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-io-tree.h"
18 #include "extent_map.h"
20 #include "btrfs_inode.h"
22 #include "check-integrity.h"
24 #include "rcu-string.h"
28 static struct kmem_cache *extent_state_cache;
29 static struct kmem_cache *extent_buffer_cache;
30 static struct bio_set btrfs_bioset;
32 static inline bool extent_state_in_tree(const struct extent_state *state)
34 return !RB_EMPTY_NODE(&state->rb_node);
37 #ifdef CONFIG_BTRFS_DEBUG
38 static LIST_HEAD(buffers);
39 static LIST_HEAD(states);
41 static DEFINE_SPINLOCK(leak_lock);
44 void btrfs_leak_debug_add(struct list_head *new, struct list_head *head)
48 spin_lock_irqsave(&leak_lock, flags);
50 spin_unlock_irqrestore(&leak_lock, flags);
54 void btrfs_leak_debug_del(struct list_head *entry)
58 spin_lock_irqsave(&leak_lock, flags);
60 spin_unlock_irqrestore(&leak_lock, flags);
63 static inline void btrfs_extent_buffer_leak_debug_check(void)
65 struct extent_buffer *eb;
67 while (!list_empty(&buffers)) {
68 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
69 pr_err("BTRFS: buffer leak start %llu len %lu refs %d bflags %lu\n",
70 eb->start, eb->len, atomic_read(&eb->refs), eb->bflags);
71 list_del(&eb->leak_list);
72 kmem_cache_free(extent_buffer_cache, eb);
76 static inline void btrfs_extent_state_leak_debug_check(void)
78 struct extent_state *state;
80 while (!list_empty(&states)) {
81 state = list_entry(states.next, struct extent_state, leak_list);
82 pr_err("BTRFS: state leak: start %llu end %llu state %u in tree %d refs %d\n",
83 state->start, state->end, state->state,
84 extent_state_in_tree(state),
85 refcount_read(&state->refs));
86 list_del(&state->leak_list);
87 kmem_cache_free(extent_state_cache, state);
91 #define btrfs_debug_check_extent_io_range(tree, start, end) \
92 __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end))
93 static inline void __btrfs_debug_check_extent_io_range(const char *caller,
94 struct extent_io_tree *tree, u64 start, u64 end)
96 struct inode *inode = tree->private_data;
99 if (!inode || !is_data_inode(inode))
102 isize = i_size_read(inode);
103 if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) {
104 btrfs_debug_rl(BTRFS_I(inode)->root->fs_info,
105 "%s: ino %llu isize %llu odd range [%llu,%llu]",
106 caller, btrfs_ino(BTRFS_I(inode)), isize, start, end);
110 #define btrfs_leak_debug_add(new, head) do {} while (0)
111 #define btrfs_leak_debug_del(entry) do {} while (0)
112 #define btrfs_extent_buffer_leak_debug_check() do {} while (0)
113 #define btrfs_extent_state_leak_debug_check() do {} while (0)
114 #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0)
120 struct rb_node rb_node;
123 struct extent_page_data {
125 struct extent_io_tree *tree;
126 /* tells writepage not to lock the state bits for this range
127 * it still does the unlocking
129 unsigned int extent_locked:1;
131 /* tells the submit_bio code to use REQ_SYNC */
132 unsigned int sync_io:1;
135 static int add_extent_changeset(struct extent_state *state, unsigned bits,
136 struct extent_changeset *changeset,
143 if (set && (state->state & bits) == bits)
145 if (!set && (state->state & bits) == 0)
147 changeset->bytes_changed += state->end - state->start + 1;
148 ret = ulist_add(&changeset->range_changed, state->start, state->end,
153 static int __must_check submit_one_bio(struct bio *bio, int mirror_num,
154 unsigned long bio_flags)
156 blk_status_t ret = 0;
157 struct extent_io_tree *tree = bio->bi_private;
159 bio->bi_private = NULL;
162 ret = tree->ops->submit_bio_hook(tree->private_data, bio,
163 mirror_num, bio_flags);
165 btrfsic_submit_bio(bio);
167 return blk_status_to_errno(ret);
170 /* Cleanup unsubmitted bios */
171 static void end_write_bio(struct extent_page_data *epd, int ret)
174 epd->bio->bi_status = errno_to_blk_status(ret);
181 * Submit bio from extent page data via submit_one_bio
183 * Return 0 if everything is OK.
184 * Return <0 for error.
186 static int __must_check flush_write_bio(struct extent_page_data *epd)
191 ret = submit_one_bio(epd->bio, 0, 0);
193 * Clean up of epd->bio is handled by its endio function.
194 * And endio is either triggered by successful bio execution
195 * or the error handler of submit bio hook.
196 * So at this point, no matter what happened, we don't need
197 * to clean up epd->bio.
204 int __init extent_state_cache_init(void)
206 extent_state_cache = kmem_cache_create("btrfs_extent_state",
207 sizeof(struct extent_state), 0,
208 SLAB_MEM_SPREAD, NULL);
209 if (!extent_state_cache)
214 int __init extent_io_init(void)
216 extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer",
217 sizeof(struct extent_buffer), 0,
218 SLAB_MEM_SPREAD, NULL);
219 if (!extent_buffer_cache)
222 if (bioset_init(&btrfs_bioset, BIO_POOL_SIZE,
223 offsetof(struct btrfs_io_bio, bio),
225 goto free_buffer_cache;
227 if (bioset_integrity_create(&btrfs_bioset, BIO_POOL_SIZE))
233 bioset_exit(&btrfs_bioset);
236 kmem_cache_destroy(extent_buffer_cache);
237 extent_buffer_cache = NULL;
241 void __cold extent_state_cache_exit(void)
243 btrfs_extent_state_leak_debug_check();
244 kmem_cache_destroy(extent_state_cache);
247 void __cold extent_io_exit(void)
249 btrfs_extent_buffer_leak_debug_check();
252 * Make sure all delayed rcu free are flushed before we
256 kmem_cache_destroy(extent_buffer_cache);
257 bioset_exit(&btrfs_bioset);
260 void extent_io_tree_init(struct btrfs_fs_info *fs_info,
261 struct extent_io_tree *tree, unsigned int owner,
264 tree->fs_info = fs_info;
265 tree->state = RB_ROOT;
267 tree->dirty_bytes = 0;
268 spin_lock_init(&tree->lock);
269 tree->private_data = private_data;
273 void extent_io_tree_release(struct extent_io_tree *tree)
275 spin_lock(&tree->lock);
277 * Do a single barrier for the waitqueue_active check here, the state
278 * of the waitqueue should not change once extent_io_tree_release is
282 while (!RB_EMPTY_ROOT(&tree->state)) {
283 struct rb_node *node;
284 struct extent_state *state;
286 node = rb_first(&tree->state);
287 state = rb_entry(node, struct extent_state, rb_node);
288 rb_erase(&state->rb_node, &tree->state);
289 RB_CLEAR_NODE(&state->rb_node);
291 * btree io trees aren't supposed to have tasks waiting for
292 * changes in the flags of extent states ever.
294 ASSERT(!waitqueue_active(&state->wq));
295 free_extent_state(state);
297 cond_resched_lock(&tree->lock);
299 spin_unlock(&tree->lock);
302 static struct extent_state *alloc_extent_state(gfp_t mask)
304 struct extent_state *state;
307 * The given mask might be not appropriate for the slab allocator,
308 * drop the unsupported bits
310 mask &= ~(__GFP_DMA32|__GFP_HIGHMEM);
311 state = kmem_cache_alloc(extent_state_cache, mask);
315 state->failrec = NULL;
316 RB_CLEAR_NODE(&state->rb_node);
317 btrfs_leak_debug_add(&state->leak_list, &states);
318 refcount_set(&state->refs, 1);
319 init_waitqueue_head(&state->wq);
320 trace_alloc_extent_state(state, mask, _RET_IP_);
324 void free_extent_state(struct extent_state *state)
328 if (refcount_dec_and_test(&state->refs)) {
329 WARN_ON(extent_state_in_tree(state));
330 btrfs_leak_debug_del(&state->leak_list);
331 trace_free_extent_state(state, _RET_IP_);
332 kmem_cache_free(extent_state_cache, state);
336 static struct rb_node *tree_insert(struct rb_root *root,
337 struct rb_node *search_start,
339 struct rb_node *node,
340 struct rb_node ***p_in,
341 struct rb_node **parent_in)
344 struct rb_node *parent = NULL;
345 struct tree_entry *entry;
347 if (p_in && parent_in) {
353 p = search_start ? &search_start : &root->rb_node;
356 entry = rb_entry(parent, struct tree_entry, rb_node);
358 if (offset < entry->start)
360 else if (offset > entry->end)
367 rb_link_node(node, parent, p);
368 rb_insert_color(node, root);
373 * __etree_search - searche @tree for an entry that contains @offset. Such
374 * entry would have entry->start <= offset && entry->end >= offset.
376 * @tree - the tree to search
377 * @offset - offset that should fall within an entry in @tree
378 * @next_ret - pointer to the first entry whose range ends after @offset
379 * @prev - pointer to the first entry whose range begins before @offset
380 * @p_ret - pointer where new node should be anchored (used when inserting an
382 * @parent_ret - points to entry which would have been the parent of the entry,
385 * This function returns a pointer to the entry that contains @offset byte
386 * address. If no such entry exists, then NULL is returned and the other
387 * pointer arguments to the function are filled, otherwise the found entry is
388 * returned and other pointers are left untouched.
390 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
391 struct rb_node **next_ret,
392 struct rb_node **prev_ret,
393 struct rb_node ***p_ret,
394 struct rb_node **parent_ret)
396 struct rb_root *root = &tree->state;
397 struct rb_node **n = &root->rb_node;
398 struct rb_node *prev = NULL;
399 struct rb_node *orig_prev = NULL;
400 struct tree_entry *entry;
401 struct tree_entry *prev_entry = NULL;
405 entry = rb_entry(prev, struct tree_entry, rb_node);
408 if (offset < entry->start)
410 else if (offset > entry->end)
423 while (prev && offset > prev_entry->end) {
424 prev = rb_next(prev);
425 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
432 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
433 while (prev && offset < prev_entry->start) {
434 prev = rb_prev(prev);
435 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
442 static inline struct rb_node *
443 tree_search_for_insert(struct extent_io_tree *tree,
445 struct rb_node ***p_ret,
446 struct rb_node **parent_ret)
448 struct rb_node *next= NULL;
451 ret = __etree_search(tree, offset, &next, NULL, p_ret, parent_ret);
457 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
460 return tree_search_for_insert(tree, offset, NULL, NULL);
464 * utility function to look for merge candidates inside a given range.
465 * Any extents with matching state are merged together into a single
466 * extent in the tree. Extents with EXTENT_IO in their state field
467 * are not merged because the end_io handlers need to be able to do
468 * operations on them without sleeping (or doing allocations/splits).
470 * This should be called with the tree lock held.
472 static void merge_state(struct extent_io_tree *tree,
473 struct extent_state *state)
475 struct extent_state *other;
476 struct rb_node *other_node;
478 if (state->state & (EXTENT_LOCKED | EXTENT_BOUNDARY))
481 other_node = rb_prev(&state->rb_node);
483 other = rb_entry(other_node, struct extent_state, rb_node);
484 if (other->end == state->start - 1 &&
485 other->state == state->state) {
486 if (tree->private_data &&
487 is_data_inode(tree->private_data))
488 btrfs_merge_delalloc_extent(tree->private_data,
490 state->start = other->start;
491 rb_erase(&other->rb_node, &tree->state);
492 RB_CLEAR_NODE(&other->rb_node);
493 free_extent_state(other);
496 other_node = rb_next(&state->rb_node);
498 other = rb_entry(other_node, struct extent_state, rb_node);
499 if (other->start == state->end + 1 &&
500 other->state == state->state) {
501 if (tree->private_data &&
502 is_data_inode(tree->private_data))
503 btrfs_merge_delalloc_extent(tree->private_data,
505 state->end = other->end;
506 rb_erase(&other->rb_node, &tree->state);
507 RB_CLEAR_NODE(&other->rb_node);
508 free_extent_state(other);
513 static void set_state_bits(struct extent_io_tree *tree,
514 struct extent_state *state, unsigned *bits,
515 struct extent_changeset *changeset);
518 * insert an extent_state struct into the tree. 'bits' are set on the
519 * struct before it is inserted.
521 * This may return -EEXIST if the extent is already there, in which case the
522 * state struct is freed.
524 * The tree lock is not taken internally. This is a utility function and
525 * probably isn't what you want to call (see set/clear_extent_bit).
527 static int insert_state(struct extent_io_tree *tree,
528 struct extent_state *state, u64 start, u64 end,
530 struct rb_node **parent,
531 unsigned *bits, struct extent_changeset *changeset)
533 struct rb_node *node;
536 btrfs_err(tree->fs_info,
537 "insert state: end < start %llu %llu", end, start);
540 state->start = start;
543 set_state_bits(tree, state, bits, changeset);
545 node = tree_insert(&tree->state, NULL, end, &state->rb_node, p, parent);
547 struct extent_state *found;
548 found = rb_entry(node, struct extent_state, rb_node);
549 btrfs_err(tree->fs_info,
550 "found node %llu %llu on insert of %llu %llu",
551 found->start, found->end, start, end);
554 merge_state(tree, state);
559 * split a given extent state struct in two, inserting the preallocated
560 * struct 'prealloc' as the newly created second half. 'split' indicates an
561 * offset inside 'orig' where it should be split.
564 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
565 * are two extent state structs in the tree:
566 * prealloc: [orig->start, split - 1]
567 * orig: [ split, orig->end ]
569 * The tree locks are not taken by this function. They need to be held
572 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
573 struct extent_state *prealloc, u64 split)
575 struct rb_node *node;
577 if (tree->private_data && is_data_inode(tree->private_data))
578 btrfs_split_delalloc_extent(tree->private_data, orig, split);
580 prealloc->start = orig->start;
581 prealloc->end = split - 1;
582 prealloc->state = orig->state;
585 node = tree_insert(&tree->state, &orig->rb_node, prealloc->end,
586 &prealloc->rb_node, NULL, NULL);
588 free_extent_state(prealloc);
594 static struct extent_state *next_state(struct extent_state *state)
596 struct rb_node *next = rb_next(&state->rb_node);
598 return rb_entry(next, struct extent_state, rb_node);
604 * utility function to clear some bits in an extent state struct.
605 * it will optionally wake up anyone waiting on this state (wake == 1).
607 * If no bits are set on the state struct after clearing things, the
608 * struct is freed and removed from the tree
610 static struct extent_state *clear_state_bit(struct extent_io_tree *tree,
611 struct extent_state *state,
612 unsigned *bits, int wake,
613 struct extent_changeset *changeset)
615 struct extent_state *next;
616 unsigned bits_to_clear = *bits & ~EXTENT_CTLBITS;
619 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
620 u64 range = state->end - state->start + 1;
621 WARN_ON(range > tree->dirty_bytes);
622 tree->dirty_bytes -= range;
625 if (tree->private_data && is_data_inode(tree->private_data))
626 btrfs_clear_delalloc_extent(tree->private_data, state, bits);
628 ret = add_extent_changeset(state, bits_to_clear, changeset, 0);
630 state->state &= ~bits_to_clear;
633 if (state->state == 0) {
634 next = next_state(state);
635 if (extent_state_in_tree(state)) {
636 rb_erase(&state->rb_node, &tree->state);
637 RB_CLEAR_NODE(&state->rb_node);
638 free_extent_state(state);
643 merge_state(tree, state);
644 next = next_state(state);
649 static struct extent_state *
650 alloc_extent_state_atomic(struct extent_state *prealloc)
653 prealloc = alloc_extent_state(GFP_ATOMIC);
658 static void extent_io_tree_panic(struct extent_io_tree *tree, int err)
660 struct inode *inode = tree->private_data;
662 btrfs_panic(btrfs_sb(inode->i_sb), err,
663 "locking error: extent tree was modified by another thread while locked");
667 * clear some bits on a range in the tree. This may require splitting
668 * or inserting elements in the tree, so the gfp mask is used to
669 * indicate which allocations or sleeping are allowed.
671 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
672 * the given range from the tree regardless of state (ie for truncate).
674 * the range [start, end] is inclusive.
676 * This takes the tree lock, and returns 0 on success and < 0 on error.
678 int __clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
679 unsigned bits, int wake, int delete,
680 struct extent_state **cached_state,
681 gfp_t mask, struct extent_changeset *changeset)
683 struct extent_state *state;
684 struct extent_state *cached;
685 struct extent_state *prealloc = NULL;
686 struct rb_node *node;
691 btrfs_debug_check_extent_io_range(tree, start, end);
692 trace_btrfs_clear_extent_bit(tree, start, end - start + 1, bits);
694 if (bits & EXTENT_DELALLOC)
695 bits |= EXTENT_NORESERVE;
698 bits |= ~EXTENT_CTLBITS;
700 if (bits & (EXTENT_LOCKED | EXTENT_BOUNDARY))
703 if (!prealloc && gfpflags_allow_blocking(mask)) {
705 * Don't care for allocation failure here because we might end
706 * up not needing the pre-allocated extent state at all, which
707 * is the case if we only have in the tree extent states that
708 * cover our input range and don't cover too any other range.
709 * If we end up needing a new extent state we allocate it later.
711 prealloc = alloc_extent_state(mask);
714 spin_lock(&tree->lock);
716 cached = *cached_state;
719 *cached_state = NULL;
723 if (cached && extent_state_in_tree(cached) &&
724 cached->start <= start && cached->end > start) {
726 refcount_dec(&cached->refs);
731 free_extent_state(cached);
734 * this search will find the extents that end after
737 node = tree_search(tree, start);
740 state = rb_entry(node, struct extent_state, rb_node);
742 if (state->start > end)
744 WARN_ON(state->end < start);
745 last_end = state->end;
747 /* the state doesn't have the wanted bits, go ahead */
748 if (!(state->state & bits)) {
749 state = next_state(state);
754 * | ---- desired range ---- |
756 * | ------------- state -------------- |
758 * We need to split the extent we found, and may flip
759 * bits on second half.
