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 * split a given extent state struct in two, inserting the preallocated
574 * struct 'prealloc' as the newly created second half. 'split' indicates an
575 * offset inside 'orig' where it should be split.
578 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
579 * are two extent state structs in the tree:
580 * prealloc: [orig->start, split - 1]
581 * orig: [ split, orig->end ]
583 * The tree locks are not taken by this function. They need to be held
586 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
587 struct extent_state *prealloc, u64 split)
589 struct rb_node *parent = NULL;
590 struct rb_node **node;
592 if (tree->private_data && is_data_inode(tree->private_data))
593 btrfs_split_delalloc_extent(tree->private_data, orig, split);
595 prealloc->start = orig->start;
596 prealloc->end = split - 1;
597 prealloc->state = orig->state;
600 parent = &orig->rb_node;
603 struct tree_entry *entry;
606 entry = rb_entry(parent, struct tree_entry, rb_node);
608 if (prealloc->end < entry->start) {
609 node = &(*node)->rb_left;
610 } else if (prealloc->end > entry->end) {
611 node = &(*node)->rb_right;
613 free_extent_state(prealloc);
618 rb_link_node(&prealloc->rb_node, parent, node);
619 rb_insert_color(&prealloc->rb_node, &tree->state);
624 static struct extent_state *next_state(struct extent_state *state)
626 struct rb_node *next = rb_next(&state->rb_node);
628 return rb_entry(next, struct extent_state, rb_node);
634 * utility function to clear some bits in an extent state struct.
635 * it will optionally wake up anyone waiting on this state (wake == 1).
637 * If no bits are set on the state struct after clearing things, the
638 * struct is freed and removed from the tree
640 static struct extent_state *clear_state_bit(struct extent_io_tree *tree,
641 struct extent_state *state,
643 struct extent_changeset *changeset)
645 struct extent_state *next;
646 u32 bits_to_clear = *bits & ~EXTENT_CTLBITS;
649 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
650 u64 range = state->end - state->start + 1;
651 WARN_ON(range > tree->dirty_bytes);
652 tree->dirty_bytes -= range;
655 if (tree->private_data && is_data_inode(tree->private_data))
656 btrfs_clear_delalloc_extent(tree->private_data, state, bits);
658 ret = add_extent_changeset(state, bits_to_clear, changeset, 0);
660 state->state &= ~bits_to_clear;
663 if (state->state == 0) {
664 next = next_state(state);
665 if (extent_state_in_tree(state)) {
666 rb_erase(&state->rb_node, &tree->state);
667 RB_CLEAR_NODE(&state->rb_node);
668 free_extent_state(state);
673 merge_state(tree, state);
674 next = next_state(state);
679 static struct extent_state *
680 alloc_extent_state_atomic(struct extent_state *prealloc)
683 prealloc = alloc_extent_state(GFP_ATOMIC);
688 static void extent_io_tree_panic(struct extent_io_tree *tree, int err)
690 btrfs_panic(tree->fs_info, err,
691 "locking error: extent tree was modified by another thread while locked");
695 * clear some bits on a range in the tree. This may require splitting
696 * or inserting elements in the tree, so the gfp mask is used to
697 * indicate which allocations or sleeping are allowed.
699 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
700 * the given range from the tree regardless of state (ie for truncate).
702 * the range [start, end] is inclusive.
704 * This takes the tree lock, and returns 0 on success and < 0 on error.
706 int __clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
707 u32 bits, int wake, int delete,
708 struct extent_state **cached_state,
709 gfp_t mask, struct extent_changeset *changeset)
711 struct extent_state *state;
712 struct extent_state *cached;
713 struct extent_state *prealloc = NULL;
714 struct rb_node *node;
719 btrfs_debug_check_extent_io_range(tree, start, end);
720 trace_btrfs_clear_extent_bit(tree, start, end - start + 1, bits);
722 if (bits & EXTENT_DELALLOC)
723 bits |= EXTENT_NORESERVE;
726 bits |= ~EXTENT_CTLBITS;
728 if (bits & (EXTENT_LOCKED | EXTENT_BOUNDARY))
731 if (!prealloc && gfpflags_allow_blocking(mask)) {
733 * Don't care for allocation failure here because we might end
734 * up not needing the pre-allocated extent state at all, which
735 * is the case if we only have in the tree extent states that
736 * cover our input range and don't cover too any other range.
737 * If we end up needing a new extent state we allocate it later.
739 prealloc = alloc_extent_state(mask);
742 spin_lock(&tree->lock);
744 cached = *cached_state;
747 *cached_state = NULL;
751 if (cached && extent_state_in_tree(cached) &&
752 cached->start <= start && cached->end > start) {
754 refcount_dec(&cached->refs);
759 free_extent_state(cached);
762 * this search will find the extents that end after
765 node = tree_search(tree, start);
768 state = rb_entry(node, struct extent_state, rb_node);
770 if (state->start > end)
772 WARN_ON(state->end < start);
773 last_end = state->end;
775 /* the state doesn't have the wanted bits, go ahead */
776 if (!(state->state & bits)) {
777 state = next_state(state);
782 * | ---- desired range ---- |
784 * | ------------- state -------------- |
786 * We need to split the extent we found, and may flip
787 * bits on second half.
789 * If the extent we found extends past our range, we
790 * just split and search again. It'll get split again
791 * the next time though.
793 * If the extent we found is inside our range, we clear
794 * the desired bit on it.
797 if (state->start < start) {
798 prealloc = alloc_extent_state_atomic(prealloc);
800 err = split_state(tree, state, prealloc, start);
802 extent_io_tree_panic(tree, err);
807 if (state->end <= end) {
808 state = clear_state_bit(tree, state, &bits, wake,
815 * | ---- desired range ---- |
817 * We need to split the extent, and clear the bit
820 if (state->start <= end && state->end > end) {
821 prealloc = alloc_extent_state_atomic(prealloc);
823 err = split_state(tree, state, prealloc, end + 1);
825 extent_io_tree_panic(tree, err);
830 clear_state_bit(tree, prealloc, &bits, wake, changeset);
836 state = clear_state_bit(tree, state, &bits, wake, changeset);
838 if (last_end == (u64)-1)
840 start = last_end + 1;
841 if (start <= end && state && !need_resched())
847 spin_unlock(&tree->lock);
848 if (gfpflags_allow_blocking(mask))
853 spin_unlock(&tree->lock);
855 free_extent_state(prealloc);
861 static void wait_on_state(struct extent_io_tree *tree,
862 struct extent_state *state)
863 __releases(tree->lock)
864 __acquires(tree->lock)
867 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
868 spin_unlock(&tree->lock);
870 spin_lock(&tree->lock);
871 finish_wait(&state->wq, &wait);
875 * waits for one or more bits to clear on a range in the state tree.
876 * The range [start, end] is inclusive.
877 * The tree lock is taken by this function
879 static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
882 struct extent_state *state;
883 struct rb_node *node;
885 btrfs_debug_check_extent_io_range(tree, start, end);
887 spin_lock(&tree->lock);
891 * this search will find all the extents that end after
894 node = tree_search(tree, start);
899 state = rb_entry(node, struct extent_state, rb_node);
901 if (state->start > end)
904 if (state->state & bits) {
905 start = state->start;
906 refcount_inc(&state->refs);
907 wait_on_state(tree, state);
908 free_extent_state(state);
911 start = state->end + 1;
916 if (!cond_resched_lock(&tree->lock)) {
917 node = rb_next(node);
922 spin_unlock(&tree->lock);
925 static void set_state_bits(struct extent_io_tree *tree,
926 struct extent_state *state,
927 u32 *bits, struct extent_changeset *changeset)
929 u32 bits_to_set = *bits & ~EXTENT_CTLBITS;
932 if (tree->private_data && is_data_inode(tree->private_data))
933 btrfs_set_delalloc_extent(tree->private_data, state, bits);
935 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
936 u64 range = state->end - state->start + 1;
937 tree->dirty_bytes += range;
939 ret = add_extent_changeset(state, bits_to_set, changeset, 1);
941 state->state |= bits_to_set;
944 static void cache_state_if_flags(struct extent_state *state,
945 struct extent_state **cached_ptr,
948 if (cached_ptr && !(*cached_ptr)) {
949 if (!flags || (state->state & flags)) {
951 refcount_inc(&state->refs);
956 static void cache_state(struct extent_state *state,
957 struct extent_state **cached_ptr)
959 return cache_state_if_flags(state, cached_ptr,
960 EXTENT_LOCKED | EXTENT_BOUNDARY);
964 * set some bits on a range in the tree. This may require allocations or
965 * sleeping, so the gfp mask is used to indicate what is allowed.
967 * If any of the exclusive bits are set, this will fail with -EEXIST if some
968 * part of the range already has the desired bits set. The start of the
969 * existing range is returned in failed_start in this case.
971 * [start, end] is inclusive This takes the tree lock.
973 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, u32 bits,
974 u32 exclusive_bits, u64 *failed_start,
975 struct extent_state **cached_state, gfp_t mask,
976 struct extent_changeset *changeset)
978 struct extent_state *state;
979 struct extent_state *prealloc = NULL;
980 struct rb_node *node;
982 struct rb_node *parent;
987 btrfs_debug_check_extent_io_range(tree, start, end);
988 trace_btrfs_set_extent_bit(tree, start, end - start + 1, bits);
991 ASSERT(failed_start);
993 ASSERT(failed_start == NULL);
995 if (!prealloc && gfpflags_allow_blocking(mask)) {
997 * Don't care for allocation failure here because we might end
998 * up not needing the pre-allocated extent state at all, which
999 * is the case if we only have in the tree extent states that
1000 * cover our input range and don't cover too any other range.
1001 * If we end up needing a new extent state we allocate it later.
1003 prealloc = alloc_extent_state(mask);
1006 spin_lock(&tree->lock);
1007 if (cached_state && *cached_state) {
1008 state = *cached_state;
1009 if (state->start <= start && state->end > start &&
1010 extent_state_in_tree(state)) {
1011 node = &state->rb_node;
1016 * this search will find all the extents that end after
1019 node = tree_search_for_insert(tree, start, &p, &parent);
1021 prealloc = alloc_extent_state_atomic(prealloc);
1023 prealloc->start = start;
1024 prealloc->end = end;
1025 err = insert_state(tree, prealloc, &p, &parent, &bits, changeset);
1027 extent_io_tree_panic(tree, err);
1029 cache_state(prealloc, cached_state);
1033 state = rb_entry(node, struct extent_state, rb_node);
1035 last_start = state->start;
1036 last_end = state->end;
1039 * | ---- desired range ---- |
1042 * Just lock what we found and keep going
1044 if (state->start == start && state->end <= end) {
1045 if (state->state & exclusive_bits) {
1046 *failed_start = state->start;
1051 set_state_bits(tree, state, &bits, changeset);
1052 cache_state(state, cached_state);
1053 merge_state(tree, state);
1054 if (last_end == (u64)-1)
1056 start = last_end + 1;
1057 state = next_state(state);
1058 if (start < end && state && state->start == start &&
1065 * | ---- desired range ---- |
1068 * | ------------- state -------------- |
1070 * We need to split the extent we found, and may flip bits on
1073 * If the extent we found extends past our
1074 * range, we just split and search again. It'll get split
1075 * again the next time though.
1077 * If the extent we found is inside our range, we set the
1078 * desired bit on it.
1080 if (state->start < start) {
1081 if (state->state & exclusive_bits) {
1082 *failed_start = start;
1088 * If this extent already has all the bits we want set, then
1089 * skip it, not necessary to split it or do anything with it.
1091 if ((state->state & bits) == bits) {
1092 start = state->end + 1;
1093 cache_state(state, cached_state);
1097 prealloc = alloc_extent_state_atomic(prealloc);
1099 err = split_state(tree, state, prealloc, start);
1101 extent_io_tree_panic(tree, err);
1106 if (state->end <= end) {
1107 set_state_bits(tree, state, &bits, changeset);
1108 cache_state(state, cached_state);
1109 merge_state(tree, state);
1110 if (last_end == (u64)-1)
1112 start = last_end + 1;
1113 state = next_state(state);
1114 if (start < end && state && state->start == start &&
1121 * | ---- desired range ---- |
1122 * | state | or | state |
1124 * There's a hole, we need to insert something in it and
1125 * ignore the extent we found.
1127 if (state->start > start) {
1129 if (end < last_start)
1132 this_end = last_start - 1;
1134 prealloc = alloc_extent_state_atomic(prealloc);
1138 * Avoid to free 'prealloc' if it can be merged with
1141 prealloc->start = start;
1142 prealloc->end = this_end;
1143 err = insert_state(tree, prealloc, NULL, NULL, &bits, changeset);
1145 extent_io_tree_panic(tree, err);
1147 cache_state(prealloc, cached_state);
1149 start = this_end + 1;
1153 * | ---- desired range ---- |
1155 * We need to split the extent, and set the bit
1158 if (state->start <= end && state->end > end) {
1159 if (state->state & exclusive_bits) {
1160 *failed_start = start;
1165 prealloc = alloc_extent_state_atomic(prealloc);
1167 err = split_state(tree, state, prealloc, end + 1);
1169 extent_io_tree_panic(tree, err);
1171 set_state_bits(tree, prealloc, &bits, changeset);
1172 cache_state(prealloc, cached_state);
1173 merge_state(tree, prealloc);
1181 spin_unlock(&tree->lock);
1182 if (gfpflags_allow_blocking(mask))
1187 spin_unlock(&tree->lock);
1189 free_extent_state(prealloc);
1196 * convert_extent_bit - convert all bits in a given range from one bit to
1198 * @tree: the io tree to search
1199 * @start: the start offset in bytes
1200 * @end: the end offset in bytes (inclusive)
1201 * @bits: the bits to set in this range
1202 * @clear_bits: the bits to clear in this range
1203 * @cached_state: state that we're going to cache
1205 * This will go through and set bits for the given range. If any states exist
1206 * already in this range they are set with the given bit and cleared of the
1207 * clear_bits. This is only meant to be used by things that are mergeable, ie
1208 * converting from say DELALLOC to DIRTY. This is not meant to be used with
1209 * boundary bits like LOCK.
1211 * All allocations are done with GFP_NOFS.
1213 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1214 u32 bits, u32 clear_bits,
1215 struct extent_state **cached_state)
1217 struct extent_state *state;
1218 struct extent_state *prealloc = NULL;
1219 struct rb_node *node;
1221 struct rb_node *parent;
1225 bool first_iteration = true;
1227 btrfs_debug_check_extent_io_range(tree, start, end);
1228 trace_btrfs_convert_extent_bit(tree, start, end - start + 1, bits,
1234 * Best effort, don't worry if extent state allocation fails
1235 * here for the first iteration. We might have a cached state
1236 * that matches exactly the target range, in which case no
1237 * extent state allocations are needed. We'll only know this
1238 * after locking the tree.
1240 prealloc = alloc_extent_state(GFP_NOFS);
1241 if (!prealloc && !first_iteration)
1245 spin_lock(&tree->lock);
1246 if (cached_state && *cached_state) {
1247 state = *cached_state;
1248 if (state->start <= start && state->end > start &&
1249 extent_state_in_tree(state)) {
1250 node = &state->rb_node;
1256 * this search will find all the extents that end after
1259 node = tree_search_for_insert(tree, start, &p, &parent);
1261 prealloc = alloc_extent_state_atomic(prealloc);
1266 prealloc->start = start;
1267 prealloc->end = end;
1268 err = insert_state(tree, prealloc, &p, &parent, &bits, NULL);
1270 extent_io_tree_panic(tree, err);
1271 cache_state(prealloc, cached_state);
1275 state = rb_entry(node, struct extent_state, rb_node);
1277 last_start = state->start;
1278 last_end = state->end;
1281 * | ---- desired range ---- |
1284 * Just lock what we found and keep going
1286 if (state->start == start && state->end <= end) {
1287 set_state_bits(tree, state, &bits, NULL);
1288 cache_state(state, cached_state);
1289 state = clear_state_bit(tree, state, &clear_bits, 0, NULL);
1290 if (last_end == (u64)-1)
1292 start = last_end + 1;
1293 if (start < end && state && state->start == start &&
1300 * | ---- desired range ---- |
1303 * | ------------- state -------------- |
1305 * We need to split the extent we found, and may flip bits on
1308 * If the extent we found extends past our
1309 * range, we just split and search again. It'll get split
1310 * again the next time though.
1312 * If the extent we found is inside our range, we set the
1313 * desired bit on it.
1315 if (state->start < start) {
1316 prealloc = alloc_extent_state_atomic(prealloc);
1321 err = split_state(tree, state, prealloc, start);
1323 extent_io_tree_panic(tree, err);
1327 if (state->end <= end) {
1328 set_state_bits(tree, state, &bits, NULL);
1329 cache_state(state, cached_state);
1330 state = clear_state_bit(tree, state, &clear_bits, 0,
1332 if (last_end == (u64)-1)
1334 start = last_end + 1;
1335 if (start < end && state && state->start == start &&
1342 * | ---- desired range ---- |
1343 * | state | or | state |
1345 * There's a hole, we need to insert something in it and
1346 * ignore the extent we found.
1348 if (state->start > start) {
1350 if (end < last_start)
1353 this_end = last_start - 1;
1355 prealloc = alloc_extent_state_atomic(prealloc);
1362 * Avoid to free 'prealloc' if it can be merged with
1365 prealloc->start = start;
1366 prealloc->end = this_end;
1367 err = insert_state(tree, prealloc, NULL, NULL, &bits, NULL);
1369 extent_io_tree_panic(tree, err);
1370 cache_state(prealloc, cached_state);
1372 start = this_end + 1;
1376 * | ---- desired range ---- |
1378 * We need to split the extent, and set the bit
1381 if (state->start <= end && state->end > end) {
1382 prealloc = alloc_extent_state_atomic(prealloc);
1388 err = split_state(tree, state, prealloc, end + 1);
1390 extent_io_tree_panic(tree, err);
1392 set_state_bits(tree, prealloc, &bits, NULL);
1393 cache_state(prealloc, cached_state);
1394 clear_state_bit(tree, prealloc, &clear_bits, 0, NULL);
1402 spin_unlock(&tree->lock);
1404 first_iteration = false;
1408 spin_unlock(&tree->lock);
1410 free_extent_state(prealloc);
1415 /* wrappers around set/clear extent bit */
1416 int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1417 u32 bits, struct extent_changeset *changeset)
1420 * We don't support EXTENT_LOCKED yet, as current changeset will
1421 * record any bits changed, so for EXTENT_LOCKED case, it will
1422 * either fail with -EEXIST or changeset will record the whole
1425 BUG_ON(bits & EXTENT_LOCKED);
1427 return set_extent_bit(tree, start, end, bits, 0, NULL, NULL, GFP_NOFS,
1431 int set_extent_bits_nowait(struct extent_io_tree *tree, u64 start, u64 end,
1434 return set_extent_bit(tree, start, end, bits, 0, NULL, NULL,
1438 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1439 u32 bits, int wake, int delete,
1440 struct extent_state **cached)
1442 return __clear_extent_bit(tree, start, end, bits, wake, delete,
1443 cached, GFP_NOFS, NULL);
1446 int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1447 u32 bits, struct extent_changeset *changeset)
1450 * Don't support EXTENT_LOCKED case, same reason as
1451 * set_record_extent_bits().
1453 BUG_ON(bits & EXTENT_LOCKED);
1455 return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, GFP_NOFS,
1460 * either insert or lock state struct between start and end use mask to tell
1461 * us if waiting is desired.
1463 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1464 struct extent_state **cached_state)
1470 err = set_extent_bit(tree, start, end, EXTENT_LOCKED,
1471 EXTENT_LOCKED, &failed_start,
1472 cached_state, GFP_NOFS, NULL);
1473 if (err == -EEXIST) {
1474 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1475 start = failed_start;
1478 WARN_ON(start > end);
1483 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1488 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1489 &failed_start, NULL, GFP_NOFS, NULL);
1490 if (err == -EEXIST) {
1491 if (failed_start > start)
1492 clear_extent_bit(tree, start, failed_start - 1,
1493 EXTENT_LOCKED, 1, 0, NULL);
1499 void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
1501 unsigned long index = start >> PAGE_SHIFT;
1502 unsigned long end_index = end >> PAGE_SHIFT;
1505 while (index <= end_index) {
1506 page = find_get_page(inode->i_mapping, index);
1507 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1508 clear_page_dirty_for_io(page);
1514 void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
1516 struct address_space *mapping = inode->i_mapping;
1517 unsigned long index = start >> PAGE_SHIFT;
1518 unsigned long end_index = end >> PAGE_SHIFT;
1519 struct folio *folio;
1521 while (index <= end_index) {
1522 folio = filemap_get_folio(mapping, index);
1523 filemap_dirty_folio(mapping, folio);
1524 folio_account_redirty(folio);
1525 index += folio_nr_pages(folio);
1530 /* find the first state struct with 'bits' set after 'start', and
1531 * return it. tree->lock must be held. NULL will returned if
1532 * nothing was found after 'start'
1534 static struct extent_state *
1535 find_first_extent_bit_state(struct extent_io_tree *tree, u64 start, u32 bits)
1537 struct rb_node *node;
1538 struct extent_state *state;
1541 * this search will find all the extents that end after
1544 node = tree_search(tree, start);
1549 state = rb_entry(node, struct extent_state, rb_node);
1550 if (state->end >= start && (state->state & bits))
1553 node = rb_next(node);
1562 * Find the first offset in the io tree with one or more @bits set.
1564 * Note: If there are multiple bits set in @bits, any of them will match.
1566 * Return 0 if we find something, and update @start_ret and @end_ret.
1567 * Return 1 if we found nothing.
