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 anyone 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 * true is returned if we find something, false if nothing was in the tree
1457 static noinline bool 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);
1480 state = rb_entry(node, struct extent_state, rb_node);
1481 if (found && (state->start != cur_start ||
1482 (state->state & EXTENT_BOUNDARY))) {
1485 if (!(state->state & EXTENT_DELALLOC)) {
1491 *start = state->start;
1492 *cached_state = state;
1493 refcount_inc(&state->refs);
1497 cur_start = state->end + 1;
1498 node = rb_next(node);
1499 total_bytes += state->end - state->start + 1;
1500 if (total_bytes >= max_bytes)
1506 spin_unlock(&tree->lock);
1510 static int __process_pages_contig(struct address_space *mapping,
1511 struct page *locked_page,
1512 pgoff_t start_index, pgoff_t end_index,
1513 unsigned long page_ops, pgoff_t *index_ret);
1515 static noinline void __unlock_for_delalloc(struct inode *inode,
1516 struct page *locked_page,
1519 unsigned long index = start >> PAGE_SHIFT;
1520 unsigned long end_index = end >> PAGE_SHIFT;
1522 ASSERT(locked_page);
1523 if (index == locked_page->index && end_index == index)
1526 __process_pages_contig(inode->i_mapping, locked_page, index, end_index,
1530 static noinline int lock_delalloc_pages(struct inode *inode,
1531 struct page *locked_page,
1535 unsigned long index = delalloc_start >> PAGE_SHIFT;
1536 unsigned long index_ret = index;
1537 unsigned long end_index = delalloc_end >> PAGE_SHIFT;
1540 ASSERT(locked_page);
1541 if (index == locked_page->index && index == end_index)
1544 ret = __process_pages_contig(inode->i_mapping, locked_page, index,
1545 end_index, PAGE_LOCK, &index_ret);
1547 __unlock_for_delalloc(inode, locked_page, delalloc_start,
1548 (u64)index_ret << PAGE_SHIFT);
1553 * Find and lock a contiguous range of bytes in the file marked as delalloc, no
1554 * more than @max_bytes. @Start and @end are used to return the range,
1556 * Return: true if we find something
1557 * false if nothing was in the tree
1560 noinline_for_stack bool find_lock_delalloc_range(struct inode *inode,
1561 struct extent_io_tree *tree,
1562 struct page *locked_page, u64 *start,
1565 u64 max_bytes = BTRFS_MAX_EXTENT_SIZE;
1569 struct extent_state *cached_state = NULL;
1574 /* step one, find a bunch of delalloc bytes starting at start */
1575 delalloc_start = *start;
1577 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1578 max_bytes, &cached_state);
1579 if (!found || delalloc_end <= *start) {
1580 *start = delalloc_start;
1581 *end = delalloc_end;
1582 free_extent_state(cached_state);
1587 * start comes from the offset of locked_page. We have to lock
1588 * pages in order, so we can't process delalloc bytes before
1591 if (delalloc_start < *start)
1592 delalloc_start = *start;
1595 * make sure to limit the number of pages we try to lock down
1597 if (delalloc_end + 1 - delalloc_start > max_bytes)
1598 delalloc_end = delalloc_start + max_bytes - 1;
1600 /* step two, lock all the pages after the page that has start */
1601 ret = lock_delalloc_pages(inode, locked_page,
1602 delalloc_start, delalloc_end);
1603 ASSERT(!ret || ret == -EAGAIN);
1604 if (ret == -EAGAIN) {
1605 /* some of the pages are gone, lets avoid looping by
1606 * shortening the size of the delalloc range we're searching
1608 free_extent_state(cached_state);
1609 cached_state = NULL;
1611 max_bytes = PAGE_SIZE;
1620 /* step three, lock the state bits for the whole range */
1621 lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state);
1623 /* then test to make sure it is all still delalloc */
1624 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1625 EXTENT_DELALLOC, 1, cached_state);
1627 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1629 __unlock_for_delalloc(inode, locked_page,
1630 delalloc_start, delalloc_end);
1634 free_extent_state(cached_state);
1635 *start = delalloc_start;
1636 *end = delalloc_end;
1641 static int __process_pages_contig(struct address_space *mapping,
1642 struct page *locked_page,
1643 pgoff_t start_index, pgoff_t end_index,
1644 unsigned long page_ops, pgoff_t *index_ret)
1646 unsigned long nr_pages = end_index - start_index + 1;
1647 unsigned long pages_locked = 0;
1648 pgoff_t index = start_index;
1649 struct page *pages[16];
1654 if (page_ops & PAGE_LOCK) {
1655 ASSERT(page_ops == PAGE_LOCK);
1656 ASSERT(index_ret && *index_ret == start_index);
1659 if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0)
1660 mapping_set_error(mapping, -EIO);
1662 while (nr_pages > 0) {
1663 ret = find_get_pages_contig(mapping, index,
1664 min_t(unsigned long,
1665 nr_pages, ARRAY_SIZE(pages)), pages);
1668 * Only if we're going to lock these pages,
1669 * can we find nothing at @index.
1671 ASSERT(page_ops & PAGE_LOCK);
1676 for (i = 0; i < ret; i++) {
1677 if (page_ops & PAGE_SET_PRIVATE2)
1678 SetPagePrivate2(pages[i]);
1680 if (pages[i] == locked_page) {
1685 if (page_ops & PAGE_CLEAR_DIRTY)
1686 clear_page_dirty_for_io(pages[i]);
1687 if (page_ops & PAGE_SET_WRITEBACK)
1688 set_page_writeback(pages[i]);
1689 if (page_ops & PAGE_SET_ERROR)
1690 SetPageError(pages[i]);
1691 if (page_ops & PAGE_END_WRITEBACK)
1692 end_page_writeback(pages[i]);
1693 if (page_ops & PAGE_UNLOCK)
1694 unlock_page(pages[i]);
1695 if (page_ops & PAGE_LOCK) {
1696 lock_page(pages[i]);
1697 if (!PageDirty(pages[i]) ||
1698 pages[i]->mapping != mapping) {
1699 unlock_page(pages[i]);
1713 if (err && index_ret)
1714 *index_ret = start_index + pages_locked - 1;
1718 void extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end,
1719 u64 delalloc_end, struct page *locked_page,
1720 unsigned clear_bits,
1721 unsigned long page_ops)
1723 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, clear_bits, 1, 0,
1726 __process_pages_contig(inode->i_mapping, locked_page,
1727 start >> PAGE_SHIFT, end >> PAGE_SHIFT,
1732 * count the number of bytes in the tree that have a given bit(s)
1733 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1734 * cached. The total number found is returned.
1736 u64 count_range_bits(struct extent_io_tree *tree,
1737 u64 *start, u64 search_end, u64 max_bytes,
1738 unsigned bits, int contig)
1740 struct rb_node *node;
1741 struct extent_state *state;
1742 u64 cur_start = *start;
1743 u64 total_bytes = 0;
1747 if (WARN_ON(search_end <= cur_start))
1750 spin_lock(&tree->lock);
1751 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1752 total_bytes = tree->dirty_bytes;
1756 * this search will find all the extents that end after
1759 node = tree_search(tree, cur_start);
1764 state = rb_entry(node, struct extent_state, rb_node);
1765 if (state->start > search_end)
1767 if (contig && found && state->start > last + 1)
1769 if (state->end >= cur_start && (state->state & bits) == bits) {
1770 total_bytes += min(search_end, state->end) + 1 -
1771 max(cur_start, state->start);
1772 if (total_bytes >= max_bytes)
1775 *start = max(cur_start, state->start);
1779 } else if (contig && found) {
1782 node = rb_next(node);
1787 spin_unlock(&tree->lock);
1792 * set the private field for a given byte offset in the tree. If there isn't
1793 * an extent_state there already, this does nothing.
1795 static noinline int set_state_failrec(struct extent_io_tree *tree, u64 start,
1796 struct io_failure_record *failrec)
1798 struct rb_node *node;
1799 struct extent_state *state;
1802 spin_lock(&tree->lock);
1804 * this search will find all the extents that end after
1807 node = tree_search(tree, start);
1812 state = rb_entry(node, struct extent_state, rb_node);
1813 if (state->start != start) {
1817 state->failrec = failrec;
1819 spin_unlock(&tree->lock);
1823 static noinline int get_state_failrec(struct extent_io_tree *tree, u64 start,
1824 struct io_failure_record **failrec)
1826 struct rb_node *node;
1827 struct extent_state *state;
1830 spin_lock(&tree->lock);
1832 * this search will find all the extents that end after
1835 node = tree_search(tree, start);
1840 state = rb_entry(node, struct extent_state, rb_node);
1841 if (state->start != start) {
1845 *failrec = state->failrec;
1847 spin_unlock(&tree->lock);
1852 * searches a range in the state tree for a given mask.
1853 * If 'filled' == 1, this returns 1 only if every extent in the tree
1854 * has the bits set. Otherwise, 1 is returned if any bit in the
1855 * range is found set.
1857 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1858 unsigned bits, int filled, struct extent_state *cached)
1860 struct extent_state *state = NULL;
1861 struct rb_node *node;
1864 spin_lock(&tree->lock);
1865 if (cached && extent_state_in_tree(cached) && cached->start <= start &&
1866 cached->end > start)
1867 node = &cached->rb_node;
1869 node = tree_search(tree, start);
1870 while (node && start <= end) {
1871 state = rb_entry(node, struct extent_state, rb_node);
1873 if (filled && state->start > start) {
1878 if (state->start > end)
1881 if (state->state & bits) {
1885 } else if (filled) {
1890 if (state->end == (u64)-1)
1893 start = state->end + 1;
1896 node = rb_next(node);
1903 spin_unlock(&tree->lock);
1908 * helper function to set a given page up to date if all the
1909 * extents in the tree for that page are up to date
1911 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
1913 u64 start = page_offset(page);
1914 u64 end = start + PAGE_SIZE - 1;
1915 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1916 SetPageUptodate(page);
1919 int free_io_failure(struct extent_io_tree *failure_tree,
1920 struct extent_io_tree *io_tree,
1921 struct io_failure_record *rec)
1926 set_state_failrec(failure_tree, rec->start, NULL);
1927 ret = clear_extent_bits(failure_tree, rec->start,
1928 rec->start + rec->len - 1,
1929 EXTENT_LOCKED | EXTENT_DIRTY);
1933 ret = clear_extent_bits(io_tree, rec->start,
1934 rec->start + rec->len - 1,
1944 * this bypasses the standard btrfs submit functions deliberately, as
1945 * the standard behavior is to write all copies in a raid setup. here we only
1946 * want to write the one bad copy. so we do the mapping for ourselves and issue
1947 * submit_bio directly.
1948 * to avoid any synchronization issues, wait for the data after writing, which
1949 * actually prevents the read that triggered the error from finishing.
1950 * currently, there can be no more than two copies of every data bit. thus,
1951 * exactly one rewrite is required.
1953 int repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start,
1954 u64 length, u64 logical, struct page *page,
1955 unsigned int pg_offset, int mirror_num)
1958 struct btrfs_device *dev;
1961 struct btrfs_bio *bbio = NULL;
1964 ASSERT(!(fs_info->sb->s_flags & SB_RDONLY));
1965 BUG_ON(!mirror_num);
1967 bio = btrfs_io_bio_alloc(1);
1968 bio->bi_iter.bi_size = 0;
1969 map_length = length;
1972 * Avoid races with device replace and make sure our bbio has devices
1973 * associated to its stripes that don't go away while we are doing the
1974 * read repair operation.
1976 btrfs_bio_counter_inc_blocked(fs_info);
1977 if (btrfs_is_parity_mirror(fs_info, logical, length)) {
1979 * Note that we don't use BTRFS_MAP_WRITE because it's supposed
1980 * to update all raid stripes, but here we just want to correct
1981 * bad stripe, thus BTRFS_MAP_READ is abused to only get the bad
1982 * stripe's dev and sector.