761 * If the extent we found extends past our range, we
762 * just split and search again. It'll get split again
763 * the next time though.
765 * If the extent we found is inside our range, we clear
766 * the desired bit on it.
769 if (state->start < start) {
770 prealloc = alloc_extent_state_atomic(prealloc);
772 err = split_state(tree, state, prealloc, start);
774 extent_io_tree_panic(tree, err);
779 if (state->end <= end) {
780 state = clear_state_bit(tree, state, &bits, wake,
787 * | ---- desired range ---- |
789 * We need to split the extent, and clear the bit
792 if (state->start <= end && state->end > end) {
793 prealloc = alloc_extent_state_atomic(prealloc);
795 err = split_state(tree, state, prealloc, end + 1);
797 extent_io_tree_panic(tree, err);
802 clear_state_bit(tree, prealloc, &bits, wake, changeset);
808 state = clear_state_bit(tree, state, &bits, wake, changeset);
810 if (last_end == (u64)-1)
812 start = last_end + 1;
813 if (start <= end && state && !need_resched())
819 spin_unlock(&tree->lock);
820 if (gfpflags_allow_blocking(mask))
825 spin_unlock(&tree->lock);
827 free_extent_state(prealloc);
833 static void wait_on_state(struct extent_io_tree *tree,
834 struct extent_state *state)
835 __releases(tree->lock)
836 __acquires(tree->lock)
839 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
840 spin_unlock(&tree->lock);
842 spin_lock(&tree->lock);
843 finish_wait(&state->wq, &wait);
847 * waits for one or more bits to clear on a range in the state tree.
848 * The range [start, end] is inclusive.
849 * The tree lock is taken by this function
851 static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
854 struct extent_state *state;
855 struct rb_node *node;
857 btrfs_debug_check_extent_io_range(tree, start, end);
859 spin_lock(&tree->lock);
863 * this search will find all the extents that end after
866 node = tree_search(tree, start);
871 state = rb_entry(node, struct extent_state, rb_node);
873 if (state->start > end)
876 if (state->state & bits) {
877 start = state->start;
878 refcount_inc(&state->refs);
879 wait_on_state(tree, state);
880 free_extent_state(state);
883 start = state->end + 1;
888 if (!cond_resched_lock(&tree->lock)) {
889 node = rb_next(node);
894 spin_unlock(&tree->lock);
897 static void set_state_bits(struct extent_io_tree *tree,
898 struct extent_state *state,
899 unsigned *bits, struct extent_changeset *changeset)
901 unsigned bits_to_set = *bits & ~EXTENT_CTLBITS;
904 if (tree->private_data && is_data_inode(tree->private_data))
905 btrfs_set_delalloc_extent(tree->private_data, state, bits);
907 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
908 u64 range = state->end - state->start + 1;
909 tree->dirty_bytes += range;
911 ret = add_extent_changeset(state, bits_to_set, changeset, 1);
913 state->state |= bits_to_set;
916 static void cache_state_if_flags(struct extent_state *state,
917 struct extent_state **cached_ptr,
920 if (cached_ptr && !(*cached_ptr)) {
921 if (!flags || (state->state & flags)) {
923 refcount_inc(&state->refs);
928 static void cache_state(struct extent_state *state,
929 struct extent_state **cached_ptr)
931 return cache_state_if_flags(state, cached_ptr,
932 EXTENT_LOCKED | EXTENT_BOUNDARY);
936 * set some bits on a range in the tree. This may require allocations or
937 * sleeping, so the gfp mask is used to indicate what is allowed.
939 * If any of the exclusive bits are set, this will fail with -EEXIST if some
940 * part of the range already has the desired bits set. The start of the
941 * existing range is returned in failed_start in this case.
943 * [start, end] is inclusive This takes the tree lock.
946 static int __must_check
947 __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
948 unsigned bits, unsigned exclusive_bits,
949 u64 *failed_start, struct extent_state **cached_state,
950 gfp_t mask, struct extent_changeset *changeset)
952 struct extent_state *state;
953 struct extent_state *prealloc = NULL;
954 struct rb_node *node;
956 struct rb_node *parent;
961 btrfs_debug_check_extent_io_range(tree, start, end);
962 trace_btrfs_set_extent_bit(tree, start, end - start + 1, bits);
965 if (!prealloc && gfpflags_allow_blocking(mask)) {
967 * Don't care for allocation failure here because we might end
968 * up not needing the pre-allocated extent state at all, which
969 * is the case if we only have in the tree extent states that
970 * cover our input range and don't cover too any other range.
971 * If we end up needing a new extent state we allocate it later.
973 prealloc = alloc_extent_state(mask);
976 spin_lock(&tree->lock);
977 if (cached_state && *cached_state) {
978 state = *cached_state;
979 if (state->start <= start && state->end > start &&
980 extent_state_in_tree(state)) {
981 node = &state->rb_node;
986 * this search will find all the extents that end after
989 node = tree_search_for_insert(tree, start, &p, &parent);
991 prealloc = alloc_extent_state_atomic(prealloc);
993 err = insert_state(tree, prealloc, start, end,
994 &p, &parent, &bits, changeset);
996 extent_io_tree_panic(tree, err);
998 cache_state(prealloc, cached_state);
1002 state = rb_entry(node, struct extent_state, rb_node);
1004 last_start = state->start;
1005 last_end = state->end;
1008 * | ---- desired range ---- |
1011 * Just lock what we found and keep going
1013 if (state->start == start && state->end <= end) {
1014 if (state->state & exclusive_bits) {
1015 *failed_start = state->start;
1020 set_state_bits(tree, state, &bits, changeset);
1021 cache_state(state, cached_state);
1022 merge_state(tree, state);
1023 if (last_end == (u64)-1)
1025 start = last_end + 1;
1026 state = next_state(state);
1027 if (start < end && state && state->start == start &&
1034 * | ---- desired range ---- |
1037 * | ------------- state -------------- |
1039 * We need to split the extent we found, and may flip bits on
1042 * If the extent we found extends past our
1043 * range, we just split and search again. It'll get split
1044 * again the next time though.
1046 * If the extent we found is inside our range, we set the
1047 * desired bit on it.
1049 if (state->start < start) {
1050 if (state->state & exclusive_bits) {
1051 *failed_start = start;
1056 prealloc = alloc_extent_state_atomic(prealloc);
1058 err = split_state(tree, state, prealloc, start);
1060 extent_io_tree_panic(tree, err);
1065 if (state->end <= end) {
1066 set_state_bits(tree, state, &bits, changeset);
1067 cache_state(state, cached_state);
1068 merge_state(tree, state);
1069 if (last_end == (u64)-1)
1071 start = last_end + 1;
1072 state = next_state(state);
1073 if (start < end && state && state->start == start &&
1080 * | ---- desired range ---- |
1081 * | state | or | state |
1083 * There's a hole, we need to insert something in it and
1084 * ignore the extent we found.
1086 if (state->start > start) {
1088 if (end < last_start)
1091 this_end = last_start - 1;
1093 prealloc = alloc_extent_state_atomic(prealloc);
1097 * Avoid to free 'prealloc' if it can be merged with
1100 err = insert_state(tree, prealloc, start, this_end,
1101 NULL, NULL, &bits, changeset);
1103 extent_io_tree_panic(tree, err);
1105 cache_state(prealloc, cached_state);
1107 start = this_end + 1;
1111 * | ---- desired range ---- |
1113 * We need to split the extent, and set the bit
1116 if (state->start <= end && state->end > end) {
1117 if (state->state & exclusive_bits) {
1118 *failed_start = start;
1123 prealloc = alloc_extent_state_atomic(prealloc);
1125 err = split_state(tree, state, prealloc, end + 1);
1127 extent_io_tree_panic(tree, err);
1129 set_state_bits(tree, prealloc, &bits, changeset);
1130 cache_state(prealloc, cached_state);
1131 merge_state(tree, prealloc);
1139 spin_unlock(&tree->lock);
1140 if (gfpflags_allow_blocking(mask))
1145 spin_unlock(&tree->lock);
1147 free_extent_state(prealloc);
1153 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1154 unsigned bits, u64 * failed_start,
1155 struct extent_state **cached_state, gfp_t mask)
1157 return __set_extent_bit(tree, start, end, bits, 0, failed_start,
1158 cached_state, mask, NULL);
1163 * convert_extent_bit - convert all bits in a given range from one bit to
1165 * @tree: the io tree to search
1166 * @start: the start offset in bytes
1167 * @end: the end offset in bytes (inclusive)
1168 * @bits: the bits to set in this range
1169 * @clear_bits: the bits to clear in this range
1170 * @cached_state: state that we're going to cache
1172 * This will go through and set bits for the given range. If any states exist
1173 * already in this range they are set with the given bit and cleared of the
1174 * clear_bits. This is only meant to be used by things that are mergeable, ie
1175 * converting from say DELALLOC to DIRTY. This is not meant to be used with
1176 * boundary bits like LOCK.
1178 * All allocations are done with GFP_NOFS.
1180 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1181 unsigned bits, unsigned clear_bits,
1182 struct extent_state **cached_state)
1184 struct extent_state *state;
1185 struct extent_state *prealloc = NULL;
1186 struct rb_node *node;
1188 struct rb_node *parent;
1192 bool first_iteration = true;
1194 btrfs_debug_check_extent_io_range(tree, start, end);
1195 trace_btrfs_convert_extent_bit(tree, start, end - start + 1, bits,
1201 * Best effort, don't worry if extent state allocation fails
1202 * here for the first iteration. We might have a cached state
1203 * that matches exactly the target range, in which case no
1204 * extent state allocations are needed. We'll only know this
1205 * after locking the tree.
1207 prealloc = alloc_extent_state(GFP_NOFS);
1208 if (!prealloc && !first_iteration)
1212 spin_lock(&tree->lock);
1213 if (cached_state && *cached_state) {
1214 state = *cached_state;
1215 if (state->start <= start && state->end > start &&
1216 extent_state_in_tree(state)) {
1217 node = &state->rb_node;
1223 * this search will find all the extents that end after
1226 node = tree_search_for_insert(tree, start, &p, &parent);
1228 prealloc = alloc_extent_state_atomic(prealloc);
1233 err = insert_state(tree, prealloc, start, end,
1234 &p, &parent, &bits, NULL);
1236 extent_io_tree_panic(tree, err);
1237 cache_state(prealloc, cached_state);
1241 state = rb_entry(node, struct extent_state, rb_node);
1243 last_start = state->start;
1244 last_end = state->end;
1247 * | ---- desired range ---- |
1250 * Just lock what we found and keep going
1252 if (state->start == start && state->end <= end) {
1253 set_state_bits(tree, state, &bits, NULL);
1254 cache_state(state, cached_state);
1255 state = clear_state_bit(tree, state, &clear_bits, 0, NULL);
1256 if (last_end == (u64)-1)
1258 start = last_end + 1;
1259 if (start < end && state && state->start == start &&
1266 * | ---- desired range ---- |
1269 * | ------------- state -------------- |
1271 * We need to split the extent we found, and may flip bits on
1274 * If the extent we found extends past our
1275 * range, we just split and search again. It'll get split
1276 * again the next time though.
1278 * If the extent we found is inside our range, we set the
1279 * desired bit on it.
1281 if (state->start < start) {
1282 prealloc = alloc_extent_state_atomic(prealloc);
1287 err = split_state(tree, state, prealloc, start);
1289 extent_io_tree_panic(tree, err);
1293 if (state->end <= end) {
1294 set_state_bits(tree, state, &bits, NULL);
1295 cache_state(state, cached_state);
1296 state = clear_state_bit(tree, state, &clear_bits, 0,
1298 if (last_end == (u64)-1)
1300 start = last_end + 1;
1301 if (start < end && state && state->start == start &&
1308 * | ---- desired range ---- |
1309 * | state | or | state |
1311 * There's a hole, we need to insert something in it and
1312 * ignore the extent we found.
1314 if (state->start > start) {
1316 if (end < last_start)
1319 this_end = last_start - 1;
1321 prealloc = alloc_extent_state_atomic(prealloc);
1328 * Avoid to free 'prealloc' if it can be merged with
1331 err = insert_state(tree, prealloc, start, this_end,
1332 NULL, NULL, &bits, NULL);
1334 extent_io_tree_panic(tree, err);
1335 cache_state(prealloc, cached_state);
1337 start = this_end + 1;
1341 * | ---- desired range ---- |
1343 * We need to split the extent, and set the bit
1346 if (state->start <= end && state->end > end) {
1347 prealloc = alloc_extent_state_atomic(prealloc);
1353 err = split_state(tree, state, prealloc, end + 1);
1355 extent_io_tree_panic(tree, err);
1357 set_state_bits(tree, prealloc, &bits, NULL);
1358 cache_state(prealloc, cached_state);
1359 clear_state_bit(tree, prealloc, &clear_bits, 0, NULL);
1367 spin_unlock(&tree->lock);
1369 first_iteration = false;
1373 spin_unlock(&tree->lock);
1375 free_extent_state(prealloc);
1380 /* wrappers around set/clear extent bit */
1381 int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1382 unsigned bits, struct extent_changeset *changeset)
1385 * We don't support EXTENT_LOCKED yet, as current changeset will
1386 * record any bits changed, so for EXTENT_LOCKED case, it will
1387 * either fail with -EEXIST or changeset will record the whole
1390 BUG_ON(bits & EXTENT_LOCKED);
1392 return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL, GFP_NOFS,
1396 int set_extent_bits_nowait(struct extent_io_tree *tree, u64 start, u64 end,
1399 return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL,
1403 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1404 unsigned bits, int wake, int delete,
1405 struct extent_state **cached)
1407 return __clear_extent_bit(tree, start, end, bits, wake, delete,
1408 cached, GFP_NOFS, NULL);
1411 int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1412 unsigned bits, struct extent_changeset *changeset)
1415 * Don't support EXTENT_LOCKED case, same reason as
1416 * set_record_extent_bits().
1418 BUG_ON(bits & EXTENT_LOCKED);
1420 return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, GFP_NOFS,
1425 * either insert or lock state struct between start and end use mask to tell
1426 * us if waiting is desired.
1428 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1429 struct extent_state **cached_state)
1435 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED,
1436 EXTENT_LOCKED, &failed_start,
1437 cached_state, GFP_NOFS, NULL);
1438 if (err == -EEXIST) {
1439 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1440 start = failed_start;
1443 WARN_ON(start > end);
1448 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1453 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1454 &failed_start, NULL, GFP_NOFS, NULL);
1455 if (err == -EEXIST) {
1456 if (failed_start > start)
1457 clear_extent_bit(tree, start, failed_start - 1,
1458 EXTENT_LOCKED, 1, 0, NULL);
1464 void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
1466 unsigned long index = start >> PAGE_SHIFT;
1467 unsigned long end_index = end >> PAGE_SHIFT;
1470 while (index <= end_index) {
1471 page = find_get_page(inode->i_mapping, index);
1472 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1473 clear_page_dirty_for_io(page);
1479 void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
1481 unsigned long index = start >> PAGE_SHIFT;
1482 unsigned long end_index = end >> PAGE_SHIFT;
1485 while (index <= end_index) {
1486 page = find_get_page(inode->i_mapping, index);
1487 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1488 __set_page_dirty_nobuffers(page);
1489 account_page_redirty(page);
1495 /* find the first state struct with 'bits' set after 'start', and
1496 * return it. tree->lock must be held. NULL will returned if
1497 * nothing was found after 'start'
1499 static struct extent_state *
1500 find_first_extent_bit_state(struct extent_io_tree *tree,
1501 u64 start, unsigned bits)
1503 struct rb_node *node;
1504 struct extent_state *state;
1507 * this search will find all the extents that end after
1510 node = tree_search(tree, start);
1515 state = rb_entry(node, struct extent_state, rb_node);
1516 if (state->end >= start && (state->state & bits))
1519 node = rb_next(node);
1528 * find the first offset in the io tree with 'bits' set. zero is
1529 * returned if we find something, and *start_ret and *end_ret are
1530 * set to reflect the state struct that was found.
1532 * If nothing was found, 1 is returned. If found something, return 0.
1534 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1535 u64 *start_ret, u64 *end_ret, unsigned bits,
1536 struct extent_state **cached_state)
1538 struct extent_state *state;
1541 spin_lock(&tree->lock);
1542 if (cached_state && *cached_state) {
1543 state = *cached_state;
1544 if (state->end == start - 1 && extent_state_in_tree(state)) {
1545 while ((state = next_state(state)) != NULL) {
1546 if (state->state & bits)
1549 free_extent_state(*cached_state);
1550 *cached_state = NULL;
1553 free_extent_state(*cached_state);
1554 *cached_state = NULL;
1557 state = find_first_extent_bit_state(tree, start, bits);
1560 cache_state_if_flags(state, cached_state, 0);
1561 *start_ret = state->start;
1562 *end_ret = state->end;
1566 spin_unlock(&tree->lock);
1571 * find_first_clear_extent_bit - find the first range that has @bits not set.
1572 * This range could start before @start.
1574 * @tree - the tree to search
1575 * @start - the offset at/after which the found extent should start
1576 * @start_ret - records the beginning of the range
1577 * @end_ret - records the end of the range (inclusive)
1578 * @bits - the set of bits which must be unset
1580 * Since unallocated range is also considered one which doesn't have the bits
1581 * set it's possible that @end_ret contains -1, this happens in case the range
1582 * spans (last_range_end, end of device]. In this case it's up to the caller to
1583 * trim @end_ret to the appropriate size.
