 tag = PAGECACHE_TAG_DIRTY;
3923 if (wbc->sync_mode == WB_SYNC_ALL)
3924 tag_pages_for_writeback(mapping, index, end);
3926 while (!done && !nr_to_write_done && (index <= end) &&
3927 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping,
3928 &index, end, tag))) {
3932 for (i = 0; i < nr_pages; i++) {
3933 struct page *page = pvec.pages[i];
3935 done_index = page->index;
3937 * At this point we hold neither the i_pages lock nor
3938 * the page lock: the page may be truncated or
3939 * invalidated (changing page->mapping to NULL),
3940 * or even swizzled back from swapper_space to
3941 * tmpfs file mapping
3943 if (!trylock_page(page)) {
3944 flush_write_bio(epd);
3948 if (unlikely(page->mapping != mapping)) {
3953 if (wbc->sync_mode != WB_SYNC_NONE) {
3954 if (PageWriteback(page))
3955 flush_write_bio(epd);
3956 wait_on_page_writeback(page);
3959 if (PageWriteback(page) ||
3960 !clear_page_dirty_for_io(page)) {
3965 ret = __extent_writepage(page, wbc, epd);
3967 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
3973 * done_index is set past this page,
3974 * so media errors will not choke
3975 * background writeout for the entire
3976 * file. This has consequences for
3977 * range_cyclic semantics (ie. it may
3978 * not be suitable for data integrity
3981 done_index = page->index + 1;
3987 * the filesystem may choose to bump up nr_to_write.
3988 * We have to make sure to honor the new nr_to_write
3991 nr_to_write_done = wbc->nr_to_write <= 0;
3993 pagevec_release(&pvec);
3996 if (!scanned && !done) {
3998 * We hit the last page and there is more work to be done: wrap
3999 * back to the start of the file
4006 if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole))
4007 mapping->writeback_index = done_index;
4009 btrfs_add_delayed_iput(inode);
4013 static void flush_write_bio(struct extent_page_data *epd)
4018 ret = submit_one_bio(epd->bio, 0, 0);
4019 BUG_ON(ret < 0); /* -ENOMEM */
4024 int extent_write_full_page(struct page *page, struct writeback_control *wbc)
4027 struct extent_page_data epd = {
4029 .tree = &BTRFS_I(page->mapping->host)->io_tree,
4031 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4034 ret = __extent_writepage(page, wbc, &epd);
4036 flush_write_bio(&epd);
4040 int extent_write_locked_range(struct inode *inode, u64 start, u64 end,
4044 struct address_space *mapping = inode->i_mapping;
4045 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
4047 unsigned long nr_pages = (end - start + PAGE_SIZE) >>
4050 struct extent_page_data epd = {
4054 .sync_io = mode == WB_SYNC_ALL,
4056 struct writeback_control wbc_writepages = {
4058 .nr_to_write = nr_pages * 2,
4059 .range_start = start,
4060 .range_end = end + 1,
4063 while (start <= end) {
4064 page = find_get_page(mapping, start >> PAGE_SHIFT);
4065 if (clear_page_dirty_for_io(page))
4066 ret = __extent_writepage(page, &wbc_writepages, &epd);
4068 btrfs_writepage_endio_finish_ordered(page, start,
4069 start + PAGE_SIZE - 1,
4077 flush_write_bio(&epd);
4081 int extent_writepages(struct address_space *mapping,
4082 struct writeback_control *wbc)
4085 struct extent_page_data epd = {
4087 .tree = &BTRFS_I(mapping->host)->io_tree,
4089 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4092 ret = extent_write_cache_pages(mapping, wbc, &epd);
4093 flush_write_bio(&epd);
4097 int extent_readpages(struct address_space *mapping, struct list_head *pages,
4100 struct bio *bio = NULL;
4102 unsigned long bio_flags = 0;
4103 struct page *pagepool[16];
4105 struct extent_map *em_cached = NULL;
4106 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
4108 u64 prev_em_start = (u64)-1;
4110 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
4111 page = list_entry(pages->prev, struct page, lru);
4113 prefetchw(&page->flags);
4114 list_del(&page->lru);
4115 if (add_to_page_cache_lru(page, mapping,
4117 readahead_gfp_mask(mapping))) {
4122 pagepool[nr++] = page;
4123 if (nr < ARRAY_SIZE(pagepool))
4125 __extent_readpages(tree, pagepool, nr, &em_cached, &bio,
4126 &bio_flags, &prev_em_start);
4130 __extent_readpages(tree, pagepool, nr, &em_cached, &bio,
4131 &bio_flags, &prev_em_start);
4134 free_extent_map(em_cached);
4136 BUG_ON(!list_empty(pages));
4138 return submit_one_bio(bio, 0, bio_flags);
4143 * basic invalidatepage code, this waits on any locked or writeback
4144 * ranges corresponding to the page, and then deletes any extent state
4145 * records from the tree
4147 int extent_invalidatepage(struct extent_io_tree *tree,
4148 struct page *page, unsigned long offset)
4150 struct extent_state *cached_state = NULL;
4151 u64 start = page_offset(page);
4152 u64 end = start + PAGE_SIZE - 1;
4153 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
4155 start += ALIGN(offset, blocksize);
4159 lock_extent_bits(tree, start, end, &cached_state);
4160 wait_on_page_writeback(page);
4161 clear_extent_bit(tree, start, end,
4162 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4163 EXTENT_DO_ACCOUNTING,
4164 1, 1, &cached_state);
4169 * a helper for releasepage, this tests for areas of the page that
4170 * are locked or under IO and drops the related state bits if it is safe
4173 static int try_release_extent_state(struct extent_io_tree *tree,
4174 struct page *page, gfp_t mask)
4176 u64 start = page_offset(page);
4177 u64 end = start + PAGE_SIZE - 1;
4180 if (test_range_bit(tree, start, end,
4181 EXTENT_IOBITS, 0, NULL))
4185 * at this point we can safely clear everything except the
4186 * locked bit and the nodatasum bit
4188 ret = __clear_extent_bit(tree, start, end,
4189 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
4190 0, 0, NULL, mask, NULL);
4192 /* if clear_extent_bit failed for enomem reasons,
4193 * we can't allow the release to continue.
