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/sched/mm.h>
10 #include <linux/spinlock.h>
11 #include <linux/blkdev.h>
12 #include <linux/swap.h>
13 #include <linux/writeback.h>
14 #include <linux/pagevec.h>
15 #include <linux/prefetch.h>
16 #include <linux/fsverity.h>
18 #include "extent_io.h"
19 #include "extent-io-tree.h"
20 #include "extent_map.h"
22 #include "btrfs_inode.h"
24 #include "check-integrity.h"
26 #include "rcu-string.h"
31 #include "block-group.h"
32 #include "compression.h"
34 static struct kmem_cache *extent_state_cache;
35 static struct kmem_cache *extent_buffer_cache;
36 static struct bio_set btrfs_bioset;
38 static inline bool extent_state_in_tree(const struct extent_state *state)
40 return !RB_EMPTY_NODE(&state->rb_node);
43 #ifdef CONFIG_BTRFS_DEBUG
44 static LIST_HEAD(states);
45 static DEFINE_SPINLOCK(leak_lock);
47 static inline void btrfs_leak_debug_add(spinlock_t *lock,
48 struct list_head *new,
49 struct list_head *head)
53 spin_lock_irqsave(lock, flags);
55 spin_unlock_irqrestore(lock, flags);
58 static inline void btrfs_leak_debug_del(spinlock_t *lock,
59 struct list_head *entry)
63 spin_lock_irqsave(lock, flags);
65 spin_unlock_irqrestore(lock, flags);
68 void btrfs_extent_buffer_leak_debug_check(struct btrfs_fs_info *fs_info)
70 struct extent_buffer *eb;
74 * If we didn't get into open_ctree our allocated_ebs will not be
75 * initialized, so just skip this.
77 if (!fs_info->allocated_ebs.next)
80 WARN_ON(!list_empty(&fs_info->allocated_ebs));
81 spin_lock_irqsave(&fs_info->eb_leak_lock, flags);
82 while (!list_empty(&fs_info->allocated_ebs)) {
83 eb = list_first_entry(&fs_info->allocated_ebs,
84 struct extent_buffer, leak_list);
86 "BTRFS: buffer leak start %llu len %lu refs %d bflags %lu owner %llu\n",
87 eb->start, eb->len, atomic_read(&eb->refs), eb->bflags,
88 btrfs_header_owner(eb));
89 list_del(&eb->leak_list);
90 kmem_cache_free(extent_buffer_cache, eb);
92 spin_unlock_irqrestore(&fs_info->eb_leak_lock, flags);
95 static inline void btrfs_extent_state_leak_debug_check(void)
97 struct extent_state *state;
99 while (!list_empty(&states)) {
100 state = list_entry(states.next, struct extent_state, leak_list);
101 pr_err("BTRFS: state leak: start %llu end %llu state %u in tree %d refs %d\n",
102 state->start, state->end, state->state,
103 extent_state_in_tree(state),
104 refcount_read(&state->refs));
105 list_del(&state->leak_list);
106 kmem_cache_free(extent_state_cache, state);
110 #define btrfs_debug_check_extent_io_range(tree, start, end) \
111 __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end))
112 static inline void __btrfs_debug_check_extent_io_range(const char *caller,
113 struct extent_io_tree *tree, u64 start, u64 end)
115 struct inode *inode = tree->private_data;
118 if (!inode || !is_data_inode(inode))
121 isize = i_size_read(inode);
122 if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) {
123 btrfs_debug_rl(BTRFS_I(inode)->root->fs_info,
124 "%s: ino %llu isize %llu odd range [%llu,%llu]",
125 caller, btrfs_ino(BTRFS_I(inode)), isize, start, end);
129 #define btrfs_leak_debug_add(lock, new, head) do {} while (0)
130 #define btrfs_leak_debug_del(lock, entry) do {} while (0)
131 #define btrfs_extent_state_leak_debug_check() do {} while (0)
132 #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0)
138 struct rb_node rb_node;
142 * Structure to record info about the bio being assembled, and other info like
143 * how many bytes are there before stripe/ordered extent boundary.
145 struct btrfs_bio_ctrl {
148 enum btrfs_compression_type compress_type;
149 u32 len_to_stripe_boundary;
150 u32 len_to_oe_boundary;
153 struct extent_page_data {
154 struct btrfs_bio_ctrl bio_ctrl;
155 /* tells writepage not to lock the state bits for this range
156 * it still does the unlocking
158 unsigned int extent_locked:1;
160 /* tells the submit_bio code to use REQ_SYNC */
161 unsigned int sync_io:1;
164 static int add_extent_changeset(struct extent_state *state, u32 bits,
165 struct extent_changeset *changeset,
172 if (set && (state->state & bits) == bits)
174 if (!set && (state->state & bits) == 0)
176 changeset->bytes_changed += state->end - state->start + 1;
177 ret = ulist_add(&changeset->range_changed, state->start, state->end,
182 static void submit_one_bio(struct btrfs_bio_ctrl *bio_ctrl)
192 inode = bio_first_page_all(bio)->mapping->host;
193 mirror_num = bio_ctrl->mirror_num;
195 /* Caller should ensure the bio has at least some range added */
196 ASSERT(bio->bi_iter.bi_size);
198 if (!is_data_inode(inode))
199 btrfs_submit_metadata_bio(inode, bio, mirror_num);
200 else if (btrfs_op(bio) == BTRFS_MAP_WRITE)
201 btrfs_submit_data_write_bio(inode, bio, mirror_num);
203 btrfs_submit_data_read_bio(inode, bio, mirror_num,
204 bio_ctrl->compress_type);
206 /* The bio is owned by the bi_end_io handler now */
207 bio_ctrl->bio = NULL;
211 * Submit or fail the current bio in an extent_page_data structure.
213 static void submit_write_bio(struct extent_page_data *epd, int ret)
215 struct bio *bio = epd->bio_ctrl.bio;
222 bio->bi_status = errno_to_blk_status(ret);
224 /* The bio is owned by the bi_end_io handler now */
225 epd->bio_ctrl.bio = NULL;
227 submit_one_bio(&epd->bio_ctrl);
231 int __init extent_state_cache_init(void)
233 extent_state_cache = kmem_cache_create("btrfs_extent_state",
234 sizeof(struct extent_state), 0,
235 SLAB_MEM_SPREAD, NULL);
236 if (!extent_state_cache)
241 int __init extent_io_init(void)
243 extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer",
244 sizeof(struct extent_buffer), 0,
245 SLAB_MEM_SPREAD, NULL);
246 if (!extent_buffer_cache)
249 if (bioset_init(&btrfs_bioset, BIO_POOL_SIZE,
250 offsetof(struct btrfs_bio, bio),
252 goto free_buffer_cache;
254 if (bioset_integrity_create(&btrfs_bioset, BIO_POOL_SIZE))
260 bioset_exit(&btrfs_bioset);
263 kmem_cache_destroy(extent_buffer_cache);
264 extent_buffer_cache = NULL;
268 void __cold extent_state_cache_exit(void)
270 btrfs_extent_state_leak_debug_check();
271 kmem_cache_destroy(extent_state_cache);
274 void __cold extent_io_exit(void)
277 * Make sure all delayed rcu free are flushed before we
281 kmem_cache_destroy(extent_buffer_cache);
282 bioset_exit(&btrfs_bioset);
286 * For the file_extent_tree, we want to hold the inode lock when we lookup and
287 * update the disk_i_size, but lockdep will complain because our io_tree we hold
288 * the tree lock and get the inode lock when setting delalloc. These two things
289 * are unrelated, so make a class for the file_extent_tree so we don't get the
290 * two locking patterns mixed up.
292 static struct lock_class_key file_extent_tree_class;
294 void extent_io_tree_init(struct btrfs_fs_info *fs_info,
295 struct extent_io_tree *tree, unsigned int owner,
298 tree->fs_info = fs_info;
299 tree->state = RB_ROOT;
300 tree->dirty_bytes = 0;
301 spin_lock_init(&tree->lock);
302 tree->private_data = private_data;
304 if (owner == IO_TREE_INODE_FILE_EXTENT)
305 lockdep_set_class(&tree->lock, &file_extent_tree_class);
308 void extent_io_tree_release(struct extent_io_tree *tree)
310 spin_lock(&tree->lock);
312 * Do a single barrier for the waitqueue_active check here, the state
313 * of the waitqueue should not change once extent_io_tree_release is
317 while (!RB_EMPTY_ROOT(&tree->state)) {
318 struct rb_node *node;
319 struct extent_state *state;
321 node = rb_first(&tree->state);
322 state = rb_entry(node, struct extent_state, rb_node);
323 rb_erase(&state->rb_node, &tree->state);
324 RB_CLEAR_NODE(&state->rb_node);
326 * btree io trees aren't supposed to have tasks waiting for
327 * changes in the flags of extent states ever.
329 ASSERT(!waitqueue_active(&state->wq));
330 free_extent_state(state);
332 cond_resched_lock(&tree->lock);
334 spin_unlock(&tree->lock);
337 static struct extent_state *alloc_extent_state(gfp_t mask)
339 struct extent_state *state;
342 * The given mask might be not appropriate for the slab allocator,
343 * drop the unsupported bits
345 mask &= ~(__GFP_DMA32|__GFP_HIGHMEM);
346 state = kmem_cache_alloc(extent_state_cache, mask);
350 state->failrec = NULL;
351 RB_CLEAR_NODE(&state->rb_node);
352 btrfs_leak_debug_add(&leak_lock, &state->leak_list, &states);
353 refcount_set(&state->refs, 1);
354 init_waitqueue_head(&state->wq);
355 trace_alloc_extent_state(state, mask, _RET_IP_);
359 void free_extent_state(struct extent_state *state)
363 if (refcount_dec_and_test(&state->refs)) {
364 WARN_ON(extent_state_in_tree(state));
365 btrfs_leak_debug_del(&leak_lock, &state->leak_list);
366 trace_free_extent_state(state, _RET_IP_);
367 kmem_cache_free(extent_state_cache, state);
372 * Search @tree for an entry that contains @offset. Such entry would have
373 * entry->start <= offset && entry->end >= offset.
375 * @tree: the tree to search
376 * @offset: offset that should fall within an entry in @tree
377 * @next_ret: pointer to the first entry whose range ends after @offset
378 * @prev_ret: pointer to the first entry whose range begins before @offset
379 * @p_ret: pointer where new node should be anchored (used when inserting an
381 * @parent_ret: points to entry which would have been the parent of the entry,
384 * This function returns a pointer to the entry that contains @offset byte
385 * address. If no such entry exists, then NULL is returned and the other
386 * pointer arguments to the function are filled, otherwise the found entry is
387 * returned and other pointers are left untouched.
389 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
390 struct rb_node **next_ret,
391 struct rb_node **prev_ret,
392 struct rb_node ***p_ret,
393 struct rb_node **parent_ret)
395 struct rb_root *root = &tree->state;
396 struct rb_node **n = &root->rb_node;
397 struct rb_node *prev = NULL;
398 struct rb_node *orig_prev = NULL;
399 struct tree_entry *entry;
400 struct tree_entry *prev_entry = NULL;
404 entry = rb_entry(prev, struct tree_entry, rb_node);
407 if (offset < entry->start)
409 else if (offset > entry->end)
422 while (prev && offset > prev_entry->end) {
423 prev = rb_next(prev);
424 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
431 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
432 while (prev && offset < prev_entry->start) {
433 prev = rb_prev(prev);
434 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
441 static inline struct rb_node *
442 tree_search_for_insert(struct extent_io_tree *tree,
444 struct rb_node ***p_ret,
445 struct rb_node **parent_ret)
447 struct rb_node *next= NULL;
450 ret = __etree_search(tree, offset, &next, NULL, p_ret, parent_ret);
456 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
459 return tree_search_for_insert(tree, offset, NULL, NULL);
463 * utility function to look for merge candidates inside a given range.
464 * Any extents with matching state are merged together into a single
465 * extent in the tree. Extents with EXTENT_IO in their state field
466 * are not merged because the end_io handlers need to be able to do
467 * operations on them without sleeping (or doing allocations/splits).
469 * This should be called with the tree lock held.
471 static void merge_state(struct extent_io_tree *tree,
472 struct extent_state *state)
474 struct extent_state *other;
475 struct rb_node *other_node;
477 if (state->state & (EXTENT_LOCKED | EXTENT_BOUNDARY))
480 other_node = rb_prev(&state->rb_node);
482 other = rb_entry(other_node, struct extent_state, rb_node);
483 if (other->end == state->start - 1 &&
484 other->state == state->state) {
485 if (tree->private_data &&
486 is_data_inode(tree->private_data))
487 btrfs_merge_delalloc_extent(tree->private_data,
489 state->start = other->start;
490 rb_erase(&other->rb_node, &tree->state);
491 RB_CLEAR_NODE(&other->rb_node);
492 free_extent_state(other);
495 other_node = rb_next(&state->rb_node);
497 other = rb_entry(other_node, struct extent_state, rb_node);
498 if (other->start == state->end + 1 &&
499 other->state == state->state) {
500 if (tree->private_data &&
501 is_data_inode(tree->private_data))
502 btrfs_merge_delalloc_extent(tree->private_data,
504 state->end = other->end;
505 rb_erase(&other->rb_node, &tree->state);
506 RB_CLEAR_NODE(&other->rb_node);
507 free_extent_state(other);
512 static void set_state_bits(struct extent_io_tree *tree,
513 struct extent_state *state, u32 bits,
514 struct extent_changeset *changeset);
517 * insert an extent_state struct into the tree. 'bits' are set on the
518 * struct before it is inserted.
520 * This may return -EEXIST if the extent is already there, in which case the
521 * state struct is freed.
523 * The tree lock is not taken internally. This is a utility function and
524 * probably isn't what you want to call (see set/clear_extent_bit).
526 static int insert_state(struct extent_io_tree *tree,
527 struct extent_state *state,
528 struct rb_node ***node_in,
529 struct rb_node **parent_in,
530 u32 bits, struct extent_changeset *changeset)
532 struct rb_node **node;
533 struct rb_node *parent;
534 const u64 end = state->end;
536 set_state_bits(tree, state, bits, changeset);
538 /* Caller provides the exact tree location */
539 if (node_in && parent_in) {
545 node = &tree->state.rb_node;
547 struct tree_entry *entry;
550 entry = rb_entry(parent, struct tree_entry, rb_node);
552 if (end < entry->start) {
553 node = &(*node)->rb_left;
554 } else if (end > entry->end) {
555 node = &(*node)->rb_right;
557 btrfs_err(tree->fs_info,
558 "found node %llu %llu on insert of %llu %llu",
559 entry->start, entry->end, state->start, end);
565 rb_link_node(&state->rb_node, parent, node);
566 rb_insert_color(&state->rb_node, &tree->state);
568 merge_state(tree, state);
573 * Insert state to @tree to the location given by @node and @parent.
575 static void insert_state_fast(struct extent_io_tree *tree,
576 struct extent_state *state, struct rb_node **node,
577 struct rb_node *parent, unsigned bits,
578 struct extent_changeset *changeset)
580 set_state_bits(tree, state, bits, changeset);
581 rb_link_node(&state->rb_node, parent, node);
582 rb_insert_color(&state->rb_node, &tree->state);
583 merge_state(tree, state);
587 * split a given extent state struct in two, inserting the preallocated
588 * struct 'prealloc' as the newly created second half. 'split' indicates an
589 * offset inside 'orig' where it should be split.
592 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
593 * are two extent state structs in the tree:
594 * prealloc: [orig->start, split - 1]
595 * orig: [ split, orig->end ]
597 * The tree locks are not taken by this function. They need to be held
600 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
601 struct extent_state *prealloc, u64 split)
603 struct rb_node *parent = NULL;
604 struct rb_node **node;
606 if (tree->private_data && is_data_inode(tree->private_data))
607 btrfs_split_delalloc_extent(tree->private_data, orig, split);
609 prealloc->start = orig->start;
610 prealloc->end = split - 1;
611 prealloc->state = orig->state;
614 parent = &orig->rb_node;
617 struct tree_entry *entry;
620 entry = rb_entry(parent, struct tree_entry, rb_node);
622 if (prealloc->end < entry->start) {
623 node = &(*node)->rb_left;
624 } else if (prealloc->end > entry->end) {
625 node = &(*node)->rb_right;
627 free_extent_state(prealloc);
632 rb_link_node(&prealloc->rb_node, parent, node);
633 rb_insert_color(&prealloc->rb_node, &tree->state);
638 static struct extent_state *next_state(struct extent_state *state)
640 struct rb_node *next = rb_next(&state->rb_node);
642 return rb_entry(next, struct extent_state, rb_node);
648 * utility function to clear some bits in an extent state struct.
649 * it will optionally wake up anyone waiting on this state (wake == 1).
651 * If no bits are set on the state struct after clearing things, the
652 * struct is freed and removed from the tree
654 static struct extent_state *clear_state_bit(struct extent_io_tree *tree,
655 struct extent_state *state,
657 struct extent_changeset *changeset)
659 struct extent_state *next;
660 u32 bits_to_clear = bits & ~EXTENT_CTLBITS;
663 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
664 u64 range = state->end - state->start + 1;
665 WARN_ON(range > tree->dirty_bytes);
666 tree->dirty_bytes -= range;
669 if (tree->private_data && is_data_inode(tree->private_data))
670 btrfs_clear_delalloc_extent(tree->private_data, state, bits);
672 ret = add_extent_changeset(state, bits_to_clear, changeset, 0);
674 state->state &= ~bits_to_clear;
677 if (state->state == 0) {
678 next = next_state(state);
679 if (extent_state_in_tree(state)) {
680 rb_erase(&state->rb_node, &tree->state);
681 RB_CLEAR_NODE(&state->rb_node);
682 free_extent_state(state);
687 merge_state(tree, state);
688 next = next_state(state);
693 static struct extent_state *
694 alloc_extent_state_atomic(struct extent_state *prealloc)
697 prealloc = alloc_extent_state(GFP_ATOMIC);
702 static void extent_io_tree_panic(struct extent_io_tree *tree, int err)
704 btrfs_panic(tree->fs_info, err,
705 "locking error: extent tree was modified by another thread while locked");
709 * clear some bits on a range in the tree. This may require splitting
710 * or inserting elements in the tree, so the gfp mask is used to
711 * indicate which allocations or sleeping are allowed.
713 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
714 * the given range from the tree regardless of state (ie for truncate).
716 * the range [start, end] is inclusive.
718 * This takes the tree lock, and returns 0 on success and < 0 on error.
720 int __clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
721 u32 bits, int wake, int delete,
722 struct extent_state **cached_state,
723 gfp_t mask, struct extent_changeset *changeset)
725 struct extent_state *state;
726 struct extent_state *cached;
727 struct extent_state *prealloc = NULL;
728 struct rb_node *node;
733 btrfs_debug_check_extent_io_range(tree, start, end);
734 trace_btrfs_clear_extent_bit(tree, start, end - start + 1, bits);
736 if (bits & EXTENT_DELALLOC)
737 bits |= EXTENT_NORESERVE;
740 bits |= ~EXTENT_CTLBITS;
742 if (bits & (EXTENT_LOCKED | EXTENT_BOUNDARY))
745 if (!prealloc && gfpflags_allow_blocking(mask)) {
747 * Don't care for allocation failure here because we might end
748 * up not needing the pre-allocated extent state at all, which
749 * is the case if we only have in the tree extent states that
750 * cover our input range and don't cover too any other range.
751 * If we end up needing a new extent state we allocate it later.
753 prealloc = alloc_extent_state(mask);
756 spin_lock(&tree->lock);
758 cached = *cached_state;
761 *cached_state = NULL;
765 if (cached && extent_state_in_tree(cached) &&
766 cached->start <= start && cached->end > start) {
768 refcount_dec(&cached->refs);
773 free_extent_state(cached);
776 * this search will find the extents that end after
779 node = tree_search(tree, start);
782 state = rb_entry(node, struct extent_state, rb_node);
784 if (state->start > end)
786 WARN_ON(state->end < start);
787 last_end = state->end;
789 /* the state doesn't have the wanted bits, go ahead */
790 if (!(state->state & bits)) {
791 state = next_state(state);
796 * | ---- desired range ---- |
798 * | ------------- state -------------- |
800 * We need to split the extent we found, and may flip
801 * bits on second half.
803 * If the extent we found extends past our range, we
804 * just split and search again. It'll get split again
805 * the next time though.
807 * If the extent we found is inside our range, we clear
808 * the desired bit on it.
811 if (state->start < start) {
812 prealloc = alloc_extent_state_atomic(prealloc);
814 err = split_state(tree, state, prealloc, start);
816 extent_io_tree_panic(tree, err);
821 if (state->end <= end) {
822 state = clear_state_bit(tree, state, bits, wake, changeset);
828 * | ---- desired range ---- |
830 * We need to split the extent, and clear the bit
833 if (state->start <= end && state->end > end) {
834 prealloc = alloc_extent_state_atomic(prealloc);
836 err = split_state(tree, state, prealloc, end + 1);
838 extent_io_tree_panic(tree, err);
843 clear_state_bit(tree, prealloc, bits, wake, changeset);
849 state = clear_state_bit(tree, state, bits, wake, changeset);
851 if (last_end == (u64)-1)
853 start = last_end + 1;
854 if (start <= end && state && !need_resched())
860 spin_unlock(&tree->lock);
861 if (gfpflags_allow_blocking(mask))
866 spin_unlock(&tree->lock);
868 free_extent_state(prealloc);
874 static void wait_on_state(struct extent_io_tree *tree,
875 struct extent_state *state)
876 __releases(tree->lock)
877 __acquires(tree->lock)
880 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
881 spin_unlock(&tree->lock);
883 spin_lock(&tree->lock);
884 finish_wait(&state->wq, &wait);
888 * waits for one or more bits to clear on a range in the state tree.
889 * The range [start, end] is inclusive.
890 * The tree lock is taken by this function
892 static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
895 struct extent_state *state;
896 struct rb_node *node;
898 btrfs_debug_check_extent_io_range(tree, start, end);
900 spin_lock(&tree->lock);
904 * this search will find all the extents that end after
907 node = tree_search(tree, start);
912 state = rb_entry(node, struct extent_state, rb_node);
914 if (state->start > end)
917 if (state->state & bits) {
918 start = state->start;
919 refcount_inc(&state->refs);
920 wait_on_state(tree, state);
921 free_extent_state(state);
924 start = state->end + 1;
929 if (!cond_resched_lock(&tree->lock)) {
930 node = rb_next(node);
935 spin_unlock(&tree->lock);
938 static void set_state_bits(struct extent_io_tree *tree,
939 struct extent_state *state,
940 u32 bits, struct extent_changeset *changeset)
942 u32 bits_to_set = bits & ~EXTENT_CTLBITS;
945 if (tree->private_data && is_data_inode(tree->private_data))
946 btrfs_set_delalloc_extent(tree->private_data, state, bits);
948 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
949 u64 range = state->end - state->start + 1;
950 tree->dirty_bytes += range;
952 ret = add_extent_changeset(state, bits_to_set, changeset, 1);
954 state->state |= bits_to_set;
957 static void cache_state_if_flags(struct extent_state *state,
958 struct extent_state **cached_ptr,
961 if (cached_ptr && !(*cached_ptr)) {
962 if (!flags || (state->state & flags)) {
964 refcount_inc(&state->refs);
969 static void cache_state(struct extent_state *state,
970 struct extent_state **cached_ptr)
972 return cache_state_if_flags(state, cached_ptr,
973 EXTENT_LOCKED | EXTENT_BOUNDARY);
977 * set some bits on a range in the tree. This may require allocations or
978 * sleeping, so the gfp mask is used to indicate what is allowed.
980 * If any of the exclusive bits are set, this will fail with -EEXIST if some
981 * part of the range already has the desired bits set. The start of the
982 * existing range is returned in failed_start in this case.
984 * [start, end] is inclusive This takes the tree lock.