1585 void find_first_clear_extent_bit(struct extent_io_tree *tree, u64 start,
1586 u64 *start_ret, u64 *end_ret, unsigned bits)
1588 struct extent_state *state;
1589 struct rb_node *node, *prev = NULL, *next;
1591 spin_lock(&tree->lock);
1593 /* Find first extent with bits cleared */
1595 node = __etree_search(tree, start, &next, &prev, NULL, NULL);
1600 * We are past the last allocated chunk,
1601 * set start at the end of the last extent. The
1602 * device alloc tree should never be empty so
1603 * prev is always set.
1606 state = rb_entry(prev, struct extent_state, rb_node);
1607 *start_ret = state->end + 1;
1613 * At this point 'node' either contains 'start' or start is
1616 state = rb_entry(node, struct extent_state, rb_node);
1618 if (in_range(start, state->start, state->end - state->start + 1)) {
1619 if (state->state & bits) {
1621 * |--range with bits sets--|
1625 start = state->end + 1;
1628 * 'start' falls within a range that doesn't
1629 * have the bits set, so take its start as
1630 * the beginning of the desired range
1632 * |--range with bits cleared----|
1636 *start_ret = state->start;
1641 * |---prev range---|---hole/unset---|---node range---|
1647 * |---hole/unset--||--first node--|
1652 state = rb_entry(prev, struct extent_state,
1654 *start_ret = state->end + 1;
1663 * Find the longest stretch from start until an entry which has the
1667 state = rb_entry(node, struct extent_state, rb_node);
1668 if (state->end >= start && !(state->state & bits)) {
1669 *end_ret = state->end;
1671 *end_ret = state->start - 1;
1675 node = rb_next(node);
1680 spin_unlock(&tree->lock);
1684 * find a contiguous range of bytes in the file marked as delalloc, not
1685 * more than 'max_bytes'. start and end are used to return the range,
1687 * true is returned if we find something, false if nothing was in the tree
1689 bool btrfs_find_delalloc_range(struct extent_io_tree *tree, u64 *start,
1690 u64 *end, u64 max_bytes,
1691 struct extent_state **cached_state)
1693 struct rb_node *node;
1694 struct extent_state *state;
1695 u64 cur_start = *start;
1697 u64 total_bytes = 0;
1699 spin_lock(&tree->lock);
1702 * this search will find all the extents that end after
1705 node = tree_search(tree, cur_start);
1712 state = rb_entry(node, struct extent_state, rb_node);
1713 if (found && (state->start != cur_start ||
1714 (state->state & EXTENT_BOUNDARY))) {
1717 if (!(state->state & EXTENT_DELALLOC)) {
1723 *start = state->start;
1724 *cached_state = state;
1725 refcount_inc(&state->refs);
1729 cur_start = state->end + 1;
1730 node = rb_next(node);
1731 total_bytes += state->end - state->start + 1;
1732 if (total_bytes >= max_bytes)
1738 spin_unlock(&tree->lock);
1742 static int __process_pages_contig(struct address_space *mapping,
1743 struct page *locked_page,
1744 pgoff_t start_index, pgoff_t end_index,
1745 unsigned long page_ops, pgoff_t *index_ret);
1747 static noinline void __unlock_for_delalloc(struct inode *inode,
1748 struct page *locked_page,
1751 unsigned long index = start >> PAGE_SHIFT;
1752 unsigned long end_index = end >> PAGE_SHIFT;
1754 ASSERT(locked_page);
1755 if (index == locked_page->index && end_index == index)
1758 __process_pages_contig(inode->i_mapping, locked_page, index, end_index,
1762 static noinline int lock_delalloc_pages(struct inode *inode,
1763 struct page *locked_page,
1767 unsigned long index = delalloc_start >> PAGE_SHIFT;
1768 unsigned long index_ret = index;
1769 unsigned long end_index = delalloc_end >> PAGE_SHIFT;
1772 ASSERT(locked_page);
1773 if (index == locked_page->index && index == end_index)
1776 ret = __process_pages_contig(inode->i_mapping, locked_page, index,
1777 end_index, PAGE_LOCK, &index_ret);
1779 __unlock_for_delalloc(inode, locked_page, delalloc_start,
1780 (u64)index_ret << PAGE_SHIFT);
1785 * Find and lock a contiguous range of bytes in the file marked as delalloc, no
1786 * more than @max_bytes. @Start and @end are used to return the range,
1788 * Return: true if we find something
1789 * false if nothing was in the tree
1792 noinline_for_stack bool find_lock_delalloc_range(struct inode *inode,
1793 struct page *locked_page, u64 *start,
1796 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
1797 u64 max_bytes = BTRFS_MAX_EXTENT_SIZE;
1801 struct extent_state *cached_state = NULL;
1806 /* step one, find a bunch of delalloc bytes starting at start */
1807 delalloc_start = *start;
1809 found = btrfs_find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1810 max_bytes, &cached_state);
1811 if (!found || delalloc_end <= *start) {
1812 *start = delalloc_start;
1813 *end = delalloc_end;
1814 free_extent_state(cached_state);
1819 * start comes from the offset of locked_page. We have to lock
1820 * pages in order, so we can't process delalloc bytes before
1823 if (delalloc_start < *start)
1824 delalloc_start = *start;
1827 * make sure to limit the number of pages we try to lock down
1829 if (delalloc_end + 1 - delalloc_start > max_bytes)
1830 delalloc_end = delalloc_start + max_bytes - 1;
1832 /* step two, lock all the pages after the page that has start */
1833 ret = lock_delalloc_pages(inode, locked_page,
1834 delalloc_start, delalloc_end);
1835 ASSERT(!ret || ret == -EAGAIN);
1836 if (ret == -EAGAIN) {
1837 /* some of the pages are gone, lets avoid looping by
1838 * shortening the size of the delalloc range we're searching
1840 free_extent_state(cached_state);
1841 cached_state = NULL;
1843 max_bytes = PAGE_SIZE;
1852 /* step three, lock the state bits for the whole range */
1853 lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state);
1855 /* then test to make sure it is all still delalloc */
1856 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1857 EXTENT_DELALLOC, 1, cached_state);
1859 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1861 __unlock_for_delalloc(inode, locked_page,
1862 delalloc_start, delalloc_end);
1866 free_extent_state(cached_state);
1867 *start = delalloc_start;
1868 *end = delalloc_end;
1873 static int __process_pages_contig(struct address_space *mapping,
1874 struct page *locked_page,
1875 pgoff_t start_index, pgoff_t end_index,
1876 unsigned long page_ops, pgoff_t *index_ret)
1878 unsigned long nr_pages = end_index - start_index + 1;
1879 unsigned long pages_locked = 0;
1880 pgoff_t index = start_index;
1881 struct page *pages[16];
1886 if (page_ops & PAGE_LOCK) {
1887 ASSERT(page_ops == PAGE_LOCK);
1888 ASSERT(index_ret && *index_ret == start_index);
1891 if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0)
1892 mapping_set_error(mapping, -EIO);
1894 while (nr_pages > 0) {
1895 ret = find_get_pages_contig(mapping, index,
1896 min_t(unsigned long,
1897 nr_pages, ARRAY_SIZE(pages)), pages);
1900 * Only if we're going to lock these pages,
1901 * can we find nothing at @index.
1903 ASSERT(page_ops & PAGE_LOCK);
1908 for (i = 0; i < ret; i++) {
1909 if (page_ops & PAGE_SET_PRIVATE2)
1910 SetPagePrivate2(pages[i]);
1912 if (locked_page && pages[i] == locked_page) {
1917 if (page_ops & PAGE_CLEAR_DIRTY)
1918 clear_page_dirty_for_io(pages[i]);
1919 if (page_ops & PAGE_SET_WRITEBACK)
1920 set_page_writeback(pages[i]);
1921 if (page_ops & PAGE_SET_ERROR)
1922 SetPageError(pages[i]);
1923 if (page_ops & PAGE_END_WRITEBACK)
1924 end_page_writeback(pages[i]);
1925 if (page_ops & PAGE_UNLOCK)
1926 unlock_page(pages[i]);
1927 if (page_ops & PAGE_LOCK) {
1928 lock_page(pages[i]);
1929 if (!PageDirty(pages[i]) ||
1930 pages[i]->mapping != mapping) {
1931 unlock_page(pages[i]);
1945 if (err && index_ret)
1946 *index_ret = start_index + pages_locked - 1;
1950 void extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end,
1951 struct page *locked_page,
1952 unsigned clear_bits,
1953 unsigned long page_ops)
1955 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, clear_bits, 1, 0,
1958 __process_pages_contig(inode->i_mapping, locked_page,
1959 start >> PAGE_SHIFT, end >> PAGE_SHIFT,
1964 * count the number of bytes in the tree that have a given bit(s)
1965 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1966 * cached. The total number found is returned.
1968 u64 count_range_bits(struct extent_io_tree *tree,
1969 u64 *start, u64 search_end, u64 max_bytes,
1970 unsigned bits, int contig)
1972 struct rb_node *node;
1973 struct extent_state *state;
1974 u64 cur_start = *start;
1975 u64 total_bytes = 0;
1979 if (WARN_ON(search_end <= cur_start))
1982 spin_lock(&tree->lock);
1983 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1984 total_bytes = tree->dirty_bytes;
1988 * this search will find all the extents that end after
1991 node = tree_search(tree, cur_start);
1996 state = rb_entry(node, struct extent_state, rb_node);
1997 if (state->start > search_end)
1999 if (contig && found && state->start > last + 1)
2001 if (state->end >= cur_start && (state->state & bits) == bits) {
2002 total_bytes += min(search_end, state->end) + 1 -
2003 max(cur_start, state->start);
2004 if (total_bytes >= max_bytes)
2007 *start = max(cur_start, state->start);
2011 } else if (contig && found) {
2014 node = rb_next(node);
2019 spin_unlock(&tree->lock);
2024 * set the private field for a given byte offset in the tree. If there isn't
2025 * an extent_state there already, this does nothing.
2027 int set_state_failrec(struct extent_io_tree *tree, u64 start,
2028 struct io_failure_record *failrec)
2030 struct rb_node *node;
2031 struct extent_state *state;
2034 spin_lock(&tree->lock);
2036 * this search will find all the extents that end after
2039 node = tree_search(tree, start);
2044 state = rb_entry(node, struct extent_state, rb_node);
2045 if (state->start != start) {
2049 state->failrec = failrec;
2051 spin_unlock(&tree->lock);
2055 int get_state_failrec(struct extent_io_tree *tree, u64 start,
2056 struct io_failure_record **failrec)
2058 struct rb_node *node;
2059 struct extent_state *state;
2062 spin_lock(&tree->lock);
2064 * this search will find all the extents that end after
2067 node = tree_search(tree, start);
2072 state = rb_entry(node, struct extent_state, rb_node);
2073 if (state->start != start) {
2077 *failrec = state->failrec;
2079 spin_unlock(&tree->lock);
2084 * searches a range in the state tree for a given mask.
2085 * If 'filled' == 1, this returns 1 only if every extent in the tree
2086 * has the bits set. Otherwise, 1 is returned if any bit in the
2087 * range is found set.
2089 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
2090 unsigned bits, int filled, struct extent_state *cached)
2092 struct extent_state *state = NULL;
2093 struct rb_node *node;
2096 spin_lock(&tree->lock);
2097 if (cached && extent_state_in_tree(cached) && cached->start <= start &&
2098 cached->end > start)
2099 node = &cached->rb_node;
2101 node = tree_search(tree, start);
2102 while (node && start <= end) {
2103 state = rb_entry(node, struct extent_state, rb_node);
2105 if (filled && state->start > start) {
2110 if (state->start > end)
2113 if (state->state & bits) {
2117 } else if (filled) {
2122 if (state->end == (u64)-1)
2125 start = state->end + 1;
2128 node = rb_next(node);
2135 spin_unlock(&tree->lock);
2140 * helper function to set a given page up to date if all the
2141 * extents in the tree for that page are up to date
2143 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
2145 u64 start = page_offset(page);
2146 u64 end = start + PAGE_SIZE - 1;
2147 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
2148 SetPageUptodate(page);
2151 int free_io_failure(struct extent_io_tree *failure_tree,
2152 struct extent_io_tree *io_tree,
2153 struct io_failure_record *rec)
2158 set_state_failrec(failure_tree, rec->start, NULL);
2159 ret = clear_extent_bits(failure_tree, rec->start,
2160 rec->start + rec->len - 1,
2161 EXTENT_LOCKED | EXTENT_DIRTY);
2165 ret = clear_extent_bits(io_tree, rec->start,
2166 rec->start + rec->len - 1,
2176 * this bypasses the standard btrfs submit functions deliberately, as
2177 * the standard behavior is to write all copies in a raid setup. here we only
2178 * want to write the one bad copy. so we do the mapping for ourselves and issue
2179 * submit_bio directly.
2180 * to avoid any synchronization issues, wait for the data after writing, which
2181 * actually prevents the read that triggered the error from finishing.
2182 * currently, there can be no more than two copies of every data bit. thus,
2183 * exactly one rewrite is required.
2185 int repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start,
2186 u64 length, u64 logical, struct page *page,
2187 unsigned int pg_offset, int mirror_num)
2190 struct btrfs_device *dev;
2193 struct btrfs_bio *bbio = NULL;
2196 ASSERT(!(fs_info->sb->s_flags & SB_RDONLY));
2197 BUG_ON(!mirror_num);
2199 bio = btrfs_io_bio_alloc(1);
2200 bio->bi_iter.bi_size = 0;
2201 map_length = length;
2204 * Avoid races with device replace and make sure our bbio has devices
2205 * associated to its stripes that don't go away while we are doing the
2206 * read repair operation.
2208 btrfs_bio_counter_inc_blocked(fs_info);
2209 if (btrfs_is_parity_mirror(fs_info, logical, length)) {
2211 * Note that we don't use BTRFS_MAP_WRITE because it's supposed
2212 * to update all raid stripes, but here we just want to correct
2213 * bad stripe, thus BTRFS_MAP_READ is abused to only get the bad
2214 * stripe's dev and sector.
2216 ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, logical,
2217 &map_length, &bbio, 0);
2219 btrfs_bio_counter_dec(fs_info);
2223 ASSERT(bbio->mirror_num == 1);
2225 ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical,
2226 &map_length, &bbio, mirror_num);
2228 btrfs_bio_counter_dec(fs_info);
2232 BUG_ON(mirror_num != bbio->mirror_num);
2235 sector = bbio->stripes[bbio->mirror_num - 1].physical >> 9;
2236 bio->bi_iter.bi_sector = sector;
2237 dev = bbio->stripes[bbio->mirror_num - 1].dev;
2238 btrfs_put_bbio(bbio);
2239 if (!dev || !dev->bdev ||
2240 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) {
2241 btrfs_bio_counter_dec(fs_info);
2245 bio_set_dev(bio, dev->bdev);
2246 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC;
2247 bio_add_page(bio, page, length, pg_offset);
2249 if (btrfsic_submit_bio_wait(bio)) {
2250 /* try to remap that extent elsewhere? */
2251 btrfs_bio_counter_dec(fs_info);
2253 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
2257 btrfs_info_rl_in_rcu(fs_info,
2258 "read error corrected: ino %llu off %llu (dev %s sector %llu)",
2260 rcu_str_deref(dev->name), sector);
2261 btrfs_bio_counter_dec(fs_info);
2266 int btrfs_repair_eb_io_failure(struct extent_buffer *eb, int mirror_num)
2268 struct btrfs_fs_info *fs_info = eb->fs_info;
2269 u64 start = eb->start;
2270 int i, num_pages = num_extent_pages(eb);
2273 if (sb_rdonly(fs_info->sb))
2276 for (i = 0; i < num_pages; i++) {
2277 struct page *p = eb->pages[i];
2279 ret = repair_io_failure(fs_info, 0, start, PAGE_SIZE, start, p,
2280 start - page_offset(p), mirror_num);
2290 * each time an IO finishes, we do a fast check in the IO failure tree
2291 * to see if we need to process or clean up an io_failure_record
2293 int clean_io_failure(struct btrfs_fs_info *fs_info,
2294 struct extent_io_tree *failure_tree,
2295 struct extent_io_tree *io_tree, u64 start,
2296 struct page *page, u64 ino, unsigned int pg_offset)
2299 struct io_failure_record *failrec;
2300 struct extent_state *state;
2305 ret = count_range_bits(failure_tree, &private, (u64)-1, 1,
2310 ret = get_state_failrec(failure_tree, start, &failrec);
2314 BUG_ON(!failrec->this_mirror);
2316 if (failrec->in_validation) {
2317 /* there was no real error, just free the record */
2318 btrfs_debug(fs_info,
2319 "clean_io_failure: freeing dummy error at %llu",
2323 if (sb_rdonly(fs_info->sb))
2326 spin_lock(&io_tree->lock);
2327 state = find_first_extent_bit_state(io_tree,
2330 spin_unlock(&io_tree->lock);
2332 if (state && state->start <= failrec->start &&
2333 state->end >= failrec->start + failrec->len - 1) {
2334 num_copies = btrfs_num_copies(fs_info, failrec->logical,
2336 if (num_copies > 1) {
2337 repair_io_failure(fs_info, ino, start, failrec->len,
2338 failrec->logical, page, pg_offset,
2339 failrec->failed_mirror);
2344 free_io_failure(failure_tree, io_tree, failrec);
2350 * Can be called when
2351 * - hold extent lock
2352 * - under ordered extent
2353 * - the inode is freeing
2355 void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start, u64 end)
2357 struct extent_io_tree *failure_tree = &inode->io_failure_tree;
2358 struct io_failure_record *failrec;
2359 struct extent_state *state, *next;
2361 if (RB_EMPTY_ROOT(&failure_tree->state))
2364 spin_lock(&failure_tree->lock);
2365 state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY);
2367 if (state->start > end)
2370 ASSERT(state->end <= end);
2372 next = next_state(state);
2374 failrec = state->failrec;
2375 free_extent_state(state);
2380 spin_unlock(&failure_tree->lock);
2383 int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end,
2384 struct io_failure_record **failrec_ret)
2386 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2387 struct io_failure_record *failrec;
2388 struct extent_map *em;
2389 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2390 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2391 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2395 ret = get_state_failrec(failure_tree, start, &failrec);
2397 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2401 failrec->start = start;
2402 failrec->len = end - start + 1;
2403 failrec->this_mirror = 0;
2404 failrec->bio_flags = 0;
2405 failrec->in_validation = 0;
2407 read_lock(&em_tree->lock);
2408 em = lookup_extent_mapping(em_tree, start, failrec->len);
2410 read_unlock(&em_tree->lock);
2415 if (em->start > start || em->start + em->len <= start) {
2416 free_extent_map(em);
2419 read_unlock(&em_tree->lock);
2425 logical = start - em->start;
2426 logical = em->block_start + logical;
2427 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2428 logical = em->block_start;
2429 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2430 extent_set_compress_type(&failrec->bio_flags,
2434 btrfs_debug(fs_info,
2435 "Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu",
2436 logical, start, failrec->len);
2438 failrec->logical = logical;
2439 free_extent_map(em);
2441 /* set the bits in the private failure tree */
2442 ret = set_extent_bits(failure_tree, start, end,
2443 EXTENT_LOCKED | EXTENT_DIRTY);
2445 ret = set_state_failrec(failure_tree, start, failrec);
2446 /* set the bits in the inode's tree */
2448 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED);
2454 btrfs_debug(fs_info,
2455 "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d",
2456 failrec->logical, failrec->start, failrec->len,
2457 failrec->in_validation);
2459 * when data can be on disk more than twice, add to failrec here
2460 * (e.g. with a list for failed_mirror) to make
2461 * clean_io_failure() clean all those errors at once.