4204 * a helper for releasepage. As long as there are no locked extents
4205 * in the range corresponding to the page, both state records and extent
4206 * map records are removed
4208 int try_release_extent_mapping(struct page *page, gfp_t mask)
4210 struct extent_map *em;
4211 u64 start = page_offset(page);
4212 u64 end = start + PAGE_SIZE - 1;
4213 struct btrfs_inode *btrfs_inode = BTRFS_I(page->mapping->host);
4214 struct extent_io_tree *tree = &btrfs_inode->io_tree;
4215 struct extent_map_tree *map = &btrfs_inode->extent_tree;
4217 if (gfpflags_allow_blocking(mask) &&
4218 page->mapping->host->i_size > SZ_16M) {
4220 while (start <= end) {
4221 len = end - start + 1;
4222 write_lock(&map->lock);
4223 em = lookup_extent_mapping(map, start, len);
4225 write_unlock(&map->lock);
4228 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
4229 em->start != start) {
4230 write_unlock(&map->lock);
4231 free_extent_map(em);
4234 if (!test_range_bit(tree, em->start,
4235 extent_map_end(em) - 1,
4236 EXTENT_LOCKED | EXTENT_WRITEBACK,
4238 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4239 &btrfs_inode->runtime_flags);
4240 remove_extent_mapping(map, em);
4241 /* once for the rb tree */
4242 free_extent_map(em);
4244 start = extent_map_end(em);
4245 write_unlock(&map->lock);
4248 free_extent_map(em);
4251 return try_release_extent_state(tree, page, mask);
4255 * helper function for fiemap, which doesn't want to see any holes.
4256 * This maps until we find something past 'last'
4258 static struct extent_map *get_extent_skip_holes(struct inode *inode,
4259 u64 offset, u64 last)
4261 u64 sectorsize = btrfs_inode_sectorsize(inode);
4262 struct extent_map *em;
4269 len = last - offset;
4272 len = ALIGN(len, sectorsize);
4273 em = btrfs_get_extent_fiemap(BTRFS_I(inode), NULL, 0, offset,
4275 if (IS_ERR_OR_NULL(em))
4278 /* if this isn't a hole return it */
4279 if (em->block_start != EXTENT_MAP_HOLE)
4282 /* this is a hole, advance to the next extent */
4283 offset = extent_map_end(em);
4284 free_extent_map(em);
4292 * To cache previous fiemap extent
4294 * Will be used for merging fiemap extent
4296 struct fiemap_cache {
4305 * Helper to submit fiemap extent.
4307 * Will try to merge current fiemap extent specified by @offset, @phys,
4308 * @len and @flags with cached one.
4309 * And only when we fails to merge, cached one will be submitted as
4312 * Return value is the same as fiemap_fill_next_extent().
4314 static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
4315 struct fiemap_cache *cache,
4316 u64 offset, u64 phys, u64 len, u32 flags)
4324 * Sanity check, extent_fiemap() should have ensured that new
4325 * fiemap extent won't overlap with cahced one.
4328 * NOTE: Physical address can overlap, due to compression
4330 if (cache->offset + cache->len > offset) {
4336 * Only merges fiemap extents if
4337 * 1) Their logical addresses are continuous
4339 * 2) Their physical addresses are continuous
4340 * So truly compressed (physical size smaller than logical size)
4341 * extents won't get merged with each other
4343 * 3) Share same flags except FIEMAP_EXTENT_LAST
4344 * So regular extent won't get merged with prealloc extent
4346 if (cache->offset + cache->len == offset &&
4347 cache->phys + cache->len == phys &&
4348 (cache->flags & ~FIEMAP_EXTENT_LAST) ==
4349 (flags & ~FIEMAP_EXTENT_LAST)) {
4351 cache->flags |= flags;
4352 goto try_submit_last;
4355 /* Not mergeable, need to submit cached one */
4356 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
4357 cache->len, cache->flags);
4358 cache->cached = false;
4362 cache->cached = true;
4363 cache->offset = offset;
4366 cache->flags = flags;
4368 if (cache->flags & FIEMAP_EXTENT_LAST) {
4369 ret = fiemap_fill_next_extent(fieinfo, cache->offset,
4370 cache->phys, cache->len, cache->flags);
4371 cache->cached = false;
4377 * Emit last fiemap cache
4379 * The last fiemap cache may still be cached in the following case:
4381 * |<- Fiemap range ->|
4382 * |<------------ First extent ----------->|
4384 * In this case, the first extent range will be cached but not emitted.
4385 * So we must emit it before ending extent_fiemap().