986 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, u32 bits,
987 u32 exclusive_bits, u64 *failed_start,
988 struct extent_state **cached_state, gfp_t mask,
989 struct extent_changeset *changeset)
991 struct extent_state *state;
992 struct extent_state *prealloc = NULL;
993 struct rb_node *node;
995 struct rb_node *parent;
1000 btrfs_debug_check_extent_io_range(tree, start, end);
1001 trace_btrfs_set_extent_bit(tree, start, end - start + 1, bits);
1004 ASSERT(failed_start);
1006 ASSERT(failed_start == NULL);
1008 if (!prealloc && gfpflags_allow_blocking(mask)) {
1010 * Don't care for allocation failure here because we might end
1011 * up not needing the pre-allocated extent state at all, which
1012 * is the case if we only have in the tree extent states that
1013 * cover our input range and don't cover too any other range.
1014 * If we end up needing a new extent state we allocate it later.
1016 prealloc = alloc_extent_state(mask);
1019 spin_lock(&tree->lock);
1020 if (cached_state && *cached_state) {
1021 state = *cached_state;
1022 if (state->start <= start && state->end > start &&
1023 extent_state_in_tree(state)) {
1024 node = &state->rb_node;
1029 * this search will find all the extents that end after
1032 node = tree_search_for_insert(tree, start, &p, &parent);
1034 prealloc = alloc_extent_state_atomic(prealloc);
1036 prealloc->start = start;
1037 prealloc->end = end;
1038 insert_state_fast(tree, prealloc, p, parent, bits, changeset);
1039 cache_state(prealloc, cached_state);
1043 state = rb_entry(node, struct extent_state, rb_node);
1045 last_start = state->start;
1046 last_end = state->end;
1049 * | ---- desired range ---- |
1052 * Just lock what we found and keep going
1054 if (state->start == start && state->end <= end) {
1055 if (state->state & exclusive_bits) {
1056 *failed_start = state->start;
1061 set_state_bits(tree, state, bits, changeset);
1062 cache_state(state, cached_state);
1063 merge_state(tree, state);
1064 if (last_end == (u64)-1)
1066 start = last_end + 1;
1067 state = next_state(state);
1068 if (start < end && state && state->start == start &&
1075 * | ---- desired range ---- |
1078 * | ------------- state -------------- |
1080 * We need to split the extent we found, and may flip bits on
1083 * If the extent we found extends past our
1084 * range, we just split and search again. It'll get split
1085 * again the next time though.
1087 * If the extent we found is inside our range, we set the
1088 * desired bit on it.
1090 if (state->start < start) {
1091 if (state->state & exclusive_bits) {
1092 *failed_start = start;
1098 * If this extent already has all the bits we want set, then
1099 * skip it, not necessary to split it or do anything with it.
1101 if ((state->state & bits) == bits) {
1102 start = state->end + 1;
1103 cache_state(state, cached_state);
1107 prealloc = alloc_extent_state_atomic(prealloc);
1109 err = split_state(tree, state, prealloc, start);
1111 extent_io_tree_panic(tree, err);
1116 if (state->end <= end) {
1117 set_state_bits(tree, state, bits, changeset);
1118 cache_state(state, cached_state);
1119 merge_state(tree, state);
1120 if (last_end == (u64)-1)
1122 start = last_end + 1;
1123 state = next_state(state);
1124 if (start < end && state && state->start == start &&
1131 * | ---- desired range ---- |
1132 * | state | or | state |
1134 * There's a hole, we need to insert something in it and
1135 * ignore the extent we found.
1137 if (state->start > start) {
1139 if (end < last_start)
1142 this_end = last_start - 1;
1144 prealloc = alloc_extent_state_atomic(prealloc);
1148 * Avoid to free 'prealloc' if it can be merged with
1151 prealloc->start = start;
1152 prealloc->end = this_end;
1153 err = insert_state(tree, prealloc, NULL, NULL, bits, changeset);
1155 extent_io_tree_panic(tree, err);
1157 cache_state(prealloc, cached_state);
1159 start = this_end + 1;
1163 * | ---- desired range ---- |
1165 * We need to split the extent, and set the bit
1168 if (state->start <= end && state->end > end) {
1169 if (state->state & exclusive_bits) {
1170 *failed_start = start;
1175 prealloc = alloc_extent_state_atomic(prealloc);
1177 err = split_state(tree, state, prealloc, end + 1);
1179 extent_io_tree_panic(tree, err);
1181 set_state_bits(tree, prealloc, bits, changeset);
1182 cache_state(prealloc, cached_state);
1183 merge_state(tree, prealloc);
1191 spin_unlock(&tree->lock);
1192 if (gfpflags_allow_blocking(mask))
1197 spin_unlock(&tree->lock);
1199 free_extent_state(prealloc);
1206 * convert_extent_bit - convert all bits in a given range from one bit to
1208 * @tree: the io tree to search
1209 * @start: the start offset in bytes
1210 * @end: the end offset in bytes (inclusive)
1211 * @bits: the bits to set in this range
1212 * @clear_bits: the bits to clear in this range
1213 * @cached_state: state that we're going to cache
1215 * This will go through and set bits for the given range. If any states exist
1216 * already in this range they are set with the given bit and cleared of the
1217 * clear_bits. This is only meant to be used by things that are mergeable, ie
1218 * converting from say DELALLOC to DIRTY. This is not meant to be used with
1219 * boundary bits like LOCK.
1221 * All allocations are done with GFP_NOFS.
1223 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1224 u32 bits, u32 clear_bits,
1225 struct extent_state **cached_state)
1227 struct extent_state *state;
1228 struct extent_state *prealloc = NULL;
1229 struct rb_node *node;
1231 struct rb_node *parent;
1235 bool first_iteration = true;
1237 btrfs_debug_check_extent_io_range(tree, start, end);
1238 trace_btrfs_convert_extent_bit(tree, start, end - start + 1, bits,
1244 * Best effort, don't worry if extent state allocation fails
1245 * here for the first iteration. We might have a cached state
1246 * that matches exactly the target range, in which case no
1247 * extent state allocations are needed. We'll only know this
1248 * after locking the tree.
1250 prealloc = alloc_extent_state(GFP_NOFS);
1251 if (!prealloc && !first_iteration)
1255 spin_lock(&tree->lock);
1256 if (cached_state && *cached_state) {
1257 state = *cached_state;
1258 if (state->start <= start && state->end > start &&
1259 extent_state_in_tree(state)) {
1260 node = &state->rb_node;
1266 * this search will find all the extents that end after
1269 node = tree_search_for_insert(tree, start, &p, &parent);
1271 prealloc = alloc_extent_state_atomic(prealloc);
1276 prealloc->start = start;
1277 prealloc->end = end;
1278 insert_state_fast(tree, prealloc, p, parent, bits, NULL);
1279 cache_state(prealloc, cached_state);
1283 state = rb_entry(node, struct extent_state, rb_node);
1285 last_start = state->start;
1286 last_end = state->end;
1289 * | ---- desired range ---- |
1292 * Just lock what we found and keep going
1294 if (state->start == start && state->end <= end) {
1295 set_state_bits(tree, state, bits, NULL);
1296 cache_state(state, cached_state);
1297 state = clear_state_bit(tree, state, clear_bits, 0, NULL);
1298 if (last_end == (u64)-1)
1300 start = last_end + 1;
1301 if (start < end && state && state->start == start &&
1308 * | ---- desired range ---- |
1311 * | ------------- state -------------- |
1313 * We need to split the extent we found, and may flip bits on
1316 * If the extent we found extends past our
1317 * range, we just split and search again. It'll get split
1318 * again the next time though.
1320 * If the extent we found is inside our range, we set the
1321 * desired bit on it.
1323 if (state->start < start) {
1324 prealloc = alloc_extent_state_atomic(prealloc);
1329 err = split_state(tree, state, prealloc, start);
1331 extent_io_tree_panic(tree, err);
1335 if (state->end <= end) {
1336 set_state_bits(tree, state, bits, NULL);
1337 cache_state(state, cached_state);
1338 state = clear_state_bit(tree, state, clear_bits, 0, NULL);
1339 if (last_end == (u64)-1)
1341 start = last_end + 1;
1342 if (start < end && state && state->start == start &&
1349 * | ---- desired range ---- |
1350 * | state | or | state |
1352 * There's a hole, we need to insert something in it and
1353 * ignore the extent we found.
1355 if (state->start > start) {
1357 if (end < last_start)
1360 this_end = last_start - 1;
1362 prealloc = alloc_extent_state_atomic(prealloc);
1369 * Avoid to free 'prealloc' if it can be merged with
1372 prealloc->start = start;
1373 prealloc->end = this_end;
1374 err = insert_state(tree, prealloc, NULL, NULL, bits, NULL);
1376 extent_io_tree_panic(tree, err);
1377 cache_state(prealloc, cached_state);
1379 start = this_end + 1;
1383 * | ---- desired range ---- |
1385 * We need to split the extent, and set the bit
1388 if (state->start <= end && state->end > end) {
1389 prealloc = alloc_extent_state_atomic(prealloc);
1395 err = split_state(tree, state, prealloc, end + 1);
1397 extent_io_tree_panic(tree, err);
1399 set_state_bits(tree, prealloc, bits, NULL);
1400 cache_state(prealloc, cached_state);
1401 clear_state_bit(tree, prealloc, clear_bits, 0, NULL);
1409 spin_unlock(&tree->lock);
1411 first_iteration = false;
1415 spin_unlock(&tree->lock);
1417 free_extent_state(prealloc);
1422 /* wrappers around set/clear extent bit */
1423 int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1424 u32 bits, struct extent_changeset *changeset)
1427 * We don't support EXTENT_LOCKED yet, as current changeset will
1428 * record any bits changed, so for EXTENT_LOCKED case, it will
1429 * either fail with -EEXIST or changeset will record the whole
1432 BUG_ON(bits & EXTENT_LOCKED);
1434 return set_extent_bit(tree, start, end, bits, 0, NULL, NULL, GFP_NOFS,
1438 int set_extent_bits_nowait(struct extent_io_tree *tree, u64 start, u64 end,
1441 return set_extent_bit(tree, start, end, bits, 0, NULL, NULL,
1445 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1446 u32 bits, int wake, int delete,
1447 struct extent_state **cached)
1449 return __clear_extent_bit(tree, start, end, bits, wake, delete,
1450 cached, GFP_NOFS, NULL);
1453 int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1454 u32 bits, struct extent_changeset *changeset)
1457 * Don't support EXTENT_LOCKED case, same reason as
1458 * set_record_extent_bits().
1460 BUG_ON(bits & EXTENT_LOCKED);
1462 return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, GFP_NOFS,
1467 * either insert or lock state struct between start and end use mask to tell
1468 * us if waiting is desired.
1470 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1471 struct extent_state **cached_state)
1477 err = set_extent_bit(tree, start, end, EXTENT_LOCKED,
1478 EXTENT_LOCKED, &failed_start,
1479 cached_state, GFP_NOFS, NULL);
1480 if (err == -EEXIST) {
1481 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1482 start = failed_start;
1485 WARN_ON(start > end);
1490 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1495 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1496 &failed_start, NULL, GFP_NOFS, NULL);
1497 if (err == -EEXIST) {
1498 if (failed_start > start)
1499 clear_extent_bit(tree, start, failed_start - 1,
1500 EXTENT_LOCKED, 1, 0, NULL);
1506 void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
1508 unsigned long index = start >> PAGE_SHIFT;
1509 unsigned long end_index = end >> PAGE_SHIFT;
1512 while (index <= end_index) {
1513 page = find_get_page(inode->i_mapping, index);
1514 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1515 clear_page_dirty_for_io(page);
1521 void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
1523 struct address_space *mapping = inode->i_mapping;
1524 unsigned long index = start >> PAGE_SHIFT;
1525 unsigned long end_index = end >> PAGE_SHIFT;
1526 struct folio *folio;
1528 while (index <= end_index) {
1529 folio = filemap_get_folio(mapping, index);
1530 filemap_dirty_folio(mapping, folio);
1531 folio_account_redirty(folio);
1532 index += folio_nr_pages(folio);
1537 /* find the first state struct with 'bits' set after 'start', and
1538 * return it. tree->lock must be held. NULL will returned if
1539 * nothing was found after 'start'
1541 static struct extent_state *
1542 find_first_extent_bit_state(struct extent_io_tree *tree, u64 start, u32 bits)
1544 struct rb_node *node;
1545 struct extent_state *state;
1548 * this search will find all the extents that end after
1551 node = tree_search(tree, start);
1556 state = rb_entry(node, struct extent_state, rb_node);
1557 if (state->end >= start && (state->state & bits))
1560 node = rb_next(node);
1569 * Find the first offset in the io tree with one or more @bits set.
1571 * Note: If there are multiple bits set in @bits, any of them will match.
1573 * Return 0 if we find something, and update @start_ret and @end_ret.
1574 * Return 1 if we found nothing.
1576 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1577 u64 *start_ret, u64 *end_ret, u32 bits,
1578 struct extent_state **cached_state)
1580 struct extent_state *state;
1583 spin_lock(&tree->lock);
1584 if (cached_state && *cached_state) {
1585 state = *cached_state;
1586 if (state->end == start - 1 && extent_state_in_tree(state)) {
1587 while ((state = next_state(state)) != NULL) {
1588 if (state->state & bits)
1591 free_extent_state(*cached_state);
1592 *cached_state = NULL;
1595 free_extent_state(*cached_state);
1596 *cached_state = NULL;
1599 state = find_first_extent_bit_state(tree, start, bits);
1602 cache_state_if_flags(state, cached_state, 0);
1603 *start_ret = state->start;
1604 *end_ret = state->end;
1608 spin_unlock(&tree->lock);
1613 * Find a contiguous area of bits
1615 * @tree: io tree to check
1616 * @start: offset to start the search from
1617 * @start_ret: the first offset we found with the bits set
1618 * @end_ret: the final contiguous range of the bits that were set
1619 * @bits: bits to look for
1621 * set_extent_bit and clear_extent_bit can temporarily split contiguous ranges
1622 * to set bits appropriately, and then merge them again. During this time it
1623 * will drop the tree->lock, so use this helper if you want to find the actual
1624 * contiguous area for given bits. We will search to the first bit we find, and
1625 * then walk down the tree until we find a non-contiguous area. The area
1626 * returned will be the full contiguous area with the bits set.
1628 int find_contiguous_extent_bit(struct extent_io_tree *tree, u64 start,
1629 u64 *start_ret, u64 *end_ret, u32 bits)
1631 struct extent_state *state;
1634 spin_lock(&tree->lock);
1635 state = find_first_extent_bit_state(tree, start, bits);
1637 *start_ret = state->start;
1638 *end_ret = state->end;
1639 while ((state = next_state(state)) != NULL) {
1640 if (state->start > (*end_ret + 1))
1642 *end_ret = state->end;
1646 spin_unlock(&tree->lock);
1651 * Find the first range that has @bits not set. This range could start before
1654 * @tree: the tree to search
1655 * @start: offset at/after which the found extent should start
1656 * @start_ret: records the beginning of the range
1657 * @end_ret: records the end of the range (inclusive)
1658 * @bits: the set of bits which must be unset
1660 * Since unallocated range is also considered one which doesn't have the bits
1661 * set it's possible that @end_ret contains -1, this happens in case the range
1662 * spans (last_range_end, end of device]. In this case it's up to the caller to
1663 * trim @end_ret to the appropriate size.
1665 void find_first_clear_extent_bit(struct extent_io_tree *tree, u64 start,
1666 u64 *start_ret, u64 *end_ret, u32 bits)
1668 struct extent_state *state;
1669 struct rb_node *node, *prev = NULL, *next;
1671 spin_lock(&tree->lock);
1673 /* Find first extent with bits cleared */
1675 node = __etree_search(tree, start, &next, &prev, NULL, NULL);
1676 if (!node && !next && !prev) {
1678 * Tree is completely empty, send full range and let
1679 * caller deal with it
1684 } else if (!node && !next) {
1686 * We are past the last allocated chunk, set start at
1687 * the end of the last extent.
1689 state = rb_entry(prev, struct extent_state, rb_node);
1690 *start_ret = state->end + 1;
1697 * At this point 'node' either contains 'start' or start is
1700 state = rb_entry(node, struct extent_state, rb_node);
1702 if (in_range(start, state->start, state->end - state->start + 1)) {
1703 if (state->state & bits) {
1705 * |--range with bits sets--|
1709 start = state->end + 1;
1712 * 'start' falls within a range that doesn't
1713 * have the bits set, so take its start as
1714 * the beginning of the desired range
1716 * |--range with bits cleared----|
1720 *start_ret = state->start;
1725 * |---prev range---|---hole/unset---|---node range---|
1731 * |---hole/unset--||--first node--|
1736 state = rb_entry(prev, struct extent_state,
1738 *start_ret = state->end + 1;
1747 * Find the longest stretch from start until an entry which has the
1751 state = rb_entry(node, struct extent_state, rb_node);
1752 if (state->end >= start && !(state->state & bits)) {
1753 *end_ret = state->end;
1755 *end_ret = state->start - 1;
1759 node = rb_next(node);
1764 spin_unlock(&tree->lock);
1768 * find a contiguous range of bytes in the file marked as delalloc, not
1769 * more than 'max_bytes'. start and end are used to return the range,
1771 * true is returned if we find something, false if nothing was in the tree
1773 bool btrfs_find_delalloc_range(struct extent_io_tree *tree, u64 *start,
1774 u64 *end, u64 max_bytes,
1775 struct extent_state **cached_state)
1777 struct rb_node *node;
1778 struct extent_state *state;
1779 u64 cur_start = *start;
1781 u64 total_bytes = 0;
1783 spin_lock(&tree->lock);
1786 * this search will find all the extents that end after
1789 node = tree_search(tree, cur_start);
1796 state = rb_entry(node, struct extent_state, rb_node);
1797 if (found && (state->start != cur_start ||
1798 (state->state & EXTENT_BOUNDARY))) {
1801 if (!(state->state & EXTENT_DELALLOC)) {
1807 *start = state->start;
1808 *cached_state = state;
1809 refcount_inc(&state->refs);
1813 cur_start = state->end + 1;
1814 node = rb_next(node);
1815 total_bytes += state->end - state->start + 1;
1816 if (total_bytes >= max_bytes)
1822 spin_unlock(&tree->lock);
1827 * Process one page for __process_pages_contig().
1829 * Return >0 if we hit @page == @locked_page.
1830 * Return 0 if we updated the page status.
1831 * Return -EGAIN if the we need to try again.
1832 * (For PAGE_LOCK case but got dirty page or page not belong to mapping)
1834 static int process_one_page(struct btrfs_fs_info *fs_info,
1835 struct address_space *mapping,
1836 struct page *page, struct page *locked_page,
1837 unsigned long page_ops, u64 start, u64 end)
1841 ASSERT(end + 1 - start != 0 && end + 1 - start < U32_MAX);
1842 len = end + 1 - start;
1844 if (page_ops & PAGE_SET_ORDERED)
1845 btrfs_page_clamp_set_ordered(fs_info, page, start, len);
1846 if (page_ops & PAGE_SET_ERROR)
1847 btrfs_page_clamp_set_error(fs_info, page, start, len);
1848 if (page_ops & PAGE_START_WRITEBACK) {
1849 btrfs_page_clamp_clear_dirty(fs_info, page, start, len);
1850 btrfs_page_clamp_set_writeback(fs_info, page, start, len);
1852 if (page_ops & PAGE_END_WRITEBACK)
1853 btrfs_page_clamp_clear_writeback(fs_info, page, start, len);
1855 if (page == locked_page)
1858 if (page_ops & PAGE_LOCK) {
1861 ret = btrfs_page_start_writer_lock(fs_info, page, start, len);
1864 if (!PageDirty(page) || page->mapping != mapping) {
1865 btrfs_page_end_writer_lock(fs_info, page, start, len);
1869 if (page_ops & PAGE_UNLOCK)
1870 btrfs_page_end_writer_lock(fs_info, page, start, len);
1874 static int __process_pages_contig(struct address_space *mapping,
1875 struct page *locked_page,
1876 u64 start, u64 end, unsigned long page_ops,
1879 struct btrfs_fs_info *fs_info = btrfs_sb(mapping->host->i_sb);
1880 pgoff_t start_index = start >> PAGE_SHIFT;
1881 pgoff_t end_index = end >> PAGE_SHIFT;
1882 pgoff_t index = start_index;
1883 unsigned long nr_pages = end_index - start_index + 1;
1884 unsigned long pages_processed = 0;
1885 struct page *pages[16];
1889 if (page_ops & PAGE_LOCK) {
1890 ASSERT(page_ops == PAGE_LOCK);
1891 ASSERT(processed_end && *processed_end == start);
1894 if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0)
1895 mapping_set_error(mapping, -EIO);
1897 while (nr_pages > 0) {
1900 found_pages = find_get_pages_contig(mapping, index,
1901 min_t(unsigned long,
1902 nr_pages, ARRAY_SIZE(pages)), pages);
1903 if (found_pages == 0) {
1905 * Only if we're going to lock these pages, we can find
1906 * nothing at @index.
1908 ASSERT(page_ops & PAGE_LOCK);
1913 for (i = 0; i < found_pages; i++) {
1916 process_ret = process_one_page(fs_info, mapping,
1917 pages[i], locked_page, page_ops,
1919 if (process_ret < 0) {
1920 for (; i < found_pages; i++)
1928 nr_pages -= found_pages;
1929 index += found_pages;
1933 if (err && processed_end) {
1935 * Update @processed_end. I know this is awful since it has
1936 * two different return value patterns (inclusive vs exclusive).
1938 * But the exclusive pattern is necessary if @start is 0, or we
1939 * underflow and check against processed_end won't work as
1942 if (pages_processed)
1943 *processed_end = min(end,
1944 ((u64)(start_index + pages_processed) << PAGE_SHIFT) - 1);
1946 *processed_end = start;
1951 static noinline void __unlock_for_delalloc(struct inode *inode,
1952 struct page *locked_page,
1955 unsigned long index = start >> PAGE_SHIFT;
1956 unsigned long end_index = end >> PAGE_SHIFT;
1958 ASSERT(locked_page);
1959 if (index == locked_page->index && end_index == index)
1962 __process_pages_contig(inode->i_mapping, locked_page, start, end,
1966 static noinline int lock_delalloc_pages(struct inode *inode,
1967 struct page *locked_page,
1971 unsigned long index = delalloc_start >> PAGE_SHIFT;
1972 unsigned long end_index = delalloc_end >> PAGE_SHIFT;
1973 u64 processed_end = delalloc_start;
1976 ASSERT(locked_page);
1977 if (index == locked_page->index && index == end_index)
1980 ret = __process_pages_contig(inode->i_mapping, locked_page, delalloc_start,
1981 delalloc_end, PAGE_LOCK, &processed_end);
1982 if (ret == -EAGAIN && processed_end > delalloc_start)
1983 __unlock_for_delalloc(inode, locked_page, delalloc_start,
1989 * Find and lock a contiguous range of bytes in the file marked as delalloc, no
1990 * more than @max_bytes.
1992 * @start: The original start bytenr to search.
1993 * Will store the extent range start bytenr.
1994 * @end: The original end bytenr of the search range
1995 * Will store the extent range end bytenr.
1997 * Return true if we find a delalloc range which starts inside the original
1998 * range, and @start/@end will store the delalloc range start/end.
2000 * Return false if we can't find any delalloc range which starts inside the
2001 * original range, and @start/@end will be the non-delalloc range start/end.