1984 ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, logical,
1985 &map_length, &bbio, 0);
1987 btrfs_bio_counter_dec(fs_info);
1991 ASSERT(bbio->mirror_num == 1);
1993 ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical,
1994 &map_length, &bbio, mirror_num);
1996 btrfs_bio_counter_dec(fs_info);
2000 BUG_ON(mirror_num != bbio->mirror_num);
2003 sector = bbio->stripes[bbio->mirror_num - 1].physical >> 9;
2004 bio->bi_iter.bi_sector = sector;
2005 dev = bbio->stripes[bbio->mirror_num - 1].dev;
2006 btrfs_put_bbio(bbio);
2007 if (!dev || !dev->bdev ||
2008 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) {
2009 btrfs_bio_counter_dec(fs_info);
2013 bio_set_dev(bio, dev->bdev);
2014 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC;
2015 bio_add_page(bio, page, length, pg_offset);
2017 if (btrfsic_submit_bio_wait(bio)) {
2018 /* try to remap that extent elsewhere? */
2019 btrfs_bio_counter_dec(fs_info);
2021 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
2025 btrfs_info_rl_in_rcu(fs_info,
2026 "read error corrected: ino %llu off %llu (dev %s sector %llu)",
2028 rcu_str_deref(dev->name), sector);
2029 btrfs_bio_counter_dec(fs_info);
2034 int repair_eb_io_failure(struct btrfs_fs_info *fs_info,
2035 struct extent_buffer *eb, int mirror_num)
2037 u64 start = eb->start;
2038 int i, num_pages = num_extent_pages(eb);
2041 if (sb_rdonly(fs_info->sb))
2044 for (i = 0; i < num_pages; i++) {
2045 struct page *p = eb->pages[i];
2047 ret = repair_io_failure(fs_info, 0, start, PAGE_SIZE, start, p,
2048 start - page_offset(p), mirror_num);
2058 * each time an IO finishes, we do a fast check in the IO failure tree
2059 * to see if we need to process or clean up an io_failure_record
2061 int clean_io_failure(struct btrfs_fs_info *fs_info,
2062 struct extent_io_tree *failure_tree,
2063 struct extent_io_tree *io_tree, u64 start,
2064 struct page *page, u64 ino, unsigned int pg_offset)
2067 struct io_failure_record *failrec;
2068 struct extent_state *state;
2073 ret = count_range_bits(failure_tree, &private, (u64)-1, 1,
2078 ret = get_state_failrec(failure_tree, start, &failrec);
2082 BUG_ON(!failrec->this_mirror);
2084 if (failrec->in_validation) {
2085 /* there was no real error, just free the record */
2086 btrfs_debug(fs_info,
2087 "clean_io_failure: freeing dummy error at %llu",
2091 if (sb_rdonly(fs_info->sb))
2094 spin_lock(&io_tree->lock);
2095 state = find_first_extent_bit_state(io_tree,
2098 spin_unlock(&io_tree->lock);
2100 if (state && state->start <= failrec->start &&
2101 state->end >= failrec->start + failrec->len - 1) {
2102 num_copies = btrfs_num_copies(fs_info, failrec->logical,
2104 if (num_copies > 1) {
2105 repair_io_failure(fs_info, ino, start, failrec->len,
2106 failrec->logical, page, pg_offset,
2107 failrec->failed_mirror);
2112 free_io_failure(failure_tree, io_tree, failrec);
2118 * Can be called when
2119 * - hold extent lock
2120 * - under ordered extent
2121 * - the inode is freeing
2123 void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start, u64 end)
2125 struct extent_io_tree *failure_tree = &inode->io_failure_tree;
2126 struct io_failure_record *failrec;
2127 struct extent_state *state, *next;
2129 if (RB_EMPTY_ROOT(&failure_tree->state))
2132 spin_lock(&failure_tree->lock);
2133 state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY);
2135 if (state->start > end)
2138 ASSERT(state->end <= end);
2140 next = next_state(state);
2142 failrec = state->failrec;
2143 free_extent_state(state);
2148 spin_unlock(&failure_tree->lock);
2151 int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end,
2152 struct io_failure_record **failrec_ret)
2154 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2155 struct io_failure_record *failrec;
2156 struct extent_map *em;
2157 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2158 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2159 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2163 ret = get_state_failrec(failure_tree, start, &failrec);
2165 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2169 failrec->start = start;
2170 failrec->len = end - start + 1;
2171 failrec->this_mirror = 0;
2172 failrec->bio_flags = 0;
2173 failrec->in_validation = 0;
2175 read_lock(&em_tree->lock);
2176 em = lookup_extent_mapping(em_tree, start, failrec->len);
2178 read_unlock(&em_tree->lock);
2183 if (em->start > start || em->start + em->len <= start) {
2184 free_extent_map(em);
2187 read_unlock(&em_tree->lock);
2193 logical = start - em->start;
2194 logical = em->block_start + logical;
2195 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2196 logical = em->block_start;
2197 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2198 extent_set_compress_type(&failrec->bio_flags,
2202 btrfs_debug(fs_info,
2203 "Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu",
2204 logical, start, failrec->len);
2206 failrec->logical = logical;
2207 free_extent_map(em);
2209 /* set the bits in the private failure tree */
2210 ret = set_extent_bits(failure_tree, start, end,
2211 EXTENT_LOCKED | EXTENT_DIRTY);
2213 ret = set_state_failrec(failure_tree, start, failrec);
2214 /* set the bits in the inode's tree */
2216 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED);
2222 btrfs_debug(fs_info,
2223 "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d",
2224 failrec->logical, failrec->start, failrec->len,
2225 failrec->in_validation);
2227 * when data can be on disk more than twice, add to failrec here
2228 * (e.g. with a list for failed_mirror) to make
2229 * clean_io_failure() clean all those errors at once.
2233 *failrec_ret = failrec;
2238 bool btrfs_check_repairable(struct inode *inode, unsigned failed_bio_pages,
2239 struct io_failure_record *failrec, int failed_mirror)
2241 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2244 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
2245 if (num_copies == 1) {
2247 * we only have a single copy of the data, so don't bother with
2248 * all the retry and error correction code that follows. no
2249 * matter what the error is, it is very likely to persist.
2251 btrfs_debug(fs_info,
2252 "Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
2253 num_copies, failrec->this_mirror, failed_mirror);
2258 * there are two premises:
2259 * a) deliver good data to the caller
2260 * b) correct the bad sectors on disk
2262 if (failed_bio_pages > 1) {
2264 * to fulfill b), we need to know the exact failing sectors, as
2265 * we don't want to rewrite any more than the failed ones. thus,
2266 * we need separate read requests for the failed bio
2268 * if the following BUG_ON triggers, our validation request got
2269 * merged. we need separate requests for our algorithm to work.
2271 BUG_ON(failrec->in_validation);
2272 failrec->in_validation = 1;
2273 failrec->this_mirror = failed_mirror;
2276 * we're ready to fulfill a) and b) alongside. get a good copy
2277 * of the failed sector and if we succeed, we have setup
2278 * everything for repair_io_failure to do the rest for us.
2280 if (failrec->in_validation) {
2281 BUG_ON(failrec->this_mirror != failed_mirror);
2282 failrec->in_validation = 0;
2283 failrec->this_mirror = 0;
2285 failrec->failed_mirror = failed_mirror;
2286 failrec->this_mirror++;
2287 if (failrec->this_mirror == failed_mirror)
2288 failrec->this_mirror++;
2291 if (failrec->this_mirror > num_copies) {
2292 btrfs_debug(fs_info,
2293 "Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
2294 num_copies, failrec->this_mirror, failed_mirror);
2302 struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio,
2303 struct io_failure_record *failrec,
2304 struct page *page, int pg_offset, int icsum,
2305 bio_end_io_t *endio_func, void *data)
2307 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2309 struct btrfs_io_bio *btrfs_failed_bio;
2310 struct btrfs_io_bio *btrfs_bio;
2312 bio = btrfs_io_bio_alloc(1);
2313 bio->bi_end_io = endio_func;
2314 bio->bi_iter.bi_sector = failrec->logical >> 9;
2315 bio_set_dev(bio, fs_info->fs_devices->latest_bdev);
2316 bio->bi_iter.bi_size = 0;
2317 bio->bi_private = data;
2319 btrfs_failed_bio = btrfs_io_bio(failed_bio);
2320 if (btrfs_failed_bio->csum) {
2321 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
2323 btrfs_bio = btrfs_io_bio(bio);
2324 btrfs_bio->csum = btrfs_bio->csum_inline;
2326 memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + icsum,
2330 bio_add_page(bio, page, failrec->len, pg_offset);
2336 * This is a generic handler for readpage errors. If other copies exist, read
2337 * those and write back good data to the failed position. Does not investigate
2338 * in remapping the failed extent elsewhere, hoping the device will be smart
2339 * enough to do this as needed
2341 static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset,
2342 struct page *page, u64 start, u64 end,
2345 struct io_failure_record *failrec;
2346 struct inode *inode = page->mapping->host;
2347 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2348 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2351 blk_status_t status;
2353 unsigned failed_bio_pages = bio_pages_all(failed_bio);
2355 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
2357 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
2361 if (!btrfs_check_repairable(inode, failed_bio_pages, failrec,
2363 free_io_failure(failure_tree, tree, failrec);
2367 if (failed_bio_pages > 1)
2368 read_mode |= REQ_FAILFAST_DEV;
2370 phy_offset >>= inode->i_sb->s_blocksize_bits;
2371 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
2372 start - page_offset(page),
2373 (int)phy_offset, failed_bio->bi_end_io,
2375 bio->bi_opf = REQ_OP_READ | read_mode;
2377 btrfs_debug(btrfs_sb(inode->i_sb),
2378 "Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d",
2379 read_mode, failrec->this_mirror, failrec->in_validation);
2381 status = tree->ops->submit_bio_hook(tree->private_data, bio, failrec->this_mirror,
2382 failrec->bio_flags, 0);
2384 free_io_failure(failure_tree, tree, failrec);
2386 ret = blk_status_to_errno(status);
2392 /* lots and lots of room for performance fixes in the end_bio funcs */
2394 void end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2396 int uptodate = (err == 0);
2399 btrfs_writepage_endio_finish_ordered(page, start, end, uptodate);
2402 ClearPageUptodate(page);
2404 ret = err < 0 ? err : -EIO;
2405 mapping_set_error(page->mapping, ret);
2410 * after a writepage IO is done, we need to:
2411 * clear the uptodate bits on error
2412 * clear the writeback bits in the extent tree for this IO
2413 * end_page_writeback if the page has no more pending IO
2415 * Scheduling is not allowed, so the extent state tree is expected
2416 * to have one and only one object corresponding to this IO.
2418 static void end_bio_extent_writepage(struct bio *bio)
2420 int error = blk_status_to_errno(bio->bi_status);
2421 struct bio_vec *bvec;
2426 ASSERT(!bio_flagged(bio, BIO_CLONED));
2427 bio_for_each_segment_all(bvec, bio, i) {
2428 struct page *page = bvec->bv_page;
2429 struct inode *inode = page->mapping->host;
2430 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2432 /* We always issue full-page reads, but if some block
2433 * in a page fails to read, blk_update_request() will
2434 * advance bv_offset and adjust bv_len to compensate.
2435 * Print a warning for nonzero offsets, and an error
2436 * if they don't add up to a full page. */
2437 if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) {
2438 if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE)
2440 "partial page write in btrfs with offset %u and length %u",
2441 bvec->bv_offset, bvec->bv_len);
2444 "incomplete page write in btrfs with offset %u and length %u",
2445 bvec->bv_offset, bvec->bv_len);
2448 start = page_offset(page);
2449 end = start + bvec->bv_offset + bvec->bv_len - 1;
2451 end_extent_writepage(page, error, start, end);
2452 end_page_writeback(page);
2459 endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len,
2462 struct extent_state *cached = NULL;
2463 u64 end = start + len - 1;
2465 if (uptodate && tree->track_uptodate)
2466 set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC);
2467 unlock_extent_cached_atomic(tree, start, end, &cached);
2471 * after a readpage IO is done, we need to:
2472 * clear the uptodate bits on error
2473 * set the uptodate bits if things worked
2474 * set the page up to date if all extents in the tree are uptodate
2475 * clear the lock bit in the extent tree
2476 * unlock the page if there are no other extents locked for it
2478 * Scheduling is not allowed, so the extent state tree is expected
2479 * to have one and only one object corresponding to this IO.
2481 static void end_bio_extent_readpage(struct bio *bio)
2483 struct bio_vec *bvec;
2484 int uptodate = !bio->bi_status;
2485 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
2486 struct extent_io_tree *tree, *failure_tree;
2491 u64 extent_start = 0;
2497 ASSERT(!bio_flagged(bio, BIO_CLONED));
2498 bio_for_each_segment_all(bvec, bio, i) {
2499 struct page *page = bvec->bv_page;
2500 struct inode *inode = page->mapping->host;
2501 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2502 bool data_inode = btrfs_ino(BTRFS_I(inode))
2503 != BTRFS_BTREE_INODE_OBJECTID;
2505 btrfs_debug(fs_info,
2506 "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u",
2507 (u64)bio->bi_iter.bi_sector, bio->bi_status,
2508 io_bio->mirror_num);
2509 tree = &BTRFS_I(inode)->io_tree;
2510 failure_tree = &BTRFS_I(inode)->io_failure_tree;
2512 /* We always issue full-page reads, but if some block
2513 * in a page fails to read, blk_update_request() will
2514 * advance bv_offset and adjust bv_len to compensate.
2515 * Print a warning for nonzero offsets, and an error
2516 * if they don't add up to a full page. */
2517 if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) {
2518 if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE)
2520 "partial page read in btrfs with offset %u and length %u",
2521 bvec->bv_offset, bvec->bv_len);
2524 "incomplete page read in btrfs with offset %u and length %u",
2525 bvec->bv_offset, bvec->bv_len);
2528 start = page_offset(page);
2529 end = start + bvec->bv_offset + bvec->bv_len - 1;
2532 mirror = io_bio->mirror_num;
2533 if (likely(uptodate)) {
2534 ret = tree->ops->readpage_end_io_hook(io_bio, offset,
2540 clean_io_failure(BTRFS_I(inode)->root->fs_info,
2541 failure_tree, tree, start,
2543 btrfs_ino(BTRFS_I(inode)), 0);
2546 if (likely(uptodate))
2552 * The generic bio_readpage_error handles errors the
2553 * following way: If possible, new read requests are
2554 * created and submitted and will end up in
2555 * end_bio_extent_readpage as well (if we're lucky,
2556 * not in the !uptodate case). In that case it returns
2557 * 0 and we just go on with the next page in our bio.