1569 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1570 u64 *start_ret, u64 *end_ret, u32 bits,
1571 struct extent_state **cached_state)
1573 struct extent_state *state;
1576 spin_lock(&tree->lock);
1577 if (cached_state && *cached_state) {
1578 state = *cached_state;
1579 if (state->end == start - 1 && extent_state_in_tree(state)) {
1580 while ((state = next_state(state)) != NULL) {
1581 if (state->state & bits)
1584 free_extent_state(*cached_state);
1585 *cached_state = NULL;
1588 free_extent_state(*cached_state);
1589 *cached_state = NULL;
1592 state = find_first_extent_bit_state(tree, start, bits);
1595 cache_state_if_flags(state, cached_state, 0);
1596 *start_ret = state->start;
1597 *end_ret = state->end;
1601 spin_unlock(&tree->lock);
1606 * Find a contiguous area of bits
1608 * @tree: io tree to check
1609 * @start: offset to start the search from
1610 * @start_ret: the first offset we found with the bits set
1611 * @end_ret: the final contiguous range of the bits that were set
1612 * @bits: bits to look for
1614 * set_extent_bit and clear_extent_bit can temporarily split contiguous ranges
1615 * to set bits appropriately, and then merge them again. During this time it
1616 * will drop the tree->lock, so use this helper if you want to find the actual
1617 * contiguous area for given bits. We will search to the first bit we find, and
1618 * then walk down the tree until we find a non-contiguous area. The area
1619 * returned will be the full contiguous area with the bits set.
1621 int find_contiguous_extent_bit(struct extent_io_tree *tree, u64 start,
1622 u64 *start_ret, u64 *end_ret, u32 bits)
1624 struct extent_state *state;
1627 spin_lock(&tree->lock);
1628 state = find_first_extent_bit_state(tree, start, bits);
1630 *start_ret = state->start;
1631 *end_ret = state->end;
1632 while ((state = next_state(state)) != NULL) {
1633 if (state->start > (*end_ret + 1))
1635 *end_ret = state->end;
1639 spin_unlock(&tree->lock);
1644 * Find the first range that has @bits not set. This range could start before
1647 * @tree: the tree to search
1648 * @start: offset at/after which the found extent should start
1649 * @start_ret: records the beginning of the range
1650 * @end_ret: records the end of the range (inclusive)
1651 * @bits: the set of bits which must be unset
1653 * Since unallocated range is also considered one which doesn't have the bits
1654 * set it's possible that @end_ret contains -1, this happens in case the range
1655 * spans (last_range_end, end of device]. In this case it's up to the caller to
1656 * trim @end_ret to the appropriate size.
1658 void find_first_clear_extent_bit(struct extent_io_tree *tree, u64 start,
1659 u64 *start_ret, u64 *end_ret, u32 bits)
1661 struct extent_state *state;
1662 struct rb_node *node, *prev = NULL, *next;
1664 spin_lock(&tree->lock);
1666 /* Find first extent with bits cleared */
1668 node = __etree_search(tree, start, &next, &prev, NULL, NULL);
1669 if (!node && !next && !prev) {
1671 * Tree is completely empty, send full range and let
1672 * caller deal with it
1677 } else if (!node && !next) {
1679 * We are past the last allocated chunk, set start at
1680 * the end of the last extent.
1682 state = rb_entry(prev, struct extent_state, rb_node);
1683 *start_ret = state->end + 1;
1690 * At this point 'node' either contains 'start' or start is
1693 state = rb_entry(node, struct extent_state, rb_node);
1695 if (in_range(start, state->start, state->end - state->start + 1)) {
1696 if (state->state & bits) {
1698 * |--range with bits sets--|
1702 start = state->end + 1;
1705 * 'start' falls within a range that doesn't
1706 * have the bits set, so take its start as
1707 * the beginning of the desired range
1709 * |--range with bits cleared----|
1713 *start_ret = state->start;
1718 * |---prev range---|---hole/unset---|---node range---|
1724 * |---hole/unset--||--first node--|
1729 state = rb_entry(prev, struct extent_state,
1731 *start_ret = state->end + 1;
1740 * Find the longest stretch from start until an entry which has the
1744 state = rb_entry(node, struct extent_state, rb_node);
1745 if (state->end >= start && !(state->state & bits)) {
1746 *end_ret = state->end;
1748 *end_ret = state->start - 1;
1752 node = rb_next(node);
1757 spin_unlock(&tree->lock);
1761 * find a contiguous range of bytes in the file marked as delalloc, not
1762 * more than 'max_bytes'. start and end are used to return the range,
1764 * true is returned if we find something, false if nothing was in the tree
1766 bool btrfs_find_delalloc_range(struct extent_io_tree *tree, u64 *start,
1767 u64 *end, u64 max_bytes,
1768 struct extent_state **cached_state)
1770 struct rb_node *node;
1771 struct extent_state *state;
1772 u64 cur_start = *start;
1774 u64 total_bytes = 0;
1776 spin_lock(&tree->lock);
1779 * this search will find all the extents that end after
1782 node = tree_search(tree, cur_start);
1789 state = rb_entry(node, struct extent_state, rb_node);
1790 if (found && (state->start != cur_start ||
1791 (state->state & EXTENT_BOUNDARY))) {
1794 if (!(state->state & EXTENT_DELALLOC)) {
1800 *start = state->start;
1801 *cached_state = state;
1802 refcount_inc(&state->refs);
1806 cur_start = state->end + 1;
1807 node = rb_next(node);
1808 total_bytes += state->end - state->start + 1;
1809 if (total_bytes >= max_bytes)
1815 spin_unlock(&tree->lock);
1820 * Process one page for __process_pages_contig().
1822 * Return >0 if we hit @page == @locked_page.
1823 * Return 0 if we updated the page status.
1824 * Return -EGAIN if the we need to try again.
1825 * (For PAGE_LOCK case but got dirty page or page not belong to mapping)
1827 static int process_one_page(struct btrfs_fs_info *fs_info,
1828 struct address_space *mapping,
1829 struct page *page, struct page *locked_page,
1830 unsigned long page_ops, u64 start, u64 end)
1834 ASSERT(end + 1 - start != 0 && end + 1 - start < U32_MAX);
1835 len = end + 1 - start;
1837 if (page_ops & PAGE_SET_ORDERED)
1838 btrfs_page_clamp_set_ordered(fs_info, page, start, len);
1839 if (page_ops & PAGE_SET_ERROR)
1840 btrfs_page_clamp_set_error(fs_info, page, start, len);
1841 if (page_ops & PAGE_START_WRITEBACK) {
1842 btrfs_page_clamp_clear_dirty(fs_info, page, start, len);
1843 btrfs_page_clamp_set_writeback(fs_info, page, start, len);
1845 if (page_ops & PAGE_END_WRITEBACK)
1846 btrfs_page_clamp_clear_writeback(fs_info, page, start, len);
1848 if (page == locked_page)
1851 if (page_ops & PAGE_LOCK) {
1854 ret = btrfs_page_start_writer_lock(fs_info, page, start, len);
1857 if (!PageDirty(page) || page->mapping != mapping) {
1858 btrfs_page_end_writer_lock(fs_info, page, start, len);
1862 if (page_ops & PAGE_UNLOCK)
1863 btrfs_page_end_writer_lock(fs_info, page, start, len);
1867 static int __process_pages_contig(struct address_space *mapping,
1868 struct page *locked_page,
1869 u64 start, u64 end, unsigned long page_ops,
1872 struct btrfs_fs_info *fs_info = btrfs_sb(mapping->host->i_sb);
1873 pgoff_t start_index = start >> PAGE_SHIFT;
1874 pgoff_t end_index = end >> PAGE_SHIFT;
1875 pgoff_t index = start_index;
1876 unsigned long nr_pages = end_index - start_index + 1;
1877 unsigned long pages_processed = 0;
1878 struct page *pages[16];
1882 if (page_ops & PAGE_LOCK) {
1883 ASSERT(page_ops == PAGE_LOCK);
1884 ASSERT(processed_end && *processed_end == start);
1887 if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0)
1888 mapping_set_error(mapping, -EIO);
1890 while (nr_pages > 0) {
1893 found_pages = find_get_pages_contig(mapping, index,
1894 min_t(unsigned long,
1895 nr_pages, ARRAY_SIZE(pages)), pages);
1896 if (found_pages == 0) {
1898 * Only if we're going to lock these pages, we can find
1899 * nothing at @index.
1901 ASSERT(page_ops & PAGE_LOCK);
1906 for (i = 0; i < found_pages; i++) {
1909 process_ret = process_one_page(fs_info, mapping,
1910 pages[i], locked_page, page_ops,
1912 if (process_ret < 0) {
1913 for (; i < found_pages; i++)
1921 nr_pages -= found_pages;
1922 index += found_pages;
1926 if (err && processed_end) {
1928 * Update @processed_end. I know this is awful since it has
1929 * two different return value patterns (inclusive vs exclusive).
1931 * But the exclusive pattern is necessary if @start is 0, or we
1932 * underflow and check against processed_end won't work as
1935 if (pages_processed)
1936 *processed_end = min(end,
1937 ((u64)(start_index + pages_processed) << PAGE_SHIFT) - 1);
1939 *processed_end = start;
1944 static noinline void __unlock_for_delalloc(struct inode *inode,
1945 struct page *locked_page,
1948 unsigned long index = start >> PAGE_SHIFT;
1949 unsigned long end_index = end >> PAGE_SHIFT;
1951 ASSERT(locked_page);
1952 if (index == locked_page->index && end_index == index)
1955 __process_pages_contig(inode->i_mapping, locked_page, start, end,
1959 static noinline int lock_delalloc_pages(struct inode *inode,
1960 struct page *locked_page,
1964 unsigned long index = delalloc_start >> PAGE_SHIFT;
1965 unsigned long end_index = delalloc_end >> PAGE_SHIFT;
1966 u64 processed_end = delalloc_start;
1969 ASSERT(locked_page);
1970 if (index == locked_page->index && index == end_index)
1973 ret = __process_pages_contig(inode->i_mapping, locked_page, delalloc_start,
1974 delalloc_end, PAGE_LOCK, &processed_end);
1975 if (ret == -EAGAIN && processed_end > delalloc_start)
1976 __unlock_for_delalloc(inode, locked_page, delalloc_start,
1982 * Find and lock a contiguous range of bytes in the file marked as delalloc, no
1983 * more than @max_bytes.
1985 * @start: The original start bytenr to search.
1986 * Will store the extent range start bytenr.
1987 * @end: The original end bytenr of the search range
1988 * Will store the extent range end bytenr.
1990 * Return true if we find a delalloc range which starts inside the original
1991 * range, and @start/@end will store the delalloc range start/end.
1993 * Return false if we can't find any delalloc range which starts inside the
1994 * original range, and @start/@end will be the non-delalloc range start/end.
1997 noinline_for_stack bool find_lock_delalloc_range(struct inode *inode,
1998 struct page *locked_page, u64 *start,
2001 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2002 const u64 orig_start = *start;
2003 const u64 orig_end = *end;
2004 u64 max_bytes = BTRFS_MAX_EXTENT_SIZE;
2008 struct extent_state *cached_state = NULL;
2012 /* Caller should pass a valid @end to indicate the search range end */
2013 ASSERT(orig_end > orig_start);
2015 /* The range should at least cover part of the page */
2016 ASSERT(!(orig_start >= page_offset(locked_page) + PAGE_SIZE ||
2017 orig_end <= page_offset(locked_page)));
2019 /* step one, find a bunch of delalloc bytes starting at start */
2020 delalloc_start = *start;
2022 found = btrfs_find_delalloc_range(tree, &delalloc_start, &delalloc_end,
2023 max_bytes, &cached_state);
2024 if (!found || delalloc_end <= *start || delalloc_start > orig_end) {
2025 *start = delalloc_start;
2027 /* @delalloc_end can be -1, never go beyond @orig_end */
2028 *end = min(delalloc_end, orig_end);
2029 free_extent_state(cached_state);
2034 * start comes from the offset of locked_page. We have to lock
2035 * pages in order, so we can't process delalloc bytes before
2038 if (delalloc_start < *start)
2039 delalloc_start = *start;
2042 * make sure to limit the number of pages we try to lock down
2044 if (delalloc_end + 1 - delalloc_start > max_bytes)
2045 delalloc_end = delalloc_start + max_bytes - 1;
2047 /* step two, lock all the pages after the page that has start */
2048 ret = lock_delalloc_pages(inode, locked_page,
2049 delalloc_start, delalloc_end);
2050 ASSERT(!ret || ret == -EAGAIN);
2051 if (ret == -EAGAIN) {
2052 /* some of the pages are gone, lets avoid looping by
2053 * shortening the size of the delalloc range we're searching
2055 free_extent_state(cached_state);
2056 cached_state = NULL;
2058 max_bytes = PAGE_SIZE;
2067 /* step three, lock the state bits for the whole range */
2068 lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state);
2070 /* then test to make sure it is all still delalloc */
2071 ret = test_range_bit(tree, delalloc_start, delalloc_end,
2072 EXTENT_DELALLOC, 1, cached_state);
2074 unlock_extent_cached(tree, delalloc_start, delalloc_end,
2076 __unlock_for_delalloc(inode, locked_page,
2077 delalloc_start, delalloc_end);
2081 free_extent_state(cached_state);
2082 *start = delalloc_start;
2083 *end = delalloc_end;
2088 void extent_clear_unlock_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2089 struct page *locked_page,
2090 u32 clear_bits, unsigned long page_ops)
2092 clear_extent_bit(&inode->io_tree, start, end, clear_bits, 1, 0, NULL);
2094 __process_pages_contig(inode->vfs_inode.i_mapping, locked_page,
2095 start, end, page_ops, NULL);
2099 * count the number of bytes in the tree that have a given bit(s)
2100 * set. This can be fairly slow, except for EXTENT_DIRTY which is
2101 * cached. The total number found is returned.
2103 u64 count_range_bits(struct extent_io_tree *tree,
2104 u64 *start, u64 search_end, u64 max_bytes,
2105 u32 bits, int contig)
2107 struct rb_node *node;
2108 struct extent_state *state;
2109 u64 cur_start = *start;
2110 u64 total_bytes = 0;
2114 if (WARN_ON(search_end <= cur_start))
2117 spin_lock(&tree->lock);
2118 if (cur_start == 0 && bits == EXTENT_DIRTY) {
2119 total_bytes = tree->dirty_bytes;
2123 * this search will find all the extents that end after
2126 node = tree_search(tree, cur_start);
2131 state = rb_entry(node, struct extent_state, rb_node);
2132 if (state->start > search_end)
2134 if (contig && found && state->start > last + 1)
2136 if (state->end >= cur_start && (state->state & bits) == bits) {
2137 total_bytes += min(search_end, state->end) + 1 -
2138 max(cur_start, state->start);
2139 if (total_bytes >= max_bytes)
2142 *start = max(cur_start, state->start);
2146 } else if (contig && found) {
2149 node = rb_next(node);
2154 spin_unlock(&tree->lock);
2159 * set the private field for a given byte offset in the tree. If there isn't
2160 * an extent_state there already, this does nothing.
2162 int set_state_failrec(struct extent_io_tree *tree, u64 start,
2163 struct io_failure_record *failrec)
2165 struct rb_node *node;
2166 struct extent_state *state;
2169 spin_lock(&tree->lock);
2171 * this search will find all the extents that end after
2174 node = tree_search(tree, start);
2179 state = rb_entry(node, struct extent_state, rb_node);
2180 if (state->start != start) {
2184 state->failrec = failrec;
2186 spin_unlock(&tree->lock);
2190 struct io_failure_record *get_state_failrec(struct extent_io_tree *tree, u64 start)
2192 struct rb_node *node;
2193 struct extent_state *state;
2194 struct io_failure_record *failrec;
2196 spin_lock(&tree->lock);
2198 * this search will find all the extents that end after
2201 node = tree_search(tree, start);
2203 failrec = ERR_PTR(-ENOENT);
2206 state = rb_entry(node, struct extent_state, rb_node);
2207 if (state->start != start) {
2208 failrec = ERR_PTR(-ENOENT);
2212 failrec = state->failrec;
2214 spin_unlock(&tree->lock);
2219 * searches a range in the state tree for a given mask.
2220 * If 'filled' == 1, this returns 1 only if every extent in the tree
2221 * has the bits set. Otherwise, 1 is returned if any bit in the
2222 * range is found set.
2224 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
2225 u32 bits, int filled, struct extent_state *cached)
2227 struct extent_state *state = NULL;
2228 struct rb_node *node;
2231 spin_lock(&tree->lock);
2232 if (cached && extent_state_in_tree(cached) && cached->start <= start &&
2233 cached->end > start)
2234 node = &cached->rb_node;
2236 node = tree_search(tree, start);
2237 while (node && start <= end) {
2238 state = rb_entry(node, struct extent_state, rb_node);
2240 if (filled && state->start > start) {
2245 if (state->start > end)
2248 if (state->state & bits) {
2252 } else if (filled) {
2257 if (state->end == (u64)-1)
2260 start = state->end + 1;
2263 node = rb_next(node);
2270 spin_unlock(&tree->lock);
2274 int free_io_failure(struct extent_io_tree *failure_tree,
2275 struct extent_io_tree *io_tree,
2276 struct io_failure_record *rec)
2281 set_state_failrec(failure_tree, rec->start, NULL);
2282 ret = clear_extent_bits(failure_tree, rec->start,
2283 rec->start + rec->len - 1,
2284 EXTENT_LOCKED | EXTENT_DIRTY);
2288 ret = clear_extent_bits(io_tree, rec->start,
2289 rec->start + rec->len - 1,
2299 * this bypasses the standard btrfs submit functions deliberately, as
2300 * the standard behavior is to write all copies in a raid setup. here we only
2301 * want to write the one bad copy. so we do the mapping for ourselves and issue
2302 * submit_bio directly.
2303 * to avoid any synchronization issues, wait for the data after writing, which
2304 * actually prevents the read that triggered the error from finishing.
2305 * currently, there can be no more than two copies of every data bit. thus,
2306 * exactly one rewrite is required.
2308 static int repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start,
2309 u64 length, u64 logical, struct page *page,
2310 unsigned int pg_offset, int mirror_num)
2312 struct btrfs_device *dev;
2313 struct bio_vec bvec;
2317 struct btrfs_io_context *bioc = NULL;
2320 ASSERT(!(fs_info->sb->s_flags & SB_RDONLY));
2321 BUG_ON(!mirror_num);
2323 if (btrfs_repair_one_zone(fs_info, logical))
2326 map_length = length;
2329 * Avoid races with device replace and make sure our bioc has devices
2330 * associated to its stripes that don't go away while we are doing the
2331 * read repair operation.
2333 btrfs_bio_counter_inc_blocked(fs_info);
2334 if (btrfs_is_parity_mirror(fs_info, logical, length)) {
2336 * Note that we don't use BTRFS_MAP_WRITE because it's supposed
2337 * to update all raid stripes, but here we just want to correct
2338 * bad stripe, thus BTRFS_MAP_READ is abused to only get the bad
2339 * stripe's dev and sector.
2341 ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, logical,
2342 &map_length, &bioc, 0);
2344 goto out_counter_dec;
2345 ASSERT(bioc->mirror_num == 1);
2347 ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical,
2348 &map_length, &bioc, mirror_num);
2350 goto out_counter_dec;
2351 BUG_ON(mirror_num != bioc->mirror_num);
2354 sector = bioc->stripes[bioc->mirror_num - 1].physical >> 9;
2355 dev = bioc->stripes[bioc->mirror_num - 1].dev;
2356 btrfs_put_bioc(bioc);
2358 if (!dev || !dev->bdev ||
2359 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) {
2361 goto out_counter_dec;
2364 bio_init(&bio, dev->bdev, &bvec, 1, REQ_OP_WRITE | REQ_SYNC);
2365 bio.bi_iter.bi_sector = sector;
2366 __bio_add_page(&bio, page, length, pg_offset);
2368 btrfsic_check_bio(&bio);
2369 ret = submit_bio_wait(&bio);
2371 /* try to remap that extent elsewhere? */
2372 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
2373 goto out_bio_uninit;
2376 btrfs_info_rl_in_rcu(fs_info,
2377 "read error corrected: ino %llu off %llu (dev %s sector %llu)",
2379 rcu_str_deref(dev->name), sector);
2385 btrfs_bio_counter_dec(fs_info);
2389 int btrfs_repair_eb_io_failure(const struct extent_buffer *eb, int mirror_num)
2391 struct btrfs_fs_info *fs_info = eb->fs_info;
2392 u64 start = eb->start;
2393 int i, num_pages = num_extent_pages(eb);
2396 if (sb_rdonly(fs_info->sb))
2399 for (i = 0; i < num_pages; i++) {
2400 struct page *p = eb->pages[i];
2402 ret = repair_io_failure(fs_info, 0, start, PAGE_SIZE, start, p,
2403 start - page_offset(p), mirror_num);
2413 * each time an IO finishes, we do a fast check in the IO failure tree
2414 * to see if we need to process or clean up an io_failure_record
2416 int clean_io_failure(struct btrfs_fs_info *fs_info,
2417 struct extent_io_tree *failure_tree,
2418 struct extent_io_tree *io_tree, u64 start,
2419 struct page *page, u64 ino, unsigned int pg_offset)
2422 struct io_failure_record *failrec;
2423 struct extent_state *state;
2428 ret = count_range_bits(failure_tree, &private, (u64)-1, 1,
2433 failrec = get_state_failrec(failure_tree, start);
2434 if (IS_ERR(failrec))
2437 BUG_ON(!failrec->this_mirror);
2439 if (sb_rdonly(fs_info->sb))
2442 spin_lock(&io_tree->lock);
2443 state = find_first_extent_bit_state(io_tree,
2446 spin_unlock(&io_tree->lock);
2448 if (state && state->start <= failrec->start &&
2449 state->end >= failrec->start + failrec->len - 1) {
2450 num_copies = btrfs_num_copies(fs_info, failrec->logical,
2452 if (num_copies > 1) {
2453 repair_io_failure(fs_info, ino, start, failrec->len,
2454 failrec->logical, page, pg_offset,
2455 failrec->failed_mirror);
2460 free_io_failure(failure_tree, io_tree, failrec);
2466 * Can be called when
2467 * - hold extent lock
2468 * - under ordered extent
2469 * - the inode is freeing
2471 void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start, u64 end)
2473 struct extent_io_tree *failure_tree = &inode->io_failure_tree;
2474 struct io_failure_record *failrec;
2475 struct extent_state *state, *next;
2477 if (RB_EMPTY_ROOT(&failure_tree->state))
2480 spin_lock(&failure_tree->lock);
2481 state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY);
2483 if (state->start > end)
2486 ASSERT(state->end <= end);
2488 next = next_state(state);
2490 failrec = state->failrec;
2491 free_extent_state(state);
2496 spin_unlock(&failure_tree->lock);
2499 static struct io_failure_record *btrfs_get_io_failure_record(struct inode *inode,
2502 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2503 struct io_failure_record *failrec;
2504 struct extent_map *em;
2505 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2506 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2507 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2508 const u32 sectorsize = fs_info->sectorsize;
2512 failrec = get_state_failrec(failure_tree, start);
2513 if (!IS_ERR(failrec)) {
2514 btrfs_debug(fs_info,
2515 "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu",
2516 failrec->logical, failrec->start, failrec->len);
2518 * when data can be on disk more than twice, add to failrec here
2519 * (e.g. with a list for failed_mirror) to make
2520 * clean_io_failure() clean all those errors at once.