4387 static int emit_last_fiemap_cache(struct btrfs_fs_info *fs_info,
4388 struct fiemap_extent_info *fieinfo,
4389 struct fiemap_cache *cache)
4396 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
4397 cache->len, cache->flags);
4398 cache->cached = false;
4404 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4405 __u64 start, __u64 len)
4409 u64 max = start + len;
4413 u64 last_for_get_extent = 0;
4415 u64 isize = i_size_read(inode);
4416 struct btrfs_key found_key;
4417 struct extent_map *em = NULL;
4418 struct extent_state *cached_state = NULL;
4419 struct btrfs_path *path;
4420 struct btrfs_root *root = BTRFS_I(inode)->root;
4421 struct fiemap_cache cache = { 0 };
4430 path = btrfs_alloc_path();
4433 path->leave_spinning = 1;
4435 start = round_down(start, btrfs_inode_sectorsize(inode));
4436 len = round_up(max, btrfs_inode_sectorsize(inode)) - start;
4439 * lookup the last file extent. We're not using i_size here
4440 * because there might be preallocation past i_size
4442 ret = btrfs_lookup_file_extent(NULL, root, path,
4443 btrfs_ino(BTRFS_I(inode)), -1, 0);
4445 btrfs_free_path(path);
4454 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
4455 found_type = found_key.type;
4457 /* No extents, but there might be delalloc bits */
4458 if (found_key.objectid != btrfs_ino(BTRFS_I(inode)) ||
4459 found_type != BTRFS_EXTENT_DATA_KEY) {
4460 /* have to trust i_size as the end */
4462 last_for_get_extent = isize;
4465 * remember the start of the last extent. There are a
4466 * bunch of different factors that go into the length of the
4467 * extent, so its much less complex to remember where it started
4469 last = found_key.offset;
4470 last_for_get_extent = last + 1;
4472 btrfs_release_path(path);
4475 * we might have some extents allocated but more delalloc past those
4476 * extents. so, we trust isize unless the start of the last extent is
4481 last_for_get_extent = isize;
4484 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4487 em = get_extent_skip_holes(inode, start, last_for_get_extent);
4496 u64 offset_in_extent = 0;
4498 /* break if the extent we found is outside the range */
4499 if (em->start >= max || extent_map_end(em) < off)
4503 * get_extent may return an extent that starts before our
4504 * requested range. We have to make sure the ranges
4505 * we return to fiemap always move forward and don't
4506 * overlap, so adjust the offsets here
4508 em_start = max(em->start, off);
4511 * record the offset from the start of the extent
4512 * for adjusting the disk offset below. Only do this if the
4513 * extent isn't compressed since our in ram offset may be past
4514 * what we have actually allocated on disk.
4516 if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4517 offset_in_extent = em_start - em->start;
4518 em_end = extent_map_end(em);
4519 em_len = em_end - em_start;
4521 if (em->block_start < EXTENT_MAP_LAST_BYTE)
4522 disko = em->block_start + offset_in_extent;
4527 * bump off for our next call to get_extent
4529 off = extent_map_end(em);
4533 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
4535 flags |= FIEMAP_EXTENT_LAST;
4536 } else if (em->block_start == EXTENT_MAP_INLINE) {
4537 flags |= (FIEMAP_EXTENT_DATA_INLINE |
4538 FIEMAP_EXTENT_NOT_ALIGNED);
4539 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
4540 flags |= (FIEMAP_EXTENT_DELALLOC |
4541 FIEMAP_EXTENT_UNKNOWN);
4542 } else if (fieinfo->fi_extents_max) {
4543 u64 bytenr = em->block_start -
4544 (em->start - em->orig_start);
4547 * As btrfs supports shared space, this information
4548 * can be exported to userspace tools via
4549 * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0
4550 * then we're just getting a count and we can skip the
4553 ret = btrfs_check_shared(root,
4554 btrfs_ino(BTRFS_I(inode)),
4559 flags |= FIEMAP_EXTENT_SHARED;
4562 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4563 flags |= FIEMAP_EXTENT_ENCODED;
4564 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4565 flags |= FIEMAP_EXTENT_UNWRITTEN;
4567 free_extent_map(em);
4569 if ((em_start >= last) || em_len == (u64)-1 ||
4570 (last == (u64)-1 && isize <= em_end)) {
4571 flags |= FIEMAP_EXTENT_LAST;
4575 /* now scan forward to see if this is really the last extent. */
4576 em = get_extent_skip_holes(inode, off, last_for_get_extent);
4582 flags |= FIEMAP_EXTENT_LAST;
4585 ret = emit_fiemap_extent(fieinfo, &cache, em_start, disko,
4595 ret = emit_last_fiemap_cache(root->fs_info, fieinfo, &cache);
4596 free_extent_map(em);
4598 btrfs_free_path(path);
4599 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4604 static void __free_extent_buffer(struct extent_buffer *eb)
4606 btrfs_leak_debug_del(&eb->leak_list);
4607 kmem_cache_free(extent_buffer_cache, eb);
4610 int extent_buffer_under_io(struct extent_buffer *eb)
4612 return (atomic_read(&eb->io_pages) ||
4613 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4614 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4618 * Release all pages attached to the extent buffer.
4620 static void btrfs_release_extent_buffer_pages(struct extent_buffer *eb)
4624 int mapped = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
4626 BUG_ON(extent_buffer_under_io(eb));
4628 num_pages = num_extent_pages(eb);
4629 for (i = 0; i < num_pages; i++) {
4630 struct page *page = eb->pages[i];
4635 spin_lock(&page->mapping->private_lock);
4637 * We do this since we'll remove the pages after we've
4638 * removed the eb from the radix tree, so we could race
4639 * and have this page now attached to the new eb. So
4640 * only clear page_private if it's still connected to
4643 if (PagePrivate(page) &&
4644 page->private == (unsigned long)eb) {
4645 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4646 BUG_ON(PageDirty(page));
4647 BUG_ON(PageWriteback(page));
4649 * We need to make sure we haven't be attached
4652 ClearPagePrivate(page);
4653 set_page_private(page, 0);
4654 /* One for the page private */
4659 spin_unlock(&page->mapping->private_lock);
4661 /* One for when we allocated the page */
4667 * Helper for releasing the extent buffer.