2558 * If it can't handle the error it will return -EIO and
2559 * we remain responsible for that page.
2561 ret = bio_readpage_error(bio, offset, page, start, end,
2564 uptodate = !bio->bi_status;
2569 struct extent_buffer *eb;
2571 eb = (struct extent_buffer *)page->private;
2572 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
2573 eb->read_mirror = mirror;
2574 atomic_dec(&eb->io_pages);
2575 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD,
2577 btree_readahead_hook(eb, -EIO);
2582 if (likely(uptodate)) {
2583 loff_t i_size = i_size_read(inode);
2584 pgoff_t end_index = i_size >> PAGE_SHIFT;
2587 /* Zero out the end if this page straddles i_size */
2588 off = offset_in_page(i_size);
2589 if (page->index == end_index && off)
2590 zero_user_segment(page, off, PAGE_SIZE);
2591 SetPageUptodate(page);
2593 ClearPageUptodate(page);
2599 if (unlikely(!uptodate)) {
2601 endio_readpage_release_extent(tree,
2607 endio_readpage_release_extent(tree, start,
2608 end - start + 1, 0);
2609 } else if (!extent_len) {
2610 extent_start = start;
2611 extent_len = end + 1 - start;
2612 } else if (extent_start + extent_len == start) {
2613 extent_len += end + 1 - start;
2615 endio_readpage_release_extent(tree, extent_start,
2616 extent_len, uptodate);
2617 extent_start = start;
2618 extent_len = end + 1 - start;
2623 endio_readpage_release_extent(tree, extent_start, extent_len,
2625 btrfs_io_bio_free_csum(io_bio);
2630 * Initialize the members up to but not including 'bio'. Use after allocating a
2631 * new bio by bio_alloc_bioset as it does not initialize the bytes outside of
2632 * 'bio' because use of __GFP_ZERO is not supported.
2634 static inline void btrfs_io_bio_init(struct btrfs_io_bio *btrfs_bio)
2636 memset(btrfs_bio, 0, offsetof(struct btrfs_io_bio, bio));
2640 * The following helpers allocate a bio. As it's backed by a bioset, it'll
2641 * never fail. We're returning a bio right now but you can call btrfs_io_bio
2642 * for the appropriate container_of magic
2644 struct bio *btrfs_bio_alloc(struct block_device *bdev, u64 first_byte)
2648 bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, &btrfs_bioset);
2649 bio_set_dev(bio, bdev);
2650 bio->bi_iter.bi_sector = first_byte >> 9;
2651 btrfs_io_bio_init(btrfs_io_bio(bio));
2655 struct bio *btrfs_bio_clone(struct bio *bio)
2657 struct btrfs_io_bio *btrfs_bio;
2660 /* Bio allocation backed by a bioset does not fail */
2661 new = bio_clone_fast(bio, GFP_NOFS, &btrfs_bioset);
2662 btrfs_bio = btrfs_io_bio(new);
2663 btrfs_io_bio_init(btrfs_bio);
2664 btrfs_bio->iter = bio->bi_iter;
2668 struct bio *btrfs_io_bio_alloc(unsigned int nr_iovecs)
2672 /* Bio allocation backed by a bioset does not fail */
2673 bio = bio_alloc_bioset(GFP_NOFS, nr_iovecs, &btrfs_bioset);
2674 btrfs_io_bio_init(btrfs_io_bio(bio));
2678 struct bio *btrfs_bio_clone_partial(struct bio *orig, int offset, int size)
2681 struct btrfs_io_bio *btrfs_bio;
2683 /* this will never fail when it's backed by a bioset */
2684 bio = bio_clone_fast(orig, GFP_NOFS, &btrfs_bioset);
2687 btrfs_bio = btrfs_io_bio(bio);
2688 btrfs_io_bio_init(btrfs_bio);
2690 bio_trim(bio, offset >> 9, size >> 9);
2691 btrfs_bio->iter = bio->bi_iter;
2695 static int __must_check submit_one_bio(struct bio *bio, int mirror_num,
2696 unsigned long bio_flags)
2698 blk_status_t ret = 0;
2699 struct bio_vec *bvec = bio_last_bvec_all(bio);
2700 struct page *page = bvec->bv_page;
2701 struct extent_io_tree *tree = bio->bi_private;
2704 start = page_offset(page) + bvec->bv_offset;
2706 bio->bi_private = NULL;
2709 ret = tree->ops->submit_bio_hook(tree->private_data, bio,
2710 mirror_num, bio_flags, start);
2712 btrfsic_submit_bio(bio);
2714 return blk_status_to_errno(ret);
2718 * @opf: bio REQ_OP_* and REQ_* flags as one value
2719 * @tree: tree so we can call our merge_bio hook
2720 * @wbc: optional writeback control for io accounting
2721 * @page: page to add to the bio
2722 * @pg_offset: offset of the new bio or to check whether we are adding
2723 * a contiguous page to the previous one
2724 * @size: portion of page that we want to write
2725 * @offset: starting offset in the page
2726 * @bdev: attach newly created bios to this bdev
2727 * @bio_ret: must be valid pointer, newly allocated bio will be stored there
2728 * @end_io_func: end_io callback for new bio
2729 * @mirror_num: desired mirror to read/write
2730 * @prev_bio_flags: flags of previous bio to see if we can merge the current one
2731 * @bio_flags: flags of the current bio to see if we can merge them
2733 static int submit_extent_page(unsigned int opf, struct extent_io_tree *tree,
2734 struct writeback_control *wbc,
2735 struct page *page, u64 offset,
2736 size_t size, unsigned long pg_offset,
2737 struct block_device *bdev,
2738 struct bio **bio_ret,
2739 bio_end_io_t end_io_func,
2741 unsigned long prev_bio_flags,
2742 unsigned long bio_flags,
2743 bool force_bio_submit)
2747 size_t page_size = min_t(size_t, size, PAGE_SIZE);
2748 sector_t sector = offset >> 9;
2754 bool can_merge = true;
2757 if (prev_bio_flags & EXTENT_BIO_COMPRESSED)
2758 contig = bio->bi_iter.bi_sector == sector;
2760 contig = bio_end_sector(bio) == sector;
2763 if (btrfs_bio_fits_in_stripe(page, page_size, bio, bio_flags))
2766 if (prev_bio_flags != bio_flags || !contig || !can_merge ||
2768 bio_add_page(bio, page, page_size, pg_offset) < page_size) {
2769 ret = submit_one_bio(bio, mirror_num, prev_bio_flags);
2777 wbc_account_io(wbc, page, page_size);
2782 bio = btrfs_bio_alloc(bdev, offset);
2783 bio_add_page(bio, page, page_size, pg_offset);
2784 bio->bi_end_io = end_io_func;
2785 bio->bi_private = tree;
2786 bio->bi_write_hint = page->mapping->host->i_write_hint;
2789 wbc_init_bio(wbc, bio);
2790 wbc_account_io(wbc, page, page_size);
2798 static void attach_extent_buffer_page(struct extent_buffer *eb,
2801 if (!PagePrivate(page)) {
2802 SetPagePrivate(page);
2804 set_page_private(page, (unsigned long)eb);
2806 WARN_ON(page->private != (unsigned long)eb);
2810 void set_page_extent_mapped(struct page *page)
2812 if (!PagePrivate(page)) {
2813 SetPagePrivate(page);
2815 set_page_private(page, EXTENT_PAGE_PRIVATE);
2819 static struct extent_map *
2820 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
2821 u64 start, u64 len, get_extent_t *get_extent,
2822 struct extent_map **em_cached)
2824 struct extent_map *em;
2826 if (em_cached && *em_cached) {
2828 if (extent_map_in_tree(em) && start >= em->start &&
2829 start < extent_map_end(em)) {
2830 refcount_inc(&em->refs);
2834 free_extent_map(em);
2838 em = get_extent(BTRFS_I(inode), page, pg_offset, start, len, 0);
2839 if (em_cached && !IS_ERR_OR_NULL(em)) {
2841 refcount_inc(&em->refs);
2847 * basic readpage implementation. Locked extent state structs are inserted
2848 * into the tree that are removed when the IO is done (by the end_io
2850 * XXX JDM: This needs looking at to ensure proper page locking
2851 * return 0 on success, otherwise return error
2853 static int __do_readpage(struct extent_io_tree *tree,
2855 get_extent_t *get_extent,
2856 struct extent_map **em_cached,
2857 struct bio **bio, int mirror_num,
2858 unsigned long *bio_flags, unsigned int read_flags,
2861 struct inode *inode = page->mapping->host;
2862 u64 start = page_offset(page);
2863 const u64 end = start + PAGE_SIZE - 1;
2866 u64 last_byte = i_size_read(inode);
2869 struct extent_map *em;
2870 struct block_device *bdev;
2873 size_t pg_offset = 0;
2875 size_t disk_io_size;
2876 size_t blocksize = inode->i_sb->s_blocksize;
2877 unsigned long this_bio_flag = 0;
2879 set_page_extent_mapped(page);
2881 if (!PageUptodate(page)) {
2882 if (cleancache_get_page(page) == 0) {
2883 BUG_ON(blocksize != PAGE_SIZE);
2884 unlock_extent(tree, start, end);
2889 if (page->index == last_byte >> PAGE_SHIFT) {
2891 size_t zero_offset = offset_in_page(last_byte);
2894 iosize = PAGE_SIZE - zero_offset;
2895 userpage = kmap_atomic(page);
2896 memset(userpage + zero_offset, 0, iosize);
2897 flush_dcache_page(page);
2898 kunmap_atomic(userpage);
2901 while (cur <= end) {
2902 bool force_bio_submit = false;
2905 if (cur >= last_byte) {
2907 struct extent_state *cached = NULL;
2909 iosize = PAGE_SIZE - pg_offset;
2910 userpage = kmap_atomic(page);
2911 memset(userpage + pg_offset, 0, iosize);
2912 flush_dcache_page(page);
2913 kunmap_atomic(userpage);
2914 set_extent_uptodate(tree, cur, cur + iosize - 1,
2916 unlock_extent_cached(tree, cur,
2917 cur + iosize - 1, &cached);
2920 em = __get_extent_map(inode, page, pg_offset, cur,
2921 end - cur + 1, get_extent, em_cached);
2922 if (IS_ERR_OR_NULL(em)) {
2924 unlock_extent(tree, cur, end);
2927 extent_offset = cur - em->start;
2928 BUG_ON(extent_map_end(em) <= cur);
2931 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2932 this_bio_flag |= EXTENT_BIO_COMPRESSED;
2933 extent_set_compress_type(&this_bio_flag,
2937 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2938 cur_end = min(extent_map_end(em) - 1, end);
2939 iosize = ALIGN(iosize, blocksize);
2940 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2941 disk_io_size = em->block_len;
2942 offset = em->block_start;
2944 offset = em->block_start + extent_offset;
2945 disk_io_size = iosize;
2948 block_start = em->block_start;
2949 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2950 block_start = EXTENT_MAP_HOLE;
2953 * If we have a file range that points to a compressed extent
2954 * and it's followed by a consecutive file range that points to
2955 * to the same compressed extent (possibly with a different
2956 * offset and/or length, so it either points to the whole extent
2957 * or only part of it), we must make sure we do not submit a
2958 * single bio to populate the pages for the 2 ranges because
2959 * this makes the compressed extent read zero out the pages
2960 * belonging to the 2nd range. Imagine the following scenario:
2963 * [0 - 8K] [8K - 24K]
2966 * points to extent X, points to extent X,
2967 * offset 4K, length of 8K offset 0, length 16K
2969 * [extent X, compressed length = 4K uncompressed length = 16K]
2971 * If the bio to read the compressed extent covers both ranges,
2972 * it will decompress extent X into the pages belonging to the
2973 * first range and then it will stop, zeroing out the remaining
2974 * pages that belong to the other range that points to extent X.
2975 * So here we make sure we submit 2 bios, one for the first
2976 * range and another one for the third range. Both will target
2977 * the same physical extent from disk, but we can't currently
2978 * make the compressed bio endio callback populate the pages
2979 * for both ranges because each compressed bio is tightly
2980 * coupled with a single extent map, and each range can have
2981 * an extent map with a different offset value relative to the
2982 * uncompressed data of our extent and different lengths. This
2983 * is a corner case so we prioritize correctness over
2984 * non-optimal behavior (submitting 2 bios for the same extent).