2526 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2528 return ERR_PTR(-ENOMEM);
2530 failrec->start = start;
2531 failrec->len = sectorsize;
2532 failrec->this_mirror = 0;
2533 failrec->compress_type = BTRFS_COMPRESS_NONE;
2535 read_lock(&em_tree->lock);
2536 em = lookup_extent_mapping(em_tree, start, failrec->len);
2538 read_unlock(&em_tree->lock);
2540 return ERR_PTR(-EIO);
2543 if (em->start > start || em->start + em->len <= start) {
2544 free_extent_map(em);
2547 read_unlock(&em_tree->lock);
2550 return ERR_PTR(-EIO);
2553 logical = start - em->start;
2554 logical = em->block_start + logical;
2555 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2556 logical = em->block_start;
2557 failrec->compress_type = em->compress_type;
2560 btrfs_debug(fs_info,
2561 "Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu",
2562 logical, start, failrec->len);
2564 failrec->logical = logical;
2565 free_extent_map(em);
2567 /* Set the bits in the private failure tree */
2568 ret = set_extent_bits(failure_tree, start, start + sectorsize - 1,
2569 EXTENT_LOCKED | EXTENT_DIRTY);
2571 ret = set_state_failrec(failure_tree, start, failrec);
2572 /* Set the bits in the inode's tree */
2573 ret = set_extent_bits(tree, start, start + sectorsize - 1,
2575 } else if (ret < 0) {
2577 return ERR_PTR(ret);
2583 static bool btrfs_check_repairable(struct inode *inode,
2584 struct io_failure_record *failrec,
2587 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2590 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
2591 if (num_copies == 1) {
2593 * we only have a single copy of the data, so don't bother with
2594 * all the retry and error correction code that follows. no
2595 * matter what the error is, it is very likely to persist.
2597 btrfs_debug(fs_info,
2598 "Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
2599 num_copies, failrec->this_mirror, failed_mirror);
2603 /* The failure record should only contain one sector */
2604 ASSERT(failrec->len == fs_info->sectorsize);
2607 * There are two premises:
2608 * a) deliver good data to the caller
2609 * b) correct the bad sectors on disk
2611 * Since we're only doing repair for one sector, we only need to get
2612 * a good copy of the failed sector and if we succeed, we have setup
2613 * everything for repair_io_failure to do the rest for us.
2615 ASSERT(failed_mirror);
2616 failrec->failed_mirror = failed_mirror;
2617 failrec->this_mirror++;
2618 if (failrec->this_mirror == failed_mirror)
2619 failrec->this_mirror++;
2621 if (failrec->this_mirror > num_copies) {
2622 btrfs_debug(fs_info,
2623 "Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
2624 num_copies, failrec->this_mirror, failed_mirror);
2631 int btrfs_repair_one_sector(struct inode *inode,
2632 struct bio *failed_bio, u32 bio_offset,
2633 struct page *page, unsigned int pgoff,
2634 u64 start, int failed_mirror,
2635 submit_bio_hook_t *submit_bio_hook)
2637 struct io_failure_record *failrec;
2638 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2639 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2640 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2641 struct btrfs_bio *failed_bbio = btrfs_bio(failed_bio);
2642 const int icsum = bio_offset >> fs_info->sectorsize_bits;
2643 struct bio *repair_bio;
2644 struct btrfs_bio *repair_bbio;
2646 btrfs_debug(fs_info,
2647 "repair read error: read error at %llu", start);
2649 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
2651 failrec = btrfs_get_io_failure_record(inode, start);
2652 if (IS_ERR(failrec))
2653 return PTR_ERR(failrec);
2656 if (!btrfs_check_repairable(inode, failrec, failed_mirror)) {
2657 free_io_failure(failure_tree, tree, failrec);
2661 repair_bio = btrfs_bio_alloc(1);
2662 repair_bbio = btrfs_bio(repair_bio);
2663 repair_bbio->file_offset = start;
2664 repair_bio->bi_opf = REQ_OP_READ;
2665 repair_bio->bi_end_io = failed_bio->bi_end_io;
2666 repair_bio->bi_iter.bi_sector = failrec->logical >> 9;
2667 repair_bio->bi_private = failed_bio->bi_private;
2669 if (failed_bbio->csum) {
2670 const u32 csum_size = fs_info->csum_size;
2672 repair_bbio->csum = repair_bbio->csum_inline;
2673 memcpy(repair_bbio->csum,
2674 failed_bbio->csum + csum_size * icsum, csum_size);
2677 bio_add_page(repair_bio, page, failrec->len, pgoff);
2678 repair_bbio->iter = repair_bio->bi_iter;
2680 btrfs_debug(btrfs_sb(inode->i_sb),
2681 "repair read error: submitting new read to mirror %d",
2682 failrec->this_mirror);
2685 * At this point we have a bio, so any errors from submit_bio_hook()
2686 * will be handled by the endio on the repair_bio, so we can't return an
2689 submit_bio_hook(inode, repair_bio, failrec->this_mirror, failrec->compress_type);
2693 static void end_page_read(struct page *page, bool uptodate, u64 start, u32 len)
2695 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
2697 ASSERT(page_offset(page) <= start &&
2698 start + len <= page_offset(page) + PAGE_SIZE);
2701 if (fsverity_active(page->mapping->host) &&
2703 !PageUptodate(page) &&
2704 start < i_size_read(page->mapping->host) &&
2705 !fsverity_verify_page(page)) {
2706 btrfs_page_set_error(fs_info, page, start, len);
2708 btrfs_page_set_uptodate(fs_info, page, start, len);
2711 btrfs_page_clear_uptodate(fs_info, page, start, len);
2712 btrfs_page_set_error(fs_info, page, start, len);
2715 if (!btrfs_is_subpage(fs_info, page))
2718 btrfs_subpage_end_reader(fs_info, page, start, len);
2721 static void end_sector_io(struct page *page, u64 offset, bool uptodate)
2723 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
2724 const u32 sectorsize = inode->root->fs_info->sectorsize;
2725 struct extent_state *cached = NULL;
2727 end_page_read(page, uptodate, offset, sectorsize);
2729 set_extent_uptodate(&inode->io_tree, offset,
2730 offset + sectorsize - 1, &cached, GFP_ATOMIC);
2731 unlock_extent_cached_atomic(&inode->io_tree, offset,
2732 offset + sectorsize - 1, &cached);
2735 static void submit_data_read_repair(struct inode *inode, struct bio *failed_bio,
2736 u32 bio_offset, const struct bio_vec *bvec,
2737 int failed_mirror, unsigned int error_bitmap)
2739 const unsigned int pgoff = bvec->bv_offset;
2740 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2741 struct page *page = bvec->bv_page;
2742 const u64 start = page_offset(bvec->bv_page) + bvec->bv_offset;
2743 const u64 end = start + bvec->bv_len - 1;
2744 const u32 sectorsize = fs_info->sectorsize;
2745 const int nr_bits = (end + 1 - start) >> fs_info->sectorsize_bits;
2748 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
2750 /* This repair is only for data */
2751 ASSERT(is_data_inode(inode));
2753 /* We're here because we had some read errors or csum mismatch */
2754 ASSERT(error_bitmap);
2757 * We only get called on buffered IO, thus page must be mapped and bio
2758 * must not be cloned.
2760 ASSERT(page->mapping && !bio_flagged(failed_bio, BIO_CLONED));
2762 /* Iterate through all the sectors in the range */
2763 for (i = 0; i < nr_bits; i++) {
2764 const unsigned int offset = i * sectorsize;
2765 bool uptodate = false;
2768 if (!(error_bitmap & (1U << i))) {
2770 * This sector has no error, just end the page read
2771 * and unlock the range.
2777 ret = btrfs_repair_one_sector(inode, failed_bio,
2778 bio_offset + offset,
2779 page, pgoff + offset, start + offset,
2780 failed_mirror, btrfs_submit_data_read_bio);
2783 * We have submitted the read repair, the page release
2784 * will be handled by the endio function of the
2785 * submitted repair bio.
2786 * Thus we don't need to do any thing here.
2791 * Continue on failed repair, otherwise the remaining sectors
2792 * will not be properly unlocked.
2795 end_sector_io(page, start + offset, uptodate);
2799 /* lots and lots of room for performance fixes in the end_bio funcs */
2801 void end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2803 struct btrfs_inode *inode;
2804 const bool uptodate = (err == 0);
2807 ASSERT(page && page->mapping);
2808 inode = BTRFS_I(page->mapping->host);
2809 btrfs_writepage_endio_finish_ordered(inode, page, start, end, uptodate);
2812 const struct btrfs_fs_info *fs_info = inode->root->fs_info;
2815 ASSERT(end + 1 - start <= U32_MAX);
2816 len = end + 1 - start;
2818 btrfs_page_clear_uptodate(fs_info, page, start, len);
2819 btrfs_page_set_error(fs_info, page, start, len);
2820 ret = err < 0 ? err : -EIO;
2821 mapping_set_error(page->mapping, ret);
2826 * after a writepage IO is done, we need to:
2827 * clear the uptodate bits on error
2828 * clear the writeback bits in the extent tree for this IO
2829 * end_page_writeback if the page has no more pending IO
2831 * Scheduling is not allowed, so the extent state tree is expected
2832 * to have one and only one object corresponding to this IO.
2834 static void end_bio_extent_writepage(struct bio *bio)
2836 int error = blk_status_to_errno(bio->bi_status);
2837 struct bio_vec *bvec;
2840 struct bvec_iter_all iter_all;
2841 bool first_bvec = true;
2843 ASSERT(!bio_flagged(bio, BIO_CLONED));
2844 bio_for_each_segment_all(bvec, bio, iter_all) {
2845 struct page *page = bvec->bv_page;
2846 struct inode *inode = page->mapping->host;
2847 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2848 const u32 sectorsize = fs_info->sectorsize;
2850 /* Our read/write should always be sector aligned. */
2851 if (!IS_ALIGNED(bvec->bv_offset, sectorsize))
2853 "partial page write in btrfs with offset %u and length %u",
2854 bvec->bv_offset, bvec->bv_len);
2855 else if (!IS_ALIGNED(bvec->bv_len, sectorsize))
2857 "incomplete page write with offset %u and length %u",
2858 bvec->bv_offset, bvec->bv_len);
2860 start = page_offset(page) + bvec->bv_offset;
2861 end = start + bvec->bv_len - 1;
2864 btrfs_record_physical_zoned(inode, start, bio);
2868 end_extent_writepage(page, error, start, end);
2870 btrfs_page_clear_writeback(fs_info, page, start, bvec->bv_len);
2877 * Record previously processed extent range
2879 * For endio_readpage_release_extent() to handle a full extent range, reducing
2880 * the extent io operations.
2882 struct processed_extent {
2883 struct btrfs_inode *inode;
2884 /* Start of the range in @inode */
2886 /* End of the range in @inode */
2892 * Try to release processed extent range
2894 * May not release the extent range right now if the current range is
2895 * contiguous to processed extent.
2897 * Will release processed extent when any of @inode, @uptodate, the range is
2898 * no longer contiguous to the processed range.
2900 * Passing @inode == NULL will force processed extent to be released.
2902 static void endio_readpage_release_extent(struct processed_extent *processed,
2903 struct btrfs_inode *inode, u64 start, u64 end,
2906 struct extent_state *cached = NULL;
2907 struct extent_io_tree *tree;
2909 /* The first extent, initialize @processed */
2910 if (!processed->inode)
2914 * Contiguous to processed extent, just uptodate the end.
2916 * Several things to notice:
2918 * - bio can be merged as long as on-disk bytenr is contiguous
2919 * This means we can have page belonging to other inodes, thus need to
2920 * check if the inode still matches.
2921 * - bvec can contain range beyond current page for multi-page bvec
2922 * Thus we need to do processed->end + 1 >= start check
2924 if (processed->inode == inode && processed->uptodate == uptodate &&
2925 processed->end + 1 >= start && end >= processed->end) {
2926 processed->end = end;
2930 tree = &processed->inode->io_tree;
2932 * Now we don't have range contiguous to the processed range, release
2933 * the processed range now.
2935 if (processed->uptodate && tree->track_uptodate)
2936 set_extent_uptodate(tree, processed->start, processed->end,
2937 &cached, GFP_ATOMIC);
2938 unlock_extent_cached_atomic(tree, processed->start, processed->end,
2942 /* Update processed to current range */
2943 processed->inode = inode;
2944 processed->start = start;
2945 processed->end = end;
2946 processed->uptodate = uptodate;
2949 static void begin_page_read(struct btrfs_fs_info *fs_info, struct page *page)
2951 ASSERT(PageLocked(page));
2952 if (!btrfs_is_subpage(fs_info, page))
2955 ASSERT(PagePrivate(page));
2956 btrfs_subpage_start_reader(fs_info, page, page_offset(page), PAGE_SIZE);
2960 * Find extent buffer for a givne bytenr.
2962 * This is for end_bio_extent_readpage(), thus we can't do any unsafe locking
2965 static struct extent_buffer *find_extent_buffer_readpage(
2966 struct btrfs_fs_info *fs_info, struct page *page, u64 bytenr)
2968 struct extent_buffer *eb;
2971 * For regular sectorsize, we can use page->private to grab extent
2974 if (fs_info->nodesize >= PAGE_SIZE) {
2975 ASSERT(PagePrivate(page) && page->private);
2976 return (struct extent_buffer *)page->private;
2979 /* For subpage case, we need to lookup buffer radix tree */
2981 eb = radix_tree_lookup(&fs_info->buffer_radix,
2982 bytenr >> fs_info->sectorsize_bits);
2989 * after a readpage IO is done, we need to:
2990 * clear the uptodate bits on error
2991 * set the uptodate bits if things worked
2992 * set the page up to date if all extents in the tree are uptodate
2993 * clear the lock bit in the extent tree
2994 * unlock the page if there are no other extents locked for it
2996 * Scheduling is not allowed, so the extent state tree is expected
2997 * to have one and only one object corresponding to this IO.
2999 static void end_bio_extent_readpage(struct bio *bio)
3001 struct bio_vec *bvec;
3002 struct btrfs_bio *bbio = btrfs_bio(bio);
3003 struct extent_io_tree *tree, *failure_tree;
3004 struct processed_extent processed = { 0 };
3006 * The offset to the beginning of a bio, since one bio can never be
3007 * larger than UINT_MAX, u32 here is enough.
3011 struct bvec_iter_all iter_all;
3013 ASSERT(!bio_flagged(bio, BIO_CLONED));
3014 bio_for_each_segment_all(bvec, bio, iter_all) {
3015 bool uptodate = !bio->bi_status;
3016 struct page *page = bvec->bv_page;
3017 struct inode *inode = page->mapping->host;
3018 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3019 const u32 sectorsize = fs_info->sectorsize;
3020 unsigned int error_bitmap = (unsigned int)-1;
3021 bool repair = false;
3026 btrfs_debug(fs_info,
3027 "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u",
3028 bio->bi_iter.bi_sector, bio->bi_status,
3030 tree = &BTRFS_I(inode)->io_tree;
3031 failure_tree = &BTRFS_I(inode)->io_failure_tree;
3034 * We always issue full-sector reads, but if some block in a
3035 * page fails to read, blk_update_request() will advance
3036 * bv_offset and adjust bv_len to compensate. Print a warning
3037 * for unaligned offsets, and an error if they don't add up to
3040 if (!IS_ALIGNED(bvec->bv_offset, sectorsize))
3042 "partial page read in btrfs with offset %u and length %u",
3043 bvec->bv_offset, bvec->bv_len);
3044 else if (!IS_ALIGNED(bvec->bv_offset + bvec->bv_len,
3047 "incomplete page read with offset %u and length %u",
3048 bvec->bv_offset, bvec->bv_len);
3050 start = page_offset(page) + bvec->bv_offset;
3051 end = start + bvec->bv_len - 1;
3054 mirror = bbio->mirror_num;
3055 if (likely(uptodate)) {
3056 if (is_data_inode(inode)) {
3057 error_bitmap = btrfs_verify_data_csum(bbio,
3058 bio_offset, page, start, end);
3062 if (btrfs_validate_metadata_buffer(bbio,
3063 page, start, end, mirror))
3068 if (likely(uptodate)) {
3069 loff_t i_size = i_size_read(inode);
3070 pgoff_t end_index = i_size >> PAGE_SHIFT;
3072 clean_io_failure(BTRFS_I(inode)->root->fs_info,
3073 failure_tree, tree, start, page,
3074 btrfs_ino(BTRFS_I(inode)), 0);
3077 * Zero out the remaining part if this range straddles
3080 * Here we should only zero the range inside the bvec,
3081 * not touch anything else.
3083 * NOTE: i_size is exclusive while end is inclusive.
3085 if (page->index == end_index && i_size <= end) {
3086 u32 zero_start = max(offset_in_page(i_size),
3087 offset_in_page(start));
3089 zero_user_segment(page, zero_start,
3090 offset_in_page(end) + 1);
3092 } else if (is_data_inode(inode)) {
3094 * Only try to repair bios that actually made it to a
3095 * device. If the bio failed to be submitted mirror
3096 * is 0 and we need to fail it without retrying.
3101 struct extent_buffer *eb;
3103 eb = find_extent_buffer_readpage(fs_info, page, start);
3104 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
3105 eb->read_mirror = mirror;
3106 atomic_dec(&eb->io_pages);
3111 * submit_data_read_repair() will handle all the good
3112 * and bad sectors, we just continue to the next bvec.
3114 submit_data_read_repair(inode, bio, bio_offset, bvec,
3115 mirror, error_bitmap);
3117 /* Update page status and unlock */
3118 end_page_read(page, uptodate, start, len);
3119 endio_readpage_release_extent(&processed, BTRFS_I(inode),
3120 start, end, PageUptodate(page));
3123 ASSERT(bio_offset + len > bio_offset);
3127 /* Release the last extent */
3128 endio_readpage_release_extent(&processed, NULL, 0, 0, false);
3129 btrfs_bio_free_csum(bbio);
3134 * Populate every free slot in a provided array with pages.
3136 * @nr_pages: number of pages to allocate
3137 * @page_array: the array to fill with pages; any existing non-null entries in
3138 * the array will be skipped
3140 * Return: 0 if all pages were able to be allocated;
3141 * -ENOMEM otherwise, and the caller is responsible for freeing all
3142 * non-null page pointers in the array.
3144 int btrfs_alloc_page_array(unsigned int nr_pages, struct page **page_array)
3146 unsigned int allocated;
3148 for (allocated = 0; allocated < nr_pages;) {
3149 unsigned int last = allocated;
3151 allocated = alloc_pages_bulk_array(GFP_NOFS, nr_pages, page_array);
3153 if (allocated == nr_pages)
3157 * During this iteration, no page could be allocated, even
3158 * though alloc_pages_bulk_array() falls back to alloc_page()
3159 * if it could not bulk-allocate. So we must be out of memory.
3161 if (allocated == last)
3164 memalloc_retry_wait(GFP_NOFS);
3170 * Initialize the members up to but not including 'bio'. Use after allocating a
3171 * new bio by bio_alloc_bioset as it does not initialize the bytes outside of
3172 * 'bio' because use of __GFP_ZERO is not supported.
3174 static inline void btrfs_bio_init(struct btrfs_bio *bbio)
3176 memset(bbio, 0, offsetof(struct btrfs_bio, bio));
3180 * Allocate a btrfs_io_bio, with @nr_iovecs as maximum number of iovecs.
3182 * The bio allocation is backed by bioset and does not fail.
3184 struct bio *btrfs_bio_alloc(unsigned int nr_iovecs)
3188 ASSERT(0 < nr_iovecs && nr_iovecs <= BIO_MAX_VECS);
3189 bio = bio_alloc_bioset(NULL, nr_iovecs, 0, GFP_NOFS, &btrfs_bioset);
3190 btrfs_bio_init(btrfs_bio(bio));
3194 struct bio *btrfs_bio_clone_partial(struct bio *orig, u64 offset, u64 size)
3197 struct btrfs_bio *bbio;
3199 ASSERT(offset <= UINT_MAX && size <= UINT_MAX);
3201 /* this will never fail when it's backed by a bioset */
3202 bio = bio_alloc_clone(orig->bi_bdev, orig, GFP_NOFS, &btrfs_bioset);
3205 bbio = btrfs_bio(bio);
3206 btrfs_bio_init(bbio);
3208 bio_trim(bio, offset >> 9, size >> 9);
3209 bbio->iter = bio->bi_iter;
3214 * Attempt to add a page to bio
3216 * @bio_ctrl: record both the bio, and its bio_flags
3217 * @page: page to add to the bio
3218 * @disk_bytenr: offset of the new bio or to check whether we are adding
3219 * a contiguous page to the previous one
3220 * @size: portion of page that we want to write
3221 * @pg_offset: starting offset in the page
3222 * @compress_type: compression type of the current bio to see if we can merge them
3224 * Attempt to add a page to bio considering stripe alignment etc.
3226 * Return >= 0 for the number of bytes added to the bio.
3227 * Can return 0 if the current bio is already at stripe/zone boundary.
3228 * Return <0 for error.
3230 static int btrfs_bio_add_page(struct btrfs_bio_ctrl *bio_ctrl,
3232 u64 disk_bytenr, unsigned int size,
3233 unsigned int pg_offset,
3234 enum btrfs_compression_type compress_type)
3236 struct bio *bio = bio_ctrl->bio;
3237 u32 bio_size = bio->bi_iter.bi_size;
3239 const sector_t sector = disk_bytenr >> SECTOR_SHIFT;
3244 /* The limit should be calculated when bio_ctrl->bio is allocated */
3245 ASSERT(bio_ctrl->len_to_oe_boundary && bio_ctrl->len_to_stripe_boundary);
3246 if (bio_ctrl->compress_type != compress_type)
3249 if (bio_ctrl->compress_type != BTRFS_COMPRESS_NONE)
3250 contig = bio->bi_iter.bi_sector == sector;
3252 contig = bio_end_sector(bio) == sector;
3256 real_size = min(bio_ctrl->len_to_oe_boundary,
3257 bio_ctrl->len_to_stripe_boundary) - bio_size;
3258 real_size = min(real_size, size);
3261 * If real_size is 0, never call bio_add_*_page(), as even size is 0,
3262 * bio will still execute its endio function on the page!