2004 noinline_for_stack bool find_lock_delalloc_range(struct inode *inode,
2005 struct page *locked_page, u64 *start,
2008 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2009 const u64 orig_start = *start;
2010 const u64 orig_end = *end;
2011 u64 max_bytes = BTRFS_MAX_EXTENT_SIZE;
2015 struct extent_state *cached_state = NULL;
2019 /* Caller should pass a valid @end to indicate the search range end */
2020 ASSERT(orig_end > orig_start);
2022 /* The range should at least cover part of the page */
2023 ASSERT(!(orig_start >= page_offset(locked_page) + PAGE_SIZE ||
2024 orig_end <= page_offset(locked_page)));
2026 /* step one, find a bunch of delalloc bytes starting at start */
2027 delalloc_start = *start;
2029 found = btrfs_find_delalloc_range(tree, &delalloc_start, &delalloc_end,
2030 max_bytes, &cached_state);
2031 if (!found || delalloc_end <= *start || delalloc_start > orig_end) {
2032 *start = delalloc_start;
2034 /* @delalloc_end can be -1, never go beyond @orig_end */
2035 *end = min(delalloc_end, orig_end);
2036 free_extent_state(cached_state);
2041 * start comes from the offset of locked_page. We have to lock
2042 * pages in order, so we can't process delalloc bytes before
2045 if (delalloc_start < *start)
2046 delalloc_start = *start;
2049 * make sure to limit the number of pages we try to lock down
2051 if (delalloc_end + 1 - delalloc_start > max_bytes)
2052 delalloc_end = delalloc_start + max_bytes - 1;
2054 /* step two, lock all the pages after the page that has start */
2055 ret = lock_delalloc_pages(inode, locked_page,
2056 delalloc_start, delalloc_end);
2057 ASSERT(!ret || ret == -EAGAIN);
2058 if (ret == -EAGAIN) {
2059 /* some of the pages are gone, lets avoid looping by
2060 * shortening the size of the delalloc range we're searching
2062 free_extent_state(cached_state);
2063 cached_state = NULL;
2065 max_bytes = PAGE_SIZE;
2074 /* step three, lock the state bits for the whole range */
2075 lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state);
2077 /* then test to make sure it is all still delalloc */
2078 ret = test_range_bit(tree, delalloc_start, delalloc_end,
2079 EXTENT_DELALLOC, 1, cached_state);
2081 unlock_extent_cached(tree, delalloc_start, delalloc_end,
2083 __unlock_for_delalloc(inode, locked_page,
2084 delalloc_start, delalloc_end);
2088 free_extent_state(cached_state);
2089 *start = delalloc_start;
2090 *end = delalloc_end;
2095 void extent_clear_unlock_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2096 struct page *locked_page,
2097 u32 clear_bits, unsigned long page_ops)
2099 clear_extent_bit(&inode->io_tree, start, end, clear_bits, 1, 0, NULL);
2101 __process_pages_contig(inode->vfs_inode.i_mapping, locked_page,
2102 start, end, page_ops, NULL);
2106 * count the number of bytes in the tree that have a given bit(s)
2107 * set. This can be fairly slow, except for EXTENT_DIRTY which is
2108 * cached. The total number found is returned.
2110 u64 count_range_bits(struct extent_io_tree *tree,
2111 u64 *start, u64 search_end, u64 max_bytes,
2112 u32 bits, int contig)
2114 struct rb_node *node;
2115 struct extent_state *state;
2116 u64 cur_start = *start;
2117 u64 total_bytes = 0;
2121 if (WARN_ON(search_end <= cur_start))
2124 spin_lock(&tree->lock);
2125 if (cur_start == 0 && bits == EXTENT_DIRTY) {
2126 total_bytes = tree->dirty_bytes;
2130 * this search will find all the extents that end after
2133 node = tree_search(tree, cur_start);
2138 state = rb_entry(node, struct extent_state, rb_node);
2139 if (state->start > search_end)
2141 if (contig && found && state->start > last + 1)
2143 if (state->end >= cur_start && (state->state & bits) == bits) {
2144 total_bytes += min(search_end, state->end) + 1 -
2145 max(cur_start, state->start);
2146 if (total_bytes >= max_bytes)
2149 *start = max(cur_start, state->start);
2153 } else if (contig && found) {
2156 node = rb_next(node);
2161 spin_unlock(&tree->lock);
2166 * set the private field for a given byte offset in the tree. If there isn't
2167 * an extent_state there already, this does nothing.
2169 int set_state_failrec(struct extent_io_tree *tree, u64 start,
2170 struct io_failure_record *failrec)
2172 struct rb_node *node;
2173 struct extent_state *state;
2176 spin_lock(&tree->lock);
2178 * this search will find all the extents that end after
2181 node = tree_search(tree, start);
2186 state = rb_entry(node, struct extent_state, rb_node);
2187 if (state->start != start) {
2191 state->failrec = failrec;
2193 spin_unlock(&tree->lock);
2197 struct io_failure_record *get_state_failrec(struct extent_io_tree *tree, u64 start)
2199 struct rb_node *node;
2200 struct extent_state *state;
2201 struct io_failure_record *failrec;
2203 spin_lock(&tree->lock);
2205 * this search will find all the extents that end after
2208 node = tree_search(tree, start);
2210 failrec = ERR_PTR(-ENOENT);
2213 state = rb_entry(node, struct extent_state, rb_node);
2214 if (state->start != start) {
2215 failrec = ERR_PTR(-ENOENT);
2219 failrec = state->failrec;
2221 spin_unlock(&tree->lock);
2226 * searches a range in the state tree for a given mask.
2227 * If 'filled' == 1, this returns 1 only if every extent in the tree
2228 * has the bits set. Otherwise, 1 is returned if any bit in the
2229 * range is found set.
2231 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
2232 u32 bits, int filled, struct extent_state *cached)
2234 struct extent_state *state = NULL;
2235 struct rb_node *node;
2238 spin_lock(&tree->lock);
2239 if (cached && extent_state_in_tree(cached) && cached->start <= start &&
2240 cached->end > start)
2241 node = &cached->rb_node;
2243 node = tree_search(tree, start);
2244 while (node && start <= end) {
2245 state = rb_entry(node, struct extent_state, rb_node);
2247 if (filled && state->start > start) {
2252 if (state->start > end)
2255 if (state->state & bits) {
2259 } else if (filled) {
2264 if (state->end == (u64)-1)
2267 start = state->end + 1;
2270 node = rb_next(node);
2277 spin_unlock(&tree->lock);
2281 int free_io_failure(struct extent_io_tree *failure_tree,
2282 struct extent_io_tree *io_tree,
2283 struct io_failure_record *rec)
2288 set_state_failrec(failure_tree, rec->start, NULL);
2289 ret = clear_extent_bits(failure_tree, rec->start,
2290 rec->start + rec->len - 1,
2291 EXTENT_LOCKED | EXTENT_DIRTY);
2295 ret = clear_extent_bits(io_tree, rec->start,
2296 rec->start + rec->len - 1,
2306 * this bypasses the standard btrfs submit functions deliberately, as
2307 * the standard behavior is to write all copies in a raid setup. here we only
2308 * want to write the one bad copy. so we do the mapping for ourselves and issue
2309 * submit_bio directly.
2310 * to avoid any synchronization issues, wait for the data after writing, which
2311 * actually prevents the read that triggered the error from finishing.
2312 * currently, there can be no more than two copies of every data bit. thus,
2313 * exactly one rewrite is required.
2315 static int repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start,
2316 u64 length, u64 logical, struct page *page,
2317 unsigned int pg_offset, int mirror_num)
2319 struct btrfs_device *dev;
2320 struct bio_vec bvec;
2324 struct btrfs_io_context *bioc = NULL;
2327 ASSERT(!(fs_info->sb->s_flags & SB_RDONLY));
2328 BUG_ON(!mirror_num);
2330 if (btrfs_repair_one_zone(fs_info, logical))
2333 map_length = length;
2336 * Avoid races with device replace and make sure our bioc has devices
2337 * associated to its stripes that don't go away while we are doing the
2338 * read repair operation.
2340 btrfs_bio_counter_inc_blocked(fs_info);
2341 if (btrfs_is_parity_mirror(fs_info, logical, length)) {
2343 * Note that we don't use BTRFS_MAP_WRITE because it's supposed
2344 * to update all raid stripes, but here we just want to correct
2345 * bad stripe, thus BTRFS_MAP_READ is abused to only get the bad
2346 * stripe's dev and sector.
2348 ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, logical,
2349 &map_length, &bioc, 0);
2351 goto out_counter_dec;
2352 ASSERT(bioc->mirror_num == 1);
2354 ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical,
2355 &map_length, &bioc, mirror_num);
2357 goto out_counter_dec;
2358 BUG_ON(mirror_num != bioc->mirror_num);
2361 sector = bioc->stripes[bioc->mirror_num - 1].physical >> 9;
2362 dev = bioc->stripes[bioc->mirror_num - 1].dev;
2363 btrfs_put_bioc(bioc);
2365 if (!dev || !dev->bdev ||
2366 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) {
2368 goto out_counter_dec;
2371 bio_init(&bio, dev->bdev, &bvec, 1, REQ_OP_WRITE | REQ_SYNC);
2372 bio.bi_iter.bi_sector = sector;
2373 __bio_add_page(&bio, page, length, pg_offset);
2375 btrfsic_check_bio(&bio);
2376 ret = submit_bio_wait(&bio);
2378 /* try to remap that extent elsewhere? */
2379 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
2380 goto out_bio_uninit;
2383 btrfs_info_rl_in_rcu(fs_info,
2384 "read error corrected: ino %llu off %llu (dev %s sector %llu)",
2386 rcu_str_deref(dev->name), sector);
2392 btrfs_bio_counter_dec(fs_info);
2396 int btrfs_repair_eb_io_failure(const struct extent_buffer *eb, int mirror_num)
2398 struct btrfs_fs_info *fs_info = eb->fs_info;
2399 u64 start = eb->start;
2400 int i, num_pages = num_extent_pages(eb);
2403 if (sb_rdonly(fs_info->sb))
2406 for (i = 0; i < num_pages; i++) {
2407 struct page *p = eb->pages[i];
2409 ret = repair_io_failure(fs_info, 0, start, PAGE_SIZE, start, p,
2410 start - page_offset(p), mirror_num);
2420 * each time an IO finishes, we do a fast check in the IO failure tree
2421 * to see if we need to process or clean up an io_failure_record
2423 int clean_io_failure(struct btrfs_fs_info *fs_info,
2424 struct extent_io_tree *failure_tree,
2425 struct extent_io_tree *io_tree, u64 start,
2426 struct page *page, u64 ino, unsigned int pg_offset)
2429 struct io_failure_record *failrec;
2430 struct extent_state *state;
2435 ret = count_range_bits(failure_tree, &private, (u64)-1, 1,
2440 failrec = get_state_failrec(failure_tree, start);
2441 if (IS_ERR(failrec))
2444 BUG_ON(!failrec->this_mirror);
2446 if (sb_rdonly(fs_info->sb))
2449 spin_lock(&io_tree->lock);
2450 state = find_first_extent_bit_state(io_tree,
2453 spin_unlock(&io_tree->lock);
2455 if (state && state->start <= failrec->start &&
2456 state->end >= failrec->start + failrec->len - 1) {
2457 num_copies = btrfs_num_copies(fs_info, failrec->logical,
2459 if (num_copies > 1) {
2460 repair_io_failure(fs_info, ino, start, failrec->len,
2461 failrec->logical, page, pg_offset,
2462 failrec->failed_mirror);
2467 free_io_failure(failure_tree, io_tree, failrec);
2473 * Can be called when
2474 * - hold extent lock
2475 * - under ordered extent
2476 * - the inode is freeing
2478 void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start, u64 end)
2480 struct extent_io_tree *failure_tree = &inode->io_failure_tree;
2481 struct io_failure_record *failrec;
2482 struct extent_state *state, *next;
2484 if (RB_EMPTY_ROOT(&failure_tree->state))
2487 spin_lock(&failure_tree->lock);
2488 state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY);
2490 if (state->start > end)
2493 ASSERT(state->end <= end);
2495 next = next_state(state);
2497 failrec = state->failrec;
2498 free_extent_state(state);
2503 spin_unlock(&failure_tree->lock);
2506 static struct io_failure_record *btrfs_get_io_failure_record(struct inode *inode,
2509 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2510 struct io_failure_record *failrec;
2511 struct extent_map *em;
2512 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2513 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2514 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2515 const u32 sectorsize = fs_info->sectorsize;
2519 failrec = get_state_failrec(failure_tree, start);
2520 if (!IS_ERR(failrec)) {
2521 btrfs_debug(fs_info,
2522 "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu",
2523 failrec->logical, failrec->start, failrec->len);
2525 * when data can be on disk more than twice, add to failrec here
2526 * (e.g. with a list for failed_mirror) to make
2527 * clean_io_failure() clean all those errors at once.
2533 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2535 return ERR_PTR(-ENOMEM);
2537 failrec->start = start;
2538 failrec->len = sectorsize;
2539 failrec->this_mirror = 0;
2540 failrec->compress_type = BTRFS_COMPRESS_NONE;
2542 read_lock(&em_tree->lock);
2543 em = lookup_extent_mapping(em_tree, start, failrec->len);
2545 read_unlock(&em_tree->lock);
2547 return ERR_PTR(-EIO);
2550 if (em->start > start || em->start + em->len <= start) {
2551 free_extent_map(em);
2554 read_unlock(&em_tree->lock);
2557 return ERR_PTR(-EIO);
2560 logical = start - em->start;
2561 logical = em->block_start + logical;
2562 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2563 logical = em->block_start;
2564 failrec->compress_type = em->compress_type;
2567 btrfs_debug(fs_info,
2568 "Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu",
2569 logical, start, failrec->len);
2571 failrec->logical = logical;
2572 free_extent_map(em);
2574 /* Set the bits in the private failure tree */
2575 ret = set_extent_bits(failure_tree, start, start + sectorsize - 1,
2576 EXTENT_LOCKED | EXTENT_DIRTY);
2578 ret = set_state_failrec(failure_tree, start, failrec);
2579 /* Set the bits in the inode's tree */
2580 ret = set_extent_bits(tree, start, start + sectorsize - 1,
2582 } else if (ret < 0) {
2584 return ERR_PTR(ret);
2590 static bool btrfs_check_repairable(struct inode *inode,
2591 struct io_failure_record *failrec,
2594 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2597 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
2598 if (num_copies == 1) {
2600 * we only have a single copy of the data, so don't bother with
2601 * all the retry and error correction code that follows. no
2602 * matter what the error is, it is very likely to persist.
2604 btrfs_debug(fs_info,
2605 "Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
2606 num_copies, failrec->this_mirror, failed_mirror);
2610 /* The failure record should only contain one sector */
2611 ASSERT(failrec->len == fs_info->sectorsize);
2614 * There are two premises:
2615 * a) deliver good data to the caller
2616 * b) correct the bad sectors on disk
2618 * Since we're only doing repair for one sector, we only need to get
2619 * a good copy of the failed sector and if we succeed, we have setup
2620 * everything for repair_io_failure to do the rest for us.
2622 ASSERT(failed_mirror);
2623 failrec->failed_mirror = failed_mirror;
2624 failrec->this_mirror++;
2625 if (failrec->this_mirror == failed_mirror)
2626 failrec->this_mirror++;
2628 if (failrec->this_mirror > num_copies) {
2629 btrfs_debug(fs_info,
2630 "Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
2631 num_copies, failrec->this_mirror, failed_mirror);
2638 int btrfs_repair_one_sector(struct inode *inode,
2639 struct bio *failed_bio, u32 bio_offset,
2640 struct page *page, unsigned int pgoff,
2641 u64 start, int failed_mirror,
2642 submit_bio_hook_t *submit_bio_hook)
2644 struct io_failure_record *failrec;
2645 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2646 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2647 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2648 struct btrfs_bio *failed_bbio = btrfs_bio(failed_bio);
2649 const int icsum = bio_offset >> fs_info->sectorsize_bits;
2650 struct bio *repair_bio;
2651 struct btrfs_bio *repair_bbio;
2653 btrfs_debug(fs_info,
2654 "repair read error: read error at %llu", start);
2656 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
2658 failrec = btrfs_get_io_failure_record(inode, start);
2659 if (IS_ERR(failrec))
2660 return PTR_ERR(failrec);
2663 if (!btrfs_check_repairable(inode, failrec, failed_mirror)) {
2664 free_io_failure(failure_tree, tree, failrec);
2668 repair_bio = btrfs_bio_alloc(1);
2669 repair_bbio = btrfs_bio(repair_bio);
2670 repair_bbio->file_offset = start;
2671 repair_bio->bi_opf = REQ_OP_READ;
2672 repair_bio->bi_end_io = failed_bio->bi_end_io;
2673 repair_bio->bi_iter.bi_sector = failrec->logical >> 9;
2674 repair_bio->bi_private = failed_bio->bi_private;
2676 if (failed_bbio->csum) {
2677 const u32 csum_size = fs_info->csum_size;
2679 repair_bbio->csum = repair_bbio->csum_inline;
2680 memcpy(repair_bbio->csum,
2681 failed_bbio->csum + csum_size * icsum, csum_size);
2684 bio_add_page(repair_bio, page, failrec->len, pgoff);
2685 repair_bbio->iter = repair_bio->bi_iter;
2687 btrfs_debug(btrfs_sb(inode->i_sb),
2688 "repair read error: submitting new read to mirror %d",
2689 failrec->this_mirror);
2692 * At this point we have a bio, so any errors from submit_bio_hook()
2693 * will be handled by the endio on the repair_bio, so we can't return an
2696 submit_bio_hook(inode, repair_bio, failrec->this_mirror, failrec->compress_type);
2700 static void end_page_read(struct page *page, bool uptodate, u64 start, u32 len)
2702 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
2704 ASSERT(page_offset(page) <= start &&
2705 start + len <= page_offset(page) + PAGE_SIZE);
2708 if (fsverity_active(page->mapping->host) &&
2710 !PageUptodate(page) &&
2711 start < i_size_read(page->mapping->host) &&
2712 !fsverity_verify_page(page)) {
2713 btrfs_page_set_error(fs_info, page, start, len);
2715 btrfs_page_set_uptodate(fs_info, page, start, len);
2718 btrfs_page_clear_uptodate(fs_info, page, start, len);
2719 btrfs_page_set_error(fs_info, page, start, len);
2722 if (!btrfs_is_subpage(fs_info, page))
2725 btrfs_subpage_end_reader(fs_info, page, start, len);
2728 static void end_sector_io(struct page *page, u64 offset, bool uptodate)
2730 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
2731 const u32 sectorsize = inode->root->fs_info->sectorsize;
2732 struct extent_state *cached = NULL;
2734 end_page_read(page, uptodate, offset, sectorsize);
2736 set_extent_uptodate(&inode->io_tree, offset,
2737 offset + sectorsize - 1, &cached, GFP_ATOMIC);
2738 unlock_extent_cached_atomic(&inode->io_tree, offset,
2739 offset + sectorsize - 1, &cached);
2742 static void submit_data_read_repair(struct inode *inode, struct bio *failed_bio,
2743 u32 bio_offset, const struct bio_vec *bvec,
2744 int failed_mirror, unsigned int error_bitmap)
2746 const unsigned int pgoff = bvec->bv_offset;
2747 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2748 struct page *page = bvec->bv_page;
2749 const u64 start = page_offset(bvec->bv_page) + bvec->bv_offset;
2750 const u64 end = start + bvec->bv_len - 1;
2751 const u32 sectorsize = fs_info->sectorsize;
2752 const int nr_bits = (end + 1 - start) >> fs_info->sectorsize_bits;
2755 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
2757 /* This repair is only for data */
2758 ASSERT(is_data_inode(inode));
2760 /* We're here because we had some read errors or csum mismatch */
2761 ASSERT(error_bitmap);
2764 * We only get called on buffered IO, thus page must be mapped and bio
2765 * must not be cloned.
2767 ASSERT(page->mapping && !bio_flagged(failed_bio, BIO_CLONED));
2769 /* Iterate through all the sectors in the range */
2770 for (i = 0; i < nr_bits; i++) {
2771 const unsigned int offset = i * sectorsize;
2772 bool uptodate = false;
2775 if (!(error_bitmap & (1U << i))) {
2777 * This sector has no error, just end the page read
2778 * and unlock the range.
2784 ret = btrfs_repair_one_sector(inode, failed_bio,
2785 bio_offset + offset,
2786 page, pgoff + offset, start + offset,
2787 failed_mirror, btrfs_submit_data_read_bio);
2790 * We have submitted the read repair, the page release
2791 * will be handled by the endio function of the
2792 * submitted repair bio.
2793 * Thus we don't need to do any thing here.
2798 * Continue on failed repair, otherwise the remaining sectors
2799 * will not be properly unlocked.
2802 end_sector_io(page, start + offset, uptodate);
2806 /* lots and lots of room for performance fixes in the end_bio funcs */
2808 void end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2810 struct btrfs_inode *inode;
2811 const bool uptodate = (err == 0);
2814 ASSERT(page && page->mapping);
2815 inode = BTRFS_I(page->mapping->host);
2816 btrfs_writepage_endio_finish_ordered(inode, page, start, end, uptodate);
2819 const struct btrfs_fs_info *fs_info = inode->root->fs_info;
2822 ASSERT(end + 1 - start <= U32_MAX);
2823 len = end + 1 - start;
2825 btrfs_page_clear_uptodate(fs_info, page, start, len);
2826 btrfs_page_set_error(fs_info, page, start, len);
2827 ret = err < 0 ? err : -EIO;
2828 mapping_set_error(page->mapping, ret);
2833 * after a writepage IO is done, we need to:
2834 * clear the uptodate bits on error
2835 * clear the writeback bits in the extent tree for this IO
2836 * end_page_writeback if the page has no more pending IO
2838 * Scheduling is not allowed, so the extent state tree is expected
2839 * to have one and only one object corresponding to this IO.
2841 static void end_bio_extent_writepage(struct bio *bio)
2843 int error = blk_status_to_errno(bio->bi_status);
2844 struct bio_vec *bvec;
2847 struct bvec_iter_all iter_all;
2848 bool first_bvec = true;
2850 ASSERT(!bio_flagged(bio, BIO_CLONED));
2851 bio_for_each_segment_all(bvec, bio, iter_all) {
2852 struct page *page = bvec->bv_page;
2853 struct inode *inode = page->mapping->host;
2854 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2855 const u32 sectorsize = fs_info->sectorsize;
2857 /* Our read/write should always be sector aligned. */
2858 if (!IS_ALIGNED(bvec->bv_offset, sectorsize))
2860 "partial page write in btrfs with offset %u and length %u",
2861 bvec->bv_offset, bvec->bv_len);
2862 else if (!IS_ALIGNED(bvec->bv_len, sectorsize))
2864 "incomplete page write with offset %u and length %u",
2865 bvec->bv_offset, bvec->bv_len);
2867 start = page_offset(page) + bvec->bv_offset;
2868 end = start + bvec->bv_len - 1;
2871 btrfs_record_physical_zoned(inode, start, bio);
2875 end_extent_writepage(page, error, start, end);
2877 btrfs_page_clear_writeback(fs_info, page, start, bvec->bv_len);
2884 * Record previously processed extent range
2886 * For endio_readpage_release_extent() to handle a full extent range, reducing
2887 * the extent io operations.
2889 struct processed_extent {
2890 struct btrfs_inode *inode;
2891 /* Start of the range in @inode */
2893 /* End of the range in @inode */
2899 * Try to release processed extent range
2901 * May not release the extent range right now if the current range is
2902 * contiguous to processed extent.
2904 * Will release processed extent when any of @inode, @uptodate, the range is
2905 * no longer contiguous to the processed range.
2907 * Passing @inode == NULL will force processed extent to be released.
2909 static void endio_readpage_release_extent(struct processed_extent *processed,
2910 struct btrfs_inode *inode, u64 start, u64 end,
2913 struct extent_state *cached = NULL;
2914 struct extent_io_tree *tree;
2916 /* The first extent, initialize @processed */
2917 if (!processed->inode)
2921 * Contiguous to processed extent, just uptodate the end.
2923 * Several things to notice:
2925 * - bio can be merged as long as on-disk bytenr is contiguous
2926 * This means we can have page belonging to other inodes, thus need to
2927 * check if the inode still matches.
2928 * - bvec can contain range beyond current page for multi-page bvec
2929 * Thus we need to do processed->end + 1 >= start check
2931 if (processed->inode == inode && processed->uptodate == uptodate &&
2932 processed->end + 1 >= start && end >= processed->end) {
2933 processed->end = end;
2937 tree = &processed->inode->io_tree;
2939 * Now we don't have range contiguous to the processed range, release
2940 * the processed range now.