2986 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) &&
2987 prev_em_start && *prev_em_start != (u64)-1 &&
2988 *prev_em_start != em->orig_start)
2989 force_bio_submit = true;
2992 *prev_em_start = em->orig_start;
2994 free_extent_map(em);
2997 /* we've found a hole, just zero and go on */
2998 if (block_start == EXTENT_MAP_HOLE) {
3000 struct extent_state *cached = NULL;
3002 userpage = kmap_atomic(page);
3003 memset(userpage + pg_offset, 0, iosize);
3004 flush_dcache_page(page);
3005 kunmap_atomic(userpage);
3007 set_extent_uptodate(tree, cur, cur + iosize - 1,
3009 unlock_extent_cached(tree, cur,
3010 cur + iosize - 1, &cached);
3012 pg_offset += iosize;
3015 /* the get_extent function already copied into the page */
3016 if (test_range_bit(tree, cur, cur_end,
3017 EXTENT_UPTODATE, 1, NULL)) {
3018 check_page_uptodate(tree, page);
3019 unlock_extent(tree, cur, cur + iosize - 1);
3021 pg_offset += iosize;
3024 /* we have an inline extent but it didn't get marked up
3025 * to date. Error out
3027 if (block_start == EXTENT_MAP_INLINE) {
3029 unlock_extent(tree, cur, cur + iosize - 1);
3031 pg_offset += iosize;
3035 ret = submit_extent_page(REQ_OP_READ | read_flags, tree, NULL,
3036 page, offset, disk_io_size,
3037 pg_offset, bdev, bio,
3038 end_bio_extent_readpage, mirror_num,
3044 *bio_flags = this_bio_flag;
3047 unlock_extent(tree, cur, cur + iosize - 1);
3051 pg_offset += iosize;
3055 if (!PageError(page))
3056 SetPageUptodate(page);
3062 static inline void __do_contiguous_readpages(struct extent_io_tree *tree,
3063 struct page *pages[], int nr_pages,
3065 struct extent_map **em_cached,
3067 unsigned long *bio_flags,
3070 struct inode *inode;
3071 struct btrfs_ordered_extent *ordered;
3074 inode = pages[0]->mapping->host;
3076 lock_extent(tree, start, end);
3077 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
3081 unlock_extent(tree, start, end);
3082 btrfs_start_ordered_extent(inode, ordered, 1);
3083 btrfs_put_ordered_extent(ordered);
3086 for (index = 0; index < nr_pages; index++) {
3087 __do_readpage(tree, pages[index], btrfs_get_extent, em_cached,
3088 bio, 0, bio_flags, REQ_RAHEAD, prev_em_start);
3089 put_page(pages[index]);
3093 static void __extent_readpages(struct extent_io_tree *tree,
3094 struct page *pages[],
3096 struct extent_map **em_cached,
3097 struct bio **bio, unsigned long *bio_flags,
3104 int first_index = 0;
3106 for (index = 0; index < nr_pages; index++) {
3107 page_start = page_offset(pages[index]);
3110 end = start + PAGE_SIZE - 1;
3111 first_index = index;
3112 } else if (end + 1 == page_start) {
3115 __do_contiguous_readpages(tree, &pages[first_index],
3116 index - first_index, start,
3121 end = start + PAGE_SIZE - 1;
3122 first_index = index;
3127 __do_contiguous_readpages(tree, &pages[first_index],
3128 index - first_index, start,
3129 end, em_cached, bio,
3130 bio_flags, prev_em_start);
3133 static int __extent_read_full_page(struct extent_io_tree *tree,
3135 get_extent_t *get_extent,
3136 struct bio **bio, int mirror_num,
3137 unsigned long *bio_flags,
3138 unsigned int read_flags)
3140 struct inode *inode = page->mapping->host;
3141 struct btrfs_ordered_extent *ordered;
3142 u64 start = page_offset(page);
3143 u64 end = start + PAGE_SIZE - 1;
3147 lock_extent(tree, start, end);
3148 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
3152 unlock_extent(tree, start, end);
3153 btrfs_start_ordered_extent(inode, ordered, 1);
3154 btrfs_put_ordered_extent(ordered);
3157 ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num,
3158 bio_flags, read_flags, NULL);
3162 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
3163 get_extent_t *get_extent, int mirror_num)
3165 struct bio *bio = NULL;
3166 unsigned long bio_flags = 0;
3169 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
3172 ret = submit_one_bio(bio, mirror_num, bio_flags);
3176 static void update_nr_written(struct writeback_control *wbc,
3177 unsigned long nr_written)
3179 wbc->nr_to_write -= nr_written;
3183 * helper for __extent_writepage, doing all of the delayed allocation setup.
3185 * This returns 1 if btrfs_run_delalloc_range function did all the work required
3186 * to write the page (copy into inline extent). In this case the IO has
3187 * been started and the page is already unlocked.
3189 * This returns 0 if all went well (page still locked)
3190 * This returns < 0 if there were errors (page still locked)
3192 static noinline_for_stack int writepage_delalloc(struct inode *inode,
3193 struct page *page, struct writeback_control *wbc,
3194 u64 delalloc_start, unsigned long *nr_written)
3196 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
3197 u64 page_end = delalloc_start + PAGE_SIZE - 1;
3199 u64 delalloc_to_write = 0;
3200 u64 delalloc_end = 0;
3202 int page_started = 0;
3205 while (delalloc_end < page_end) {
3206 found = find_lock_delalloc_range(inode, tree,
3211 delalloc_start = delalloc_end + 1;
3214 ret = btrfs_run_delalloc_range(inode, page, delalloc_start,
3215 delalloc_end, &page_started, nr_written, wbc);
3216 /* File system has been set read-only */
3220 * btrfs_run_delalloc_range should return < 0 for error
3221 * but just in case, we use > 0 here meaning the IO is
3222 * started, so we don't want to return > 0 unless
3223 * things are going well.
3225 ret = ret < 0 ? ret : -EIO;
3229 * delalloc_end is already one less than the total length, so
3230 * we don't subtract one from PAGE_SIZE
3232 delalloc_to_write += (delalloc_end - delalloc_start +
3233 PAGE_SIZE) >> PAGE_SHIFT;
3234 delalloc_start = delalloc_end + 1;
3236 if (wbc->nr_to_write < delalloc_to_write) {
3239 if (delalloc_to_write < thresh * 2)
3240 thresh = delalloc_to_write;
3241 wbc->nr_to_write = min_t(u64, delalloc_to_write,
3245 /* did the fill delalloc function already unlock and start
3250 * we've unlocked the page, so we can't update
3251 * the mapping's writeback index, just update
3254 wbc->nr_to_write -= *nr_written;
3265 * helper for __extent_writepage. This calls the writepage start hooks,
3266 * and does the loop to map the page into extents and bios.
3268 * We return 1 if the IO is started and the page is unlocked,
3269 * 0 if all went well (page still locked)
3270 * < 0 if there were errors (page still locked)
3272 static noinline_for_stack int __extent_writepage_io(struct inode *inode,
3274 struct writeback_control *wbc,
3275 struct extent_page_data *epd,
3277 unsigned long nr_written,
3278 unsigned int write_flags, int *nr_ret)
3280 struct extent_io_tree *tree = epd->tree;
3281 u64 start = page_offset(page);
3282 u64 page_end = start + PAGE_SIZE - 1;
3288 struct extent_map *em;
3289 struct block_device *bdev;
3290 size_t pg_offset = 0;
3296 ret = btrfs_writepage_cow_fixup(page, start, page_end);
3298 /* Fixup worker will requeue */
3300 wbc->pages_skipped++;
3302 redirty_page_for_writepage(wbc, page);
3304 update_nr_written(wbc, nr_written);
3310 * we don't want to touch the inode after unlocking the page,
3311 * so we update the mapping writeback index now
3313 update_nr_written(wbc, nr_written + 1);
3316 if (i_size <= start) {
3317 btrfs_writepage_endio_finish_ordered(page, start, page_end, 1);
3321 blocksize = inode->i_sb->s_blocksize;
3323 while (cur <= end) {
3327 if (cur >= i_size) {
3328 btrfs_writepage_endio_finish_ordered(page, cur,
3332 em = btrfs_get_extent(BTRFS_I(inode), page, pg_offset, cur,
3334 if (IS_ERR_OR_NULL(em)) {
3336 ret = PTR_ERR_OR_ZERO(em);
3340 extent_offset = cur - em->start;
3341 em_end = extent_map_end(em);
3342 BUG_ON(em_end <= cur);
3344 iosize = min(em_end - cur, end - cur + 1);
3345 iosize = ALIGN(iosize, blocksize);
3346 offset = em->block_start + extent_offset;
3348 block_start = em->block_start;
3349 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3350 free_extent_map(em);
3354 * compressed and inline extents are written through other
3357 if (compressed || block_start == EXTENT_MAP_HOLE ||
3358 block_start == EXTENT_MAP_INLINE) {
3360 * end_io notification does not happen here for
3361 * compressed extents
3364 btrfs_writepage_endio_finish_ordered(page, cur,
3367 else if (compressed) {
3368 /* we don't want to end_page_writeback on
3369 * a compressed extent. this happens
3376 pg_offset += iosize;
3380 btrfs_set_range_writeback(tree, cur, cur + iosize - 1);
3381 if (!PageWriteback(page)) {
3382 btrfs_err(BTRFS_I(inode)->root->fs_info,
3383 "page %lu not writeback, cur %llu end %llu",
3384 page->index, cur, end);
3387 ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc,
3388 page, offset, iosize, pg_offset,
3390 end_bio_extent_writepage,
3394 if (PageWriteback(page))
3395 end_page_writeback(page);
3399 pg_offset += iosize;
3408 * the writepage semantics are similar to regular writepage. extent
3409 * records are inserted to lock ranges in the tree, and as dirty areas
3410 * are found, they are marked writeback. Then the lock bits are removed
3411 * and the end_io handler clears the writeback ranges
3413 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
3414 struct extent_page_data *epd)
3416 struct inode *inode = page->mapping->host;
3417 u64 start = page_offset(page);
3418 u64 page_end = start + PAGE_SIZE - 1;
3421 size_t pg_offset = 0;
3422 loff_t i_size = i_size_read(inode);
3423 unsigned long end_index = i_size >> PAGE_SHIFT;
3424 unsigned int write_flags = 0;
3425 unsigned long nr_written = 0;
3427 write_flags = wbc_to_write_flags(wbc);
3429 trace___extent_writepage(page, inode, wbc);
3431 WARN_ON(!PageLocked(page));
3433 ClearPageError(page);
3435 pg_offset = offset_in_page(i_size);
3436 if (page->index > end_index ||
3437 (page->index == end_index && !pg_offset)) {
3438 page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
3443 if (page->index == end_index) {
3446 userpage = kmap_atomic(page);
3447 memset(userpage + pg_offset, 0,
3448 PAGE_SIZE - pg_offset);
3449 kunmap_atomic(userpage);
3450 flush_dcache_page(page);
3455 set_page_extent_mapped(page);
3457 if (!epd->extent_locked) {
3458 ret = writepage_delalloc(inode, page, wbc, start, &nr_written);
3465 ret = __extent_writepage_io(inode, page, wbc, epd,
3466 i_size, nr_written, write_flags, &nr);
3472 /* make sure the mapping tag for page dirty gets cleared */
3473 set_page_writeback(page);
3474 end_page_writeback(page);
3476 if (PageError(page)) {
3477 ret = ret < 0 ? ret : -EIO;
3478 end_extent_writepage(page, ret, start, page_end);
3487 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3489 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
3490 TASK_UNINTERRUPTIBLE);
3493 static noinline_for_stack int
3494 lock_extent_buffer_for_io(struct extent_buffer *eb,
3495 struct btrfs_fs_info *fs_info,
3496 struct extent_page_data *epd)
3502 if (!btrfs_try_tree_write_lock(eb)) {
3504 flush_write_bio(epd);
3505 btrfs_tree_lock(eb);
3508 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3509 btrfs_tree_unlock(eb);
3513 flush_write_bio(epd);
3517 wait_on_extent_buffer_writeback(eb);
3518 btrfs_tree_lock(eb);
3519 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3521 btrfs_tree_unlock(eb);
3526 * We need to do this to prevent races in people who check if the eb is
3527 * under IO since we can end up having no IO bits set for a short period
3530 spin_lock(&eb->refs_lock);
3531 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3532 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3533 spin_unlock(&eb->refs_lock);
3534 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3535 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
3537 fs_info->dirty_metadata_batch);
3540 spin_unlock(&eb->refs_lock);
3543 btrfs_tree_unlock(eb);
3548 num_pages = num_extent_pages(eb);
3549 for (i = 0; i < num_pages; i++) {
3550 struct page *p = eb->pages[i];
3552 if (!trylock_page(p)) {
3554 flush_write_bio(epd);
3564 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3566 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3567 smp_mb__after_atomic();
3568 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3571 static void set_btree_ioerr(struct page *page)
3573 struct extent_buffer *eb = (struct extent_buffer *)page->private;
3576 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))
3580 * If writeback for a btree extent that doesn't belong to a log tree
3581 * failed, increment the counter transaction->eb_write_errors.
3582 * We do this because while the transaction is running and before it's
3583 * committing (when we call filemap_fdata[write|wait]_range against
3584 * the btree inode), we might have
3585 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
3586 * returns an error or an error happens during writeback, when we're
3587 * committing the transaction we wouldn't know about it, since the pages
3588 * can be no longer dirty nor marked anymore for writeback (if a
3589 * subsequent modification to the extent buffer didn't happen before the
3590 * transaction commit), which makes filemap_fdata[write|wait]_range not
3591 * able to find the pages tagged with SetPageError at transaction
3592 * commit time. So if this happens we must abort the transaction,
3593 * otherwise we commit a super block with btree roots that point to
3594 * btree nodes/leafs whose content on disk is invalid - either garbage
3595 * or the content of some node/leaf from a past generation that got
3596 * cowed or deleted and is no longer valid.