4669 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4671 btrfs_release_extent_buffer_pages(eb);
4672 __free_extent_buffer(eb);
4675 static struct extent_buffer *
4676 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
4679 struct extent_buffer *eb = NULL;
4681 eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL);
4684 eb->fs_info = fs_info;
4686 rwlock_init(&eb->lock);
4687 atomic_set(&eb->write_locks, 0);
4688 atomic_set(&eb->read_locks, 0);
4689 atomic_set(&eb->blocking_readers, 0);
4690 atomic_set(&eb->blocking_writers, 0);
4691 atomic_set(&eb->spinning_readers, 0);
4692 atomic_set(&eb->spinning_writers, 0);
4693 eb->lock_nested = 0;
4694 init_waitqueue_head(&eb->write_lock_wq);
4695 init_waitqueue_head(&eb->read_lock_wq);
4697 btrfs_leak_debug_add(&eb->leak_list, &buffers);
4699 spin_lock_init(&eb->refs_lock);
4700 atomic_set(&eb->refs, 1);
4701 atomic_set(&eb->io_pages, 0);
4704 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4706 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4707 > MAX_INLINE_EXTENT_BUFFER_SIZE);
4708 BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE);
4713 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src)
4717 struct extent_buffer *new;
4718 int num_pages = num_extent_pages(src);
4720 new = __alloc_extent_buffer(src->fs_info, src->start, src->len);
4724 for (i = 0; i < num_pages; i++) {
4725 p = alloc_page(GFP_NOFS);
4727 btrfs_release_extent_buffer(new);
4730 attach_extent_buffer_page(new, p);
4731 WARN_ON(PageDirty(p));
4734 copy_page(page_address(p), page_address(src->pages[i]));
4737 set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
4738 set_bit(EXTENT_BUFFER_UNMAPPED, &new->bflags);
4743 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4744 u64 start, unsigned long len)
4746 struct extent_buffer *eb;
4750 eb = __alloc_extent_buffer(fs_info, start, len);
4754 num_pages = num_extent_pages(eb);
4755 for (i = 0; i < num_pages; i++) {
4756 eb->pages[i] = alloc_page(GFP_NOFS);
4760 set_extent_buffer_uptodate(eb);
4761 btrfs_set_header_nritems(eb, 0);
4762 set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
4767 __free_page(eb->pages[i - 1]);
4768 __free_extent_buffer(eb);
4772 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4775 return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize);
4778 static void check_buffer_tree_ref(struct extent_buffer *eb)
4781 /* the ref bit is tricky. We have to make sure it is set
4782 * if we have the buffer dirty. Otherwise the
4783 * code to free a buffer can end up dropping a dirty
4786 * Once the ref bit is set, it won't go away while the
4787 * buffer is dirty or in writeback, and it also won't
4788 * go away while we have the reference count on the
4791 * We can't just set the ref bit without bumping the
4792 * ref on the eb because free_extent_buffer might
4793 * see the ref bit and try to clear it. If this happens
4794 * free_extent_buffer might end up dropping our original
4795 * ref by mistake and freeing the page before we are able
4796 * to add one more ref.
4798 * So bump the ref count first, then set the bit. If someone
4799 * beat us to it, drop the ref we added.
4801 refs = atomic_read(&eb->refs);
4802 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4805 spin_lock(&eb->refs_lock);
4806 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4807 atomic_inc(&eb->refs);
4808 spin_unlock(&eb->refs_lock);
4811 static void mark_extent_buffer_accessed(struct extent_buffer *eb,
4812 struct page *accessed)
4816 check_buffer_tree_ref(eb);
4818 num_pages = num_extent_pages(eb);
4819 for (i = 0; i < num_pages; i++) {
4820 struct page *p = eb->pages[i];
4823 mark_page_accessed(p);
4827 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
4830 struct extent_buffer *eb;
4833 eb = radix_tree_lookup(&fs_info->buffer_radix,
4834 start >> PAGE_SHIFT);
4835 if (eb && atomic_inc_not_zero(&eb->refs)) {
4838 * Lock our eb's refs_lock to avoid races with
4839 * free_extent_buffer. When we get our eb it might be flagged
4840 * with EXTENT_BUFFER_STALE and another task running
4841 * free_extent_buffer might have seen that flag set,
4842 * eb->refs == 2, that the buffer isn't under IO (dirty and
4843 * writeback flags not set) and it's still in the tree (flag
4844 * EXTENT_BUFFER_TREE_REF set), therefore being in the process
4845 * of decrementing the extent buffer's reference count twice.
4846 * So here we could race and increment the eb's reference count,
4847 * clear its stale flag, mark it as dirty and drop our reference
4848 * before the other task finishes executing free_extent_buffer,
4849 * which would later result in an attempt to free an extent
4850 * buffer that is dirty.