3267 if (bio_op(bio) == REQ_OP_ZONE_APPEND)
3268 ret = bio_add_zone_append_page(bio, page, real_size, pg_offset);
3270 ret = bio_add_page(bio, page, real_size, pg_offset);
3275 static int calc_bio_boundaries(struct btrfs_bio_ctrl *bio_ctrl,
3276 struct btrfs_inode *inode, u64 file_offset)
3278 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3279 struct btrfs_io_geometry geom;
3280 struct btrfs_ordered_extent *ordered;
3281 struct extent_map *em;
3282 u64 logical = (bio_ctrl->bio->bi_iter.bi_sector << SECTOR_SHIFT);
3286 * Pages for compressed extent are never submitted to disk directly,
3287 * thus it has no real boundary, just set them to U32_MAX.
3289 * The split happens for real compressed bio, which happens in
3290 * btrfs_submit_compressed_read/write().
3292 if (bio_ctrl->compress_type != BTRFS_COMPRESS_NONE) {
3293 bio_ctrl->len_to_oe_boundary = U32_MAX;
3294 bio_ctrl->len_to_stripe_boundary = U32_MAX;
3297 em = btrfs_get_chunk_map(fs_info, logical, fs_info->sectorsize);
3300 ret = btrfs_get_io_geometry(fs_info, em, btrfs_op(bio_ctrl->bio),
3302 free_extent_map(em);
3306 if (geom.len > U32_MAX)
3307 bio_ctrl->len_to_stripe_boundary = U32_MAX;
3309 bio_ctrl->len_to_stripe_boundary = (u32)geom.len;
3311 if (bio_op(bio_ctrl->bio) != REQ_OP_ZONE_APPEND) {
3312 bio_ctrl->len_to_oe_boundary = U32_MAX;
3316 /* Ordered extent not yet created, so we're good */
3317 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
3319 bio_ctrl->len_to_oe_boundary = U32_MAX;
3323 bio_ctrl->len_to_oe_boundary = min_t(u32, U32_MAX,
3324 ordered->disk_bytenr + ordered->disk_num_bytes - logical);
3325 btrfs_put_ordered_extent(ordered);
3329 static int alloc_new_bio(struct btrfs_inode *inode,
3330 struct btrfs_bio_ctrl *bio_ctrl,
3331 struct writeback_control *wbc,
3333 bio_end_io_t end_io_func,
3334 u64 disk_bytenr, u32 offset, u64 file_offset,
3335 enum btrfs_compression_type compress_type)
3337 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3341 bio = btrfs_bio_alloc(BIO_MAX_VECS);
3343 * For compressed page range, its disk_bytenr is always @disk_bytenr
3344 * passed in, no matter if we have added any range into previous bio.
3346 if (compress_type != BTRFS_COMPRESS_NONE)
3347 bio->bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
3349 bio->bi_iter.bi_sector = (disk_bytenr + offset) >> SECTOR_SHIFT;
3350 bio_ctrl->bio = bio;
3351 bio_ctrl->compress_type = compress_type;
3352 bio->bi_end_io = end_io_func;
3354 ret = calc_bio_boundaries(bio_ctrl, inode, file_offset);
3360 * For Zone append we need the correct block_device that we are
3361 * going to write to set in the bio to be able to respect the
3362 * hardware limitation. Look it up here:
3364 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
3365 struct btrfs_device *dev;
3367 dev = btrfs_zoned_get_device(fs_info, disk_bytenr,
3368 fs_info->sectorsize);
3374 bio_set_dev(bio, dev->bdev);
3377 * Otherwise pick the last added device to support
3378 * cgroup writeback. For multi-device file systems this
3379 * means blk-cgroup policies have to always be set on the
3380 * last added/replaced device. This is a bit odd but has
3381 * been like that for a long time.
3383 bio_set_dev(bio, fs_info->fs_devices->latest_dev->bdev);
3385 wbc_init_bio(wbc, bio);
3387 ASSERT(bio_op(bio) != REQ_OP_ZONE_APPEND);
3391 bio_ctrl->bio = NULL;
3392 bio->bi_status = errno_to_blk_status(ret);
3398 * @opf: bio REQ_OP_* and REQ_* flags as one value
3399 * @wbc: optional writeback control for io accounting
3400 * @page: page to add to the bio
3401 * @disk_bytenr: logical bytenr where the write will be
3402 * @size: portion of page that we want to write to
3403 * @pg_offset: offset of the new bio or to check whether we are adding
3404 * a contiguous page to the previous one
3405 * @bio_ret: must be valid pointer, newly allocated bio will be stored there
3406 * @end_io_func: end_io callback for new bio
3407 * @mirror_num: desired mirror to read/write
3408 * @prev_bio_flags: flags of previous bio to see if we can merge the current one
3409 * @compress_type: compress type for current bio
3411 static int submit_extent_page(unsigned int opf,
3412 struct writeback_control *wbc,
3413 struct btrfs_bio_ctrl *bio_ctrl,
3414 struct page *page, u64 disk_bytenr,
3415 size_t size, unsigned long pg_offset,
3416 bio_end_io_t end_io_func,
3417 enum btrfs_compression_type compress_type,
3418 bool force_bio_submit)
3421 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
3422 unsigned int cur = pg_offset;
3426 ASSERT(pg_offset < PAGE_SIZE && size <= PAGE_SIZE &&
3427 pg_offset + size <= PAGE_SIZE);
3428 if (force_bio_submit)
3429 submit_one_bio(bio_ctrl);
3431 while (cur < pg_offset + size) {
3432 u32 offset = cur - pg_offset;
3435 /* Allocate new bio if needed */
3436 if (!bio_ctrl->bio) {
3437 ret = alloc_new_bio(inode, bio_ctrl, wbc, opf,
3438 end_io_func, disk_bytenr, offset,
3439 page_offset(page) + cur,
3445 * We must go through btrfs_bio_add_page() to ensure each
3446 * page range won't cross various boundaries.
3448 if (compress_type != BTRFS_COMPRESS_NONE)
3449 added = btrfs_bio_add_page(bio_ctrl, page, disk_bytenr,
3450 size - offset, pg_offset + offset,
3453 added = btrfs_bio_add_page(bio_ctrl, page,
3454 disk_bytenr + offset, size - offset,
3455 pg_offset + offset, compress_type);
3457 /* Metadata page range should never be split */
3458 if (!is_data_inode(&inode->vfs_inode))
3459 ASSERT(added == 0 || added == size - offset);
3461 /* At least we added some page, update the account */
3463 wbc_account_cgroup_owner(wbc, page, added);
3465 /* We have reached boundary, submit right now */
3466 if (added < size - offset) {
3467 /* The bio should contain some page(s) */
3468 ASSERT(bio_ctrl->bio->bi_iter.bi_size);
3469 submit_one_bio(bio_ctrl);
3476 static int attach_extent_buffer_page(struct extent_buffer *eb,
3478 struct btrfs_subpage *prealloc)
3480 struct btrfs_fs_info *fs_info = eb->fs_info;
3484 * If the page is mapped to btree inode, we should hold the private
3485 * lock to prevent race.
3486 * For cloned or dummy extent buffers, their pages are not mapped and
3487 * will not race with any other ebs.
3490 lockdep_assert_held(&page->mapping->private_lock);
3492 if (fs_info->nodesize >= PAGE_SIZE) {
3493 if (!PagePrivate(page))
3494 attach_page_private(page, eb);
3496 WARN_ON(page->private != (unsigned long)eb);
3500 /* Already mapped, just free prealloc */
3501 if (PagePrivate(page)) {
3502 btrfs_free_subpage(prealloc);
3507 /* Has preallocated memory for subpage */
3508 attach_page_private(page, prealloc);
3510 /* Do new allocation to attach subpage */
3511 ret = btrfs_attach_subpage(fs_info, page,
3512 BTRFS_SUBPAGE_METADATA);
3516 int set_page_extent_mapped(struct page *page)
3518 struct btrfs_fs_info *fs_info;
3520 ASSERT(page->mapping);
3522 if (PagePrivate(page))
3525 fs_info = btrfs_sb(page->mapping->host->i_sb);
3527 if (btrfs_is_subpage(fs_info, page))
3528 return btrfs_attach_subpage(fs_info, page, BTRFS_SUBPAGE_DATA);
3530 attach_page_private(page, (void *)EXTENT_PAGE_PRIVATE);
3534 void clear_page_extent_mapped(struct page *page)
3536 struct btrfs_fs_info *fs_info;
3538 ASSERT(page->mapping);
3540 if (!PagePrivate(page))
3543 fs_info = btrfs_sb(page->mapping->host->i_sb);
3544 if (btrfs_is_subpage(fs_info, page))
3545 return btrfs_detach_subpage(fs_info, page);
3547 detach_page_private(page);
3550 static struct extent_map *
3551 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
3552 u64 start, u64 len, struct extent_map **em_cached)
3554 struct extent_map *em;
3556 if (em_cached && *em_cached) {
3558 if (extent_map_in_tree(em) && start >= em->start &&
3559 start < extent_map_end(em)) {
3560 refcount_inc(&em->refs);
3564 free_extent_map(em);
3568 em = btrfs_get_extent(BTRFS_I(inode), page, pg_offset, start, len);
3569 if (em_cached && !IS_ERR(em)) {
3571 refcount_inc(&em->refs);
3577 * basic readpage implementation. Locked extent state structs are inserted
3578 * into the tree that are removed when the IO is done (by the end_io
3580 * XXX JDM: This needs looking at to ensure proper page locking
3581 * return 0 on success, otherwise return error
3583 static int btrfs_do_readpage(struct page *page, struct extent_map **em_cached,
3584 struct btrfs_bio_ctrl *bio_ctrl,
3585 unsigned int read_flags, u64 *prev_em_start)
3587 struct inode *inode = page->mapping->host;
3588 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3589 u64 start = page_offset(page);
3590 const u64 end = start + PAGE_SIZE - 1;
3593 u64 last_byte = i_size_read(inode);
3596 struct extent_map *em;
3598 size_t pg_offset = 0;
3600 size_t blocksize = inode->i_sb->s_blocksize;
3601 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
3603 ret = set_page_extent_mapped(page);
3605 unlock_extent(tree, start, end);
3606 btrfs_page_set_error(fs_info, page, start, PAGE_SIZE);
3611 if (page->index == last_byte >> PAGE_SHIFT) {
3612 size_t zero_offset = offset_in_page(last_byte);
3615 iosize = PAGE_SIZE - zero_offset;
3616 memzero_page(page, zero_offset, iosize);
3619 begin_page_read(fs_info, page);
3620 while (cur <= end) {
3621 unsigned long this_bio_flag = 0;
3622 bool force_bio_submit = false;
3625 ASSERT(IS_ALIGNED(cur, fs_info->sectorsize));
3626 if (cur >= last_byte) {
3627 struct extent_state *cached = NULL;
3629 iosize = PAGE_SIZE - pg_offset;
3630 memzero_page(page, pg_offset, iosize);
3631 set_extent_uptodate(tree, cur, cur + iosize - 1,
3633 unlock_extent_cached(tree, cur,
3634 cur + iosize - 1, &cached);
3635 end_page_read(page, true, cur, iosize);
3638 em = __get_extent_map(inode, page, pg_offset, cur,
3639 end - cur + 1, em_cached);
3641 unlock_extent(tree, cur, end);
3642 end_page_read(page, false, cur, end + 1 - cur);
3646 extent_offset = cur - em->start;
3647 BUG_ON(extent_map_end(em) <= cur);
3650 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
3651 this_bio_flag = em->compress_type;
3653 iosize = min(extent_map_end(em) - cur, end - cur + 1);
3654 cur_end = min(extent_map_end(em) - 1, end);
3655 iosize = ALIGN(iosize, blocksize);
3656 if (this_bio_flag != BTRFS_COMPRESS_NONE)
3657 disk_bytenr = em->block_start;
3659 disk_bytenr = em->block_start + extent_offset;
3660 block_start = em->block_start;
3661 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
3662 block_start = EXTENT_MAP_HOLE;
3665 * If we have a file range that points to a compressed extent
3666 * and it's followed by a consecutive file range that points
3667 * to the same compressed extent (possibly with a different
3668 * offset and/or length, so it either points to the whole extent
3669 * or only part of it), we must make sure we do not submit a
3670 * single bio to populate the pages for the 2 ranges because
3671 * this makes the compressed extent read zero out the pages
3672 * belonging to the 2nd range. Imagine the following scenario:
3675 * [0 - 8K] [8K - 24K]
3678 * points to extent X, points to extent X,
3679 * offset 4K, length of 8K offset 0, length 16K
3681 * [extent X, compressed length = 4K uncompressed length = 16K]
3683 * If the bio to read the compressed extent covers both ranges,
3684 * it will decompress extent X into the pages belonging to the
3685 * first range and then it will stop, zeroing out the remaining
3686 * pages that belong to the other range that points to extent X.
3687 * So here we make sure we submit 2 bios, one for the first
3688 * range and another one for the third range. Both will target
3689 * the same physical extent from disk, but we can't currently
3690 * make the compressed bio endio callback populate the pages
3691 * for both ranges because each compressed bio is tightly
3692 * coupled with a single extent map, and each range can have
3693 * an extent map with a different offset value relative to the
3694 * uncompressed data of our extent and different lengths. This
3695 * is a corner case so we prioritize correctness over
3696 * non-optimal behavior (submitting 2 bios for the same extent).
3698 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) &&
3699 prev_em_start && *prev_em_start != (u64)-1 &&
3700 *prev_em_start != em->start)
3701 force_bio_submit = true;
3704 *prev_em_start = em->start;
3706 free_extent_map(em);
3709 /* we've found a hole, just zero and go on */
3710 if (block_start == EXTENT_MAP_HOLE) {
3711 struct extent_state *cached = NULL;
3713 memzero_page(page, pg_offset, iosize);
3715 set_extent_uptodate(tree, cur, cur + iosize - 1,
3717 unlock_extent_cached(tree, cur,
3718 cur + iosize - 1, &cached);
3719 end_page_read(page, true, cur, iosize);
3721 pg_offset += iosize;
3724 /* the get_extent function already copied into the page */
3725 if (test_range_bit(tree, cur, cur_end,
3726 EXTENT_UPTODATE, 1, NULL)) {
3727 unlock_extent(tree, cur, cur + iosize - 1);
3728 end_page_read(page, true, cur, iosize);
3730 pg_offset += iosize;
3733 /* we have an inline extent but it didn't get marked up
3734 * to date. Error out
3736 if (block_start == EXTENT_MAP_INLINE) {
3737 unlock_extent(tree, cur, cur + iosize - 1);
3738 end_page_read(page, false, cur, iosize);
3740 pg_offset += iosize;
3744 ret = submit_extent_page(REQ_OP_READ | read_flags, NULL,
3745 bio_ctrl, page, disk_bytenr, iosize,
3746 pg_offset, end_bio_extent_readpage,
3747 this_bio_flag, force_bio_submit);
3750 * We have to unlock the remaining range, or the page
3751 * will never be unlocked.
3753 unlock_extent(tree, cur, end);
3754 end_page_read(page, false, cur, end + 1 - cur);
3758 pg_offset += iosize;
3764 int btrfs_read_folio(struct file *file, struct folio *folio)
3766 struct page *page = &folio->page;
3767 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
3768 u64 start = page_offset(page);
3769 u64 end = start + PAGE_SIZE - 1;
3770 struct btrfs_bio_ctrl bio_ctrl = { 0 };
3773 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
3775 ret = btrfs_do_readpage(page, NULL, &bio_ctrl, 0, NULL);
3777 * If btrfs_do_readpage() failed we will want to submit the assembled
3778 * bio to do the cleanup.
3780 submit_one_bio(&bio_ctrl);
3784 static inline void contiguous_readpages(struct page *pages[], int nr_pages,
3786 struct extent_map **em_cached,
3787 struct btrfs_bio_ctrl *bio_ctrl,
3790 struct btrfs_inode *inode = BTRFS_I(pages[0]->mapping->host);
3793 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
3795 for (index = 0; index < nr_pages; index++) {
3796 btrfs_do_readpage(pages[index], em_cached, bio_ctrl,
3797 REQ_RAHEAD, prev_em_start);
3798 put_page(pages[index]);
3803 * helper for __extent_writepage, doing all of the delayed allocation setup.
3805 * This returns 1 if btrfs_run_delalloc_range function did all the work required
3806 * to write the page (copy into inline extent). In this case the IO has
3807 * been started and the page is already unlocked.
3809 * This returns 0 if all went well (page still locked)
3810 * This returns < 0 if there were errors (page still locked)
3812 static noinline_for_stack int writepage_delalloc(struct btrfs_inode *inode,
3813 struct page *page, struct writeback_control *wbc)
3815 const u64 page_end = page_offset(page) + PAGE_SIZE - 1;
3816 u64 delalloc_start = page_offset(page);
3817 u64 delalloc_to_write = 0;
3818 /* How many pages are started by btrfs_run_delalloc_range() */
3819 unsigned long nr_written = 0;
3821 int page_started = 0;
3823 while (delalloc_start < page_end) {
3824 u64 delalloc_end = page_end;
3827 found = find_lock_delalloc_range(&inode->vfs_inode, page,
3831 delalloc_start = delalloc_end + 1;
3834 ret = btrfs_run_delalloc_range(inode, page, delalloc_start,
3835 delalloc_end, &page_started, &nr_written, wbc);
3837 btrfs_page_set_error(inode->root->fs_info, page,
3838 page_offset(page), PAGE_SIZE);
3842 * delalloc_end is already one less than the total length, so
3843 * we don't subtract one from PAGE_SIZE
3845 delalloc_to_write += (delalloc_end - delalloc_start +
3846 PAGE_SIZE) >> PAGE_SHIFT;
3847 delalloc_start = delalloc_end + 1;
3849 if (wbc->nr_to_write < delalloc_to_write) {
3852 if (delalloc_to_write < thresh * 2)
3853 thresh = delalloc_to_write;
3854 wbc->nr_to_write = min_t(u64, delalloc_to_write,
3858 /* Did btrfs_run_dealloc_range() already unlock and start the IO? */
3861 * We've unlocked the page, so we can't update the mapping's
3862 * writeback index, just update nr_to_write.
3864 wbc->nr_to_write -= nr_written;
3872 * Find the first byte we need to write.
3874 * For subpage, one page can contain several sectors, and
3875 * __extent_writepage_io() will just grab all extent maps in the page
3876 * range and try to submit all non-inline/non-compressed extents.
3878 * This is a big problem for subpage, we shouldn't re-submit already written
3880 * This function will lookup subpage dirty bit to find which range we really
3883 * Return the next dirty range in [@start, @end).
3884 * If no dirty range is found, @start will be page_offset(page) + PAGE_SIZE.
3886 static void find_next_dirty_byte(struct btrfs_fs_info *fs_info,
3887 struct page *page, u64 *start, u64 *end)
3889 struct btrfs_subpage *subpage = (struct btrfs_subpage *)page->private;
3890 struct btrfs_subpage_info *spi = fs_info->subpage_info;
3891 u64 orig_start = *start;
3892 /* Declare as unsigned long so we can use bitmap ops */
3893 unsigned long flags;
3894 int range_start_bit;
3898 * For regular sector size == page size case, since one page only
3899 * contains one sector, we return the page offset directly.
3901 if (!btrfs_is_subpage(fs_info, page)) {
3902 *start = page_offset(page);
3903 *end = page_offset(page) + PAGE_SIZE;
3907 range_start_bit = spi->dirty_offset +
3908 (offset_in_page(orig_start) >> fs_info->sectorsize_bits);
3910 /* We should have the page locked, but just in case */
3911 spin_lock_irqsave(&subpage->lock, flags);
3912 bitmap_next_set_region(subpage->bitmaps, &range_start_bit, &range_end_bit,
3913 spi->dirty_offset + spi->bitmap_nr_bits);
3914 spin_unlock_irqrestore(&subpage->lock, flags);
3916 range_start_bit -= spi->dirty_offset;
3917 range_end_bit -= spi->dirty_offset;
3919 *start = page_offset(page) + range_start_bit * fs_info->sectorsize;
3920 *end = page_offset(page) + range_end_bit * fs_info->sectorsize;
3924 * helper for __extent_writepage. This calls the writepage start hooks,
3925 * and does the loop to map the page into extents and bios.
3927 * We return 1 if the IO is started and the page is unlocked,
3928 * 0 if all went well (page still locked)
3929 * < 0 if there were errors (page still locked)
3931 static noinline_for_stack int __extent_writepage_io(struct btrfs_inode *inode,
3933 struct writeback_control *wbc,
3934 struct extent_page_data *epd,
3938 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3939 u64 cur = page_offset(page);
3940 u64 end = cur + PAGE_SIZE - 1;
3943 struct extent_map *em;
3947 u32 opf = REQ_OP_WRITE;
3948 const unsigned int write_flags = wbc_to_write_flags(wbc);
3949 bool has_error = false;
3952 ret = btrfs_writepage_cow_fixup(page);
3954 /* Fixup worker will requeue */
3955 redirty_page_for_writepage(wbc, page);
3961 * we don't want to touch the inode after unlocking the page,
3962 * so we update the mapping writeback index now
3966 while (cur <= end) {
3969 u64 dirty_range_start = cur;
3970 u64 dirty_range_end;
3973 if (cur >= i_size) {
3974 btrfs_writepage_endio_finish_ordered(inode, page, cur,
3977 * This range is beyond i_size, thus we don't need to
3978 * bother writing back.
3979 * But we still need to clear the dirty subpage bit, or
3980 * the next time the page gets dirtied, we will try to
3981 * writeback the sectors with subpage dirty bits,
3982 * causing writeback without ordered extent.