2465 *failrec_ret = failrec;
2470 bool btrfs_check_repairable(struct inode *inode, unsigned failed_bio_pages,
2471 struct io_failure_record *failrec, int failed_mirror)
2473 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2476 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
2477 if (num_copies == 1) {
2479 * we only have a single copy of the data, so don't bother with
2480 * all the retry and error correction code that follows. no
2481 * matter what the error is, it is very likely to persist.
2483 btrfs_debug(fs_info,
2484 "Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
2485 num_copies, failrec->this_mirror, failed_mirror);
2490 * there are two premises:
2491 * a) deliver good data to the caller
2492 * b) correct the bad sectors on disk
2494 if (failed_bio_pages > 1) {
2496 * to fulfill b), we need to know the exact failing sectors, as
2497 * we don't want to rewrite any more than the failed ones. thus,
2498 * we need separate read requests for the failed bio
2500 * if the following BUG_ON triggers, our validation request got
2501 * merged. we need separate requests for our algorithm to work.
2503 BUG_ON(failrec->in_validation);
2504 failrec->in_validation = 1;
2505 failrec->this_mirror = failed_mirror;
2508 * we're ready to fulfill a) and b) alongside. get a good copy
2509 * of the failed sector and if we succeed, we have setup
2510 * everything for repair_io_failure to do the rest for us.
2512 if (failrec->in_validation) {
2513 BUG_ON(failrec->this_mirror != failed_mirror);
2514 failrec->in_validation = 0;
2515 failrec->this_mirror = 0;
2517 failrec->failed_mirror = failed_mirror;
2518 failrec->this_mirror++;
2519 if (failrec->this_mirror == failed_mirror)
2520 failrec->this_mirror++;
2523 if (failrec->this_mirror > num_copies) {
2524 btrfs_debug(fs_info,
2525 "Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
2526 num_copies, failrec->this_mirror, failed_mirror);
2534 struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio,
2535 struct io_failure_record *failrec,
2536 struct page *page, int pg_offset, int icsum,
2537 bio_end_io_t *endio_func, void *data)
2539 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2541 struct btrfs_io_bio *btrfs_failed_bio;
2542 struct btrfs_io_bio *btrfs_bio;
2544 bio = btrfs_io_bio_alloc(1);
2545 bio->bi_end_io = endio_func;
2546 bio->bi_iter.bi_sector = failrec->logical >> 9;
2547 bio->bi_iter.bi_size = 0;
2548 bio->bi_private = data;
2550 btrfs_failed_bio = btrfs_io_bio(failed_bio);
2551 if (btrfs_failed_bio->csum) {
2552 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
2554 btrfs_bio = btrfs_io_bio(bio);
2555 btrfs_bio->csum = btrfs_bio->csum_inline;
2557 memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + icsum,
2561 bio_add_page(bio, page, failrec->len, pg_offset);
2567 * This is a generic handler for readpage errors. If other copies exist, read
2568 * those and write back good data to the failed position. Does not investigate
2569 * in remapping the failed extent elsewhere, hoping the device will be smart
2570 * enough to do this as needed
2572 static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset,
2573 struct page *page, u64 start, u64 end,
2576 struct io_failure_record *failrec;
2577 struct inode *inode = page->mapping->host;
2578 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2579 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2582 blk_status_t status;
2584 unsigned failed_bio_pages = failed_bio->bi_iter.bi_size >> PAGE_SHIFT;
2586 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
2588 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
2592 if (!btrfs_check_repairable(inode, failed_bio_pages, failrec,
2594 free_io_failure(failure_tree, tree, failrec);
2598 if (failed_bio_pages > 1)
2599 read_mode |= REQ_FAILFAST_DEV;
2601 phy_offset >>= inode->i_sb->s_blocksize_bits;
2602 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
2603 start - page_offset(page),
2604 (int)phy_offset, failed_bio->bi_end_io,
2606 bio->bi_opf = REQ_OP_READ | read_mode;
2608 btrfs_debug(btrfs_sb(inode->i_sb),
2609 "Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d",
2610 read_mode, failrec->this_mirror, failrec->in_validation);
2612 status = tree->ops->submit_bio_hook(tree->private_data, bio, failrec->this_mirror,
2613 failrec->bio_flags);
2615 free_io_failure(failure_tree, tree, failrec);
2617 ret = blk_status_to_errno(status);
2623 /* lots and lots of room for performance fixes in the end_bio funcs */
2625 void end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2627 int uptodate = (err == 0);
2630 btrfs_writepage_endio_finish_ordered(page, start, end, uptodate);
2633 ClearPageUptodate(page);
2635 ret = err < 0 ? err : -EIO;
2636 mapping_set_error(page->mapping, ret);
2641 * after a writepage IO is done, we need to:
2642 * clear the uptodate bits on error
2643 * clear the writeback bits in the extent tree for this IO
2644 * end_page_writeback if the page has no more pending IO
2646 * Scheduling is not allowed, so the extent state tree is expected
2647 * to have one and only one object corresponding to this IO.
2649 static void end_bio_extent_writepage(struct bio *bio)
2651 int error = blk_status_to_errno(bio->bi_status);
2652 struct bio_vec *bvec;
2655 struct bvec_iter_all iter_all;
2657 ASSERT(!bio_flagged(bio, BIO_CLONED));
2658 bio_for_each_segment_all(bvec, bio, iter_all) {
2659 struct page *page = bvec->bv_page;
2660 struct inode *inode = page->mapping->host;
2661 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2663 /* We always issue full-page reads, but if some block
2664 * in a page fails to read, blk_update_request() will
2665 * advance bv_offset and adjust bv_len to compensate.
2666 * Print a warning for nonzero offsets, and an error
2667 * if they don't add up to a full page. */
2668 if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) {
2669 if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE)
2671 "partial page write in btrfs with offset %u and length %u",
2672 bvec->bv_offset, bvec->bv_len);
2675 "incomplete page write in btrfs with offset %u and length %u",
2676 bvec->bv_offset, bvec->bv_len);
2679 start = page_offset(page);
2680 end = start + bvec->bv_offset + bvec->bv_len - 1;
2682 end_extent_writepage(page, error, start, end);
2683 end_page_writeback(page);
2690 endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len,
2693 struct extent_state *cached = NULL;
2694 u64 end = start + len - 1;
2696 if (uptodate && tree->track_uptodate)
2697 set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC);
2698 unlock_extent_cached_atomic(tree, start, end, &cached);
2702 * after a readpage IO is done, we need to:
2703 * clear the uptodate bits on error
2704 * set the uptodate bits if things worked
2705 * set the page up to date if all extents in the tree are uptodate
2706 * clear the lock bit in the extent tree
2707 * unlock the page if there are no other extents locked for it
2709 * Scheduling is not allowed, so the extent state tree is expected
2710 * to have one and only one object corresponding to this IO.
2712 static void end_bio_extent_readpage(struct bio *bio)
2714 struct bio_vec *bvec;
2715 int uptodate = !bio->bi_status;
2716 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
2717 struct extent_io_tree *tree, *failure_tree;
2722 u64 extent_start = 0;
2726 struct bvec_iter_all iter_all;
2728 ASSERT(!bio_flagged(bio, BIO_CLONED));
2729 bio_for_each_segment_all(bvec, bio, iter_all) {
2730 struct page *page = bvec->bv_page;
2731 struct inode *inode = page->mapping->host;
2732 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2733 bool data_inode = btrfs_ino(BTRFS_I(inode))
2734 != BTRFS_BTREE_INODE_OBJECTID;
2736 btrfs_debug(fs_info,
2737 "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u",
2738 (u64)bio->bi_iter.bi_sector, bio->bi_status,
2739 io_bio->mirror_num);
2740 tree = &BTRFS_I(inode)->io_tree;
2741 failure_tree = &BTRFS_I(inode)->io_failure_tree;
2743 /* We always issue full-page reads, but if some block
2744 * in a page fails to read, blk_update_request() will
2745 * advance bv_offset and adjust bv_len to compensate.
2746 * Print a warning for nonzero offsets, and an error
2747 * if they don't add up to a full page. */
2748 if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) {
2749 if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE)
2751 "partial page read in btrfs with offset %u and length %u",
2752 bvec->bv_offset, bvec->bv_len);
2755 "incomplete page read in btrfs with offset %u and length %u",
2756 bvec->bv_offset, bvec->bv_len);
2759 start = page_offset(page);
2760 end = start + bvec->bv_offset + bvec->bv_len - 1;
2763 mirror = io_bio->mirror_num;
2764 if (likely(uptodate)) {
2765 ret = tree->ops->readpage_end_io_hook(io_bio, offset,
2771 clean_io_failure(BTRFS_I(inode)->root->fs_info,
2772 failure_tree, tree, start,
2774 btrfs_ino(BTRFS_I(inode)), 0);
2777 if (likely(uptodate))
2783 * The generic bio_readpage_error handles errors the
2784 * following way: If possible, new read requests are
2785 * created and submitted and will end up in
2786 * end_bio_extent_readpage as well (if we're lucky,
2787 * not in the !uptodate case). In that case it returns
2788 * 0 and we just go on with the next page in our bio.
2789 * If it can't handle the error it will return -EIO and
2790 * we remain responsible for that page.
2792 ret = bio_readpage_error(bio, offset, page, start, end,
2795 uptodate = !bio->bi_status;
2800 struct extent_buffer *eb;
2802 eb = (struct extent_buffer *)page->private;
2803 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
2804 eb->read_mirror = mirror;
2805 atomic_dec(&eb->io_pages);
2806 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD,
2808 btree_readahead_hook(eb, -EIO);
2811 if (likely(uptodate)) {
2812 loff_t i_size = i_size_read(inode);
2813 pgoff_t end_index = i_size >> PAGE_SHIFT;
2816 /* Zero out the end if this page straddles i_size */
2817 off = offset_in_page(i_size);
2818 if (page->index == end_index && off)
2819 zero_user_segment(page, off, PAGE_SIZE);
2820 SetPageUptodate(page);
2822 ClearPageUptodate(page);
2828 if (unlikely(!uptodate)) {
2830 endio_readpage_release_extent(tree,
2836 endio_readpage_release_extent(tree, start,
2837 end - start + 1, 0);
2838 } else if (!extent_len) {
2839 extent_start = start;
2840 extent_len = end + 1 - start;
2841 } else if (extent_start + extent_len == start) {
2842 extent_len += end + 1 - start;
2844 endio_readpage_release_extent(tree, extent_start,
2845 extent_len, uptodate);
2846 extent_start = start;
2847 extent_len = end + 1 - start;
2852 endio_readpage_release_extent(tree, extent_start, extent_len,
2854 btrfs_io_bio_free_csum(io_bio);
2859 * Initialize the members up to but not including 'bio'. Use after allocating a
2860 * new bio by bio_alloc_bioset as it does not initialize the bytes outside of
2861 * 'bio' because use of __GFP_ZERO is not supported.
2863 static inline void btrfs_io_bio_init(struct btrfs_io_bio *btrfs_bio)
2865 memset(btrfs_bio, 0, offsetof(struct btrfs_io_bio, bio));
2869 * The following helpers allocate a bio. As it's backed by a bioset, it'll
2870 * never fail. We're returning a bio right now but you can call btrfs_io_bio
2871 * for the appropriate container_of magic
2873 struct bio *btrfs_bio_alloc(u64 first_byte)
2877 bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, &btrfs_bioset);
2878 bio->bi_iter.bi_sector = first_byte >> 9;
2879 btrfs_io_bio_init(btrfs_io_bio(bio));
2883 struct bio *btrfs_bio_clone(struct bio *bio)
2885 struct btrfs_io_bio *btrfs_bio;
2888 /* Bio allocation backed by a bioset does not fail */
2889 new = bio_clone_fast(bio, GFP_NOFS, &btrfs_bioset);
2890 btrfs_bio = btrfs_io_bio(new);
2891 btrfs_io_bio_init(btrfs_bio);
2892 btrfs_bio->iter = bio->bi_iter;
2896 struct bio *btrfs_io_bio_alloc(unsigned int nr_iovecs)
2900 /* Bio allocation backed by a bioset does not fail */
2901 bio = bio_alloc_bioset(GFP_NOFS, nr_iovecs, &btrfs_bioset);
2902 btrfs_io_bio_init(btrfs_io_bio(bio));
2906 struct bio *btrfs_bio_clone_partial(struct bio *orig, int offset, int size)
2909 struct btrfs_io_bio *btrfs_bio;
2911 /* this will never fail when it's backed by a bioset */
2912 bio = bio_clone_fast(orig, GFP_NOFS, &btrfs_bioset);
2915 btrfs_bio = btrfs_io_bio(bio);
2916 btrfs_io_bio_init(btrfs_bio);
2918 bio_trim(bio, offset >> 9, size >> 9);
2919 btrfs_bio->iter = bio->bi_iter;
2924 * @opf: bio REQ_OP_* and REQ_* flags as one value
2925 * @tree: tree so we can call our merge_bio hook
2926 * @wbc: optional writeback control for io accounting
2927 * @page: page to add to the bio
2928 * @pg_offset: offset of the new bio or to check whether we are adding
2929 * a contiguous page to the previous one
2930 * @size: portion of page that we want to write
2931 * @offset: starting offset in the page
2932 * @bio_ret: must be valid pointer, newly allocated bio will be stored there
2933 * @end_io_func: end_io callback for new bio
2934 * @mirror_num: desired mirror to read/write
2935 * @prev_bio_flags: flags of previous bio to see if we can merge the current one
2936 * @bio_flags: flags of the current bio to see if we can merge them
2938 static int submit_extent_page(unsigned int opf, struct extent_io_tree *tree,
2939 struct writeback_control *wbc,
2940 struct page *page, u64 offset,
2941 size_t size, unsigned long pg_offset,
2942 struct bio **bio_ret,
2943 bio_end_io_t end_io_func,
2945 unsigned long prev_bio_flags,
2946 unsigned long bio_flags,
2947 bool force_bio_submit)
2951 size_t page_size = min_t(size_t, size, PAGE_SIZE);
2952 sector_t sector = offset >> 9;
2958 bool can_merge = true;
2961 if (prev_bio_flags & EXTENT_BIO_COMPRESSED)
2962 contig = bio->bi_iter.bi_sector == sector;
2964 contig = bio_end_sector(bio) == sector;
2967 if (btrfs_bio_fits_in_stripe(page, page_size, bio, bio_flags))
2970 if (prev_bio_flags != bio_flags || !contig || !can_merge ||
2972 bio_add_page(bio, page, page_size, pg_offset) < page_size) {
2973 ret = submit_one_bio(bio, mirror_num, prev_bio_flags);
2981 wbc_account_cgroup_owner(wbc, page, page_size);
2986 bio = btrfs_bio_alloc(offset);
2987 bio_add_page(bio, page, page_size, pg_offset);
2988 bio->bi_end_io = end_io_func;
2989 bio->bi_private = tree;
2990 bio->bi_write_hint = page->mapping->host->i_write_hint;
2993 struct block_device *bdev;
2995 bdev = BTRFS_I(page->mapping->host)->root->fs_info->fs_devices->latest_bdev;
2996 bio_set_dev(bio, bdev);
2997 wbc_init_bio(wbc, bio);
2998 wbc_account_cgroup_owner(wbc, page, page_size);
3006 static void attach_extent_buffer_page(struct extent_buffer *eb,
3009 if (!PagePrivate(page)) {
3010 SetPagePrivate(page);
3012 set_page_private(page, (unsigned long)eb);
3014 WARN_ON(page->private != (unsigned long)eb);
3018 void set_page_extent_mapped(struct page *page)
3020 if (!PagePrivate(page)) {
3021 SetPagePrivate(page);
3023 set_page_private(page, EXTENT_PAGE_PRIVATE);
3027 static struct extent_map *
3028 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
3029 u64 start, u64 len, get_extent_t *get_extent,
3030 struct extent_map **em_cached)
3032 struct extent_map *em;
3034 if (em_cached && *em_cached) {
3036 if (extent_map_in_tree(em) && start >= em->start &&
3037 start < extent_map_end(em)) {
3038 refcount_inc(&em->refs);
3042 free_extent_map(em);
3046 em = get_extent(BTRFS_I(inode), page, pg_offset, start, len);
3047 if (em_cached && !IS_ERR_OR_NULL(em)) {
3049 refcount_inc(&em->refs);
3055 * basic readpage implementation. Locked extent state structs are inserted
3056 * into the tree that are removed when the IO is done (by the end_io
3058 * XXX JDM: This needs looking at to ensure proper page locking
3059 * return 0 on success, otherwise return error
3061 static int __do_readpage(struct extent_io_tree *tree,
3063 get_extent_t *get_extent,
3064 struct extent_map **em_cached,
3065 struct bio **bio, int mirror_num,
3066 unsigned long *bio_flags, unsigned int read_flags,
3069 struct inode *inode = page->mapping->host;
3070 u64 start = page_offset(page);
3071 const u64 end = start + PAGE_SIZE - 1;
3074 u64 last_byte = i_size_read(inode);
3077 struct extent_map *em;
3080 size_t pg_offset = 0;
3082 size_t disk_io_size;
3083 size_t blocksize = inode->i_sb->s_blocksize;
3084 unsigned long this_bio_flag = 0;
3086 set_page_extent_mapped(page);
3088 if (!PageUptodate(page)) {
3089 if (cleancache_get_page(page) == 0) {
3090 BUG_ON(blocksize != PAGE_SIZE);
3091 unlock_extent(tree, start, end);
3096 if (page->index == last_byte >> PAGE_SHIFT) {
3098 size_t zero_offset = offset_in_page(last_byte);
3101 iosize = PAGE_SIZE - zero_offset;
3102 userpage = kmap_atomic(page);
3103 memset(userpage + zero_offset, 0, iosize);
3104 flush_dcache_page(page);
3105 kunmap_atomic(userpage);
3108 while (cur <= end) {
3109 bool force_bio_submit = false;
3112 if (cur >= last_byte) {
3114 struct extent_state *cached = NULL;
3116 iosize = PAGE_SIZE - pg_offset;
3117 userpage = kmap_atomic(page);
3118 memset(userpage + pg_offset, 0, iosize);
3119 flush_dcache_page(page);
3120 kunmap_atomic(userpage);
3121 set_extent_uptodate(tree, cur, cur + iosize - 1,
3123 unlock_extent_cached(tree, cur,
3124 cur + iosize - 1, &cached);
3127 em = __get_extent_map(inode, page, pg_offset, cur,
3128 end - cur + 1, get_extent, em_cached);
3129 if (IS_ERR_OR_NULL(em)) {
3131 unlock_extent(tree, cur, end);
3134 extent_offset = cur - em->start;
3135 BUG_ON(extent_map_end(em) <= cur);
3138 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
3139 this_bio_flag |= EXTENT_BIO_COMPRESSED;
3140 extent_set_compress_type(&this_bio_flag,
3144 iosize = min(extent_map_end(em) - cur, end - cur + 1);
3145 cur_end = min(extent_map_end(em) - 1, end);
3146 iosize = ALIGN(iosize, blocksize);
3147 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
3148 disk_io_size = em->block_len;
3149 offset = em->block_start;
3151 offset = em->block_start + extent_offset;
3152 disk_io_size = iosize;
3154 block_start = em->block_start;
3155 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
3156 block_start = EXTENT_MAP_HOLE;
3159 * If we have a file range that points to a compressed extent
3160 * and it's followed by a consecutive file range that points to
3161 * to the same compressed extent (possibly with a different
3162 * offset and/or length, so it either points to the whole extent
3163 * or only part of it), we must make sure we do not submit a
3164 * single bio to populate the pages for the 2 ranges because
3165 * this makes the compressed extent read zero out the pages
3166 * belonging to the 2nd range. Imagine the following scenario:
3169 * [0 - 8K] [8K - 24K]
3172 * points to extent X, points to extent X,
3173 * offset 4K, length of 8K offset 0, length 16K
3175 * [extent X, compressed length = 4K uncompressed length = 16K]
3177 * If the bio to read the compressed extent covers both ranges,
3178 * it will decompress extent X into the pages belonging to the
3179 * first range and then it will stop, zeroing out the remaining
3180 * pages that belong to the other range that points to extent X.