4852 if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) {
4853 spin_lock(&eb->refs_lock);
4854 spin_unlock(&eb->refs_lock);
4856 mark_extent_buffer_accessed(eb, NULL);
4864 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4865 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
4868 struct extent_buffer *eb, *exists = NULL;
4871 eb = find_extent_buffer(fs_info, start);
4874 eb = alloc_dummy_extent_buffer(fs_info, start);
4877 eb->fs_info = fs_info;
4879 ret = radix_tree_preload(GFP_NOFS);
4882 spin_lock(&fs_info->buffer_lock);
4883 ret = radix_tree_insert(&fs_info->buffer_radix,
4884 start >> PAGE_SHIFT, eb);
4885 spin_unlock(&fs_info->buffer_lock);
4886 radix_tree_preload_end();
4887 if (ret == -EEXIST) {
4888 exists = find_extent_buffer(fs_info, start);
4894 check_buffer_tree_ref(eb);
4895 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
4899 btrfs_release_extent_buffer(eb);
4904 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
4907 unsigned long len = fs_info->nodesize;
4910 unsigned long index = start >> PAGE_SHIFT;
4911 struct extent_buffer *eb;
4912 struct extent_buffer *exists = NULL;
4914 struct address_space *mapping = fs_info->btree_inode->i_mapping;
4918 if (!IS_ALIGNED(start, fs_info->sectorsize)) {
4919 btrfs_err(fs_info, "bad tree block start %llu", start);
4920 return ERR_PTR(-EINVAL);
4923 eb = find_extent_buffer(fs_info, start);
4927 eb = __alloc_extent_buffer(fs_info, start, len);
4929 return ERR_PTR(-ENOMEM);
4931 num_pages = num_extent_pages(eb);
4932 for (i = 0; i < num_pages; i++, index++) {
4933 p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL);
4935 exists = ERR_PTR(-ENOMEM);
4939 spin_lock(&mapping->private_lock);
4940 if (PagePrivate(p)) {
4942 * We could have already allocated an eb for this page
4943 * and attached one so lets see if we can get a ref on
4944 * the existing eb, and if we can we know it's good and
4945 * we can just return that one, else we know we can just
4946 * overwrite page->private.
4948 exists = (struct extent_buffer *)p->private;
4949 if (atomic_inc_not_zero(&exists->refs)) {
4950 spin_unlock(&mapping->private_lock);
4953 mark_extent_buffer_accessed(exists, p);
4959 * Do this so attach doesn't complain and we need to
4960 * drop the ref the old guy had.
4962 ClearPagePrivate(p);
4963 WARN_ON(PageDirty(p));
4966 attach_extent_buffer_page(eb, p);
4967 spin_unlock(&mapping->private_lock);
4968 WARN_ON(PageDirty(p));
4970 if (!PageUptodate(p))
4974 * We can't unlock the pages just yet since the extent buffer
4975 * hasn't been properly inserted in the radix tree, this
4976 * opens a race with btree_releasepage which can free a page
4977 * while we are still filling in all pages for the buffer and
4982 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4984 ret = radix_tree_preload(GFP_NOFS);
4986 exists = ERR_PTR(ret);
4990 spin_lock(&fs_info->buffer_lock);
4991 ret = radix_tree_insert(&fs_info->buffer_radix,
4992 start >> PAGE_SHIFT, eb);
4993 spin_unlock(&fs_info->buffer_lock);
4994 radix_tree_preload_end();
4995 if (ret == -EEXIST) {
4996 exists = find_extent_buffer(fs_info, start);
5002 /* add one reference for the tree */
5003 check_buffer_tree_ref(eb);
5004 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
5007 * Now it's safe to unlock the pages because any calls to
5008 * btree_releasepage will correctly detect that a page belongs to a
5009 * live buffer and won't free them prematurely.
5011 for (i = 0; i < num_pages; i++)
5012 unlock_page(eb->pages[i]);
5016 WARN_ON(!atomic_dec_and_test(&eb->refs));
5017 for (i = 0; i < num_pages; i++) {
5019 unlock_page(eb->pages[i]);
5022 btrfs_release_extent_buffer(eb);
5026 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
5028 struct extent_buffer *eb =
5029 container_of(head, struct extent_buffer, rcu_head);
5031 __free_extent_buffer(eb);
5034 static int release_extent_buffer(struct extent_buffer *eb)
5036 lockdep_assert_held(&eb->refs_lock);
5038 WARN_ON(atomic_read(&eb->refs) == 0);
5039 if (atomic_dec_and_test(&eb->refs)) {
5040 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
5041 struct btrfs_fs_info *fs_info = eb->fs_info;
5043 spin_unlock(&eb->refs_lock);
5045 spin_lock(&fs_info->buffer_lock);
5046 radix_tree_delete(&fs_info->buffer_radix,
5047 eb->start >> PAGE_SHIFT);
5048 spin_unlock(&fs_info->buffer_lock);
5050 spin_unlock(&eb->refs_lock);
5053 /* Should be safe to release our pages at this point */
5054 btrfs_release_extent_buffer_pages(eb);
5055 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5056 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags))) {
5057 __free_extent_buffer(eb);
5061 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
5064 spin_unlock(&eb->refs_lock);
5069 void free_extent_buffer(struct extent_buffer *eb)
5077 refs = atomic_read(&eb->refs);
5078 if ((!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && refs <= 3)
5079 || (test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) &&
5082 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
5087 spin_lock(&eb->refs_lock);
5088 if (atomic_read(&eb->refs) == 2 &&
5089 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
5090 !extent_buffer_under_io(eb) &&
5091 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5092 atomic_dec(&eb->refs);
5095 * I know this is terrible, but it's temporary until we stop tracking
5096 * the uptodate bits and such for the extent buffers.