3984 btrfs_page_clear_dirty(fs_info, page, cur, end + 1 - cur);
3988 find_next_dirty_byte(fs_info, page, &dirty_range_start,
3990 if (cur < dirty_range_start) {
3991 cur = dirty_range_start;
3995 em = btrfs_get_extent(inode, NULL, 0, cur, end - cur + 1);
3997 btrfs_page_set_error(fs_info, page, cur, end - cur + 1);
3998 ret = PTR_ERR_OR_ZERO(em);
4005 extent_offset = cur - em->start;
4006 em_end = extent_map_end(em);
4007 ASSERT(cur <= em_end);
4009 ASSERT(IS_ALIGNED(em->start, fs_info->sectorsize));
4010 ASSERT(IS_ALIGNED(em->len, fs_info->sectorsize));
4011 block_start = em->block_start;
4012 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4013 disk_bytenr = em->block_start + extent_offset;
4016 * Note that em_end from extent_map_end() and dirty_range_end from
4017 * find_next_dirty_byte() are all exclusive
4019 iosize = min(min(em_end, end + 1), dirty_range_end) - cur;
4021 if (btrfs_use_zone_append(inode, em->block_start))
4022 opf = REQ_OP_ZONE_APPEND;
4024 free_extent_map(em);
4028 * compressed and inline extents are written through other
4031 if (compressed || block_start == EXTENT_MAP_HOLE ||
4032 block_start == EXTENT_MAP_INLINE) {
4036 btrfs_writepage_endio_finish_ordered(inode,
4037 page, cur, cur + iosize - 1, true);
4038 btrfs_page_clear_dirty(fs_info, page, cur, iosize);
4043 btrfs_set_range_writeback(inode, cur, cur + iosize - 1);
4044 if (!PageWriteback(page)) {
4045 btrfs_err(inode->root->fs_info,
4046 "page %lu not writeback, cur %llu end %llu",
4047 page->index, cur, end);
4051 * Although the PageDirty bit is cleared before entering this
4052 * function, subpage dirty bit is not cleared.
4053 * So clear subpage dirty bit here so next time we won't submit
4054 * page for range already written to disk.
4056 btrfs_page_clear_dirty(fs_info, page, cur, iosize);
4058 ret = submit_extent_page(opf | write_flags, wbc,
4059 &epd->bio_ctrl, page,
4060 disk_bytenr, iosize,
4061 cur - page_offset(page),
4062 end_bio_extent_writepage,
4069 btrfs_page_set_error(fs_info, page, cur, iosize);
4070 if (PageWriteback(page))
4071 btrfs_page_clear_writeback(fs_info, page, cur,
4079 * If we finish without problem, we should not only clear page dirty,
4080 * but also empty subpage dirty bits
4083 btrfs_page_assert_not_dirty(fs_info, page);
4091 * the writepage semantics are similar to regular writepage. extent
4092 * records are inserted to lock ranges in the tree, and as dirty areas
4093 * are found, they are marked writeback. Then the lock bits are removed
4094 * and the end_io handler clears the writeback ranges
4096 * Return 0 if everything goes well.
4097 * Return <0 for error.
4099 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
4100 struct extent_page_data *epd)
4102 struct folio *folio = page_folio(page);
4103 struct inode *inode = page->mapping->host;
4104 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4105 const u64 page_start = page_offset(page);
4106 const u64 page_end = page_start + PAGE_SIZE - 1;
4110 loff_t i_size = i_size_read(inode);
4111 unsigned long end_index = i_size >> PAGE_SHIFT;
4113 trace___extent_writepage(page, inode, wbc);
4115 WARN_ON(!PageLocked(page));
4117 btrfs_page_clear_error(btrfs_sb(inode->i_sb), page,
4118 page_offset(page), PAGE_SIZE);
4120 pg_offset = offset_in_page(i_size);
4121 if (page->index > end_index ||
4122 (page->index == end_index && !pg_offset)) {
4123 folio_invalidate(folio, 0, folio_size(folio));
4124 folio_unlock(folio);
4128 if (page->index == end_index)
4129 memzero_page(page, pg_offset, PAGE_SIZE - pg_offset);
4131 ret = set_page_extent_mapped(page);
4137 if (!epd->extent_locked) {
4138 ret = writepage_delalloc(BTRFS_I(inode), page, wbc);
4145 ret = __extent_writepage_io(BTRFS_I(inode), page, wbc, epd, i_size,
4152 /* make sure the mapping tag for page dirty gets cleared */
4153 set_page_writeback(page);
4154 end_page_writeback(page);
4157 * Here we used to have a check for PageError() and then set @ret and
4158 * call end_extent_writepage().
4160 * But in fact setting @ret here will cause different error paths
4161 * between subpage and regular sectorsize.
4163 * For regular page size, we never submit current page, but only add
4164 * current page to current bio.
4165 * The bio submission can only happen in next page.
4166 * Thus if we hit the PageError() branch, @ret is already set to
4167 * non-zero value and will not get updated for regular sectorsize.
4169 * But for subpage case, it's possible we submit part of current page,
4170 * thus can get PageError() set by submitted bio of the same page,
4171 * while our @ret is still 0.
4173 * So here we unify the behavior and don't set @ret.
4174 * Error can still be properly passed to higher layer as page will
4175 * be set error, here we just don't handle the IO failure.
4177 * NOTE: This is just a hotfix for subpage.
4178 * The root fix will be properly ending ordered extent when we hit
4179 * an error during writeback.
4181 * But that needs a bigger refactoring, as we not only need to grab the
4182 * submitted OE, but also need to know exactly at which bytenr we hit
4184 * Currently the full page based __extent_writepage_io() is not
4187 if (PageError(page))
4188 end_extent_writepage(page, ret, page_start, page_end);
4189 if (epd->extent_locked) {
4191 * If epd->extent_locked, it's from extent_write_locked_range(),
4192 * the page can either be locked by lock_page() or
4193 * process_one_page().
4194 * Let btrfs_page_unlock_writer() handle both cases.
4197 btrfs_page_unlock_writer(fs_info, page, wbc->range_start,
4198 wbc->range_end + 1 - wbc->range_start);
4206 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
4208 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
4209 TASK_UNINTERRUPTIBLE);
4212 static void end_extent_buffer_writeback(struct extent_buffer *eb)
4214 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
4215 smp_mb__after_atomic();
4216 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
4220 * Lock extent buffer status and pages for writeback.
4222 * May try to flush write bio if we can't get the lock.
4224 * Return 0 if the extent buffer doesn't need to be submitted.
4225 * (E.g. the extent buffer is not dirty)
4226 * Return >0 is the extent buffer is submitted to bio.
4227 * Return <0 if something went wrong, no page is locked.
4229 static noinline_for_stack int lock_extent_buffer_for_io(struct extent_buffer *eb,
4230 struct extent_page_data *epd)
4232 struct btrfs_fs_info *fs_info = eb->fs_info;
4237 if (!btrfs_try_tree_write_lock(eb)) {
4238 submit_write_bio(epd, 0);
4240 btrfs_tree_lock(eb);
4243 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
4244 btrfs_tree_unlock(eb);
4248 submit_write_bio(epd, 0);
4252 wait_on_extent_buffer_writeback(eb);
4253 btrfs_tree_lock(eb);
4254 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
4256 btrfs_tree_unlock(eb);
4261 * We need to do this to prevent races in people who check if the eb is
4262 * under IO since we can end up having no IO bits set for a short period
4265 spin_lock(&eb->refs_lock);
4266 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
4267 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
4268 spin_unlock(&eb->refs_lock);
4269 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
4270 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4272 fs_info->dirty_metadata_batch);
4275 spin_unlock(&eb->refs_lock);
4278 btrfs_tree_unlock(eb);
4281 * Either we don't need to submit any tree block, or we're submitting
4283 * Subpage metadata doesn't use page locking at all, so we can skip
4286 if (!ret || fs_info->nodesize < PAGE_SIZE)
4289 num_pages = num_extent_pages(eb);
4290 for (i = 0; i < num_pages; i++) {
4291 struct page *p = eb->pages[i];
4293 if (!trylock_page(p)) {
4295 submit_write_bio(epd, 0);
4305 static void set_btree_ioerr(struct page *page, struct extent_buffer *eb)
4307 struct btrfs_fs_info *fs_info = eb->fs_info;
4309 btrfs_page_set_error(fs_info, page, eb->start, eb->len);
4310 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))
4314 * A read may stumble upon this buffer later, make sure that it gets an
4315 * error and knows there was an error.
4317 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4320 * We need to set the mapping with the io error as well because a write
4321 * error will flip the file system readonly, and then syncfs() will
4322 * return a 0 because we are readonly if we don't modify the err seq for
4325 mapping_set_error(page->mapping, -EIO);
4328 * If we error out, we should add back the dirty_metadata_bytes
4329 * to make it consistent.
4331 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4332 eb->len, fs_info->dirty_metadata_batch);
4335 * If writeback for a btree extent that doesn't belong to a log tree
4336 * failed, increment the counter transaction->eb_write_errors.
4337 * We do this because while the transaction is running and before it's
4338 * committing (when we call filemap_fdata[write|wait]_range against
4339 * the btree inode), we might have
4340 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
4341 * returns an error or an error happens during writeback, when we're
4342 * committing the transaction we wouldn't know about it, since the pages
4343 * can be no longer dirty nor marked anymore for writeback (if a
4344 * subsequent modification to the extent buffer didn't happen before the
4345 * transaction commit), which makes filemap_fdata[write|wait]_range not
4346 * able to find the pages tagged with SetPageError at transaction
4347 * commit time. So if this happens we must abort the transaction,
4348 * otherwise we commit a super block with btree roots that point to
4349 * btree nodes/leafs whose content on disk is invalid - either garbage
4350 * or the content of some node/leaf from a past generation that got
4351 * cowed or deleted and is no longer valid.
4353 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
4354 * not be enough - we need to distinguish between log tree extents vs
4355 * non-log tree extents, and the next filemap_fdatawait_range() call
4356 * will catch and clear such errors in the mapping - and that call might
4357 * be from a log sync and not from a transaction commit. Also, checking
4358 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
4359 * not done and would not be reliable - the eb might have been released
4360 * from memory and reading it back again means that flag would not be
4361 * set (since it's a runtime flag, not persisted on disk).
4363 * Using the flags below in the btree inode also makes us achieve the
4364 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
4365 * writeback for all dirty pages and before filemap_fdatawait_range()
4366 * is called, the writeback for all dirty pages had already finished
4367 * with errors - because we were not using AS_EIO/AS_ENOSPC,
4368 * filemap_fdatawait_range() would return success, as it could not know
4369 * that writeback errors happened (the pages were no longer tagged for
4372 switch (eb->log_index) {
4374 set_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags);
4377 set_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags);
4380 set_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags);
4383 BUG(); /* unexpected, logic error */
4388 * The endio specific version which won't touch any unsafe spinlock in endio
4391 static struct extent_buffer *find_extent_buffer_nolock(
4392 struct btrfs_fs_info *fs_info, u64 start)
4394 struct extent_buffer *eb;
4397 eb = radix_tree_lookup(&fs_info->buffer_radix,
4398 start >> fs_info->sectorsize_bits);
4399 if (eb && atomic_inc_not_zero(&eb->refs)) {
4408 * The endio function for subpage extent buffer write.
4410 * Unlike end_bio_extent_buffer_writepage(), we only call end_page_writeback()
4411 * after all extent buffers in the page has finished their writeback.
4413 static void end_bio_subpage_eb_writepage(struct bio *bio)
4415 struct btrfs_fs_info *fs_info;
4416 struct bio_vec *bvec;
4417 struct bvec_iter_all iter_all;
4419 fs_info = btrfs_sb(bio_first_page_all(bio)->mapping->host->i_sb);
4420 ASSERT(fs_info->nodesize < PAGE_SIZE);
4422 ASSERT(!bio_flagged(bio, BIO_CLONED));
4423 bio_for_each_segment_all(bvec, bio, iter_all) {
4424 struct page *page = bvec->bv_page;
4425 u64 bvec_start = page_offset(page) + bvec->bv_offset;
4426 u64 bvec_end = bvec_start + bvec->bv_len - 1;
4427 u64 cur_bytenr = bvec_start;
4429 ASSERT(IS_ALIGNED(bvec->bv_len, fs_info->nodesize));
4431 /* Iterate through all extent buffers in the range */
4432 while (cur_bytenr <= bvec_end) {
4433 struct extent_buffer *eb;
4437 * Here we can't use find_extent_buffer(), as it may
4438 * try to lock eb->refs_lock, which is not safe in endio
4441 eb = find_extent_buffer_nolock(fs_info, cur_bytenr);
4444 cur_bytenr = eb->start + eb->len;
4446 ASSERT(test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags));
4447 done = atomic_dec_and_test(&eb->io_pages);
4450 if (bio->bi_status ||
4451 test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) {
4452 ClearPageUptodate(page);
4453 set_btree_ioerr(page, eb);
4456 btrfs_subpage_clear_writeback(fs_info, page, eb->start,
4458 end_extent_buffer_writeback(eb);
4460 * free_extent_buffer() will grab spinlock which is not
4461 * safe in endio context. Thus here we manually dec
4464 atomic_dec(&eb->refs);
4470 static void end_bio_extent_buffer_writepage(struct bio *bio)
4472 struct bio_vec *bvec;
4473 struct extent_buffer *eb;
4475 struct bvec_iter_all iter_all;
4477 ASSERT(!bio_flagged(bio, BIO_CLONED));
4478 bio_for_each_segment_all(bvec, bio, iter_all) {
4479 struct page *page = bvec->bv_page;
4481 eb = (struct extent_buffer *)page->private;
4483 done = atomic_dec_and_test(&eb->io_pages);
4485 if (bio->bi_status ||
4486 test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) {
4487 ClearPageUptodate(page);
4488 set_btree_ioerr(page, eb);
4491 end_page_writeback(page);
4496 end_extent_buffer_writeback(eb);
4502 static void prepare_eb_write(struct extent_buffer *eb)
4505 unsigned long start;
4508 clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
4509 atomic_set(&eb->io_pages, num_extent_pages(eb));
4511 /* Set btree blocks beyond nritems with 0 to avoid stale content */
4512 nritems = btrfs_header_nritems(eb);
4513 if (btrfs_header_level(eb) > 0) {
4514 end = btrfs_node_key_ptr_offset(nritems);
4515 memzero_extent_buffer(eb, end, eb->len - end);
4519 * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0
4521 start = btrfs_item_nr_offset(nritems);
4522 end = BTRFS_LEAF_DATA_OFFSET + leaf_data_end(eb);
4523 memzero_extent_buffer(eb, start, end - start);
4528 * Unlike the work in write_one_eb(), we rely completely on extent locking.
4529 * Page locking is only utilized at minimum to keep the VMM code happy.
4531 static int write_one_subpage_eb(struct extent_buffer *eb,
4532 struct writeback_control *wbc,
4533 struct extent_page_data *epd)
4535 struct btrfs_fs_info *fs_info = eb->fs_info;
4536 struct page *page = eb->pages[0];
4537 unsigned int write_flags = wbc_to_write_flags(wbc);
4538 bool no_dirty_ebs = false;
4541 prepare_eb_write(eb);
4543 /* clear_page_dirty_for_io() in subpage helper needs page locked */
4545 btrfs_subpage_set_writeback(fs_info, page, eb->start, eb->len);
4547 /* Check if this is the last dirty bit to update nr_written */
4548 no_dirty_ebs = btrfs_subpage_clear_and_test_dirty(fs_info, page,
4549 eb->start, eb->len);
4551 clear_page_dirty_for_io(page);
4553 ret = submit_extent_page(REQ_OP_WRITE | write_flags, wbc,
4554 &epd->bio_ctrl, page, eb->start, eb->len,
4555 eb->start - page_offset(page),
4556 end_bio_subpage_eb_writepage, 0, false);
4558 btrfs_subpage_clear_writeback(fs_info, page, eb->start, eb->len);
4559 set_btree_ioerr(page, eb);
4562 if (atomic_dec_and_test(&eb->io_pages))
4563 end_extent_buffer_writeback(eb);
4568 * Submission finished without problem, if no range of the page is
4569 * dirty anymore, we have submitted a page. Update nr_written in wbc.
4576 static noinline_for_stack int write_one_eb(struct extent_buffer *eb,
4577 struct writeback_control *wbc,
4578 struct extent_page_data *epd)
4580 u64 disk_bytenr = eb->start;
4582 unsigned int write_flags = wbc_to_write_flags(wbc);
4585 prepare_eb_write(eb);
4587 num_pages = num_extent_pages(eb);
4588 for (i = 0; i < num_pages; i++) {
4589 struct page *p = eb->pages[i];
4591 clear_page_dirty_for_io(p);
4592 set_page_writeback(p);
4593 ret = submit_extent_page(REQ_OP_WRITE | write_flags, wbc,
4594 &epd->bio_ctrl, p, disk_bytenr,
4596 end_bio_extent_buffer_writepage,
4599 set_btree_ioerr(p, eb);
4600 if (PageWriteback(p))
4601 end_page_writeback(p);
4602 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
4603 end_extent_buffer_writeback(eb);
4607 disk_bytenr += PAGE_SIZE;
4612 if (unlikely(ret)) {
4613 for (; i < num_pages; i++) {
4614 struct page *p = eb->pages[i];
4615 clear_page_dirty_for_io(p);
4624 * Submit one subpage btree page.
4626 * The main difference to submit_eb_page() is:
4628 * For subpage, we don't rely on page locking at all.
4631 * We only flush bio if we may be unable to fit current extent buffers into
4634 * Return >=0 for the number of submitted extent buffers.
4635 * Return <0 for fatal error.
4637 static int submit_eb_subpage(struct page *page,
4638 struct writeback_control *wbc,
4639 struct extent_page_data *epd)
4641 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
4643 u64 page_start = page_offset(page);
4645 int sectors_per_node = fs_info->nodesize >> fs_info->sectorsize_bits;
4648 /* Lock and write each dirty extent buffers in the range */
4649 while (bit_start < fs_info->subpage_info->bitmap_nr_bits) {
4650 struct btrfs_subpage *subpage = (struct btrfs_subpage *)page->private;
4651 struct extent_buffer *eb;
4652 unsigned long flags;
4656 * Take private lock to ensure the subpage won't be detached
4659 spin_lock(&page->mapping->private_lock);
4660 if (!PagePrivate(page)) {
4661 spin_unlock(&page->mapping->private_lock);
4664 spin_lock_irqsave(&subpage->lock, flags);
4665 if (!test_bit(bit_start + fs_info->subpage_info->dirty_offset,
4666 subpage->bitmaps)) {
4667 spin_unlock_irqrestore(&subpage->lock, flags);
4668 spin_unlock(&page->mapping->private_lock);
4673 start = page_start + bit_start * fs_info->sectorsize;
4674 bit_start += sectors_per_node;
4677 * Here we just want to grab the eb without touching extra
4678 * spin locks, so call find_extent_buffer_nolock().
4680 eb = find_extent_buffer_nolock(fs_info, start);
4681 spin_unlock_irqrestore(&subpage->lock, flags);
4682 spin_unlock(&page->mapping->private_lock);
4685 * The eb has already reached 0 refs thus find_extent_buffer()
4686 * doesn't return it. We don't need to write back such eb
4692 ret = lock_extent_buffer_for_io(eb, epd);
4694 free_extent_buffer(eb);
4698 free_extent_buffer(eb);
4701 ret = write_one_subpage_eb(eb, wbc, epd);
4702 free_extent_buffer(eb);
4710 /* We hit error, end bio for the submitted extent buffers */
4711 submit_write_bio(epd, ret);
4716 * Submit all page(s) of one extent buffer.
4718 * @page: the page of one extent buffer
4719 * @eb_context: to determine if we need to submit this page, if current page
4720 * belongs to this eb, we don't need to submit
4722 * The caller should pass each page in their bytenr order, and here we use
4723 * @eb_context to determine if we have submitted pages of one extent buffer.
4725 * If we have, we just skip until we hit a new page that doesn't belong to
4726 * current @eb_context.
4728 * If not, we submit all the page(s) of the extent buffer.
4730 * Return >0 if we have submitted the extent buffer successfully.
4731 * Return 0 if we don't need to submit the page, as it's already submitted by
4733 * Return <0 for fatal error.
4735 static int submit_eb_page(struct page *page, struct writeback_control *wbc,
4736 struct extent_page_data *epd,
4737 struct extent_buffer **eb_context)
4739 struct address_space *mapping = page->mapping;
4740 struct btrfs_block_group *cache = NULL;
4741 struct extent_buffer *eb;
4744 if (!PagePrivate(page))
4747 if (btrfs_sb(page->mapping->host->i_sb)->nodesize < PAGE_SIZE)
4748 return submit_eb_subpage(page, wbc, epd);
4750 spin_lock(&mapping->private_lock);
4751 if (!PagePrivate(page)) {
4752 spin_unlock(&mapping->private_lock);
4756 eb = (struct extent_buffer *)page->private;
4759 * Shouldn't happen and normally this would be a BUG_ON but no point
4760 * crashing the machine for something we can survive anyway.
4763 spin_unlock(&mapping->private_lock);
4767 if (eb == *eb_context) {
4768 spin_unlock(&mapping->private_lock);
4771 ret = atomic_inc_not_zero(&eb->refs);
4772 spin_unlock(&mapping->private_lock);
4776 if (!btrfs_check_meta_write_pointer(eb->fs_info, eb, &cache)) {
4778 * If for_sync, this hole will be filled with
4779 * trasnsaction commit.
4781 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
4785 free_extent_buffer(eb);
4791 ret = lock_extent_buffer_for_io(eb, epd);
4793 btrfs_revert_meta_write_pointer(cache, eb);
4795 btrfs_put_block_group(cache);
4796 free_extent_buffer(eb);
4801 * Implies write in zoned mode. Mark the last eb in a block group.
4803 btrfs_schedule_zone_finish_bg(cache, eb);
4804 btrfs_put_block_group(cache);
4806 ret = write_one_eb(eb, wbc, epd);
4807 free_extent_buffer(eb);
4813 int btree_write_cache_pages(struct address_space *mapping,
4814 struct writeback_control *wbc)
4816 struct extent_buffer *eb_context = NULL;
4817 struct extent_page_data epd = {
4820 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4822 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
4825 int nr_to_write_done = 0;
4826 struct pagevec pvec;
4829 pgoff_t end; /* Inclusive */
4833 pagevec_init(&pvec);
4834 if (wbc->range_cyclic) {
4835 index = mapping->writeback_index; /* Start from prev offset */
4838 * Start from the beginning does not need to cycle over the
4839 * range, mark it as scanned.