3181 * So here we make sure we submit 2 bios, one for the first
3182 * range and another one for the third range. Both will target
3183 * the same physical extent from disk, but we can't currently
3184 * make the compressed bio endio callback populate the pages
3185 * for both ranges because each compressed bio is tightly
3186 * coupled with a single extent map, and each range can have
3187 * an extent map with a different offset value relative to the
3188 * uncompressed data of our extent and different lengths. This
3189 * is a corner case so we prioritize correctness over
3190 * non-optimal behavior (submitting 2 bios for the same extent).
3192 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) &&
3193 prev_em_start && *prev_em_start != (u64)-1 &&
3194 *prev_em_start != em->start)
3195 force_bio_submit = true;
3198 *prev_em_start = em->start;
3200 free_extent_map(em);
3203 /* we've found a hole, just zero and go on */
3204 if (block_start == EXTENT_MAP_HOLE) {
3206 struct extent_state *cached = NULL;
3208 userpage = kmap_atomic(page);
3209 memset(userpage + pg_offset, 0, iosize);
3210 flush_dcache_page(page);
3211 kunmap_atomic(userpage);
3213 set_extent_uptodate(tree, cur, cur + iosize - 1,
3215 unlock_extent_cached(tree, cur,
3216 cur + iosize - 1, &cached);
3218 pg_offset += iosize;
3221 /* the get_extent function already copied into the page */
3222 if (test_range_bit(tree, cur, cur_end,
3223 EXTENT_UPTODATE, 1, NULL)) {
3224 check_page_uptodate(tree, page);
3225 unlock_extent(tree, cur, cur + iosize - 1);
3227 pg_offset += iosize;
3230 /* we have an inline extent but it didn't get marked up
3231 * to date. Error out
3233 if (block_start == EXTENT_MAP_INLINE) {
3235 unlock_extent(tree, cur, cur + iosize - 1);
3237 pg_offset += iosize;
3241 ret = submit_extent_page(REQ_OP_READ | read_flags, tree, NULL,
3242 page, offset, disk_io_size,
3244 end_bio_extent_readpage, mirror_num,
3250 *bio_flags = this_bio_flag;
3253 unlock_extent(tree, cur, cur + iosize - 1);
3257 pg_offset += iosize;
3261 if (!PageError(page))
3262 SetPageUptodate(page);
3268 static inline void contiguous_readpages(struct extent_io_tree *tree,
3269 struct page *pages[], int nr_pages,
3271 struct extent_map **em_cached,
3273 unsigned long *bio_flags,
3276 struct btrfs_inode *inode = BTRFS_I(pages[0]->mapping->host);
3279 btrfs_lock_and_flush_ordered_range(tree, inode, start, end, NULL);
3281 for (index = 0; index < nr_pages; index++) {
3282 __do_readpage(tree, pages[index], btrfs_get_extent, em_cached,
3283 bio, 0, bio_flags, REQ_RAHEAD, prev_em_start);
3284 put_page(pages[index]);
3288 static int __extent_read_full_page(struct extent_io_tree *tree,
3290 get_extent_t *get_extent,
3291 struct bio **bio, int mirror_num,
3292 unsigned long *bio_flags,
3293 unsigned int read_flags)
3295 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
3296 u64 start = page_offset(page);
3297 u64 end = start + PAGE_SIZE - 1;
3300 btrfs_lock_and_flush_ordered_range(tree, inode, start, end, NULL);
3302 ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num,
3303 bio_flags, read_flags, NULL);
3307 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
3308 get_extent_t *get_extent, int mirror_num)
3310 struct bio *bio = NULL;
3311 unsigned long bio_flags = 0;
3314 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
3317 ret = submit_one_bio(bio, mirror_num, bio_flags);
3321 static void update_nr_written(struct writeback_control *wbc,
3322 unsigned long nr_written)
3324 wbc->nr_to_write -= nr_written;
3328 * helper for __extent_writepage, doing all of the delayed allocation setup.
3330 * This returns 1 if btrfs_run_delalloc_range function did all the work required
3331 * to write the page (copy into inline extent). In this case the IO has
3332 * been started and the page is already unlocked.
3334 * This returns 0 if all went well (page still locked)
3335 * This returns < 0 if there were errors (page still locked)
3337 static noinline_for_stack int writepage_delalloc(struct inode *inode,
3338 struct page *page, struct writeback_control *wbc,
3339 u64 delalloc_start, unsigned long *nr_written)
3341 u64 page_end = delalloc_start + PAGE_SIZE - 1;
3343 u64 delalloc_to_write = 0;
3344 u64 delalloc_end = 0;
3346 int page_started = 0;
3349 while (delalloc_end < page_end) {
3350 found = find_lock_delalloc_range(inode, page,
3354 delalloc_start = delalloc_end + 1;
3357 ret = btrfs_run_delalloc_range(inode, page, delalloc_start,
3358 delalloc_end, &page_started, nr_written, wbc);
3362 * btrfs_run_delalloc_range should return < 0 for error
3363 * but just in case, we use > 0 here meaning the IO is
3364 * started, so we don't want to return > 0 unless
3365 * things are going well.
3367 ret = ret < 0 ? ret : -EIO;
3371 * delalloc_end is already one less than the total length, so
3372 * we don't subtract one from PAGE_SIZE
3374 delalloc_to_write += (delalloc_end - delalloc_start +
3375 PAGE_SIZE) >> PAGE_SHIFT;
3376 delalloc_start = delalloc_end + 1;
3378 if (wbc->nr_to_write < delalloc_to_write) {
3381 if (delalloc_to_write < thresh * 2)
3382 thresh = delalloc_to_write;
3383 wbc->nr_to_write = min_t(u64, delalloc_to_write,
3387 /* did the fill delalloc function already unlock and start
3392 * we've unlocked the page, so we can't update
3393 * the mapping's writeback index, just update
3396 wbc->nr_to_write -= *nr_written;
3407 * helper for __extent_writepage. This calls the writepage start hooks,
3408 * and does the loop to map the page into extents and bios.
3410 * We return 1 if the IO is started and the page is unlocked,
3411 * 0 if all went well (page still locked)
3412 * < 0 if there were errors (page still locked)
3414 static noinline_for_stack int __extent_writepage_io(struct inode *inode,
3416 struct writeback_control *wbc,
3417 struct extent_page_data *epd,
3419 unsigned long nr_written,
3422 struct extent_io_tree *tree = epd->tree;
3423 u64 start = page_offset(page);
3424 u64 page_end = start + PAGE_SIZE - 1;
3430 struct extent_map *em;
3431 size_t pg_offset = 0;
3435 const unsigned int write_flags = wbc_to_write_flags(wbc);
3438 ret = btrfs_writepage_cow_fixup(page, start, page_end);
3440 /* Fixup worker will requeue */
3442 wbc->pages_skipped++;
3444 redirty_page_for_writepage(wbc, page);
3446 update_nr_written(wbc, nr_written);
3452 * we don't want to touch the inode after unlocking the page,
3453 * so we update the mapping writeback index now
3455 update_nr_written(wbc, nr_written + 1);
3458 blocksize = inode->i_sb->s_blocksize;
3460 while (cur <= end) {
3464 if (cur >= i_size) {
3465 btrfs_writepage_endio_finish_ordered(page, cur,
3469 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur,
3471 if (IS_ERR_OR_NULL(em)) {
3473 ret = PTR_ERR_OR_ZERO(em);
3477 extent_offset = cur - em->start;
3478 em_end = extent_map_end(em);
3479 BUG_ON(em_end <= cur);
3481 iosize = min(em_end - cur, end - cur + 1);
3482 iosize = ALIGN(iosize, blocksize);
3483 offset = em->block_start + extent_offset;
3484 block_start = em->block_start;
3485 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3486 free_extent_map(em);
3490 * compressed and inline extents are written through other
3493 if (compressed || block_start == EXTENT_MAP_HOLE ||
3494 block_start == EXTENT_MAP_INLINE) {
3496 * end_io notification does not happen here for
3497 * compressed extents
3500 btrfs_writepage_endio_finish_ordered(page, cur,
3503 else if (compressed) {
3504 /* we don't want to end_page_writeback on
3505 * a compressed extent. this happens
3512 pg_offset += iosize;
3516 btrfs_set_range_writeback(tree, cur, cur + iosize - 1);
3517 if (!PageWriteback(page)) {
3518 btrfs_err(BTRFS_I(inode)->root->fs_info,
3519 "page %lu not writeback, cur %llu end %llu",
3520 page->index, cur, end);
3523 ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc,
3524 page, offset, iosize, pg_offset,
3526 end_bio_extent_writepage,
3530 if (PageWriteback(page))
3531 end_page_writeback(page);
3535 pg_offset += iosize;
3543 * the writepage semantics are similar to regular writepage. extent
3544 * records are inserted to lock ranges in the tree, and as dirty areas
3545 * are found, they are marked writeback. Then the lock bits are removed
3546 * and the end_io handler clears the writeback ranges
3548 * Return 0 if everything goes well.
3549 * Return <0 for error.
3551 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
3552 struct extent_page_data *epd)
3554 struct inode *inode = page->mapping->host;
3555 u64 start = page_offset(page);
3556 u64 page_end = start + PAGE_SIZE - 1;
3560 loff_t i_size = i_size_read(inode);
3561 unsigned long end_index = i_size >> PAGE_SHIFT;
3562 unsigned long nr_written = 0;
3564 trace___extent_writepage(page, inode, wbc);
3566 WARN_ON(!PageLocked(page));
3568 ClearPageError(page);
3570 pg_offset = offset_in_page(i_size);
3571 if (page->index > end_index ||
3572 (page->index == end_index && !pg_offset)) {
3573 page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
3578 if (page->index == end_index) {
3581 userpage = kmap_atomic(page);
3582 memset(userpage + pg_offset, 0,
3583 PAGE_SIZE - pg_offset);
3584 kunmap_atomic(userpage);
3585 flush_dcache_page(page);
3588 set_page_extent_mapped(page);
3590 if (!epd->extent_locked) {
3591 ret = writepage_delalloc(inode, page, wbc, start, &nr_written);
3598 ret = __extent_writepage_io(inode, page, wbc, epd,
3599 i_size, nr_written, &nr);
3605 /* make sure the mapping tag for page dirty gets cleared */
3606 set_page_writeback(page);
3607 end_page_writeback(page);
3609 if (PageError(page)) {
3610 ret = ret < 0 ? ret : -EIO;
3611 end_extent_writepage(page, ret, start, page_end);
3618 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3620 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
3621 TASK_UNINTERRUPTIBLE);
3624 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3626 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3627 smp_mb__after_atomic();
3628 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3632 * Lock eb pages and flush the bio if we can't the locks
3634 * Return 0 if nothing went wrong
3635 * Return >0 is same as 0, except bio is not submitted
3636 * Return <0 if something went wrong, no page is locked
3638 static noinline_for_stack int lock_extent_buffer_for_io(struct extent_buffer *eb,
3639 struct extent_page_data *epd)
3641 struct btrfs_fs_info *fs_info = eb->fs_info;
3642 int i, num_pages, failed_page_nr;
3646 if (!btrfs_try_tree_write_lock(eb)) {
3647 ret = flush_write_bio(epd);
3651 btrfs_tree_lock(eb);
3654 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3655 btrfs_tree_unlock(eb);
3659 ret = flush_write_bio(epd);
3665 wait_on_extent_buffer_writeback(eb);
3666 btrfs_tree_lock(eb);
3667 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3669 btrfs_tree_unlock(eb);
3674 * We need to do this to prevent races in people who check if the eb is
3675 * under IO since we can end up having no IO bits set for a short period
3678 spin_lock(&eb->refs_lock);
3679 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3680 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3681 spin_unlock(&eb->refs_lock);
3682 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3683 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
3685 fs_info->dirty_metadata_batch);
3688 spin_unlock(&eb->refs_lock);
3691 btrfs_tree_unlock(eb);
3696 num_pages = num_extent_pages(eb);
3697 for (i = 0; i < num_pages; i++) {
3698 struct page *p = eb->pages[i];
3700 if (!trylock_page(p)) {
3704 err = flush_write_bio(epd);
3718 /* Unlock already locked pages */
3719 for (i = 0; i < failed_page_nr; i++)
3720 unlock_page(eb->pages[i]);
3722 * Clear EXTENT_BUFFER_WRITEBACK and wake up anyone waiting on it.
3723 * Also set back EXTENT_BUFFER_DIRTY so future attempts to this eb can
3724 * be made and undo everything done before.
3726 btrfs_tree_lock(eb);
3727 spin_lock(&eb->refs_lock);
3728 set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
3729 end_extent_buffer_writeback(eb);
3730 spin_unlock(&eb->refs_lock);
3731 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, eb->len,
3732 fs_info->dirty_metadata_batch);
3733 btrfs_clear_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3734 btrfs_tree_unlock(eb);
3738 static void set_btree_ioerr(struct page *page)
3740 struct extent_buffer *eb = (struct extent_buffer *)page->private;
3741 struct btrfs_fs_info *fs_info;
3744 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))
3748 * If we error out, we should add back the dirty_metadata_bytes
3749 * to make it consistent.
3751 fs_info = eb->fs_info;
3752 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
3753 eb->len, fs_info->dirty_metadata_batch);
3756 * If writeback for a btree extent that doesn't belong to a log tree
3757 * failed, increment the counter transaction->eb_write_errors.