3598 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
3599 * not be enough - we need to distinguish between log tree extents vs
3600 * non-log tree extents, and the next filemap_fdatawait_range() call
3601 * will catch and clear such errors in the mapping - and that call might
3602 * be from a log sync and not from a transaction commit. Also, checking
3603 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
3604 * not done and would not be reliable - the eb might have been released
3605 * from memory and reading it back again means that flag would not be
3606 * set (since it's a runtime flag, not persisted on disk).
3608 * Using the flags below in the btree inode also makes us achieve the
3609 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
3610 * writeback for all dirty pages and before filemap_fdatawait_range()
3611 * is called, the writeback for all dirty pages had already finished
3612 * with errors - because we were not using AS_EIO/AS_ENOSPC,
3613 * filemap_fdatawait_range() would return success, as it could not know
3614 * that writeback errors happened (the pages were no longer tagged for
3617 switch (eb->log_index) {
3619 set_bit(BTRFS_FS_BTREE_ERR, &eb->fs_info->flags);
3622 set_bit(BTRFS_FS_LOG1_ERR, &eb->fs_info->flags);
3625 set_bit(BTRFS_FS_LOG2_ERR, &eb->fs_info->flags);
3628 BUG(); /* unexpected, logic error */
3632 static void end_bio_extent_buffer_writepage(struct bio *bio)
3634 struct bio_vec *bvec;
3635 struct extent_buffer *eb;
3638 ASSERT(!bio_flagged(bio, BIO_CLONED));
3639 bio_for_each_segment_all(bvec, bio, i) {
3640 struct page *page = bvec->bv_page;
3642 eb = (struct extent_buffer *)page->private;
3644 done = atomic_dec_and_test(&eb->io_pages);
3646 if (bio->bi_status ||
3647 test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) {
3648 ClearPageUptodate(page);
3649 set_btree_ioerr(page);
3652 end_page_writeback(page);
3657 end_extent_buffer_writeback(eb);
3663 static noinline_for_stack int write_one_eb(struct extent_buffer *eb,
3664 struct btrfs_fs_info *fs_info,
3665 struct writeback_control *wbc,
3666 struct extent_page_data *epd)
3668 struct block_device *bdev = fs_info->fs_devices->latest_bdev;
3669 struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
3670 u64 offset = eb->start;
3673 unsigned long start, end;
3674 unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META;
3677 clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
3678 num_pages = num_extent_pages(eb);
3679 atomic_set(&eb->io_pages, num_pages);
3681 /* set btree blocks beyond nritems with 0 to avoid stale content. */
3682 nritems = btrfs_header_nritems(eb);
3683 if (btrfs_header_level(eb) > 0) {
3684 end = btrfs_node_key_ptr_offset(nritems);
3686 memzero_extent_buffer(eb, end, eb->len - end);
3690 * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0
3692 start = btrfs_item_nr_offset(nritems);
3693 end = BTRFS_LEAF_DATA_OFFSET + leaf_data_end(fs_info, eb);
3694 memzero_extent_buffer(eb, start, end - start);
3697 for (i = 0; i < num_pages; i++) {
3698 struct page *p = eb->pages[i];
3700 clear_page_dirty_for_io(p);
3701 set_page_writeback(p);
3702 ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc,
3703 p, offset, PAGE_SIZE, 0, bdev,
3705 end_bio_extent_buffer_writepage,
3709 if (PageWriteback(p))
3710 end_page_writeback(p);
3711 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3712 end_extent_buffer_writeback(eb);
3716 offset += PAGE_SIZE;
3717 update_nr_written(wbc, 1);
3721 if (unlikely(ret)) {
3722 for (; i < num_pages; i++) {
3723 struct page *p = eb->pages[i];
3724 clear_page_dirty_for_io(p);
3732 int btree_write_cache_pages(struct address_space *mapping,
3733 struct writeback_control *wbc)
3735 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
3736 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
3737 struct extent_buffer *eb, *prev_eb = NULL;
3738 struct extent_page_data epd = {
3742 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3746 int nr_to_write_done = 0;
3747 struct pagevec pvec;
3750 pgoff_t end; /* Inclusive */
3754 pagevec_init(&pvec);
3755 if (wbc->range_cyclic) {
3756 index = mapping->writeback_index; /* Start from prev offset */
3759 index = wbc->range_start >> PAGE_SHIFT;
3760 end = wbc->range_end >> PAGE_SHIFT;
3763 if (wbc->sync_mode == WB_SYNC_ALL)
3764 tag = PAGECACHE_TAG_TOWRITE;
3766 tag = PAGECACHE_TAG_DIRTY;
3768 if (wbc->sync_mode == WB_SYNC_ALL)
3769 tag_pages_for_writeback(mapping, index, end);
3770 while (!done && !nr_to_write_done && (index <= end) &&
3771 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
3776 for (i = 0; i < nr_pages; i++) {
3777 struct page *page = pvec.pages[i];
3779 if (!PagePrivate(page))
3782 spin_lock(&mapping->private_lock);
3783 if (!PagePrivate(page)) {
3784 spin_unlock(&mapping->private_lock);
3788 eb = (struct extent_buffer *)page->private;
3791 * Shouldn't happen and normally this would be a BUG_ON
3792 * but no sense in crashing the users box for something
3793 * we can survive anyway.
3796 spin_unlock(&mapping->private_lock);
3800 if (eb == prev_eb) {
3801 spin_unlock(&mapping->private_lock);
3805 ret = atomic_inc_not_zero(&eb->refs);
3806 spin_unlock(&mapping->private_lock);
3811 ret = lock_extent_buffer_for_io(eb, fs_info, &epd);
3813 free_extent_buffer(eb);
3817 ret = write_one_eb(eb, fs_info, wbc, &epd);
3820 free_extent_buffer(eb);
3823 free_extent_buffer(eb);
3826 * the filesystem may choose to bump up nr_to_write.
3827 * We have to make sure to honor the new nr_to_write
3830 nr_to_write_done = wbc->nr_to_write <= 0;
3832 pagevec_release(&pvec);
3835 if (!scanned && !done) {
3837 * We hit the last page and there is more work to be done: wrap
3838 * back to the start of the file
3844 flush_write_bio(&epd);
3849 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3850 * @mapping: address space structure to write
3851 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3852 * @data: data passed to __extent_writepage function
3854 * If a page is already under I/O, write_cache_pages() skips it, even
3855 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3856 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3857 * and msync() need to guarantee that all the data which was dirty at the time
3858 * the call was made get new I/O started against them. If wbc->sync_mode is
3859 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3860 * existing IO to complete.
3862 static int extent_write_cache_pages(struct address_space *mapping,
3863 struct writeback_control *wbc,
3864 struct extent_page_data *epd)
3866 struct inode *inode = mapping->host;
3869 int nr_to_write_done = 0;
3870 struct pagevec pvec;
3873 pgoff_t end; /* Inclusive */
3875 int range_whole = 0;
3880 * We have to hold onto the inode so that ordered extents can do their
3881 * work when the IO finishes. The alternative to this is failing to add
3882 * an ordered extent if the igrab() fails there and that is a huge pain
3883 * to deal with, so instead just hold onto the inode throughout the
3884 * writepages operation. If it fails here we are freeing up the inode
3885 * anyway and we'd rather not waste our time writing out stuff that is
3886 * going to be truncated anyway.
3891 pagevec_init(&pvec);
3892 if (wbc->range_cyclic) {
3893 index = mapping->writeback_index; /* Start from prev offset */
3896 index = wbc->range_start >> PAGE_SHIFT;
3897 end = wbc->range_end >> PAGE_SHIFT;
3898 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
3904 * We do the tagged writepage as long as the snapshot flush bit is set
3905 * and we are the first one who do the filemap_flush() on this inode.
3907 * The nr_to_write == LONG_MAX is needed to make sure other flushers do
3908 * not race in and drop the bit.
3910 if (range_whole && wbc->nr_to_write == LONG_MAX &&
3911 test_and_clear_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
3912 &BTRFS_I(inode)->runtime_flags))
3913 wbc->tagged_writepages = 1;
3915 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
3916 tag = PAGECACHE_TAG_TOWRITE;
3918 tag = PAGECACHE_TAG_DIRTY;
3920 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
3921 tag_pages_for_writeback(mapping, index, end);
3923 while (!done && !nr_to_write_done && (index <= end) &&
3924 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping,
3925 &index, end, tag))) {
3929 for (i = 0; i < nr_pages; i++) {
3930 struct page *page = pvec.pages[i];
3932 done_index = page->index;
3934 * At this point we hold neither the i_pages lock nor
3935 * the page lock: the page may be truncated or
3936 * invalidated (changing page->mapping to NULL),
3937 * or even swizzled back from swapper_space to
3938 * tmpfs file mapping
3940 if (!trylock_page(page)) {
3941 flush_write_bio(epd);
3945 if (unlikely(page->mapping != mapping)) {
3950 if (wbc->sync_mode != WB_SYNC_NONE) {
3951 if (PageWriteback(page))
3952 flush_write_bio(epd);
3953 wait_on_page_writeback(page);
3956 if (PageWriteback(page) ||
3957 !clear_page_dirty_for_io(page)) {
3962 ret = __extent_writepage(page, wbc, epd);
3964 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
3970 * done_index is set past this page,
3971 * so media errors will not choke
3972 * background writeout for the entire
3973 * file. This has consequences for
3974 * range_cyclic semantics (ie. it may
3975 * not be suitable for data integrity
3978 done_index = page->index + 1;
3984 * the filesystem may choose to bump up nr_to_write.
3985 * We have to make sure to honor the new nr_to_write
3988 nr_to_write_done = wbc->nr_to_write <= 0;
3990 pagevec_release(&pvec);
3993 if (!scanned && !done) {
3995 * We hit the last page and there is more work to be done: wrap
3996 * back to the start of the file
4003 if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole))
4004 mapping->writeback_index = done_index;
4006 btrfs_add_delayed_iput(inode);
4010 static void flush_write_bio(struct extent_page_data *epd)
4015 ret = submit_one_bio(epd->bio, 0, 0);
4016 BUG_ON(ret < 0); /* -ENOMEM */
4021 int extent_write_full_page(struct page *page, struct writeback_control *wbc)
4024 struct extent_page_data epd = {
4026 .tree = &BTRFS_I(page->mapping->host)->io_tree,
4028 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4031 ret = __extent_writepage(page, wbc, &epd);
4033 flush_write_bio(&epd);
4037 int extent_write_locked_range(struct inode *inode, u64 start, u64 end,
4041 struct address_space *mapping = inode->i_mapping;
4042 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
4044 unsigned long nr_pages = (end - start + PAGE_SIZE) >>
4047 struct extent_page_data epd = {
4051 .sync_io = mode == WB_SYNC_ALL,
4053 struct writeback_control wbc_writepages = {
4055 .nr_to_write = nr_pages * 2,
4056 .range_start = start,
4057 .range_end = end + 1,
4060 while (start <= end) {
4061 page = find_get_page(mapping, start >> PAGE_SHIFT);
4062 if (clear_page_dirty_for_io(page))
4063 ret = __extent_writepage(page, &wbc_writepages, &epd);
4065 btrfs_writepage_endio_finish_ordered(page, start,
4066 start + PAGE_SIZE - 1, 1);
4073 flush_write_bio(&epd);
4077 int extent_writepages(struct address_space *mapping,
4078 struct writeback_control *wbc)
4081 struct extent_page_data epd = {
4083 .tree = &BTRFS_I(mapping->host)->io_tree,
4085 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4088 ret = extent_write_cache_pages(mapping, wbc, &epd);
4089 flush_write_bio(&epd);
4093 int extent_readpages(struct address_space *mapping, struct list_head *pages,
4096 struct bio *bio = NULL;
4097 unsigned long bio_flags = 0;
4098 struct page *pagepool[16];
4099 struct extent_map *em_cached = NULL;
4100 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
4102 u64 prev_em_start = (u64)-1;
4104 while (!list_empty(pages)) {
4105 for (nr = 0; nr < ARRAY_SIZE(pagepool) && !list_empty(pages);) {
4106 struct page *page = lru_to_page(pages);
4108 prefetchw(&page->flags);
4109 list_del(&page->lru);
4110 if (add_to_page_cache_lru(page, mapping, page->index,
4111 readahead_gfp_mask(mapping))) {
4116 pagepool[nr++] = page;
4119 __extent_readpages(tree, pagepool, nr, &em_cached, &bio,
4120 &bio_flags, &prev_em_start);
4124 free_extent_map(em_cached);
4127 return submit_one_bio(bio, 0, bio_flags);
4132 * basic invalidatepage code, this waits on any locked or writeback
4133 * ranges corresponding to the page, and then deletes any extent state
4134 * records from the tree
4136 int extent_invalidatepage(struct extent_io_tree *tree,
4137 struct page *page, unsigned long offset)
4139 struct extent_state *cached_state = NULL;
4140 u64 start = page_offset(page);
4141 u64 end = start + PAGE_SIZE - 1;
4142 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
4144 start += ALIGN(offset, blocksize);
4148 lock_extent_bits(tree, start, end, &cached_state);
4149 wait_on_page_writeback(page);
4150 clear_extent_bit(tree, start, end,
4151 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4152 EXTENT_DO_ACCOUNTING,
4153 1, 1, &cached_state);
4158 * a helper for releasepage, this tests for areas of the page that
4159 * are locked or under IO and drops the related state bits if it is safe
4162 static int try_release_extent_state(struct extent_io_tree *tree,
4163 struct page *page, gfp_t mask)
4165 u64 start = page_offset(page);
4166 u64 end = start + PAGE_SIZE - 1;
4169 if (test_range_bit(tree, start, end,
4170 EXTENT_IOBITS, 0, NULL))
4174 * at this point we can safely clear everything except the
4175 * locked bit and the nodatasum bit
4177 ret = __clear_extent_bit(tree, start, end,
4178 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
4179 0, 0, NULL, mask, NULL);
4181 /* if clear_extent_bit failed for enomem reasons,
4182 * we can't allow the release to continue.