5098 release_extent_buffer(eb);
5101 void free_extent_buffer_stale(struct extent_buffer *eb)
5106 spin_lock(&eb->refs_lock);
5107 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
5109 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
5110 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5111 atomic_dec(&eb->refs);
5112 release_extent_buffer(eb);
5115 void clear_extent_buffer_dirty(struct extent_buffer *eb)
5121 num_pages = num_extent_pages(eb);
5123 for (i = 0; i < num_pages; i++) {
5124 page = eb->pages[i];
5125 if (!PageDirty(page))
5129 WARN_ON(!PagePrivate(page));
5131 clear_page_dirty_for_io(page);
5132 xa_lock_irq(&page->mapping->i_pages);
5133 if (!PageDirty(page))
5134 __xa_clear_mark(&page->mapping->i_pages,
5135 page_index(page), PAGECACHE_TAG_DIRTY);
5136 xa_unlock_irq(&page->mapping->i_pages);
5137 ClearPageError(page);
5140 WARN_ON(atomic_read(&eb->refs) == 0);
5143 bool set_extent_buffer_dirty(struct extent_buffer *eb)
5149 check_buffer_tree_ref(eb);
5151 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
5153 num_pages = num_extent_pages(eb);
5154 WARN_ON(atomic_read(&eb->refs) == 0);
5155 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
5158 for (i = 0; i < num_pages; i++)
5159 set_page_dirty(eb->pages[i]);
5161 #ifdef CONFIG_BTRFS_DEBUG
5162 for (i = 0; i < num_pages; i++)
5163 ASSERT(PageDirty(eb->pages[i]));
5169 void clear_extent_buffer_uptodate(struct extent_buffer *eb)
5175 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5176 num_pages = num_extent_pages(eb);
5177 for (i = 0; i < num_pages; i++) {
5178 page = eb->pages[i];
5180 ClearPageUptodate(page);
5184 void set_extent_buffer_uptodate(struct extent_buffer *eb)
5190 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5191 num_pages = num_extent_pages(eb);
5192 for (i = 0; i < num_pages; i++) {
5193 page = eb->pages[i];
5194 SetPageUptodate(page);
5198 int read_extent_buffer_pages(struct extent_io_tree *tree,
5199 struct extent_buffer *eb, int wait, int mirror_num)
5205 int locked_pages = 0;
5206 int all_uptodate = 1;
5208 unsigned long num_reads = 0;
5209 struct bio *bio = NULL;
5210 unsigned long bio_flags = 0;
5212 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
5215 num_pages = num_extent_pages(eb);
5216 for (i = 0; i < num_pages; i++) {
5217 page = eb->pages[i];
5218 if (wait == WAIT_NONE) {
5219 if (!trylock_page(page))
5227 * We need to firstly lock all pages to make sure that
5228 * the uptodate bit of our pages won't be affected by
5229 * clear_extent_buffer_uptodate().
5231 for (i = 0; i < num_pages; i++) {
5232 page = eb->pages[i];
5233 if (!PageUptodate(page)) {
5240 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5244 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
5245 eb->read_mirror = 0;
5246 atomic_set(&eb->io_pages, num_reads);
5247 for (i = 0; i < num_pages; i++) {
5248 page = eb->pages[i];
5250 if (!PageUptodate(page)) {
5252 atomic_dec(&eb->io_pages);
5257 ClearPageError(page);
5258 err = __extent_read_full_page(tree, page,
5259 btree_get_extent, &bio,
5260 mirror_num, &bio_flags,
5265 * We use &bio in above __extent_read_full_page,
5266 * so we ensure that if it returns error, the
5267 * current page fails to add itself to bio and
5268 * it's been unlocked.
5270 * We must dec io_pages by ourselves.
5272 atomic_dec(&eb->io_pages);
5280 err = submit_one_bio(bio, mirror_num, bio_flags);
5285 if (ret || wait != WAIT_COMPLETE)
5288 for (i = 0; i < num_pages; i++) {
5289 page = eb->pages[i];
5290 wait_on_page_locked(page);
5291 if (!PageUptodate(page))
5298 while (locked_pages > 0) {
5300 page = eb->pages[locked_pages];
5306 void read_extent_buffer(const struct extent_buffer *eb, void *dstv,
5307 unsigned long start, unsigned long len)
5313 char *dst = (char *)dstv;
5314 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5315 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5317 if (start + len > eb->len) {
5318 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n",
5319 eb->start, eb->len, start, len);
5320 memset(dst, 0, len);
5324 offset = (start_offset + start) & (PAGE_SIZE - 1);
5327 page = eb->pages[i];
5329 cur = min(len, (PAGE_SIZE - offset));
5330 kaddr = page_address(page);
5331 memcpy(dst, kaddr + offset, cur);
5340 int read_extent_buffer_to_user(const struct extent_buffer *eb,
5342 unsigned long start, unsigned long len)
5348 char __user *dst = (char __user *)dstv;
5349 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5350 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5353 WARN_ON(start > eb->len);
5354 WARN_ON(start + len > eb->start + eb->len);
5356 offset = (start_offset + start) & (PAGE_SIZE - 1);
5359 page = eb->pages[i];
5361 cur = min(len, (PAGE_SIZE - offset));
5362 kaddr = page_address(page);
5363 if (copy_to_user(dst, kaddr + offset, cur)) {
5378 * return 0 if the item is found within a page.
5379 * return 1 if the item spans two pages.
5380 * return -EINVAL otherwise.