3758 * We do this because while the transaction is running and before it's
3759 * committing (when we call filemap_fdata[write|wait]_range against
3760 * the btree inode), we might have
3761 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
3762 * returns an error or an error happens during writeback, when we're
3763 * committing the transaction we wouldn't know about it, since the pages
3764 * can be no longer dirty nor marked anymore for writeback (if a
3765 * subsequent modification to the extent buffer didn't happen before the
3766 * transaction commit), which makes filemap_fdata[write|wait]_range not
3767 * able to find the pages tagged with SetPageError at transaction
3768 * commit time. So if this happens we must abort the transaction,
3769 * otherwise we commit a super block with btree roots that point to
3770 * btree nodes/leafs whose content on disk is invalid - either garbage
3771 * or the content of some node/leaf from a past generation that got
3772 * cowed or deleted and is no longer valid.
3774 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
3775 * not be enough - we need to distinguish between log tree extents vs
3776 * non-log tree extents, and the next filemap_fdatawait_range() call
3777 * will catch and clear such errors in the mapping - and that call might
3778 * be from a log sync and not from a transaction commit. Also, checking
3779 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
3780 * not done and would not be reliable - the eb might have been released
3781 * from memory and reading it back again means that flag would not be
3782 * set (since it's a runtime flag, not persisted on disk).
3784 * Using the flags below in the btree inode also makes us achieve the
3785 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
3786 * writeback for all dirty pages and before filemap_fdatawait_range()
3787 * is called, the writeback for all dirty pages had already finished
3788 * with errors - because we were not using AS_EIO/AS_ENOSPC,
3789 * filemap_fdatawait_range() would return success, as it could not know
3790 * that writeback errors happened (the pages were no longer tagged for
3793 switch (eb->log_index) {
3795 set_bit(BTRFS_FS_BTREE_ERR, &eb->fs_info->flags);
3798 set_bit(BTRFS_FS_LOG1_ERR, &eb->fs_info->flags);
3801 set_bit(BTRFS_FS_LOG2_ERR, &eb->fs_info->flags);
3804 BUG(); /* unexpected, logic error */
3808 static void end_bio_extent_buffer_writepage(struct bio *bio)
3810 struct bio_vec *bvec;
3811 struct extent_buffer *eb;
3813 struct bvec_iter_all iter_all;
3815 ASSERT(!bio_flagged(bio, BIO_CLONED));
3816 bio_for_each_segment_all(bvec, bio, iter_all) {
3817 struct page *page = bvec->bv_page;
3819 eb = (struct extent_buffer *)page->private;
3821 done = atomic_dec_and_test(&eb->io_pages);
3823 if (bio->bi_status ||
3824 test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) {
3825 ClearPageUptodate(page);
3826 set_btree_ioerr(page);
3829 end_page_writeback(page);
3834 end_extent_buffer_writeback(eb);
3840 static noinline_for_stack int write_one_eb(struct extent_buffer *eb,
3841 struct writeback_control *wbc,
3842 struct extent_page_data *epd)
3844 struct btrfs_fs_info *fs_info = eb->fs_info;
3845 struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
3846 u64 offset = eb->start;
3849 unsigned long start, end;
3850 unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META;
3853 clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
3854 num_pages = num_extent_pages(eb);
3855 atomic_set(&eb->io_pages, num_pages);
3857 /* set btree blocks beyond nritems with 0 to avoid stale content. */
3858 nritems = btrfs_header_nritems(eb);
3859 if (btrfs_header_level(eb) > 0) {
3860 end = btrfs_node_key_ptr_offset(nritems);
3862 memzero_extent_buffer(eb, end, eb->len - end);
3866 * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0
3868 start = btrfs_item_nr_offset(nritems);
3869 end = BTRFS_LEAF_DATA_OFFSET + leaf_data_end(eb);
3870 memzero_extent_buffer(eb, start, end - start);
3873 for (i = 0; i < num_pages; i++) {
3874 struct page *p = eb->pages[i];
3876 clear_page_dirty_for_io(p);
3877 set_page_writeback(p);
3878 ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc,
3879 p, offset, PAGE_SIZE, 0,
3881 end_bio_extent_buffer_writepage,
3885 if (PageWriteback(p))
3886 end_page_writeback(p);
3887 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3888 end_extent_buffer_writeback(eb);
3892 offset += PAGE_SIZE;
3893 update_nr_written(wbc, 1);
3897 if (unlikely(ret)) {
3898 for (; i < num_pages; i++) {
3899 struct page *p = eb->pages[i];
3900 clear_page_dirty_for_io(p);
3908 int btree_write_cache_pages(struct address_space *mapping,
3909 struct writeback_control *wbc)
3911 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
3912 struct extent_buffer *eb, *prev_eb = NULL;
3913 struct extent_page_data epd = {
3917 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3921 int nr_to_write_done = 0;
3922 struct pagevec pvec;
3925 pgoff_t end; /* Inclusive */
3929 pagevec_init(&pvec);
3930 if (wbc->range_cyclic) {
3931 index = mapping->writeback_index; /* Start from prev offset */
3934 index = wbc->range_start >> PAGE_SHIFT;
3935 end = wbc->range_end >> PAGE_SHIFT;
3938 if (wbc->sync_mode == WB_SYNC_ALL)
3939 tag = PAGECACHE_TAG_TOWRITE;
3941 tag = PAGECACHE_TAG_DIRTY;
3943 if (wbc->sync_mode == WB_SYNC_ALL)
3944 tag_pages_for_writeback(mapping, index, end);
3945 while (!done && !nr_to_write_done && (index <= end) &&
3946 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
3951 for (i = 0; i < nr_pages; i++) {
3952 struct page *page = pvec.pages[i];
3954 if (!PagePrivate(page))
3957 spin_lock(&mapping->private_lock);
3958 if (!PagePrivate(page)) {
3959 spin_unlock(&mapping->private_lock);
3963 eb = (struct extent_buffer *)page->private;
3966 * Shouldn't happen and normally this would be a BUG_ON
3967 * but no sense in crashing the users box for something
3968 * we can survive anyway.
3971 spin_unlock(&mapping->private_lock);
3975 if (eb == prev_eb) {
3976 spin_unlock(&mapping->private_lock);
3980 ret = atomic_inc_not_zero(&eb->refs);
3981 spin_unlock(&mapping->private_lock);
3986 ret = lock_extent_buffer_for_io(eb, &epd);
3988 free_extent_buffer(eb);
3990 } else if (ret < 0) {
3992 free_extent_buffer(eb);
3996 ret = write_one_eb(eb, wbc, &epd);
3999 free_extent_buffer(eb);
4002 free_extent_buffer(eb);
4005 * the filesystem may choose to bump up nr_to_write.
4006 * We have to make sure to honor the new nr_to_write
4009 nr_to_write_done = wbc->nr_to_write <= 0;
4011 pagevec_release(&pvec);
4014 if (!scanned && !done) {
4016 * We hit the last page and there is more work to be done: wrap
4017 * back to the start of the file
4025 end_write_bio(&epd, ret);
4028 ret = flush_write_bio(&epd);
4033 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
4034 * @mapping: address space structure to write
4035 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
4036 * @data: data passed to __extent_writepage function
4038 * If a page is already under I/O, write_cache_pages() skips it, even
4039 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
4040 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
4041 * and msync() need to guarantee that all the data which was dirty at the time
4042 * the call was made get new I/O started against them. If wbc->sync_mode is
4043 * WB_SYNC_ALL then we were called for data integrity and we must wait for
4044 * existing IO to complete.
4046 static int extent_write_cache_pages(struct address_space *mapping,
4047 struct writeback_control *wbc,
4048 struct extent_page_data *epd)
4050 struct inode *inode = mapping->host;
4053 int nr_to_write_done = 0;
4054 struct pagevec pvec;
4057 pgoff_t end; /* Inclusive */
4059 int range_whole = 0;
4064 * We have to hold onto the inode so that ordered extents can do their
4065 * work when the IO finishes. The alternative to this is failing to add
4066 * an ordered extent if the igrab() fails there and that is a huge pain
4067 * to deal with, so instead just hold onto the inode throughout the
4068 * writepages operation. If it fails here we are freeing up the inode
4069 * anyway and we'd rather not waste our time writing out stuff that is
4070 * going to be truncated anyway.
4075 pagevec_init(&pvec);
4076 if (wbc->range_cyclic) {
4077 index = mapping->writeback_index; /* Start from prev offset */
4080 index = wbc->range_start >> PAGE_SHIFT;
4081 end = wbc->range_end >> PAGE_SHIFT;
4082 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
4088 * We do the tagged writepage as long as the snapshot flush bit is set
4089 * and we are the first one who do the filemap_flush() on this inode.
4091 * The nr_to_write == LONG_MAX is needed to make sure other flushers do
4092 * not race in and drop the bit.
4094 if (range_whole && wbc->nr_to_write == LONG_MAX &&
4095 test_and_clear_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
4096 &BTRFS_I(inode)->runtime_flags))
4097 wbc->tagged_writepages = 1;
4099 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
4100 tag = PAGECACHE_TAG_TOWRITE;
4102 tag = PAGECACHE_TAG_DIRTY;
4104 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
4105 tag_pages_for_writeback(mapping, index, end);
4107 while (!done && !nr_to_write_done && (index <= end) &&
4108 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping,
4109 &index, end, tag))) {
4113 for (i = 0; i < nr_pages; i++) {
4114 struct page *page = pvec.pages[i];
4116 done_index = page->index + 1;
4118 * At this point we hold neither the i_pages lock nor
4119 * the page lock: the page may be truncated or
4120 * invalidated (changing page->mapping to NULL),
4121 * or even swizzled back from swapper_space to
4122 * tmpfs file mapping
4124 if (!trylock_page(page)) {
4125 ret = flush_write_bio(epd);
4130 if (unlikely(page->mapping != mapping)) {
4135 if (wbc->sync_mode != WB_SYNC_NONE) {
4136 if (PageWriteback(page)) {
4137 ret = flush_write_bio(epd);
4140 wait_on_page_writeback(page);
4143 if (PageWriteback(page) ||
4144 !clear_page_dirty_for_io(page)) {
4149 ret = __extent_writepage(page, wbc, epd);
4156 * the filesystem may choose to bump up nr_to_write.
4157 * We have to make sure to honor the new nr_to_write
4160 nr_to_write_done = wbc->nr_to_write <= 0;
4162 pagevec_release(&pvec);
4165 if (!scanned && !done) {
4167 * We hit the last page and there is more work to be done: wrap
4168 * back to the start of the file
4175 if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole))
4176 mapping->writeback_index = done_index;
4178 btrfs_add_delayed_iput(inode);
4182 int extent_write_full_page(struct page *page, struct writeback_control *wbc)
4185 struct extent_page_data epd = {
4187 .tree = &BTRFS_I(page->mapping->host)->io_tree,
4189 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4192 ret = __extent_writepage(page, wbc, &epd);
4195 end_write_bio(&epd, ret);
4199 ret = flush_write_bio(&epd);
4204 int extent_write_locked_range(struct inode *inode, u64 start, u64 end,
4208 struct address_space *mapping = inode->i_mapping;
4209 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
4211 unsigned long nr_pages = (end - start + PAGE_SIZE) >>
4214 struct extent_page_data epd = {
4218 .sync_io = mode == WB_SYNC_ALL,
4220 struct writeback_control wbc_writepages = {
4222 .nr_to_write = nr_pages * 2,
4223 .range_start = start,
4224 .range_end = end + 1,
4225 /* We're called from an async helper function */
4226 .punt_to_cgroup = 1,
4227 .no_cgroup_owner = 1,
4230 wbc_attach_fdatawrite_inode(&wbc_writepages, inode);
4231 while (start <= end) {
4232 page = find_get_page(mapping, start >> PAGE_SHIFT);
4233 if (clear_page_dirty_for_io(page))
4234 ret = __extent_writepage(page, &wbc_writepages, &epd);
4236 btrfs_writepage_endio_finish_ordered(page, start,
4237 start + PAGE_SIZE - 1, 1);
4246 ret = flush_write_bio(&epd);
4248 end_write_bio(&epd, ret);
4250 wbc_detach_inode(&wbc_writepages);
4254 int extent_writepages(struct address_space *mapping,
4255 struct writeback_control *wbc)
4258 struct extent_page_data epd = {
4260 .tree = &BTRFS_I(mapping->host)->io_tree,
4262 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4265 ret = extent_write_cache_pages(mapping, wbc, &epd);
4268 end_write_bio(&epd, ret);
4271 ret = flush_write_bio(&epd);
4275 int extent_readpages(struct address_space *mapping, struct list_head *pages,
4278 struct bio *bio = NULL;
4279 unsigned long bio_flags = 0;
4280 struct page *pagepool[16];
4281 struct extent_map *em_cached = NULL;
4282 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
4284 u64 prev_em_start = (u64)-1;
4286 while (!list_empty(pages)) {
4289 for (nr = 0; nr < ARRAY_SIZE(pagepool) && !list_empty(pages);) {
4290 struct page *page = lru_to_page(pages);
4292 prefetchw(&page->flags);
4293 list_del(&page->lru);
4294 if (add_to_page_cache_lru(page, mapping, page->index,
4295 readahead_gfp_mask(mapping))) {
4300 pagepool[nr++] = page;
4301 contig_end = page_offset(page) + PAGE_SIZE - 1;
4305 u64 contig_start = page_offset(pagepool[0]);
4307 ASSERT(contig_start + nr * PAGE_SIZE - 1 == contig_end);
4309 contiguous_readpages(tree, pagepool, nr, contig_start,
4310 contig_end, &em_cached, &bio, &bio_flags,
4316 free_extent_map(em_cached);
4319 return submit_one_bio(bio, 0, bio_flags);
4324 * basic invalidatepage code, this waits on any locked or writeback
4325 * ranges corresponding to the page, and then deletes any extent state
4326 * records from the tree
4328 int extent_invalidatepage(struct extent_io_tree *tree,
4329 struct page *page, unsigned long offset)
4331 struct extent_state *cached_state = NULL;
4332 u64 start = page_offset(page);
4333 u64 end = start + PAGE_SIZE - 1;
4334 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
4336 start += ALIGN(offset, blocksize);
4340 lock_extent_bits(tree, start, end, &cached_state);
4341 wait_on_page_writeback(page);
4342 clear_extent_bit(tree, start, end, EXTENT_LOCKED | EXTENT_DELALLOC |
4343 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state);
4348 * a helper for releasepage, this tests for areas of the page that
4349 * are locked or under IO and drops the related state bits if it is safe
4352 static int try_release_extent_state(struct extent_io_tree *tree,
4353 struct page *page, gfp_t mask)
4355 u64 start = page_offset(page);
4356 u64 end = start + PAGE_SIZE - 1;
4359 if (test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL)) {
4363 * at this point we can safely clear everything except the
4364 * locked bit and the nodatasum bit
4366 ret = __clear_extent_bit(tree, start, end,
4367 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
4368 0, 0, NULL, mask, NULL);
4370 /* if clear_extent_bit failed for enomem reasons,
4371 * we can't allow the release to continue.
4382 * a helper for releasepage. As long as there are no locked extents
4383 * in the range corresponding to the page, both state records and extent
4384 * map records are removed
4386 int try_release_extent_mapping(struct page *page, gfp_t mask)
4388 struct extent_map *em;
4389 u64 start = page_offset(page);
4390 u64 end = start + PAGE_SIZE - 1;
4391 struct btrfs_inode *btrfs_inode = BTRFS_I(page->mapping->host);
4392 struct extent_io_tree *tree = &btrfs_inode->io_tree;
4393 struct extent_map_tree *map = &btrfs_inode->extent_tree;
4395 if (gfpflags_allow_blocking(mask) &&
4396 page->mapping->host->i_size > SZ_16M) {
4398 while (start <= end) {
4399 len = end - start + 1;
4400 write_lock(&map->lock);
4401 em = lookup_extent_mapping(map, start, len);
4403 write_unlock(&map->lock);
4406 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
4407 em->start != start) {
4408 write_unlock(&map->lock);
4409 free_extent_map(em);
4412 if (!test_range_bit(tree, em->start,
4413 extent_map_end(em) - 1,
4414 EXTENT_LOCKED, 0, NULL)) {
4415 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4416 &btrfs_inode->runtime_flags);
4417 remove_extent_mapping(map, em);
4418 /* once for the rb tree */
4419 free_extent_map(em);
4421 start = extent_map_end(em);
4422 write_unlock(&map->lock);
4425 free_extent_map(em);
4428 return try_release_extent_state(tree, page, mask);
4432 * helper function for fiemap, which doesn't want to see any holes.
4433 * This maps until we find something past 'last'
4435 static struct extent_map *get_extent_skip_holes(struct inode *inode,
4436 u64 offset, u64 last)
4438 u64 sectorsize = btrfs_inode_sectorsize(inode);
4439 struct extent_map *em;
4446 len = last - offset;
4449 len = ALIGN(len, sectorsize);
4450 em = btrfs_get_extent_fiemap(BTRFS_I(inode), offset, len);
4451 if (IS_ERR_OR_NULL(em))
4454 /* if this isn't a hole return it */
4455 if (em->block_start != EXTENT_MAP_HOLE)
4458 /* this is a hole, advance to the next extent */
4459 offset = extent_map_end(em);
4460 free_extent_map(em);
4468 * To cache previous fiemap extent
4470 * Will be used for merging fiemap extent
4472 struct fiemap_cache {
4481 * Helper to submit fiemap extent.
4483 * Will try to merge current fiemap extent specified by @offset, @phys,
4484 * @len and @flags with cached one.
4485 * And only when we fails to merge, cached one will be submitted as
4488 * Return value is the same as fiemap_fill_next_extent().
4490 static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
4491 struct fiemap_cache *cache,
4492 u64 offset, u64 phys, u64 len, u32 flags)
4500 * Sanity check, extent_fiemap() should have ensured that new
4501 * fiemap extent won't overlap with cached one.