4193 * a helper for releasepage. As long as there are no locked extents
4194 * in the range corresponding to the page, both state records and extent
4195 * map records are removed
4197 int try_release_extent_mapping(struct page *page, gfp_t mask)
4199 struct extent_map *em;
4200 u64 start = page_offset(page);
4201 u64 end = start + PAGE_SIZE - 1;
4202 struct btrfs_inode *btrfs_inode = BTRFS_I(page->mapping->host);
4203 struct extent_io_tree *tree = &btrfs_inode->io_tree;
4204 struct extent_map_tree *map = &btrfs_inode->extent_tree;
4206 if (gfpflags_allow_blocking(mask) &&
4207 page->mapping->host->i_size > SZ_16M) {
4209 while (start <= end) {
4210 len = end - start + 1;
4211 write_lock(&map->lock);
4212 em = lookup_extent_mapping(map, start, len);
4214 write_unlock(&map->lock);
4217 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
4218 em->start != start) {
4219 write_unlock(&map->lock);
4220 free_extent_map(em);
4223 if (!test_range_bit(tree, em->start,
4224 extent_map_end(em) - 1,
4225 EXTENT_LOCKED | EXTENT_WRITEBACK,
4227 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4228 &btrfs_inode->runtime_flags);
4229 remove_extent_mapping(map, em);
4230 /* once for the rb tree */
4231 free_extent_map(em);
4233 start = extent_map_end(em);
4234 write_unlock(&map->lock);
4237 free_extent_map(em);
4240 return try_release_extent_state(tree, page, mask);
4244 * helper function for fiemap, which doesn't want to see any holes.
4245 * This maps until we find something past 'last'
4247 static struct extent_map *get_extent_skip_holes(struct inode *inode,
4248 u64 offset, u64 last)
4250 u64 sectorsize = btrfs_inode_sectorsize(inode);
4251 struct extent_map *em;
4258 len = last - offset;
4261 len = ALIGN(len, sectorsize);
4262 em = btrfs_get_extent_fiemap(BTRFS_I(inode), NULL, 0, offset,
4264 if (IS_ERR_OR_NULL(em))
4267 /* if this isn't a hole return it */
4268 if (em->block_start != EXTENT_MAP_HOLE)
4271 /* this is a hole, advance to the next extent */
4272 offset = extent_map_end(em);
4273 free_extent_map(em);
4281 * To cache previous fiemap extent
4283 * Will be used for merging fiemap extent
4285 struct fiemap_cache {
4294 * Helper to submit fiemap extent.
4296 * Will try to merge current fiemap extent specified by @offset, @phys,
4297 * @len and @flags with cached one.
4298 * And only when we fails to merge, cached one will be submitted as
4301 * Return value is the same as fiemap_fill_next_extent().
4303 static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
4304 struct fiemap_cache *cache,
4305 u64 offset, u64 phys, u64 len, u32 flags)
4313 * Sanity check, extent_fiemap() should have ensured that new
4314 * fiemap extent won't overlap with cached one.
4317 * NOTE: Physical address can overlap, due to compression
4319 if (cache->offset + cache->len > offset) {
4325 * Only merges fiemap extents if
4326 * 1) Their logical addresses are continuous
4328 * 2) Their physical addresses are continuous
4329 * So truly compressed (physical size smaller than logical size)
4330 * extents won't get merged with each other
4332 * 3) Share same flags except FIEMAP_EXTENT_LAST
4333 * So regular extent won't get merged with prealloc extent
4335 if (cache->offset + cache->len == offset &&
4336 cache->phys + cache->len == phys &&
4337 (cache->flags & ~FIEMAP_EXTENT_LAST) ==
4338 (flags & ~FIEMAP_EXTENT_LAST)) {
4340 cache->flags |= flags;
4341 goto try_submit_last;
4344 /* Not mergeable, need to submit cached one */
4345 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
4346 cache->len, cache->flags);
4347 cache->cached = false;
4351 cache->cached = true;
4352 cache->offset = offset;
4355 cache->flags = flags;
4357 if (cache->flags & FIEMAP_EXTENT_LAST) {
4358 ret = fiemap_fill_next_extent(fieinfo, cache->offset,
4359 cache->phys, cache->len, cache->flags);
4360 cache->cached = false;
4366 * Emit last fiemap cache
4368 * The last fiemap cache may still be cached in the following case:
4370 * |<- Fiemap range ->|
4371 * |<------------ First extent ----------->|
4373 * In this case, the first extent range will be cached but not emitted.
4374 * So we must emit it before ending extent_fiemap().
4376 static int emit_last_fiemap_cache(struct btrfs_fs_info *fs_info,
4377 struct fiemap_extent_info *fieinfo,
4378 struct fiemap_cache *cache)
4385 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
4386 cache->len, cache->flags);
4387 cache->cached = false;
4393 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4394 __u64 start, __u64 len)
4398 u64 max = start + len;
4402 u64 last_for_get_extent = 0;
4404 u64 isize = i_size_read(inode);
4405 struct btrfs_key found_key;
4406 struct extent_map *em = NULL;
4407 struct extent_state *cached_state = NULL;
4408 struct btrfs_path *path;
4409 struct btrfs_root *root = BTRFS_I(inode)->root;
4410 struct fiemap_cache cache = { 0 };
4419 path = btrfs_alloc_path();
4422 path->leave_spinning = 1;
4424 start = round_down(start, btrfs_inode_sectorsize(inode));
4425 len = round_up(max, btrfs_inode_sectorsize(inode)) - start;
4428 * lookup the last file extent. We're not using i_size here
4429 * because there might be preallocation past i_size
4431 ret = btrfs_lookup_file_extent(NULL, root, path,
4432 btrfs_ino(BTRFS_I(inode)), -1, 0);
4434 btrfs_free_path(path);
4443 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
4444 found_type = found_key.type;
4446 /* No extents, but there might be delalloc bits */
4447 if (found_key.objectid != btrfs_ino(BTRFS_I(inode)) ||
4448 found_type != BTRFS_EXTENT_DATA_KEY) {
4449 /* have to trust i_size as the end */
4451 last_for_get_extent = isize;
4454 * remember the start of the last extent. There are a
4455 * bunch of different factors that go into the length of the
4456 * extent, so its much less complex to remember where it started
4458 last = found_key.offset;
4459 last_for_get_extent = last + 1;
4461 btrfs_release_path(path);
4464 * we might have some extents allocated but more delalloc past those
4465 * extents. so, we trust isize unless the start of the last extent is
4470 last_for_get_extent = isize;
4473 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4476 em = get_extent_skip_holes(inode, start, last_for_get_extent);
4485 u64 offset_in_extent = 0;
4487 /* break if the extent we found is outside the range */
4488 if (em->start >= max || extent_map_end(em) < off)
4492 * get_extent may return an extent that starts before our
4493 * requested range. We have to make sure the ranges
4494 * we return to fiemap always move forward and don't
4495 * overlap, so adjust the offsets here
4497 em_start = max(em->start, off);
4500 * record the offset from the start of the extent
4501 * for adjusting the disk offset below. Only do this if the
4502 * extent isn't compressed since our in ram offset may be past
4503 * what we have actually allocated on disk.
4505 if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4506 offset_in_extent = em_start - em->start;
4507 em_end = extent_map_end(em);
4508 em_len = em_end - em_start;
4510 if (em->block_start < EXTENT_MAP_LAST_BYTE)
4511 disko = em->block_start + offset_in_extent;
4516 * bump off for our next call to get_extent
4518 off = extent_map_end(em);
4522 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
4524 flags |= FIEMAP_EXTENT_LAST;
4525 } else if (em->block_start == EXTENT_MAP_INLINE) {
4526 flags |= (FIEMAP_EXTENT_DATA_INLINE |
4527 FIEMAP_EXTENT_NOT_ALIGNED);
4528 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
4529 flags |= (FIEMAP_EXTENT_DELALLOC |
4530 FIEMAP_EXTENT_UNKNOWN);
4531 } else if (fieinfo->fi_extents_max) {
4532 u64 bytenr = em->block_start -
4533 (em->start - em->orig_start);
4536 * As btrfs supports shared space, this information
4537 * can be exported to userspace tools via
4538 * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0
4539 * then we're just getting a count and we can skip the
4542 ret = btrfs_check_shared(root,
4543 btrfs_ino(BTRFS_I(inode)),
4548 flags |= FIEMAP_EXTENT_SHARED;
4551 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4552 flags |= FIEMAP_EXTENT_ENCODED;
4553 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4554 flags |= FIEMAP_EXTENT_UNWRITTEN;
4556 free_extent_map(em);
4558 if ((em_start >= last) || em_len == (u64)-1 ||
4559 (last == (u64)-1 && isize <= em_end)) {
4560 flags |= FIEMAP_EXTENT_LAST;
4564 /* now scan forward to see if this is really the last extent. */
4565 em = get_extent_skip_holes(inode, off, last_for_get_extent);
4571 flags |= FIEMAP_EXTENT_LAST;
4574 ret = emit_fiemap_extent(fieinfo, &cache, em_start, disko,
4584 ret = emit_last_fiemap_cache(root->fs_info, fieinfo, &cache);
4585 free_extent_map(em);
4587 btrfs_free_path(path);
4588 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4593 static void __free_extent_buffer(struct extent_buffer *eb)
4595 btrfs_leak_debug_del(&eb->leak_list);
4596 kmem_cache_free(extent_buffer_cache, eb);
4599 int extent_buffer_under_io(struct extent_buffer *eb)
4601 return (atomic_read(&eb->io_pages) ||
4602 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4603 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4607 * Release all pages attached to the extent buffer.
4609 static void btrfs_release_extent_buffer_pages(struct extent_buffer *eb)
4613 int mapped = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
4615 BUG_ON(extent_buffer_under_io(eb));
4617 num_pages = num_extent_pages(eb);
4618 for (i = 0; i < num_pages; i++) {
4619 struct page *page = eb->pages[i];
4624 spin_lock(&page->mapping->private_lock);
4626 * We do this since we'll remove the pages after we've
4627 * removed the eb from the radix tree, so we could race
4628 * and have this page now attached to the new eb. So
4629 * only clear page_private if it's still connected to
4632 if (PagePrivate(page) &&
4633 page->private == (unsigned long)eb) {
4634 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4635 BUG_ON(PageDirty(page));
4636 BUG_ON(PageWriteback(page));
4638 * We need to make sure we haven't be attached
4641 ClearPagePrivate(page);
4642 set_page_private(page, 0);
4643 /* One for the page private */
4648 spin_unlock(&page->mapping->private_lock);
4650 /* One for when we allocated the page */
4656 * Helper for releasing the extent buffer.