2942 if (processed->uptodate && tree->track_uptodate)
2943 set_extent_uptodate(tree, processed->start, processed->end,
2944 &cached, GFP_ATOMIC);
2945 unlock_extent_cached_atomic(tree, processed->start, processed->end,
2949 /* Update processed to current range */
2950 processed->inode = inode;
2951 processed->start = start;
2952 processed->end = end;
2953 processed->uptodate = uptodate;
2956 static void begin_page_read(struct btrfs_fs_info *fs_info, struct page *page)
2958 ASSERT(PageLocked(page));
2959 if (!btrfs_is_subpage(fs_info, page))
2962 ASSERT(PagePrivate(page));
2963 btrfs_subpage_start_reader(fs_info, page, page_offset(page), PAGE_SIZE);
2967 * Find extent buffer for a givne bytenr.
2969 * This is for end_bio_extent_readpage(), thus we can't do any unsafe locking
2972 static struct extent_buffer *find_extent_buffer_readpage(
2973 struct btrfs_fs_info *fs_info, struct page *page, u64 bytenr)
2975 struct extent_buffer *eb;
2978 * For regular sectorsize, we can use page->private to grab extent
2981 if (fs_info->nodesize >= PAGE_SIZE) {
2982 ASSERT(PagePrivate(page) && page->private);
2983 return (struct extent_buffer *)page->private;
2986 /* For subpage case, we need to lookup buffer radix tree */
2988 eb = radix_tree_lookup(&fs_info->buffer_radix,
2989 bytenr >> fs_info->sectorsize_bits);
2996 * after a readpage IO is done, we need to:
2997 * clear the uptodate bits on error
2998 * set the uptodate bits if things worked
2999 * set the page up to date if all extents in the tree are uptodate
3000 * clear the lock bit in the extent tree
3001 * unlock the page if there are no other extents locked for it
3003 * Scheduling is not allowed, so the extent state tree is expected
3004 * to have one and only one object corresponding to this IO.
3006 static void end_bio_extent_readpage(struct bio *bio)
3008 struct bio_vec *bvec;
3009 struct btrfs_bio *bbio = btrfs_bio(bio);
3010 struct extent_io_tree *tree, *failure_tree;
3011 struct processed_extent processed = { 0 };
3013 * The offset to the beginning of a bio, since one bio can never be
3014 * larger than UINT_MAX, u32 here is enough.
3018 struct bvec_iter_all iter_all;
3020 ASSERT(!bio_flagged(bio, BIO_CLONED));
3021 bio_for_each_segment_all(bvec, bio, iter_all) {
3022 bool uptodate = !bio->bi_status;
3023 struct page *page = bvec->bv_page;
3024 struct inode *inode = page->mapping->host;
3025 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3026 const u32 sectorsize = fs_info->sectorsize;
3027 unsigned int error_bitmap = (unsigned int)-1;
3028 bool repair = false;
3033 btrfs_debug(fs_info,
3034 "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u",
3035 bio->bi_iter.bi_sector, bio->bi_status,
3037 tree = &BTRFS_I(inode)->io_tree;
3038 failure_tree = &BTRFS_I(inode)->io_failure_tree;
3041 * We always issue full-sector reads, but if some block in a
3042 * page fails to read, blk_update_request() will advance
3043 * bv_offset and adjust bv_len to compensate. Print a warning
3044 * for unaligned offsets, and an error if they don't add up to
3047 if (!IS_ALIGNED(bvec->bv_offset, sectorsize))
3049 "partial page read in btrfs with offset %u and length %u",
3050 bvec->bv_offset, bvec->bv_len);
3051 else if (!IS_ALIGNED(bvec->bv_offset + bvec->bv_len,
3054 "incomplete page read with offset %u and length %u",
3055 bvec->bv_offset, bvec->bv_len);
3057 start = page_offset(page) + bvec->bv_offset;
3058 end = start + bvec->bv_len - 1;
3061 mirror = bbio->mirror_num;
3062 if (likely(uptodate)) {
3063 if (is_data_inode(inode)) {
3064 error_bitmap = btrfs_verify_data_csum(bbio,
3065 bio_offset, page, start, end);
3069 if (btrfs_validate_metadata_buffer(bbio,
3070 page, start, end, mirror))
3075 if (likely(uptodate)) {
3076 loff_t i_size = i_size_read(inode);
3077 pgoff_t end_index = i_size >> PAGE_SHIFT;
3079 clean_io_failure(BTRFS_I(inode)->root->fs_info,
3080 failure_tree, tree, start, page,
3081 btrfs_ino(BTRFS_I(inode)), 0);
3084 * Zero out the remaining part if this range straddles
3087 * Here we should only zero the range inside the bvec,
3088 * not touch anything else.
3090 * NOTE: i_size is exclusive while end is inclusive.
3092 if (page->index == end_index && i_size <= end) {
3093 u32 zero_start = max(offset_in_page(i_size),
3094 offset_in_page(start));
3096 zero_user_segment(page, zero_start,
3097 offset_in_page(end) + 1);
3099 } else if (is_data_inode(inode)) {
3101 * Only try to repair bios that actually made it to a
3102 * device. If the bio failed to be submitted mirror
3103 * is 0 and we need to fail it without retrying.
3108 struct extent_buffer *eb;
3110 eb = find_extent_buffer_readpage(fs_info, page, start);
3111 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
3112 eb->read_mirror = mirror;
3113 atomic_dec(&eb->io_pages);
3118 * submit_data_read_repair() will handle all the good
3119 * and bad sectors, we just continue to the next bvec.
3121 submit_data_read_repair(inode, bio, bio_offset, bvec,
3122 mirror, error_bitmap);
3124 /* Update page status and unlock */
3125 end_page_read(page, uptodate, start, len);
3126 endio_readpage_release_extent(&processed, BTRFS_I(inode),
3127 start, end, PageUptodate(page));
3130 ASSERT(bio_offset + len > bio_offset);
3134 /* Release the last extent */
3135 endio_readpage_release_extent(&processed, NULL, 0, 0, false);
3136 btrfs_bio_free_csum(bbio);
3141 * Populate every free slot in a provided array with pages.
3143 * @nr_pages: number of pages to allocate
3144 * @page_array: the array to fill with pages; any existing non-null entries in
3145 * the array will be skipped
3147 * Return: 0 if all pages were able to be allocated;
3148 * -ENOMEM otherwise, and the caller is responsible for freeing all
3149 * non-null page pointers in the array.
3151 int btrfs_alloc_page_array(unsigned int nr_pages, struct page **page_array)
3153 unsigned int allocated;
3155 for (allocated = 0; allocated < nr_pages;) {
3156 unsigned int last = allocated;
3158 allocated = alloc_pages_bulk_array(GFP_NOFS, nr_pages, page_array);
3160 if (allocated == nr_pages)
3164 * During this iteration, no page could be allocated, even
3165 * though alloc_pages_bulk_array() falls back to alloc_page()
3166 * if it could not bulk-allocate. So we must be out of memory.
3168 if (allocated == last)
3171 memalloc_retry_wait(GFP_NOFS);
3177 * Initialize the members up to but not including 'bio'. Use after allocating a
3178 * new bio by bio_alloc_bioset as it does not initialize the bytes outside of
3179 * 'bio' because use of __GFP_ZERO is not supported.
3181 static inline void btrfs_bio_init(struct btrfs_bio *bbio)
3183 memset(bbio, 0, offsetof(struct btrfs_bio, bio));
3187 * Allocate a btrfs_io_bio, with @nr_iovecs as maximum number of iovecs.
3189 * The bio allocation is backed by bioset and does not fail.
3191 struct bio *btrfs_bio_alloc(unsigned int nr_iovecs)
3195 ASSERT(0 < nr_iovecs && nr_iovecs <= BIO_MAX_VECS);
3196 bio = bio_alloc_bioset(NULL, nr_iovecs, 0, GFP_NOFS, &btrfs_bioset);
3197 btrfs_bio_init(btrfs_bio(bio));
3201 struct bio *btrfs_bio_clone_partial(struct bio *orig, u64 offset, u64 size)
3204 struct btrfs_bio *bbio;
3206 ASSERT(offset <= UINT_MAX && size <= UINT_MAX);
3208 /* this will never fail when it's backed by a bioset */
3209 bio = bio_alloc_clone(orig->bi_bdev, orig, GFP_NOFS, &btrfs_bioset);
3212 bbio = btrfs_bio(bio);
3213 btrfs_bio_init(bbio);
3215 bio_trim(bio, offset >> 9, size >> 9);
3216 bbio->iter = bio->bi_iter;
3221 * Attempt to add a page to bio
3223 * @bio_ctrl: record both the bio, and its bio_flags
3224 * @page: page to add to the bio
3225 * @disk_bytenr: offset of the new bio or to check whether we are adding
3226 * a contiguous page to the previous one
3227 * @size: portion of page that we want to write
3228 * @pg_offset: starting offset in the page
3229 * @compress_type: compression type of the current bio to see if we can merge them
3231 * Attempt to add a page to bio considering stripe alignment etc.
3233 * Return >= 0 for the number of bytes added to the bio.
3234 * Can return 0 if the current bio is already at stripe/zone boundary.
3235 * Return <0 for error.
3237 static int btrfs_bio_add_page(struct btrfs_bio_ctrl *bio_ctrl,
3239 u64 disk_bytenr, unsigned int size,
3240 unsigned int pg_offset,
3241 enum btrfs_compression_type compress_type)
3243 struct bio *bio = bio_ctrl->bio;
3244 u32 bio_size = bio->bi_iter.bi_size;
3246 const sector_t sector = disk_bytenr >> SECTOR_SHIFT;
3251 /* The limit should be calculated when bio_ctrl->bio is allocated */
3252 ASSERT(bio_ctrl->len_to_oe_boundary && bio_ctrl->len_to_stripe_boundary);
3253 if (bio_ctrl->compress_type != compress_type)
3256 if (bio_ctrl->compress_type != BTRFS_COMPRESS_NONE)
3257 contig = bio->bi_iter.bi_sector == sector;
3259 contig = bio_end_sector(bio) == sector;
3263 real_size = min(bio_ctrl->len_to_oe_boundary,
3264 bio_ctrl->len_to_stripe_boundary) - bio_size;
3265 real_size = min(real_size, size);
3268 * If real_size is 0, never call bio_add_*_page(), as even size is 0,
3269 * bio will still execute its endio function on the page!
3274 if (bio_op(bio) == REQ_OP_ZONE_APPEND)
3275 ret = bio_add_zone_append_page(bio, page, real_size, pg_offset);
3277 ret = bio_add_page(bio, page, real_size, pg_offset);
3282 static int calc_bio_boundaries(struct btrfs_bio_ctrl *bio_ctrl,
3283 struct btrfs_inode *inode, u64 file_offset)
3285 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3286 struct btrfs_io_geometry geom;
3287 struct btrfs_ordered_extent *ordered;
3288 struct extent_map *em;
3289 u64 logical = (bio_ctrl->bio->bi_iter.bi_sector << SECTOR_SHIFT);
3293 * Pages for compressed extent are never submitted to disk directly,
3294 * thus it has no real boundary, just set them to U32_MAX.
3296 * The split happens for real compressed bio, which happens in
3297 * btrfs_submit_compressed_read/write().
3299 if (bio_ctrl->compress_type != BTRFS_COMPRESS_NONE) {
3300 bio_ctrl->len_to_oe_boundary = U32_MAX;
3301 bio_ctrl->len_to_stripe_boundary = U32_MAX;
3304 em = btrfs_get_chunk_map(fs_info, logical, fs_info->sectorsize);
3307 ret = btrfs_get_io_geometry(fs_info, em, btrfs_op(bio_ctrl->bio),
3309 free_extent_map(em);
3313 if (geom.len > U32_MAX)
3314 bio_ctrl->len_to_stripe_boundary = U32_MAX;
3316 bio_ctrl->len_to_stripe_boundary = (u32)geom.len;
3318 if (bio_op(bio_ctrl->bio) != REQ_OP_ZONE_APPEND) {
3319 bio_ctrl->len_to_oe_boundary = U32_MAX;
3323 /* Ordered extent not yet created, so we're good */
3324 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
3326 bio_ctrl->len_to_oe_boundary = U32_MAX;
3330 bio_ctrl->len_to_oe_boundary = min_t(u32, U32_MAX,
3331 ordered->disk_bytenr + ordered->disk_num_bytes - logical);
3332 btrfs_put_ordered_extent(ordered);
3336 static int alloc_new_bio(struct btrfs_inode *inode,
3337 struct btrfs_bio_ctrl *bio_ctrl,
3338 struct writeback_control *wbc,
3340 bio_end_io_t end_io_func,
3341 u64 disk_bytenr, u32 offset, u64 file_offset,
3342 enum btrfs_compression_type compress_type)
3344 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3348 bio = btrfs_bio_alloc(BIO_MAX_VECS);
3350 * For compressed page range, its disk_bytenr is always @disk_bytenr
3351 * passed in, no matter if we have added any range into previous bio.
3353 if (compress_type != BTRFS_COMPRESS_NONE)
3354 bio->bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
3356 bio->bi_iter.bi_sector = (disk_bytenr + offset) >> SECTOR_SHIFT;
3357 bio_ctrl->bio = bio;
3358 bio_ctrl->compress_type = compress_type;
3359 bio->bi_end_io = end_io_func;
3361 ret = calc_bio_boundaries(bio_ctrl, inode, file_offset);
3367 * For Zone append we need the correct block_device that we are
3368 * going to write to set in the bio to be able to respect the
3369 * hardware limitation. Look it up here:
3371 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
3372 struct btrfs_device *dev;
3374 dev = btrfs_zoned_get_device(fs_info, disk_bytenr,
3375 fs_info->sectorsize);
3381 bio_set_dev(bio, dev->bdev);
3384 * Otherwise pick the last added device to support
3385 * cgroup writeback. For multi-device file systems this
3386 * means blk-cgroup policies have to always be set on the
3387 * last added/replaced device. This is a bit odd but has
3388 * been like that for a long time.
3390 bio_set_dev(bio, fs_info->fs_devices->latest_dev->bdev);
3392 wbc_init_bio(wbc, bio);
3394 ASSERT(bio_op(bio) != REQ_OP_ZONE_APPEND);
3398 bio_ctrl->bio = NULL;
3399 bio->bi_status = errno_to_blk_status(ret);
3405 * @opf: bio REQ_OP_* and REQ_* flags as one value
3406 * @wbc: optional writeback control for io accounting
3407 * @page: page to add to the bio
3408 * @disk_bytenr: logical bytenr where the write will be
3409 * @size: portion of page that we want to write to
3410 * @pg_offset: offset of the new bio or to check whether we are adding
3411 * a contiguous page to the previous one
3412 * @bio_ret: must be valid pointer, newly allocated bio will be stored there
3413 * @end_io_func: end_io callback for new bio
3414 * @mirror_num: desired mirror to read/write
3415 * @prev_bio_flags: flags of previous bio to see if we can merge the current one
3416 * @compress_type: compress type for current bio
3418 static int submit_extent_page(unsigned int opf,
3419 struct writeback_control *wbc,
3420 struct btrfs_bio_ctrl *bio_ctrl,
3421 struct page *page, u64 disk_bytenr,
3422 size_t size, unsigned long pg_offset,
3423 bio_end_io_t end_io_func,
3424 enum btrfs_compression_type compress_type,
3425 bool force_bio_submit)
3428 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
3429 unsigned int cur = pg_offset;
3433 ASSERT(pg_offset < PAGE_SIZE && size <= PAGE_SIZE &&
3434 pg_offset + size <= PAGE_SIZE);
3435 if (force_bio_submit)
3436 submit_one_bio(bio_ctrl);
3438 while (cur < pg_offset + size) {
3439 u32 offset = cur - pg_offset;
3442 /* Allocate new bio if needed */
3443 if (!bio_ctrl->bio) {
3444 ret = alloc_new_bio(inode, bio_ctrl, wbc, opf,
3445 end_io_func, disk_bytenr, offset,
3446 page_offset(page) + cur,
3452 * We must go through btrfs_bio_add_page() to ensure each
3453 * page range won't cross various boundaries.
3455 if (compress_type != BTRFS_COMPRESS_NONE)
3456 added = btrfs_bio_add_page(bio_ctrl, page, disk_bytenr,
3457 size - offset, pg_offset + offset,
3460 added = btrfs_bio_add_page(bio_ctrl, page,
3461 disk_bytenr + offset, size - offset,
3462 pg_offset + offset, compress_type);
3464 /* Metadata page range should never be split */
3465 if (!is_data_inode(&inode->vfs_inode))
3466 ASSERT(added == 0 || added == size - offset);
3468 /* At least we added some page, update the account */
3470 wbc_account_cgroup_owner(wbc, page, added);
3472 /* We have reached boundary, submit right now */
3473 if (added < size - offset) {
3474 /* The bio should contain some page(s) */
3475 ASSERT(bio_ctrl->bio->bi_iter.bi_size);
3476 submit_one_bio(bio_ctrl);
3483 static int attach_extent_buffer_page(struct extent_buffer *eb,
3485 struct btrfs_subpage *prealloc)
3487 struct btrfs_fs_info *fs_info = eb->fs_info;
3491 * If the page is mapped to btree inode, we should hold the private
3492 * lock to prevent race.
3493 * For cloned or dummy extent buffers, their pages are not mapped and
3494 * will not race with any other ebs.
3497 lockdep_assert_held(&page->mapping->private_lock);
3499 if (fs_info->nodesize >= PAGE_SIZE) {
3500 if (!PagePrivate(page))
3501 attach_page_private(page, eb);
3503 WARN_ON(page->private != (unsigned long)eb);
3507 /* Already mapped, just free prealloc */
3508 if (PagePrivate(page)) {
3509 btrfs_free_subpage(prealloc);
3514 /* Has preallocated memory for subpage */
3515 attach_page_private(page, prealloc);
3517 /* Do new allocation to attach subpage */
3518 ret = btrfs_attach_subpage(fs_info, page,
3519 BTRFS_SUBPAGE_METADATA);
3523 int set_page_extent_mapped(struct page *page)
3525 struct btrfs_fs_info *fs_info;
3527 ASSERT(page->mapping);
3529 if (PagePrivate(page))
3532 fs_info = btrfs_sb(page->mapping->host->i_sb);
3534 if (btrfs_is_subpage(fs_info, page))
3535 return btrfs_attach_subpage(fs_info, page, BTRFS_SUBPAGE_DATA);
3537 attach_page_private(page, (void *)EXTENT_PAGE_PRIVATE);
3541 void clear_page_extent_mapped(struct page *page)
3543 struct btrfs_fs_info *fs_info;
3545 ASSERT(page->mapping);
3547 if (!PagePrivate(page))
3550 fs_info = btrfs_sb(page->mapping->host->i_sb);
3551 if (btrfs_is_subpage(fs_info, page))
3552 return btrfs_detach_subpage(fs_info, page);
3554 detach_page_private(page);
3557 static struct extent_map *
3558 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
3559 u64 start, u64 len, struct extent_map **em_cached)
3561 struct extent_map *em;
3563 if (em_cached && *em_cached) {
3565 if (extent_map_in_tree(em) && start >= em->start &&
3566 start < extent_map_end(em)) {
3567 refcount_inc(&em->refs);
3571 free_extent_map(em);
3575 em = btrfs_get_extent(BTRFS_I(inode), page, pg_offset, start, len);
3576 if (em_cached && !IS_ERR(em)) {
3578 refcount_inc(&em->refs);
3584 * basic readpage implementation. Locked extent state structs are inserted
3585 * into the tree that are removed when the IO is done (by the end_io
3587 * XXX JDM: This needs looking at to ensure proper page locking
3588 * return 0 on success, otherwise return error
3590 static int btrfs_do_readpage(struct page *page, struct extent_map **em_cached,
3591 struct btrfs_bio_ctrl *bio_ctrl,
3592 unsigned int read_flags, u64 *prev_em_start)
3594 struct inode *inode = page->mapping->host;
3595 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3596 u64 start = page_offset(page);
3597 const u64 end = start + PAGE_SIZE - 1;
3600 u64 last_byte = i_size_read(inode);
3603 struct extent_map *em;
3605 size_t pg_offset = 0;
3607 size_t blocksize = inode->i_sb->s_blocksize;
3608 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
3610 ret = set_page_extent_mapped(page);
3612 unlock_extent(tree, start, end);
3613 btrfs_page_set_error(fs_info, page, start, PAGE_SIZE);
3618 if (page->index == last_byte >> PAGE_SHIFT) {
3619 size_t zero_offset = offset_in_page(last_byte);
3622 iosize = PAGE_SIZE - zero_offset;
3623 memzero_page(page, zero_offset, iosize);
3626 begin_page_read(fs_info, page);
3627 while (cur <= end) {
3628 unsigned long this_bio_flag = 0;
3629 bool force_bio_submit = false;
3632 ASSERT(IS_ALIGNED(cur, fs_info->sectorsize));
3633 if (cur >= last_byte) {
3634 struct extent_state *cached = NULL;
3636 iosize = PAGE_SIZE - pg_offset;
3637 memzero_page(page, pg_offset, iosize);
3638 set_extent_uptodate(tree, cur, cur + iosize - 1,
3640 unlock_extent_cached(tree, cur,
3641 cur + iosize - 1, &cached);
3642 end_page_read(page, true, cur, iosize);
3645 em = __get_extent_map(inode, page, pg_offset, cur,
3646 end - cur + 1, em_cached);
3648 unlock_extent(tree, cur, end);
3649 end_page_read(page, false, cur, end + 1 - cur);
3653 extent_offset = cur - em->start;
3654 BUG_ON(extent_map_end(em) <= cur);
3657 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
3658 this_bio_flag = em->compress_type;
3660 iosize = min(extent_map_end(em) - cur, end - cur + 1);
3661 cur_end = min(extent_map_end(em) - 1, end);
3662 iosize = ALIGN(iosize, blocksize);
3663 if (this_bio_flag != BTRFS_COMPRESS_NONE)
3664 disk_bytenr = em->block_start;
3666 disk_bytenr = em->block_start + extent_offset;
3667 block_start = em->block_start;
3668 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
3669 block_start = EXTENT_MAP_HOLE;
3672 * If we have a file range that points to a compressed extent
3673 * and it's followed by a consecutive file range that points
3674 * to the same compressed extent (possibly with a different
3675 * offset and/or length, so it either points to the whole extent
3676 * or only part of it), we must make sure we do not submit a
3677 * single bio to populate the pages for the 2 ranges because
3678 * this makes the compressed extent read zero out the pages
3679 * belonging to the 2nd range. Imagine the following scenario:
3682 * [0 - 8K] [8K - 24K]
3685 * points to extent X, points to extent X,
3686 * offset 4K, length of 8K offset 0, length 16K
3688 * [extent X, compressed length = 4K uncompressed length = 16K]
3690 * If the bio to read the compressed extent covers both ranges,
3691 * it will decompress extent X into the pages belonging to the
3692 * first range and then it will stop, zeroing out the remaining
3693 * pages that belong to the other range that points to extent X.
3694 * So here we make sure we submit 2 bios, one for the first
3695 * range and another one for the third range. Both will target
3696 * the same physical extent from disk, but we can't currently
3697 * make the compressed bio endio callback populate the pages
3698 * for both ranges because each compressed bio is tightly
3699 * coupled with a single extent map, and each range can have
3700 * an extent map with a different offset value relative to the
3701 * uncompressed data of our extent and different lengths. This
3702 * is a corner case so we prioritize correctness over
3703 * non-optimal behavior (submitting 2 bios for the same extent).
3705 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) &&
3706 prev_em_start && *prev_em_start != (u64)-1 &&
3707 *prev_em_start != em->start)
3708 force_bio_submit = true;
3711 *prev_em_start = em->start;
3713 free_extent_map(em);
3716 /* we've found a hole, just zero and go on */
3717 if (block_start == EXTENT_MAP_HOLE) {
3718 struct extent_state *cached = NULL;
3720 memzero_page(page, pg_offset, iosize);
3722 set_extent_uptodate(tree, cur, cur + iosize - 1,
3724 unlock_extent_cached(tree, cur,
3725 cur + iosize - 1, &cached);
3726 end_page_read(page, true, cur, iosize);
3728 pg_offset += iosize;
3731 /* the get_extent function already copied into the page */
3732 if (test_range_bit(tree, cur, cur_end,
3733 EXTENT_UPTODATE, 1, NULL)) {
3734 unlock_extent(tree, cur, cur + iosize - 1);
3735 end_page_read(page, true, cur, iosize);
3737 pg_offset += iosize;
3740 /* we have an inline extent but it didn't get marked up
3741 * to date. Error out
3743 if (block_start == EXTENT_MAP_INLINE) {
3744 unlock_extent(tree, cur, cur + iosize - 1);
3745 end_page_read(page, false, cur, iosize);
3747 pg_offset += iosize;
3751 ret = submit_extent_page(REQ_OP_READ | read_flags, NULL,
3752 bio_ctrl, page, disk_bytenr, iosize,
3753 pg_offset, end_bio_extent_readpage,
3754 this_bio_flag, force_bio_submit);
3757 * We have to unlock the remaining range, or the page
3758 * will never be unlocked.