4504 * NOTE: Physical address can overlap, due to compression
4506 if (cache->offset + cache->len > offset) {
4512 * Only merges fiemap extents if
4513 * 1) Their logical addresses are continuous
4515 * 2) Their physical addresses are continuous
4516 * So truly compressed (physical size smaller than logical size)
4517 * extents won't get merged with each other
4519 * 3) Share same flags except FIEMAP_EXTENT_LAST
4520 * So regular extent won't get merged with prealloc extent
4522 if (cache->offset + cache->len == offset &&
4523 cache->phys + cache->len == phys &&
4524 (cache->flags & ~FIEMAP_EXTENT_LAST) ==
4525 (flags & ~FIEMAP_EXTENT_LAST)) {
4527 cache->flags |= flags;
4528 goto try_submit_last;
4531 /* Not mergeable, need to submit cached one */
4532 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
4533 cache->len, cache->flags);
4534 cache->cached = false;
4538 cache->cached = true;
4539 cache->offset = offset;
4542 cache->flags = flags;
4544 if (cache->flags & FIEMAP_EXTENT_LAST) {
4545 ret = fiemap_fill_next_extent(fieinfo, cache->offset,
4546 cache->phys, cache->len, cache->flags);
4547 cache->cached = false;
4553 * Emit last fiemap cache
4555 * The last fiemap cache may still be cached in the following case:
4557 * |<- Fiemap range ->|
4558 * |<------------ First extent ----------->|
4560 * In this case, the first extent range will be cached but not emitted.
4561 * So we must emit it before ending extent_fiemap().
4563 static int emit_last_fiemap_cache(struct fiemap_extent_info *fieinfo,
4564 struct fiemap_cache *cache)
4571 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
4572 cache->len, cache->flags);
4573 cache->cached = false;
4579 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4580 __u64 start, __u64 len)
4584 u64 max = start + len;
4588 u64 last_for_get_extent = 0;
4590 u64 isize = i_size_read(inode);
4591 struct btrfs_key found_key;
4592 struct extent_map *em = NULL;
4593 struct extent_state *cached_state = NULL;
4594 struct btrfs_path *path;
4595 struct btrfs_root *root = BTRFS_I(inode)->root;
4596 struct fiemap_cache cache = { 0 };
4597 struct ulist *roots;
4598 struct ulist *tmp_ulist;
4607 path = btrfs_alloc_path();
4610 path->leave_spinning = 1;
4612 roots = ulist_alloc(GFP_KERNEL);
4613 tmp_ulist = ulist_alloc(GFP_KERNEL);
4614 if (!roots || !tmp_ulist) {
4616 goto out_free_ulist;
4619 start = round_down(start, btrfs_inode_sectorsize(inode));
4620 len = round_up(max, btrfs_inode_sectorsize(inode)) - start;
4623 * lookup the last file extent. We're not using i_size here
4624 * because there might be preallocation past i_size
4626 ret = btrfs_lookup_file_extent(NULL, root, path,
4627 btrfs_ino(BTRFS_I(inode)), -1, 0);
4629 goto out_free_ulist;
4637 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
4638 found_type = found_key.type;
4640 /* No extents, but there might be delalloc bits */
4641 if (found_key.objectid != btrfs_ino(BTRFS_I(inode)) ||
4642 found_type != BTRFS_EXTENT_DATA_KEY) {
4643 /* have to trust i_size as the end */
4645 last_for_get_extent = isize;
4648 * remember the start of the last extent. There are a
4649 * bunch of different factors that go into the length of the
4650 * extent, so its much less complex to remember where it started
4652 last = found_key.offset;
4653 last_for_get_extent = last + 1;
4655 btrfs_release_path(path);
4658 * we might have some extents allocated but more delalloc past those
4659 * extents. so, we trust isize unless the start of the last extent is
4664 last_for_get_extent = isize;
4667 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4670 em = get_extent_skip_holes(inode, start, last_for_get_extent);
4679 u64 offset_in_extent = 0;
4681 /* break if the extent we found is outside the range */
4682 if (em->start >= max || extent_map_end(em) < off)
4686 * get_extent may return an extent that starts before our
4687 * requested range. We have to make sure the ranges
4688 * we return to fiemap always move forward and don't
4689 * overlap, so adjust the offsets here
4691 em_start = max(em->start, off);
4694 * record the offset from the start of the extent
4695 * for adjusting the disk offset below. Only do this if the
4696 * extent isn't compressed since our in ram offset may be past
4697 * what we have actually allocated on disk.
4699 if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4700 offset_in_extent = em_start - em->start;
4701 em_end = extent_map_end(em);
4702 em_len = em_end - em_start;
4704 if (em->block_start < EXTENT_MAP_LAST_BYTE)
4705 disko = em->block_start + offset_in_extent;
4710 * bump off for our next call to get_extent
4712 off = extent_map_end(em);
4716 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
4718 flags |= FIEMAP_EXTENT_LAST;
4719 } else if (em->block_start == EXTENT_MAP_INLINE) {
4720 flags |= (FIEMAP_EXTENT_DATA_INLINE |
4721 FIEMAP_EXTENT_NOT_ALIGNED);
4722 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
4723 flags |= (FIEMAP_EXTENT_DELALLOC |
4724 FIEMAP_EXTENT_UNKNOWN);
4725 } else if (fieinfo->fi_extents_max) {
4726 u64 bytenr = em->block_start -
4727 (em->start - em->orig_start);
4730 * As btrfs supports shared space, this information
4731 * can be exported to userspace tools via
4732 * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0
4733 * then we're just getting a count and we can skip the
4736 ret = btrfs_check_shared(root,
4737 btrfs_ino(BTRFS_I(inode)),
4738 bytenr, roots, tmp_ulist);
4742 flags |= FIEMAP_EXTENT_SHARED;
4745 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4746 flags |= FIEMAP_EXTENT_ENCODED;
4747 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4748 flags |= FIEMAP_EXTENT_UNWRITTEN;
4750 free_extent_map(em);
4752 if ((em_start >= last) || em_len == (u64)-1 ||
4753 (last == (u64)-1 && isize <= em_end)) {
4754 flags |= FIEMAP_EXTENT_LAST;
4758 /* now scan forward to see if this is really the last extent. */
4759 em = get_extent_skip_holes(inode, off, last_for_get_extent);
4765 flags |= FIEMAP_EXTENT_LAST;
4768 ret = emit_fiemap_extent(fieinfo, &cache, em_start, disko,
4778 ret = emit_last_fiemap_cache(fieinfo, &cache);
4779 free_extent_map(em);
4781 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4785 btrfs_free_path(path);
4787 ulist_free(tmp_ulist);
4791 static void __free_extent_buffer(struct extent_buffer *eb)
4793 btrfs_leak_debug_del(&eb->leak_list);
4794 kmem_cache_free(extent_buffer_cache, eb);
4797 int extent_buffer_under_io(struct extent_buffer *eb)
4799 return (atomic_read(&eb->io_pages) ||
4800 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4801 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4805 * Release all pages attached to the extent buffer.
4807 static void btrfs_release_extent_buffer_pages(struct extent_buffer *eb)
4811 int mapped = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
4813 BUG_ON(extent_buffer_under_io(eb));
4815 num_pages = num_extent_pages(eb);
4816 for (i = 0; i < num_pages; i++) {
4817 struct page *page = eb->pages[i];
4822 spin_lock(&page->mapping->private_lock);
4824 * We do this since we'll remove the pages after we've
4825 * removed the eb from the radix tree, so we could race
4826 * and have this page now attached to the new eb. So
4827 * only clear page_private if it's still connected to
4830 if (PagePrivate(page) &&
4831 page->private == (unsigned long)eb) {
4832 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4833 BUG_ON(PageDirty(page));
4834 BUG_ON(PageWriteback(page));
4836 * We need to make sure we haven't be attached
4839 ClearPagePrivate(page);
4840 set_page_private(page, 0);
4841 /* One for the page private */
4846 spin_unlock(&page->mapping->private_lock);
4848 /* One for when we allocated the page */
4854 * Helper for releasing the extent buffer.
4856 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4858 btrfs_release_extent_buffer_pages(eb);
4859 __free_extent_buffer(eb);
4862 static struct extent_buffer *
4863 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
4866 struct extent_buffer *eb = NULL;
4868 eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL);
4871 eb->fs_info = fs_info;
4873 rwlock_init(&eb->lock);
4874 atomic_set(&eb->blocking_readers, 0);
4875 eb->blocking_writers = 0;
4876 eb->lock_nested = false;
4877 init_waitqueue_head(&eb->write_lock_wq);
4878 init_waitqueue_head(&eb->read_lock_wq);
4880 btrfs_leak_debug_add(&eb->leak_list, &buffers);
4882 spin_lock_init(&eb->refs_lock);
4883 atomic_set(&eb->refs, 1);
4884 atomic_set(&eb->io_pages, 0);
4887 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4889 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4890 > MAX_INLINE_EXTENT_BUFFER_SIZE);
4891 BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE);
4893 #ifdef CONFIG_BTRFS_DEBUG
4894 eb->spinning_writers = 0;
4895 atomic_set(&eb->spinning_readers, 0);
4896 atomic_set(&eb->read_locks, 0);
4897 eb->write_locks = 0;
4903 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src)
4907 struct extent_buffer *new;
4908 int num_pages = num_extent_pages(src);
4910 new = __alloc_extent_buffer(src->fs_info, src->start, src->len);
4914 for (i = 0; i < num_pages; i++) {
4915 p = alloc_page(GFP_NOFS);
4917 btrfs_release_extent_buffer(new);
4920 attach_extent_buffer_page(new, p);
4921 WARN_ON(PageDirty(p));
4924 copy_page(page_address(p), page_address(src->pages[i]));
4927 set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
4928 set_bit(EXTENT_BUFFER_UNMAPPED, &new->bflags);
4933 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4934 u64 start, unsigned long len)
4936 struct extent_buffer *eb;
4940 eb = __alloc_extent_buffer(fs_info, start, len);
4944 num_pages = num_extent_pages(eb);
4945 for (i = 0; i < num_pages; i++) {
4946 eb->pages[i] = alloc_page(GFP_NOFS);
4950 set_extent_buffer_uptodate(eb);
4951 btrfs_set_header_nritems(eb, 0);
4952 set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
4957 __free_page(eb->pages[i - 1]);
4958 __free_extent_buffer(eb);
4962 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4965 return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize);
4968 static void check_buffer_tree_ref(struct extent_buffer *eb)
4971 /* the ref bit is tricky. We have to make sure it is set
4972 * if we have the buffer dirty. Otherwise the
4973 * code to free a buffer can end up dropping a dirty
4976 * Once the ref bit is set, it won't go away while the
4977 * buffer is dirty or in writeback, and it also won't
4978 * go away while we have the reference count on the
4981 * We can't just set the ref bit without bumping the
4982 * ref on the eb because free_extent_buffer might
4983 * see the ref bit and try to clear it. If this happens
4984 * free_extent_buffer might end up dropping our original
4985 * ref by mistake and freeing the page before we are able
4986 * to add one more ref.
4988 * So bump the ref count first, then set the bit. If someone
4989 * beat us to it, drop the ref we added.
4991 refs = atomic_read(&eb->refs);
4992 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4995 spin_lock(&eb->refs_lock);
4996 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4997 atomic_inc(&eb->refs);
4998 spin_unlock(&eb->refs_lock);
5001 static void mark_extent_buffer_accessed(struct extent_buffer *eb,
5002 struct page *accessed)
5006 check_buffer_tree_ref(eb);
5008 num_pages = num_extent_pages(eb);
5009 for (i = 0; i < num_pages; i++) {
5010 struct page *p = eb->pages[i];
5013 mark_page_accessed(p);
5017 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
5020 struct extent_buffer *eb;
5023 eb = radix_tree_lookup(&fs_info->buffer_radix,
5024 start >> PAGE_SHIFT);
5025 if (eb && atomic_inc_not_zero(&eb->refs)) {
5028 * Lock our eb's refs_lock to avoid races with
5029 * free_extent_buffer. When we get our eb it might be flagged
5030 * with EXTENT_BUFFER_STALE and another task running
5031 * free_extent_buffer might have seen that flag set,
5032 * eb->refs == 2, that the buffer isn't under IO (dirty and
5033 * writeback flags not set) and it's still in the tree (flag
5034 * EXTENT_BUFFER_TREE_REF set), therefore being in the process
5035 * of decrementing the extent buffer's reference count twice.
5036 * So here we could race and increment the eb's reference count,
5037 * clear its stale flag, mark it as dirty and drop our reference
5038 * before the other task finishes executing free_extent_buffer,
5039 * which would later result in an attempt to free an extent
5040 * buffer that is dirty.
5042 if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) {
5043 spin_lock(&eb->refs_lock);
5044 spin_unlock(&eb->refs_lock);
5046 mark_extent_buffer_accessed(eb, NULL);
5054 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5055 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
5058 struct extent_buffer *eb, *exists = NULL;
5061 eb = find_extent_buffer(fs_info, start);
5064 eb = alloc_dummy_extent_buffer(fs_info, start);
5066 return ERR_PTR(-ENOMEM);
5067 eb->fs_info = fs_info;
5069 ret = radix_tree_preload(GFP_NOFS);
5071 exists = ERR_PTR(ret);
5074 spin_lock(&fs_info->buffer_lock);
5075 ret = radix_tree_insert(&fs_info->buffer_radix,
5076 start >> PAGE_SHIFT, eb);
5077 spin_unlock(&fs_info->buffer_lock);
5078 radix_tree_preload_end();
5079 if (ret == -EEXIST) {
5080 exists = find_extent_buffer(fs_info, start);
5086 check_buffer_tree_ref(eb);
5087 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
5091 btrfs_release_extent_buffer(eb);
5096 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
5099 unsigned long len = fs_info->nodesize;
5102 unsigned long index = start >> PAGE_SHIFT;
5103 struct extent_buffer *eb;
5104 struct extent_buffer *exists = NULL;
5106 struct address_space *mapping = fs_info->btree_inode->i_mapping;
5110 if (!IS_ALIGNED(start, fs_info->sectorsize)) {
5111 btrfs_err(fs_info, "bad tree block start %llu", start);
5112 return ERR_PTR(-EINVAL);
5115 eb = find_extent_buffer(fs_info, start);
5119 eb = __alloc_extent_buffer(fs_info, start, len);
5121 return ERR_PTR(-ENOMEM);
5123 num_pages = num_extent_pages(eb);
5124 for (i = 0; i < num_pages; i++, index++) {
5125 p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL);
5127 exists = ERR_PTR(-ENOMEM);
5131 spin_lock(&mapping->private_lock);
5132 if (PagePrivate(p)) {
5134 * We could have already allocated an eb for this page
5135 * and attached one so lets see if we can get a ref on
5136 * the existing eb, and if we can we know it's good and
5137 * we can just return that one, else we know we can just
5138 * overwrite page->private.
5140 exists = (struct extent_buffer *)p->private;
5141 if (atomic_inc_not_zero(&exists->refs)) {
5142 spin_unlock(&mapping->private_lock);
5145 mark_extent_buffer_accessed(exists, p);
5151 * Do this so attach doesn't complain and we need to
5152 * drop the ref the old guy had.
5154 ClearPagePrivate(p);
5155 WARN_ON(PageDirty(p));
5158 attach_extent_buffer_page(eb, p);
5159 spin_unlock(&mapping->private_lock);
5160 WARN_ON(PageDirty(p));
5162 if (!PageUptodate(p))
5166 * We can't unlock the pages just yet since the extent buffer
5167 * hasn't been properly inserted in the radix tree, this
5168 * opens a race with btree_releasepage which can free a page
5169 * while we are still filling in all pages for the buffer and
5174 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5176 ret = radix_tree_preload(GFP_NOFS);
5178 exists = ERR_PTR(ret);
5182 spin_lock(&fs_info->buffer_lock);
5183 ret = radix_tree_insert(&fs_info->buffer_radix,
5184 start >> PAGE_SHIFT, eb);
5185 spin_unlock(&fs_info->buffer_lock);
5186 radix_tree_preload_end();
5187 if (ret == -EEXIST) {
5188 exists = find_extent_buffer(fs_info, start);
5194 /* add one reference for the tree */
5195 check_buffer_tree_ref(eb);
5196 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
5199 * Now it's safe to unlock the pages because any calls to
5200 * btree_releasepage will correctly detect that a page belongs to a
5201 * live buffer and won't free them prematurely.
5203 for (i = 0; i < num_pages; i++)
5204 unlock_page(eb->pages[i]);
5208 WARN_ON(!atomic_dec_and_test(&eb->refs));
5209 for (i = 0; i < num_pages; i++) {
5211 unlock_page(eb->pages[i]);
5214 btrfs_release_extent_buffer(eb);
5218 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
5220 struct extent_buffer *eb =
5221 container_of(head, struct extent_buffer, rcu_head);
5223 __free_extent_buffer(eb);
5226 static int release_extent_buffer(struct extent_buffer *eb)
5228 lockdep_assert_held(&eb->refs_lock);
5230 WARN_ON(atomic_read(&eb->refs) == 0);
5231 if (atomic_dec_and_test(&eb->refs)) {
5232 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
5233 struct btrfs_fs_info *fs_info = eb->fs_info;
5235 spin_unlock(&eb->refs_lock);
5237 spin_lock(&fs_info->buffer_lock);
5238 radix_tree_delete(&fs_info->buffer_radix,
5239 eb->start >> PAGE_SHIFT);
5240 spin_unlock(&fs_info->buffer_lock);
5242 spin_unlock(&eb->refs_lock);
5245 /* Should be safe to release our pages at this point */
5246 btrfs_release_extent_buffer_pages(eb);
5247 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5248 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags))) {
5249 __free_extent_buffer(eb);
5253 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
5256 spin_unlock(&eb->refs_lock);
5261 void free_extent_buffer(struct extent_buffer *eb)
5269 refs = atomic_read(&eb->refs);
5270 if ((!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && refs <= 3)
5271 || (test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) &&
5274 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
5279 spin_lock(&eb->refs_lock);
5280 if (atomic_read(&eb->refs) == 2 &&
5281 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
5282 !extent_buffer_under_io(eb) &&
5283 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5284 atomic_dec(&eb->refs);
5287 * I know this is terrible, but it's temporary until we stop tracking
5288 * the uptodate bits and such for the extent buffers.