4658 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4660 btrfs_release_extent_buffer_pages(eb);
4661 __free_extent_buffer(eb);
4664 static struct extent_buffer *
4665 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
4668 struct extent_buffer *eb = NULL;
4670 eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL);
4673 eb->fs_info = fs_info;
4675 rwlock_init(&eb->lock);
4676 atomic_set(&eb->write_locks, 0);
4677 atomic_set(&eb->read_locks, 0);
4678 atomic_set(&eb->blocking_readers, 0);
4679 atomic_set(&eb->blocking_writers, 0);
4680 atomic_set(&eb->spinning_readers, 0);
4681 atomic_set(&eb->spinning_writers, 0);
4682 eb->lock_nested = 0;
4683 init_waitqueue_head(&eb->write_lock_wq);
4684 init_waitqueue_head(&eb->read_lock_wq);
4686 btrfs_leak_debug_add(&eb->leak_list, &buffers);
4688 spin_lock_init(&eb->refs_lock);
4689 atomic_set(&eb->refs, 1);
4690 atomic_set(&eb->io_pages, 0);
4693 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4695 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4696 > MAX_INLINE_EXTENT_BUFFER_SIZE);
4697 BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE);
4702 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src)
4706 struct extent_buffer *new;
4707 int num_pages = num_extent_pages(src);
4709 new = __alloc_extent_buffer(src->fs_info, src->start, src->len);
4713 for (i = 0; i < num_pages; i++) {
4714 p = alloc_page(GFP_NOFS);
4716 btrfs_release_extent_buffer(new);
4719 attach_extent_buffer_page(new, p);
4720 WARN_ON(PageDirty(p));
4723 copy_page(page_address(p), page_address(src->pages[i]));
4726 set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
4727 set_bit(EXTENT_BUFFER_UNMAPPED, &new->bflags);
4732 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4733 u64 start, unsigned long len)
4735 struct extent_buffer *eb;
4739 eb = __alloc_extent_buffer(fs_info, start, len);
4743 num_pages = num_extent_pages(eb);
4744 for (i = 0; i < num_pages; i++) {
4745 eb->pages[i] = alloc_page(GFP_NOFS);
4749 set_extent_buffer_uptodate(eb);
4750 btrfs_set_header_nritems(eb, 0);
4751 set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
4756 __free_page(eb->pages[i - 1]);
4757 __free_extent_buffer(eb);
4761 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4764 return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize);
4767 static void check_buffer_tree_ref(struct extent_buffer *eb)
4770 /* the ref bit is tricky. We have to make sure it is set
4771 * if we have the buffer dirty. Otherwise the
4772 * code to free a buffer can end up dropping a dirty
4775 * Once the ref bit is set, it won't go away while the
4776 * buffer is dirty or in writeback, and it also won't
4777 * go away while we have the reference count on the
4780 * We can't just set the ref bit without bumping the
4781 * ref on the eb because free_extent_buffer might
4782 * see the ref bit and try to clear it. If this happens
4783 * free_extent_buffer might end up dropping our original
4784 * ref by mistake and freeing the page before we are able
4785 * to add one more ref.
4787 * So bump the ref count first, then set the bit. If someone
4788 * beat us to it, drop the ref we added.
4790 refs = atomic_read(&eb->refs);
4791 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4794 spin_lock(&eb->refs_lock);
4795 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4796 atomic_inc(&eb->refs);
4797 spin_unlock(&eb->refs_lock);
4800 static void mark_extent_buffer_accessed(struct extent_buffer *eb,
4801 struct page *accessed)
4805 check_buffer_tree_ref(eb);
4807 num_pages = num_extent_pages(eb);
4808 for (i = 0; i < num_pages; i++) {
4809 struct page *p = eb->pages[i];
4812 mark_page_accessed(p);
4816 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
4819 struct extent_buffer *eb;
4822 eb = radix_tree_lookup(&fs_info->buffer_radix,
4823 start >> PAGE_SHIFT);
4824 if (eb && atomic_inc_not_zero(&eb->refs)) {
4827 * Lock our eb's refs_lock to avoid races with
4828 * free_extent_buffer. When we get our eb it might be flagged
4829 * with EXTENT_BUFFER_STALE and another task running
4830 * free_extent_buffer might have seen that flag set,
4831 * eb->refs == 2, that the buffer isn't under IO (dirty and
4832 * writeback flags not set) and it's still in the tree (flag
4833 * EXTENT_BUFFER_TREE_REF set), therefore being in the process
4834 * of decrementing the extent buffer's reference count twice.
4835 * So here we could race and increment the eb's reference count,
4836 * clear its stale flag, mark it as dirty and drop our reference
4837 * before the other task finishes executing free_extent_buffer,
4838 * which would later result in an attempt to free an extent
4839 * buffer that is dirty.
4841 if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) {
4842 spin_lock(&eb->refs_lock);
4843 spin_unlock(&eb->refs_lock);
4845 mark_extent_buffer_accessed(eb, NULL);
4853 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4854 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
4857 struct extent_buffer *eb, *exists = NULL;
4860 eb = find_extent_buffer(fs_info, start);
4863 eb = alloc_dummy_extent_buffer(fs_info, start);
4866 eb->fs_info = fs_info;
4868 ret = radix_tree_preload(GFP_NOFS);
4871 spin_lock(&fs_info->buffer_lock);
4872 ret = radix_tree_insert(&fs_info->buffer_radix,
4873 start >> PAGE_SHIFT, eb);
4874 spin_unlock(&fs_info->buffer_lock);
4875 radix_tree_preload_end();
4876 if (ret == -EEXIST) {
4877 exists = find_extent_buffer(fs_info, start);
4883 check_buffer_tree_ref(eb);
4884 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
4888 btrfs_release_extent_buffer(eb);
4893 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
4896 unsigned long len = fs_info->nodesize;
4899 unsigned long index = start >> PAGE_SHIFT;
4900 struct extent_buffer *eb;
4901 struct extent_buffer *exists = NULL;
4903 struct address_space *mapping = fs_info->btree_inode->i_mapping;
4907 if (!IS_ALIGNED(start, fs_info->sectorsize)) {
4908 btrfs_err(fs_info, "bad tree block start %llu", start);
4909 return ERR_PTR(-EINVAL);
4912 eb = find_extent_buffer(fs_info, start);
4916 eb = __alloc_extent_buffer(fs_info, start, len);
4918 return ERR_PTR(-ENOMEM);
4920 num_pages = num_extent_pages(eb);
4921 for (i = 0; i < num_pages; i++, index++) {
4922 p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL);
4924 exists = ERR_PTR(-ENOMEM);
4928 spin_lock(&mapping->private_lock);
4929 if (PagePrivate(p)) {
4931 * We could have already allocated an eb for this page
4932 * and attached one so lets see if we can get a ref on
4933 * the existing eb, and if we can we know it's good and
4934 * we can just return that one, else we know we can just
4935 * overwrite page->private.
4937 exists = (struct extent_buffer *)p->private;
4938 if (atomic_inc_not_zero(&exists->refs)) {
4939 spin_unlock(&mapping->private_lock);
4942 mark_extent_buffer_accessed(exists, p);
4948 * Do this so attach doesn't complain and we need to
4949 * drop the ref the old guy had.
4951 ClearPagePrivate(p);
4952 WARN_ON(PageDirty(p));
4955 attach_extent_buffer_page(eb, p);
4956 spin_unlock(&mapping->private_lock);
4957 WARN_ON(PageDirty(p));
4959 if (!PageUptodate(p))
4963 * We can't unlock the pages just yet since the extent buffer
4964 * hasn't been properly inserted in the radix tree, this
4965 * opens a race with btree_releasepage which can free a page
4966 * while we are still filling in all pages for the buffer and
4971 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4973 ret = radix_tree_preload(GFP_NOFS);
4975 exists = ERR_PTR(ret);
4979 spin_lock(&fs_info->buffer_lock);
4980 ret = radix_tree_insert(&fs_info->buffer_radix,
4981 start >> PAGE_SHIFT, eb);
4982 spin_unlock(&fs_info->buffer_lock);
4983 radix_tree_preload_end();
4984 if (ret == -EEXIST) {
4985 exists = find_extent_buffer(fs_info, start);
4991 /* add one reference for the tree */
4992 check_buffer_tree_ref(eb);
4993 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
4996 * Now it's safe to unlock the pages because any calls to
4997 * btree_releasepage will correctly detect that a page belongs to a
4998 * live buffer and won't free them prematurely.
5000 for (i = 0; i < num_pages; i++)
5001 unlock_page(eb->pages[i]);
5005 WARN_ON(!atomic_dec_and_test(&eb->refs));
5006 for (i = 0; i < num_pages; i++) {
5008 unlock_page(eb->pages[i]);
5011 btrfs_release_extent_buffer(eb);
5015 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
5017 struct extent_buffer *eb =
5018 container_of(head, struct extent_buffer, rcu_head);
5020 __free_extent_buffer(eb);
5023 static int release_extent_buffer(struct extent_buffer *eb)
5025 lockdep_assert_held(&eb->refs_lock);
5027 WARN_ON(atomic_read(&eb->refs) == 0);
5028 if (atomic_dec_and_test(&eb->refs)) {
5029 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
5030 struct btrfs_fs_info *fs_info = eb->fs_info;
5032 spin_unlock(&eb->refs_lock);
5034 spin_lock(&fs_info->buffer_lock);
5035 radix_tree_delete(&fs_info->buffer_radix,
5036 eb->start >> PAGE_SHIFT);
5037 spin_unlock(&fs_info->buffer_lock);
5039 spin_unlock(&eb->refs_lock);
5042 /* Should be safe to release our pages at this point */
5043 btrfs_release_extent_buffer_pages(eb);
5044 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5045 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags))) {
5046 __free_extent_buffer(eb);
5050 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
5053 spin_unlock(&eb->refs_lock);
5058 void free_extent_buffer(struct extent_buffer *eb)
5066 refs = atomic_read(&eb->refs);
5067 if ((!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && refs <= 3)
5068 || (test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) &&
5071 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
5076 spin_lock(&eb->refs_lock);
5077 if (atomic_read(&eb->refs) == 2 &&
5078 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
5079 !extent_buffer_under_io(eb) &&
5080 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5081 atomic_dec(&eb->refs);
5084 * I know this is terrible, but it's temporary until we stop tracking
5085 * the uptodate bits and such for the extent buffers.
5087 release_extent_buffer(eb);
5090 void free_extent_buffer_stale(struct extent_buffer *eb)
5095 spin_lock(&eb->refs_lock);
5096 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
5098 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
5099 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5100 atomic_dec(&eb->refs);
5101 release_extent_buffer(eb);
5104 void clear_extent_buffer_dirty(struct extent_buffer *eb)
5110 num_pages = num_extent_pages(eb);
5112 for (i = 0; i < num_pages; i++) {
5113 page = eb->pages[i];
5114 if (!PageDirty(page))
5118 WARN_ON(!PagePrivate(page));
5120 clear_page_dirty_for_io(page);
5121 xa_lock_irq(&page->mapping->i_pages);
5122 if (!PageDirty(page))
5123 __xa_clear_mark(&page->mapping->i_pages,
5124 page_index(page), PAGECACHE_TAG_DIRTY);
5125 xa_unlock_irq(&page->mapping->i_pages);
5126 ClearPageError(page);
5129 WARN_ON(atomic_read(&eb->refs) == 0);
5132 bool set_extent_buffer_dirty(struct extent_buffer *eb)
5138 check_buffer_tree_ref(eb);
5140 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
5142 num_pages = num_extent_pages(eb);
5143 WARN_ON(atomic_read(&eb->refs) == 0);
5144 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
5147 for (i = 0; i < num_pages; i++)
5148 set_page_dirty(eb->pages[i]);
5150 #ifdef CONFIG_BTRFS_DEBUG
5151 for (i = 0; i < num_pages; i++)
5152 ASSERT(PageDirty(eb->pages[i]));
5158 void clear_extent_buffer_uptodate(struct extent_buffer *eb)
5164 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5165 num_pages = num_extent_pages(eb);
5166 for (i = 0; i < num_pages; i++) {
5167 page = eb->pages[i];
5169 ClearPageUptodate(page);
5173 void set_extent_buffer_uptodate(struct extent_buffer *eb)
5179 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5180 num_pages = num_extent_pages(eb);
5181 for (i = 0; i < num_pages; i++) {
5182 page = eb->pages[i];
5183 SetPageUptodate(page);
5187 int read_extent_buffer_pages(struct extent_io_tree *tree,
5188 struct extent_buffer *eb, int wait, int mirror_num)
5194 int locked_pages = 0;
5195 int all_uptodate = 1;
5197 unsigned long num_reads = 0;
5198 struct bio *bio = NULL;
5199 unsigned long bio_flags = 0;
5201 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
5204 num_pages = num_extent_pages(eb);
5205 for (i = 0; i < num_pages; i++) {
5206 page = eb->pages[i];
5207 if (wait == WAIT_NONE) {
5208 if (!trylock_page(page))
5216 * We need to firstly lock all pages to make sure that
5217 * the uptodate bit of our pages won't be affected by
5218 * clear_extent_buffer_uptodate().
5220 for (i = 0; i < num_pages; i++) {
5221 page = eb->pages[i];
5222 if (!PageUptodate(page)) {
5229 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5233 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
5234 eb->read_mirror = 0;
5235 atomic_set(&eb->io_pages, num_reads);
5236 for (i = 0; i < num_pages; i++) {
5237 page = eb->pages[i];
5239 if (!PageUptodate(page)) {
5241 atomic_dec(&eb->io_pages);
5246 ClearPageError(page);
5247 err = __extent_read_full_page(tree, page,
5248 btree_get_extent, &bio,
5249 mirror_num, &bio_flags,
5254 * We use &bio in above __extent_read_full_page,
5255 * so we ensure that if it returns error, the
5256 * current page fails to add itself to bio and
5257 * it's been unlocked.
5259 * We must dec io_pages by ourselves.