4841 scanned = (index == 0);
4843 index = wbc->range_start >> PAGE_SHIFT;
4844 end = wbc->range_end >> PAGE_SHIFT;
4847 if (wbc->sync_mode == WB_SYNC_ALL)
4848 tag = PAGECACHE_TAG_TOWRITE;
4850 tag = PAGECACHE_TAG_DIRTY;
4851 btrfs_zoned_meta_io_lock(fs_info);
4853 if (wbc->sync_mode == WB_SYNC_ALL)
4854 tag_pages_for_writeback(mapping, index, end);
4855 while (!done && !nr_to_write_done && (index <= end) &&
4856 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
4860 for (i = 0; i < nr_pages; i++) {
4861 struct page *page = pvec.pages[i];
4863 ret = submit_eb_page(page, wbc, &epd, &eb_context);
4872 * the filesystem may choose to bump up nr_to_write.
4873 * We have to make sure to honor the new nr_to_write
4876 nr_to_write_done = wbc->nr_to_write <= 0;
4878 pagevec_release(&pvec);
4881 if (!scanned && !done) {
4883 * We hit the last page and there is more work to be done: wrap
4884 * back to the start of the file
4891 * If something went wrong, don't allow any metadata write bio to be
4894 * This would prevent use-after-free if we had dirty pages not
4895 * cleaned up, which can still happen by fuzzed images.
4898 * Allowing existing tree block to be allocated for other trees.
4900 * - Log tree operations
4901 * Exiting tree blocks get allocated to log tree, bumps its
4902 * generation, then get cleaned in tree re-balance.
4903 * Such tree block will not be written back, since it's clean,
4904 * thus no WRITTEN flag set.
4905 * And after log writes back, this tree block is not traced by
4906 * any dirty extent_io_tree.
4908 * - Offending tree block gets re-dirtied from its original owner
4909 * Since it has bumped generation, no WRITTEN flag, it can be
4910 * reused without COWing. This tree block will not be traced
4911 * by btrfs_transaction::dirty_pages.
4913 * Now such dirty tree block will not be cleaned by any dirty
4914 * extent io tree. Thus we don't want to submit such wild eb
4915 * if the fs already has error.
4917 * We can get ret > 0 from submit_extent_page() indicating how many ebs
4918 * were submitted. Reset it to 0 to avoid false alerts for the caller.
4922 if (!ret && BTRFS_FS_ERROR(fs_info))
4924 submit_write_bio(&epd, ret);
4926 btrfs_zoned_meta_io_unlock(fs_info);
4931 * Walk the list of dirty pages of the given address space and write all of them.
4933 * @mapping: address space structure to write
4934 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
4935 * @epd: holds context for the write, namely the bio
4937 * If a page is already under I/O, write_cache_pages() skips it, even
4938 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
4939 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
4940 * and msync() need to guarantee that all the data which was dirty at the time
4941 * the call was made get new I/O started against them. If wbc->sync_mode is
4942 * WB_SYNC_ALL then we were called for data integrity and we must wait for
4943 * existing IO to complete.
4945 static int extent_write_cache_pages(struct address_space *mapping,
4946 struct writeback_control *wbc,
4947 struct extent_page_data *epd)
4949 struct inode *inode = mapping->host;
4952 int nr_to_write_done = 0;
4953 struct pagevec pvec;
4956 pgoff_t end; /* Inclusive */
4958 int range_whole = 0;
4963 * We have to hold onto the inode so that ordered extents can do their
4964 * work when the IO finishes. The alternative to this is failing to add
4965 * an ordered extent if the igrab() fails there and that is a huge pain
4966 * to deal with, so instead just hold onto the inode throughout the
4967 * writepages operation. If it fails here we are freeing up the inode
4968 * anyway and we'd rather not waste our time writing out stuff that is
4969 * going to be truncated anyway.
4974 pagevec_init(&pvec);
4975 if (wbc->range_cyclic) {
4976 index = mapping->writeback_index; /* Start from prev offset */
4979 * Start from the beginning does not need to cycle over the
4980 * range, mark it as scanned.
4982 scanned = (index == 0);
4984 index = wbc->range_start >> PAGE_SHIFT;
4985 end = wbc->range_end >> PAGE_SHIFT;
4986 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
4992 * We do the tagged writepage as long as the snapshot flush bit is set
4993 * and we are the first one who do the filemap_flush() on this inode.
4995 * The nr_to_write == LONG_MAX is needed to make sure other flushers do
4996 * not race in and drop the bit.
4998 if (range_whole && wbc->nr_to_write == LONG_MAX &&
4999 test_and_clear_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
5000 &BTRFS_I(inode)->runtime_flags))
5001 wbc->tagged_writepages = 1;
5003 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
5004 tag = PAGECACHE_TAG_TOWRITE;
5006 tag = PAGECACHE_TAG_DIRTY;
5008 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
5009 tag_pages_for_writeback(mapping, index, end);
5011 while (!done && !nr_to_write_done && (index <= end) &&
5012 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping,
5013 &index, end, tag))) {
5016 for (i = 0; i < nr_pages; i++) {
5017 struct page *page = pvec.pages[i];
5019 done_index = page->index + 1;
5021 * At this point we hold neither the i_pages lock nor
5022 * the page lock: the page may be truncated or
5023 * invalidated (changing page->mapping to NULL),
5024 * or even swizzled back from swapper_space to
5025 * tmpfs file mapping
5027 if (!trylock_page(page)) {
5028 submit_write_bio(epd, 0);
5032 if (unlikely(page->mapping != mapping)) {
5037 if (wbc->sync_mode != WB_SYNC_NONE) {
5038 if (PageWriteback(page))
5039 submit_write_bio(epd, 0);
5040 wait_on_page_writeback(page);
5043 if (PageWriteback(page) ||
5044 !clear_page_dirty_for_io(page)) {
5049 ret = __extent_writepage(page, wbc, epd);
5056 * the filesystem may choose to bump up nr_to_write.
5057 * We have to make sure to honor the new nr_to_write
5060 nr_to_write_done = wbc->nr_to_write <= 0;
5062 pagevec_release(&pvec);
5065 if (!scanned && !done) {
5067 * We hit the last page and there is more work to be done: wrap
5068 * back to the start of the file
5074 * If we're looping we could run into a page that is locked by a
5075 * writer and that writer could be waiting on writeback for a
5076 * page in our current bio, and thus deadlock, so flush the
5079 submit_write_bio(epd, 0);
5083 if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole))
5084 mapping->writeback_index = done_index;
5086 btrfs_add_delayed_iput(inode);
5090 int extent_write_full_page(struct page *page, struct writeback_control *wbc)
5093 struct extent_page_data epd = {
5096 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
5099 ret = __extent_writepage(page, wbc, &epd);
5100 submit_write_bio(&epd, ret);
5105 * Submit the pages in the range to bio for call sites which delalloc range has
5106 * already been ran (aka, ordered extent inserted) and all pages are still
5109 int extent_write_locked_range(struct inode *inode, u64 start, u64 end)
5111 bool found_error = false;
5112 int first_error = 0;
5114 struct address_space *mapping = inode->i_mapping;
5117 unsigned long nr_pages;
5118 const u32 sectorsize = btrfs_sb(inode->i_sb)->sectorsize;
5119 struct extent_page_data epd = {
5124 struct writeback_control wbc_writepages = {
5125 .sync_mode = WB_SYNC_ALL,
5126 .range_start = start,
5127 .range_end = end + 1,
5128 /* We're called from an async helper function */
5129 .punt_to_cgroup = 1,
5130 .no_cgroup_owner = 1,
5133 ASSERT(IS_ALIGNED(start, sectorsize) && IS_ALIGNED(end + 1, sectorsize));
5134 nr_pages = (round_up(end, PAGE_SIZE) - round_down(start, PAGE_SIZE)) >>
5136 wbc_writepages.nr_to_write = nr_pages * 2;
5138 wbc_attach_fdatawrite_inode(&wbc_writepages, inode);
5139 while (cur <= end) {
5140 u64 cur_end = min(round_down(cur, PAGE_SIZE) + PAGE_SIZE - 1, end);
5142 page = find_get_page(mapping, cur >> PAGE_SHIFT);
5144 * All pages in the range are locked since
5145 * btrfs_run_delalloc_range(), thus there is no way to clear
5146 * the page dirty flag.
5148 ASSERT(PageLocked(page));
5149 ASSERT(PageDirty(page));
5150 clear_page_dirty_for_io(page);
5151 ret = __extent_writepage(page, &wbc_writepages, &epd);
5161 submit_write_bio(&epd, found_error ? ret : 0);
5163 wbc_detach_inode(&wbc_writepages);
5169 int extent_writepages(struct address_space *mapping,
5170 struct writeback_control *wbc)
5172 struct inode *inode = mapping->host;
5174 struct extent_page_data epd = {
5177 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
5181 * Allow only a single thread to do the reloc work in zoned mode to
5182 * protect the write pointer updates.
5184 btrfs_zoned_data_reloc_lock(BTRFS_I(inode));
5185 ret = extent_write_cache_pages(mapping, wbc, &epd);
5186 submit_write_bio(&epd, ret);
5187 btrfs_zoned_data_reloc_unlock(BTRFS_I(inode));
5191 void extent_readahead(struct readahead_control *rac)
5193 struct btrfs_bio_ctrl bio_ctrl = { 0 };
5194 struct page *pagepool[16];
5195 struct extent_map *em_cached = NULL;
5196 u64 prev_em_start = (u64)-1;
5199 while ((nr = readahead_page_batch(rac, pagepool))) {
5200 u64 contig_start = readahead_pos(rac);
5201 u64 contig_end = contig_start + readahead_batch_length(rac) - 1;
5203 contiguous_readpages(pagepool, nr, contig_start, contig_end,
5204 &em_cached, &bio_ctrl, &prev_em_start);
5208 free_extent_map(em_cached);
5209 submit_one_bio(&bio_ctrl);
5213 * basic invalidate_folio code, this waits on any locked or writeback
5214 * ranges corresponding to the folio, and then deletes any extent state
5215 * records from the tree
5217 int extent_invalidate_folio(struct extent_io_tree *tree,
5218 struct folio *folio, size_t offset)
5220 struct extent_state *cached_state = NULL;
5221 u64 start = folio_pos(folio);
5222 u64 end = start + folio_size(folio) - 1;
5223 size_t blocksize = folio->mapping->host->i_sb->s_blocksize;
5225 /* This function is only called for the btree inode */
5226 ASSERT(tree->owner == IO_TREE_BTREE_INODE_IO);
5228 start += ALIGN(offset, blocksize);
5232 lock_extent_bits(tree, start, end, &cached_state);
5233 folio_wait_writeback(folio);
5236 * Currently for btree io tree, only EXTENT_LOCKED is utilized,
5237 * so here we only need to unlock the extent range to free any
5238 * existing extent state.
5240 unlock_extent_cached(tree, start, end, &cached_state);
5245 * a helper for release_folio, this tests for areas of the page that
5246 * are locked or under IO and drops the related state bits if it is safe
5249 static int try_release_extent_state(struct extent_io_tree *tree,
5250 struct page *page, gfp_t mask)
5252 u64 start = page_offset(page);
5253 u64 end = start + PAGE_SIZE - 1;
5256 if (test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL)) {
5260 * At this point we can safely clear everything except the
5261 * locked bit, the nodatasum bit and the delalloc new bit.
5262 * The delalloc new bit will be cleared by ordered extent
5265 ret = __clear_extent_bit(tree, start, end,
5266 ~(EXTENT_LOCKED | EXTENT_NODATASUM | EXTENT_DELALLOC_NEW),
5267 0, 0, NULL, mask, NULL);
5269 /* if clear_extent_bit failed for enomem reasons,
5270 * we can't allow the release to continue.
5281 * a helper for release_folio. As long as there are no locked extents
5282 * in the range corresponding to the page, both state records and extent
5283 * map records are removed
5285 int try_release_extent_mapping(struct page *page, gfp_t mask)
5287 struct extent_map *em;
5288 u64 start = page_offset(page);
5289 u64 end = start + PAGE_SIZE - 1;
5290 struct btrfs_inode *btrfs_inode = BTRFS_I(page->mapping->host);
5291 struct extent_io_tree *tree = &btrfs_inode->io_tree;
5292 struct extent_map_tree *map = &btrfs_inode->extent_tree;
5294 if (gfpflags_allow_blocking(mask) &&
5295 page->mapping->host->i_size > SZ_16M) {
5297 while (start <= end) {
5298 struct btrfs_fs_info *fs_info;
5301 len = end - start + 1;
5302 write_lock(&map->lock);
5303 em = lookup_extent_mapping(map, start, len);
5305 write_unlock(&map->lock);
5308 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
5309 em->start != start) {
5310 write_unlock(&map->lock);
5311 free_extent_map(em);
5314 if (test_range_bit(tree, em->start,
5315 extent_map_end(em) - 1,
5316 EXTENT_LOCKED, 0, NULL))
5319 * If it's not in the list of modified extents, used
5320 * by a fast fsync, we can remove it. If it's being
5321 * logged we can safely remove it since fsync took an
5322 * extra reference on the em.
5324 if (list_empty(&em->list) ||
5325 test_bit(EXTENT_FLAG_LOGGING, &em->flags))
5328 * If it's in the list of modified extents, remove it
5329 * only if its generation is older then the current one,
5330 * in which case we don't need it for a fast fsync.
5331 * Otherwise don't remove it, we could be racing with an
5332 * ongoing fast fsync that could miss the new extent.
5334 fs_info = btrfs_inode->root->fs_info;
5335 spin_lock(&fs_info->trans_lock);
5336 cur_gen = fs_info->generation;
5337 spin_unlock(&fs_info->trans_lock);
5338 if (em->generation >= cur_gen)
5342 * We only remove extent maps that are not in the list of
5343 * modified extents or that are in the list but with a
5344 * generation lower then the current generation, so there
5345 * is no need to set the full fsync flag on the inode (it
5346 * hurts the fsync performance for workloads with a data
5347 * size that exceeds or is close to the system's memory).
5349 remove_extent_mapping(map, em);
5350 /* once for the rb tree */
5351 free_extent_map(em);
5353 start = extent_map_end(em);
5354 write_unlock(&map->lock);
5357 free_extent_map(em);
5359 cond_resched(); /* Allow large-extent preemption. */
5362 return try_release_extent_state(tree, page, mask);
5366 * helper function for fiemap, which doesn't want to see any holes.
5367 * This maps until we find something past 'last'
5369 static struct extent_map *get_extent_skip_holes(struct btrfs_inode *inode,
5370 u64 offset, u64 last)
5372 u64 sectorsize = btrfs_inode_sectorsize(inode);
5373 struct extent_map *em;
5380 len = last - offset;
5383 len = ALIGN(len, sectorsize);
5384 em = btrfs_get_extent_fiemap(inode, offset, len);
5388 /* if this isn't a hole return it */
5389 if (em->block_start != EXTENT_MAP_HOLE)
5392 /* this is a hole, advance to the next extent */
5393 offset = extent_map_end(em);
5394 free_extent_map(em);
5402 * To cache previous fiemap extent
5404 * Will be used for merging fiemap extent
5406 struct fiemap_cache {
5415 * Helper to submit fiemap extent.
5417 * Will try to merge current fiemap extent specified by @offset, @phys,
5418 * @len and @flags with cached one.
5419 * And only when we fails to merge, cached one will be submitted as
5422 * Return value is the same as fiemap_fill_next_extent().
5424 static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
5425 struct fiemap_cache *cache,
5426 u64 offset, u64 phys, u64 len, u32 flags)
5434 * Sanity check, extent_fiemap() should have ensured that new
5435 * fiemap extent won't overlap with cached one.
5438 * NOTE: Physical address can overlap, due to compression
5440 if (cache->offset + cache->len > offset) {
5446 * Only merges fiemap extents if
5447 * 1) Their logical addresses are continuous
5449 * 2) Their physical addresses are continuous
5450 * So truly compressed (physical size smaller than logical size)
5451 * extents won't get merged with each other
5453 * 3) Share same flags except FIEMAP_EXTENT_LAST
5454 * So regular extent won't get merged with prealloc extent
5456 if (cache->offset + cache->len == offset &&
5457 cache->phys + cache->len == phys &&
5458 (cache->flags & ~FIEMAP_EXTENT_LAST) ==
5459 (flags & ~FIEMAP_EXTENT_LAST)) {
5461 cache->flags |= flags;
5462 goto try_submit_last;
5465 /* Not mergeable, need to submit cached one */
5466 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
5467 cache->len, cache->flags);
5468 cache->cached = false;
5472 cache->cached = true;
5473 cache->offset = offset;
5476 cache->flags = flags;
5478 if (cache->flags & FIEMAP_EXTENT_LAST) {
5479 ret = fiemap_fill_next_extent(fieinfo, cache->offset,
5480 cache->phys, cache->len, cache->flags);
5481 cache->cached = false;
5487 * Emit last fiemap cache
5489 * The last fiemap cache may still be cached in the following case:
5491 * |<- Fiemap range ->|
5492 * |<------------ First extent ----------->|
5494 * In this case, the first extent range will be cached but not emitted.
5495 * So we must emit it before ending extent_fiemap().
5497 static int emit_last_fiemap_cache(struct fiemap_extent_info *fieinfo,
5498 struct fiemap_cache *cache)
5505 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
5506 cache->len, cache->flags);
5507 cache->cached = false;
5513 int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,
5518 u64 max = start + len;
5522 u64 last_for_get_extent = 0;
5524 u64 isize = i_size_read(&inode->vfs_inode);
5525 struct btrfs_key found_key;
5526 struct extent_map *em = NULL;
5527 struct extent_state *cached_state = NULL;
5528 struct btrfs_path *path;
5529 struct btrfs_root *root = inode->root;
5530 struct fiemap_cache cache = { 0 };
5531 struct ulist *roots;
5532 struct ulist *tmp_ulist;
5541 path = btrfs_alloc_path();
5545 roots = ulist_alloc(GFP_KERNEL);
5546 tmp_ulist = ulist_alloc(GFP_KERNEL);
5547 if (!roots || !tmp_ulist) {
5549 goto out_free_ulist;
5553 * We can't initialize that to 'start' as this could miss extents due
5554 * to extent item merging
5557 start = round_down(start, btrfs_inode_sectorsize(inode));
5558 len = round_up(max, btrfs_inode_sectorsize(inode)) - start;
5561 * lookup the last file extent. We're not using i_size here
5562 * because there might be preallocation past i_size
5564 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode), -1,
5567 goto out_free_ulist;
5575 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
5576 found_type = found_key.type;
5578 /* No extents, but there might be delalloc bits */
5579 if (found_key.objectid != btrfs_ino(inode) ||
5580 found_type != BTRFS_EXTENT_DATA_KEY) {
5581 /* have to trust i_size as the end */
5583 last_for_get_extent = isize;
5586 * remember the start of the last extent. There are a
5587 * bunch of different factors that go into the length of the
5588 * extent, so its much less complex to remember where it started
5590 last = found_key.offset;
5591 last_for_get_extent = last + 1;
5593 btrfs_release_path(path);
5596 * we might have some extents allocated but more delalloc past those
5597 * extents. so, we trust isize unless the start of the last extent is
5602 last_for_get_extent = isize;
5605 lock_extent_bits(&inode->io_tree, start, start + len - 1,
5608 em = get_extent_skip_holes(inode, start, last_for_get_extent);
5617 u64 offset_in_extent = 0;
5619 /* break if the extent we found is outside the range */
5620 if (em->start >= max || extent_map_end(em) < off)
5624 * get_extent may return an extent that starts before our
5625 * requested range. We have to make sure the ranges
5626 * we return to fiemap always move forward and don't
5627 * overlap, so adjust the offsets here
5629 em_start = max(em->start, off);
5632 * record the offset from the start of the extent
5633 * for adjusting the disk offset below. Only do this if the
5634 * extent isn't compressed since our in ram offset may be past
5635 * what we have actually allocated on disk.
5637 if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
5638 offset_in_extent = em_start - em->start;
5639 em_end = extent_map_end(em);
5640 em_len = em_end - em_start;
5642 if (em->block_start < EXTENT_MAP_LAST_BYTE)
5643 disko = em->block_start + offset_in_extent;
5648 * bump off for our next call to get_extent
5650 off = extent_map_end(em);
5654 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
5656 flags |= FIEMAP_EXTENT_LAST;
5657 } else if (em->block_start == EXTENT_MAP_INLINE) {
5658 flags |= (FIEMAP_EXTENT_DATA_INLINE |
5659 FIEMAP_EXTENT_NOT_ALIGNED);
5660 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
5661 flags |= (FIEMAP_EXTENT_DELALLOC |
5662 FIEMAP_EXTENT_UNKNOWN);
5663 } else if (fieinfo->fi_extents_max) {
5664 u64 bytenr = em->block_start -
5665 (em->start - em->orig_start);
5668 * As btrfs supports shared space, this information
5669 * can be exported to userspace tools via
5670 * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0
5671 * then we're just getting a count and we can skip the
5674 ret = btrfs_check_shared(root, btrfs_ino(inode),
5675 bytenr, roots, tmp_ulist);
5679 flags |= FIEMAP_EXTENT_SHARED;
5682 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
5683 flags |= FIEMAP_EXTENT_ENCODED;
5684 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5685 flags |= FIEMAP_EXTENT_UNWRITTEN;
5687 free_extent_map(em);
5689 if ((em_start >= last) || em_len == (u64)-1 ||
5690 (last == (u64)-1 && isize <= em_end)) {
5691 flags |= FIEMAP_EXTENT_LAST;
5695 /* now scan forward to see if this is really the last extent. */
5696 em = get_extent_skip_holes(inode, off, last_for_get_extent);
5702 flags |= FIEMAP_EXTENT_LAST;
5705 ret = emit_fiemap_extent(fieinfo, &cache, em_start, disko,
5715 ret = emit_last_fiemap_cache(fieinfo, &cache);
5716 free_extent_map(em);
5718 unlock_extent_cached(&inode->io_tree, start, start + len - 1,
5722 btrfs_free_path(path);
5724 ulist_free(tmp_ulist);
5728 static void __free_extent_buffer(struct extent_buffer *eb)
5730 kmem_cache_free(extent_buffer_cache, eb);
5733 int extent_buffer_under_io(const struct extent_buffer *eb)
5735 return (atomic_read(&eb->io_pages) ||
5736 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
5737 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
5740 static bool page_range_has_eb(struct btrfs_fs_info *fs_info, struct page *page)
5742 struct btrfs_subpage *subpage;
5744 lockdep_assert_held(&page->mapping->private_lock);
5746 if (PagePrivate(page)) {
5747 subpage = (struct btrfs_subpage *)page->private;
5748 if (atomic_read(&subpage->eb_refs))
5751 * Even there is no eb refs here, we may still have
5752 * end_page_read() call relying on page::private.