5290 release_extent_buffer(eb);
5293 void free_extent_buffer_stale(struct extent_buffer *eb)
5298 spin_lock(&eb->refs_lock);
5299 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
5301 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
5302 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5303 atomic_dec(&eb->refs);
5304 release_extent_buffer(eb);
5307 void clear_extent_buffer_dirty(struct extent_buffer *eb)
5313 num_pages = num_extent_pages(eb);
5315 for (i = 0; i < num_pages; i++) {
5316 page = eb->pages[i];
5317 if (!PageDirty(page))
5321 WARN_ON(!PagePrivate(page));
5323 clear_page_dirty_for_io(page);
5324 xa_lock_irq(&page->mapping->i_pages);
5325 if (!PageDirty(page))
5326 __xa_clear_mark(&page->mapping->i_pages,
5327 page_index(page), PAGECACHE_TAG_DIRTY);
5328 xa_unlock_irq(&page->mapping->i_pages);
5329 ClearPageError(page);
5332 WARN_ON(atomic_read(&eb->refs) == 0);
5335 bool set_extent_buffer_dirty(struct extent_buffer *eb)
5341 check_buffer_tree_ref(eb);
5343 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
5345 num_pages = num_extent_pages(eb);
5346 WARN_ON(atomic_read(&eb->refs) == 0);
5347 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
5350 for (i = 0; i < num_pages; i++)
5351 set_page_dirty(eb->pages[i]);
5353 #ifdef CONFIG_BTRFS_DEBUG
5354 for (i = 0; i < num_pages; i++)
5355 ASSERT(PageDirty(eb->pages[i]));
5361 void clear_extent_buffer_uptodate(struct extent_buffer *eb)
5367 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5368 num_pages = num_extent_pages(eb);
5369 for (i = 0; i < num_pages; i++) {
5370 page = eb->pages[i];
5372 ClearPageUptodate(page);
5376 void set_extent_buffer_uptodate(struct extent_buffer *eb)
5382 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5383 num_pages = num_extent_pages(eb);
5384 for (i = 0; i < num_pages; i++) {
5385 page = eb->pages[i];
5386 SetPageUptodate(page);
5390 int read_extent_buffer_pages(struct extent_buffer *eb, int wait, int mirror_num)
5396 int locked_pages = 0;
5397 int all_uptodate = 1;
5399 unsigned long num_reads = 0;
5400 struct bio *bio = NULL;
5401 unsigned long bio_flags = 0;
5402 struct extent_io_tree *tree = &BTRFS_I(eb->fs_info->btree_inode)->io_tree;
5404 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
5407 num_pages = num_extent_pages(eb);
5408 for (i = 0; i < num_pages; i++) {
5409 page = eb->pages[i];
5410 if (wait == WAIT_NONE) {
5411 if (!trylock_page(page))
5419 * We need to firstly lock all pages to make sure that
5420 * the uptodate bit of our pages won't be affected by
5421 * clear_extent_buffer_uptodate().
5423 for (i = 0; i < num_pages; i++) {
5424 page = eb->pages[i];
5425 if (!PageUptodate(page)) {
5432 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5436 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
5437 eb->read_mirror = 0;
5438 atomic_set(&eb->io_pages, num_reads);
5439 for (i = 0; i < num_pages; i++) {
5440 page = eb->pages[i];
5442 if (!PageUptodate(page)) {
5444 atomic_dec(&eb->io_pages);
5449 ClearPageError(page);
5450 err = __extent_read_full_page(tree, page,
5451 btree_get_extent, &bio,
5452 mirror_num, &bio_flags,
5457 * We use &bio in above __extent_read_full_page,
5458 * so we ensure that if it returns error, the
5459 * current page fails to add itself to bio and
5460 * it's been unlocked.
5462 * We must dec io_pages by ourselves.
5464 atomic_dec(&eb->io_pages);
5472 err = submit_one_bio(bio, mirror_num, bio_flags);
5477 if (ret || wait != WAIT_COMPLETE)
5480 for (i = 0; i < num_pages; i++) {
5481 page = eb->pages[i];
5482 wait_on_page_locked(page);
5483 if (!PageUptodate(page))
5490 while (locked_pages > 0) {
5492 page = eb->pages[locked_pages];
5498 void read_extent_buffer(const struct extent_buffer *eb, void *dstv,
5499 unsigned long start, unsigned long len)
5505 char *dst = (char *)dstv;
5506 size_t start_offset = offset_in_page(eb->start);
5507 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5509 if (start + len > eb->len) {
5510 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n",
5511 eb->start, eb->len, start, len);
5512 memset(dst, 0, len);
5516 offset = offset_in_page(start_offset + start);
5519 page = eb->pages[i];
5521 cur = min(len, (PAGE_SIZE - offset));
5522 kaddr = page_address(page);
5523 memcpy(dst, kaddr + offset, cur);
5532 int read_extent_buffer_to_user(const struct extent_buffer *eb,
5534 unsigned long start, unsigned long len)
5540 char __user *dst = (char __user *)dstv;
5541 size_t start_offset = offset_in_page(eb->start);
5542 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5545 WARN_ON(start > eb->len);
5546 WARN_ON(start + len > eb->start + eb->len);
5548 offset = offset_in_page(start_offset + start);
5551 page = eb->pages[i];
5553 cur = min(len, (PAGE_SIZE - offset));
5554 kaddr = page_address(page);
5555 if (copy_to_user(dst, kaddr + offset, cur)) {
5570 * return 0 if the item is found within a page.
5571 * return 1 if the item spans two pages.
5572 * return -EINVAL otherwise.
5574 int map_private_extent_buffer(const struct extent_buffer *eb,
5575 unsigned long start, unsigned long min_len,
5576 char **map, unsigned long *map_start,
5577 unsigned long *map_len)
5582 size_t start_offset = offset_in_page(eb->start);
5583 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5584 unsigned long end_i = (start_offset + start + min_len - 1) >>
5587 if (start + min_len > eb->len) {
5588 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n",
5589 eb->start, eb->len, start, min_len);
5597 offset = start_offset;
5601 *map_start = ((u64)i << PAGE_SHIFT) - start_offset;
5605 kaddr = page_address(p);
5606 *map = kaddr + offset;
5607 *map_len = PAGE_SIZE - offset;
5611 int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,
5612 unsigned long start, unsigned long len)
5618 char *ptr = (char *)ptrv;
5619 size_t start_offset = offset_in_page(eb->start);
5620 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5623 WARN_ON(start > eb->len);
5624 WARN_ON(start + len > eb->start + eb->len);
5626 offset = offset_in_page(start_offset + start);
5629 page = eb->pages[i];
5631 cur = min(len, (PAGE_SIZE - offset));
5633 kaddr = page_address(page);
5634 ret = memcmp(ptr, kaddr + offset, cur);
5646 void write_extent_buffer_chunk_tree_uuid(struct extent_buffer *eb,
5651 WARN_ON(!PageUptodate(eb->pages[0]));
5652 kaddr = page_address(eb->pages[0]);
5653 memcpy(kaddr + offsetof(struct btrfs_header, chunk_tree_uuid), srcv,
5657 void write_extent_buffer_fsid(struct extent_buffer *eb, const void *srcv)
5661 WARN_ON(!PageUptodate(eb->pages[0]));
5662 kaddr = page_address(eb->pages[0]);
5663 memcpy(kaddr + offsetof(struct btrfs_header, fsid), srcv,
5667 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
5668 unsigned long start, unsigned long len)
5674 char *src = (char *)srcv;
5675 size_t start_offset = offset_in_page(eb->start);
5676 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5678 WARN_ON(start > eb->len);
5679 WARN_ON(start + len > eb->start + eb->len);
5681 offset = offset_in_page(start_offset + start);
5684 page = eb->pages[i];
5685 WARN_ON(!PageUptodate(page));
5687 cur = min(len, PAGE_SIZE - offset);
5688 kaddr = page_address(page);
5689 memcpy(kaddr + offset, src, cur);
5698 void memzero_extent_buffer(struct extent_buffer *eb, unsigned long start,
5705 size_t start_offset = offset_in_page(eb->start);
5706 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5708 WARN_ON(start > eb->len);
5709 WARN_ON(start + len > eb->start + eb->len);
5711 offset = offset_in_page(start_offset + start);
5714 page = eb->pages[i];
5715 WARN_ON(!PageUptodate(page));
5717 cur = min(len, PAGE_SIZE - offset);
5718 kaddr = page_address(page);
5719 memset(kaddr + offset, 0, cur);
5727 void copy_extent_buffer_full(struct extent_buffer *dst,
5728 struct extent_buffer *src)
5733 ASSERT(dst->len == src->len);
5735 num_pages = num_extent_pages(dst);
5736 for (i = 0; i < num_pages; i++)
5737 copy_page(page_address(dst->pages[i]),
5738 page_address(src->pages[i]));
5741 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
5742 unsigned long dst_offset, unsigned long src_offset,
5745 u64 dst_len = dst->len;
5750 size_t start_offset = offset_in_page(dst->start);
5751 unsigned long i = (start_offset + dst_offset) >> PAGE_SHIFT;
5753 WARN_ON(src->len != dst_len);
5755 offset = offset_in_page(start_offset + dst_offset);
5758 page = dst->pages[i];
5759 WARN_ON(!PageUptodate(page));
5761 cur = min(len, (unsigned long)(PAGE_SIZE - offset));
5763 kaddr = page_address(page);
5764 read_extent_buffer(src, kaddr + offset, src_offset, cur);
5774 * eb_bitmap_offset() - calculate the page and offset of the byte containing the
5776 * @eb: the extent buffer
5777 * @start: offset of the bitmap item in the extent buffer
5779 * @page_index: return index of the page in the extent buffer that contains the
5781 * @page_offset: return offset into the page given by page_index
5783 * This helper hides the ugliness of finding the byte in an extent buffer which
5784 * contains a given bit.
5786 static inline void eb_bitmap_offset(struct extent_buffer *eb,
5787 unsigned long start, unsigned long nr,
5788 unsigned long *page_index,
5789 size_t *page_offset)
5791 size_t start_offset = offset_in_page(eb->start);
5792 size_t byte_offset = BIT_BYTE(nr);
5796 * The byte we want is the offset of the extent buffer + the offset of
5797 * the bitmap item in the extent buffer + the offset of the byte in the
5800 offset = start_offset + start + byte_offset;
5802 *page_index = offset >> PAGE_SHIFT;
5803 *page_offset = offset_in_page(offset);
5807 * extent_buffer_test_bit - determine whether a bit in a bitmap item is set
5808 * @eb: the extent buffer
5809 * @start: offset of the bitmap item in the extent buffer
5810 * @nr: bit number to test
5812 int extent_buffer_test_bit(struct extent_buffer *eb, unsigned long start,
5820 eb_bitmap_offset(eb, start, nr, &i, &offset);
5821 page = eb->pages[i];
5822 WARN_ON(!PageUptodate(page));
5823 kaddr = page_address(page);
5824 return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1)));
5828 * extent_buffer_bitmap_set - set an area of a bitmap
5829 * @eb: the extent buffer
5830 * @start: offset of the bitmap item in the extent buffer
5831 * @pos: bit number of the first bit
5832 * @len: number of bits to set
5834 void extent_buffer_bitmap_set(struct extent_buffer *eb, unsigned long start,
5835 unsigned long pos, unsigned long len)
5841 const unsigned int size = pos + len;
5842 int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5843 u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos);
5845 eb_bitmap_offset(eb, start, pos, &i, &offset);
5846 page = eb->pages[i];
5847 WARN_ON(!PageUptodate(page));
5848 kaddr = page_address(page);
5850 while (len >= bits_to_set) {
5851 kaddr[offset] |= mask_to_set;
5853 bits_to_set = BITS_PER_BYTE;
5855 if (++offset >= PAGE_SIZE && len > 0) {
5857 page = eb->pages[++i];
5858 WARN_ON(!PageUptodate(page));
5859 kaddr = page_address(page);
5863 mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
5864 kaddr[offset] |= mask_to_set;
5870 * extent_buffer_bitmap_clear - clear an area of a bitmap
5871 * @eb: the extent buffer
5872 * @start: offset of the bitmap item in the extent buffer
5873 * @pos: bit number of the first bit
5874 * @len: number of bits to clear
5876 void extent_buffer_bitmap_clear(struct extent_buffer *eb, unsigned long start,
5877 unsigned long pos, unsigned long len)
5883 const unsigned int size = pos + len;
5884 int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5885 u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos);
5887 eb_bitmap_offset(eb, start, pos, &i, &offset);
5888 page = eb->pages[i];
5889 WARN_ON(!PageUptodate(page));
5890 kaddr = page_address(page);
5892 while (len >= bits_to_clear) {
5893 kaddr[offset] &= ~mask_to_clear;
5894 len -= bits_to_clear;
5895 bits_to_clear = BITS_PER_BYTE;
5897 if (++offset >= PAGE_SIZE && len > 0) {
5899 page = eb->pages[++i];
5900 WARN_ON(!PageUptodate(page));
5901 kaddr = page_address(page);
5905 mask_to_clear &= BITMAP_LAST_BYTE_MASK(size);
5906 kaddr[offset] &= ~mask_to_clear;
5910 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
5912 unsigned long distance = (src > dst) ? src - dst : dst - src;
5913 return distance < len;
5916 static void copy_pages(struct page *dst_page, struct page *src_page,
5917 unsigned long dst_off, unsigned long src_off,
5920 char *dst_kaddr = page_address(dst_page);
5922 int must_memmove = 0;
5924 if (dst_page != src_page) {
5925 src_kaddr = page_address(src_page);
5927 src_kaddr = dst_kaddr;
5928 if (areas_overlap(src_off, dst_off, len))
5933 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
5935 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
5938 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5939 unsigned long src_offset, unsigned long len)
5941 struct btrfs_fs_info *fs_info = dst->fs_info;
5943 size_t dst_off_in_page;
5944 size_t src_off_in_page;
5945 size_t start_offset = offset_in_page(dst->start);
5946 unsigned long dst_i;
5947 unsigned long src_i;
5949 if (src_offset + len > dst->len) {
5951 "memmove bogus src_offset %lu move len %lu dst len %lu",
5952 src_offset, len, dst->len);
5955 if (dst_offset + len > dst->len) {
5957 "memmove bogus dst_offset %lu move len %lu dst len %lu",
5958 dst_offset, len, dst->len);
5963 dst_off_in_page = offset_in_page(start_offset + dst_offset);
5964 src_off_in_page = offset_in_page(start_offset + src_offset);
5966 dst_i = (start_offset + dst_offset) >> PAGE_SHIFT;
5967 src_i = (start_offset + src_offset) >> PAGE_SHIFT;
5969 cur = min(len, (unsigned long)(PAGE_SIZE -
5971 cur = min_t(unsigned long, cur,
5972 (unsigned long)(PAGE_SIZE - dst_off_in_page));
5974 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5975 dst_off_in_page, src_off_in_page, cur);
5983 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5984 unsigned long src_offset, unsigned long len)
5986 struct btrfs_fs_info *fs_info = dst->fs_info;
5988 size_t dst_off_in_page;
5989 size_t src_off_in_page;
5990 unsigned long dst_end = dst_offset + len - 1;
5991 unsigned long src_end = src_offset + len - 1;
5992 size_t start_offset = offset_in_page(dst->start);
5993 unsigned long dst_i;
5994 unsigned long src_i;
5996 if (src_offset + len > dst->len) {
5998 "memmove bogus src_offset %lu move len %lu len %lu",
5999 src_offset, len, dst->len);
6002 if (dst_offset + len > dst->len) {
6004 "memmove bogus dst_offset %lu move len %lu len %lu",
6005 dst_offset, len, dst->len);
6008 if (dst_offset < src_offset) {
6009 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
6013 dst_i = (start_offset + dst_end) >> PAGE_SHIFT;
6014 src_i = (start_offset + src_end) >> PAGE_SHIFT;
6016 dst_off_in_page = offset_in_page(start_offset + dst_end);
6017 src_off_in_page = offset_in_page(start_offset + src_end);
6019 cur = min_t(unsigned long, len, src_off_in_page + 1);
6020 cur = min(cur, dst_off_in_page + 1);
6021 copy_pages(dst->pages[dst_i], dst->pages[src_i],
6022 dst_off_in_page - cur + 1,
6023 src_off_in_page - cur + 1, cur);
6031 int try_release_extent_buffer(struct page *page)
6033 struct extent_buffer *eb;
6036 * We need to make sure nobody is attaching this page to an eb right
6039 spin_lock(&page->mapping->private_lock);
6040 if (!PagePrivate(page)) {
6041 spin_unlock(&page->mapping->private_lock);
6045 eb = (struct extent_buffer *)page->private;
6049 * This is a little awful but should be ok, we need to make sure that
6050 * the eb doesn't disappear out from under us while we're looking at
6053 spin_lock(&eb->refs_lock);
6054 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
6055 spin_unlock(&eb->refs_lock);
6056 spin_unlock(&page->mapping->private_lock);
6059 spin_unlock(&page->mapping->private_lock);
6062 * If tree ref isn't set then we know the ref on this eb is a real ref,
6063 * so just return, this page will likely be freed soon anyway.
6065 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
6066 spin_unlock(&eb->refs_lock);
6070 return release_extent_buffer(eb);