5261 atomic_dec(&eb->io_pages);
5269 err = submit_one_bio(bio, mirror_num, bio_flags);
5274 if (ret || wait != WAIT_COMPLETE)
5277 for (i = 0; i < num_pages; i++) {
5278 page = eb->pages[i];
5279 wait_on_page_locked(page);
5280 if (!PageUptodate(page))
5287 while (locked_pages > 0) {
5289 page = eb->pages[locked_pages];
5295 void read_extent_buffer(const struct extent_buffer *eb, void *dstv,
5296 unsigned long start, unsigned long len)
5302 char *dst = (char *)dstv;
5303 size_t start_offset = offset_in_page(eb->start);
5304 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5306 if (start + len > eb->len) {
5307 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n",
5308 eb->start, eb->len, start, len);
5309 memset(dst, 0, len);
5313 offset = offset_in_page(start_offset + start);
5316 page = eb->pages[i];
5318 cur = min(len, (PAGE_SIZE - offset));
5319 kaddr = page_address(page);
5320 memcpy(dst, kaddr + offset, cur);
5329 int read_extent_buffer_to_user(const struct extent_buffer *eb,
5331 unsigned long start, unsigned long len)
5337 char __user *dst = (char __user *)dstv;
5338 size_t start_offset = offset_in_page(eb->start);
5339 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5342 WARN_ON(start > eb->len);
5343 WARN_ON(start + len > eb->start + eb->len);
5345 offset = offset_in_page(start_offset + start);
5348 page = eb->pages[i];
5350 cur = min(len, (PAGE_SIZE - offset));
5351 kaddr = page_address(page);
5352 if (copy_to_user(dst, kaddr + offset, cur)) {
5367 * return 0 if the item is found within a page.
5368 * return 1 if the item spans two pages.
5369 * return -EINVAL otherwise.
5371 int map_private_extent_buffer(const struct extent_buffer *eb,
5372 unsigned long start, unsigned long min_len,
5373 char **map, unsigned long *map_start,
5374 unsigned long *map_len)
5379 size_t start_offset = offset_in_page(eb->start);
5380 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5381 unsigned long end_i = (start_offset + start + min_len - 1) >>
5384 if (start + min_len > eb->len) {
5385 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n",
5386 eb->start, eb->len, start, min_len);
5394 offset = start_offset;
5398 *map_start = ((u64)i << PAGE_SHIFT) - start_offset;
5402 kaddr = page_address(p);
5403 *map = kaddr + offset;
5404 *map_len = PAGE_SIZE - offset;
5408 int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,
5409 unsigned long start, unsigned long len)
5415 char *ptr = (char *)ptrv;
5416 size_t start_offset = offset_in_page(eb->start);
5417 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5420 WARN_ON(start > eb->len);
5421 WARN_ON(start + len > eb->start + eb->len);
5423 offset = offset_in_page(start_offset + start);
5426 page = eb->pages[i];
5428 cur = min(len, (PAGE_SIZE - offset));
5430 kaddr = page_address(page);
5431 ret = memcmp(ptr, kaddr + offset, cur);
5443 void write_extent_buffer_chunk_tree_uuid(struct extent_buffer *eb,
5448 WARN_ON(!PageUptodate(eb->pages[0]));
5449 kaddr = page_address(eb->pages[0]);
5450 memcpy(kaddr + offsetof(struct btrfs_header, chunk_tree_uuid), srcv,
5454 void write_extent_buffer_fsid(struct extent_buffer *eb, const void *srcv)
5458 WARN_ON(!PageUptodate(eb->pages[0]));
5459 kaddr = page_address(eb->pages[0]);
5460 memcpy(kaddr + offsetof(struct btrfs_header, fsid), srcv,
5464 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
5465 unsigned long start, unsigned long len)
5471 char *src = (char *)srcv;
5472 size_t start_offset = offset_in_page(eb->start);
5473 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5475 WARN_ON(start > eb->len);
5476 WARN_ON(start + len > eb->start + eb->len);
5478 offset = offset_in_page(start_offset + start);
5481 page = eb->pages[i];
5482 WARN_ON(!PageUptodate(page));
5484 cur = min(len, PAGE_SIZE - offset);
5485 kaddr = page_address(page);
5486 memcpy(kaddr + offset, src, cur);
5495 void memzero_extent_buffer(struct extent_buffer *eb, unsigned long start,
5502 size_t start_offset = offset_in_page(eb->start);
5503 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5505 WARN_ON(start > eb->len);
5506 WARN_ON(start + len > eb->start + eb->len);
5508 offset = offset_in_page(start_offset + start);
5511 page = eb->pages[i];
5512 WARN_ON(!PageUptodate(page));
5514 cur = min(len, PAGE_SIZE - offset);
5515 kaddr = page_address(page);
5516 memset(kaddr + offset, 0, cur);
5524 void copy_extent_buffer_full(struct extent_buffer *dst,
5525 struct extent_buffer *src)
5530 ASSERT(dst->len == src->len);
5532 num_pages = num_extent_pages(dst);
5533 for (i = 0; i < num_pages; i++)
5534 copy_page(page_address(dst->pages[i]),
5535 page_address(src->pages[i]));
5538 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
5539 unsigned long dst_offset, unsigned long src_offset,
5542 u64 dst_len = dst->len;
5547 size_t start_offset = offset_in_page(dst->start);
5548 unsigned long i = (start_offset + dst_offset) >> PAGE_SHIFT;
5550 WARN_ON(src->len != dst_len);
5552 offset = offset_in_page(start_offset + dst_offset);
5555 page = dst->pages[i];
5556 WARN_ON(!PageUptodate(page));
5558 cur = min(len, (unsigned long)(PAGE_SIZE - offset));
5560 kaddr = page_address(page);
5561 read_extent_buffer(src, kaddr + offset, src_offset, cur);
5571 * eb_bitmap_offset() - calculate the page and offset of the byte containing the
5573 * @eb: the extent buffer
5574 * @start: offset of the bitmap item in the extent buffer
5576 * @page_index: return index of the page in the extent buffer that contains the
5578 * @page_offset: return offset into the page given by page_index
5580 * This helper hides the ugliness of finding the byte in an extent buffer which
5581 * contains a given bit.
5583 static inline void eb_bitmap_offset(struct extent_buffer *eb,
5584 unsigned long start, unsigned long nr,
5585 unsigned long *page_index,
5586 size_t *page_offset)
5588 size_t start_offset = offset_in_page(eb->start);
5589 size_t byte_offset = BIT_BYTE(nr);
5593 * The byte we want is the offset of the extent buffer + the offset of
5594 * the bitmap item in the extent buffer + the offset of the byte in the
5597 offset = start_offset + start + byte_offset;
5599 *page_index = offset >> PAGE_SHIFT;
5600 *page_offset = offset_in_page(offset);
5604 * extent_buffer_test_bit - determine whether a bit in a bitmap item is set
5605 * @eb: the extent buffer
5606 * @start: offset of the bitmap item in the extent buffer
5607 * @nr: bit number to test
5609 int extent_buffer_test_bit(struct extent_buffer *eb, unsigned long start,
5617 eb_bitmap_offset(eb, start, nr, &i, &offset);
5618 page = eb->pages[i];
5619 WARN_ON(!PageUptodate(page));
5620 kaddr = page_address(page);
5621 return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1)));
5625 * extent_buffer_bitmap_set - set an area of a bitmap
5626 * @eb: the extent buffer
5627 * @start: offset of the bitmap item in the extent buffer
5628 * @pos: bit number of the first bit
5629 * @len: number of bits to set
5631 void extent_buffer_bitmap_set(struct extent_buffer *eb, unsigned long start,
5632 unsigned long pos, unsigned long len)
5638 const unsigned int size = pos + len;
5639 int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5640 u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos);
5642 eb_bitmap_offset(eb, start, pos, &i, &offset);
5643 page = eb->pages[i];
5644 WARN_ON(!PageUptodate(page));
5645 kaddr = page_address(page);
5647 while (len >= bits_to_set) {
5648 kaddr[offset] |= mask_to_set;
5650 bits_to_set = BITS_PER_BYTE;
5652 if (++offset >= PAGE_SIZE && len > 0) {
5654 page = eb->pages[++i];
5655 WARN_ON(!PageUptodate(page));
5656 kaddr = page_address(page);
5660 mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
5661 kaddr[offset] |= mask_to_set;
5667 * extent_buffer_bitmap_clear - clear an area of a bitmap
5668 * @eb: the extent buffer
5669 * @start: offset of the bitmap item in the extent buffer
5670 * @pos: bit number of the first bit
5671 * @len: number of bits to clear
5673 void extent_buffer_bitmap_clear(struct extent_buffer *eb, unsigned long start,
5674 unsigned long pos, unsigned long len)
5680 const unsigned int size = pos + len;
5681 int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5682 u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos);
5684 eb_bitmap_offset(eb, start, pos, &i, &offset);
5685 page = eb->pages[i];
5686 WARN_ON(!PageUptodate(page));
5687 kaddr = page_address(page);
5689 while (len >= bits_to_clear) {
5690 kaddr[offset] &= ~mask_to_clear;
5691 len -= bits_to_clear;
5692 bits_to_clear = BITS_PER_BYTE;
5694 if (++offset >= PAGE_SIZE && len > 0) {
5696 page = eb->pages[++i];
5697 WARN_ON(!PageUptodate(page));
5698 kaddr = page_address(page);
5702 mask_to_clear &= BITMAP_LAST_BYTE_MASK(size);
5703 kaddr[offset] &= ~mask_to_clear;
5707 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
5709 unsigned long distance = (src > dst) ? src - dst : dst - src;
5710 return distance < len;
5713 static void copy_pages(struct page *dst_page, struct page *src_page,
5714 unsigned long dst_off, unsigned long src_off,
5717 char *dst_kaddr = page_address(dst_page);
5719 int must_memmove = 0;
5721 if (dst_page != src_page) {
5722 src_kaddr = page_address(src_page);
5724 src_kaddr = dst_kaddr;
5725 if (areas_overlap(src_off, dst_off, len))
5730 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
5732 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
5735 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5736 unsigned long src_offset, unsigned long len)
5738 struct btrfs_fs_info *fs_info = dst->fs_info;
5740 size_t dst_off_in_page;
5741 size_t src_off_in_page;
5742 size_t start_offset = offset_in_page(dst->start);
5743 unsigned long dst_i;
5744 unsigned long src_i;
5746 if (src_offset + len > dst->len) {
5748 "memmove bogus src_offset %lu move len %lu dst len %lu",
5749 src_offset, len, dst->len);
5752 if (dst_offset + len > dst->len) {
5754 "memmove bogus dst_offset %lu move len %lu dst len %lu",
5755 dst_offset, len, dst->len);
5760 dst_off_in_page = offset_in_page(start_offset + dst_offset);
5761 src_off_in_page = offset_in_page(start_offset + src_offset);
5763 dst_i = (start_offset + dst_offset) >> PAGE_SHIFT;
5764 src_i = (start_offset + src_offset) >> PAGE_SHIFT;
5766 cur = min(len, (unsigned long)(PAGE_SIZE -
5768 cur = min_t(unsigned long, cur,
5769 (unsigned long)(PAGE_SIZE - dst_off_in_page));
5771 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5772 dst_off_in_page, src_off_in_page, cur);
5780 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5781 unsigned long src_offset, unsigned long len)
5783 struct btrfs_fs_info *fs_info = dst->fs_info;
5785 size_t dst_off_in_page;
5786 size_t src_off_in_page;
5787 unsigned long dst_end = dst_offset + len - 1;
5788 unsigned long src_end = src_offset + len - 1;
5789 size_t start_offset = offset_in_page(dst->start);
5790 unsigned long dst_i;
5791 unsigned long src_i;
5793 if (src_offset + len > dst->len) {
5795 "memmove bogus src_offset %lu move len %lu len %lu",
5796 src_offset, len, dst->len);
5799 if (dst_offset + len > dst->len) {
5801 "memmove bogus dst_offset %lu move len %lu len %lu",
5802 dst_offset, len, dst->len);
5805 if (dst_offset < src_offset) {
5806 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
5810 dst_i = (start_offset + dst_end) >> PAGE_SHIFT;
5811 src_i = (start_offset + src_end) >> PAGE_SHIFT;
5813 dst_off_in_page = offset_in_page(start_offset + dst_end);
5814 src_off_in_page = offset_in_page(start_offset + src_end);
5816 cur = min_t(unsigned long, len, src_off_in_page + 1);
5817 cur = min(cur, dst_off_in_page + 1);
5818 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5819 dst_off_in_page - cur + 1,
5820 src_off_in_page - cur + 1, cur);
5828 int try_release_extent_buffer(struct page *page)
5830 struct extent_buffer *eb;
5833 * We need to make sure nobody is attaching this page to an eb right
5836 spin_lock(&page->mapping->private_lock);
5837 if (!PagePrivate(page)) {
5838 spin_unlock(&page->mapping->private_lock);
5842 eb = (struct extent_buffer *)page->private;
5846 * This is a little awful but should be ok, we need to make sure that
5847 * the eb doesn't disappear out from under us while we're looking at
5850 spin_lock(&eb->refs_lock);
5851 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
5852 spin_unlock(&eb->refs_lock);
5853 spin_unlock(&page->mapping->private_lock);
5856 spin_unlock(&page->mapping->private_lock);
5859 * If tree ref isn't set then we know the ref on this eb is a real ref,
5860 * so just return, this page will likely be freed soon anyway.
5862 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
5863 spin_unlock(&eb->refs_lock);
5867 return release_extent_buffer(eb);