3760 unlock_extent(tree, cur, end);
3761 end_page_read(page, false, cur, end + 1 - cur);
3765 pg_offset += iosize;
3771 int btrfs_read_folio(struct file *file, struct folio *folio)
3773 struct page *page = &folio->page;
3774 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
3775 u64 start = page_offset(page);
3776 u64 end = start + PAGE_SIZE - 1;
3777 struct btrfs_bio_ctrl bio_ctrl = { 0 };
3780 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
3782 ret = btrfs_do_readpage(page, NULL, &bio_ctrl, 0, NULL);
3784 * If btrfs_do_readpage() failed we will want to submit the assembled
3785 * bio to do the cleanup.
3787 submit_one_bio(&bio_ctrl);
3791 static inline void contiguous_readpages(struct page *pages[], int nr_pages,
3793 struct extent_map **em_cached,
3794 struct btrfs_bio_ctrl *bio_ctrl,
3797 struct btrfs_inode *inode = BTRFS_I(pages[0]->mapping->host);
3800 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
3802 for (index = 0; index < nr_pages; index++) {
3803 btrfs_do_readpage(pages[index], em_cached, bio_ctrl,
3804 REQ_RAHEAD, prev_em_start);
3805 put_page(pages[index]);
3810 * helper for __extent_writepage, doing all of the delayed allocation setup.
3812 * This returns 1 if btrfs_run_delalloc_range function did all the work required
3813 * to write the page (copy into inline extent). In this case the IO has
3814 * been started and the page is already unlocked.
3816 * This returns 0 if all went well (page still locked)
3817 * This returns < 0 if there were errors (page still locked)
3819 static noinline_for_stack int writepage_delalloc(struct btrfs_inode *inode,
3820 struct page *page, struct writeback_control *wbc)
3822 const u64 page_end = page_offset(page) + PAGE_SIZE - 1;
3823 u64 delalloc_start = page_offset(page);
3824 u64 delalloc_to_write = 0;
3825 /* How many pages are started by btrfs_run_delalloc_range() */
3826 unsigned long nr_written = 0;
3828 int page_started = 0;
3830 while (delalloc_start < page_end) {
3831 u64 delalloc_end = page_end;
3834 found = find_lock_delalloc_range(&inode->vfs_inode, page,
3838 delalloc_start = delalloc_end + 1;
3841 ret = btrfs_run_delalloc_range(inode, page, delalloc_start,
3842 delalloc_end, &page_started, &nr_written, wbc);
3844 btrfs_page_set_error(inode->root->fs_info, page,
3845 page_offset(page), PAGE_SIZE);
3849 * delalloc_end is already one less than the total length, so
3850 * we don't subtract one from PAGE_SIZE
3852 delalloc_to_write += (delalloc_end - delalloc_start +
3853 PAGE_SIZE) >> PAGE_SHIFT;
3854 delalloc_start = delalloc_end + 1;
3856 if (wbc->nr_to_write < delalloc_to_write) {
3859 if (delalloc_to_write < thresh * 2)
3860 thresh = delalloc_to_write;
3861 wbc->nr_to_write = min_t(u64, delalloc_to_write,
3865 /* Did btrfs_run_dealloc_range() already unlock and start the IO? */
3868 * We've unlocked the page, so we can't update the mapping's
3869 * writeback index, just update nr_to_write.
3871 wbc->nr_to_write -= nr_written;
3879 * Find the first byte we need to write.
3881 * For subpage, one page can contain several sectors, and
3882 * __extent_writepage_io() will just grab all extent maps in the page
3883 * range and try to submit all non-inline/non-compressed extents.
3885 * This is a big problem for subpage, we shouldn't re-submit already written
3887 * This function will lookup subpage dirty bit to find which range we really
3890 * Return the next dirty range in [@start, @end).
3891 * If no dirty range is found, @start will be page_offset(page) + PAGE_SIZE.
3893 static void find_next_dirty_byte(struct btrfs_fs_info *fs_info,
3894 struct page *page, u64 *start, u64 *end)
3896 struct btrfs_subpage *subpage = (struct btrfs_subpage *)page->private;
3897 struct btrfs_subpage_info *spi = fs_info->subpage_info;
3898 u64 orig_start = *start;
3899 /* Declare as unsigned long so we can use bitmap ops */
3900 unsigned long flags;
3901 int range_start_bit;
3905 * For regular sector size == page size case, since one page only
3906 * contains one sector, we return the page offset directly.
3908 if (!btrfs_is_subpage(fs_info, page)) {
3909 *start = page_offset(page);
3910 *end = page_offset(page) + PAGE_SIZE;
3914 range_start_bit = spi->dirty_offset +
3915 (offset_in_page(orig_start) >> fs_info->sectorsize_bits);
3917 /* We should have the page locked, but just in case */
3918 spin_lock_irqsave(&subpage->lock, flags);
3919 bitmap_next_set_region(subpage->bitmaps, &range_start_bit, &range_end_bit,
3920 spi->dirty_offset + spi->bitmap_nr_bits);
3921 spin_unlock_irqrestore(&subpage->lock, flags);
3923 range_start_bit -= spi->dirty_offset;
3924 range_end_bit -= spi->dirty_offset;
3926 *start = page_offset(page) + range_start_bit * fs_info->sectorsize;
3927 *end = page_offset(page) + range_end_bit * fs_info->sectorsize;
3931 * helper for __extent_writepage. This calls the writepage start hooks,
3932 * and does the loop to map the page into extents and bios.
3934 * We return 1 if the IO is started and the page is unlocked,
3935 * 0 if all went well (page still locked)
3936 * < 0 if there were errors (page still locked)
3938 static noinline_for_stack int __extent_writepage_io(struct btrfs_inode *inode,
3940 struct writeback_control *wbc,
3941 struct extent_page_data *epd,
3945 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3946 u64 cur = page_offset(page);
3947 u64 end = cur + PAGE_SIZE - 1;
3950 struct extent_map *em;
3954 u32 opf = REQ_OP_WRITE;
3955 const unsigned int write_flags = wbc_to_write_flags(wbc);
3956 bool has_error = false;
3959 ret = btrfs_writepage_cow_fixup(page);
3961 /* Fixup worker will requeue */
3962 redirty_page_for_writepage(wbc, page);
3968 * we don't want to touch the inode after unlocking the page,
3969 * so we update the mapping writeback index now
3973 while (cur <= end) {
3976 u64 dirty_range_start = cur;
3977 u64 dirty_range_end;
3980 if (cur >= i_size) {
3981 btrfs_writepage_endio_finish_ordered(inode, page, cur,
3984 * This range is beyond i_size, thus we don't need to
3985 * bother writing back.
3986 * But we still need to clear the dirty subpage bit, or
3987 * the next time the page gets dirtied, we will try to
3988 * writeback the sectors with subpage dirty bits,
3989 * causing writeback without ordered extent.
3991 btrfs_page_clear_dirty(fs_info, page, cur, end + 1 - cur);
3995 find_next_dirty_byte(fs_info, page, &dirty_range_start,
3997 if (cur < dirty_range_start) {
3998 cur = dirty_range_start;
4002 em = btrfs_get_extent(inode, NULL, 0, cur, end - cur + 1);
4004 btrfs_page_set_error(fs_info, page, cur, end - cur + 1);
4005 ret = PTR_ERR_OR_ZERO(em);
4012 extent_offset = cur - em->start;
4013 em_end = extent_map_end(em);
4014 ASSERT(cur <= em_end);
4016 ASSERT(IS_ALIGNED(em->start, fs_info->sectorsize));
4017 ASSERT(IS_ALIGNED(em->len, fs_info->sectorsize));
4018 block_start = em->block_start;
4019 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4020 disk_bytenr = em->block_start + extent_offset;
4023 * Note that em_end from extent_map_end() and dirty_range_end from
4024 * find_next_dirty_byte() are all exclusive
4026 iosize = min(min(em_end, end + 1), dirty_range_end) - cur;
4028 if (btrfs_use_zone_append(inode, em->block_start))
4029 opf = REQ_OP_ZONE_APPEND;
4031 free_extent_map(em);
4035 * compressed and inline extents are written through other
4038 if (compressed || block_start == EXTENT_MAP_HOLE ||
4039 block_start == EXTENT_MAP_INLINE) {
4043 btrfs_writepage_endio_finish_ordered(inode,
4044 page, cur, cur + iosize - 1, true);
4045 btrfs_page_clear_dirty(fs_info, page, cur, iosize);
4050 btrfs_set_range_writeback(inode, cur, cur + iosize - 1);
4051 if (!PageWriteback(page)) {
4052 btrfs_err(inode->root->fs_info,
4053 "page %lu not writeback, cur %llu end %llu",
4054 page->index, cur, end);
4058 * Although the PageDirty bit is cleared before entering this
4059 * function, subpage dirty bit is not cleared.
4060 * So clear subpage dirty bit here so next time we won't submit
4061 * page for range already written to disk.
4063 btrfs_page_clear_dirty(fs_info, page, cur, iosize);
4065 ret = submit_extent_page(opf | write_flags, wbc,
4066 &epd->bio_ctrl, page,
4067 disk_bytenr, iosize,
4068 cur - page_offset(page),
4069 end_bio_extent_writepage,
4076 btrfs_page_set_error(fs_info, page, cur, iosize);
4077 if (PageWriteback(page))
4078 btrfs_page_clear_writeback(fs_info, page, cur,
4086 * If we finish without problem, we should not only clear page dirty,
4087 * but also empty subpage dirty bits
4090 btrfs_page_assert_not_dirty(fs_info, page);
4098 * the writepage semantics are similar to regular writepage. extent
4099 * records are inserted to lock ranges in the tree, and as dirty areas
4100 * are found, they are marked writeback. Then the lock bits are removed
4101 * and the end_io handler clears the writeback ranges
4103 * Return 0 if everything goes well.
4104 * Return <0 for error.
4106 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
4107 struct extent_page_data *epd)
4109 struct folio *folio = page_folio(page);
4110 struct inode *inode = page->mapping->host;
4111 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4112 const u64 page_start = page_offset(page);
4113 const u64 page_end = page_start + PAGE_SIZE - 1;
4117 loff_t i_size = i_size_read(inode);
4118 unsigned long end_index = i_size >> PAGE_SHIFT;
4120 trace___extent_writepage(page, inode, wbc);
4122 WARN_ON(!PageLocked(page));
4124 btrfs_page_clear_error(btrfs_sb(inode->i_sb), page,
4125 page_offset(page), PAGE_SIZE);
4127 pg_offset = offset_in_page(i_size);
4128 if (page->index > end_index ||
4129 (page->index == end_index && !pg_offset)) {
4130 folio_invalidate(folio, 0, folio_size(folio));
4131 folio_unlock(folio);
4135 if (page->index == end_index)
4136 memzero_page(page, pg_offset, PAGE_SIZE - pg_offset);
4138 ret = set_page_extent_mapped(page);
4144 if (!epd->extent_locked) {
4145 ret = writepage_delalloc(BTRFS_I(inode), page, wbc);
4152 ret = __extent_writepage_io(BTRFS_I(inode), page, wbc, epd, i_size,
4159 /* make sure the mapping tag for page dirty gets cleared */
4160 set_page_writeback(page);
4161 end_page_writeback(page);
4164 * Here we used to have a check for PageError() and then set @ret and
4165 * call end_extent_writepage().
4167 * But in fact setting @ret here will cause different error paths
4168 * between subpage and regular sectorsize.
4170 * For regular page size, we never submit current page, but only add
4171 * current page to current bio.
4172 * The bio submission can only happen in next page.
4173 * Thus if we hit the PageError() branch, @ret is already set to
4174 * non-zero value and will not get updated for regular sectorsize.
4176 * But for subpage case, it's possible we submit part of current page,
4177 * thus can get PageError() set by submitted bio of the same page,
4178 * while our @ret is still 0.
4180 * So here we unify the behavior and don't set @ret.
4181 * Error can still be properly passed to higher layer as page will
4182 * be set error, here we just don't handle the IO failure.
4184 * NOTE: This is just a hotfix for subpage.
4185 * The root fix will be properly ending ordered extent when we hit
4186 * an error during writeback.
4188 * But that needs a bigger refactoring, as we not only need to grab the
4189 * submitted OE, but also need to know exactly at which bytenr we hit
4191 * Currently the full page based __extent_writepage_io() is not
4194 if (PageError(page))
4195 end_extent_writepage(page, ret, page_start, page_end);
4196 if (epd->extent_locked) {
4198 * If epd->extent_locked, it's from extent_write_locked_range(),
4199 * the page can either be locked by lock_page() or
4200 * process_one_page().
4201 * Let btrfs_page_unlock_writer() handle both cases.
4204 btrfs_page_unlock_writer(fs_info, page, wbc->range_start,
4205 wbc->range_end + 1 - wbc->range_start);
4213 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
4215 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
4216 TASK_UNINTERRUPTIBLE);
4219 static void end_extent_buffer_writeback(struct extent_buffer *eb)
4221 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
4222 smp_mb__after_atomic();
4223 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
4227 * Lock extent buffer status and pages for writeback.
4229 * May try to flush write bio if we can't get the lock.
4231 * Return 0 if the extent buffer doesn't need to be submitted.
4232 * (E.g. the extent buffer is not dirty)
4233 * Return >0 is the extent buffer is submitted to bio.
4234 * Return <0 if something went wrong, no page is locked.
4236 static noinline_for_stack int lock_extent_buffer_for_io(struct extent_buffer *eb,
4237 struct extent_page_data *epd)
4239 struct btrfs_fs_info *fs_info = eb->fs_info;
4244 if (!btrfs_try_tree_write_lock(eb)) {
4245 submit_write_bio(epd, 0);
4247 btrfs_tree_lock(eb);
4250 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
4251 btrfs_tree_unlock(eb);
4255 submit_write_bio(epd, 0);
4259 wait_on_extent_buffer_writeback(eb);
4260 btrfs_tree_lock(eb);
4261 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
4263 btrfs_tree_unlock(eb);
4268 * We need to do this to prevent races in people who check if the eb is
4269 * under IO since we can end up having no IO bits set for a short period
4272 spin_lock(&eb->refs_lock);
4273 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
4274 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
4275 spin_unlock(&eb->refs_lock);
4276 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
4277 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4279 fs_info->dirty_metadata_batch);
4282 spin_unlock(&eb->refs_lock);
4285 btrfs_tree_unlock(eb);
4288 * Either we don't need to submit any tree block, or we're submitting
4290 * Subpage metadata doesn't use page locking at all, so we can skip
4293 if (!ret || fs_info->nodesize < PAGE_SIZE)
4296 num_pages = num_extent_pages(eb);
4297 for (i = 0; i < num_pages; i++) {
4298 struct page *p = eb->pages[i];
4300 if (!trylock_page(p)) {
4302 submit_write_bio(epd, 0);
4312 static void set_btree_ioerr(struct page *page, struct extent_buffer *eb)
4314 struct btrfs_fs_info *fs_info = eb->fs_info;
4316 btrfs_page_set_error(fs_info, page, eb->start, eb->len);
4317 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))
4321 * A read may stumble upon this buffer later, make sure that it gets an
4322 * error and knows there was an error.
4324 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4327 * We need to set the mapping with the io error as well because a write
4328 * error will flip the file system readonly, and then syncfs() will
4329 * return a 0 because we are readonly if we don't modify the err seq for
4332 mapping_set_error(page->mapping, -EIO);
4335 * If we error out, we should add back the dirty_metadata_bytes
4336 * to make it consistent.
4338 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4339 eb->len, fs_info->dirty_metadata_batch);
4342 * If writeback for a btree extent that doesn't belong to a log tree
4343 * failed, increment the counter transaction->eb_write_errors.
4344 * We do this because while the transaction is running and before it's
4345 * committing (when we call filemap_fdata[write|wait]_range against
4346 * the btree inode), we might have
4347 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
4348 * returns an error or an error happens during writeback, when we're
4349 * committing the transaction we wouldn't know about it, since the pages
4350 * can be no longer dirty nor marked anymore for writeback (if a
4351 * subsequent modification to the extent buffer didn't happen before the
4352 * transaction commit), which makes filemap_fdata[write|wait]_range not
4353 * able to find the pages tagged with SetPageError at transaction
4354 * commit time. So if this happens we must abort the transaction,
4355 * otherwise we commit a super block with btree roots that point to
4356 * btree nodes/leafs whose content on disk is invalid - either garbage
4357 * or the content of some node/leaf from a past generation that got
4358 * cowed or deleted and is no longer valid.
4360 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
4361 * not be enough - we need to distinguish between log tree extents vs
4362 * non-log tree extents, and the next filemap_fdatawait_range() call
4363 * will catch and clear such errors in the mapping - and that call might
4364 * be from a log sync and not from a transaction commit. Also, checking
4365 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
4366 * not done and would not be reliable - the eb might have been released
4367 * from memory and reading it back again means that flag would not be
4368 * set (since it's a runtime flag, not persisted on disk).
4370 * Using the flags below in the btree inode also makes us achieve the
4371 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
4372 * writeback for all dirty pages and before filemap_fdatawait_range()
4373 * is called, the writeback for all dirty pages had already finished
4374 * with errors - because we were not using AS_EIO/AS_ENOSPC,
4375 * filemap_fdatawait_range() would return success, as it could not know
4376 * that writeback errors happened (the pages were no longer tagged for
4379 switch (eb->log_index) {
4381 set_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags);
4384 set_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags);
4387 set_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags);
4390 BUG(); /* unexpected, logic error */
4395 * The endio specific version which won't touch any unsafe spinlock in endio
4398 static struct extent_buffer *find_extent_buffer_nolock(
4399 struct btrfs_fs_info *fs_info, u64 start)
4401 struct extent_buffer *eb;
4404 eb = radix_tree_lookup(&fs_info->buffer_radix,
4405 start >> fs_info->sectorsize_bits);
4406 if (eb && atomic_inc_not_zero(&eb->refs)) {
4415 * The endio function for subpage extent buffer write.
4417 * Unlike end_bio_extent_buffer_writepage(), we only call end_page_writeback()
4418 * after all extent buffers in the page has finished their writeback.
4420 static void end_bio_subpage_eb_writepage(struct bio *bio)
4422 struct btrfs_fs_info *fs_info;
4423 struct bio_vec *bvec;
4424 struct bvec_iter_all iter_all;
4426 fs_info = btrfs_sb(bio_first_page_all(bio)->mapping->host->i_sb);
4427 ASSERT(fs_info->nodesize < PAGE_SIZE);
4429 ASSERT(!bio_flagged(bio, BIO_CLONED));
4430 bio_for_each_segment_all(bvec, bio, iter_all) {
4431 struct page *page = bvec->bv_page;
4432 u64 bvec_start = page_offset(page) + bvec->bv_offset;
4433 u64 bvec_end = bvec_start + bvec->bv_len - 1;
4434 u64 cur_bytenr = bvec_start;
4436 ASSERT(IS_ALIGNED(bvec->bv_len, fs_info->nodesize));
4438 /* Iterate through all extent buffers in the range */
4439 while (cur_bytenr <= bvec_end) {
4440 struct extent_buffer *eb;
4444 * Here we can't use find_extent_buffer(), as it may
4445 * try to lock eb->refs_lock, which is not safe in endio
4448 eb = find_extent_buffer_nolock(fs_info, cur_bytenr);
4451 cur_bytenr = eb->start + eb->len;
4453 ASSERT(test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags));
4454 done = atomic_dec_and_test(&eb->io_pages);
4457 if (bio->bi_status ||
4458 test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) {
4459 ClearPageUptodate(page);
4460 set_btree_ioerr(page, eb);
4463 btrfs_subpage_clear_writeback(fs_info, page, eb->start,
4465 end_extent_buffer_writeback(eb);
4467 * free_extent_buffer() will grab spinlock which is not
4468 * safe in endio context. Thus here we manually dec
4471 atomic_dec(&eb->refs);
4477 static void end_bio_extent_buffer_writepage(struct bio *bio)
4479 struct bio_vec *bvec;
4480 struct extent_buffer *eb;
4482 struct bvec_iter_all iter_all;
4484 ASSERT(!bio_flagged(bio, BIO_CLONED));
4485 bio_for_each_segment_all(bvec, bio, iter_all) {
4486 struct page *page = bvec->bv_page;
4488 eb = (struct extent_buffer *)page->private;
4490 done = atomic_dec_and_test(&eb->io_pages);
4492 if (bio->bi_status ||
4493 test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) {
4494 ClearPageUptodate(page);
4495 set_btree_ioerr(page, eb);
4498 end_page_writeback(page);
4503 end_extent_buffer_writeback(eb);
4509 static void prepare_eb_write(struct extent_buffer *eb)
4512 unsigned long start;
4515 clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
4516 atomic_set(&eb->io_pages, num_extent_pages(eb));
4518 /* Set btree blocks beyond nritems with 0 to avoid stale content */
4519 nritems = btrfs_header_nritems(eb);
4520 if (btrfs_header_level(eb) > 0) {
4521 end = btrfs_node_key_ptr_offset(nritems);
4522 memzero_extent_buffer(eb, end, eb->len - end);
4526 * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0
4528 start = btrfs_item_nr_offset(nritems);
4529 end = BTRFS_LEAF_DATA_OFFSET + leaf_data_end(eb);
4530 memzero_extent_buffer(eb, start, end - start);
4535 * Unlike the work in write_one_eb(), we rely completely on extent locking.
4536 * Page locking is only utilized at minimum to keep the VMM code happy.
4538 static int write_one_subpage_eb(struct extent_buffer *eb,
4539 struct writeback_control *wbc,
4540 struct extent_page_data *epd)
4542 struct btrfs_fs_info *fs_info = eb->fs_info;
4543 struct page *page = eb->pages[0];
4544 unsigned int write_flags = wbc_to_write_flags(wbc);
4545 bool no_dirty_ebs = false;
4548 prepare_eb_write(eb);
4550 /* clear_page_dirty_for_io() in subpage helper needs page locked */
4552 btrfs_subpage_set_writeback(fs_info, page, eb->start, eb->len);
4554 /* Check if this is the last dirty bit to update nr_written */
4555 no_dirty_ebs = btrfs_subpage_clear_and_test_dirty(fs_info, page,
4556 eb->start, eb->len);
4558 clear_page_dirty_for_io(page);
4560 ret = submit_extent_page(REQ_OP_WRITE | write_flags, wbc,
4561 &epd->bio_ctrl, page, eb->start, eb->len,
4562 eb->start - page_offset(page),
4563 end_bio_subpage_eb_writepage, 0, false);
4565 btrfs_subpage_clear_writeback(fs_info, page, eb->start, eb->len);
4566 set_btree_ioerr(page, eb);
4569 if (atomic_dec_and_test(&eb->io_pages))
4570 end_extent_buffer_writeback(eb);
4575 * Submission finished without problem, if no range of the page is
4576 * dirty anymore, we have submitted a page. Update nr_written in wbc.