5754 if (atomic_read(&subpage->readers))
5760 static void detach_extent_buffer_page(struct extent_buffer *eb, struct page *page)
5762 struct btrfs_fs_info *fs_info = eb->fs_info;
5763 const bool mapped = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
5766 * For mapped eb, we're going to change the page private, which should
5767 * be done under the private_lock.
5770 spin_lock(&page->mapping->private_lock);
5772 if (!PagePrivate(page)) {
5774 spin_unlock(&page->mapping->private_lock);
5778 if (fs_info->nodesize >= PAGE_SIZE) {
5780 * We do this since we'll remove the pages after we've
5781 * removed the eb from the radix tree, so we could race
5782 * and have this page now attached to the new eb. So
5783 * only clear page_private if it's still connected to
5786 if (PagePrivate(page) &&
5787 page->private == (unsigned long)eb) {
5788 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
5789 BUG_ON(PageDirty(page));
5790 BUG_ON(PageWriteback(page));
5792 * We need to make sure we haven't be attached
5795 detach_page_private(page);
5798 spin_unlock(&page->mapping->private_lock);
5803 * For subpage, we can have dummy eb with page private. In this case,
5804 * we can directly detach the private as such page is only attached to
5805 * one dummy eb, no sharing.
5808 btrfs_detach_subpage(fs_info, page);
5812 btrfs_page_dec_eb_refs(fs_info, page);
5815 * We can only detach the page private if there are no other ebs in the
5816 * page range and no unfinished IO.
5818 if (!page_range_has_eb(fs_info, page))
5819 btrfs_detach_subpage(fs_info, page);
5821 spin_unlock(&page->mapping->private_lock);
5824 /* Release all pages attached to the extent buffer */
5825 static void btrfs_release_extent_buffer_pages(struct extent_buffer *eb)
5830 ASSERT(!extent_buffer_under_io(eb));
5832 num_pages = num_extent_pages(eb);
5833 for (i = 0; i < num_pages; i++) {
5834 struct page *page = eb->pages[i];
5839 detach_extent_buffer_page(eb, page);
5841 /* One for when we allocated the page */
5847 * Helper for releasing the extent buffer.
5849 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
5851 btrfs_release_extent_buffer_pages(eb);
5852 btrfs_leak_debug_del(&eb->fs_info->eb_leak_lock, &eb->leak_list);
5853 __free_extent_buffer(eb);
5856 static struct extent_buffer *
5857 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
5860 struct extent_buffer *eb = NULL;
5862 eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL);
5865 eb->fs_info = fs_info;
5867 init_rwsem(&eb->lock);
5869 btrfs_leak_debug_add(&fs_info->eb_leak_lock, &eb->leak_list,
5870 &fs_info->allocated_ebs);
5871 INIT_LIST_HEAD(&eb->release_list);
5873 spin_lock_init(&eb->refs_lock);
5874 atomic_set(&eb->refs, 1);
5875 atomic_set(&eb->io_pages, 0);
5877 ASSERT(len <= BTRFS_MAX_METADATA_BLOCKSIZE);
5882 struct extent_buffer *btrfs_clone_extent_buffer(const struct extent_buffer *src)
5885 struct extent_buffer *new;
5886 int num_pages = num_extent_pages(src);
5889 new = __alloc_extent_buffer(src->fs_info, src->start, src->len);
5894 * Set UNMAPPED before calling btrfs_release_extent_buffer(), as
5895 * btrfs_release_extent_buffer() have different behavior for
5896 * UNMAPPED subpage extent buffer.
5898 set_bit(EXTENT_BUFFER_UNMAPPED, &new->bflags);
5900 memset(new->pages, 0, sizeof(*new->pages) * num_pages);
5901 ret = btrfs_alloc_page_array(num_pages, new->pages);
5903 btrfs_release_extent_buffer(new);
5907 for (i = 0; i < num_pages; i++) {
5909 struct page *p = new->pages[i];
5911 ret = attach_extent_buffer_page(new, p, NULL);
5913 btrfs_release_extent_buffer(new);
5916 WARN_ON(PageDirty(p));
5917 copy_page(page_address(p), page_address(src->pages[i]));
5919 set_extent_buffer_uptodate(new);
5924 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
5925 u64 start, unsigned long len)
5927 struct extent_buffer *eb;
5932 eb = __alloc_extent_buffer(fs_info, start, len);
5936 num_pages = num_extent_pages(eb);
5937 ret = btrfs_alloc_page_array(num_pages, eb->pages);
5941 for (i = 0; i < num_pages; i++) {
5942 struct page *p = eb->pages[i];
5944 ret = attach_extent_buffer_page(eb, p, NULL);
5949 set_extent_buffer_uptodate(eb);
5950 btrfs_set_header_nritems(eb, 0);
5951 set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
5955 for (i = 0; i < num_pages; i++) {
5957 detach_extent_buffer_page(eb, eb->pages[i]);
5958 __free_page(eb->pages[i]);
5961 __free_extent_buffer(eb);
5965 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
5968 return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize);
5971 static void check_buffer_tree_ref(struct extent_buffer *eb)
5975 * The TREE_REF bit is first set when the extent_buffer is added
5976 * to the radix tree. It is also reset, if unset, when a new reference
5977 * is created by find_extent_buffer.
5979 * It is only cleared in two cases: freeing the last non-tree
5980 * reference to the extent_buffer when its STALE bit is set or
5981 * calling release_folio when the tree reference is the only reference.
5983 * In both cases, care is taken to ensure that the extent_buffer's
5984 * pages are not under io. However, release_folio can be concurrently
5985 * called with creating new references, which is prone to race
5986 * conditions between the calls to check_buffer_tree_ref in those
5987 * codepaths and clearing TREE_REF in try_release_extent_buffer.
5989 * The actual lifetime of the extent_buffer in the radix tree is
5990 * adequately protected by the refcount, but the TREE_REF bit and
5991 * its corresponding reference are not. To protect against this
5992 * class of races, we call check_buffer_tree_ref from the codepaths
5993 * which trigger io after they set eb->io_pages. Note that once io is
5994 * initiated, TREE_REF can no longer be cleared, so that is the
5995 * moment at which any such race is best fixed.
5997 refs = atomic_read(&eb->refs);
5998 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
6001 spin_lock(&eb->refs_lock);
6002 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
6003 atomic_inc(&eb->refs);
6004 spin_unlock(&eb->refs_lock);
6007 static void mark_extent_buffer_accessed(struct extent_buffer *eb,
6008 struct page *accessed)
6012 check_buffer_tree_ref(eb);
6014 num_pages = num_extent_pages(eb);
6015 for (i = 0; i < num_pages; i++) {
6016 struct page *p = eb->pages[i];
6019 mark_page_accessed(p);
6023 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
6026 struct extent_buffer *eb;
6028 eb = find_extent_buffer_nolock(fs_info, start);
6032 * Lock our eb's refs_lock to avoid races with free_extent_buffer().
6033 * When we get our eb it might be flagged with EXTENT_BUFFER_STALE and
6034 * another task running free_extent_buffer() might have seen that flag
6035 * set, eb->refs == 2, that the buffer isn't under IO (dirty and
6036 * writeback flags not set) and it's still in the tree (flag
6037 * EXTENT_BUFFER_TREE_REF set), therefore being in the process of
6038 * decrementing the extent buffer's reference count twice. So here we
6039 * could race and increment the eb's reference count, clear its stale
6040 * flag, mark it as dirty and drop our reference before the other task
6041 * finishes executing free_extent_buffer, which would later result in
6042 * an attempt to free an extent buffer that is dirty.
6044 if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) {
6045 spin_lock(&eb->refs_lock);
6046 spin_unlock(&eb->refs_lock);
6048 mark_extent_buffer_accessed(eb, NULL);
6052 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
6053 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
6056 struct extent_buffer *eb, *exists = NULL;
6059 eb = find_extent_buffer(fs_info, start);
6062 eb = alloc_dummy_extent_buffer(fs_info, start);
6064 return ERR_PTR(-ENOMEM);
6065 eb->fs_info = fs_info;
6067 ret = radix_tree_preload(GFP_NOFS);
6069 exists = ERR_PTR(ret);
6072 spin_lock(&fs_info->buffer_lock);
6073 ret = radix_tree_insert(&fs_info->buffer_radix,
6074 start >> fs_info->sectorsize_bits, eb);
6075 spin_unlock(&fs_info->buffer_lock);
6076 radix_tree_preload_end();
6077 if (ret == -EEXIST) {
6078 exists = find_extent_buffer(fs_info, start);
6084 check_buffer_tree_ref(eb);
6085 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
6089 btrfs_release_extent_buffer(eb);
6094 static struct extent_buffer *grab_extent_buffer(
6095 struct btrfs_fs_info *fs_info, struct page *page)
6097 struct extent_buffer *exists;
6100 * For subpage case, we completely rely on radix tree to ensure we
6101 * don't try to insert two ebs for the same bytenr. So here we always
6102 * return NULL and just continue.
6104 if (fs_info->nodesize < PAGE_SIZE)
6107 /* Page not yet attached to an extent buffer */
6108 if (!PagePrivate(page))
6112 * We could have already allocated an eb for this page and attached one
6113 * so lets see if we can get a ref on the existing eb, and if we can we
6114 * know it's good and we can just return that one, else we know we can
6115 * just overwrite page->private.
6117 exists = (struct extent_buffer *)page->private;
6118 if (atomic_inc_not_zero(&exists->refs))
6121 WARN_ON(PageDirty(page));
6122 detach_page_private(page);
6126 static int check_eb_alignment(struct btrfs_fs_info *fs_info, u64 start)
6128 if (!IS_ALIGNED(start, fs_info->sectorsize)) {
6129 btrfs_err(fs_info, "bad tree block start %llu", start);
6133 if (fs_info->nodesize < PAGE_SIZE &&
6134 offset_in_page(start) + fs_info->nodesize > PAGE_SIZE) {
6136 "tree block crosses page boundary, start %llu nodesize %u",
6137 start, fs_info->nodesize);
6140 if (fs_info->nodesize >= PAGE_SIZE &&
6141 !PAGE_ALIGNED(start)) {
6143 "tree block is not page aligned, start %llu nodesize %u",
6144 start, fs_info->nodesize);
6150 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
6151 u64 start, u64 owner_root, int level)
6153 unsigned long len = fs_info->nodesize;
6156 unsigned long index = start >> PAGE_SHIFT;
6157 struct extent_buffer *eb;
6158 struct extent_buffer *exists = NULL;
6160 struct address_space *mapping = fs_info->btree_inode->i_mapping;
6164 if (check_eb_alignment(fs_info, start))
6165 return ERR_PTR(-EINVAL);
6167 #if BITS_PER_LONG == 32
6168 if (start >= MAX_LFS_FILESIZE) {
6169 btrfs_err_rl(fs_info,
6170 "extent buffer %llu is beyond 32bit page cache limit", start);
6171 btrfs_err_32bit_limit(fs_info);
6172 return ERR_PTR(-EOVERFLOW);
6174 if (start >= BTRFS_32BIT_EARLY_WARN_THRESHOLD)
6175 btrfs_warn_32bit_limit(fs_info);
6178 eb = find_extent_buffer(fs_info, start);
6182 eb = __alloc_extent_buffer(fs_info, start, len);
6184 return ERR_PTR(-ENOMEM);
6185 btrfs_set_buffer_lockdep_class(owner_root, eb, level);
6187 num_pages = num_extent_pages(eb);
6188 for (i = 0; i < num_pages; i++, index++) {
6189 struct btrfs_subpage *prealloc = NULL;
6191 p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL);
6193 exists = ERR_PTR(-ENOMEM);
6198 * Preallocate page->private for subpage case, so that we won't
6199 * allocate memory with private_lock hold. The memory will be
6200 * freed by attach_extent_buffer_page() or freed manually if
6203 * Although we have ensured one subpage eb can only have one
6204 * page, but it may change in the future for 16K page size
6205 * support, so we still preallocate the memory in the loop.
6207 if (fs_info->nodesize < PAGE_SIZE) {
6208 prealloc = btrfs_alloc_subpage(fs_info, BTRFS_SUBPAGE_METADATA);
6209 if (IS_ERR(prealloc)) {
6210 ret = PTR_ERR(prealloc);
6213 exists = ERR_PTR(ret);
6218 spin_lock(&mapping->private_lock);
6219 exists = grab_extent_buffer(fs_info, p);
6221 spin_unlock(&mapping->private_lock);
6224 mark_extent_buffer_accessed(exists, p);
6225 btrfs_free_subpage(prealloc);
6228 /* Should not fail, as we have preallocated the memory */
6229 ret = attach_extent_buffer_page(eb, p, prealloc);
6232 * To inform we have extra eb under allocation, so that
6233 * detach_extent_buffer_page() won't release the page private
6234 * when the eb hasn't yet been inserted into radix tree.
6236 * The ref will be decreased when the eb released the page, in
6237 * detach_extent_buffer_page().
6238 * Thus needs no special handling in error path.
6240 btrfs_page_inc_eb_refs(fs_info, p);
6241 spin_unlock(&mapping->private_lock);
6243 WARN_ON(btrfs_page_test_dirty(fs_info, p, eb->start, eb->len));
6245 if (!PageUptodate(p))
6249 * We can't unlock the pages just yet since the extent buffer
6250 * hasn't been properly inserted in the radix tree, this
6251 * opens a race with btree_release_folio which can free a page
6252 * while we are still filling in all pages for the buffer and
6257 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
6259 ret = radix_tree_preload(GFP_NOFS);
6261 exists = ERR_PTR(ret);
6265 spin_lock(&fs_info->buffer_lock);
6266 ret = radix_tree_insert(&fs_info->buffer_radix,
6267 start >> fs_info->sectorsize_bits, eb);
6268 spin_unlock(&fs_info->buffer_lock);
6269 radix_tree_preload_end();
6270 if (ret == -EEXIST) {
6271 exists = find_extent_buffer(fs_info, start);
6277 /* add one reference for the tree */
6278 check_buffer_tree_ref(eb);
6279 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
6282 * Now it's safe to unlock the pages because any calls to
6283 * btree_release_folio will correctly detect that a page belongs to a
6284 * live buffer and won't free them prematurely.
6286 for (i = 0; i < num_pages; i++)
6287 unlock_page(eb->pages[i]);
6291 WARN_ON(!atomic_dec_and_test(&eb->refs));
6292 for (i = 0; i < num_pages; i++) {
6294 unlock_page(eb->pages[i]);
6297 btrfs_release_extent_buffer(eb);
6301 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
6303 struct extent_buffer *eb =
6304 container_of(head, struct extent_buffer, rcu_head);
6306 __free_extent_buffer(eb);
6309 static int release_extent_buffer(struct extent_buffer *eb)
6310 __releases(&eb->refs_lock)
6312 lockdep_assert_held(&eb->refs_lock);
6314 WARN_ON(atomic_read(&eb->refs) == 0);
6315 if (atomic_dec_and_test(&eb->refs)) {
6316 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
6317 struct btrfs_fs_info *fs_info = eb->fs_info;
6319 spin_unlock(&eb->refs_lock);
6321 spin_lock(&fs_info->buffer_lock);
6322 radix_tree_delete(&fs_info->buffer_radix,
6323 eb->start >> fs_info->sectorsize_bits);
6324 spin_unlock(&fs_info->buffer_lock);
6326 spin_unlock(&eb->refs_lock);
6329 btrfs_leak_debug_del(&eb->fs_info->eb_leak_lock, &eb->leak_list);
6330 /* Should be safe to release our pages at this point */
6331 btrfs_release_extent_buffer_pages(eb);
6332 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
6333 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags))) {
6334 __free_extent_buffer(eb);
6338 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
6341 spin_unlock(&eb->refs_lock);
6346 void free_extent_buffer(struct extent_buffer *eb)
6354 refs = atomic_read(&eb->refs);
6355 if ((!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && refs <= 3)
6356 || (test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) &&
6359 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
6364 spin_lock(&eb->refs_lock);
6365 if (atomic_read(&eb->refs) == 2 &&
6366 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
6367 !extent_buffer_under_io(eb) &&
6368 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
6369 atomic_dec(&eb->refs);
6372 * I know this is terrible, but it's temporary until we stop tracking
6373 * the uptodate bits and such for the extent buffers.
6375 release_extent_buffer(eb);
6378 void free_extent_buffer_stale(struct extent_buffer *eb)
6383 spin_lock(&eb->refs_lock);
6384 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
6386 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
6387 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
6388 atomic_dec(&eb->refs);
6389 release_extent_buffer(eb);
6392 static void btree_clear_page_dirty(struct page *page)
6394 ASSERT(PageDirty(page));
6395 ASSERT(PageLocked(page));
6396 clear_page_dirty_for_io(page);
6397 xa_lock_irq(&page->mapping->i_pages);
6398 if (!PageDirty(page))
6399 __xa_clear_mark(&page->mapping->i_pages,
6400 page_index(page), PAGECACHE_TAG_DIRTY);
6401 xa_unlock_irq(&page->mapping->i_pages);
6404 static void clear_subpage_extent_buffer_dirty(const struct extent_buffer *eb)
6406 struct btrfs_fs_info *fs_info = eb->fs_info;
6407 struct page *page = eb->pages[0];
6410 /* btree_clear_page_dirty() needs page locked */
6412 last = btrfs_subpage_clear_and_test_dirty(fs_info, page, eb->start,
6415 btree_clear_page_dirty(page);
6417 WARN_ON(atomic_read(&eb->refs) == 0);
6420 void clear_extent_buffer_dirty(const struct extent_buffer *eb)
6426 if (eb->fs_info->nodesize < PAGE_SIZE)
6427 return clear_subpage_extent_buffer_dirty(eb);
6429 num_pages = num_extent_pages(eb);
6431 for (i = 0; i < num_pages; i++) {
6432 page = eb->pages[i];
6433 if (!PageDirty(page))
6436 btree_clear_page_dirty(page);
6437 ClearPageError(page);
6440 WARN_ON(atomic_read(&eb->refs) == 0);
6443 bool set_extent_buffer_dirty(struct extent_buffer *eb)
6449 check_buffer_tree_ref(eb);
6451 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
6453 num_pages = num_extent_pages(eb);
6454 WARN_ON(atomic_read(&eb->refs) == 0);
6455 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
6458 bool subpage = eb->fs_info->nodesize < PAGE_SIZE;
6461 * For subpage case, we can have other extent buffers in the
6462 * same page, and in clear_subpage_extent_buffer_dirty() we
6463 * have to clear page dirty without subpage lock held.
6464 * This can cause race where our page gets dirty cleared after
6467 * Thankfully, clear_subpage_extent_buffer_dirty() has locked
6468 * its page for other reasons, we can use page lock to prevent
6472 lock_page(eb->pages[0]);
6473 for (i = 0; i < num_pages; i++)
6474 btrfs_page_set_dirty(eb->fs_info, eb->pages[i],
6475 eb->start, eb->len);
6477 unlock_page(eb->pages[0]);
6479 #ifdef CONFIG_BTRFS_DEBUG
6480 for (i = 0; i < num_pages; i++)
6481 ASSERT(PageDirty(eb->pages[i]));
6487 void clear_extent_buffer_uptodate(struct extent_buffer *eb)
6489 struct btrfs_fs_info *fs_info = eb->fs_info;
6494 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
6495 num_pages = num_extent_pages(eb);
6496 for (i = 0; i < num_pages; i++) {
6497 page = eb->pages[i];
6502 * This is special handling for metadata subpage, as regular
6503 * btrfs_is_subpage() can not handle cloned/dummy metadata.
6505 if (fs_info->nodesize >= PAGE_SIZE)
6506 ClearPageUptodate(page);
6508 btrfs_subpage_clear_uptodate(fs_info, page, eb->start,
6513 void set_extent_buffer_uptodate(struct extent_buffer *eb)
6515 struct btrfs_fs_info *fs_info = eb->fs_info;
6520 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
6521 num_pages = num_extent_pages(eb);
6522 for (i = 0; i < num_pages; i++) {
6523 page = eb->pages[i];
6526 * This is special handling for metadata subpage, as regular
6527 * btrfs_is_subpage() can not handle cloned/dummy metadata.
6529 if (fs_info->nodesize >= PAGE_SIZE)
6530 SetPageUptodate(page);
6532 btrfs_subpage_set_uptodate(fs_info, page, eb->start,
6537 static int read_extent_buffer_subpage(struct extent_buffer *eb, int wait,
6540 struct btrfs_fs_info *fs_info = eb->fs_info;
6541 struct extent_io_tree *io_tree;
6542 struct page *page = eb->pages[0];
6543 struct btrfs_bio_ctrl bio_ctrl = {
6544 .mirror_num = mirror_num,
6548 ASSERT(!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags));
6549 ASSERT(PagePrivate(page));
6550 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
6552 if (wait == WAIT_NONE) {
6553 if (!try_lock_extent(io_tree, eb->start, eb->start + eb->len - 1))
6556 ret = lock_extent(io_tree, eb->start, eb->start + eb->len - 1);
6562 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags) ||
6563 PageUptodate(page) ||
6564 btrfs_subpage_test_uptodate(fs_info, page, eb->start, eb->len)) {
6565 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
6566 unlock_extent(io_tree, eb->start, eb->start + eb->len - 1);
6570 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
6571 eb->read_mirror = 0;
6572 atomic_set(&eb->io_pages, 1);
6573 check_buffer_tree_ref(eb);
6574 btrfs_subpage_clear_error(fs_info, page, eb->start, eb->len);
6576 btrfs_subpage_start_reader(fs_info, page, eb->start, eb->len);
6577 ret = submit_extent_page(REQ_OP_READ, NULL, &bio_ctrl,
6578 page, eb->start, eb->len,
6579 eb->start - page_offset(page),
6580 end_bio_extent_readpage, 0, true);
6583 * In the endio function, if we hit something wrong we will
6584 * increase the io_pages, so here we need to decrease it for
6587 atomic_dec(&eb->io_pages);
6589 submit_one_bio(&bio_ctrl);
6590 if (ret || wait != WAIT_COMPLETE)
6593 wait_extent_bit(io_tree, eb->start, eb->start + eb->len - 1, EXTENT_LOCKED);
6594 if (!test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
6599 int read_extent_buffer_pages(struct extent_buffer *eb, int wait, int mirror_num)
6605 int locked_pages = 0;
6606 int all_uptodate = 1;
6608 unsigned long num_reads = 0;
6609 struct btrfs_bio_ctrl bio_ctrl = {
6610 .mirror_num = mirror_num,
6613 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
6617 * We could have had EXTENT_BUFFER_UPTODATE cleared by the write
6618 * operation, which could potentially still be in flight. In this case
6619 * we simply want to return an error.