5382 int map_private_extent_buffer(const struct extent_buffer *eb,
5383 unsigned long start, unsigned long min_len,
5384 char **map, unsigned long *map_start,
5385 unsigned long *map_len)
5387 size_t offset = start & (PAGE_SIZE - 1);
5390 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5391 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5392 unsigned long end_i = (start_offset + start + min_len - 1) >>
5395 if (start + min_len > eb->len) {
5396 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n",
5397 eb->start, eb->len, start, min_len);
5405 offset = start_offset;
5409 *map_start = ((u64)i << PAGE_SHIFT) - start_offset;
5413 kaddr = page_address(p);
5414 *map = kaddr + offset;
5415 *map_len = PAGE_SIZE - offset;
5419 int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,
5420 unsigned long start, unsigned long len)
5426 char *ptr = (char *)ptrv;
5427 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5428 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5431 WARN_ON(start > eb->len);
5432 WARN_ON(start + len > eb->start + eb->len);
5434 offset = (start_offset + start) & (PAGE_SIZE - 1);
5437 page = eb->pages[i];
5439 cur = min(len, (PAGE_SIZE - offset));
5441 kaddr = page_address(page);
5442 ret = memcmp(ptr, kaddr + offset, cur);
5454 void write_extent_buffer_chunk_tree_uuid(struct extent_buffer *eb,
5459 WARN_ON(!PageUptodate(eb->pages[0]));
5460 kaddr = page_address(eb->pages[0]);
5461 memcpy(kaddr + offsetof(struct btrfs_header, chunk_tree_uuid), srcv,
5465 void write_extent_buffer_fsid(struct extent_buffer *eb, const void *srcv)
5469 WARN_ON(!PageUptodate(eb->pages[0]));
5470 kaddr = page_address(eb->pages[0]);
5471 memcpy(kaddr + offsetof(struct btrfs_header, fsid), srcv,
5475 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
5476 unsigned long start, unsigned long len)
5482 char *src = (char *)srcv;
5483 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5484 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5486 WARN_ON(start > eb->len);
5487 WARN_ON(start + len > eb->start + eb->len);
5489 offset = (start_offset + start) & (PAGE_SIZE - 1);
5492 page = eb->pages[i];
5493 WARN_ON(!PageUptodate(page));
5495 cur = min(len, PAGE_SIZE - offset);
5496 kaddr = page_address(page);
5497 memcpy(kaddr + offset, src, cur);
5506 void memzero_extent_buffer(struct extent_buffer *eb, unsigned long start,
5513 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5514 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5516 WARN_ON(start > eb->len);
5517 WARN_ON(start + len > eb->start + eb->len);
5519 offset = (start_offset + start) & (PAGE_SIZE - 1);
5522 page = eb->pages[i];
5523 WARN_ON(!PageUptodate(page));
5525 cur = min(len, PAGE_SIZE - offset);
5526 kaddr = page_address(page);
5527 memset(kaddr + offset, 0, cur);
5535 void copy_extent_buffer_full(struct extent_buffer *dst,
5536 struct extent_buffer *src)
5541 ASSERT(dst->len == src->len);
5543 num_pages = num_extent_pages(dst);
5544 for (i = 0; i < num_pages; i++)
5545 copy_page(page_address(dst->pages[i]),
5546 page_address(src->pages[i]));
5549 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
5550 unsigned long dst_offset, unsigned long src_offset,
5553 u64 dst_len = dst->len;
5558 size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1);
5559 unsigned long i = (start_offset + dst_offset) >> PAGE_SHIFT;
5561 WARN_ON(src->len != dst_len);
5563 offset = (start_offset + dst_offset) &
5567 page = dst->pages[i];
5568 WARN_ON(!PageUptodate(page));
5570 cur = min(len, (unsigned long)(PAGE_SIZE - offset));
5572 kaddr = page_address(page);
5573 read_extent_buffer(src, kaddr + offset, src_offset, cur);
5583 * eb_bitmap_offset() - calculate the page and offset of the byte containing the
5585 * @eb: the extent buffer
5586 * @start: offset of the bitmap item in the extent buffer
5588 * @page_index: return index of the page in the extent buffer that contains the
5590 * @page_offset: return offset into the page given by page_index
5592 * This helper hides the ugliness of finding the byte in an extent buffer which
5593 * contains a given bit.
5595 static inline void eb_bitmap_offset(struct extent_buffer *eb,
5596 unsigned long start, unsigned long nr,
5597 unsigned long *page_index,
5598 size_t *page_offset)
5600 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5601 size_t byte_offset = BIT_BYTE(nr);
5605 * The byte we want is the offset of the extent buffer + the offset of
5606 * the bitmap item in the extent buffer + the offset of the byte in the
5609 offset = start_offset + start + byte_offset;
5611 *page_index = offset >> PAGE_SHIFT;
5612 *page_offset = offset & (PAGE_SIZE - 1);
5616 * extent_buffer_test_bit - determine whether a bit in a bitmap item is set
5617 * @eb: the extent buffer
5618 * @start: offset of the bitmap item in the extent buffer
5619 * @nr: bit number to test
5621 int extent_buffer_test_bit(struct extent_buffer *eb, unsigned long start,
5629 eb_bitmap_offset(eb, start, nr, &i, &offset);
5630 page = eb->pages[i];
5631 WARN_ON(!PageUptodate(page));
5632 kaddr = page_address(page);
5633 return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1)));
5637 * extent_buffer_bitmap_set - set an area of a bitmap
5638 * @eb: the extent buffer
5639 * @start: offset of the bitmap item in the extent buffer
5640 * @pos: bit number of the first bit
5641 * @len: number of bits to set
5643 void extent_buffer_bitmap_set(struct extent_buffer *eb, unsigned long start,
5644 unsigned long pos, unsigned long len)
5650 const unsigned int size = pos + len;