4583 static noinline_for_stack int write_one_eb(struct extent_buffer *eb,
4584 struct writeback_control *wbc,
4585 struct extent_page_data *epd)
4587 u64 disk_bytenr = eb->start;
4589 unsigned int write_flags = wbc_to_write_flags(wbc);
4592 prepare_eb_write(eb);
4594 num_pages = num_extent_pages(eb);
4595 for (i = 0; i < num_pages; i++) {
4596 struct page *p = eb->pages[i];
4598 clear_page_dirty_for_io(p);
4599 set_page_writeback(p);
4600 ret = submit_extent_page(REQ_OP_WRITE | write_flags, wbc,
4601 &epd->bio_ctrl, p, disk_bytenr,
4603 end_bio_extent_buffer_writepage,
4606 set_btree_ioerr(p, eb);
4607 if (PageWriteback(p))
4608 end_page_writeback(p);
4609 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
4610 end_extent_buffer_writeback(eb);
4614 disk_bytenr += PAGE_SIZE;
4619 if (unlikely(ret)) {
4620 for (; i < num_pages; i++) {
4621 struct page *p = eb->pages[i];
4622 clear_page_dirty_for_io(p);
4631 * Submit one subpage btree page.
4633 * The main difference to submit_eb_page() is:
4635 * For subpage, we don't rely on page locking at all.
4638 * We only flush bio if we may be unable to fit current extent buffers into
4641 * Return >=0 for the number of submitted extent buffers.
4642 * Return <0 for fatal error.
4644 static int submit_eb_subpage(struct page *page,
4645 struct writeback_control *wbc,
4646 struct extent_page_data *epd)
4648 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
4650 u64 page_start = page_offset(page);
4652 int sectors_per_node = fs_info->nodesize >> fs_info->sectorsize_bits;
4655 /* Lock and write each dirty extent buffers in the range */
4656 while (bit_start < fs_info->subpage_info->bitmap_nr_bits) {
4657 struct btrfs_subpage *subpage = (struct btrfs_subpage *)page->private;
4658 struct extent_buffer *eb;
4659 unsigned long flags;
4663 * Take private lock to ensure the subpage won't be detached
4666 spin_lock(&page->mapping->private_lock);
4667 if (!PagePrivate(page)) {
4668 spin_unlock(&page->mapping->private_lock);
4671 spin_lock_irqsave(&subpage->lock, flags);
4672 if (!test_bit(bit_start + fs_info->subpage_info->dirty_offset,
4673 subpage->bitmaps)) {
4674 spin_unlock_irqrestore(&subpage->lock, flags);
4675 spin_unlock(&page->mapping->private_lock);
4680 start = page_start + bit_start * fs_info->sectorsize;
4681 bit_start += sectors_per_node;
4684 * Here we just want to grab the eb without touching extra
4685 * spin locks, so call find_extent_buffer_nolock().
4687 eb = find_extent_buffer_nolock(fs_info, start);
4688 spin_unlock_irqrestore(&subpage->lock, flags);
4689 spin_unlock(&page->mapping->private_lock);
4692 * The eb has already reached 0 refs thus find_extent_buffer()
4693 * doesn't return it. We don't need to write back such eb
4699 ret = lock_extent_buffer_for_io(eb, epd);
4701 free_extent_buffer(eb);
4705 free_extent_buffer(eb);
4708 ret = write_one_subpage_eb(eb, wbc, epd);
4709 free_extent_buffer(eb);
4717 /* We hit error, end bio for the submitted extent buffers */
4718 submit_write_bio(epd, ret);
4723 * Submit all page(s) of one extent buffer.
4725 * @page: the page of one extent buffer
4726 * @eb_context: to determine if we need to submit this page, if current page
4727 * belongs to this eb, we don't need to submit
4729 * The caller should pass each page in their bytenr order, and here we use
4730 * @eb_context to determine if we have submitted pages of one extent buffer.
4732 * If we have, we just skip until we hit a new page that doesn't belong to
4733 * current @eb_context.
4735 * If not, we submit all the page(s) of the extent buffer.
4737 * Return >0 if we have submitted the extent buffer successfully.
4738 * Return 0 if we don't need to submit the page, as it's already submitted by
4740 * Return <0 for fatal error.
4742 static int submit_eb_page(struct page *page, struct writeback_control *wbc,
4743 struct extent_page_data *epd,
4744 struct extent_buffer **eb_context)
4746 struct address_space *mapping = page->mapping;
4747 struct btrfs_block_group *cache = NULL;
4748 struct extent_buffer *eb;
4751 if (!PagePrivate(page))
4754 if (btrfs_sb(page->mapping->host->i_sb)->nodesize < PAGE_SIZE)
4755 return submit_eb_subpage(page, wbc, epd);
4757 spin_lock(&mapping->private_lock);
4758 if (!PagePrivate(page)) {
4759 spin_unlock(&mapping->private_lock);
4763 eb = (struct extent_buffer *)page->private;
4766 * Shouldn't happen and normally this would be a BUG_ON but no point
4767 * crashing the machine for something we can survive anyway.
4770 spin_unlock(&mapping->private_lock);
4774 if (eb == *eb_context) {
4775 spin_unlock(&mapping->private_lock);
4778 ret = atomic_inc_not_zero(&eb->refs);
4779 spin_unlock(&mapping->private_lock);
4783 if (!btrfs_check_meta_write_pointer(eb->fs_info, eb, &cache)) {
4785 * If for_sync, this hole will be filled with
4786 * trasnsaction commit.
4788 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
4792 free_extent_buffer(eb);
4798 ret = lock_extent_buffer_for_io(eb, epd);
4800 btrfs_revert_meta_write_pointer(cache, eb);
4802 btrfs_put_block_group(cache);
4803 free_extent_buffer(eb);
4808 * Implies write in zoned mode. Mark the last eb in a block group.
4810 btrfs_schedule_zone_finish_bg(cache, eb);
4811 btrfs_put_block_group(cache);
4813 ret = write_one_eb(eb, wbc, epd);
4814 free_extent_buffer(eb);
4820 int btree_write_cache_pages(struct address_space *mapping,
4821 struct writeback_control *wbc)
4823 struct extent_buffer *eb_context = NULL;
4824 struct extent_page_data epd = {
4827 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4829 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
4832 int nr_to_write_done = 0;
4833 struct pagevec pvec;
4836 pgoff_t end; /* Inclusive */
4840 pagevec_init(&pvec);
4841 if (wbc->range_cyclic) {
4842 index = mapping->writeback_index; /* Start from prev offset */
4845 * Start from the beginning does not need to cycle over the
4846 * range, mark it as scanned.
4848 scanned = (index == 0);
4850 index = wbc->range_start >> PAGE_SHIFT;
4851 end = wbc->range_end >> PAGE_SHIFT;
4854 if (wbc->sync_mode == WB_SYNC_ALL)
4855 tag = PAGECACHE_TAG_TOWRITE;
4857 tag = PAGECACHE_TAG_DIRTY;
4858 btrfs_zoned_meta_io_lock(fs_info);
4860 if (wbc->sync_mode == WB_SYNC_ALL)
4861 tag_pages_for_writeback(mapping, index, end);
4862 while (!done && !nr_to_write_done && (index <= end) &&
4863 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
4867 for (i = 0; i < nr_pages; i++) {
4868 struct page *page = pvec.pages[i];
4870 ret = submit_eb_page(page, wbc, &epd, &eb_context);
4879 * the filesystem may choose to bump up nr_to_write.
4880 * We have to make sure to honor the new nr_to_write
4883 nr_to_write_done = wbc->nr_to_write <= 0;
4885 pagevec_release(&pvec);
4888 if (!scanned && !done) {
4890 * We hit the last page and there is more work to be done: wrap
4891 * back to the start of the file
4898 * If something went wrong, don't allow any metadata write bio to be
4901 * This would prevent use-after-free if we had dirty pages not
4902 * cleaned up, which can still happen by fuzzed images.
4905 * Allowing existing tree block to be allocated for other trees.
4907 * - Log tree operations
4908 * Exiting tree blocks get allocated to log tree, bumps its
4909 * generation, then get cleaned in tree re-balance.
4910 * Such tree block will not be written back, since it's clean,
4911 * thus no WRITTEN flag set.
4912 * And after log writes back, this tree block is not traced by
4913 * any dirty extent_io_tree.
4915 * - Offending tree block gets re-dirtied from its original owner
4916 * Since it has bumped generation, no WRITTEN flag, it can be
4917 * reused without COWing. This tree block will not be traced
4918 * by btrfs_transaction::dirty_pages.
4920 * Now such dirty tree block will not be cleaned by any dirty
4921 * extent io tree. Thus we don't want to submit such wild eb
4922 * if the fs already has error.
4924 * We can get ret > 0 from submit_extent_page() indicating how many ebs
4925 * were submitted. Reset it to 0 to avoid false alerts for the caller.
4929 if (!ret && BTRFS_FS_ERROR(fs_info))
4931 submit_write_bio(&epd, ret);
4933 btrfs_zoned_meta_io_unlock(fs_info);
4938 * Walk the list of dirty pages of the given address space and write all of them.
4940 * @mapping: address space structure to write
4941 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
4942 * @epd: holds context for the write, namely the bio
4944 * If a page is already under I/O, write_cache_pages() skips it, even
4945 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
4946 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
4947 * and msync() need to guarantee that all the data which was dirty at the time
4948 * the call was made get new I/O started against them. If wbc->sync_mode is
4949 * WB_SYNC_ALL then we were called for data integrity and we must wait for
4950 * existing IO to complete.
4952 static int extent_write_cache_pages(struct address_space *mapping,
4953 struct writeback_control *wbc,
4954 struct extent_page_data *epd)
4956 struct inode *inode = mapping->host;
4959 int nr_to_write_done = 0;
4960 struct pagevec pvec;
4963 pgoff_t end; /* Inclusive */
4965 int range_whole = 0;
4970 * We have to hold onto the inode so that ordered extents can do their
4971 * work when the IO finishes. The alternative to this is failing to add
4972 * an ordered extent if the igrab() fails there and that is a huge pain
4973 * to deal with, so instead just hold onto the inode throughout the
4974 * writepages operation. If it fails here we are freeing up the inode
4975 * anyway and we'd rather not waste our time writing out stuff that is
4976 * going to be truncated anyway.
4981 pagevec_init(&pvec);
4982 if (wbc->range_cyclic) {
4983 index = mapping->writeback_index; /* Start from prev offset */
4986 * Start from the beginning does not need to cycle over the
4987 * range, mark it as scanned.
4989 scanned = (index == 0);
4991 index = wbc->range_start >> PAGE_SHIFT;
4992 end = wbc->range_end >> PAGE_SHIFT;
4993 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
4999 * We do the tagged writepage as long as the snapshot flush bit is set
5000 * and we are the first one who do the filemap_flush() on this inode.
5002 * The nr_to_write == LONG_MAX is needed to make sure other flushers do
5003 * not race in and drop the bit.
5005 if (range_whole && wbc->nr_to_write == LONG_MAX &&
5006 test_and_clear_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
5007 &BTRFS_I(inode)->runtime_flags))
5008 wbc->tagged_writepages = 1;
5010 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
5011 tag = PAGECACHE_TAG_TOWRITE;
5013 tag = PAGECACHE_TAG_DIRTY;
5015 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
5016 tag_pages_for_writeback(mapping, index, end);
5018 while (!done && !nr_to_write_done && (index <= end) &&
5019 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping,
5020 &index, end, tag))) {
5023 for (i = 0; i < nr_pages; i++) {
5024 struct page *page = pvec.pages[i];
5026 done_index = page->index + 1;
5028 * At this point we hold neither the i_pages lock nor
5029 * the page lock: the page may be truncated or
5030 * invalidated (changing page->mapping to NULL),
5031 * or even swizzled back from swapper_space to
5032 * tmpfs file mapping
5034 if (!trylock_page(page)) {
5035 submit_write_bio(epd, 0);
5039 if (unlikely(page->mapping != mapping)) {
5044 if (wbc->sync_mode != WB_SYNC_NONE) {
5045 if (PageWriteback(page))
5046 submit_write_bio(epd, 0);
5047 wait_on_page_writeback(page);
5050 if (PageWriteback(page) ||
5051 !clear_page_dirty_for_io(page)) {
5056 ret = __extent_writepage(page, wbc, epd);
5063 * the filesystem may choose to bump up nr_to_write.
5064 * We have to make sure to honor the new nr_to_write
5067 nr_to_write_done = wbc->nr_to_write <= 0;
5069 pagevec_release(&pvec);
5072 if (!scanned && !done) {
5074 * We hit the last page and there is more work to be done: wrap
5075 * back to the start of the file
5081 * If we're looping we could run into a page that is locked by a
5082 * writer and that writer could be waiting on writeback for a
5083 * page in our current bio, and thus deadlock, so flush the
5086 submit_write_bio(epd, 0);
5090 if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole))
5091 mapping->writeback_index = done_index;
5093 btrfs_add_delayed_iput(inode);
5097 int extent_write_full_page(struct page *page, struct writeback_control *wbc)
5100 struct extent_page_data epd = {
5103 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
5106 ret = __extent_writepage(page, wbc, &epd);
5107 submit_write_bio(&epd, ret);
5112 * Submit the pages in the range to bio for call sites which delalloc range has
5113 * already been ran (aka, ordered extent inserted) and all pages are still
5116 int extent_write_locked_range(struct inode *inode, u64 start, u64 end)
5118 bool found_error = false;
5119 int first_error = 0;
5121 struct address_space *mapping = inode->i_mapping;
5124 unsigned long nr_pages;
5125 const u32 sectorsize = btrfs_sb(inode->i_sb)->sectorsize;
5126 struct extent_page_data epd = {
5131 struct writeback_control wbc_writepages = {
5132 .sync_mode = WB_SYNC_ALL,
5133 .range_start = start,
5134 .range_end = end + 1,
5135 /* We're called from an async helper function */
5136 .punt_to_cgroup = 1,
5137 .no_cgroup_owner = 1,
5140 ASSERT(IS_ALIGNED(start, sectorsize) && IS_ALIGNED(end + 1, sectorsize));
5141 nr_pages = (round_up(end, PAGE_SIZE) - round_down(start, PAGE_SIZE)) >>
5143 wbc_writepages.nr_to_write = nr_pages * 2;
5145 wbc_attach_fdatawrite_inode(&wbc_writepages, inode);
5146 while (cur <= end) {
5147 u64 cur_end = min(round_down(cur, PAGE_SIZE) + PAGE_SIZE - 1, end);
5149 page = find_get_page(mapping, cur >> PAGE_SHIFT);
5151 * All pages in the range are locked since
5152 * btrfs_run_delalloc_range(), thus there is no way to clear
5153 * the page dirty flag.
5155 ASSERT(PageLocked(page));
5156 ASSERT(PageDirty(page));
5157 clear_page_dirty_for_io(page);
5158 ret = __extent_writepage(page, &wbc_writepages, &epd);
5168 submit_write_bio(&epd, found_error ? ret : 0);
5170 wbc_detach_inode(&wbc_writepages);
5176 int extent_writepages(struct address_space *mapping,
5177 struct writeback_control *wbc)
5179 struct inode *inode = mapping->host;
5181 struct extent_page_data epd = {
5184 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
5188 * Allow only a single thread to do the reloc work in zoned mode to
5189 * protect the write pointer updates.
5191 btrfs_zoned_data_reloc_lock(BTRFS_I(inode));
5192 ret = extent_write_cache_pages(mapping, wbc, &epd);
5193 submit_write_bio(&epd, ret);
5194 btrfs_zoned_data_reloc_unlock(BTRFS_I(inode));
5198 void extent_readahead(struct readahead_control *rac)
5200 struct btrfs_bio_ctrl bio_ctrl = { 0 };
5201 struct page *pagepool[16];
5202 struct extent_map *em_cached = NULL;
5203 u64 prev_em_start = (u64)-1;
5206 while ((nr = readahead_page_batch(rac, pagepool))) {
5207 u64 contig_start = readahead_pos(rac);
5208 u64 contig_end = contig_start + readahead_batch_length(rac) - 1;
5210 contiguous_readpages(pagepool, nr, contig_start, contig_end,
5211 &em_cached, &bio_ctrl, &prev_em_start);
5215 free_extent_map(em_cached);
5216 submit_one_bio(&bio_ctrl);
5220 * basic invalidate_folio code, this waits on any locked or writeback
5221 * ranges corresponding to the folio, and then deletes any extent state
5222 * records from the tree
5224 int extent_invalidate_folio(struct extent_io_tree *tree,
5225 struct folio *folio, size_t offset)
5227 struct extent_state *cached_state = NULL;
5228 u64 start = folio_pos(folio);
5229 u64 end = start + folio_size(folio) - 1;
5230 size_t blocksize = folio->mapping->host->i_sb->s_blocksize;
5232 /* This function is only called for the btree inode */
5233 ASSERT(tree->owner == IO_TREE_BTREE_INODE_IO);
5235 start += ALIGN(offset, blocksize);
5239 lock_extent_bits(tree, start, end, &cached_state);
5240 folio_wait_writeback(folio);
5243 * Currently for btree io tree, only EXTENT_LOCKED is utilized,
5244 * so here we only need to unlock the extent range to free any
5245 * existing extent state.
5247 unlock_extent_cached(tree, start, end, &cached_state);
5252 * a helper for release_folio, this tests for areas of the page that
5253 * are locked or under IO and drops the related state bits if it is safe
5256 static int try_release_extent_state(struct extent_io_tree *tree,
5257 struct page *page, gfp_t mask)
5259 u64 start = page_offset(page);
5260 u64 end = start + PAGE_SIZE - 1;
5263 if (test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL)) {
5267 * At this point we can safely clear everything except the
5268 * locked bit, the nodatasum bit and the delalloc new bit.
5269 * The delalloc new bit will be cleared by ordered extent
5272 ret = __clear_extent_bit(tree, start, end,
5273 ~(EXTENT_LOCKED | EXTENT_NODATASUM | EXTENT_DELALLOC_NEW),
5274 0, 0, NULL, mask, NULL);
5276 /* if clear_extent_bit failed for enomem reasons,
5277 * we can't allow the release to continue.
5288 * a helper for release_folio. As long as there are no locked extents
5289 * in the range corresponding to the page, both state records and extent
5290 * map records are removed
5292 int try_release_extent_mapping(struct page *page, gfp_t mask)
5294 struct extent_map *em;
5295 u64 start = page_offset(page);
5296 u64 end = start + PAGE_SIZE - 1;
5297 struct btrfs_inode *btrfs_inode = BTRFS_I(page->mapping->host);
5298 struct extent_io_tree *tree = &btrfs_inode->io_tree;
5299 struct extent_map_tree *map = &btrfs_inode->extent_tree;
5301 if (gfpflags_allow_blocking(mask) &&
5302 page->mapping->host->i_size > SZ_16M) {
5304 while (start <= end) {
5305 struct btrfs_fs_info *fs_info;
5308 len = end - start + 1;
5309 write_lock(&map->lock);
5310 em = lookup_extent_mapping(map, start, len);
5312 write_unlock(&map->lock);
5315 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
5316 em->start != start) {
5317 write_unlock(&map->lock);
5318 free_extent_map(em);
5321 if (test_range_bit(tree, em->start,
5322 extent_map_end(em) - 1,
5323 EXTENT_LOCKED, 0, NULL))
5326 * If it's not in the list of modified extents, used
5327 * by a fast fsync, we can remove it. If it's being
5328 * logged we can safely remove it since fsync took an
5329 * extra reference on the em.
5331 if (list_empty(&em->list) ||
5332 test_bit(EXTENT_FLAG_LOGGING, &em->flags))
5335 * If it's in the list of modified extents, remove it
5336 * only if its generation is older then the current one,
5337 * in which case we don't need it for a fast fsync.
5338 * Otherwise don't remove it, we could be racing with an
5339 * ongoing fast fsync that could miss the new extent.
5341 fs_info = btrfs_inode->root->fs_info;
5342 spin_lock(&fs_info->trans_lock);
5343 cur_gen = fs_info->generation;
5344 spin_unlock(&fs_info->trans_lock);
5345 if (em->generation >= cur_gen)
5349 * We only remove extent maps that are not in the list of
5350 * modified extents or that are in the list but with a
5351 * generation lower then the current generation, so there
5352 * is no need to set the full fsync flag on the inode (it
5353 * hurts the fsync performance for workloads with a data
5354 * size that exceeds or is close to the system's memory).
5356 remove_extent_mapping(map, em);
5357 /* once for the rb tree */
5358 free_extent_map(em);
5360 start = extent_map_end(em);
5361 write_unlock(&map->lock);
5364 free_extent_map(em);
5366 cond_resched(); /* Allow large-extent preemption. */
5369 return try_release_extent_state(tree, page, mask);
5373 * helper function for fiemap, which doesn't want to see any holes.
5374 * This maps until we find something past 'last'
5376 static struct extent_map *get_extent_skip_holes(struct btrfs_inode *inode,
5377 u64 offset, u64 last)
5379 u64 sectorsize = btrfs_inode_sectorsize(inode);
5380 struct extent_map *em;
5387 len = last - offset;
5390 len = ALIGN(len, sectorsize);
5391 em = btrfs_get_extent_fiemap(inode, offset, len);
5395 /* if this isn't a hole return it */
5396 if (em->block_start != EXTENT_MAP_HOLE)
5399 /* this is a hole, advance to the next extent */
5400 offset = extent_map_end(em);
5401 free_extent_map(em);
5409 * To cache previous fiemap extent
5411 * Will be used for merging fiemap extent
5413 struct fiemap_cache {
5422 * Helper to submit fiemap extent.
5424 * Will try to merge current fiemap extent specified by @offset, @phys,
5425 * @len and @flags with cached one.
5426 * And only when we fails to merge, cached one will be submitted as
5429 * Return value is the same as fiemap_fill_next_extent().
5431 static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
5432 struct fiemap_cache *cache,
5433 u64 offset, u64 phys, u64 len, u32 flags)
5441 * Sanity check, extent_fiemap() should have ensured that new
5442 * fiemap extent won't overlap with cached one.
5445 * NOTE: Physical address can overlap, due to compression
5447 if (cache->offset + cache->len > offset) {
5453 * Only merges fiemap extents if
5454 * 1) Their logical addresses are continuous
5456 * 2) Their physical addresses are continuous
5457 * So truly compressed (physical size smaller than logical size)
5458 * extents won't get merged with each other
5460 * 3) Share same flags except FIEMAP_EXTENT_LAST
5461 * So regular extent won't get merged with prealloc extent
5463 if (cache->offset + cache->len == offset &&
5464 cache->phys + cache->len == phys &&
5465 (cache->flags & ~FIEMAP_EXTENT_LAST) ==
5466 (flags & ~FIEMAP_EXTENT_LAST)) {
5468 cache->flags |= flags;
5469 goto try_submit_last;
5472 /* Not mergeable, need to submit cached one */
5473 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
5474 cache->len, cache->flags);
5475 cache->cached = false;
5479 cache->cached = true;
5480 cache->offset = offset;
5483 cache->flags = flags;
5485 if (cache->flags & FIEMAP_EXTENT_LAST) {
5486 ret = fiemap_fill_next_extent(fieinfo, cache->offset,
5487 cache->phys, cache->len, cache->flags);
5488 cache->cached = false;
5494 * Emit last fiemap cache
5496 * The last fiemap cache may still be cached in the following case:
5498 * |<- Fiemap range ->|
5499 * |<------------ First extent ----------->|
5501 * In this case, the first extent range will be cached but not emitted.
5502 * So we must emit it before ending extent_fiemap().