6621 if (unlikely(test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)))
6624 if (eb->fs_info->nodesize < PAGE_SIZE)
6625 return read_extent_buffer_subpage(eb, wait, mirror_num);
6627 num_pages = num_extent_pages(eb);
6628 for (i = 0; i < num_pages; i++) {
6629 page = eb->pages[i];
6630 if (wait == WAIT_NONE) {
6632 * WAIT_NONE is only utilized by readahead. If we can't
6633 * acquire the lock atomically it means either the eb
6634 * is being read out or under modification.
6635 * Either way the eb will be or has been cached,
6636 * readahead can exit safely.
6638 if (!trylock_page(page))
6646 * We need to firstly lock all pages to make sure that
6647 * the uptodate bit of our pages won't be affected by
6648 * clear_extent_buffer_uptodate().
6650 for (i = 0; i < num_pages; i++) {
6651 page = eb->pages[i];
6652 if (!PageUptodate(page)) {
6659 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
6663 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
6664 eb->read_mirror = 0;
6665 atomic_set(&eb->io_pages, num_reads);
6667 * It is possible for release_folio to clear the TREE_REF bit before we
6668 * set io_pages. See check_buffer_tree_ref for a more detailed comment.
6670 check_buffer_tree_ref(eb);
6671 for (i = 0; i < num_pages; i++) {
6672 page = eb->pages[i];
6674 if (!PageUptodate(page)) {
6676 atomic_dec(&eb->io_pages);
6681 ClearPageError(page);
6682 err = submit_extent_page(REQ_OP_READ, NULL,
6683 &bio_ctrl, page, page_offset(page),
6684 PAGE_SIZE, 0, end_bio_extent_readpage,
6688 * We failed to submit the bio so it's the
6689 * caller's responsibility to perform cleanup
6690 * i.e unlock page/set error bit.
6695 atomic_dec(&eb->io_pages);
6702 submit_one_bio(&bio_ctrl);
6704 if (ret || wait != WAIT_COMPLETE)
6707 for (i = 0; i < num_pages; i++) {
6708 page = eb->pages[i];
6709 wait_on_page_locked(page);
6710 if (!PageUptodate(page))
6717 while (locked_pages > 0) {
6719 page = eb->pages[locked_pages];
6725 static bool report_eb_range(const struct extent_buffer *eb, unsigned long start,
6728 btrfs_warn(eb->fs_info,
6729 "access to eb bytenr %llu len %lu out of range start %lu len %lu",
6730 eb->start, eb->len, start, len);
6731 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
6737 * Check if the [start, start + len) range is valid before reading/writing
6739 * NOTE: @start and @len are offset inside the eb, not logical address.
6741 * Caller should not touch the dst/src memory if this function returns error.
6743 static inline int check_eb_range(const struct extent_buffer *eb,
6744 unsigned long start, unsigned long len)
6746 unsigned long offset;
6748 /* start, start + len should not go beyond eb->len nor overflow */
6749 if (unlikely(check_add_overflow(start, len, &offset) || offset > eb->len))
6750 return report_eb_range(eb, start, len);
6755 void read_extent_buffer(const struct extent_buffer *eb, void *dstv,
6756 unsigned long start, unsigned long len)
6762 char *dst = (char *)dstv;
6763 unsigned long i = get_eb_page_index(start);
6765 if (check_eb_range(eb, start, len))
6768 offset = get_eb_offset_in_page(eb, start);
6771 page = eb->pages[i];
6773 cur = min(len, (PAGE_SIZE - offset));
6774 kaddr = page_address(page);
6775 memcpy(dst, kaddr + offset, cur);
6784 int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb,
6786 unsigned long start, unsigned long len)
6792 char __user *dst = (char __user *)dstv;
6793 unsigned long i = get_eb_page_index(start);
6796 WARN_ON(start > eb->len);
6797 WARN_ON(start + len > eb->start + eb->len);
6799 offset = get_eb_offset_in_page(eb, start);
6802 page = eb->pages[i];
6804 cur = min(len, (PAGE_SIZE - offset));
6805 kaddr = page_address(page);
6806 if (copy_to_user_nofault(dst, kaddr + offset, cur)) {
6820 int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,
6821 unsigned long start, unsigned long len)
6827 char *ptr = (char *)ptrv;
6828 unsigned long i = get_eb_page_index(start);
6831 if (check_eb_range(eb, start, len))
6834 offset = get_eb_offset_in_page(eb, start);
6837 page = eb->pages[i];
6839 cur = min(len, (PAGE_SIZE - offset));
6841 kaddr = page_address(page);
6842 ret = memcmp(ptr, kaddr + offset, cur);
6855 * Check that the extent buffer is uptodate.
6857 * For regular sector size == PAGE_SIZE case, check if @page is uptodate.
6858 * For subpage case, check if the range covered by the eb has EXTENT_UPTODATE.
6860 static void assert_eb_page_uptodate(const struct extent_buffer *eb,
6863 struct btrfs_fs_info *fs_info = eb->fs_info;
6866 * If we are using the commit root we could potentially clear a page
6867 * Uptodate while we're using the extent buffer that we've previously
6868 * looked up. We don't want to complain in this case, as the page was
6869 * valid before, we just didn't write it out. Instead we want to catch
6870 * the case where we didn't actually read the block properly, which
6871 * would have !PageUptodate && !PageError, as we clear PageError before
6874 if (fs_info->nodesize < PAGE_SIZE) {
6875 bool uptodate, error;
6877 uptodate = btrfs_subpage_test_uptodate(fs_info, page,
6878 eb->start, eb->len);
6879 error = btrfs_subpage_test_error(fs_info, page, eb->start, eb->len);
6880 WARN_ON(!uptodate && !error);
6882 WARN_ON(!PageUptodate(page) && !PageError(page));
6886 void write_extent_buffer_chunk_tree_uuid(const struct extent_buffer *eb,
6891 assert_eb_page_uptodate(eb, eb->pages[0]);
6892 kaddr = page_address(eb->pages[0]) +
6893 get_eb_offset_in_page(eb, offsetof(struct btrfs_header,
6895 memcpy(kaddr, srcv, BTRFS_FSID_SIZE);
6898 void write_extent_buffer_fsid(const struct extent_buffer *eb, const void *srcv)
6902 assert_eb_page_uptodate(eb, eb->pages[0]);
6903 kaddr = page_address(eb->pages[0]) +
6904 get_eb_offset_in_page(eb, offsetof(struct btrfs_header, fsid));
6905 memcpy(kaddr, srcv, BTRFS_FSID_SIZE);
6908 void write_extent_buffer(const struct extent_buffer *eb, const void *srcv,
6909 unsigned long start, unsigned long len)
6915 char *src = (char *)srcv;
6916 unsigned long i = get_eb_page_index(start);
6918 WARN_ON(test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags));
6920 if (check_eb_range(eb, start, len))
6923 offset = get_eb_offset_in_page(eb, start);
6926 page = eb->pages[i];
6927 assert_eb_page_uptodate(eb, page);
6929 cur = min(len, PAGE_SIZE - offset);
6930 kaddr = page_address(page);
6931 memcpy(kaddr + offset, src, cur);
6940 void memzero_extent_buffer(const struct extent_buffer *eb, unsigned long start,
6947 unsigned long i = get_eb_page_index(start);
6949 if (check_eb_range(eb, start, len))
6952 offset = get_eb_offset_in_page(eb, start);
6955 page = eb->pages[i];
6956 assert_eb_page_uptodate(eb, page);
6958 cur = min(len, PAGE_SIZE - offset);
6959 kaddr = page_address(page);
6960 memset(kaddr + offset, 0, cur);
6968 void copy_extent_buffer_full(const struct extent_buffer *dst,
6969 const struct extent_buffer *src)
6974 ASSERT(dst->len == src->len);
6976 if (dst->fs_info->nodesize >= PAGE_SIZE) {
6977 num_pages = num_extent_pages(dst);
6978 for (i = 0; i < num_pages; i++)
6979 copy_page(page_address(dst->pages[i]),
6980 page_address(src->pages[i]));
6982 size_t src_offset = get_eb_offset_in_page(src, 0);
6983 size_t dst_offset = get_eb_offset_in_page(dst, 0);
6985 ASSERT(src->fs_info->nodesize < PAGE_SIZE);
6986 memcpy(page_address(dst->pages[0]) + dst_offset,
6987 page_address(src->pages[0]) + src_offset,
6992 void copy_extent_buffer(const struct extent_buffer *dst,
6993 const struct extent_buffer *src,
6994 unsigned long dst_offset, unsigned long src_offset,
6997 u64 dst_len = dst->len;
7002 unsigned long i = get_eb_page_index(dst_offset);
7004 if (check_eb_range(dst, dst_offset, len) ||
7005 check_eb_range(src, src_offset, len))
7008 WARN_ON(src->len != dst_len);
7010 offset = get_eb_offset_in_page(dst, dst_offset);
7013 page = dst->pages[i];
7014 assert_eb_page_uptodate(dst, page);
7016 cur = min(len, (unsigned long)(PAGE_SIZE - offset));
7018 kaddr = page_address(page);
7019 read_extent_buffer(src, kaddr + offset, src_offset, cur);
7029 * eb_bitmap_offset() - calculate the page and offset of the byte containing the
7031 * @eb: the extent buffer
7032 * @start: offset of the bitmap item in the extent buffer
7034 * @page_index: return index of the page in the extent buffer that contains the
7036 * @page_offset: return offset into the page given by page_index
7038 * This helper hides the ugliness of finding the byte in an extent buffer which
7039 * contains a given bit.
7041 static inline void eb_bitmap_offset(const struct extent_buffer *eb,
7042 unsigned long start, unsigned long nr,
7043 unsigned long *page_index,
7044 size_t *page_offset)
7046 size_t byte_offset = BIT_BYTE(nr);
7050 * The byte we want is the offset of the extent buffer + the offset of
7051 * the bitmap item in the extent buffer + the offset of the byte in the
7054 offset = start + offset_in_page(eb->start) + byte_offset;
7056 *page_index = offset >> PAGE_SHIFT;
7057 *page_offset = offset_in_page(offset);
7061 * extent_buffer_test_bit - determine whether a bit in a bitmap item is set
7062 * @eb: the extent buffer
7063 * @start: offset of the bitmap item in the extent buffer
7064 * @nr: bit number to test
7066 int extent_buffer_test_bit(const struct extent_buffer *eb, unsigned long start,
7074 eb_bitmap_offset(eb, start, nr, &i, &offset);
7075 page = eb->pages[i];
7076 assert_eb_page_uptodate(eb, page);
7077 kaddr = page_address(page);
7078 return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1)));
7082 * extent_buffer_bitmap_set - set an area of a bitmap
7083 * @eb: the extent buffer
7084 * @start: offset of the bitmap item in the extent buffer
7085 * @pos: bit number of the first bit
7086 * @len: number of bits to set
7088 void extent_buffer_bitmap_set(const struct extent_buffer *eb, unsigned long start,
7089 unsigned long pos, unsigned long len)
7095 const unsigned int size = pos + len;
7096 int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
7097 u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos);
7099 eb_bitmap_offset(eb, start, pos, &i, &offset);
7100 page = eb->pages[i];
7101 assert_eb_page_uptodate(eb, page);
7102 kaddr = page_address(page);
7104 while (len >= bits_to_set) {
7105 kaddr[offset] |= mask_to_set;
7107 bits_to_set = BITS_PER_BYTE;
7109 if (++offset >= PAGE_SIZE && len > 0) {
7111 page = eb->pages[++i];
7112 assert_eb_page_uptodate(eb, page);
7113 kaddr = page_address(page);
7117 mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
7118 kaddr[offset] |= mask_to_set;
7124 * extent_buffer_bitmap_clear - clear an area of a bitmap
7125 * @eb: the extent buffer
7126 * @start: offset of the bitmap item in the extent buffer
7127 * @pos: bit number of the first bit
7128 * @len: number of bits to clear
7130 void extent_buffer_bitmap_clear(const struct extent_buffer *eb,
7131 unsigned long start, unsigned long pos,
7138 const unsigned int size = pos + len;
7139 int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
7140 u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos);
7142 eb_bitmap_offset(eb, start, pos, &i, &offset);
7143 page = eb->pages[i];
7144 assert_eb_page_uptodate(eb, page);
7145 kaddr = page_address(page);
7147 while (len >= bits_to_clear) {
7148 kaddr[offset] &= ~mask_to_clear;
7149 len -= bits_to_clear;
7150 bits_to_clear = BITS_PER_BYTE;
7152 if (++offset >= PAGE_SIZE && len > 0) {
7154 page = eb->pages[++i];
7155 assert_eb_page_uptodate(eb, page);
7156 kaddr = page_address(page);
7160 mask_to_clear &= BITMAP_LAST_BYTE_MASK(size);
7161 kaddr[offset] &= ~mask_to_clear;
7165 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
7167 unsigned long distance = (src > dst) ? src - dst : dst - src;
7168 return distance < len;
7171 static void copy_pages(struct page *dst_page, struct page *src_page,
7172 unsigned long dst_off, unsigned long src_off,
7175 char *dst_kaddr = page_address(dst_page);
7177 int must_memmove = 0;
7179 if (dst_page != src_page) {
7180 src_kaddr = page_address(src_page);
7182 src_kaddr = dst_kaddr;
7183 if (areas_overlap(src_off, dst_off, len))
7188 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
7190 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
7193 void memcpy_extent_buffer(const struct extent_buffer *dst,
7194 unsigned long dst_offset, unsigned long src_offset,
7198 size_t dst_off_in_page;
7199 size_t src_off_in_page;
7200 unsigned long dst_i;
7201 unsigned long src_i;
7203 if (check_eb_range(dst, dst_offset, len) ||
7204 check_eb_range(dst, src_offset, len))
7208 dst_off_in_page = get_eb_offset_in_page(dst, dst_offset);
7209 src_off_in_page = get_eb_offset_in_page(dst, src_offset);
7211 dst_i = get_eb_page_index(dst_offset);
7212 src_i = get_eb_page_index(src_offset);
7214 cur = min(len, (unsigned long)(PAGE_SIZE -
7216 cur = min_t(unsigned long, cur,
7217 (unsigned long)(PAGE_SIZE - dst_off_in_page));
7219 copy_pages(dst->pages[dst_i], dst->pages[src_i],
7220 dst_off_in_page, src_off_in_page, cur);
7228 void memmove_extent_buffer(const struct extent_buffer *dst,
7229 unsigned long dst_offset, unsigned long src_offset,
7233 size_t dst_off_in_page;
7234 size_t src_off_in_page;
7235 unsigned long dst_end = dst_offset + len - 1;
7236 unsigned long src_end = src_offset + len - 1;
7237 unsigned long dst_i;
7238 unsigned long src_i;
7240 if (check_eb_range(dst, dst_offset, len) ||
7241 check_eb_range(dst, src_offset, len))
7243 if (dst_offset < src_offset) {
7244 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
7248 dst_i = get_eb_page_index(dst_end);
7249 src_i = get_eb_page_index(src_end);
7251 dst_off_in_page = get_eb_offset_in_page(dst, dst_end);
7252 src_off_in_page = get_eb_offset_in_page(dst, src_end);
7254 cur = min_t(unsigned long, len, src_off_in_page + 1);
7255 cur = min(cur, dst_off_in_page + 1);
7256 copy_pages(dst->pages[dst_i], dst->pages[src_i],
7257 dst_off_in_page - cur + 1,
7258 src_off_in_page - cur + 1, cur);
7266 #define GANG_LOOKUP_SIZE 16
7267 static struct extent_buffer *get_next_extent_buffer(
7268 struct btrfs_fs_info *fs_info, struct page *page, u64 bytenr)
7270 struct extent_buffer *gang[GANG_LOOKUP_SIZE];
7271 struct extent_buffer *found = NULL;
7272 u64 page_start = page_offset(page);
7273 u64 cur = page_start;
7275 ASSERT(in_range(bytenr, page_start, PAGE_SIZE));
7276 lockdep_assert_held(&fs_info->buffer_lock);
7278 while (cur < page_start + PAGE_SIZE) {
7282 ret = radix_tree_gang_lookup(&fs_info->buffer_radix,
7283 (void **)gang, cur >> fs_info->sectorsize_bits,
7284 min_t(unsigned int, GANG_LOOKUP_SIZE,
7285 PAGE_SIZE / fs_info->nodesize));
7288 for (i = 0; i < ret; i++) {
7289 /* Already beyond page end */
7290 if (gang[i]->start >= page_start + PAGE_SIZE)
7293 if (gang[i]->start >= bytenr) {
7298 cur = gang[ret - 1]->start + gang[ret - 1]->len;
7304 static int try_release_subpage_extent_buffer(struct page *page)
7306 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
7307 u64 cur = page_offset(page);
7308 const u64 end = page_offset(page) + PAGE_SIZE;
7312 struct extent_buffer *eb = NULL;
7315 * Unlike try_release_extent_buffer() which uses page->private
7316 * to grab buffer, for subpage case we rely on radix tree, thus
7317 * we need to ensure radix tree consistency.
7319 * We also want an atomic snapshot of the radix tree, thus go
7320 * with spinlock rather than RCU.
7322 spin_lock(&fs_info->buffer_lock);
7323 eb = get_next_extent_buffer(fs_info, page, cur);
7325 /* No more eb in the page range after or at cur */
7326 spin_unlock(&fs_info->buffer_lock);
7329 cur = eb->start + eb->len;
7332 * The same as try_release_extent_buffer(), to ensure the eb
7333 * won't disappear out from under us.
7335 spin_lock(&eb->refs_lock);
7336 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
7337 spin_unlock(&eb->refs_lock);
7338 spin_unlock(&fs_info->buffer_lock);
7341 spin_unlock(&fs_info->buffer_lock);
7344 * If tree ref isn't set then we know the ref on this eb is a
7345 * real ref, so just return, this eb will likely be freed soon
7348 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
7349 spin_unlock(&eb->refs_lock);
7354 * Here we don't care about the return value, we will always
7355 * check the page private at the end. And
7356 * release_extent_buffer() will release the refs_lock.
7358 release_extent_buffer(eb);
7361 * Finally to check if we have cleared page private, as if we have
7362 * released all ebs in the page, the page private should be cleared now.
7364 spin_lock(&page->mapping->private_lock);
7365 if (!PagePrivate(page))
7369 spin_unlock(&page->mapping->private_lock);
7374 int try_release_extent_buffer(struct page *page)
7376 struct extent_buffer *eb;
7378 if (btrfs_sb(page->mapping->host->i_sb)->nodesize < PAGE_SIZE)
7379 return try_release_subpage_extent_buffer(page);
7382 * We need to make sure nobody is changing page->private, as we rely on
7383 * page->private as the pointer to extent buffer.
7385 spin_lock(&page->mapping->private_lock);
7386 if (!PagePrivate(page)) {
7387 spin_unlock(&page->mapping->private_lock);
7391 eb = (struct extent_buffer *)page->private;
7395 * This is a little awful but should be ok, we need to make sure that
7396 * the eb doesn't disappear out from under us while we're looking at
7399 spin_lock(&eb->refs_lock);
7400 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
7401 spin_unlock(&eb->refs_lock);
7402 spin_unlock(&page->mapping->private_lock);
7405 spin_unlock(&page->mapping->private_lock);
7408 * If tree ref isn't set then we know the ref on this eb is a real ref,
7409 * so just return, this page will likely be freed soon anyway.
7411 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
7412 spin_unlock(&eb->refs_lock);
7416 return release_extent_buffer(eb);
7420 * btrfs_readahead_tree_block - attempt to readahead a child block
7421 * @fs_info: the fs_info
7422 * @bytenr: bytenr to read
7423 * @owner_root: objectid of the root that owns this eb
7424 * @gen: generation for the uptodate check, can be 0
7425 * @level: level for the eb
7427 * Attempt to readahead a tree block at @bytenr. If @gen is 0 then we do a
7428 * normal uptodate check of the eb, without checking the generation. If we have
7429 * to read the block we will not block on anything.
7431 void btrfs_readahead_tree_block(struct btrfs_fs_info *fs_info,
7432 u64 bytenr, u64 owner_root, u64 gen, int level)
7434 struct extent_buffer *eb;
7437 eb = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level);
7441 if (btrfs_buffer_uptodate(eb, gen, 1)) {
7442 free_extent_buffer(eb);
7446 ret = read_extent_buffer_pages(eb, WAIT_NONE, 0);
7448 free_extent_buffer_stale(eb);
7450 free_extent_buffer(eb);
7454 * btrfs_readahead_node_child - readahead a node's child block
7455 * @node: parent node we're reading from
7456 * @slot: slot in the parent node for the child we want to read
7458 * A helper for btrfs_readahead_tree_block, we simply read the bytenr pointed at
7459 * the slot in the node provided.
7461 void btrfs_readahead_node_child(struct extent_buffer *node, int slot)
7463 btrfs_readahead_tree_block(node->fs_info,
7464 btrfs_node_blockptr(node, slot),
7465 btrfs_header_owner(node),
7466 btrfs_node_ptr_generation(node, slot),
7467 btrfs_header_level(node) - 1);