5651 int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5652 u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos);
5654 eb_bitmap_offset(eb, start, pos, &i, &offset);
5655 page = eb->pages[i];
5656 WARN_ON(!PageUptodate(page));
5657 kaddr = page_address(page);
5659 while (len >= bits_to_set) {
5660 kaddr[offset] |= mask_to_set;
5662 bits_to_set = BITS_PER_BYTE;
5664 if (++offset >= PAGE_SIZE && len > 0) {
5666 page = eb->pages[++i];
5667 WARN_ON(!PageUptodate(page));
5668 kaddr = page_address(page);
5672 mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
5673 kaddr[offset] |= mask_to_set;
5679 * extent_buffer_bitmap_clear - clear an area of a bitmap
5680 * @eb: the extent buffer
5681 * @start: offset of the bitmap item in the extent buffer
5682 * @pos: bit number of the first bit
5683 * @len: number of bits to clear
5685 void extent_buffer_bitmap_clear(struct extent_buffer *eb, unsigned long start,
5686 unsigned long pos, unsigned long len)
5692 const unsigned int size = pos + len;
5693 int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5694 u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos);
5696 eb_bitmap_offset(eb, start, pos, &i, &offset);
5697 page = eb->pages[i];
5698 WARN_ON(!PageUptodate(page));
5699 kaddr = page_address(page);
5701 while (len >= bits_to_clear) {
5702 kaddr[offset] &= ~mask_to_clear;
5703 len -= bits_to_clear;
5704 bits_to_clear = BITS_PER_BYTE;
5706 if (++offset >= PAGE_SIZE && len > 0) {
5708 page = eb->pages[++i];
5709 WARN_ON(!PageUptodate(page));
5710 kaddr = page_address(page);
5714 mask_to_clear &= BITMAP_LAST_BYTE_MASK(size);
5715 kaddr[offset] &= ~mask_to_clear;
5719 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
5721 unsigned long distance = (src > dst) ? src - dst : dst - src;
5722 return distance < len;
5725 static void copy_pages(struct page *dst_page, struct page *src_page,
5726 unsigned long dst_off, unsigned long src_off,
5729 char *dst_kaddr = page_address(dst_page);
5731 int must_memmove = 0;
5733 if (dst_page != src_page) {
5734 src_kaddr = page_address(src_page);
5736 src_kaddr = dst_kaddr;
5737 if (areas_overlap(src_off, dst_off, len))
5742 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
5744 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
5747 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5748 unsigned long src_offset, unsigned long len)
5750 struct btrfs_fs_info *fs_info = dst->fs_info;
5752 size_t dst_off_in_page;
5753 size_t src_off_in_page;
5754 size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1);
5755 unsigned long dst_i;
5756 unsigned long src_i;
5758 if (src_offset + len > dst->len) {
5760 "memmove bogus src_offset %lu move len %lu dst len %lu",
5761 src_offset, len, dst->len);
5764 if (dst_offset + len > dst->len) {
5766 "memmove bogus dst_offset %lu move len %lu dst len %lu",
5767 dst_offset, len, dst->len);
5772 dst_off_in_page = (start_offset + dst_offset) &
5774 src_off_in_page = (start_offset + src_offset) &
5777 dst_i = (start_offset + dst_offset) >> PAGE_SHIFT;
5778 src_i = (start_offset + src_offset) >> PAGE_SHIFT;
5780 cur = min(len, (unsigned long)(PAGE_SIZE -
5782 cur = min_t(unsigned long, cur,
5783 (unsigned long)(PAGE_SIZE - dst_off_in_page));
5785 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5786 dst_off_in_page, src_off_in_page, cur);
5794 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5795 unsigned long src_offset, unsigned long len)
5797 struct btrfs_fs_info *fs_info = dst->fs_info;
5799 size_t dst_off_in_page;
5800 size_t src_off_in_page;
5801 unsigned long dst_end = dst_offset + len - 1;
5802 unsigned long src_end = src_offset + len - 1;
5803 size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1);
5804 unsigned long dst_i;
5805 unsigned long src_i;
5807 if (src_offset + len > dst->len) {
5809 "memmove bogus src_offset %lu move len %lu len %lu",
5810 src_offset, len, dst->len);
5813 if (dst_offset + len > dst->len) {
5815 "memmove bogus dst_offset %lu move len %lu len %lu",
5816 dst_offset, len, dst->len);
5819 if (dst_offset < src_offset) {
5820 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
5824 dst_i = (start_offset + dst_end) >> PAGE_SHIFT;
5825 src_i = (start_offset + src_end) >> PAGE_SHIFT;
5827 dst_off_in_page = (start_offset + dst_end) &
5829 src_off_in_page = (start_offset + src_end) &
5832 cur = min_t(unsigned long, len, src_off_in_page + 1);
5833 cur = min(cur, dst_off_in_page + 1);
5834 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5835 dst_off_in_page - cur + 1,
5836 src_off_in_page - cur + 1, cur);
5844 int try_release_extent_buffer(struct page *page)
5846 struct extent_buffer *eb;
5849 * We need to make sure nobody is attaching this page to an eb right
5852 spin_lock(&page->mapping->private_lock);
5853 if (!PagePrivate(page)) {
5854 spin_unlock(&page->mapping->private_lock);
5858 eb = (struct extent_buffer *)page->private;
5862 * This is a little awful but should be ok, we need to make sure that
5863 * the eb doesn't disappear out from under us while we're looking at
5866 spin_lock(&eb->refs_lock);
5867 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
5868 spin_unlock(&eb->refs_lock);
5869 spin_unlock(&page->mapping->private_lock);
5872 spin_unlock(&page->mapping->private_lock);
5875 * If tree ref isn't set then we know the ref on this eb is a real ref,
5876 * so just return, this page will likely be freed soon anyway.
5878 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
5879 spin_unlock(&eb->refs_lock);
5883 return release_extent_buffer(eb);