5504 static int emit_last_fiemap_cache(struct fiemap_extent_info *fieinfo,
5505 struct fiemap_cache *cache)
5512 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
5513 cache->len, cache->flags);
5514 cache->cached = false;
5520 int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,
5525 u64 max = start + len;
5529 u64 last_for_get_extent = 0;
5531 u64 isize = i_size_read(&inode->vfs_inode);
5532 struct btrfs_key found_key;
5533 struct extent_map *em = NULL;
5534 struct extent_state *cached_state = NULL;
5535 struct btrfs_path *path;
5536 struct btrfs_root *root = inode->root;
5537 struct fiemap_cache cache = { 0 };
5538 struct ulist *roots;
5539 struct ulist *tmp_ulist;
5548 path = btrfs_alloc_path();
5552 roots = ulist_alloc(GFP_KERNEL);
5553 tmp_ulist = ulist_alloc(GFP_KERNEL);
5554 if (!roots || !tmp_ulist) {
5556 goto out_free_ulist;
5560 * We can't initialize that to 'start' as this could miss extents due
5561 * to extent item merging
5564 start = round_down(start, btrfs_inode_sectorsize(inode));
5565 len = round_up(max, btrfs_inode_sectorsize(inode)) - start;
5568 * lookup the last file extent. We're not using i_size here
5569 * because there might be preallocation past i_size
5571 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode), -1,
5574 goto out_free_ulist;
5582 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
5583 found_type = found_key.type;
5585 /* No extents, but there might be delalloc bits */
5586 if (found_key.objectid != btrfs_ino(inode) ||
5587 found_type != BTRFS_EXTENT_DATA_KEY) {
5588 /* have to trust i_size as the end */
5590 last_for_get_extent = isize;
5593 * remember the start of the last extent. There are a
5594 * bunch of different factors that go into the length of the
5595 * extent, so its much less complex to remember where it started
5597 last = found_key.offset;
5598 last_for_get_extent = last + 1;
5600 btrfs_release_path(path);
5603 * we might have some extents allocated but more delalloc past those
5604 * extents. so, we trust isize unless the start of the last extent is
5609 last_for_get_extent = isize;
5612 lock_extent_bits(&inode->io_tree, start, start + len - 1,
5615 em = get_extent_skip_holes(inode, start, last_for_get_extent);
5624 u64 offset_in_extent = 0;
5626 /* break if the extent we found is outside the range */
5627 if (em->start >= max || extent_map_end(em) < off)
5631 * get_extent may return an extent that starts before our
5632 * requested range. We have to make sure the ranges
5633 * we return to fiemap always move forward and don't
5634 * overlap, so adjust the offsets here
5636 em_start = max(em->start, off);
5639 * record the offset from the start of the extent
5640 * for adjusting the disk offset below. Only do this if the
5641 * extent isn't compressed since our in ram offset may be past
5642 * what we have actually allocated on disk.
5644 if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
5645 offset_in_extent = em_start - em->start;
5646 em_end = extent_map_end(em);
5647 em_len = em_end - em_start;
5649 if (em->block_start < EXTENT_MAP_LAST_BYTE)
5650 disko = em->block_start + offset_in_extent;
5655 * bump off for our next call to get_extent
5657 off = extent_map_end(em);
5661 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
5663 flags |= FIEMAP_EXTENT_LAST;
5664 } else if (em->block_start == EXTENT_MAP_INLINE) {
5665 flags |= (FIEMAP_EXTENT_DATA_INLINE |
5666 FIEMAP_EXTENT_NOT_ALIGNED);
5667 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
5668 flags |= (FIEMAP_EXTENT_DELALLOC |
5669 FIEMAP_EXTENT_UNKNOWN);
5670 } else if (fieinfo->fi_extents_max) {
5671 u64 bytenr = em->block_start -
5672 (em->start - em->orig_start);
5675 * As btrfs supports shared space, this information
5676 * can be exported to userspace tools via
5677 * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0
5678 * then we're just getting a count and we can skip the
5681 ret = btrfs_check_shared(root, btrfs_ino(inode),
5682 bytenr, roots, tmp_ulist);
5686 flags |= FIEMAP_EXTENT_SHARED;
5689 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
5690 flags |= FIEMAP_EXTENT_ENCODED;
5691 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5692 flags |= FIEMAP_EXTENT_UNWRITTEN;
5694 free_extent_map(em);
5696 if ((em_start >= last) || em_len == (u64)-1 ||
5697 (last == (u64)-1 && isize <= em_end)) {
5698 flags |= FIEMAP_EXTENT_LAST;
5702 /* now scan forward to see if this is really the last extent. */
5703 em = get_extent_skip_holes(inode, off, last_for_get_extent);
5709 flags |= FIEMAP_EXTENT_LAST;
5712 ret = emit_fiemap_extent(fieinfo, &cache, em_start, disko,
5722 ret = emit_last_fiemap_cache(fieinfo, &cache);
5723 free_extent_map(em);
5725 unlock_extent_cached(&inode->io_tree, start, start + len - 1,
5729 btrfs_free_path(path);
5731 ulist_free(tmp_ulist);
5735 static void __free_extent_buffer(struct extent_buffer *eb)
5737 kmem_cache_free(extent_buffer_cache, eb);
5740 int extent_buffer_under_io(const struct extent_buffer *eb)
5742 return (atomic_read(&eb->io_pages) ||
5743 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
5744 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
5747 static bool page_range_has_eb(struct btrfs_fs_info *fs_info, struct page *page)
5749 struct btrfs_subpage *subpage;
5751 lockdep_assert_held(&page->mapping->private_lock);
5753 if (PagePrivate(page)) {
5754 subpage = (struct btrfs_subpage *)page->private;
5755 if (atomic_read(&subpage->eb_refs))
5758 * Even there is no eb refs here, we may still have
5759 * end_page_read() call relying on page::private.
5761 if (atomic_read(&subpage->readers))
5767 static void detach_extent_buffer_page(struct extent_buffer *eb, struct page *page)
5769 struct btrfs_fs_info *fs_info = eb->fs_info;
5770 const bool mapped = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
5773 * For mapped eb, we're going to change the page private, which should
5774 * be done under the private_lock.
5777 spin_lock(&page->mapping->private_lock);
5779 if (!PagePrivate(page)) {
5781 spin_unlock(&page->mapping->private_lock);
5785 if (fs_info->nodesize >= PAGE_SIZE) {
5787 * We do this since we'll remove the pages after we've
5788 * removed the eb from the radix tree, so we could race
5789 * and have this page now attached to the new eb. So
5790 * only clear page_private if it's still connected to
5793 if (PagePrivate(page) &&
5794 page->private == (unsigned long)eb) {
5795 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
5796 BUG_ON(PageDirty(page));
5797 BUG_ON(PageWriteback(page));
5799 * We need to make sure we haven't be attached
5802 detach_page_private(page);
5805 spin_unlock(&page->mapping->private_lock);
5810 * For subpage, we can have dummy eb with page private. In this case,
5811 * we can directly detach the private as such page is only attached to
5812 * one dummy eb, no sharing.
5815 btrfs_detach_subpage(fs_info, page);
5819 btrfs_page_dec_eb_refs(fs_info, page);
5822 * We can only detach the page private if there are no other ebs in the
5823 * page range and no unfinished IO.
5825 if (!page_range_has_eb(fs_info, page))
5826 btrfs_detach_subpage(fs_info, page);
5828 spin_unlock(&page->mapping->private_lock);
5831 /* Release all pages attached to the extent buffer */
5832 static void btrfs_release_extent_buffer_pages(struct extent_buffer *eb)
5837 ASSERT(!extent_buffer_under_io(eb));
5839 num_pages = num_extent_pages(eb);
5840 for (i = 0; i < num_pages; i++) {
5841 struct page *page = eb->pages[i];
5846 detach_extent_buffer_page(eb, page);
5848 /* One for when we allocated the page */
5854 * Helper for releasing the extent buffer.
5856 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
5858 btrfs_release_extent_buffer_pages(eb);
5859 btrfs_leak_debug_del(&eb->fs_info->eb_leak_lock, &eb->leak_list);
5860 __free_extent_buffer(eb);
5863 static struct extent_buffer *
5864 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
5867 struct extent_buffer *eb = NULL;
5869 eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL);
5872 eb->fs_info = fs_info;
5874 init_rwsem(&eb->lock);
5876 btrfs_leak_debug_add(&fs_info->eb_leak_lock, &eb->leak_list,
5877 &fs_info->allocated_ebs);
5878 INIT_LIST_HEAD(&eb->release_list);
5880 spin_lock_init(&eb->refs_lock);
5881 atomic_set(&eb->refs, 1);
5882 atomic_set(&eb->io_pages, 0);
5884 ASSERT(len <= BTRFS_MAX_METADATA_BLOCKSIZE);
5889 struct extent_buffer *btrfs_clone_extent_buffer(const struct extent_buffer *src)
5892 struct extent_buffer *new;
5893 int num_pages = num_extent_pages(src);
5896 new = __alloc_extent_buffer(src->fs_info, src->start, src->len);
5901 * Set UNMAPPED before calling btrfs_release_extent_buffer(), as
5902 * btrfs_release_extent_buffer() have different behavior for
5903 * UNMAPPED subpage extent buffer.
5905 set_bit(EXTENT_BUFFER_UNMAPPED, &new->bflags);
5907 memset(new->pages, 0, sizeof(*new->pages) * num_pages);
5908 ret = btrfs_alloc_page_array(num_pages, new->pages);
5910 btrfs_release_extent_buffer(new);
5914 for (i = 0; i < num_pages; i++) {
5916 struct page *p = new->pages[i];
5918 ret = attach_extent_buffer_page(new, p, NULL);
5920 btrfs_release_extent_buffer(new);
5923 WARN_ON(PageDirty(p));
5924 copy_page(page_address(p), page_address(src->pages[i]));
5926 set_extent_buffer_uptodate(new);
5931 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
5932 u64 start, unsigned long len)
5934 struct extent_buffer *eb;
5939 eb = __alloc_extent_buffer(fs_info, start, len);
5943 num_pages = num_extent_pages(eb);
5944 ret = btrfs_alloc_page_array(num_pages, eb->pages);
5948 for (i = 0; i < num_pages; i++) {
5949 struct page *p = eb->pages[i];
5951 ret = attach_extent_buffer_page(eb, p, NULL);
5956 set_extent_buffer_uptodate(eb);
5957 btrfs_set_header_nritems(eb, 0);
5958 set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
5962 for (i = 0; i < num_pages; i++) {
5964 detach_extent_buffer_page(eb, eb->pages[i]);
5965 __free_page(eb->pages[i]);
5968 __free_extent_buffer(eb);
5972 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
5975 return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize);
5978 static void check_buffer_tree_ref(struct extent_buffer *eb)
5982 * The TREE_REF bit is first set when the extent_buffer is added
5983 * to the radix tree. It is also reset, if unset, when a new reference
5984 * is created by find_extent_buffer.
5986 * It is only cleared in two cases: freeing the last non-tree
5987 * reference to the extent_buffer when its STALE bit is set or
5988 * calling release_folio when the tree reference is the only reference.
5990 * In both cases, care is taken to ensure that the extent_buffer's
5991 * pages are not under io. However, release_folio can be concurrently
5992 * called with creating new references, which is prone to race
5993 * conditions between the calls to check_buffer_tree_ref in those
5994 * codepaths and clearing TREE_REF in try_release_extent_buffer.
5996 * The actual lifetime of the extent_buffer in the radix tree is
5997 * adequately protected by the refcount, but the TREE_REF bit and
5998 * its corresponding reference are not. To protect against this
5999 * class of races, we call check_buffer_tree_ref from the codepaths
6000 * which trigger io after they set eb->io_pages. Note that once io is
6001 * initiated, TREE_REF can no longer be cleared, so that is the
6002 * moment at which any such race is best fixed.
6004 refs = atomic_read(&eb->refs);
6005 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
6008 spin_lock(&eb->refs_lock);
6009 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
6010 atomic_inc(&eb->refs);
6011 spin_unlock(&eb->refs_lock);
6014 static void mark_extent_buffer_accessed(struct extent_buffer *eb,
6015 struct page *accessed)
6019 check_buffer_tree_ref(eb);
6021 num_pages = num_extent_pages(eb);
6022 for (i = 0; i < num_pages; i++) {
6023 struct page *p = eb->pages[i];
6026 mark_page_accessed(p);
6030 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
6033 struct extent_buffer *eb;
6035 eb = find_extent_buffer_nolock(fs_info, start);
6039 * Lock our eb's refs_lock to avoid races with free_extent_buffer().
6040 * When we get our eb it might be flagged with EXTENT_BUFFER_STALE and
6041 * another task running free_extent_buffer() might have seen that flag
6042 * set, eb->refs == 2, that the buffer isn't under IO (dirty and
6043 * writeback flags not set) and it's still in the tree (flag
6044 * EXTENT_BUFFER_TREE_REF set), therefore being in the process of
6045 * decrementing the extent buffer's reference count twice. So here we
6046 * could race and increment the eb's reference count, clear its stale
6047 * flag, mark it as dirty and drop our reference before the other task
6048 * finishes executing free_extent_buffer, which would later result in
6049 * an attempt to free an extent buffer that is dirty.
6051 if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) {
6052 spin_lock(&eb->refs_lock);
6053 spin_unlock(&eb->refs_lock);
6055 mark_extent_buffer_accessed(eb, NULL);
6059 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
6060 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
6063 struct extent_buffer *eb, *exists = NULL;
6066 eb = find_extent_buffer(fs_info, start);
6069 eb = alloc_dummy_extent_buffer(fs_info, start);
6071 return ERR_PTR(-ENOMEM);
6072 eb->fs_info = fs_info;
6074 ret = radix_tree_preload(GFP_NOFS);
6076 exists = ERR_PTR(ret);
6079 spin_lock(&fs_info->buffer_lock);
6080 ret = radix_tree_insert(&fs_info->buffer_radix,
6081 start >> fs_info->sectorsize_bits, eb);
6082 spin_unlock(&fs_info->buffer_lock);
6083 radix_tree_preload_end();
6084 if (ret == -EEXIST) {
6085 exists = find_extent_buffer(fs_info, start);
6091 check_buffer_tree_ref(eb);
6092 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
6096 btrfs_release_extent_buffer(eb);
6101 static struct extent_buffer *grab_extent_buffer(
6102 struct btrfs_fs_info *fs_info, struct page *page)
6104 struct extent_buffer *exists;
6107 * For subpage case, we completely rely on radix tree to ensure we
6108 * don't try to insert two ebs for the same bytenr. So here we always
6109 * return NULL and just continue.
6111 if (fs_info->nodesize < PAGE_SIZE)
6114 /* Page not yet attached to an extent buffer */
6115 if (!PagePrivate(page))
6119 * We could have already allocated an eb for this page and attached one
6120 * so lets see if we can get a ref on the existing eb, and if we can we
6121 * know it's good and we can just return that one, else we know we can
6122 * just overwrite page->private.
6124 exists = (struct extent_buffer *)page->private;
6125 if (atomic_inc_not_zero(&exists->refs))
6128 WARN_ON(PageDirty(page));
6129 detach_page_private(page);
6133 static int check_eb_alignment(struct btrfs_fs_info *fs_info, u64 start)
6135 if (!IS_ALIGNED(start, fs_info->sectorsize)) {
6136 btrfs_err(fs_info, "bad tree block start %llu", start);
6140 if (fs_info->nodesize < PAGE_SIZE &&
6141 offset_in_page(start) + fs_info->nodesize > PAGE_SIZE) {
6143 "tree block crosses page boundary, start %llu nodesize %u",
6144 start, fs_info->nodesize);
6147 if (fs_info->nodesize >= PAGE_SIZE &&
6148 !PAGE_ALIGNED(start)) {
6150 "tree block is not page aligned, start %llu nodesize %u",
6151 start, fs_info->nodesize);
6157 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
6158 u64 start, u64 owner_root, int level)
6160 unsigned long len = fs_info->nodesize;
6163 unsigned long index = start >> PAGE_SHIFT;
6164 struct extent_buffer *eb;
6165 struct extent_buffer *exists = NULL;
6167 struct address_space *mapping = fs_info->btree_inode->i_mapping;
6171 if (check_eb_alignment(fs_info, start))
6172 return ERR_PTR(-EINVAL);
6174 #if BITS_PER_LONG == 32
6175 if (start >= MAX_LFS_FILESIZE) {
6176 btrfs_err_rl(fs_info,
6177 "extent buffer %llu is beyond 32bit page cache limit", start);
6178 btrfs_err_32bit_limit(fs_info);
6179 return ERR_PTR(-EOVERFLOW);
6181 if (start >= BTRFS_32BIT_EARLY_WARN_THRESHOLD)
6182 btrfs_warn_32bit_limit(fs_info);
6185 eb = find_extent_buffer(fs_info, start);
6189 eb = __alloc_extent_buffer(fs_info, start, len);
6191 return ERR_PTR(-ENOMEM);
6192 btrfs_set_buffer_lockdep_class(owner_root, eb, level);
6194 num_pages = num_extent_pages(eb);
6195 for (i = 0; i < num_pages; i++, index++) {
6196 struct btrfs_subpage *prealloc = NULL;
6198 p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL);
6200 exists = ERR_PTR(-ENOMEM);
6205 * Preallocate page->private for subpage case, so that we won't
6206 * allocate memory with private_lock hold. The memory will be
6207 * freed by attach_extent_buffer_page() or freed manually if
6210 * Although we have ensured one subpage eb can only have one
6211 * page, but it may change in the future for 16K page size
6212 * support, so we still preallocate the memory in the loop.
6214 if (fs_info->nodesize < PAGE_SIZE) {
6215 prealloc = btrfs_alloc_subpage(fs_info, BTRFS_SUBPAGE_METADATA);
6216 if (IS_ERR(prealloc)) {
6217 ret = PTR_ERR(prealloc);
6220 exists = ERR_PTR(ret);
6225 spin_lock(&mapping->private_lock);
6226 exists = grab_extent_buffer(fs_info, p);
6228 spin_unlock(&mapping->private_lock);
6231 mark_extent_buffer_accessed(exists, p);
6232 btrfs_free_subpage(prealloc);
6235 /* Should not fail, as we have preallocated the memory */
6236 ret = attach_extent_buffer_page(eb, p, prealloc);
6239 * To inform we have extra eb under allocation, so that
6240 * detach_extent_buffer_page() won't release the page private
6241 * when the eb hasn't yet been inserted into radix tree.
6243 * The ref will be decreased when the eb released the page, in
6244 * detach_extent_buffer_page().
6245 * Thus needs no special handling in error path.
6247 btrfs_page_inc_eb_refs(fs_info, p);
6248 spin_unlock(&mapping->private_lock);
6250 WARN_ON(btrfs_page_test_dirty(fs_info, p, eb->start, eb->len));
6252 if (!PageUptodate(p))
6256 * We can't unlock the pages just yet since the extent buffer
6257 * hasn't been properly inserted in the radix tree, this
6258 * opens a race with btree_release_folio which can free a page
6259 * while we are still filling in all pages for the buffer and
6264 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
6266 ret = radix_tree_preload(GFP_NOFS);
6268 exists = ERR_PTR(ret);
6272 spin_lock(&fs_info->buffer_lock);
6273 ret = radix_tree_insert(&fs_info->buffer_radix,
6274 start >> fs_info->sectorsize_bits, eb);
6275 spin_unlock(&fs_info->buffer_lock);
6276 radix_tree_preload_end();
6277 if (ret == -EEXIST) {
6278 exists = find_extent_buffer(fs_info, start);
6284 /* add one reference for the tree */
6285 check_buffer_tree_ref(eb);
6286 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
6289 * Now it's safe to unlock the pages because any calls to
6290 * btree_release_folio will correctly detect that a page belongs to a
6291 * live buffer and won't free them prematurely.
6293 for (i = 0; i < num_pages; i++)
6294 unlock_page(eb->pages[i]);
6298 WARN_ON(!atomic_dec_and_test(&eb->refs));
6299 for (i = 0; i < num_pages; i++) {
6301 unlock_page(eb->pages[i]);
6304 btrfs_release_extent_buffer(eb);
6308 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
6310 struct extent_buffer *eb =
6311 container_of(head, struct extent_buffer, rcu_head);
6313 __free_extent_buffer(eb);
6316 static int release_extent_buffer(struct extent_buffer *eb)
6317 __releases(&eb->refs_lock)
6319 lockdep_assert_held(&eb->refs_lock);
6321 WARN_ON(atomic_read(&eb->refs) == 0);
6322 if (atomic_dec_and_test(&eb->refs)) {
6323 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
6324 struct btrfs_fs_info *fs_info = eb->fs_info;
6326 spin_unlock(&eb->refs_lock);
6328 spin_lock(&fs_info->buffer_lock);
6329 radix_tree_delete(&fs_info->buffer_radix,
6330 eb->start >> fs_info->sectorsize_bits);
6331 spin_unlock(&fs_info->buffer_lock);
6333 spin_unlock(&eb->refs_lock);
6336 btrfs_leak_debug_del(&eb->fs_info->eb_leak_lock, &eb->leak_list);
6337 /* Should be safe to release our pages at this point */
6338 btrfs_release_extent_buffer_pages(eb);
6339 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
6340 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags))) {
6341 __free_extent_buffer(eb);
6345 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
6348 spin_unlock(&eb->refs_lock);
6353 void free_extent_buffer(struct extent_buffer *eb)
6361 refs = atomic_read(&eb->refs);
6362 if ((!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && refs <= 3)
6363 || (test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) &&
6366 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
6371 spin_lock(&eb->refs_lock);
6372 if (atomic_read(&eb->refs) == 2 &&
6373 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
6374 !extent_buffer_under_io(eb) &&
6375 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
6376 atomic_dec(&eb->refs);
6379 * I know this is terrible, but it's temporary until we stop tracking
6380 * the uptodate bits and such for the extent buffers.
6382 release_extent_buffer(eb);
6385 void free_extent_buffer_stale(struct extent_buffer *eb)
6390 spin_lock(&eb->refs_lock);
6391 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
6393 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
6394 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
6395 atomic_dec(&eb->refs);
6396 release_extent_buffer(eb);
6399 static void btree_clear_page_dirty(struct page *page)
6401 ASSERT(PageDirty(page));
6402 ASSERT(PageLocked(page));
6403 clear_page_dirty_for_io(page);
6404 xa_lock_irq(&page->mapping->i_pages);
6405 if (!PageDirty(page))
6406 __xa_clear_mark(&page->mapping->i_pages,
6407 page_index(page), PAGECACHE_TAG_DIRTY);
6408 xa_unlock_irq(&page->mapping->i_pages);
6411 static void clear_subpage_extent_buffer_dirty(const struct extent_buffer *eb)
6413 struct btrfs_fs_info *fs_info = eb->fs_info;
6414 struct page *page = eb->pages[0];
6417 /* btree_clear_page_dirty() needs page locked */
6419 last = btrfs_subpage_clear_and_test_dirty(fs_info, page, eb->start,
6422 btree_clear_page_dirty(page);
6424 WARN_ON(atomic_read(&eb->refs) == 0);
6427 void clear_extent_buffer_dirty(const struct extent_buffer *eb)
6433 if (eb->fs_info->nodesize < PAGE_SIZE)
6434 return clear_subpage_extent_buffer_dirty(eb);
6436 num_pages = num_extent_pages(eb);
6438 for (i = 0; i < num_pages; i++) {
6439 page = eb->pages[i];
6440 if (!PageDirty(page))
6443 btree_clear_page_dirty(page);
6444 ClearPageError(page);
6447 WARN_ON(atomic_read(&eb->refs) == 0);
6450 bool set_extent_buffer_dirty(struct extent_buffer *eb)
6456 check_buffer_tree_ref(eb);
6458 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
6460 num_pages = num_extent_pages(eb);
6461 WARN_ON(atomic_read(&eb->refs) == 0);
6462 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
6465 bool subpage = eb->fs_info->nodesize < PAGE_SIZE;
6468 * For subpage case, we can have other extent buffers in the
6469 * same page, and in clear_subpage_extent_buffer_dirty() we
6470 * have to clear page dirty without subpage lock held.
6471 * This can cause race where our page gets dirty cleared after
6474 * Thankfully, clear_subpage_extent_buffer_dirty() has locked
6475 * its page for other reasons, we can use page lock to prevent
6479 lock_page(eb->pages[0]);
6480 for (i = 0; i < num_pages; i++)
6481 btrfs_page_set_dirty(eb->fs_info, eb->pages[i],
6482 eb->start, eb->len);
6484 unlock_page(eb->pages[0]);
6486 #ifdef CONFIG_BTRFS_DEBUG
6487 for (i = 0; i < num_pages; i++)
6488 ASSERT(PageDirty(eb->pages[i]));
6494 void clear_extent_buffer_uptodate(struct extent_buffer *eb)
6496 struct btrfs_fs_info *fs_info = eb->fs_info;
6501 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
6502 num_pages = num_extent_pages(eb);
6503 for (i = 0; i < num_pages; i++) {
6504 page = eb->pages[i];
6509 * This is special handling for metadata subpage, as regular
6510 * btrfs_is_subpage() can not handle cloned/dummy metadata.
6512 if (fs_info->nodesize >= PAGE_SIZE)
6513 ClearPageUptodate(page);
6515 btrfs_subpage_clear_uptodate(fs_info, page, eb->start,
6520 void set_extent_buffer_uptodate(struct extent_buffer *eb)
6522 struct btrfs_fs_info *fs_info = eb->fs_info;
6527 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
6528 num_pages = num_extent_pages(eb);
6529 for (i = 0; i < num_pages; i++) {
6530 page = eb->pages[i];
6533 * This is special handling for metadata subpage, as regular
6534 * btrfs_is_subpage() can not handle cloned/dummy metadata.
6536 if (fs_info->nodesize >= PAGE_SIZE)
6537 SetPageUptodate(page);
6539 btrfs_subpage_set_uptodate(fs_info, page, eb->start,
6544 static int read_extent_buffer_subpage(struct extent_buffer *eb, int wait,
6547 struct btrfs_fs_info *fs_info = eb->fs_info;
6548 struct extent_io_tree *io_tree;
6549 struct page *page = eb->pages[0];
6550 struct btrfs_bio_ctrl bio_ctrl = {
6551 .mirror_num = mirror_num,
6555 ASSERT(!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags));
6556 ASSERT(PagePrivate(page));
6557 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
6559 if (wait == WAIT_NONE) {
6560 if (!try_lock_extent(io_tree, eb->start, eb->start + eb->len - 1))
6563 ret = lock_extent(io_tree, eb->start, eb->start + eb->len - 1);
6569 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags) ||
6570 PageUptodate(page) ||
6571 btrfs_subpage_test_uptodate(fs_info, page, eb->start, eb->len)) {
6572 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
6573 unlock_extent(io_tree, eb->start, eb->start + eb->len - 1);
6577 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
6578 eb->read_mirror = 0;
6579 atomic_set(&eb->io_pages, 1);
6580 check_buffer_tree_ref(eb);
6581 btrfs_subpage_clear_error(fs_info, page, eb->start, eb->len);
6583 btrfs_subpage_start_reader(fs_info, page, eb->start, eb->len);
6584 ret = submit_extent_page(REQ_OP_READ, NULL, &bio_ctrl,
6585 page, eb->start, eb->len,
6586 eb->start - page_offset(page),
6587 end_bio_extent_readpage, 0, true);
6590 * In the endio function, if we hit something wrong we will
6591 * increase the io_pages, so here we need to decrease it for
6594 atomic_dec(&eb->io_pages);
6596 submit_one_bio(&bio_ctrl);
6597 if (ret || wait != WAIT_COMPLETE)
6600 wait_extent_bit(io_tree, eb->start, eb->start + eb->len - 1, EXTENT_LOCKED);
6601 if (!test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
6606 int read_extent_buffer_pages(struct extent_buffer *eb, int wait, int mirror_num)
6612 int locked_pages = 0;
6613 int all_uptodate = 1;
6615 unsigned long num_reads = 0;
6616 struct btrfs_bio_ctrl bio_ctrl = {
6617 .mirror_num = mirror_num,
6620 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
6624 * We could have had EXTENT_BUFFER_UPTODATE cleared by the write
6625 * operation, which could potentially still be in flight. In this case
6626 * we simply want to return an error.
6628 if (unlikely(test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)))
6631 if (eb->fs_info->nodesize < PAGE_SIZE)
6632 return read_extent_buffer_subpage(eb, wait, mirror_num);
6634 num_pages = num_extent_pages(eb);
6635 for (i = 0; i < num_pages; i++) {
6636 page = eb->pages[i];
6637 if (wait == WAIT_NONE) {
6639 * WAIT_NONE is only utilized by readahead. If we can't
6640 * acquire the lock atomically it means either the eb
6641 * is being read out or under modification.
6642 * Either way the eb will be or has been cached,
6643 * readahead can exit safely.
6645 if (!trylock_page(page))
6653 * We need to firstly lock all pages to make sure that
6654 * the uptodate bit of our pages won't be affected by
6655 * clear_extent_buffer_uptodate().
6657 for (i = 0; i < num_pages; i++) {
6658 page = eb->pages[i];
6659 if (!PageUptodate(page)) {
6666 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
6670 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
6671 eb->read_mirror = 0;
6672 atomic_set(&eb->io_pages, num_reads);
6674 * It is possible for release_folio to clear the TREE_REF bit before we
6675 * set io_pages. See check_buffer_tree_ref for a more detailed comment.
6677 check_buffer_tree_ref(eb);
6678 for (i = 0; i < num_pages; i++) {
6679 page = eb->pages[i];
6681 if (!PageUptodate(page)) {
6683 atomic_dec(&eb->io_pages);
6688 ClearPageError(page);
6689 err = submit_extent_page(REQ_OP_READ, NULL,
6690 &bio_ctrl, page, page_offset(page),
6691 PAGE_SIZE, 0, end_bio_extent_readpage,
6695 * We failed to submit the bio so it's the
6696 * caller's responsibility to perform cleanup
6697 * i.e unlock page/set error bit.
6702 atomic_dec(&eb->io_pages);
6709 submit_one_bio(&bio_ctrl);
6711 if (ret || wait != WAIT_COMPLETE)
6714 for (i = 0; i < num_pages; i++) {
6715 page = eb->pages[i];
6716 wait_on_page_locked(page);
6717 if (!PageUptodate(page))
6724 while (locked_pages > 0) {
6726 page = eb->pages[locked_pages];
6732 static bool report_eb_range(const struct extent_buffer *eb, unsigned long start,
6735 btrfs_warn(eb->fs_info,
6736 "access to eb bytenr %llu len %lu out of range start %lu len %lu",
6737 eb->start, eb->len, start, len);
6738 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
6744 * Check if the [start, start + len) range is valid before reading/writing
6746 * NOTE: @start and @len are offset inside the eb, not logical address.
6748 * Caller should not touch the dst/src memory if this function returns error.
6750 static inline int check_eb_range(const struct extent_buffer *eb,
6751 unsigned long start, unsigned long len)
6753 unsigned long offset;
6755 /* start, start + len should not go beyond eb->len nor overflow */
6756 if (unlikely(check_add_overflow(start, len, &offset) || offset > eb->len))
6757 return report_eb_range(eb, start, len);
6762 void read_extent_buffer(const struct extent_buffer *eb, void *dstv,
6763 unsigned long start, unsigned long len)
6769 char *dst = (char *)dstv;
6770 unsigned long i = get_eb_page_index(start);
6772 if (check_eb_range(eb, start, len))
6775 offset = get_eb_offset_in_page(eb, start);
6778 page = eb->pages[i];
6780 cur = min(len, (PAGE_SIZE - offset));
6781 kaddr = page_address(page);
6782 memcpy(dst, kaddr + offset, cur);
6791 int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb,
6793 unsigned long start, unsigned long len)
6799 char __user *dst = (char __user *)dstv;
6800 unsigned long i = get_eb_page_index(start);
6803 WARN_ON(start > eb->len);
6804 WARN_ON(start + len > eb->start + eb->len);
6806 offset = get_eb_offset_in_page(eb, start);
6809 page = eb->pages[i];
6811 cur = min(len, (PAGE_SIZE - offset));
6812 kaddr = page_address(page);
6813 if (copy_to_user_nofault(dst, kaddr + offset, cur)) {
6827 int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,
6828 unsigned long start, unsigned long len)
6834 char *ptr = (char *)ptrv;
6835 unsigned long i = get_eb_page_index(start);
6838 if (check_eb_range(eb, start, len))
6841 offset = get_eb_offset_in_page(eb, start);
6844 page = eb->pages[i];
6846 cur = min(len, (PAGE_SIZE - offset));
6848 kaddr = page_address(page);
6849 ret = memcmp(ptr, kaddr + offset, cur);
6862 * Check that the extent buffer is uptodate.
6864 * For regular sector size == PAGE_SIZE case, check if @page is uptodate.
6865 * For subpage case, check if the range covered by the eb has EXTENT_UPTODATE.
6867 static void assert_eb_page_uptodate(const struct extent_buffer *eb,
6870 struct btrfs_fs_info *fs_info = eb->fs_info;
6873 * If we are using the commit root we could potentially clear a page
6874 * Uptodate while we're using the extent buffer that we've previously
6875 * looked up. We don't want to complain in this case, as the page was
6876 * valid before, we just didn't write it out. Instead we want to catch
6877 * the case where we didn't actually read the block properly, which
6878 * would have !PageUptodate && !PageError, as we clear PageError before
6881 if (fs_info->nodesize < PAGE_SIZE) {
6882 bool uptodate, error;
6884 uptodate = btrfs_subpage_test_uptodate(fs_info, page,
6885 eb->start, eb->len);
6886 error = btrfs_subpage_test_error(fs_info, page, eb->start, eb->len);
6887 WARN_ON(!uptodate && !error);
6889 WARN_ON(!PageUptodate(page) && !PageError(page));
6893 void write_extent_buffer_chunk_tree_uuid(const struct extent_buffer *eb,
6898 assert_eb_page_uptodate(eb, eb->pages[0]);
6899 kaddr = page_address(eb->pages[0]) +
6900 get_eb_offset_in_page(eb, offsetof(struct btrfs_header,
6902 memcpy(kaddr, srcv, BTRFS_FSID_SIZE);
6905 void write_extent_buffer_fsid(const struct extent_buffer *eb, const void *srcv)
6909 assert_eb_page_uptodate(eb, eb->pages[0]);
6910 kaddr = page_address(eb->pages[0]) +
6911 get_eb_offset_in_page(eb, offsetof(struct btrfs_header, fsid));
6912 memcpy(kaddr, srcv, BTRFS_FSID_SIZE);
6915 void write_extent_buffer(const struct extent_buffer *eb, const void *srcv,
6916 unsigned long start, unsigned long len)
6922 char *src = (char *)srcv;
6923 unsigned long i = get_eb_page_index(start);
6925 WARN_ON(test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags));
6927 if (check_eb_range(eb, start, len))
6930 offset = get_eb_offset_in_page(eb, start);
6933 page = eb->pages[i];
6934 assert_eb_page_uptodate(eb, page);
6936 cur = min(len, PAGE_SIZE - offset);
6937 kaddr = page_address(page);
6938 memcpy(kaddr + offset, src, cur);
6947 void memzero_extent_buffer(const struct extent_buffer *eb, unsigned long start,
6954 unsigned long i = get_eb_page_index(start);
6956 if (check_eb_range(eb, start, len))
6959 offset = get_eb_offset_in_page(eb, start);
6962 page = eb->pages[i];
6963 assert_eb_page_uptodate(eb, page);
6965 cur = min(len, PAGE_SIZE - offset);
6966 kaddr = page_address(page);
6967 memset(kaddr + offset, 0, cur);
6975 void copy_extent_buffer_full(const struct extent_buffer *dst,
6976 const struct extent_buffer *src)
6981 ASSERT(dst->len == src->len);
6983 if (dst->fs_info->nodesize >= PAGE_SIZE) {
6984 num_pages = num_extent_pages(dst);
6985 for (i = 0; i < num_pages; i++)
6986 copy_page(page_address(dst->pages[i]),
6987 page_address(src->pages[i]));
6989 size_t src_offset = get_eb_offset_in_page(src, 0);
6990 size_t dst_offset = get_eb_offset_in_page(dst, 0);
6992 ASSERT(src->fs_info->nodesize < PAGE_SIZE);
6993 memcpy(page_address(dst->pages[0]) + dst_offset,
6994 page_address(src->pages[0]) + src_offset,
6999 void copy_extent_buffer(const struct extent_buffer *dst,
7000 const struct extent_buffer *src,
7001 unsigned long dst_offset, unsigned long src_offset,
7004 u64 dst_len = dst->len;
7009 unsigned long i = get_eb_page_index(dst_offset);
7011 if (check_eb_range(dst, dst_offset, len) ||
7012 check_eb_range(src, src_offset, len))
7015 WARN_ON(src->len != dst_len);
7017 offset = get_eb_offset_in_page(dst, dst_offset);
7020 page = dst->pages[i];
7021 assert_eb_page_uptodate(dst, page);
7023 cur = min(len, (unsigned long)(PAGE_SIZE - offset));
7025 kaddr = page_address(page);
7026 read_extent_buffer(src, kaddr + offset, src_offset, cur);
7036 * eb_bitmap_offset() - calculate the page and offset of the byte containing the
7038 * @eb: the extent buffer
7039 * @start: offset of the bitmap item in the extent buffer
7041 * @page_index: return index of the page in the extent buffer that contains the
7043 * @page_offset: return offset into the page given by page_index
7045 * This helper hides the ugliness of finding the byte in an extent buffer which
7046 * contains a given bit.
7048 static inline void eb_bitmap_offset(const struct extent_buffer *eb,
7049 unsigned long start, unsigned long nr,
7050 unsigned long *page_index,
7051 size_t *page_offset)
7053 size_t byte_offset = BIT_BYTE(nr);
7057 * The byte we want is the offset of the extent buffer + the offset of
7058 * the bitmap item in the extent buffer + the offset of the byte in the
7061 offset = start + offset_in_page(eb->start) + byte_offset;
7063 *page_index = offset >> PAGE_SHIFT;
7064 *page_offset = offset_in_page(offset);
7068 * extent_buffer_test_bit - determine whether a bit in a bitmap item is set
7069 * @eb: the extent buffer
7070 * @start: offset of the bitmap item in the extent buffer
7071 * @nr: bit number to test
7073 int extent_buffer_test_bit(const struct extent_buffer *eb, unsigned long start,
7081 eb_bitmap_offset(eb, start, nr, &i, &offset);
7082 page = eb->pages[i];
7083 assert_eb_page_uptodate(eb, page);
7084 kaddr = page_address(page);
7085 return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1)));
7089 * extent_buffer_bitmap_set - set an area of a bitmap
7090 * @eb: the extent buffer
7091 * @start: offset of the bitmap item in the extent buffer
7092 * @pos: bit number of the first bit
7093 * @len: number of bits to set
7095 void extent_buffer_bitmap_set(const struct extent_buffer *eb, unsigned long start,
7096 unsigned long pos, unsigned long len)
7102 const unsigned int size = pos + len;
7103 int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
7104 u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos);
7106 eb_bitmap_offset(eb, start, pos, &i, &offset);
7107 page = eb->pages[i];
7108 assert_eb_page_uptodate(eb, page);
7109 kaddr = page_address(page);
7111 while (len >= bits_to_set) {
7112 kaddr[offset] |= mask_to_set;
7114 bits_to_set = BITS_PER_BYTE;
7116 if (++offset >= PAGE_SIZE && len > 0) {
7118 page = eb->pages[++i];
7119 assert_eb_page_uptodate(eb, page);
7120 kaddr = page_address(page);
7124 mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
7125 kaddr[offset] |= mask_to_set;
7131 * extent_buffer_bitmap_clear - clear an area of a bitmap
7132 * @eb: the extent buffer
7133 * @start: offset of the bitmap item in the extent buffer
7134 * @pos: bit number of the first bit
7135 * @len: number of bits to clear
7137 void extent_buffer_bitmap_clear(const struct extent_buffer *eb,
7138 unsigned long start, unsigned long pos,
7145 const unsigned int size = pos + len;
7146 int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
7147 u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos);
7149 eb_bitmap_offset(eb, start, pos, &i, &offset);
7150 page = eb->pages[i];
7151 assert_eb_page_uptodate(eb, page);
7152 kaddr = page_address(page);
7154 while (len >= bits_to_clear) {
7155 kaddr[offset] &= ~mask_to_clear;
7156 len -= bits_to_clear;
7157 bits_to_clear = BITS_PER_BYTE;
7159 if (++offset >= PAGE_SIZE && len > 0) {
7161 page = eb->pages[++i];
7162 assert_eb_page_uptodate(eb, page);
7163 kaddr = page_address(page);
7167 mask_to_clear &= BITMAP_LAST_BYTE_MASK(size);
7168 kaddr[offset] &= ~mask_to_clear;
7172 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
7174 unsigned long distance = (src > dst) ? src - dst : dst - src;
7175 return distance < len;
7178 static void copy_pages(struct page *dst_page, struct page *src_page,
7179 unsigned long dst_off, unsigned long src_off,
7182 char *dst_kaddr = page_address(dst_page);
7184 int must_memmove = 0;
7186 if (dst_page != src_page) {
7187 src_kaddr = page_address(src_page);
7189 src_kaddr = dst_kaddr;
7190 if (areas_overlap(src_off, dst_off, len))
7195 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
7197 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
7200 void memcpy_extent_buffer(const struct extent_buffer *dst,
7201 unsigned long dst_offset, unsigned long src_offset,
7205 size_t dst_off_in_page;
7206 size_t src_off_in_page;
7207 unsigned long dst_i;
7208 unsigned long src_i;
7210 if (check_eb_range(dst, dst_offset, len) ||
7211 check_eb_range(dst, src_offset, len))
7215 dst_off_in_page = get_eb_offset_in_page(dst, dst_offset);
7216 src_off_in_page = get_eb_offset_in_page(dst, src_offset);
7218 dst_i = get_eb_page_index(dst_offset);
7219 src_i = get_eb_page_index(src_offset);
7221 cur = min(len, (unsigned long)(PAGE_SIZE -
7223 cur = min_t(unsigned long, cur,
7224 (unsigned long)(PAGE_SIZE - dst_off_in_page));
7226 copy_pages(dst->pages[dst_i], dst->pages[src_i],
7227 dst_off_in_page, src_off_in_page, cur);
7235 void memmove_extent_buffer(const struct extent_buffer *dst,
7236 unsigned long dst_offset, unsigned long src_offset,
7240 size_t dst_off_in_page;
7241 size_t src_off_in_page;
7242 unsigned long dst_end = dst_offset + len - 1;
7243 unsigned long src_end = src_offset + len - 1;
7244 unsigned long dst_i;
7245 unsigned long src_i;
7247 if (check_eb_range(dst, dst_offset, len) ||
7248 check_eb_range(dst, src_offset, len))
7250 if (dst_offset < src_offset) {
7251 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
7255 dst_i = get_eb_page_index(dst_end);
7256 src_i = get_eb_page_index(src_end);
7258 dst_off_in_page = get_eb_offset_in_page(dst, dst_end);
7259 src_off_in_page = get_eb_offset_in_page(dst, src_end);
7261 cur = min_t(unsigned long, len, src_off_in_page + 1);
7262 cur = min(cur, dst_off_in_page + 1);
7263 copy_pages(dst->pages[dst_i], dst->pages[src_i],
7264 dst_off_in_page - cur + 1,
7265 src_off_in_page - cur + 1, cur);
7273 #define GANG_LOOKUP_SIZE 16
7274 static struct extent_buffer *get_next_extent_buffer(
7275 struct btrfs_fs_info *fs_info, struct page *page, u64 bytenr)
7277 struct extent_buffer *gang[GANG_LOOKUP_SIZE];
7278 struct extent_buffer *found = NULL;
7279 u64 page_start = page_offset(page);
7280 u64 cur = page_start;
7282 ASSERT(in_range(bytenr, page_start, PAGE_SIZE));
7283 lockdep_assert_held(&fs_info->buffer_lock);
7285 while (cur < page_start + PAGE_SIZE) {
7289 ret = radix_tree_gang_lookup(&fs_info->buffer_radix,
7290 (void **)gang, cur >> fs_info->sectorsize_bits,
7291 min_t(unsigned int, GANG_LOOKUP_SIZE,
7292 PAGE_SIZE / fs_info->nodesize));
7295 for (i = 0; i < ret; i++) {
7296 /* Already beyond page end */
7297 if (gang[i]->start >= page_start + PAGE_SIZE)
7300 if (gang[i]->start >= bytenr) {
7305 cur = gang[ret - 1]->start + gang[ret - 1]->len;
7311 static int try_release_subpage_extent_buffer(struct page *page)
7313 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
7314 u64 cur = page_offset(page);
7315 const u64 end = page_offset(page) + PAGE_SIZE;
7319 struct extent_buffer *eb = NULL;
7322 * Unlike try_release_extent_buffer() which uses page->private
7323 * to grab buffer, for subpage case we rely on radix tree, thus
7324 * we need to ensure radix tree consistency.
7326 * We also want an atomic snapshot of the radix tree, thus go
7327 * with spinlock rather than RCU.
7329 spin_lock(&fs_info->buffer_lock);
7330 eb = get_next_extent_buffer(fs_info, page, cur);
7332 /* No more eb in the page range after or at cur */
7333 spin_unlock(&fs_info->buffer_lock);
7336 cur = eb->start + eb->len;
7339 * The same as try_release_extent_buffer(), to ensure the eb
7340 * won't disappear out from under us.
7342 spin_lock(&eb->refs_lock);
7343 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
7344 spin_unlock(&eb->refs_lock);
7345 spin_unlock(&fs_info->buffer_lock);
7348 spin_unlock(&fs_info->buffer_lock);
7351 * If tree ref isn't set then we know the ref on this eb is a
7352 * real ref, so just return, this eb will likely be freed soon
7355 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
7356 spin_unlock(&eb->refs_lock);
7361 * Here we don't care about the return value, we will always
7362 * check the page private at the end. And
7363 * release_extent_buffer() will release the refs_lock.
7365 release_extent_buffer(eb);
7368 * Finally to check if we have cleared page private, as if we have
7369 * released all ebs in the page, the page private should be cleared now.
7371 spin_lock(&page->mapping->private_lock);
7372 if (!PagePrivate(page))
7376 spin_unlock(&page->mapping->private_lock);
7381 int try_release_extent_buffer(struct page *page)
7383 struct extent_buffer *eb;
7385 if (btrfs_sb(page->mapping->host->i_sb)->nodesize < PAGE_SIZE)
7386 return try_release_subpage_extent_buffer(page);
7389 * We need to make sure nobody is changing page->private, as we rely on
7390 * page->private as the pointer to extent buffer.
7392 spin_lock(&page->mapping->private_lock);
7393 if (!PagePrivate(page)) {
7394 spin_unlock(&page->mapping->private_lock);
7398 eb = (struct extent_buffer *)page->private;
7402 * This is a little awful but should be ok, we need to make sure that
7403 * the eb doesn't disappear out from under us while we're looking at
7406 spin_lock(&eb->refs_lock);
7407 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
7408 spin_unlock(&eb->refs_lock);
7409 spin_unlock(&page->mapping->private_lock);
7412 spin_unlock(&page->mapping->private_lock);
7415 * If tree ref isn't set then we know the ref on this eb is a real ref,
7416 * so just return, this page will likely be freed soon anyway.
7418 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
7419 spin_unlock(&eb->refs_lock);
7423 return release_extent_buffer(eb);
7427 * btrfs_readahead_tree_block - attempt to readahead a child block
7428 * @fs_info: the fs_info
7429 * @bytenr: bytenr to read
7430 * @owner_root: objectid of the root that owns this eb
7431 * @gen: generation for the uptodate check, can be 0
7432 * @level: level for the eb
7434 * Attempt to readahead a tree block at @bytenr. If @gen is 0 then we do a
7435 * normal uptodate check of the eb, without checking the generation. If we have
7436 * to read the block we will not block on anything.
7438 void btrfs_readahead_tree_block(struct btrfs_fs_info *fs_info,
7439 u64 bytenr, u64 owner_root, u64 gen, int level)
7441 struct extent_buffer *eb;
7444 eb = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level);
7448 if (btrfs_buffer_uptodate(eb, gen, 1)) {
7449 free_extent_buffer(eb);
7453 ret = read_extent_buffer_pages(eb, WAIT_NONE, 0);
7455 free_extent_buffer_stale(eb);
7457 free_extent_buffer(eb);
7461 * btrfs_readahead_node_child - readahead a node's child block
7462 * @node: parent node we're reading from
7463 * @slot: slot in the parent node for the child we want to read
7465 * A helper for btrfs_readahead_tree_block, we simply read the bytenr pointed at
7466 * the slot in the node provided.
7468 void btrfs_readahead_node_child(struct extent_buffer *node, int slot)
7470 btrfs_readahead_tree_block(node->fs_info,
7471 btrfs_node_blockptr(node, slot),
7472 btrfs_header_owner(node),
7473 btrfs_node_ptr_generation(node, slot),
7474 btrfs_header_level(node) - 1);